Update drm expected version.

Fix typo in 3D image layout for i915.

progs/demos/Makefile

@@ -47,6 +47,7 @@ PROGS = \
 	renormal \
 	shadowtex \
 	singlebuffer \
+	streaming_rect \
 	spectex \
 	spriteblast \
 	stex3d \

progs/demos/streaming_rect.c

@@ -0,0 +1,407 @@
+
+/*
+ * GL_ARB_multitexture demo
+ *
+ * Command line options:
+ *    -info      print GL implementation information
+ *
+ *
+ * Brian Paul  November 1998  This program is in the public domain.
+ * Modified on 12 Feb 2002 for > 2 texture units.
+ */
+
+#define GL_GLEXT_PROTOTYPES
+
+#include <math.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <GL/glut.h>
+
+#include "readtex.h"
+
+enum {
+   WHOLERECT,
+   RENDER,
+   SWAPBUF,
+
+   UPLOAD_CONVENTIONAL,
+   UPLOAD_PBO_NAIVE,
+   UPLOAD_PBO_PINGPONG,
+   UPLOAD_PBO_WITH_RESET,
+   UPLOAD_NOOP,
+
+   QUIT
+};
+
+static GLint upload_style = UPLOAD_CONVENTIONAL;
+static GLboolean whole_rect = 1;
+static GLboolean do_render = 1;
+static GLboolean do_swapbuffers = 1;
+
+
+static GLuint col = 0x0;
+static GLfloat Xrot = 20.0, Yrot = 30.0;
+
+static GLuint Width = 1024;
+static GLuint Height = 512;
+
+#define NR_PBO 2
+static GLuint texObj, DrawPBO[NR_PBO];
+
+
+static void Idle( void )
+{
+   col++;
+   
+   glutPostRedisplay();
+}
+
+static int min( int a, int b ) { return a < b ? a : b; }
+
+static void DrawObject()
+{
+   GLint size = Width * Height * 4;
+   static char *static_image = NULL;
+   static int current = 0;
+
+   switch (upload_style) {
+   case UPLOAD_PBO_NAIVE:
+      /* Continually upload from the one pbo without any attempt to
+       * decouple from hardware trying to access the same data:
+       */
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, DrawPBO[0]);
+
+      {
+	 char *image = glMapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, GL_WRITE_ONLY_ARB);
+	 memset(image, col&0xff, size);
+	 glUnmapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT);
+      }
+   
+      glTexImage2D(GL_TEXTURE_RECTANGLE_ARB, 0, GL_RGBA, Width, Height, 0,
+		   GL_BGRA, GL_UNSIGNED_BYTE, NULL);
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, 0);
+      break;
+
+   case UPLOAD_PBO_PINGPONG:
+      /* Cycle through the available upload buffers to avoid waiting
+       * on buffers currently enqueued on the hardware.  Avoids
+       * waiting on buffer contents that may still be referenced by
+       * hardware.
+       */
+      current = (current + 1) % NR_PBO;
+
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, DrawPBO[current]);
+
+      {
+	 char *image = glMapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, GL_WRITE_ONLY_ARB);
+	 memset(image, col&0xff, size);
+	 glUnmapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT);
+      }
+   
+      glTexImage2D(GL_TEXTURE_RECTANGLE_ARB, 0, GL_RGBA, Width, Height, 0,
+		   GL_BGRA, GL_UNSIGNED_BYTE, NULL);
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, 0);
+      break;
+
+
+   case UPLOAD_PBO_WITH_RESET:
+      /* Alternate approach, uses BufferDataARB(NULL) to explicitly
+       * release the old image contents from the pbo prior to upload.
+       * Similar effect to double buffering pbos as above, but a
+       * little more subtle as the implications of the NULL data are
+       * not entirely clear.  I don't know if all drivers will take
+       * full advantage of the optimization or not.
+       */
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, DrawPBO[0]);
+
+      /* XXX: This is extremely important - semantically makes the buffer
+       * contents undefined, but in practice means that the driver can
+       * release the old copy of the texture and allocate a new one
+       * without waiting for outstanding rendering to complete.
+       */
+      glBufferDataARB(GL_PIXEL_UNPACK_BUFFER_EXT, size, NULL, GL_STREAM_DRAW_ARB);
+
+      {
+	 char *image = glMapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, GL_WRITE_ONLY_ARB);
+	 memset(image, col&0xff, size);
+	 glUnmapBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT);
+      }
+   
+      glTexImage2D(GL_TEXTURE_RECTANGLE_ARB, 0, GL_RGBA, Width, Height, 0,
+		   GL_BGRA, GL_UNSIGNED_BYTE, NULL);
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, 0);
+      break;
+
+
+   case UPLOAD_CONVENTIONAL: 
+      if (static_image == NULL) 
+	 static_image = malloc(size);
+
+      memset(static_image, col&0xff, size);
+
+      /* BGRA should be the fast path for regular uploads as well.
+       */
+      glTexImage2D(GL_TEXTURE_RECTANGLE_ARB, 0, GL_RGBA, Width, Height, 0,
+		   GL_BGRA, GL_UNSIGNED_BYTE, static_image);
+      break;
+
+   case UPLOAD_NOOP:
+      break;
+   }
+
+   if (do_render)
+   {
+      int x,y,w,h;
+
+      if (whole_rect) {
+	 x = y = 0;
+	 w = Width;
+	 h = Height;
+      }
+      else {
+	 x = y = 0;
+	 w = min(10, Width);
+	 h = min(10, Height);
+      }
+
+      glBegin(GL_QUADS);
+
+      glTexCoord2f( x, y);
+      glVertex2f( x, y );
+
+      glTexCoord2f( x, y + h);
+      glVertex2f( x, y + h);
+
+      glTexCoord2f( x + w, y + h);
+      glVertex2f( x + w, y + h );
+
+      glTexCoord2f( x + w, y);
+      glVertex2f( x + w, y );
+
+      glEnd();
+   }
+}
+
+
+
+static void Display( void )
+{
+   static GLint T0 = 0;
+   static GLint Frames = 0;
+   GLint t;
+
+   glClear( GL_COLOR_BUFFER_BIT );
+
+   glPushMatrix();
+      DrawObject();
+   glPopMatrix();
+
+   if (do_swapbuffers) 
+      glutSwapBuffers();
+   else
+      glFlush();
+
+   Frames++;
+
+   t = glutGet(GLUT_ELAPSED_TIME);
+   if (t - T0 >= 1000) {
+      GLfloat seconds = (t - T0) / 1000.0;
+
+      GLfloat fps = Frames / seconds;
+      printf("%d frames in %6.3f seconds = %6.3f FPS\n", Frames, seconds, fps);
+
+      T0 = t;
+      Frames = 0;
+   }
+}
+
+
+static void Reshape( int width, int height )
+{
+   glViewport( 0, 0, width, height );
+   glMatrixMode( GL_PROJECTION );
+   glLoadIdentity();
+/*    glFrustum( -1.0, 1.0, -1.0, 1.0, 10.0, 100.0 ); */
+   gluOrtho2D( 0, width, height, 0 );
+   glMatrixMode( GL_MODELVIEW );
+   glLoadIdentity();
+/*    glTranslatef(0.375, 0.375, 0); */
+}
+
+
+static void ModeMenu(int entry)
+{
+   switch (entry) {
+   case WHOLERECT:
+      whole_rect = !whole_rect;
+      break;
+   case RENDER:
+      do_render = !do_render;
+      break;
+   case SWAPBUF:
+      do_swapbuffers = !do_swapbuffers;
+      break;
+
+   case UPLOAD_CONVENTIONAL:
+   case UPLOAD_PBO_NAIVE:
+   case UPLOAD_PBO_PINGPONG:
+   case UPLOAD_PBO_WITH_RESET:
+   case UPLOAD_NOOP:
+      upload_style = entry;
+      break;
+
+   case QUIT:
+      exit(0);
+      break;
+   }
+
+   glutPostRedisplay();
+}
+
+
+static void Key( unsigned char key, int x, int y )
+{
+   (void) x;
+   (void) y;
+   switch (key) {
+      case 27:
+         exit(0);
+         break;
+   }
+   glutPostRedisplay();
+}
+
+
+static void SpecialKey( int key, int x, int y )
+{
+   float step = 3.0;
+   (void) x;
+   (void) y;
+
+   switch (key) {
+      case GLUT_KEY_UP:
+         Xrot += step;
+         break;
+      case GLUT_KEY_DOWN:
+         Xrot -= step;
+         break;
+      case GLUT_KEY_LEFT:
+         Yrot += step;
+         break;
+      case GLUT_KEY_RIGHT:
+         Yrot -= step;
+         break;
+   }
+   glutPostRedisplay();
+}
+
+
+static void Init( int argc, char *argv[] )
+{
+   const char *exten = (const char *) glGetString(GL_EXTENSIONS);
+   GLint size;
+
+
+   if (!strstr(exten, "GL_ARB_multitexture")) {
+      printf("Sorry, GL_ARB_multitexture not supported by this renderer.\n");
+      exit(1);
+   }
+
+   glGetIntegerv(GL_MAX_TEXTURE_SIZE, &size);
+   printf("%d x %d max texture size\n", size, size);
+
+   glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+
+   /* allocate two texture objects */
+   glGenTextures(1, &texObj);
+
+   /* setup the texture objects */
+   glActiveTextureARB(GL_TEXTURE0_ARB);
+   glBindTexture(GL_TEXTURE_RECTANGLE_ARB, texObj);
+
+   glTexParameteri(GL_TEXTURE_RECTANGLE_ARB, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+   glTexParameteri(GL_TEXTURE_RECTANGLE_ARB, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+   glGenBuffersARB(NR_PBO, DrawPBO);
+
+   {
+      int i;
+      for (i = 0; i < NR_PBO; i++) {
+	 glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, DrawPBO[i]);
+	 glBufferDataARB(GL_PIXEL_UNPACK_BUFFER_EXT, Width * Height * 4, NULL, GL_STREAM_DRAW_ARB);
+      }
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, 0);
+   }
+   
+   glTexEnvi(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
+
+   glEnable(GL_TEXTURE_RECTANGLE_ARB);
+
+   glShadeModel(GL_SMOOTH);
+   glClearColor(0.3, 0.3, 0.4, 1.0);
+
+   if (argc > 1 && strcmp(argv[1], "-info")==0) {
+      printf("GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+      printf("GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+      printf("GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+      printf("GL_EXTENSIONS = %s\n", (char *) glGetString(GL_EXTENSIONS));
+   }
+}
+
+
+int main( int argc, char *argv[] )
+{
+   GLint i;
+
+   glutInit( &argc, argv );
+
+   for (i = 1; i < argc; i++) {
+      if (strcmp(argv[i], "-w") == 0) {
+         Width = atoi(argv[i+1]);
+         if (Width <= 0) {
+            printf("Error, bad width\n");
+            exit(1);
+         }
+         i++;
+      }
+      else if (strcmp(argv[i], "-h") == 0) {
+         Height = atoi(argv[i+1]);
+         if (Height <= 0) {
+            printf("Error, bad height\n");
+            exit(1);
+         }
+         i++;
+      }
+   }
+
+   glutInitWindowSize( Width, Height );
+   glutInitWindowPosition( 0, 0 );
+   glutInitDisplayMode( GLUT_RGB | GLUT_DOUBLE );
+   glutCreateWindow(argv[0] );
+
+   Init( argc, argv );
+
+   glutReshapeFunc( Reshape );
+   glutKeyboardFunc( Key );
+   glutSpecialFunc( SpecialKey );
+   glutDisplayFunc( Display );
+   glutIdleFunc( Idle );
+
+   glutCreateMenu(ModeMenu);
+   glutAddMenuEntry("Wholerect", WHOLERECT);
+   glutAddMenuEntry("Render", RENDER);
+   glutAddMenuEntry("Swapbuf", SWAPBUF);
+
+   glutAddMenuEntry("Upload style CONVENTIONAL", UPLOAD_CONVENTIONAL);
+   glutAddMenuEntry("Upload style PBO_NAIVE", UPLOAD_PBO_NAIVE);
+   glutAddMenuEntry("Upload style PBO_PINGPONG", UPLOAD_PBO_PINGPONG);
+   glutAddMenuEntry("Upload style PBO_WITH_RESET", UPLOAD_PBO_WITH_RESET);
+   glutAddMenuEntry("Upload style NOOP", UPLOAD_NOOP);
+
+   glutAddMenuEntry("Quit", QUIT);
+   glutAttachMenu(GLUT_RIGHT_BUTTON);
+
+   glutMainLoop();
+   return 0;
+}

progs/demos/texdown.c

@@ -38,8 +38,8 @@
 #include <GL/glut.h>
 
 
-static GLsizei MaxSize = 1024;
-static GLsizei TexWidth = 256, TexHeight = 256, TexBorder = 0;
+static GLsizei MaxSize = 2048;
+static GLsizei TexWidth = 1024, TexHeight = 1024, TexBorder = 0;
 static GLboolean ScaleAndBias = GL_FALSE;
 static GLboolean SubImage = GL_FALSE;
 static GLdouble DownloadRate = 0.0;  /* texels/sec */
@@ -47,6 +47,32 @@ static GLdouble DownloadRate = 0.0;  /* texels/sec */
 static GLuint Mode = 0;
 
 
+/* Try and avoid L2 cache effects by cycling through a small number of
+ * textures.
+ * 
+ * At the initial size of 1024x1024x4 == 4mbyte, say 8 textures will
+ * keep us out of most caches at 32mb total.
+ *
+ * This turns into a fairly interesting question of what exactly you
+ * expect to be in cache in normal usage, and what you think should be
+ * outside.  There's no rules for this, no reason to favour one usage
+ * over another except what the application you care about happens to
+ * resemble most closely.
+ *
+ * - Should the client texture image be in L2 cache?  Has it just been
+ *   generated or read from disk?
+ * - Does the application really use >1 texture, or is it constantly 
+ *   updating one image in-place?
+ *
+ * Different answers will favour different texture upload mechanisms.
+ * To upload an image that is purely outside of cache, a DMA-based
+ * upload will probably win, whereas for small, in-cache textures,
+ * copying looks good.
+ */
+#define NR_TEXOBJ 4
+static GLuint TexObj[NR_TEXOBJ];
+
+
 struct FormatRec {
    GLenum Format;
    GLenum Type;
@@ -116,25 +142,57 @@ TypeStr(GLenum type)
    }
 }
 
+/* On x86, there is a performance cliff for memcpy to texture memory
+ * for sources below 64 byte alignment.  We do our best with this in
+ * the driver, but it is better if the images are correctly aligned to
+ * start with:
+ */
+#define ALIGN (1<<12)
+
+static unsigned align(unsigned value, unsigned a)
+{
+   return (value + a - 1) & ~(a-1);
+}
+
+static int MIN2(int a, int b)
+{
+   return a < b ? a : b;
+}
 
 static void
 MeasureDownloadRate(void)
 {
    const int w = TexWidth + 2 * TexBorder;
    const int h = TexHeight + 2 * TexBorder;
-   const int bytes = w * h * BytesPerTexel(Format);
+   const int image_bytes = align(w * h * BytesPerTexel(Format), ALIGN);
+   const int bytes = image_bytes * NR_TEXOBJ;
+   GLubyte *orig_texImage, *orig_getImage;
    GLubyte *texImage, *getImage;
    GLdouble t0, t1, time;
    int count;
    int i;
+   int offset = 0;
+   GLdouble total = 0;		/* ints will tend to overflow */
 
-   texImage = (GLubyte *) malloc(bytes);
-   getImage = (GLubyte *) malloc(bytes);
-   if (!texImage || !getImage) {
+   printf("allocating %d bytes for %d %dx%d images\n",
+	  bytes, NR_TEXOBJ, w, h);
+
+   orig_texImage = (GLubyte *) malloc(bytes + ALIGN);
+   orig_getImage = (GLubyte *) malloc(image_bytes + ALIGN);
+   if (!orig_texImage || !orig_getImage) {
       DownloadRate = 0.0;
       return;
    }
 
+   printf("alloc %p %p\n", orig_texImage, orig_getImage);
+
+   texImage = (GLubyte *)align((unsigned)orig_texImage, ALIGN);
+   getImage = (GLubyte *)align((unsigned)orig_getImage, ALIGN);   
+
+   for (i = 1; !(((unsigned)texImage) & i); i<<=1)
+      ;
+   printf("texture image alignment: %d bytes (%p)\n", i, texImage);
+      
    for (i = 0; i < bytes; i++) {
       texImage[i] = i & 0xff;
    }
@@ -166,16 +224,50 @@ MeasureDownloadRate(void)
    count = 0;
    t0 = glutGet(GLUT_ELAPSED_TIME) * 0.001;
    do {
+      int img = count%NR_TEXOBJ;
+      GLubyte *img_ptr = texImage + img * image_bytes;
+
+      glBindTexture(GL_TEXTURE_2D, TexObj[img]);
+
       if (SubImage && count > 0) {
-         glTexSubImage2D(GL_TEXTURE_2D, 0, -TexBorder, -TexBorder, w, h,
+	 /* Only update a portion of the image each iteration.  This
+	  * is presumably why you'd want to use texsubimage, otherwise
+	  * you may as well just call teximage again.
+	  *
+	  * A bigger question is whether to use a pointer that moves
+	  * with each call, ie does the incoming data come from L2
+	  * cache under normal circumstances, or is it pulled from
+	  * uncached memory?  
+	  * 
+	  * There's a good argument to say L2 cache, ie you'd expect
+	  * the data to have been recently generated.  It's possible
+	  * that it could have come from a file read, which may or may
+	  * not have gone through the cpu.
+	  */
+         glTexSubImage2D(GL_TEXTURE_2D, 0, 
+			 -TexBorder, 
+			 -TexBorder + offset * h/8, 
+			 w, 
+			 h/8,
                          FormatTable[Format].Format,
-                         FormatTable[Format].Type, texImage);
+                         FormatTable[Format].Type, 
+#if 1
+			 texImage /* likely in L2$ */
+#else
+			 img_ptr + offset * bytes/8 /* unlikely in L2$ */
+#endif
+	    );
+	 offset += 1;
+	 offset %= 8;
+	 total += w * h / 8;
       }
       else {
          glTexImage2D(GL_TEXTURE_2D, 0,
                       FormatTable[Format].IntFormat, w, h, TexBorder,
                       FormatTable[Format].Format,
-                      FormatTable[Format].Type, texImage);
+                      FormatTable[Format].Type, 
+		      img_ptr);
+	 total += w*h;
       }
 
       /* draw a tiny polygon to force texture into texram */
@@ -192,25 +284,12 @@ MeasureDownloadRate(void)
 
    glDisable(GL_TEXTURE_2D);
 
-   printf("w*h=%d  count=%d  time=%f\n", w*h, count, time);
-   DownloadRate = w * h * count / time;
+   printf("total texels=%f  time=%f\n", total, time);
+   DownloadRate = total / time;
 
-#if 0
-   if (!ScaleAndBias) {
-      /* verify texture readback */
-      glGetTexImage(GL_TEXTURE_2D, 0,
-                    FormatTable[Format].Format,
-                    FormatTable[Format].Type, getImage);
-      for (i = 0; i < w * h; i++) {
-         if (texImage[i] != getImage[i]) {
-            printf("[%d] %d != %d\n", i, texImage[i], getImage[i]);
-         }
-      }
-   }
-#endif
 
-   free(texImage);
-   free(getImage);
+   free(orig_texImage); 
+   free(orig_getImage); 
 
    {
       GLint err = glGetError();

progs/tests/arbnpot-mipmap.c

@@ -0,0 +1,184 @@
+
+/* Copyright (c) Mark J. Kilgard, 1994. */
+
+/*
+ * (c) Copyright 1993, Silicon Graphics, Inc.
+ * ALL RIGHTS RESERVED
+ * Permission to use, copy, modify, and distribute this software for
+ * any purpose and without fee is hereby granted, provided that the above
+ * copyright notice appear in all copies and that both the copyright notice
+ * and this permission notice appear in supporting documentation, and that
+ * the name of Silicon Graphics, Inc. not be used in advertising
+ * or publicity pertaining to distribution of the software without specific,
+ * written prior permission.
+ *
+ * THE MATERIAL EMBODIED ON THIS SOFTWARE IS PROVIDED TO YOU "AS-IS"
+ * AND WITHOUT WARRANTY OF ANY KIND, EXPRESS, IMPLIED OR OTHERWISE,
+ * INCLUDING WITHOUT LIMITATION, ANY WARRANTY OF MERCHANTABILITY OR
+ * FITNESS FOR A PARTICULAR PURPOSE.  IN NO EVENT SHALL SILICON
+ * GRAPHICS, INC.  BE LIABLE TO YOU OR ANYONE ELSE FOR ANY DIRECT,
+ * SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY
+ * KIND, OR ANY DAMAGES WHATSOEVER, INCLUDING WITHOUT LIMITATION,
+ * LOSS OF PROFIT, LOSS OF USE, SAVINGS OR REVENUE, OR THE CLAIMS OF
+ * THIRD PARTIES, WHETHER OR NOT SILICON GRAPHICS, INC.  HAS BEEN
+ * ADVISED OF THE POSSIBILITY OF SUCH LOSS, HOWEVER CAUSED AND ON
+ * ANY THEORY OF LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE
+ * POSSESSION, USE OR PERFORMANCE OF THIS SOFTWARE.
+ *
+ * US Government Users Restricted Rights
+ * Use, duplication, or disclosure by the Government is subject to
+ * restrictions set forth in FAR 52.227.19(c)(2) or subparagraph
+ * (c)(1)(ii) of the Rights in Technical Data and Computer Software
+ * clause at DFARS 252.227-7013 and/or in similar or successor
+ * clauses in the FAR or the DOD or NASA FAR Supplement.
+ * Unpublished-- rights reserved under the copyright laws of the
+ * United States.  Contractor/manufacturer is Silicon Graphics,
+ * Inc., 2011 N.  Shoreline Blvd., Mountain View, CA 94039-7311.
+ *
+ * OpenGL(TM) is a trademark of Silicon Graphics, Inc.
+ */
+/*  mipmap.c
+ *  This program demonstrates using mipmaps for texture maps.
+ *  To overtly show the effect of mipmaps, each mipmap reduction
+ *  level has a solidly colored, contrasting texture image.
+ *  Thus, the quadrilateral which is drawn is drawn with several
+ *  different colors.
+ */
+#include <stdlib.h>
+#include <stdio.h>
+#include <GL/glut.h>
+
+GLubyte mipmapImage32[40][46][3];
+GLubyte mipmapImage16[20][23][3];
+GLubyte mipmapImage8[10][11][3];
+GLubyte mipmapImage4[5][5][3];
+GLubyte mipmapImage2[2][2][3];
+GLubyte mipmapImage1[1][1][3];
+
+static void makeImages(void)
+{
+    int i, j;
+
+    for (i = 0; i < 40; i++) {
+	for (j = 0; j < 46; j++) {
+	    mipmapImage32[i][j][0] = 255;
+	    mipmapImage32[i][j][1] = 255;
+	    mipmapImage32[i][j][2] = 0;
+	}
+    }
+    for (i = 0; i < 20; i++) {
+	for (j = 0; j < 23; j++) {
+	    mipmapImage16[i][j][0] = 255;
+	    mipmapImage16[i][j][1] = 0;
+	    mipmapImage16[i][j][2] = 255;
+	}
+    }
+    for (i = 0; i < 10; i++) {
+	for (j = 0; j < 11; j++) {
+	    mipmapImage8[i][j][0] = 255;
+	    mipmapImage8[i][j][1] = 0;
+	    mipmapImage8[i][j][2] = 0;
+	}
+    }
+    for (i = 0; i < 5; i++) {
+	for (j = 0; j < 5; j++) {
+	    mipmapImage4[i][j][0] = 0;
+	    mipmapImage4[i][j][1] = 255;
+	    mipmapImage4[i][j][2] = 0;
+	}
+    }
+    for (i = 0; i < 2; i++) {
+	for (j = 0; j < 2; j++) {
+	    mipmapImage2[i][j][0] = 0;
+	    mipmapImage2[i][j][1] = 0;
+	    mipmapImage2[i][j][2] = 255;
+	}
+    }
+    mipmapImage1[0][0][0] = 255;
+    mipmapImage1[0][0][1] = 255;
+    mipmapImage1[0][0][2] = 255;
+}
+
+static void myinit(void)
+{
+    if (!glutExtensionSupported("GL_ARB_texture_non_power_of_two")) {
+       printf("Sorry, this program requires GL_ARB_texture_non_power_of_two\n");
+       exit(1);
+    }
+
+    glEnable(GL_DEPTH_TEST);
+    glDepthFunc(GL_LESS);
+    glShadeModel(GL_FLAT);
+
+    glTranslatef(0.0, 0.0, -3.6);
+    makeImages();
+    glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+    glTexImage2D(GL_TEXTURE_2D, 0, 3, 40, 46, 0,
+		 GL_RGB, GL_UNSIGNED_BYTE, &mipmapImage32[0][0][0]);
+    glTexImage2D(GL_TEXTURE_2D, 1, 3, 20, 23, 0,
+		 GL_RGB, GL_UNSIGNED_BYTE, &mipmapImage16[0][0][0]);
+    glTexImage2D(GL_TEXTURE_2D, 2, 3, 10, 11, 0,
+		 GL_RGB, GL_UNSIGNED_BYTE, &mipmapImage8[0][0][0]);
+    glTexImage2D(GL_TEXTURE_2D, 3, 3, 5, 5, 0,
+		 GL_RGB, GL_UNSIGNED_BYTE, &mipmapImage4[0][0][0]);
+    glTexImage2D(GL_TEXTURE_2D, 4, 3, 2, 2, 0,
+		 GL_RGB, GL_UNSIGNED_BYTE, &mipmapImage2[0][0][0]);
+    glTexImage2D(GL_TEXTURE_2D, 5, 3, 1, 1, 0,
+		 GL_RGB, GL_UNSIGNED_BYTE, &mipmapImage1[0][0][0]);
+    glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_S, GL_REPEAT);
+    glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_T, GL_REPEAT);
+    glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+    glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER,
+	GL_NEAREST_MIPMAP_NEAREST);
+    glTexEnvf(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_DECAL);
+    glEnable(GL_TEXTURE_2D);
+}
+
+static void display(void)
+{
+    glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
+    glBegin(GL_QUADS);
+    glTexCoord2f(0.0, 0.0); glVertex3f(-2.0, -1.0, 0.0);
+    glTexCoord2f(0.0, 8.0); glVertex3f(-2.0, 1.0, 0.0);
+    glTexCoord2f(8.0, 8.0); glVertex3f(2000.0, 1.0, -6000.0);
+    glTexCoord2f(8.0, 0.0); glVertex3f(2000.0, -1.0, -6000.0);
+    glEnd();
+    glFlush();
+}
+
+static void myReshape(int w, int h)
+{
+    glViewport(0, 0, w, h);
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    gluPerspective(60.0, 1.0*(GLfloat)w/(GLfloat)h, 1.0, 30000.0);
+    glMatrixMode(GL_MODELVIEW);
+    glLoadIdentity();
+}
+
+static void
+key(unsigned char k, int x, int y)
+{
+  switch (k) {
+  case 27:  /* Escape */
+    exit(0);
+    break;
+  default:
+    return;
+  }
+  glutPostRedisplay();
+}
+
+int main(int argc, char** argv)
+{
+    glutInit(&argc, argv);
+    glutInitDisplayMode (GLUT_SINGLE | GLUT_RGB | GLUT_DEPTH);
+    glutInitWindowSize (500, 500);
+    glutCreateWindow (argv[0]);
+    myinit();
+    glutReshapeFunc (myReshape);
+    glutDisplayFunc(display);
+    glutKeyboardFunc(key);
+    glutMainLoop();
+    return 0;             /* ANSI C requires main to return int. */
+}

progs/trivial/Makefile

@@ -38,7 +38,9 @@ SOURCES = \
 	quad-offset-unfilled.c \
 	quad-unfilled.c \
 	quad-tex-2d.c \
+	quad-tex-pbo.c \
 	quad-tex-3d.c \
+	quad-tex-dep.c \
 	quad.c \
 	quads.c \
 	quadstrip.c \

progs/trivial/quad-tex-dep.c

@@ -0,0 +1,245 @@
+
+#include <assert.h>
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#define GL_GLEXT_PROTOTYPES
+#include <GL/glut.h>
+
+
+
+GLenum doubleBuffer;
+
+/* Demonstrate an equivalent to paletted texturing using fragment
+ * programs and dependent texturing.  This requires at least one
+ * texture unit to be reserved for the palette lookup.
+ */
+
+static void Init( void )
+{
+   static const char *modulate2D =
+      "!!ARBfp1.0\n"
+      "TEMP index; \n"
+      "TEX index, fragment.texcoord[0], texture[0], 2D; \n"
+      "TEX result.color, index, texture[1], 1D; \n"
+      "END"
+      ;
+   GLuint modulateProg;
+   GLuint Texture;
+
+   if (!glutExtensionSupported("GL_ARB_fragment_program")) {
+      printf("Error: GL_ARB_fragment_program not supported!\n");
+      exit(1);
+   }
+   printf("GL_RENDERER = %s\n", (char *) glGetString(GL_RENDERER));
+
+   /* Setup the fragment program */
+   glGenProgramsARB(1, &modulateProg);
+   glBindProgramARB(GL_FRAGMENT_PROGRAM_ARB, modulateProg);
+   glProgramStringARB(GL_FRAGMENT_PROGRAM_ARB, GL_PROGRAM_FORMAT_ASCII_ARB,
+		      strlen(modulate2D), (const GLubyte *)modulate2D);
+
+   printf("glGetError = 0x%x\n", (int) glGetError());
+   printf("glError(GL_PROGRAM_ERROR_STRING_ARB) = %s\n",
+          (char *) glGetString(GL_PROGRAM_ERROR_STRING_ARB));
+   assert(glIsProgramARB(modulateProg));
+
+   glEnable(GL_FRAGMENT_PROGRAM_ARB);
+
+
+   {
+#define HEIGHT 8
+#define WIDTH 32
+
+#define B 0
+#define G 1
+#define R 2
+#define A 3
+
+     static char texture[HEIGHT * WIDTH + 1] =
+	"                                "
+	"    MMM    EEEE   SSS    AAA    "
+	"   M M M  E      S   S  A   A   "
+	"   M M M  EEEE    SS    A   A   "
+	"   M M M  E         SS  AAAAA   "
+	"   M   M  E      S   S  A   A   "
+	"   M   M   EEEE   SSS   A   A   ";
+      
+      GLubyte table[256][4];
+
+      
+      
+      /* load the color table for each texel-index */
+      memset(table, 0xff, 256*4);
+      table[' '][B] = 255;
+      table[' '][G] = 255;
+      table[' '][R] = 255;
+      table[' '][A] = 64;
+      table['M'][B] = 0;
+      table['M'][G] = 0;
+      table['M'][R] = 255;
+      table['M'][A] = 255;
+      table['E'][B] = 0;
+      table['E'][G] = 255;
+      table['E'][R] = 0;
+      table['E'][A] = 255;
+      table['S'][B] = 255;
+      table['S'][G] = 0;
+      table['S'][R] = 0;
+      table['S'][A] = 255;
+      table['A'][B] = 0;
+      table['A'][G] = 255;
+      table['A'][R] = 255;
+      table['A'][A] = 255;
+
+      glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+
+
+      /* Load indexed texture as a 2D I8 texture */
+      glActiveTextureARB(GL_TEXTURE0_ARB);
+      glGenTextures(1, &Texture);
+      glBindTexture(GL_TEXTURE_2D, Texture);
+
+      glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+      glTexParameteri(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+      glTexImage2D(GL_TEXTURE_2D,       /* target */
+		   0,                   /* level */
+		   GL_INTENSITY, /* internal format */
+		   WIDTH, HEIGHT,       /* width, height */
+		   0,                   /* border */
+		   GL_LUMINANCE,      /* texture format */
+		   GL_UNSIGNED_BYTE,    /* texture type */
+		   texture);            /* the texture */
+
+
+   printf("glGetError = 0x%x\n", (int) glGetError());
+   printf("glError(GL_PROGRAM_ERROR_STRING_ARB) = %s\n",
+          (char *) glGetString(GL_PROGRAM_ERROR_STRING_ARB));
+
+      glEnable(GL_TEXTURE_2D);
+
+      /* Load "pallete" as a 1D BGRA8888 texture */
+      glActiveTextureARB(GL_TEXTURE1_ARB);
+      glGenTextures(1, &Texture);
+      glBindTexture(GL_TEXTURE_1D, Texture);
+
+      glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+      glTexParameteri(GL_TEXTURE_1D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+      glTexImage1D(GL_TEXTURE_1D,       /* target */
+		   0,                   /* level */
+		   GL_RGBA,	/* internal format */
+		   256,		/* width */
+		   0,                   /* border */
+		   GL_BGRA,      /* texture format */
+		   GL_UNSIGNED_BYTE,    /* texture type */
+		   table);            /* the texture */
+
+      glEnable(GL_TEXTURE_1D);
+
+/*       glEnable(GL_BLEND); */
+/*       glBlendFunc(GL_SRC_ALPHA, GL_ONE_MINUS_SRC_ALPHA); */
+#undef HEIGHT
+#undef WIDTH
+   }
+
+
+   glClearColor(.3, .3, .3, 0);
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0);
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+
+    switch (key) {
+      case 27:
+	exit(1);
+      default:
+	return;
+    }
+
+    glutPostRedisplay();
+}
+
+static void Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glBegin(GL_QUADS);
+   glTexCoord2f(1,1); 
+   glVertex3f( 0.9, -0.9, -30.0);
+   glTexCoord2f(1,0); 
+   glVertex3f( 0.9,  0.9, -30.0);
+   glTexCoord2f(0,0); 
+   glVertex3f(-0.9,  0.9, -30.0);
+   glTexCoord2f(0,1); 
+   glVertex3f(-0.9,  -0.9, -30.0);
+   glEnd();
+
+   glFlush();
+
+   if (doubleBuffer) {
+      glutSwapBuffers();
+   }
+
+
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    GLint i;
+
+    doubleBuffer = GL_FALSE;
+
+    for (i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "-sb") == 0) {
+	    doubleBuffer = GL_FALSE;
+	} else if (strcmp(argv[i], "-db") == 0) {
+	    doubleBuffer = GL_TRUE;
+	} else {
+	    fprintf(stderr, "%s (Bad option).\n", argv[i]);
+	    return GL_FALSE;
+	}
+    }
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB;
+    type |= (doubleBuffer) ? GLUT_DOUBLE : GLUT_SINGLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+	return 0;
+}

progs/trivial/quad-tex-pbo.c

@@ -0,0 +1,181 @@
+/*
+ * Copyright (c) 1991, 1992, 1993 Silicon Graphics, Inc.
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and
+ * its documentation for any purpose is hereby granted without fee, provided
+ * that (i) the above copyright notices and this permission notice appear in
+ * all copies of the software and related documentation, and (ii) the name of
+ * Silicon Graphics may not be used in any advertising or
+ * publicity relating to the software without the specific, prior written
+ * permission of Silicon Graphics.
+ *
+ * THE SOFTWARE IS PROVIDED "AS-IS" AND WITHOUT WARRANTY OF
+ * ANY KIND,
+ * EXPRESS, IMPLIED OR OTHERWISE, INCLUDING WITHOUT LIMITATION, ANY
+ * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
+ *
+ * IN NO EVENT SHALL SILICON GRAPHICS BE LIABLE FOR
+ * ANY SPECIAL, INCIDENTAL, INDIRECT OR CONSEQUENTIAL DAMAGES OF ANY KIND,
+ * OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS,
+ * WHETHER OR NOT ADVISED OF THE POSSIBILITY OF DAMAGE, AND ON ANY THEORY OF
+ * LIABILITY, ARISING OUT OF OR IN CONNECTION WITH THE USE OR PERFORMANCE
+ * OF THIS SOFTWARE.
+ */
+
+#define GL_GLEXT_PROTOTYPES
+
+#include <stdio.h>
+#include <string.h>
+#include <stdlib.h>
+#include <GL/glut.h>
+
+
+#define CI_OFFSET_1 16
+#define CI_OFFSET_2 32
+
+GLenum doubleBuffer;
+
+static GLuint DrawPBO;
+
+static void Init(void)
+{
+   fprintf(stderr, "GL_RENDERER   = %s\n", (char *) glGetString(GL_RENDERER));
+   fprintf(stderr, "GL_VERSION    = %s\n", (char *) glGetString(GL_VERSION));
+   fprintf(stderr, "GL_VENDOR     = %s\n", (char *) glGetString(GL_VENDOR));
+
+   glClearColor(0.0, 0.0, 1.0, 0.0);
+
+#define SIZE 16
+   {
+      GLubyte tex2d[SIZE][SIZE][4];
+      GLint s, t;
+
+      for (s = 0; s < SIZE; s++) {
+	 for (t = 0; t < SIZE; t++) {
+	    /* bgra:
+	     */
+	    tex2d[t][s][0] = 0x30;
+	    tex2d[t][s][1] = t*255/(SIZE-1);
+	    tex2d[t][s][2] = s*255/(SIZE-1);
+	    tex2d[t][s][3] = 0xff;
+	 }
+      }
+
+
+      /* put image into DrawPBO */
+      glGenBuffersARB(1, &DrawPBO);
+      glBindBufferARB(GL_PIXEL_PACK_BUFFER_EXT, DrawPBO);
+      glBufferDataARB(GL_PIXEL_PACK_BUFFER_EXT,
+		      SIZE * SIZE * 4, tex2d, GL_STATIC_DRAW);
+      glBindBufferARB(GL_PIXEL_PACK_BUFFER_EXT, 0);
+
+
+      glTexEnvf(GL_TEXTURE_ENV, GL_TEXTURE_ENV_MODE, GL_REPLACE);
+      glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_WRAP_R, GL_REPEAT);
+      glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MIN_FILTER, GL_NEAREST);
+      glTexParameterf(GL_TEXTURE_2D, GL_TEXTURE_MAG_FILTER, GL_NEAREST);
+
+      glPixelStorei(GL_UNPACK_ALIGNMENT, 1);
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, DrawPBO);
+      glTexImage2D(GL_TEXTURE_2D, 0, GL_RGBA, SIZE, SIZE, 0,
+		   GL_BGRA, GL_UNSIGNED_BYTE, NULL);
+      glBindBufferARB(GL_PIXEL_UNPACK_BUFFER_EXT, 0);
+      glPixelStorei(GL_UNPACK_ALIGNMENT, 4);
+      glEnable(GL_TEXTURE_2D);
+   }
+
+}
+
+static void Reshape(int width, int height)
+{
+
+    glViewport(0, 0, (GLint)width, (GLint)height);
+
+    glMatrixMode(GL_PROJECTION);
+    glLoadIdentity();
+    glOrtho(-1.0, 1.0, -1.0, 1.0, -0.5, 1000.0);
+    glMatrixMode(GL_MODELVIEW);
+}
+
+static void Key(unsigned char key, int x, int y)
+{
+
+    switch (key) {
+      case 27:
+	exit(1);
+      default:
+	return;
+    }
+
+    glutPostRedisplay();
+}
+
+static void Draw(void)
+{
+   glClear(GL_COLOR_BUFFER_BIT); 
+
+   glBegin(GL_QUADS);
+   glTexCoord2f(1,0); 
+   glVertex3f( 0.9, -0.9, -30.0);
+   glTexCoord2f(1,1); 
+   glVertex3f( 0.9,  0.9, -30.0);
+   glTexCoord2f(0,1); 
+   glVertex3f(-0.9,  0.9, -30.0);
+   glTexCoord2f(0,0); 
+   glVertex3f(-0.9,  -0.9, -30.0);
+   glEnd();
+
+   glFlush();
+
+   if (doubleBuffer) {
+      glutSwapBuffers();
+   }
+}
+
+static GLenum Args(int argc, char **argv)
+{
+    GLint i;
+
+    doubleBuffer = GL_FALSE;
+
+    for (i = 1; i < argc; i++) {
+        if (strcmp(argv[i], "-sb") == 0) {
+	    doubleBuffer = GL_FALSE;
+	} else if (strcmp(argv[i], "-db") == 0) {
+	    doubleBuffer = GL_TRUE;
+	} else {
+	    fprintf(stderr, "%s (Bad option).\n", argv[i]);
+	    return GL_FALSE;
+	}
+    }
+    return GL_TRUE;
+}
+
+int main(int argc, char **argv)
+{
+    GLenum type;
+
+    glutInit(&argc, argv);
+
+    if (Args(argc, argv) == GL_FALSE) {
+	exit(1);
+    }
+
+    glutInitWindowPosition(0, 0); glutInitWindowSize( 250, 250);
+
+    type = GLUT_RGB;
+    type |= (doubleBuffer) ? GLUT_DOUBLE : GLUT_SINGLE;
+    glutInitDisplayMode(type);
+
+    if (glutCreateWindow("First Tri") == GL_FALSE) {
+	exit(1);
+    }
+
+    Init();
+
+    glutReshapeFunc(Reshape);
+    glutKeyboardFunc(Key);
+    glutDisplayFunc(Draw);
+    glutMainLoop();
+	return 0;
+}

src/glx/x11/glxext.c

@@ -722,6 +722,68 @@ static const __DRIinterfaceMethods interface_methods = {
     glXGetMscRateOML,
 };
 
+#define DRM_MAX_FDS 16
+static struct {
+   char *BusID;
+   int fd;
+   int refcount;
+} connection[DRM_MAX_FDS];
+
+static int nr_fds = 0;
+
+int drmOpenOnce(void *unused, 
+		const char *BusID,
+		int *newlyopened)
+{
+   int i;
+   int fd;
+   
+   for (i = 0; i < nr_fds; i++)
+      if (strcmp(BusID, connection[i].BusID) == 0) {
+	 connection[i].refcount++;
+	 *newlyopened = 0;
+	 return connection[i].fd;
+      }
+
+   fd = drmOpen(unused, BusID);
+   if (fd <= 0 || nr_fds == DRM_MAX_FDS)
+      return fd;
+   
+   connection[nr_fds].BusID = strdup(BusID);
+   connection[nr_fds].fd = fd;
+   connection[nr_fds].refcount = 1;
+   *newlyopened = 1;
+
+   fprintf(stderr, "saved connection %d for %s %d\n", 
+	  nr_fds, connection[nr_fds].BusID, 
+	  strcmp(BusID, connection[nr_fds].BusID));
+
+   nr_fds++;
+
+   return fd;   
+}
+
+void drmCloseOnce(int fd)
+{
+   int i;
+
+   
+
+   for (i = 0; i < nr_fds; i++) {
+      if (fd == connection[i].fd) {
+	 if (--connection[i].refcount == 0) {
+	    drmClose(connection[i].fd);
+	    free(connection[i].BusID);
+	    
+	    if (i < --nr_fds) 
+	       connection[i] = connection[nr_fds];
+
+	    return;
+	 }
+      }
+   }
+}
+
 
 /**
  * Perform the required libGL-side initialization and call the client-side
@@ -773,7 +835,8 @@ CallCreateNewScreen(Display *dpy, int scrn, __DRIscreen *psc,
     framebuffer.dev_priv = NULL;
 
     if (XF86DRIOpenConnection(dpy, scrn, &hSAREA, &BusID)) {
-	fd = drmOpen(NULL,BusID);
+        int newlyopened;
+	fd = drmOpenOnce(NULL,BusID, &newlyopened);
 	Xfree(BusID); /* No longer needed */
 
 	err_msg = "open DRM";
@@ -800,7 +863,7 @@ CallCreateNewScreen(Display *dpy, int scrn, __DRIscreen *psc,
 		}
 
 		err_msg = "XF86DRIAuthConnection";
-		if (XF86DRIAuthConnection(dpy, scrn, magic)) {
+		if (!newlyopened || XF86DRIAuthConnection(dpy, scrn, magic)) {
 		    char *driverName;
 
 		    /*
@@ -904,7 +967,7 @@ CallCreateNewScreen(Display *dpy, int scrn, __DRIscreen *psc,
 	}
 
 	if ( fd >= 0 ) {
-	    (void)drmClose(fd);
+	    (void)drmCloseOnce(fd);
 	}
 
 	(void)XF86DRICloseConnection(dpy, scrn);

src/mesa/drivers/dri/Makefile.template

@@ -9,7 +9,10 @@ COMMON_SOURCES = \
         ../common/vblank.c \
         ../common/dri_util.c \
         ../common/xmlconfig.c \
-        ../common/drirenderbuffer.c
+        ../common/drirenderbuffer.c \
+	../common/dri_bufmgr.c \
+	../common/dri_drmpool.c
+
 
 ifeq ($(WINDOW_SYSTEM),dri)
 WINOBJ=

src/mesa/drivers/dri/common/dri_bufmgr.c

@@ -0,0 +1,493 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellstr<74>m <thomas-at-tungstengraphics-dot-com>
+ *          Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ */
+
+#include <xf86drm.h>
+#include <stdlib.h>
+#include "glthread.h"
+#include "errno.h"
+#include "dri_bufmgr.h"
+#include "string.h"
+#include "imports.h"
+#include "dri_bufpool.h"
+
+_glthread_DECLARE_STATIC_MUTEX(bmMutex);
+
+/*
+ * TODO: Introduce fence pools in the same way as 
+ * buffer object pools.
+ */
+
+
+
+typedef struct _DriFenceObject
+{
+   int fd;
+   _glthread_Mutex mutex;
+   int refCount;
+   const char *name;
+   drmFence fence;
+} DriFenceObject;
+
+typedef struct _DriBufferObject
+{
+   DriBufferPool *pool;
+   _glthread_Mutex mutex;
+   int refCount;
+   const char *name;
+   unsigned flags;
+   unsigned hint;
+   unsigned alignment;
+   void *private;
+} DriBufferObject;
+
+
+void
+bmError(int val, const char *file, const char *function, int line)
+{
+   _mesa_printf("Fatal video memory manager error \"%s\".\n"
+                "Check kernel logs or set the LIBGL_DEBUG\n"
+                "environment variable to \"verbose\" for more info.\n"
+                "Detected in file %s, line %d, function %s.\n",
+                strerror(-val), file, line, function);
+#ifndef NDEBUG
+   abort();
+#else
+   abort();
+#endif
+}
+
+DriFenceObject *
+driFenceBuffers(int fd, char *name, unsigned flags)
+{
+   DriFenceObject *fence = (DriFenceObject *) malloc(sizeof(*fence));
+   int ret;
+
+   if (!fence)
+      BM_CKFATAL(-EINVAL);
+
+   _glthread_LOCK_MUTEX(bmMutex);
+   fence->refCount = 1;
+   fence->name = name;
+   fence->fd = fd;
+   _glthread_INIT_MUTEX(fence->mutex);
+   ret = drmFenceBuffers(fd, flags, &fence->fence);
+   _glthread_UNLOCK_MUTEX(bmMutex);
+   if (ret) {
+      free(fence);
+      BM_CKFATAL(ret);
+   }
+   return fence;
+}
+
+
+unsigned 
+driFenceType(DriFenceObject * fence)
+{
+    unsigned ret;
+
+    _glthread_LOCK_MUTEX(bmMutex);
+    ret = fence->fence.flags;
+    _glthread_UNLOCK_MUTEX(bmMutex);
+    
+    return ret;
+}
+
+
+DriFenceObject *
+driFenceReference(DriFenceObject * fence)
+{
+   _glthread_LOCK_MUTEX(bmMutex);
+   ++fence->refCount;
+   _glthread_UNLOCK_MUTEX(bmMutex);
+   return fence;
+}
+
+void
+driFenceUnReference(DriFenceObject * fence)
+{
+   if (!fence)
+      return;
+
+   _glthread_LOCK_MUTEX(bmMutex);
+   if (--fence->refCount == 0) {
+      drmFenceDestroy(fence->fd, &fence->fence);
+      free(fence);
+   }
+   _glthread_UNLOCK_MUTEX(bmMutex);
+}
+
+void
+driFenceFinish(DriFenceObject * fence, unsigned type, int lazy)
+{
+   int ret;
+   unsigned flags = (lazy) ? DRM_FENCE_FLAG_WAIT_LAZY : 0;
+
+   _glthread_LOCK_MUTEX(fence->mutex);
+   ret = drmFenceWait(fence->fd, flags, &fence->fence, type);
+   _glthread_UNLOCK_MUTEX(fence->mutex);
+   BM_CKFATAL(ret);
+}
+
+int
+driFenceSignaled(DriFenceObject * fence, unsigned type)
+{
+   int signaled;
+   int ret;
+
+   if (fence == NULL)
+      return GL_TRUE;
+
+   _glthread_LOCK_MUTEX(fence->mutex);
+   ret = drmFenceSignaled(fence->fd, &fence->fence, type, &signaled);
+   _glthread_UNLOCK_MUTEX(fence->mutex);
+   BM_CKFATAL(ret);
+   return signaled;
+}
+
+
+extern drmBO *
+driBOKernel(struct _DriBufferObject *buf)
+{
+   drmBO *ret;
+
+   assert(buf->private != NULL);
+   ret = buf->pool->kernel(buf->pool, buf->private);
+   if (!ret)
+      BM_CKFATAL(-EINVAL);
+
+   return ret;
+}
+
+void
+driBOWaitIdle(struct _DriBufferObject *buf, int lazy)
+{
+   assert(buf->private != NULL);
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   BM_CKFATAL(buf->pool->waitIdle(buf->pool, buf->private, lazy));
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+}
+
+void *
+driBOMap(struct _DriBufferObject *buf, unsigned flags, unsigned hint)
+{
+   void *virtual;
+
+   assert(buf->private != NULL);
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   BM_CKFATAL(buf->pool->map(buf->pool, buf->private, flags, hint, &virtual));
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+   return virtual;
+}
+
+void
+driBOUnmap(struct _DriBufferObject *buf)
+{
+   assert(buf->private != NULL);
+
+   buf->pool->unmap(buf->pool, buf->private);
+}
+
+unsigned long
+driBOOffset(struct _DriBufferObject *buf)
+{
+   unsigned long ret;
+
+   assert(buf->private != NULL);
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   ret = buf->pool->offset(buf->pool, buf->private);
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+   return ret;
+}
+
+unsigned
+driBOFlags(struct _DriBufferObject *buf)
+{
+   unsigned ret;
+
+   assert(buf->private != NULL);
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   ret = buf->pool->flags(buf->pool, buf->private);
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+   return ret;
+}
+
+struct _DriBufferObject *
+driBOReference(struct _DriBufferObject *buf)
+{
+   _glthread_LOCK_MUTEX(bmMutex);
+   if (++buf->refCount == 1) {
+      BM_CKFATAL(-EINVAL);
+   }
+   _glthread_UNLOCK_MUTEX(bmMutex);
+   return buf;
+}
+
+void
+driBOUnReference(struct _DriBufferObject *buf)
+{
+   int tmp;
+
+   if (!buf)
+      return;
+
+   _glthread_LOCK_MUTEX(bmMutex);
+   tmp = --buf->refCount;
+   _glthread_UNLOCK_MUTEX(bmMutex);
+   if (!tmp) {
+      buf->pool->destroy(buf->pool, buf->private);
+      free(buf);
+   }
+}
+
+void
+driBOData(struct _DriBufferObject *buf,
+          unsigned size, const void *data, unsigned flags)
+{
+   void *virtual;
+   int newBuffer;
+   struct _DriBufferPool *pool;
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   pool = buf->pool;
+   if (!pool->create) {
+      _mesa_error(NULL, GL_INVALID_OPERATION,
+                  "driBOData called on invalid buffer\n");
+      BM_CKFATAL(-EINVAL);
+   }
+   newBuffer = !buf->private || (pool->size(pool, buf->private) < size) ||
+      pool->map(pool, buf->private, DRM_BO_FLAG_WRITE,
+                DRM_BO_HINT_DONT_BLOCK, &virtual);
+
+   if (newBuffer) {
+      if (buf->private)
+         pool->destroy(pool, buf->private);
+      if (!flags)
+         flags = buf->flags;
+      buf->private = pool->create(pool, size, flags, 0, buf->alignment);
+      if (!buf->private)
+         BM_CKFATAL(-ENOMEM);
+      BM_CKFATAL(pool->map(pool, buf->private,
+                           DRM_BO_FLAG_WRITE,
+                           DRM_BO_HINT_DONT_BLOCK, &virtual));
+   }
+
+   if (data != NULL)
+      memcpy(virtual, data, size);
+
+   BM_CKFATAL(pool->unmap(pool, buf->private));
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+}
+
+void
+driBOSubData(struct _DriBufferObject *buf,
+             unsigned long offset, unsigned long size, const void *data)
+{
+   void *virtual;
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   if (size && data) {
+      BM_CKFATAL(buf->pool->map(buf->pool, buf->private,
+                                DRM_BO_FLAG_WRITE, 0, &virtual));
+      memcpy((unsigned char *) virtual + offset, data, size);
+      BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private));
+   }
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+}
+
+void
+driBOGetSubData(struct _DriBufferObject *buf,
+                unsigned long offset, unsigned long size, void *data)
+{
+   void *virtual;
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   if (size && data) {
+      BM_CKFATAL(buf->pool->map(buf->pool, buf->private,
+                                DRM_BO_FLAG_READ, 0, &virtual));
+      memcpy(data, (unsigned char *) virtual + offset, size);
+      BM_CKFATAL(buf->pool->unmap(buf->pool, buf->private));
+   }
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+}
+
+void
+driBOSetStatic(struct _DriBufferObject *buf,
+               unsigned long offset,
+               unsigned long size, void *virtual, unsigned flags)
+{
+   _glthread_LOCK_MUTEX(buf->mutex);
+   if (buf->private != NULL) {
+      _mesa_error(NULL, GL_INVALID_OPERATION,
+                  "Invalid buffer for setStatic\n");
+      BM_CKFATAL(-EINVAL);
+   }
+   if (buf->pool->setstatic == NULL) {
+      _mesa_error(NULL, GL_INVALID_OPERATION,
+                  "Invalid buffer pool for setStatic\n");
+      BM_CKFATAL(-EINVAL);
+   }
+
+   if (!flags)
+      flags = buf->flags;
+
+   buf->private = buf->pool->setstatic(buf->pool, offset, size,
+                                       virtual, flags);
+   if (!buf->private) {
+      _mesa_error(NULL, GL_OUT_OF_MEMORY,
+                  "Invalid buffer pool for setStatic\n");
+      BM_CKFATAL(-ENOMEM);
+   }
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+}
+
+
+
+void
+driGenBuffers(struct _DriBufferPool *pool,
+              const char *name,
+              unsigned n,
+              struct _DriBufferObject *buffers[],
+              unsigned alignment, unsigned flags, unsigned hint)
+{
+   struct _DriBufferObject *buf;
+   int i;
+
+   flags = (flags) ? flags : DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_MEM_VRAM |
+      DRM_BO_FLAG_MEM_LOCAL | DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE;
+
+
+   for (i = 0; i < n; ++i) {
+      buf = (struct _DriBufferObject *) calloc(1, sizeof(*buf));
+      if (!buf)
+         BM_CKFATAL(-ENOMEM);
+
+      _glthread_INIT_MUTEX(buf->mutex);
+      _glthread_LOCK_MUTEX(buf->mutex);
+      _glthread_LOCK_MUTEX(bmMutex);
+      buf->refCount = 1;
+      _glthread_UNLOCK_MUTEX(bmMutex);
+      buf->flags = flags;
+      buf->hint = hint;
+      buf->name = name;
+      buf->alignment = alignment;
+      buf->pool = pool;
+      _glthread_UNLOCK_MUTEX(buf->mutex);
+      buffers[i] = buf;
+   }
+}
+
+void
+driDeleteBuffers(unsigned n, struct _DriBufferObject *buffers[])
+{
+   int i;
+
+   for (i = 0; i < n; ++i) {
+      driBOUnReference(buffers[i]);
+   }
+}
+
+
+void
+driInitBufMgr(int fd)
+{
+   ;
+}
+
+
+void
+driBOCreateList(int target, drmBOList * list)
+{
+   _glthread_LOCK_MUTEX(bmMutex);
+   BM_CKFATAL(drmBOCreateList(20, list));
+   _glthread_UNLOCK_MUTEX(bmMutex);
+}
+
+void
+driBOResetList(drmBOList * list)
+{
+   _glthread_LOCK_MUTEX(bmMutex);
+   BM_CKFATAL(drmBOResetList(list));
+   _glthread_UNLOCK_MUTEX(bmMutex);
+}
+
+void
+driBOAddListItem(drmBOList * list, struct _DriBufferObject *buf,
+                 unsigned flags, unsigned mask)
+{
+   int newItem;
+
+   _glthread_LOCK_MUTEX(buf->mutex);
+   _glthread_LOCK_MUTEX(bmMutex);
+   BM_CKFATAL(drmAddValidateItem(list, driBOKernel(buf),
+                                 flags, mask, &newItem));
+   _glthread_UNLOCK_MUTEX(bmMutex);
+
+   /*
+    * Tell userspace pools to validate the buffer. This should be a 
+    * noop if the pool is already validated.
+    * FIXME: We should have a list for this as well.
+    */
+
+   if (buf->pool->validate) {
+      BM_CKFATAL(buf->pool->validate(buf->pool, buf->private));
+   }
+
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+}
+
+void
+driBOFence(struct _DriBufferObject *buf, struct _DriFenceObject *fence)
+{
+   _glthread_LOCK_MUTEX(buf->mutex);
+   BM_CKFATAL(buf->pool->fence(buf->pool, buf->private, fence));
+   _glthread_UNLOCK_MUTEX(buf->mutex);
+
+}
+
+void
+driBOValidateList(int fd, drmBOList * list)
+{
+   _glthread_LOCK_MUTEX(bmMutex);
+   BM_CKFATAL(drmBOValidateList(fd, list));
+   _glthread_UNLOCK_MUTEX(bmMutex);
+}
+
+void
+driPoolTakeDown(struct _DriBufferPool *pool)
+{
+   pool->takeDown(pool);
+
+}

src/mesa/drivers/dri/common/dri_bufmgr.h

@@ -0,0 +1,99 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellstr<74>m <thomas-at-tungstengraphics-dot-com>
+ *          Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ */
+
+#ifndef _DRI_BUFMGR_H_
+#define _DRI_BUFMGR_H_
+#include <xf86drm.h>
+
+
+struct _DriFenceObject;
+struct _DriBufferObject;
+struct _DriBufferPool;
+
+extern struct _DriFenceObject *driFenceBuffers(int fd, char *name,
+                                               unsigned flags);
+
+extern struct _DriFenceObject *driFenceReference(struct _DriFenceObject *fence);
+
+extern void driFenceUnReference(struct _DriFenceObject *fence);
+
+extern void
+driFenceFinish(struct _DriFenceObject *fence, unsigned type, int lazy);
+
+extern int driFenceSignaled(struct _DriFenceObject *fence, unsigned type);
+extern unsigned driFenceType(struct _DriFenceObject *fence);
+
+/*
+ * Return a pointer to the libdrm buffer object this DriBufferObject
+ * uses.
+ */
+
+extern drmBO *driBOKernel(struct _DriBufferObject *buf);
+extern void *driBOMap(struct _DriBufferObject *buf, unsigned flags,
+                      unsigned hint);
+extern void driBOUnmap(struct _DriBufferObject *buf);
+extern unsigned long driBOOffset(struct _DriBufferObject *buf);
+extern unsigned driBOFlags(struct _DriBufferObject *buf);
+extern struct _DriBufferObject *driBOReference(struct _DriBufferObject *buf);
+extern void driBOUnReference(struct _DriBufferObject *buf);
+extern void driBOData(struct _DriBufferObject *r_buf,
+                      unsigned size, const void *data, unsigned flags);
+extern void driBOSubData(struct _DriBufferObject *buf,
+                         unsigned long offset, unsigned long size,
+                         const void *data);
+extern void driBOGetSubData(struct _DriBufferObject *buf,
+                            unsigned long offset, unsigned long size,
+                            void *data);
+extern void driGenBuffers(struct _DriBufferPool *pool,
+                          const char *name,
+                          unsigned n,
+                          struct _DriBufferObject *buffers[],
+                          unsigned alignment, unsigned flags, unsigned hint);
+extern void driDeleteBuffers(unsigned n, struct _DriBufferObject *buffers[]);
+extern void driInitBufMgr(int fd);
+extern void driBOCreateList(int target, drmBOList * list);
+extern void driBOResetList(drmBOList * list);
+extern void driBOAddListItem(drmBOList * list, struct _DriBufferObject *buf,
+                             unsigned flags, unsigned mask);
+extern void driBOValidateList(int fd, drmBOList * list);
+
+extern void driBOFence(struct _DriBufferObject *buf,
+                       struct _DriFenceObject *fence);
+
+extern void driPoolTakeDown(struct _DriBufferPool *pool);
+extern void driBOSetStatic(struct _DriBufferObject *buf,
+                           unsigned long offset,
+                           unsigned long size, void *virtual, unsigned flags);
+extern void driBOWaitIdle(struct _DriBufferObject *buf, int lazy);
+extern void driPoolTakeDown(struct _DriBufferPool *pool);
+
+#endif

src/mesa/drivers/dri/common/dri_bufpool.h

@@ -0,0 +1,86 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellstr<74>m <thomas-at-tungstengraphics-dot-com>
+ */
+
+#ifndef _DRI_BUFPOOL_H_
+#define _DRI_BUFPOOL_H_
+
+#include <xf86drm.h>
+struct _DriFenceObject;
+
+typedef struct _DriBufferPool
+{
+   int fd;
+   int (*map) (struct _DriBufferPool * pool, void *private,
+               unsigned flags, int hint, void **virtual);
+   int (*unmap) (struct _DriBufferPool * pool, void *private);
+   int (*destroy) (struct _DriBufferPool * pool, void *private);
+   unsigned long (*offset) (struct _DriBufferPool * pool, void *private);
+   unsigned (*flags) (struct _DriBufferPool * pool, void *private);
+   unsigned long (*size) (struct _DriBufferPool * pool, void *private);
+   void *(*create) (struct _DriBufferPool * pool, unsigned long size,
+                    unsigned flags, unsigned hint, unsigned alignment);
+   int (*fence) (struct _DriBufferPool * pool, void *private,
+                 struct _DriFenceObject * fence);
+   drmBO *(*kernel) (struct _DriBufferPool * pool, void *private);
+   int (*validate) (struct _DriBufferPool * pool, void *private);
+   void *(*setstatic) (struct _DriBufferPool * pool, unsigned long offset,
+                       unsigned long size, void *virtual, unsigned flags);
+   int (*waitIdle) (struct _DriBufferPool *pool, void *private,
+		    int lazy);
+   void (*takeDown) (struct _DriBufferPool * pool);
+   void *data;
+} DriBufferPool;
+
+extern void bmError(int val, const char *file, const char *function,
+                    int line);
+#define BM_CKFATAL(val)					       \
+  do{							       \
+    int tstVal = (val);					       \
+    if (tstVal) 					       \
+      bmError(tstVal, __FILE__, __FUNCTION__, __LINE__);       \
+  } while(0);
+
+
+
+
+
+/*
+ * Builtin pools.
+ */
+
+/*
+ * Kernel buffer objects. Size in multiples of page size. Page size aligned.
+ */
+
+extern struct _DriBufferPool *driDRMPoolInit(int fd);
+extern struct _DriBufferPool *driDRMStaticPoolInit(int fd);
+
+#endif

src/mesa/drivers/dri/common/dri_drmpool.c

@@ -0,0 +1,221 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellstr<74>m <thomas-at-tungstengraphics-dot-com>
+ */
+
+#include <xf86drm.h>
+#include <stdlib.h>
+#include "dri_bufpool.h"
+
+/*
+ * Buffer pool implementation using DRM buffer objects as DRI buffer objects.
+ */
+
+static void *
+pool_create(struct _DriBufferPool *pool,
+            unsigned long size, unsigned flags, unsigned hint,
+            unsigned alignment)
+{
+   drmBO *buf = (drmBO *) malloc(sizeof(*buf));
+   int ret;
+
+   if (!buf)
+      return NULL;
+
+   if (alignment && ((4096 % alignment) != 0))
+      return NULL;
+
+   ret = drmBOCreate(pool->fd, NULL, 0, size, NULL, drm_bo_type_dc,
+                     flags, hint, buf);
+
+   if (ret) {
+      free(buf);
+      return NULL;
+   }
+
+   return (void *) buf;
+}
+
+static int
+pool_destroy(struct _DriBufferPool *pool, void *private)
+{
+   drmBO *buf = (drmBO *) private;
+   return drmBODestroy(pool->fd, buf);
+}
+
+static int
+pool_map(struct _DriBufferPool *pool, void *private, unsigned flags,
+         int hint, void **virtual)
+{
+   drmBO *buf = (drmBO *) private;
+
+   return drmBOMap(pool->fd, buf, flags, hint, virtual);
+}
+
+static int
+pool_unmap(struct _DriBufferPool *pool, void *private)
+{
+   drmBO *buf = (drmBO *) private;
+   return drmBOUnmap(pool->fd, buf);
+}
+
+static unsigned long
+pool_offset(struct _DriBufferPool *pool, void *private)
+{
+   drmBO *buf = (drmBO *) private;
+   return buf->offset;
+}
+
+static unsigned
+pool_flags(struct _DriBufferPool *pool, void *private)
+{
+   drmBO *buf = (drmBO *) private;
+   return buf->flags;
+}
+
+
+static unsigned long
+pool_size(struct _DriBufferPool *pool, void *private)
+{
+   drmBO *buf = (drmBO *) private;
+   return buf->size;
+}
+
+static int
+pool_fence(struct _DriBufferPool *pool, void *private,
+           struct _DriFenceObject *fence)
+{
+   /*
+    * Noop. The kernel handles all fencing.
+    */
+
+   return 0;
+}
+
+static drmBO *
+pool_kernel(struct _DriBufferPool *pool, void *private)
+{
+   return (drmBO *) private;
+}
+
+static int
+pool_waitIdle(struct _DriBufferPool *pool, void *private, int lazy)
+{
+   drmBO *buf = (drmBO *) private;
+   return drmBOWaitIdle(pool->fd, buf, (lazy) ? DRM_BO_HINT_WAIT_LAZY:0);
+}
+
+    
+static void
+pool_takedown(struct _DriBufferPool *pool)
+{
+   free(pool);
+}
+
+
+struct _DriBufferPool *
+driDRMPoolInit(int fd)
+{
+   struct _DriBufferPool *pool;
+
+   pool = (struct _DriBufferPool *) malloc(sizeof(*pool));
+
+   if (!pool)
+      return NULL;
+
+   pool->fd = fd;
+   pool->map = &pool_map;
+   pool->unmap = &pool_unmap;
+   pool->destroy = &pool_destroy;
+   pool->offset = &pool_offset;
+   pool->flags = &pool_flags;
+   pool->size = &pool_size;
+   pool->create = &pool_create;
+   pool->fence = &pool_fence;
+   pool->kernel = &pool_kernel;
+   pool->validate = NULL;
+   pool->setstatic = NULL;
+   pool->waitIdle = &pool_waitIdle;
+   pool->takeDown = &pool_takedown;
+   pool->data = NULL;
+   return pool;
+}
+
+
+static void *
+pool_setstatic(struct _DriBufferPool *pool, unsigned long offset,
+               unsigned long size, void *virtual, unsigned flags)
+{
+   drmBO *buf = (drmBO *) malloc(sizeof(*buf));
+   int ret;
+
+   if (!buf)
+      return NULL;
+
+   ret = drmBOCreate(pool->fd, NULL, offset, size, NULL, drm_bo_type_fake,
+                     flags, 0, buf);
+
+   if (ret) {
+      free(buf);
+      return NULL;
+   }
+
+   buf->virtual = virtual;
+
+   return (void *) buf;
+}
+
+
+struct _DriBufferPool *
+driDRMStaticPoolInit(int fd)
+{
+   struct _DriBufferPool *pool;
+
+   pool = (struct _DriBufferPool *) malloc(sizeof(*pool));
+
+   if (!pool)
+      return NULL;
+
+   pool->fd = fd;
+   pool->map = &pool_map;
+   pool->unmap = &pool_unmap;
+   pool->destroy = &pool_destroy;
+   pool->offset = &pool_offset;
+   pool->flags = &pool_flags;
+   pool->size = &pool_size;
+   pool->create = NULL;
+   pool->fence = &pool_fence;
+   pool->kernel = &pool_kernel;
+   pool->validate = NULL;
+   pool->setstatic = &pool_setstatic;
+   pool->waitIdle = &pool_waitIdle;
+   pool->takeDown = &pool_takedown;
+   pool->data = NULL;
+   return pool;
+}

src/mesa/drivers/dri/common/dri_util.c

@@ -403,8 +403,9 @@ __driUtilUpdateDrawableInfo(__DRIdrawablePrivate *pdp)
     __DRIcontextPrivate *pcp = pdp->driContextPriv;
     
     if (!pcp || (pdp != pcp->driDrawablePriv)) {
-	/* ERROR!!! */
-	return;
+	/* ERROR!!! 
+	 * ..but we must ignore it. There can be many contexts bound to a drawable.
+	 */
     }
 
     psp = pdp->driScreenPriv;
@@ -435,6 +436,7 @@ __driUtilUpdateDrawableInfo(__DRIdrawablePrivate *pdp)
 	/* Error -- eg the window may have been destroyed.  Keep going
 	 * with no cliprects.
 	 */
+        fprintf(stderr, "Drawable destroyed");
         pdp->pStamp = &pdp->lastStamp; /* prevent endless loop */
 	pdp->numClipRects = 0;
 	pdp->pClipRects = NULL;
@@ -841,7 +843,7 @@ static void driDestroyScreen(__DRInativeDisplay *dpy, int scrn, void *screenPriv
 	(void)drmUnmap((drmAddress)psp->pSAREA, SAREA_MAX);
 	(void)drmUnmap((drmAddress)psp->pFB, psp->fbSize);
 	_mesa_free(psp->pDevPriv);
-	(void)drmClose(psp->fd);
+	(void)drmCloseOnce(psp->fd);
 	if ( psp->modes != NULL ) {
 	    (*dri_interface->destroyContextModes)( psp->modes );
 	}

src/mesa/drivers/dri/common/dri_util.h

@@ -87,15 +87,15 @@ typedef struct __DRIutilversionRec2    __DRIutilversion2;
 #define DRI_VALIDATE_DRAWABLE_INFO(psp, pdp)                            \
 do {                                                                    \
     while (*(pdp->pStamp) != pdp->lastStamp) {                          \
-	DRM_UNLOCK(psp->fd, &psp->pSAREA->lock,                         \
-		   pdp->driContextPriv->hHWContext);                    \
+        register unsigned int hwContext = psp->pSAREA->lock.lock &      \
+		     ~(DRM_LOCK_HELD | DRM_LOCK_CONT);                  \
+	DRM_UNLOCK(psp->fd, &psp->pSAREA->lock, hwContext);             \
                                                                         \
 	DRM_SPINLOCK(&psp->pSAREA->drawable_lock, psp->drawLockID);     \
 	DRI_VALIDATE_DRAWABLE_INFO_ONCE(pdp);                           \
 	DRM_SPINUNLOCK(&psp->pSAREA->drawable_lock, psp->drawLockID);   \
                                                                         \
-	DRM_LIGHT_LOCK(psp->fd, &psp->pSAREA->lock,                     \
-		       pdp->driContextPriv->hHWContext);                \
+	DRM_LIGHT_LOCK(psp->fd, &psp->pSAREA->lock, hwContext);         \
     }                                                                   \
 } while (0)
 

src/mesa/drivers/dri/i915/Makefile

@@ -7,16 +7,6 @@ LIBNAME = i915_dri.so
 MINIGLX_SOURCES = server/intel_dri.c
 
 DRIVER_SOURCES = \
-	i915_context.c \
-	i915_debug.c \
-	i915_fragprog.c \
-	i915_metaops.c \
-	i915_program.c \
-	i915_state.c \
-	i915_tex.c \
-	i915_texprog.c \
-	i915_texstate.c \
-	i915_vtbl.c \
 	i830_context.c \
 	i830_metaops.c \
 	i830_state.c \
@@ -24,19 +14,43 @@ DRIVER_SOURCES = \
 	i830_tex.c \
 	i830_texstate.c \
 	i830_vtbl.c \
+	intel_render.c \
+	intel_regions.c \
+	intel_buffer_objects.c \
 	intel_batchbuffer.c \
+	intel_mipmap_tree.c \
+	i915_tex_layout.c \
+	intel_tex_image.c \
+	intel_tex_subimage.c \
+	intel_tex_copy.c \
+	intel_tex_validate.c \
+	intel_tex_format.c \
+	intel_tex.c \
+	intel_pixel.c \
+	intel_pixel_copy.c \
+	intel_pixel_read.c \
+	intel_pixel_draw.c \
+	intel_buffers.c \
+	intel_blit.c \
+	i915_tex.c \
+	i915_texstate.c \
+	i915_context.c \
+	i915_debug.c \
+	i915_fragprog.c \
+	i915_metaops.c \
+	i915_program.c \
+	i915_state.c \
+	i915_vtbl.c \
 	intel_context.c \
 	intel_ioctl.c \
-	intel_pixel.c \
-	intel_render.c \
 	intel_rotate.c \
 	intel_screen.c \
 	intel_span.c \
 	intel_state.c \
-	intel_tex.c \
-	intel_texmem.c \
-	intel_tris.c 
-
+	intel_tris.c \
+	intel_fbo.c \
+	intel_depthstencil.c \
+	intel_batchpool.c
 C_SOURCES = \
 	$(COMMON_SOURCES) \
 	$(DRIVER_SOURCES)

src/mesa/drivers/dri/i915/i830_context.c

@@ -38,37 +38,38 @@
  * Mesa's Driver Functions
  ***************************************/
 
-static const struct dri_extension i830_extensions[] =
-{
-    { "GL_ARB_texture_env_crossbar",       NULL },
-    { NULL,                                NULL }
+static const struct dri_extension i830_extensions[] = {
+   {"GL_ARB_texture_env_crossbar", NULL},
+   {NULL, NULL}
 };
 
 
-static void i830InitDriverFunctions( struct dd_function_table *functions )
+static void
+i830InitDriverFunctions(struct dd_function_table *functions)
 {
-   intelInitDriverFunctions( functions );
-   i830InitStateFuncs( functions );
-   i830InitTextureFuncs( functions );
+   intelInitDriverFunctions(functions);
+   i830InitStateFuncs(functions);
+   i830InitTextureFuncs(functions);
 }
 
 
-GLboolean i830CreateContext( const __GLcontextModes *mesaVis,
-			    __DRIcontextPrivate *driContextPriv,
-			    void *sharedContextPrivate)
+GLboolean
+i830CreateContext(const __GLcontextModes * mesaVis,
+                  __DRIcontextPrivate * driContextPriv,
+                  void *sharedContextPrivate)
 {
    struct dd_function_table functions;
-   i830ContextPtr i830 = (i830ContextPtr) CALLOC_STRUCT(i830_context);
-   intelContextPtr intel = &i830->intel;
+   struct i830_context *i830 = CALLOC_STRUCT(i830_context);
+   struct intel_context *intel = &i830->intel;
    GLcontext *ctx = &intel->ctx;
-   GLuint i;
-   if (!i830) return GL_FALSE;
+   if (!i830)
+      return GL_FALSE;
 
-   i830InitVtbl( i830 );
-   i830InitDriverFunctions( &functions );
+   i830InitVtbl(i830);
+   i830InitDriverFunctions(&functions);
 
-   if (!intelInitContext( intel, mesaVis, driContextPriv,
-			  sharedContextPrivate, &functions )) {
+   if (!intelInitContext(intel, mesaVis, driContextPriv,
+                         sharedContextPrivate, &functions)) {
       FREE(i830);
       return GL_FALSE;
    }
@@ -77,48 +78,27 @@ GLboolean i830CreateContext( const __GLcontextModes *mesaVis,
    intel->ctx.Const.MaxTextureImageUnits = I830_TEX_UNITS;
    intel->ctx.Const.MaxTextureCoordUnits = I830_TEX_UNITS;
 
-   intel->nr_heaps = 1;
-   intel->texture_heaps[0] = 
-      driCreateTextureHeap( 0, intel,
-			    intel->intelScreen->tex.size,
-			    12,
-			    I830_NR_TEX_REGIONS,
-			    intel->sarea->texList,
-			    (unsigned *) & intel->sarea->texAge,
-			    & intel->swapped,
-			    sizeof( struct i830_texture_object ),
-			    (destroy_texture_object_t *)intelDestroyTexObj );
-
-   /* FIXME: driCalculateMaxTextureLevels assumes that mipmaps are tightly
-    * FIXME: packed, but they're not in Intel graphics hardware.
+   /* Advertise the full hardware capabilities.  The new memory
+    * manager should cope much better with overload situations:
     */
-   intel->ctx.Const.MaxTextureUnits = I830_TEX_UNITS;
-   i = driQueryOptioni( &intel->intelScreen->optionCache, "allow_large_textures");
-   driCalculateMaxTextureLevels( intel->texture_heaps,
-				 intel->nr_heaps,
-				 &intel->ctx.Const,
-				 4,
-				 11, /* max 2D texture size is 2048x2048 */
-				 8,  /* max 3D texture size is 256^3 */
-				 10, /* max CUBE texture size is 1024x1024 */
-				 11, /* max RECT. supported */
-				 12,
-				 GL_FALSE,
-				 i );
+   ctx->Const.MaxTextureLevels = 12;
+   ctx->Const.Max3DTextureLevels = 9;
+   ctx->Const.MaxCubeTextureLevels = 11;
+   ctx->Const.MaxTextureRectSize = (1 << 11);
+   ctx->Const.MaxTextureUnits = I830_TEX_UNITS;
 
-   _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 
-		       18 * sizeof(GLfloat) );
+   _tnl_init_vertices(ctx, ctx->Const.MaxArrayLockSize + 12,
+                      18 * sizeof(GLfloat));
 
    intel->verts = TNL_CONTEXT(ctx)->clipspace.vertex_buf;
 
-   driInitExtensions( ctx, i830_extensions, GL_FALSE );
+   driInitExtensions(ctx, i830_extensions, GL_FALSE);
 
-   i830InitState( i830 );
+   i830InitState(i830);
+   i830InitMetaFuncs(i830);
 
-
-   _tnl_allow_vertex_fog( ctx, 1 ); 
-   _tnl_allow_pixel_fog( ctx, 0 ); 
+   _tnl_allow_vertex_fog(ctx, 1);
+   _tnl_allow_pixel_fog(ctx, 0);
 
    return GL_TRUE;
 }
-

src/mesa/drivers/dri/i915/i830_context.h

@@ -39,6 +39,7 @@
 #define I830_UPLOAD_CTX              0x1
 #define I830_UPLOAD_BUFFERS          0x2
 #define I830_UPLOAD_STIPPLE          0x4
+#define I830_UPLOAD_INVARIENT        0x8
 #define I830_UPLOAD_TEX(i)           (0x10<<(i))
 #define I830_UPLOAD_TEXBLEND(i)      (0x100<<(i))
 #define I830_UPLOAD_TEX_ALL          (0x0f0)
@@ -48,17 +49,15 @@
  */
 #define I830_DESTREG_CBUFADDR0 0
 #define I830_DESTREG_CBUFADDR1 1
-#define I830_DESTREG_CBUFADDR2 2
-#define I830_DESTREG_DBUFADDR0 3
-#define I830_DESTREG_DBUFADDR1 4
-#define I830_DESTREG_DBUFADDR2 5
-#define I830_DESTREG_DV0 6
-#define I830_DESTREG_DV1 7
-#define I830_DESTREG_SENABLE 8
-#define I830_DESTREG_SR0 9
-#define I830_DESTREG_SR1 10
-#define I830_DESTREG_SR2 11
-#define I830_DEST_SETUP_SIZE 12
+#define I830_DESTREG_DBUFADDR0 2
+#define I830_DESTREG_DBUFADDR1 3
+#define I830_DESTREG_DV0 4
+#define I830_DESTREG_DV1 5
+#define I830_DESTREG_SENABLE 6
+#define I830_DESTREG_SR0 7
+#define I830_DESTREG_SR1 8
+#define I830_DESTREG_SR2 9
+#define I830_DEST_SETUP_SIZE 10
 
 #define I830_CTXREG_STATE1		0
 #define I830_CTXREG_STATE2		1
@@ -72,7 +71,7 @@
 #define I830_CTXREG_AA			9
 #define I830_CTXREG_FOGCOLOR		10
 #define I830_CTXREG_BLENDCOLOR0		11
-#define I830_CTXREG_BLENDCOLOR1		12 
+#define I830_CTXREG_BLENDCOLOR1		12
 #define I830_CTXREG_VF			13
 #define I830_CTXREG_VF2			14
 #define I830_CTXREG_MCSB0		15
@@ -83,17 +82,16 @@
 #define I830_STPREG_ST1        1
 #define I830_STP_SETUP_SIZE    2
 
-#define I830_TEXREG_TM0LI      0 /* load immediate 2 texture map n */
-#define I830_TEXREG_TM0S0      1
-#define I830_TEXREG_TM0S1      2
-#define I830_TEXREG_TM0S2      3
-#define I830_TEXREG_TM0S3      4
-#define I830_TEXREG_TM0S4      5
-#define I830_TEXREG_MCS	       6	/* _3DSTATE_MAP_COORD_SETS */
-#define I830_TEXREG_CUBE       7	/* _3DSTATE_MAP_SUBE */
-#define I830_TEX_SETUP_SIZE    8
+#define I830_TEXREG_TM0LI      0        /* load immediate 2 texture map n */
+#define I830_TEXREG_TM0S1      1
+#define I830_TEXREG_TM0S2      2
+#define I830_TEXREG_TM0S3      3
+#define I830_TEXREG_TM0S4      4
+#define I830_TEXREG_MCS	       5        /* _3DSTATE_MAP_COORD_SETS */
+#define I830_TEXREG_CUBE       6        /* _3DSTATE_MAP_SUBE */
+#define I830_TEX_SETUP_SIZE    7
 
-#define I830_TEXBLEND_SIZE	12	/* (4 args + op) * 2 + COLOR_FACTOR */
+#define I830_TEXBLEND_SIZE	12      /* (4 args + op) * 2 + COLOR_FACTOR */
 
 struct i830_texture_object
 {
@@ -103,30 +101,39 @@ struct i830_texture_object
 
 #define I830_TEX_UNITS 4
 
-struct i830_hw_state {
+struct i830_hw_state
+{
    GLuint Ctx[I830_CTX_SETUP_SIZE];
    GLuint Buffer[I830_DEST_SETUP_SIZE];
    GLuint Stipple[I830_STP_SETUP_SIZE];
    GLuint Tex[I830_TEX_UNITS][I830_TEX_SETUP_SIZE];
    GLuint TexBlend[I830_TEX_UNITS][I830_TEXBLEND_SIZE];
    GLuint TexBlendWordsUsed[I830_TEX_UNITS];
-   GLuint emitted;		/* I810_UPLOAD_* */
+
+   struct intel_region *draw_region;
+   struct intel_region *depth_region;
+
+   /* Regions aren't actually that appropriate here as the memory may
+    * be from a PBO or FBO.  Just use the buffer id.  Will have to do
+    * this for draw and depth for FBO's...
+    */
+   struct _DriBufferObject *tex_buffer[I830_TEX_UNITS];
+   GLuint tex_offset[I830_TEX_UNITS];
+
+   GLuint emitted;              /* I810_UPLOAD_* */
    GLuint active;
 };
 
-struct i830_context 
+struct i830_context
 {
    struct intel_context intel;
-   
-   DECLARE_RENDERINPUTS(last_index_bitset);
+
+   GLuint lodbias_tm0s3[MAX_TEXTURE_UNITS];
+     DECLARE_RENDERINPUTS(last_index_bitset);
 
    struct i830_hw_state meta, initial, state, *current;
 };
 
-typedef struct i830_context *i830ContextPtr;
-typedef struct i830_texture_object *i830TextureObjectPtr;
-
-#define I830_CONTEXT(ctx)	((i830ContextPtr)(ctx))
 
 
 
@@ -147,71 +154,55 @@ do {						\
 
 /* i830_vtbl.c
  */
-extern void 
-i830InitVtbl( i830ContextPtr i830 );
+extern void i830InitVtbl(struct i830_context *i830);
 
 /* i830_context.c
  */
-extern GLboolean 
-i830CreateContext( const __GLcontextModes *mesaVis,
-		   __DRIcontextPrivate *driContextPriv,
-		   void *sharedContextPrivate);
+extern GLboolean
+i830CreateContext(const __GLcontextModes * mesaVis,
+                  __DRIcontextPrivate * driContextPriv,
+                  void *sharedContextPrivate);
 
 /* i830_tex.c, i830_texstate.c
  */
-extern void 
-i830UpdateTextureState( intelContextPtr intel );
+extern void i830UpdateTextureState(struct intel_context *intel);
 
-extern void 
-i830InitTextureFuncs( struct dd_function_table *functions );
-
-extern intelTextureObjectPtr
-i830AllocTexObj( struct gl_texture_object *tObj );
+extern void i830InitTextureFuncs(struct dd_function_table *functions);
 
 /* i830_texblend.c
  */
-extern GLuint i830SetTexEnvCombine(i830ContextPtr i830,
-    const struct gl_tex_env_combine_state * combine, GLint blendUnit,
-     GLuint texel_op, GLuint *state, const GLfloat *factor );
+extern GLuint i830SetTexEnvCombine(struct i830_context *i830,
+                                   const struct gl_tex_env_combine_state
+                                   *combine, GLint blendUnit, GLuint texel_op,
+                                   GLuint * state, const GLfloat * factor);
 
-extern void 
-i830EmitTextureBlend( i830ContextPtr i830 );
+extern void i830EmitTextureBlend(struct i830_context *i830);
 
 
 /* i830_state.c
  */
-extern void 
-i830InitStateFuncs( struct dd_function_table *functions );
+extern void i830InitStateFuncs(struct dd_function_table *functions);
 
-extern void 
-i830EmitState( i830ContextPtr i830 );
+extern void i830EmitState(struct i830_context *i830);
 
-extern void 
-i830InitState( i830ContextPtr i830 );
+extern void i830InitState(struct i830_context *i830);
 
 /* i830_metaops.c
  */
-extern GLboolean
-i830TryTextureReadPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *pack,
-			  GLvoid *pixels );
-
-extern GLboolean
-i830TryTextureDrawPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *unpack,
-			  const GLvoid *pixels );
-
-extern void 
-i830ClearWithTris( intelContextPtr intel, GLbitfield mask,
-		   GLboolean all, GLint cx, GLint cy, GLint cw, GLint ch);
+extern void i830InitMetaFuncs(struct i830_context *i830);
 
 extern void
-i830RotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
+i830RotateWindow(struct intel_context *intel, __DRIdrawablePrivate * dPriv,
                  GLuint srcBuf);
 
-#endif
+/*======================================================================
+ * Inline conversion functions.  These are better-typed than the
+ * macros used previously:
+ */
+static INLINE struct i830_context *
+i830_context(GLcontext * ctx)
+{
+   return (struct i830_context *) ctx;
+}
 
+#endif

src/mesa/drivers/dri/i915/i830_reg.h

@@ -407,7 +407,7 @@
 #define LOGICOP_SET			0xf
 #define MODE4_ENABLE_STENCIL_TEST_MASK	((1<<17)|(0xff00))
 #define ENABLE_STENCIL_TEST_MASK	(1<<17)
-#define STENCIL_TEST_MASK(x)		((x)<<8)
+#define STENCIL_TEST_MASK(x)		(((x)&0xff)<<8)
 #define MODE4_ENABLE_STENCIL_WRITE_MASK	((1<<16)|(0x00ff))
 #define ENABLE_STENCIL_WRITE_MASK	(1<<16)
 #define STENCIL_WRITE_MASK(x)		((x)&0xff)
@@ -554,8 +554,8 @@
 #define    MAPSURF_4BIT_INDEXED		   (7<<6)
 #define TM0S1_MT_FORMAT_MASK         (0x7 << 3)
 #define TM0S1_MT_FORMAT_SHIFT        3
-#define    MT_4BIT_IDX_ARGB8888	           (7<<3) /* SURFACE_4BIT_INDEXED */
-#define    MT_8BIT_IDX_RGB565	           (0<<3) /* SURFACE_8BIT_INDEXED */
+#define    MT_4BIT_IDX_ARGB8888	           (7<<3)       /* SURFACE_4BIT_INDEXED */
+#define    MT_8BIT_IDX_RGB565	           (0<<3)       /* SURFACE_8BIT_INDEXED */
 #define    MT_8BIT_IDX_ARGB1555	           (1<<3)
 #define    MT_8BIT_IDX_ARGB4444	           (2<<3)
 #define    MT_8BIT_IDX_AY88		   (3<<3)
@@ -563,9 +563,9 @@
 #define    MT_8BIT_IDX_BUMP_88DVDU 	   (5<<3)
 #define    MT_8BIT_IDX_BUMP_655LDVDU	   (6<<3)
 #define    MT_8BIT_IDX_ARGB8888	           (7<<3)
-#define    MT_8BIT_I8		           (0<<3) /* SURFACE_8BIT */
+#define    MT_8BIT_I8		           (0<<3)       /* SURFACE_8BIT */
 #define    MT_8BIT_L8		           (1<<3)
-#define    MT_16BIT_RGB565 		   (0<<3) /* SURFACE_16BIT */
+#define    MT_16BIT_RGB565 		   (0<<3)       /* SURFACE_16BIT */
 #define    MT_16BIT_ARGB1555		   (1<<3)
 #define    MT_16BIT_ARGB4444		   (2<<3)
 #define    MT_16BIT_AY88		   (3<<3)
@@ -573,16 +573,16 @@
 #define    MT_16BIT_BUMP_88DVDU	           (5<<3)
 #define    MT_16BIT_BUMP_655LDVDU	   (6<<3)
 #define    MT_16BIT_DIB_RGB565_8888	   (7<<3)
-#define    MT_32BIT_ARGB8888		   (0<<3) /* SURFACE_32BIT */
+#define    MT_32BIT_ARGB8888		   (0<<3)       /* SURFACE_32BIT */
 #define    MT_32BIT_ABGR8888		   (1<<3)
 #define    MT_32BIT_BUMP_XLDVDU_8888	   (6<<3)
 #define    MT_32BIT_DIB_8888		   (7<<3)
-#define    MT_411_YUV411		   (0<<3) /* SURFACE_411 */
-#define    MT_422_YCRCB_SWAPY	           (0<<3) /* SURFACE_422 */
+#define    MT_411_YUV411		   (0<<3)       /* SURFACE_411 */
+#define    MT_422_YCRCB_SWAPY	           (0<<3)       /* SURFACE_422 */
 #define    MT_422_YCRCB_NORMAL	           (1<<3)
 #define    MT_422_YCRCB_SWAPUV	           (2<<3)
 #define    MT_422_YCRCB_SWAPUVY	           (3<<3)
-#define    MT_COMPRESS_DXT1		   (0<<3) /* SURFACE_COMPRESSED */
+#define    MT_COMPRESS_DXT1		   (0<<3)       /* SURFACE_COMPRESSED */
 #define    MT_COMPRESS_DXT2_3	           (1<<3)
 #define    MT_COMPRESS_DXT4_5	           (2<<3)
 #define    MT_COMPRESS_FXT1		   (3<<3)

src/mesa/drivers/dri/i915/i830_tex.c

@@ -45,261 +45,13 @@
 
 
 
-
-/**
- * Set the texture wrap modes.
- * 
- * The i830M (and related graphics cores) do not support GL_CLAMP.  The Intel
- * drivers for "other operating systems" implement GL_CLAMP as
- * GL_CLAMP_TO_EDGE, so the same is done here.
- * 
- * \param t Texture object whose wrap modes are to be set
- * \param swrap Wrap mode for the \a s texture coordinate
- * \param twrap Wrap mode for the \a t texture coordinate
- */
-static void i830SetTexWrapping(i830TextureObjectPtr tex,
-			       GLenum swrap, 
-			       GLenum twrap)
+static void
+i830TexEnv(GLcontext * ctx, GLenum target,
+           GLenum pname, const GLfloat * param)
 {
-   tex->Setup[I830_TEXREG_MCS] &= ~(TEXCOORD_ADDR_U_MASK|TEXCOORD_ADDR_V_MASK);
-
-   switch( swrap ) {
-   case GL_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_WRAP);
-      break;
-   case GL_CLAMP:
-   case GL_CLAMP_TO_EDGE:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_CLAMP);
-      break;
-   case GL_CLAMP_TO_BORDER:
-      tex->Setup[I830_TEXREG_MCS] |= 
-			TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_CLAMP_BORDER);
-      break;
-   case GL_MIRRORED_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |= 
-			TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_MIRROR);
-      break;
-   default:
-      break;
-   }
-
-   switch( twrap ) {
-   case GL_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_WRAP);
-      break;
-   case GL_CLAMP:
-   case GL_CLAMP_TO_EDGE:
-      tex->Setup[I830_TEXREG_MCS] |= TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_CLAMP);
-      break;
-   case GL_CLAMP_TO_BORDER:
-      tex->Setup[I830_TEXREG_MCS] |= 
-			TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_CLAMP_BORDER);
-      break;
-   case GL_MIRRORED_REPEAT:
-      tex->Setup[I830_TEXREG_MCS] |=
-			TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_MIRROR);
-      break;
-   default:
-      break;
-   }
-}
-
-
-/**
- * Set the texture magnification and minification modes.
- * 
- * \param t Texture whose filter modes are to be set
- * \param minf Texture minification mode
- * \param magf Texture magnification mode
- * \param bias LOD bias for this texture unit.
- */
-
-static void i830SetTexFilter( i830TextureObjectPtr t, GLenum minf, GLenum magf,
-			      GLfloat maxanisotropy )
-{
-   int minFilt = 0, mipFilt = 0, magFilt = 0;
-
-   if(INTEL_DEBUG&DEBUG_DRI)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if ( maxanisotropy > 1.0 ) {
-      minFilt = FILTER_ANISOTROPIC;
-      magFilt = FILTER_ANISOTROPIC;
-   }
-   else {
-      switch (minf) {
-      case GL_NEAREST:
-	 minFilt = FILTER_NEAREST;
-	 mipFilt = MIPFILTER_NONE;
-	 break;
-      case GL_LINEAR:
-	 minFilt = FILTER_LINEAR;
-	 mipFilt = MIPFILTER_NONE;
-	 break;
-      case GL_NEAREST_MIPMAP_NEAREST:
-	 minFilt = FILTER_NEAREST;
-	 mipFilt = MIPFILTER_NEAREST;
-	 break;
-      case GL_LINEAR_MIPMAP_NEAREST:
-	 minFilt = FILTER_LINEAR;
-	 mipFilt = MIPFILTER_NEAREST;
-	 break;
-      case GL_NEAREST_MIPMAP_LINEAR:
-	 minFilt = FILTER_NEAREST;
-	 mipFilt = MIPFILTER_LINEAR;
-	 break;
-      case GL_LINEAR_MIPMAP_LINEAR:
-	 minFilt = FILTER_LINEAR;
-	 mipFilt = MIPFILTER_LINEAR;
-	 break;
-      default:
-	 break;
-      }
-
-      switch (magf) {
-      case GL_NEAREST:
-	 magFilt = FILTER_NEAREST;
-	 break;
-      case GL_LINEAR:
-	 magFilt = FILTER_LINEAR;
-	 break;
-      default:
-	 break;
-      }  
-   }
-
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MIN_FILTER_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MIP_FILTER_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MAG_FILTER_MASK;
-   t->Setup[I830_TEXREG_TM0S3] |= ((minFilt << TM0S3_MIN_FILTER_SHIFT) |
-				   (mipFilt << TM0S3_MIP_FILTER_SHIFT) |
-				   (magFilt << TM0S3_MAG_FILTER_SHIFT));
-}
-
-static void i830SetTexBorderColor(i830TextureObjectPtr t, GLubyte color[4])
-{
-   if(INTEL_DEBUG&DEBUG_DRI)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-    t->Setup[I830_TEXREG_TM0S4] = 
-        INTEL_PACKCOLOR8888(color[0],color[1],color[2],color[3]);
-}
-
-
-/**
- * Allocate space for and load the mesa images into the texture memory block.
- * This will happen before drawing with a new texture, or drawing with a
- * texture after it was swapped out or teximaged again.
- */
-
-intelTextureObjectPtr i830AllocTexObj( struct gl_texture_object *texObj )
-{
-   i830TextureObjectPtr t = CALLOC_STRUCT( i830_texture_object );
-   if ( !t ) 
-      return NULL;
-
-   texObj->DriverData = t;
-   t->intel.base.tObj = texObj;
-   t->intel.dirty = I830_UPLOAD_TEX_ALL;
-   make_empty_list( &t->intel.base );
-
-   t->Setup[I830_TEXREG_TM0LI] = 0; /* not used */
-   t->Setup[I830_TEXREG_TM0S0] = 0;
-   t->Setup[I830_TEXREG_TM0S1] = 0;
-   t->Setup[I830_TEXREG_TM0S2] = 0;
-   t->Setup[I830_TEXREG_TM0S3] = 0;
-   t->Setup[I830_TEXREG_MCS] = (_3DSTATE_MAP_COORD_SET_CMD |
-				MAP_UNIT(0) |
-				ENABLE_TEXCOORD_PARAMS |
-				TEXCOORDS_ARE_NORMAL |
-				TEXCOORDTYPE_CARTESIAN |
-				ENABLE_ADDR_V_CNTL |
-				TEXCOORD_ADDR_V_MODE(TEXCOORDMODE_WRAP) |
-				ENABLE_ADDR_U_CNTL |
-				TEXCOORD_ADDR_U_MODE(TEXCOORDMODE_WRAP));
-
-   
-   i830SetTexWrapping( t, texObj->WrapS, texObj->WrapT );
-   i830SetTexFilter( t, texObj->MinFilter, texObj->MagFilter, 
-		     texObj->MaxAnisotropy );
-   i830SetTexBorderColor( t, texObj->_BorderChan );
-
-   return &t->intel;
-}
-
-
-static void i830TexParameter( GLcontext *ctx, GLenum target,
-			      struct gl_texture_object *tObj,
-			      GLenum pname, const GLfloat *params )
-{
-   i830TextureObjectPtr t = (i830TextureObjectPtr) tObj->DriverData;
-   if (!t)
-      return;
 
    switch (pname) {
-   case GL_TEXTURE_MIN_FILTER:
-   case GL_TEXTURE_MAG_FILTER:
-   case GL_TEXTURE_MAX_ANISOTROPY_EXT:
-      i830SetTexFilter( t, tObj->MinFilter, tObj->MagFilter,
-			tObj->MaxAnisotropy);
-      break;
-
-   case GL_TEXTURE_WRAP_S:
-   case GL_TEXTURE_WRAP_T:
-      i830SetTexWrapping( t, tObj->WrapS, tObj->WrapT );
-      break;
-  
-   case GL_TEXTURE_BORDER_COLOR:
-      i830SetTexBorderColor( t, tObj->_BorderChan );
-      break;
-
-   case GL_TEXTURE_BASE_LEVEL:
-   case GL_TEXTURE_MAX_LEVEL:
-   case GL_TEXTURE_MIN_LOD:
-   case GL_TEXTURE_MAX_LOD:
-      /* The i830 and its successors can do a lot of this without
-       * reloading the textures.  A project for someone?
-       */
-      intelFlush( ctx );
-      driSwapOutTextureObject( (driTextureObject *) t );
-      break;
-
-   default:
-      return;
-   }
-
-   t->intel.dirty = I830_UPLOAD_TEX_ALL;
-}
-
-
-static void i830TexEnv( GLcontext *ctx, GLenum target, 
-			GLenum pname, const GLfloat *param )
-{
-   i830ContextPtr i830 = I830_CONTEXT( ctx );
-   GLuint unit = ctx->Texture.CurrentUnit;
-
-   switch (pname) {
-   case GL_TEXTURE_ENV_COLOR: 
-#if 0
-   {
-      GLubyte r, g, b, a;
-      GLuint col;
-      
-      UNCLAMPED_FLOAT_TO_UBYTE(r, param[RCOMP]);
-      UNCLAMPED_FLOAT_TO_UBYTE(g, param[GCOMP]);
-      UNCLAMPED_FLOAT_TO_UBYTE(b, param[BCOMP]);
-      UNCLAMPED_FLOAT_TO_UBYTE(a, param[ACOMP]);
-
-      col = ((a << 24) | (r << 16) | (g << 8) | b);
-
-      if (col != i830->state.TexEnv[unit][I830_TEXENVREG_COL1]) {
-	 I830_STATECHANGE(i830, I830_UPLOAD_TEXENV);
-	 i830->state.TexEnv[unit][I830_TEXENVREG_COL1] = col;
-      }
-
-      break;
-   }
-#endif
+   case GL_TEXTURE_ENV_COLOR:
    case GL_TEXTURE_ENV_MODE:
    case GL_COMBINE_RGB:
    case GL_COMBINE_ALPHA:
@@ -319,38 +71,30 @@ static void i830TexEnv( GLcontext *ctx, GLenum target,
    case GL_ALPHA_SCALE:
       break;
 
-   case GL_TEXTURE_LOD_BIAS: {
-      int b = (int) ((*param) * 16.0);
-      if (b > 63) b = 63;
-      if (b < -64) b = -64;
-      I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_TM0S3] &= ~TM0S3_LOD_BIAS_MASK;
-      i830->state.Tex[unit][I830_TEXREG_TM0S3] |= 
-	 ((b << TM0S3_LOD_BIAS_SHIFT) & TM0S3_LOD_BIAS_MASK);
-      break;
-   }
+   case GL_TEXTURE_LOD_BIAS:{
+         struct i830_context *i830 = i830_context(ctx);
+         GLuint unit = ctx->Texture.CurrentUnit;
+         int b = (int) ((*param) * 16.0);
+         if (b > 63)
+            b = 63;
+         if (b < -64)
+            b = -64;
+         I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
+         i830->lodbias_tm0s3[unit] =
+            ((b << TM0S3_LOD_BIAS_SHIFT) & TM0S3_LOD_BIAS_MASK);
+         break;
+      }
 
    default:
       break;
    }
 }
 
-static void i830BindTexture( GLcontext *ctx, GLenum target,
-			    struct gl_texture_object *texObj )
+
+
+
+void
+i830InitTextureFuncs(struct dd_function_table *functions)
 {
-   i830TextureObjectPtr tex;
-   
-   if (!texObj->DriverData)
-      i830AllocTexObj( texObj );
-   
-   tex = (i830TextureObjectPtr)texObj->DriverData;
-}
-
-
-
-void i830InitTextureFuncs( struct dd_function_table *functions )
-{
-   functions->BindTexture 		= i830BindTexture;
-   functions->TexEnv                    = i830TexEnv;
-   functions->TexParameter              = i830TexParameter;
+   functions->TexEnv = i830TexEnv;
 }

src/mesa/drivers/dri/i915/i830_texblend.c

@@ -46,46 +46,42 @@
 /* ================================================================
  * Texture combine functions
  */
-static GLuint pass_through( GLuint *state, GLuint blendUnit )
+static GLuint
+pass_through(GLuint * state, GLuint blendUnit)
 {
    state[0] = (_3DSTATE_MAP_BLEND_OP_CMD(blendUnit) |
-	       TEXPIPE_COLOR |
-	       ENABLE_TEXOUTPUT_WRT_SEL |
-	       TEXOP_OUTPUT_CURRENT |
-	       DISABLE_TEX_CNTRL_STAGE |
-	       TEXOP_SCALE_1X |
-	       TEXOP_MODIFY_PARMS |
-	       TEXBLENDOP_ARG1);
+               TEXPIPE_COLOR |
+               ENABLE_TEXOUTPUT_WRT_SEL |
+               TEXOP_OUTPUT_CURRENT |
+               DISABLE_TEX_CNTRL_STAGE |
+               TEXOP_SCALE_1X | TEXOP_MODIFY_PARMS | TEXBLENDOP_ARG1);
    state[1] = (_3DSTATE_MAP_BLEND_OP_CMD(blendUnit) |
-	       TEXPIPE_ALPHA |
-	       ENABLE_TEXOUTPUT_WRT_SEL |
-	       TEXOP_OUTPUT_CURRENT |
-	       TEXOP_SCALE_1X |
-	       TEXOP_MODIFY_PARMS |
-	       TEXBLENDOP_ARG1);
+               TEXPIPE_ALPHA |
+               ENABLE_TEXOUTPUT_WRT_SEL |
+               TEXOP_OUTPUT_CURRENT |
+               TEXOP_SCALE_1X | TEXOP_MODIFY_PARMS | TEXBLENDOP_ARG1);
    state[2] = (_3DSTATE_MAP_BLEND_ARG_CMD(blendUnit) |
-	       TEXPIPE_COLOR |
-	       TEXBLEND_ARG1 |
-	       TEXBLENDARG_MODIFY_PARMS |
-	       TEXBLENDARG_CURRENT);
+               TEXPIPE_COLOR |
+               TEXBLEND_ARG1 |
+               TEXBLENDARG_MODIFY_PARMS | TEXBLENDARG_CURRENT);
    state[3] = (_3DSTATE_MAP_BLEND_ARG_CMD(blendUnit) |
-	       TEXPIPE_ALPHA |
-	       TEXBLEND_ARG1 |
-	       TEXBLENDARG_MODIFY_PARMS |
-	       TEXBLENDARG_CURRENT);
+               TEXPIPE_ALPHA |
+               TEXBLEND_ARG1 |
+               TEXBLENDARG_MODIFY_PARMS | TEXBLENDARG_CURRENT);
 
    return 4;
 }
 
-static GLuint emit_factor( GLuint blendUnit, GLuint *state, GLuint count, 
-			   const GLfloat *factor )
+static GLuint
+emit_factor(GLuint blendUnit, GLuint * state, GLuint count,
+            const GLfloat * factor)
 {
    GLubyte r, g, b, a;
    GLuint col;
-      
+
    if (0)
       fprintf(stderr, "emit constant %d: %.2f %.2f %.2f %.2f\n",
-	  blendUnit, factor[0], factor[1], factor[2], factor[3]);
+              blendUnit, factor[0], factor[1], factor[2], factor[3]);
 
    UNCLAMPED_FLOAT_TO_UBYTE(r, factor[0]);
    UNCLAMPED_FLOAT_TO_UBYTE(g, factor[1]);
@@ -94,21 +90,27 @@ static GLuint emit_factor( GLuint blendUnit, GLuint *state, GLuint count,
 
    col = ((a << 24) | (r << 16) | (g << 8) | b);
 
-   state[count++] = _3DSTATE_COLOR_FACTOR_N_CMD(blendUnit); 
+   state[count++] = _3DSTATE_COLOR_FACTOR_N_CMD(blendUnit);
    state[count++] = col;
 
    return count;
 }
 
 
-static __inline__ GLuint GetTexelOp(GLint unit)
+static INLINE GLuint
+GetTexelOp(GLint unit)
 {
-   switch(unit) {
-   case 0: return TEXBLENDARG_TEXEL0;
-   case 1: return TEXBLENDARG_TEXEL1;
-   case 2: return TEXBLENDARG_TEXEL2;
-   case 3: return TEXBLENDARG_TEXEL3;
-   default: return TEXBLENDARG_TEXEL0;
+   switch (unit) {
+   case 0:
+      return TEXBLENDARG_TEXEL0;
+   case 1:
+      return TEXBLENDARG_TEXEL1;
+   case 2:
+      return TEXBLENDARG_TEXEL2;
+   case 3:
+      return TEXBLENDARG_TEXEL3;
+   default:
+      return TEXBLENDARG_TEXEL0;
    }
 }
 
@@ -132,12 +134,10 @@ static __inline__ GLuint GetTexelOp(GLint unit)
  * partial support for the extension?
  */
 GLuint
-i830SetTexEnvCombine(i830ContextPtr i830,
-		     const struct gl_tex_env_combine_state * combine,
-		     GLint blendUnit,
-		     GLuint texel_op,
-		     GLuint *state,
-		     const GLfloat *factor )
+i830SetTexEnvCombine(struct i830_context * i830,
+                     const struct gl_tex_env_combine_state * combine,
+                     GLint blendUnit,
+                     GLuint texel_op, GLuint * state, const GLfloat * factor)
 {
    const GLuint numColorArgs = combine->_NumArgsRGB;
    const GLuint numAlphaArgs = combine->_NumArgsA;
@@ -162,7 +162,7 @@ i830SetTexEnvCombine(i830ContextPtr i830,
       TEXPIPE_ALPHA | TEXBLEND_ARG0 | TEXBLENDARG_MODIFY_PARMS,
    };
 
-   if(INTEL_DEBUG&DEBUG_TEXTURE)
+   if (INTEL_DEBUG & DEBUG_TEXTURE)
       fprintf(stderr, "%s\n", __FUNCTION__);
 
 
@@ -188,23 +188,23 @@ i830SetTexEnvCombine(i830ContextPtr i830,
    }
 
 
-   switch(combine->ModeRGB) {
-   case GL_REPLACE: 
+   switch (combine->ModeRGB) {
+   case GL_REPLACE:
       blendop = TEXBLENDOP_ARG1;
       break;
-   case GL_MODULATE: 
+   case GL_MODULATE:
       blendop = TEXBLENDOP_MODULATE;
       break;
-   case GL_ADD: 
+   case GL_ADD:
       blendop = TEXBLENDOP_ADD;
       break;
    case GL_ADD_SIGNED:
-      blendop = TEXBLENDOP_ADDSIGNED; 
+      blendop = TEXBLENDOP_ADDSIGNED;
       break;
    case GL_INTERPOLATE:
-      blendop = TEXBLENDOP_BLEND; 
+      blendop = TEXBLENDOP_BLEND;
       break;
-   case GL_SUBTRACT: 
+   case GL_SUBTRACT:
       blendop = TEXBLENDOP_SUBTRACT;
       break;
    case GL_DOT3_RGB_EXT:
@@ -215,55 +215,54 @@ i830SetTexEnvCombine(i830ContextPtr i830,
    case GL_DOT3_RGBA:
       blendop = TEXBLENDOP_DOT3;
       break;
-   default: 
-      return pass_through( state, blendUnit );
+   default:
+      return pass_through(state, blendUnit);
    }
 
    blendop |= (rgb_shift << TEXOP_SCALE_SHIFT);
 
 
    /* Handle RGB args */
-   for(i = 0; i < 3; i++) {
-      switch(combine->SourceRGB[i]) {
-      case GL_TEXTURE: 
-	 args_RGB[i] = texel_op;
-	 break;
+   for (i = 0; i < 3; i++) {
+      switch (combine->SourceRGB[i]) {
+      case GL_TEXTURE:
+         args_RGB[i] = texel_op;
+         break;
       case GL_TEXTURE0:
       case GL_TEXTURE1:
       case GL_TEXTURE2:
       case GL_TEXTURE3:
-	 args_RGB[i] = GetTexelOp( combine->SourceRGB[i] - GL_TEXTURE0 );
-	 break;
+         args_RGB[i] = GetTexelOp(combine->SourceRGB[i] - GL_TEXTURE0);
+         break;
       case GL_CONSTANT:
-	 args_RGB[i] = TEXBLENDARG_FACTOR_N; 
-	 need_factor = 1;
-	 break;
+         args_RGB[i] = TEXBLENDARG_FACTOR_N;
+         need_factor = 1;
+         break;
       case GL_PRIMARY_COLOR:
-	 args_RGB[i] = TEXBLENDARG_DIFFUSE;
-	 break;
+         args_RGB[i] = TEXBLENDARG_DIFFUSE;
+         break;
       case GL_PREVIOUS:
-	 args_RGB[i] = TEXBLENDARG_CURRENT; 
-	 break;
-      default: 
-	 return pass_through( state, blendUnit );
+         args_RGB[i] = TEXBLENDARG_CURRENT;
+         break;
+      default:
+         return pass_through(state, blendUnit);
       }
 
-      switch(combine->OperandRGB[i]) {
-      case GL_SRC_COLOR: 
-	 args_RGB[i] |= 0;
-	 break;
-      case GL_ONE_MINUS_SRC_COLOR: 
-	 args_RGB[i] |= TEXBLENDARG_INV_ARG;
-	 break;
-      case GL_SRC_ALPHA: 
-	 args_RGB[i] |= TEXBLENDARG_REPLICATE_ALPHA;
-	 break;
-      case GL_ONE_MINUS_SRC_ALPHA: 
-	 args_RGB[i] |= (TEXBLENDARG_REPLICATE_ALPHA | 
-			 TEXBLENDARG_INV_ARG);
-	 break;
-      default: 
-	 return pass_through( state, blendUnit );
+      switch (combine->OperandRGB[i]) {
+      case GL_SRC_COLOR:
+         args_RGB[i] |= 0;
+         break;
+      case GL_ONE_MINUS_SRC_COLOR:
+         args_RGB[i] |= TEXBLENDARG_INV_ARG;
+         break;
+      case GL_SRC_ALPHA:
+         args_RGB[i] |= TEXBLENDARG_REPLICATE_ALPHA;
+         break;
+      case GL_ONE_MINUS_SRC_ALPHA:
+         args_RGB[i] |= (TEXBLENDARG_REPLICATE_ALPHA | TEXBLENDARG_INV_ARG);
+         break;
+      default:
+         return pass_through(state, blendUnit);
       }
    }
 
@@ -275,76 +274,76 @@ i830SetTexEnvCombine(i830ContextPtr i830,
     * Note - the global factor is set up with alpha == .5, so 
     * the alpha part of the DOT4 calculation should be zero.
     */
-   if ( combine->ModeRGB == GL_DOT3_RGBA_EXT || 
-	combine->ModeRGB == GL_DOT3_RGBA ) {
+   if (combine->ModeRGB == GL_DOT3_RGBA_EXT ||
+       combine->ModeRGB == GL_DOT3_RGBA) {
       ablendop = TEXBLENDOP_DOT4;
-      args_A[0] = TEXBLENDARG_FACTOR; /* the global factor */
+      args_A[0] = TEXBLENDARG_FACTOR;   /* the global factor */
       args_A[1] = TEXBLENDARG_FACTOR;
       args_A[2] = TEXBLENDARG_FACTOR;
    }
    else {
-      switch(combine->ModeA) {
-      case GL_REPLACE: 
-	 ablendop = TEXBLENDOP_ARG1;
-	 break;
-      case GL_MODULATE: 
-	 ablendop = TEXBLENDOP_MODULATE;
-	 break;
-      case GL_ADD: 
-	 ablendop = TEXBLENDOP_ADD;
-	 break;
+      switch (combine->ModeA) {
+      case GL_REPLACE:
+         ablendop = TEXBLENDOP_ARG1;
+         break;
+      case GL_MODULATE:
+         ablendop = TEXBLENDOP_MODULATE;
+         break;
+      case GL_ADD:
+         ablendop = TEXBLENDOP_ADD;
+         break;
       case GL_ADD_SIGNED:
-	 ablendop = TEXBLENDOP_ADDSIGNED; 
-	 break;
+         ablendop = TEXBLENDOP_ADDSIGNED;
+         break;
       case GL_INTERPOLATE:
-	 ablendop = TEXBLENDOP_BLEND; 
-	 break;
-      case GL_SUBTRACT: 
-	 ablendop = TEXBLENDOP_SUBTRACT;
-	 break;
+         ablendop = TEXBLENDOP_BLEND;
+         break;
+      case GL_SUBTRACT:
+         ablendop = TEXBLENDOP_SUBTRACT;
+         break;
       default:
-	 return pass_through( state, blendUnit );
+         return pass_through(state, blendUnit);
       }
 
 
       ablendop |= (alpha_shift << TEXOP_SCALE_SHIFT);
 
       /* Handle A args */
-      for(i = 0; i < 3; i++) {
-	 switch(combine->SourceA[i]) {
-	 case GL_TEXTURE: 
-	    args_A[i] = texel_op;
-	    break;
-	 case GL_TEXTURE0:
-	 case GL_TEXTURE1:
-	 case GL_TEXTURE2:
-	 case GL_TEXTURE3:
-	    args_A[i] = GetTexelOp( combine->SourceA[i] - GL_TEXTURE0 );
-	    break;
-	 case GL_CONSTANT:
-	    args_A[i] = TEXBLENDARG_FACTOR_N; 
-	    need_factor = 1;
-	    break;
-	 case GL_PRIMARY_COLOR:
-	    args_A[i] = TEXBLENDARG_DIFFUSE; 
-	    break;
-	 case GL_PREVIOUS:
-	    args_A[i] = TEXBLENDARG_CURRENT; 
-	    break;
-	 default: 
-	    return pass_through( state, blendUnit );
-	 }
+      for (i = 0; i < 3; i++) {
+         switch (combine->SourceA[i]) {
+         case GL_TEXTURE:
+            args_A[i] = texel_op;
+            break;
+         case GL_TEXTURE0:
+         case GL_TEXTURE1:
+         case GL_TEXTURE2:
+         case GL_TEXTURE3:
+            args_A[i] = GetTexelOp(combine->SourceA[i] - GL_TEXTURE0);
+            break;
+         case GL_CONSTANT:
+            args_A[i] = TEXBLENDARG_FACTOR_N;
+            need_factor = 1;
+            break;
+         case GL_PRIMARY_COLOR:
+            args_A[i] = TEXBLENDARG_DIFFUSE;
+            break;
+         case GL_PREVIOUS:
+            args_A[i] = TEXBLENDARG_CURRENT;
+            break;
+         default:
+            return pass_through(state, blendUnit);
+         }
 
-	 switch(combine->OperandA[i]) {
-	 case GL_SRC_ALPHA: 
-	    args_A[i] |= 0;
-	    break;
-	 case GL_ONE_MINUS_SRC_ALPHA: 
-	    args_A[i] |= TEXBLENDARG_INV_ARG;
-	    break;
-	 default: 
-	    return pass_through( state, blendUnit );
-	 }
+         switch (combine->OperandA[i]) {
+         case GL_SRC_ALPHA:
+            args_A[i] |= 0;
+            break;
+         case GL_ONE_MINUS_SRC_ALPHA:
+            args_A[i] |= TEXBLENDARG_INV_ARG;
+            break;
+         default:
+            return pass_through(state, blendUnit);
+         }
       }
    }
 
@@ -363,86 +362,86 @@ i830SetTexEnvCombine(i830ContextPtr i830,
 
    used = 0;
    state[used++] = (_3DSTATE_MAP_BLEND_OP_CMD(blendUnit) |
-		    TEXPIPE_COLOR |
-		    ENABLE_TEXOUTPUT_WRT_SEL |
-		    TEXOP_OUTPUT_CURRENT |
-		    DISABLE_TEX_CNTRL_STAGE |
-		    TEXOP_MODIFY_PARMS |
-		    blendop);
+                    TEXPIPE_COLOR |
+                    ENABLE_TEXOUTPUT_WRT_SEL |
+                    TEXOP_OUTPUT_CURRENT |
+                    DISABLE_TEX_CNTRL_STAGE | TEXOP_MODIFY_PARMS | blendop);
    state[used++] = (_3DSTATE_MAP_BLEND_OP_CMD(blendUnit) |
-		    TEXPIPE_ALPHA |
-		    ENABLE_TEXOUTPUT_WRT_SEL |
-		    TEXOP_OUTPUT_CURRENT |
-		    TEXOP_MODIFY_PARMS |
-		    ablendop);
+                    TEXPIPE_ALPHA |
+                    ENABLE_TEXOUTPUT_WRT_SEL |
+                    TEXOP_OUTPUT_CURRENT | TEXOP_MODIFY_PARMS | ablendop);
 
-   for ( i = 0 ; i < numColorArgs ; i++ ) {
+   for (i = 0; i < numColorArgs; i++) {
       state[used++] = (_3DSTATE_MAP_BLEND_ARG_CMD(blendUnit) |
-		       tex_blend_rgb[i] | args_RGB[i]);
+                       tex_blend_rgb[i] | args_RGB[i]);
    }
 
-   for ( i = 0 ; i < numAlphaArgs ; i++ ) {
+   for (i = 0; i < numAlphaArgs; i++) {
       state[used++] = (_3DSTATE_MAP_BLEND_ARG_CMD(blendUnit) |
-		       tex_blend_a[i] | args_A[i]);
+                       tex_blend_a[i] | args_A[i]);
    }
 
 
-   if (need_factor) 
-      return emit_factor( blendUnit, state, used, factor );
-   else 
+   if (need_factor)
+      return emit_factor(blendUnit, state, used, factor);
+   else
       return used;
 }
 
 
-static void emit_texblend( i830ContextPtr i830, GLuint unit, GLuint blendUnit,
-			   GLboolean last_stage )
+static void
+emit_texblend(struct i830_context *i830, GLuint unit, GLuint blendUnit,
+              GLboolean last_stage)
 {
    struct gl_texture_unit *texUnit = &i830->intel.ctx.Texture.Unit[unit];
    GLuint tmp[I830_TEXBLEND_SIZE], tmp_sz;
 
 
-   if (0) fprintf(stderr, "%s unit %d\n", __FUNCTION__, unit);
+   if (0)
+      fprintf(stderr, "%s unit %d\n", __FUNCTION__, unit);
 
    /* Update i830->state.TexBlend
-    */ 
-   tmp_sz = i830SetTexEnvCombine(i830, texUnit->_CurrentCombine, blendUnit, 
-				 GetTexelOp(unit), tmp,
-				 texUnit->EnvColor );
+    */
+   tmp_sz = i830SetTexEnvCombine(i830, texUnit->_CurrentCombine, blendUnit,
+                                 GetTexelOp(unit), tmp, texUnit->EnvColor);
 
-   if (last_stage) 
+   if (last_stage)
       tmp[0] |= TEXOP_LAST_STAGE;
 
    if (tmp_sz != i830->state.TexBlendWordsUsed[blendUnit] ||
-       memcmp( tmp, i830->state.TexBlend[blendUnit], tmp_sz * sizeof(GLuint))) {
-      
-      I830_STATECHANGE( i830, I830_UPLOAD_TEXBLEND(blendUnit) );
-      memcpy( i830->state.TexBlend[blendUnit], tmp, tmp_sz * sizeof(GLuint));
+       memcmp(tmp, i830->state.TexBlend[blendUnit],
+              tmp_sz * sizeof(GLuint))) {
+
+      I830_STATECHANGE(i830, I830_UPLOAD_TEXBLEND(blendUnit));
+      memcpy(i830->state.TexBlend[blendUnit], tmp, tmp_sz * sizeof(GLuint));
       i830->state.TexBlendWordsUsed[blendUnit] = tmp_sz;
    }
 
    I830_ACTIVESTATE(i830, I830_UPLOAD_TEXBLEND(blendUnit), GL_TRUE);
 }
 
-static void emit_passthrough( i830ContextPtr i830 )
+static void
+emit_passthrough(struct i830_context *i830)
 {
    GLuint tmp[I830_TEXBLEND_SIZE], tmp_sz;
    GLuint unit = 0;
 
-   tmp_sz = pass_through( tmp, unit );
+   tmp_sz = pass_through(tmp, unit);
    tmp[0] |= TEXOP_LAST_STAGE;
 
    if (tmp_sz != i830->state.TexBlendWordsUsed[unit] ||
-       memcmp( tmp, i830->state.TexBlend[unit], tmp_sz * sizeof(GLuint))) {
-      
-      I830_STATECHANGE( i830, I830_UPLOAD_TEXBLEND(unit) );
-      memcpy( i830->state.TexBlend[unit], tmp, tmp_sz * sizeof(GLuint));
+       memcmp(tmp, i830->state.TexBlend[unit], tmp_sz * sizeof(GLuint))) {
+
+      I830_STATECHANGE(i830, I830_UPLOAD_TEXBLEND(unit));
+      memcpy(i830->state.TexBlend[unit], tmp, tmp_sz * sizeof(GLuint));
       i830->state.TexBlendWordsUsed[unit] = tmp_sz;
    }
 
    I830_ACTIVESTATE(i830, I830_UPLOAD_TEXBLEND(unit), GL_TRUE);
 }
 
-void i830EmitTextureBlend( i830ContextPtr i830 )
+void
+i830EmitTextureBlend(struct i830_context *i830)
 {
    GLcontext *ctx = &i830->intel.ctx;
    GLuint unit, last_stage = 0, blendunit = 0;
@@ -450,16 +449,15 @@ void i830EmitTextureBlend( i830ContextPtr i830 )
    I830_ACTIVESTATE(i830, I830_UPLOAD_TEXBLEND_ALL, GL_FALSE);
 
    if (ctx->Texture._EnabledUnits) {
-      for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
-	 if (ctx->Texture.Unit[unit]._ReallyEnabled) 
-	    last_stage = unit;
+      for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++)
+         if (ctx->Texture.Unit[unit]._ReallyEnabled)
+            last_stage = unit;
 
-      for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
-	 if (ctx->Texture.Unit[unit]._ReallyEnabled) 
-	    emit_texblend( i830, unit, blendunit++, last_stage == unit );
+      for (unit = 0; unit < ctx->Const.MaxTextureUnits; unit++)
+         if (ctx->Texture.Unit[unit]._ReallyEnabled)
+            emit_texblend(i830, unit, blendunit++, last_stage == unit);
    }
    else {
-      emit_passthrough( i830 );
+      emit_passthrough(i830);
    }
 }
-

src/mesa/drivers/dri/i915/i830_texstate.c

@@ -38,446 +38,279 @@
 #include "intel_screen.h"
 #include "intel_ioctl.h"
 #include "intel_tex.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
 
 #include "i830_context.h"
 #include "i830_reg.h"
 
-static const GLint initial_offsets[6][2] = { {0,0},
-				       {0,2},
-				       {1,0},
-				       {1,2},
-				       {1,1},
-				       {1,3} };
 
-static const GLint step_offsets[6][2] = { {0,2},
-				    {0,2},
-				    {-1,2},
-				    {-1,2},
-				    {-1,1},
-				    {-1,1} };
 
-#define I830_TEX_UNIT_ENABLED(unit)		(1<<unit)
-
-static GLboolean i830SetTexImages( i830ContextPtr i830, 
-				  struct gl_texture_object *tObj )
+static GLuint
+translate_texture_format(GLuint mesa_format)
 {
-   GLuint total_height, pitch, i, textureFormat;
-   i830TextureObjectPtr t = (i830TextureObjectPtr) tObj->DriverData;
-   const struct gl_texture_image *baseImage = tObj->Image[0][tObj->BaseLevel];
-   GLint firstLevel, lastLevel, numLevels;
-
-   switch( baseImage->TexFormat->MesaFormat ) {
+   switch (mesa_format) {
    case MESA_FORMAT_L8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_L8;
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_L8;
    case MESA_FORMAT_I8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_I8;
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_I8;
    case MESA_FORMAT_A8:
-      t->intel.texelBytes = 1;
-      textureFormat = MAPSURF_8BIT | MT_8BIT_I8; /* Kludge -- check with conform, glean */
-      break;
-
+      return MAPSURF_8BIT | MT_8BIT_I8; /* Kludge! */
    case MESA_FORMAT_AL88:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_AY88;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_AY88;
    case MESA_FORMAT_RGB565:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_RGB565;
    case MESA_FORMAT_ARGB1555:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB1555;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_ARGB1555;
    case MESA_FORMAT_ARGB4444:
-      t->intel.texelBytes = 2;
-      textureFormat = MAPSURF_16BIT | MT_16BIT_ARGB4444;
-      break;
-
+      return MAPSURF_16BIT | MT_16BIT_ARGB4444;
    case MESA_FORMAT_ARGB8888:
-      t->intel.texelBytes = 4;
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-      break;
-
+      return MAPSURF_32BIT | MT_32BIT_ARGB8888;
    case MESA_FORMAT_YCBCR_REV:
-      t->intel.texelBytes = 2;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_NORMAL | 
-		       TM0S1_COLORSPACE_CONVERSION);
-      break;
-
+      return (MAPSURF_422 | MT_422_YCRCB_NORMAL);
    case MESA_FORMAT_YCBCR:
-      t->intel.texelBytes = 2;
-      textureFormat = (MAPSURF_422 | MT_422_YCRCB_SWAPY | /* ??? */
-		       TM0S1_COLORSPACE_CONVERSION);
-      break;
-
+      return (MAPSURF_422 | MT_422_YCRCB_SWAPY);
    case MESA_FORMAT_RGB_FXT1:
    case MESA_FORMAT_RGBA_FXT1:
-     t->intel.texelBytes = 2;
-     textureFormat = MAPSURF_COMPRESSED | MT_COMPRESS_FXT1;
-     break;
-
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_FXT1);
    case MESA_FORMAT_RGBA_DXT1:
    case MESA_FORMAT_RGB_DXT1:
-     /* 
-      * DXTn pitches are Width/4 * blocksize in bytes 
-      * for DXT1: blocksize=8 so Width/4*8 = Width * 2 
-      * for DXT3/5: blocksize=16 so Width/4*16 = Width * 4
-      */
-     t->intel.texelBytes = 2;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1);
-     break;
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT1);
    case MESA_FORMAT_RGBA_DXT3:
-     t->intel.texelBytes = 4;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3);
-     break;
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT2_3);
    case MESA_FORMAT_RGBA_DXT5:
-     t->intel.texelBytes = 4;
-     textureFormat = (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
-     break;
-
+      return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
    default:
-      fprintf(stderr, "%s: bad image format\n", __FUNCTION__);
+      fprintf(stderr, "%s: bad image format %x\n", __FUNCTION__, mesa_format);
       abort();
+      return 0;
    }
-
-   /* Compute which mipmap levels we really want to send to the hardware.
-    * This depends on the base image size, GL_TEXTURE_MIN_LOD,
-    * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
-    * Yes, this looks overly complicated, but it's all needed.
-    */
-   driCalculateTextureFirstLastLevel( (driTextureObject *) t );
+}
 
 
-   /* Figure out the amount of memory required to hold all the mipmap
-    * levels.  Choose the smallest pitch to accomodate the largest
-    * mipmap:
-    */
-   firstLevel = t->intel.base.firstLevel;
-   lastLevel = t->intel.base.lastLevel;
-   numLevels = lastLevel - firstLevel + 1;
 
 
-   /* All images must be loaded at this pitch.  Count the number of
-    * lines required:
-    */
-   switch (tObj->Target) {
-   case GL_TEXTURE_CUBE_MAP: {
-      const GLuint dim = tObj->Image[0][firstLevel]->Width;
-      GLuint face;
-
-      pitch = dim * t->intel.texelBytes;
-      pitch *= 2;		/* double pitch for cube layouts */
-      pitch = (pitch + 3) & ~3;
-      
-      total_height = dim * 4;
-
-      for ( face = 0 ; face < 6 ; face++) {
-	 GLuint x = initial_offsets[face][0] * dim;
-	 GLuint y = initial_offsets[face][1] * dim;
-	 GLuint d = dim;
-	 
-	 t->intel.base.dirty_images[face] = ~0;
-
-	 assert(tObj->Image[face][firstLevel]->Width == dim);
-	 assert(tObj->Image[face][firstLevel]->Height == dim);
-
-	 for (i = 0; i < numLevels; i++) {
-	    t->intel.image[face][i].image = tObj->Image[face][firstLevel + i];
-	    if (!t->intel.image[face][i].image) {
-	       fprintf(stderr, "no image %d %d\n", face, i);
-	       break;		/* can't happen */
-	    }
-	 
-	    t->intel.image[face][i].offset = 
-	       y * pitch + x * t->intel.texelBytes;
-	    t->intel.image[face][i].internalFormat = baseImage->_BaseFormat;
-
-	    d >>= 1;
-	    x += step_offsets[face][0] * d;
-	    y += step_offsets[face][1] * d;
-	 }
-      }
-      break;
-   }
+/* The i915 (and related graphics cores) do not support GL_CLAMP.  The
+ * Intel drivers for "other operating systems" implement GL_CLAMP as
+ * GL_CLAMP_TO_EDGE, so the same is done here.
+ */
+static GLuint
+translate_wrap_mode(GLenum wrap)
+{
+   switch (wrap) {
+   case GL_REPEAT:
+      return TEXCOORDMODE_WRAP;
+   case GL_CLAMP:
+   case GL_CLAMP_TO_EDGE:
+      return TEXCOORDMODE_CLAMP;        /* not really correct */
+   case GL_CLAMP_TO_BORDER:
+      return TEXCOORDMODE_CLAMP_BORDER;
+   case GL_MIRRORED_REPEAT:
+      return TEXCOORDMODE_MIRROR;
    default:
-      pitch = tObj->Image[0][firstLevel]->Width * t->intel.texelBytes;
-      pitch = (pitch + 3) & ~3;
-      t->intel.base.dirty_images[0] = ~0;
-
-      for ( total_height = i = 0 ; i < numLevels ; i++ ) {
-	 t->intel.image[0][i].image = tObj->Image[0][firstLevel + i];
-	 if (!t->intel.image[0][i].image) 
-	    break;
-	 
-	 t->intel.image[0][i].offset = total_height * pitch;
-	 t->intel.image[0][i].internalFormat = baseImage->_BaseFormat;
-	 if (t->intel.image[0][i].image->IsCompressed)
-	 {
-	   if (t->intel.image[0][i].image->Height > 4)
-	     total_height += t->intel.image[0][i].image->Height/4;
-	   else
-	     total_height += 1;
-	 }
-	 else
-	   total_height += MAX2(2, t->intel.image[0][i].image->Height);
-      }
-      break;
+      return TEXCOORDMODE_WRAP;
    }
-
-   t->intel.Pitch = pitch;
-   t->intel.base.totalSize = total_height*pitch;
-   t->intel.max_level = i-1;
-   t->Setup[I830_TEXREG_TM0S1] = 
-      (((tObj->Image[0][firstLevel]->Height - 1) << TM0S1_HEIGHT_SHIFT) |
-       ((tObj->Image[0][firstLevel]->Width - 1) << TM0S1_WIDTH_SHIFT) |
-       textureFormat);
-   t->Setup[I830_TEXREG_TM0S2] = 
-      (((pitch / 4) - 1) << TM0S2_PITCH_SHIFT) |
-      TM0S2_CUBE_FACE_ENA_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MAX_MIP_MASK;
-   t->Setup[I830_TEXREG_TM0S3] &= ~TM0S3_MIN_MIP_MASK;
-   t->Setup[I830_TEXREG_TM0S3] |= ((numLevels - 1)*4) << TM0S3_MIN_MIP_SHIFT;
-   t->intel.dirty = I830_UPLOAD_TEX_ALL;
-
-   return intelUploadTexImages( &i830->intel, &t->intel, 0 );
 }
 
 
-static void i830_import_tex_unit( i830ContextPtr i830, 
-			   i830TextureObjectPtr t,
-			   GLuint unit )
+/* Recalculate all state from scratch.  Perhaps not the most
+ * efficient, but this has gotten complex enough that we need
+ * something which is understandable and reliable.
+ */
+static GLboolean
+i830_update_tex_unit(struct intel_context *intel, GLuint unit, GLuint ss3)
 {
-   if(INTEL_DEBUG&DEBUG_TEXTURE)
-      fprintf(stderr, "%s unit(%d)\n", __FUNCTION__, unit);
-   
-   if (i830->intel.CurrentTexObj[unit]) 
-      i830->intel.CurrentTexObj[unit]->base.bound &= ~(1U << unit);
+   GLcontext *ctx = &intel->ctx;
+   struct i830_context *i830 = i830_context(ctx);
+   struct gl_texture_object *tObj = ctx->Texture.Unit[unit]._Current;
+   struct intel_texture_object *intelObj = intel_texture_object(tObj);
+   struct gl_texture_image *firstImage;
+   GLuint *state = i830->state.Tex[unit];
 
-   i830->intel.CurrentTexObj[unit] = (intelTextureObjectPtr)t;
-   t->intel.base.bound |= (1 << unit);
+   memset(state, 0, sizeof(state));
 
-   I830_STATECHANGE( i830, I830_UPLOAD_TEX(unit) );
-
-   i830->state.Tex[unit][I830_TEXREG_TM0LI] = (_3DSTATE_LOAD_STATE_IMMEDIATE_2 | 
-					       (LOAD_TEXTURE_MAP0 << unit) | 4);
-   i830->state.Tex[unit][I830_TEXREG_TM0S0] = (TM0S0_USE_FENCE |
-					       t->intel.TextureOffset);
-
-   i830->state.Tex[unit][I830_TEXREG_TM0S1] = t->Setup[I830_TEXREG_TM0S1];
-   i830->state.Tex[unit][I830_TEXREG_TM0S2] = t->Setup[I830_TEXREG_TM0S2];
-
-   i830->state.Tex[unit][I830_TEXREG_TM0S3] &= TM0S3_LOD_BIAS_MASK;
-   i830->state.Tex[unit][I830_TEXREG_TM0S3] |= (t->Setup[I830_TEXREG_TM0S3] &
-						~TM0S3_LOD_BIAS_MASK);
-
-   i830->state.Tex[unit][I830_TEXREG_TM0S4] = t->Setup[I830_TEXREG_TM0S4];
-   i830->state.Tex[unit][I830_TEXREG_MCS] = (t->Setup[I830_TEXREG_MCS] & 
-					     ~MAP_UNIT_MASK);   
-   i830->state.Tex[unit][I830_TEXREG_CUBE] = t->Setup[I830_TEXREG_CUBE];
-   i830->state.Tex[unit][I830_TEXREG_MCS] |= MAP_UNIT(unit);
-
-   t->intel.dirty &= ~I830_UPLOAD_TEX(unit);
-}
-
-
-
-static GLboolean enable_tex_common( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i830TextureObjectPtr t = (i830TextureObjectPtr)tObj->DriverData;
-
-   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* Fallback if there's a texture border */
-   if ( tObj->Image[0][tObj->BaseLevel]->Border > 0 ) {
-      fprintf(stderr, "Texture border\n");
+   if (!intel_finalize_mipmap_tree(intel, unit))
       return GL_FALSE;
+
+   /* Get first image here, since intelObj->firstLevel will get set in
+    * the intel_finalize_mipmap_tree() call above.
+    */
+   firstImage = tObj->Image[0][intelObj->firstLevel];
+
+   i830->state.tex_buffer[unit] = intelObj->mt->region->buffer;
+   i830->state.tex_offset[unit] = intel_miptree_image_offset(intelObj->mt, 0,
+                                                             intelObj->
+                                                             firstLevel);
+
+
+   state[I830_TEXREG_TM0LI] = (_3DSTATE_LOAD_STATE_IMMEDIATE_2 |
+                               (LOAD_TEXTURE_MAP0 << unit) | 4);
+
+/*    state[I830_TEXREG_TM0S0] = (TM0S0_USE_FENCE | */
+/* 			       t->intel.TextureOffset); */
+
+
+   state[I830_TEXREG_TM0S1] =
+      (((firstImage->Height - 1) << TM0S1_HEIGHT_SHIFT) |
+       ((firstImage->Width - 1) << TM0S1_WIDTH_SHIFT) |
+       translate_texture_format(firstImage->TexFormat->MesaFormat));
+
+   state[I830_TEXREG_TM0S2] =
+      (((((intelObj->mt->pitch * intelObj->mt->cpp) / 4) -
+         1) << TM0S2_PITCH_SHIFT) | TM0S2_CUBE_FACE_ENA_MASK);
+
+   {
+      if (tObj->Target == GL_TEXTURE_CUBE_MAP)
+         state[I830_TEXREG_CUBE] = (_3DSTATE_MAP_CUBE | MAP_UNIT(unit) |
+                                    CUBE_NEGX_ENABLE |
+                                    CUBE_POSX_ENABLE |
+                                    CUBE_NEGY_ENABLE |
+                                    CUBE_POSY_ENABLE |
+                                    CUBE_NEGZ_ENABLE | CUBE_POSZ_ENABLE);
+      else
+         state[I830_TEXREG_CUBE] = (_3DSTATE_MAP_CUBE | MAP_UNIT(unit));
    }
 
-   /* Upload teximages (not pipelined)
-    */
-   if (t->intel.base.dirty_images[0]) {
-      if (!i830SetTexImages( i830, tObj )) {
-	 return GL_FALSE;
+
+
+
+   {
+      GLuint minFilt, mipFilt, magFilt;
+
+      switch (tObj->MinFilter) {
+      case GL_NEAREST:
+         minFilt = FILTER_NEAREST;
+         mipFilt = MIPFILTER_NONE;
+         break;
+      case GL_LINEAR:
+         minFilt = FILTER_LINEAR;
+         mipFilt = MIPFILTER_NONE;
+         break;
+      case GL_NEAREST_MIPMAP_NEAREST:
+         minFilt = FILTER_NEAREST;
+         mipFilt = MIPFILTER_NEAREST;
+         break;
+      case GL_LINEAR_MIPMAP_NEAREST:
+         minFilt = FILTER_LINEAR;
+         mipFilt = MIPFILTER_NEAREST;
+         break;
+      case GL_NEAREST_MIPMAP_LINEAR:
+         minFilt = FILTER_NEAREST;
+         mipFilt = MIPFILTER_LINEAR;
+         break;
+      case GL_LINEAR_MIPMAP_LINEAR:
+         minFilt = FILTER_LINEAR;
+         mipFilt = MIPFILTER_LINEAR;
+         break;
+      default:
+         return GL_FALSE;
       }
+
+      if (tObj->MaxAnisotropy > 1.0) {
+         minFilt = FILTER_ANISOTROPIC;
+         magFilt = FILTER_ANISOTROPIC;
+      }
+      else {
+         switch (tObj->MagFilter) {
+         case GL_NEAREST:
+            magFilt = FILTER_NEAREST;
+            break;
+         case GL_LINEAR:
+            magFilt = FILTER_LINEAR;
+            break;
+         default:
+            return GL_FALSE;
+         }
+      }
+
+      state[I830_TEXREG_TM0S3] = i830->lodbias_tm0s3[unit];
+
+#if 0
+      /* YUV conversion:
+       */
+      if (firstImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR ||
+          firstImage->TexFormat->MesaFormat == MESA_FORMAT_YCBCR_REV)
+         state[I830_TEXREG_TM0S3] |= SS2_COLORSPACE_CONVERSION;
+#endif
+
+      state[I830_TEXREG_TM0S3] |= ((intelObj->lastLevel -
+                                    intelObj->firstLevel) *
+                                   4) << TM0S3_MIN_MIP_SHIFT;
+
+      state[I830_TEXREG_TM0S3] |= ((minFilt << TM0S3_MIN_FILTER_SHIFT) |
+                                   (mipFilt << TM0S3_MIP_FILTER_SHIFT) |
+                                   (magFilt << TM0S3_MAG_FILTER_SHIFT));
    }
 
-   /* Update state if this is a different texture object to last
-    * time.
-    */
-   if (i830->intel.CurrentTexObj[unit] != &t->intel || 
-       (t->intel.dirty & I830_UPLOAD_TEX(unit))) {
-      i830_import_tex_unit( i830, t, unit);
+   {
+      GLenum ws = tObj->WrapS;
+      GLenum wt = tObj->WrapT;
+
+
+      /* 3D textures not available on i830
+       */
+      if (tObj->Target == GL_TEXTURE_3D)
+         return GL_FALSE;
+
+      state[I830_TEXREG_MCS] = (_3DSTATE_MAP_COORD_SET_CMD |
+                                MAP_UNIT(unit) |
+                                ENABLE_TEXCOORD_PARAMS |
+                                ss3 |
+                                ENABLE_ADDR_V_CNTL |
+                                TEXCOORD_ADDR_V_MODE(translate_wrap_mode(wt))
+                                | ENABLE_ADDR_U_CNTL |
+                                TEXCOORD_ADDR_U_MODE(translate_wrap_mode
+                                                     (ws)));
    }
 
+
+   state[I830_TEXREG_TM0S4] = INTEL_PACKCOLOR8888(tObj->_BorderChan[0],
+                                                  tObj->_BorderChan[1],
+                                                  tObj->_BorderChan[2],
+                                                  tObj->_BorderChan[3]);
+
+
    I830_ACTIVESTATE(i830, I830_UPLOAD_TEX(unit), GL_TRUE);
-
-   return GL_TRUE;
-}
-
-static GLboolean enable_tex_rect( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   GLuint mcs = i830->state.Tex[unit][I830_TEXREG_MCS];
-
-   mcs &= ~TEXCOORDS_ARE_NORMAL;
-   mcs |= TEXCOORDS_ARE_IN_TEXELUNITS;
-
-   if ((mcs != i830->state.Tex[unit][I830_TEXREG_MCS])
-       || (0 != i830->state.Tex[unit][I830_TEXREG_CUBE])) {
-      I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_MCS] = mcs;
-      i830->state.Tex[unit][I830_TEXREG_CUBE] = 0;
-   }
-
-   return GL_TRUE;
-}
-
-
-static GLboolean enable_tex_2d( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   GLuint mcs = i830->state.Tex[unit][I830_TEXREG_MCS];
-
-   mcs &= ~TEXCOORDS_ARE_IN_TEXELUNITS;
-   mcs |= TEXCOORDS_ARE_NORMAL;
-
-   if ((mcs != i830->state.Tex[unit][I830_TEXREG_MCS])
-       || (0 != i830->state.Tex[unit][I830_TEXREG_CUBE])) {
-      I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_MCS] = mcs;
-      i830->state.Tex[unit][I830_TEXREG_CUBE] = 0;
-   }
-
-   return GL_TRUE;
-}
-
- 
-static GLboolean enable_tex_cube( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-   struct gl_texture_object *tObj = texUnit->_Current;
-   i830TextureObjectPtr t = (i830TextureObjectPtr)tObj->DriverData;
-   GLuint mcs = i830->state.Tex[unit][I830_TEXREG_MCS];
-   const GLuint cube = CUBE_NEGX_ENABLE | CUBE_POSX_ENABLE
-     | CUBE_NEGY_ENABLE | CUBE_POSY_ENABLE
-     | CUBE_NEGZ_ENABLE | CUBE_POSZ_ENABLE;
-   GLuint face;
-
-   mcs &= ~TEXCOORDS_ARE_IN_TEXELUNITS;
-   mcs |= TEXCOORDS_ARE_NORMAL;
-
-   if ((mcs != i830->state.Tex[unit][I830_TEXREG_MCS])
-       || (cube != i830->state.Tex[unit][I830_TEXREG_CUBE])) {
-      I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
-      i830->state.Tex[unit][I830_TEXREG_MCS] = mcs;
-      i830->state.Tex[unit][I830_TEXREG_CUBE] = cube;
-   }
-
-   /* Upload teximages (not pipelined)
+   /* memcmp was already disabled, but definitely won't work as the
+    * region might now change and that wouldn't be detected:
     */
-   if ( t->intel.base.dirty_images[0] || t->intel.base.dirty_images[1] ||
-        t->intel.base.dirty_images[2] || t->intel.base.dirty_images[3] ||
-        t->intel.base.dirty_images[4] || t->intel.base.dirty_images[5] ) {
-      i830SetTexImages( i830, tObj );
-   }
+   I830_STATECHANGE(i830, I830_UPLOAD_TEX(unit));
+   return GL_TRUE;
+}
 
-   /* upload (per face) */
-   for (face = 0; face < 6; face++) {
-      if (t->intel.base.dirty_images[face]) {
-	 if (!intelUploadTexImages( &i830->intel, &t->intel, face )) {
-	    return GL_FALSE;
-	 }
+
+
+
+void
+i830UpdateTextureState(struct intel_context *intel)
+{
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   GLboolean ok = GL_TRUE;
+   GLuint i;
+
+   for (i = 0; i < I830_TEX_UNITS && ok; i++) {
+      switch (intel->ctx.Texture.Unit[i]._ReallyEnabled) {
+      case TEXTURE_1D_BIT:
+      case TEXTURE_2D_BIT:
+      case TEXTURE_CUBE_BIT:
+         ok = i830_update_tex_unit(intel, i, TEXCOORDS_ARE_NORMAL);
+         break;
+      case TEXTURE_RECT_BIT:
+         ok = i830_update_tex_unit(intel, i, TEXCOORDS_ARE_IN_TEXELUNITS);
+         break;
+      case 0:
+         if (i830->state.active & I830_UPLOAD_TEX(i))
+            I830_ACTIVESTATE(i830, I830_UPLOAD_TEX(i), GL_FALSE);
+         break;
+      case TEXTURE_3D_BIT:
+      default:
+         ok = GL_FALSE;
+         break;
       }
    }
 
-
-   return GL_TRUE;
-}
-
-
-static GLboolean disable_tex( GLcontext *ctx, GLuint unit )
-{
-   i830ContextPtr i830 = I830_CONTEXT(ctx);
-
-   /* This is happening too often.  I need to conditionally send diffuse
-    * state to the card.  Perhaps a diffuse dirty flag of some kind.
-    * Will need to change this logic if more than 2 texture units are
-    * used.  We need to only do this up to the last unit enabled, or unit
-    * one if nothing is enabled.
-    */
-
-   if ( i830->intel.CurrentTexObj[unit] != NULL ) {
-      /* The old texture is no longer bound to this texture unit.
-       * Mark it as such.
-       */
-
-      i830->intel.CurrentTexObj[unit]->base.bound &= ~(1U << 0);
-      i830->intel.CurrentTexObj[unit] = NULL;
-   }
-
-   return GL_TRUE;
-}
-
-static GLboolean i830UpdateTexUnit( GLcontext *ctx, GLuint unit )
-{
-   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
-
-   if (texUnit->_ReallyEnabled &&
-       INTEL_CONTEXT(ctx)->intelScreen->tex.size < 2048 * 1024)
-      return GL_FALSE;
-
-   switch(texUnit->_ReallyEnabled) {
-   case TEXTURE_1D_BIT:
-   case TEXTURE_2D_BIT:
-      return (enable_tex_common( ctx, unit ) &&
-	      enable_tex_2d( ctx, unit ));
-   case TEXTURE_RECT_BIT:
-      return (enable_tex_common( ctx, unit ) &&
-	      enable_tex_rect( ctx, unit ));
-   case TEXTURE_CUBE_BIT:
-      return (enable_tex_common( ctx, unit ) &&
-	      enable_tex_cube( ctx, unit ));
-   case 0:
-      return disable_tex( ctx, unit );
-   default:
-      return GL_FALSE;
-   }
-}
-
-
-void i830UpdateTextureState( intelContextPtr intel )
-{
-   i830ContextPtr i830 = I830_CONTEXT(intel);
-   GLcontext *ctx = &intel->ctx;
-   GLboolean ok;
-
-   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
-
-   I830_ACTIVESTATE(i830, I830_UPLOAD_TEX_ALL, GL_FALSE);
-
-   ok = (i830UpdateTexUnit( ctx, 0 ) &&
-	 i830UpdateTexUnit( ctx, 1 ) &&
-	 i830UpdateTexUnit( ctx, 2 ) &&
-	 i830UpdateTexUnit( ctx, 3 ));
-
-   FALLBACK( intel, I830_FALLBACK_TEXTURE, !ok );
+   FALLBACK(intel, I830_FALLBACK_TEXTURE, !ok);
 
    if (ok)
-      i830EmitTextureBlend( i830 );
+      i830EmitTextureBlend(i830);
 }
-
-
-

src/mesa/drivers/dri/i915/i830_vtbl.c

@@ -28,14 +28,15 @@
 
 #include "i830_context.h"
 #include "i830_reg.h"
-
 #include "intel_batchbuffer.h"
-
+#include "intel_regions.h"
 #include "tnl/t_context.h"
 #include "tnl/t_vertex.h"
 
-static GLboolean i830_check_vertex_size( intelContextPtr intel,
-					 GLuint expected );
+#define FILE_DEBUG_FLAG DEBUG_STATE
+
+static GLboolean i830_check_vertex_size(struct intel_context *intel,
+                                        GLuint expected);
 
 #define SZ_TO_HW(sz)  ((sz-2)&0x3)
 #define EMIT_SZ(sz)   (EMIT_1F + (sz) - 1)
@@ -59,10 +60,11 @@ do {									\
 #define VRTX_TEX_SET_FMT(n, x)          ((x)<<((n)*2))
 #define TEXBIND_SET(n, x) 		((x)<<((n)*4))
 
-static void i830_render_start( intelContextPtr intel )
+static void
+i830_render_start(struct intel_context *intel)
 {
    GLcontext *ctx = &intel->ctx;
-   i830ContextPtr i830 = I830_CONTEXT(intel);
+   struct i830_context *i830 = i830_context(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    DECLARE_RENDERINPUTS(index_bitset);
@@ -70,7 +72,7 @@ static void i830_render_start( intelContextPtr intel )
    GLuint v2 = _3DSTATE_VFT1_CMD;
    GLuint mcsb1 = 0;
 
-   RENDERINPUTS_COPY( index_bitset, tnl->render_inputs_bitset );
+   RENDERINPUTS_COPY(index_bitset, tnl->render_inputs_bitset);
 
    /* Important:
     */
@@ -80,196 +82,215 @@ static void i830_render_start( intelContextPtr intel )
    /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
     * build up a hardware vertex.
     */
-   if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, VFT0_XYZW );
+   if (RENDERINPUTS_TEST_RANGE(index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX)) {
+      EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, VFT0_XYZW);
       intel->coloroffset = 4;
    }
    else {
-      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, VFT0_XYZ );
+      EMIT_ATTR(_TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, VFT0_XYZ);
       intel->coloroffset = 3;
    }
 
-   if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POINTSIZE )) {
-      EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F, VFT0_POINT_WIDTH );
+   if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_POINTSIZE)) {
+      EMIT_ATTR(_TNL_ATTRIB_POINTSIZE, EMIT_1F, VFT0_POINT_WIDTH);
    }
 
-   EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, VFT0_DIFFUSE );
-      
+   EMIT_ATTR(_TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, VFT0_DIFFUSE);
+
    intel->specoffset = 0;
-   if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 ) ||
-       RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_FOG )) {
-      if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 )) {
+   if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_COLOR1) ||
+       RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_FOG)) {
+      if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_COLOR1)) {
          intel->specoffset = intel->coloroffset + 1;
-         EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, VFT0_SPEC );
+         EMIT_ATTR(_TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, VFT0_SPEC);
       }
       else
-         EMIT_PAD( 3 );
+         EMIT_PAD(3);
 
-      if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_FOG ))
-         EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, VFT0_SPEC );
+      if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_FOG))
+         EMIT_ATTR(_TNL_ATTRIB_FOG, EMIT_1UB_1F, VFT0_SPEC);
       else
-         EMIT_PAD( 1 );
+         EMIT_PAD(1);
    }
 
-   if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
+   if (RENDERINPUTS_TEST_RANGE(index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX)) {
       int i, count = 0;
 
       for (i = 0; i < I830_TEX_UNITS; i++) {
-         if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
+         if (RENDERINPUTS_TEST(index_bitset, _TNL_ATTRIB_TEX(i))) {
             GLuint sz = VB->TexCoordPtr[i]->size;
             GLuint emit;
-            GLuint mcs = (i830->state.Tex[i][I830_TEXREG_MCS] & 
+            GLuint mcs = (i830->state.Tex[i][I830_TEXREG_MCS] &
                           ~TEXCOORDTYPE_MASK);
 
-	    switch (sz) {
-	    case 1: 
-	    case 2: 
-	       emit = EMIT_2F; 
-	       sz = 2; 
-	       mcs |= TEXCOORDTYPE_CARTESIAN; 
-	       break;
-	    case 3:
-	       emit = EMIT_3F; 
-	       sz = 3;
-	       mcs |= TEXCOORDTYPE_VECTOR;
-	       break;
-	    case 4: 
-	       emit = EMIT_3F_XYW; 
-	       sz = 3;     
-	       mcs |= TEXCOORDTYPE_HOMOGENEOUS;
-	       break;
-	    default: 
-	       continue;
-	    };
-	      
+            switch (sz) {
+            case 1:
+            case 2:
+               emit = EMIT_2F;
+               sz = 2;
+               mcs |= TEXCOORDTYPE_CARTESIAN;
+               break;
+            case 3:
+               emit = EMIT_3F;
+               sz = 3;
+               mcs |= TEXCOORDTYPE_VECTOR;
+               break;
+            case 4:
+               emit = EMIT_3F_XYW;
+               sz = 3;
+               mcs |= TEXCOORDTYPE_HOMOGENEOUS;
+               break;
+            default:
+               continue;
+            };
 
-	    EMIT_ATTR( _TNL_ATTRIB_TEX0+i, emit, 0 );	       
-	    v2 |= VRTX_TEX_SET_FMT(count, SZ_TO_HW(sz));
-	    mcsb1 |= (count+8)<<(i*4);
 
-	    if (mcs != i830->state.Tex[i][I830_TEXREG_MCS]) {
-	       I830_STATECHANGE(i830, I830_UPLOAD_TEX(i));
-	       i830->state.Tex[i][I830_TEXREG_MCS] = mcs;
-	    }
+            EMIT_ATTR(_TNL_ATTRIB_TEX0 + i, emit, 0);
+            v2 |= VRTX_TEX_SET_FMT(count, SZ_TO_HW(sz));
+            mcsb1 |= (count + 8) << (i * 4);
 
-	    count++;
-	 }
+            if (mcs != i830->state.Tex[i][I830_TEXREG_MCS]) {
+               I830_STATECHANGE(i830, I830_UPLOAD_TEX(i));
+               i830->state.Tex[i][I830_TEXREG_MCS] = mcs;
+            }
+
+            count++;
+         }
       }
 
       v0 |= VFT0_TEX_COUNT(count);
    }
-   
+
    /* Only need to change the vertex emit code if there has been a
     * statechange to a new hardware vertex format:
     */
    if (v0 != i830->state.Ctx[I830_CTXREG_VF] ||
        v2 != i830->state.Ctx[I830_CTXREG_VF2] ||
        mcsb1 != i830->state.Ctx[I830_CTXREG_MCSB1] ||
-       !RENDERINPUTS_EQUAL( index_bitset, i830->last_index_bitset )) {
-    
-      I830_STATECHANGE( i830, I830_UPLOAD_CTX );
+       !RENDERINPUTS_EQUAL(index_bitset, i830->last_index_bitset)) {
+      int k;
+
+      I830_STATECHANGE(i830, I830_UPLOAD_CTX);
 
       /* Must do this *after* statechange, so as not to affect
        * buffered vertices reliant on the old state:
        */
-      intel->vertex_size = 
-	 _tnl_install_attrs( ctx, 
-			     intel->vertex_attrs, 
-			     intel->vertex_attr_count,
-			     intel->ViewportMatrix.m, 0 );
+      intel->vertex_size =
+         _tnl_install_attrs(ctx,
+                            intel->vertex_attrs,
+                            intel->vertex_attr_count,
+                            intel->ViewportMatrix.m, 0);
 
       intel->vertex_size >>= 2;
 
       i830->state.Ctx[I830_CTXREG_VF] = v0;
       i830->state.Ctx[I830_CTXREG_VF2] = v2;
       i830->state.Ctx[I830_CTXREG_MCSB1] = mcsb1;
-      RENDERINPUTS_COPY( i830->last_index_bitset, index_bitset );
+      RENDERINPUTS_COPY(i830->last_index_bitset, index_bitset);
 
-      assert(i830_check_vertex_size( intel, intel->vertex_size ));
+      k = i830_check_vertex_size(intel, intel->vertex_size);
+      assert(k);
    }
 }
 
-static void i830_reduced_primitive_state( intelContextPtr intel,
-					  GLenum rprim )
+static void
+i830_reduced_primitive_state(struct intel_context *intel, GLenum rprim)
 {
-    i830ContextPtr i830 = I830_CONTEXT(intel);
-    GLuint st1 = i830->state.Stipple[I830_STPREG_ST1];
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   GLuint st1 = i830->state.Stipple[I830_STPREG_ST1];
 
-    st1 &= ~ST1_ENABLE;
+   st1 &= ~ST1_ENABLE;
 
-    switch (rprim) {
-    case GL_TRIANGLES:
-       if (intel->ctx.Polygon.StippleFlag &&
-	   intel->hw_stipple)
-	  st1 |= ST1_ENABLE;
-       break;
-    case GL_LINES:
-    case GL_POINTS:
-    default:
-       break;
-    }
+   switch (rprim) {
+   case GL_TRIANGLES:
+      if (intel->ctx.Polygon.StippleFlag && intel->hw_stipple)
+         st1 |= ST1_ENABLE;
+      break;
+   case GL_LINES:
+   case GL_POINTS:
+   default:
+      break;
+   }
 
-    i830->intel.reduced_primitive = rprim;
+   i830->intel.reduced_primitive = rprim;
 
-    if (st1 != i830->state.Stipple[I830_STPREG_ST1]) {
-       I830_STATECHANGE(i830, I830_UPLOAD_STIPPLE);
-       i830->state.Stipple[I830_STPREG_ST1] = st1;
-    }
+   if (st1 != i830->state.Stipple[I830_STPREG_ST1]) {
+      INTEL_FIREVERTICES(intel);
+
+      I830_STATECHANGE(i830, I830_UPLOAD_STIPPLE);
+      i830->state.Stipple[I830_STPREG_ST1] = st1;
+   }
 }
 
 /* Pull apart the vertex format registers and figure out how large a
  * vertex is supposed to be. 
  */
-static GLboolean i830_check_vertex_size( intelContextPtr intel,
-					 GLuint expected )
+static GLboolean
+i830_check_vertex_size(struct intel_context *intel, GLuint expected)
 {
-   i830ContextPtr i830 = I830_CONTEXT(intel);
+   struct i830_context *i830 = i830_context(&intel->ctx);
    int vft0 = i830->current->Ctx[I830_CTXREG_VF];
    int vft1 = i830->current->Ctx[I830_CTXREG_VF2];
    int nrtex = (vft0 & VFT0_TEX_COUNT_MASK) >> VFT0_TEX_COUNT_SHIFT;
    int i, sz = 0;
 
    switch (vft0 & VFT0_XYZW_MASK) {
-   case VFT0_XY: sz = 2; break;
-   case VFT0_XYZ: sz = 3; break;
-   case VFT0_XYW: sz = 3; break;
-   case VFT0_XYZW: sz = 4; break;
-   default: 
+   case VFT0_XY:
+      sz = 2;
+      break;
+   case VFT0_XYZ:
+      sz = 3;
+      break;
+   case VFT0_XYW:
+      sz = 3;
+      break;
+   case VFT0_XYZW:
+      sz = 4;
+      break;
+   default:
       fprintf(stderr, "no xyzw specified\n");
       return 0;
    }
 
-   if (vft0 & VFT0_SPEC) sz++;
-   if (vft0 & VFT0_DIFFUSE) sz++;
-   if (vft0 & VFT0_DEPTH_OFFSET) sz++;
-   if (vft0 & VFT0_POINT_WIDTH) sz++;
-	
-   for (i = 0 ; i < nrtex ; i++) { 
+   if (vft0 & VFT0_SPEC)
+      sz++;
+   if (vft0 & VFT0_DIFFUSE)
+      sz++;
+   if (vft0 & VFT0_DEPTH_OFFSET)
+      sz++;
+   if (vft0 & VFT0_POINT_WIDTH)
+      sz++;
+
+   for (i = 0; i < nrtex; i++) {
       switch (vft1 & VFT1_TEX0_MASK) {
-      case TEXCOORDFMT_2D: sz += 2; break;
-      case TEXCOORDFMT_3D: sz += 3; break;
-      case TEXCOORDFMT_4D: sz += 4; break;
-      case TEXCOORDFMT_1D: sz += 1; break;
+      case TEXCOORDFMT_2D:
+         sz += 2;
+         break;
+      case TEXCOORDFMT_3D:
+         sz += 3;
+         break;
+      case TEXCOORDFMT_4D:
+         sz += 4;
+         break;
+      case TEXCOORDFMT_1D:
+         sz += 1;
+         break;
       }
       vft1 >>= VFT1_TEX1_SHIFT;
    }
-	
-   if (sz != expected) 
+
+   if (sz != expected)
       fprintf(stderr, "vertex size mismatch %d/%d\n", sz, expected);
-   
+
    return sz == expected;
 }
 
-static void i830_emit_invarient_state( intelContextPtr intel )
+static void
+i830_emit_invarient_state(struct intel_context *intel)
 {
    BATCH_LOCALS;
 
-   BEGIN_BATCH( 200 );
-
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(0));
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(1));
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(2));
-   OUT_BATCH(_3DSTATE_MAP_CUBE | MAP_UNIT(3));
+   BEGIN_BATCH(40, 0);
 
    OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
    OUT_BATCH(0);
@@ -282,37 +303,35 @@ static void i830_emit_invarient_state( intelContextPtr intel )
 
    OUT_BATCH(_3DSTATE_FOG_MODE_CMD);
    OUT_BATCH(FOGFUNC_ENABLE |
-	     FOG_LINEAR_CONST | 
-	     FOGSRC_INDEX_Z | 
-	     ENABLE_FOG_DENSITY);
+             FOG_LINEAR_CONST | FOGSRC_INDEX_Z | ENABLE_FOG_DENSITY);
    OUT_BATCH(0);
    OUT_BATCH(0);
 
 
    OUT_BATCH(_3DSTATE_MAP_TEX_STREAM_CMD |
-	     MAP_UNIT(0) |
-	     DISABLE_TEX_STREAM_BUMP |
-	     ENABLE_TEX_STREAM_COORD_SET |
-	     TEX_STREAM_COORD_SET(0) |
-	     ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(0));
+             MAP_UNIT(0) |
+             DISABLE_TEX_STREAM_BUMP |
+             ENABLE_TEX_STREAM_COORD_SET |
+             TEX_STREAM_COORD_SET(0) |
+             ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(0));
    OUT_BATCH(_3DSTATE_MAP_TEX_STREAM_CMD |
-	     MAP_UNIT(1) |
-	     DISABLE_TEX_STREAM_BUMP |
-	     ENABLE_TEX_STREAM_COORD_SET |
-	     TEX_STREAM_COORD_SET(1) |
-	     ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(1));
+             MAP_UNIT(1) |
+             DISABLE_TEX_STREAM_BUMP |
+             ENABLE_TEX_STREAM_COORD_SET |
+             TEX_STREAM_COORD_SET(1) |
+             ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(1));
    OUT_BATCH(_3DSTATE_MAP_TEX_STREAM_CMD |
-	     MAP_UNIT(2) |
-	     DISABLE_TEX_STREAM_BUMP |
-	     ENABLE_TEX_STREAM_COORD_SET |
-	     TEX_STREAM_COORD_SET(2) |
-	     ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(2));
+             MAP_UNIT(2) |
+             DISABLE_TEX_STREAM_BUMP |
+             ENABLE_TEX_STREAM_COORD_SET |
+             TEX_STREAM_COORD_SET(2) |
+             ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(2));
    OUT_BATCH(_3DSTATE_MAP_TEX_STREAM_CMD |
-	     MAP_UNIT(3) |
-	     DISABLE_TEX_STREAM_BUMP |
-	     ENABLE_TEX_STREAM_COORD_SET |
-	     TEX_STREAM_COORD_SET(3) |
-	     ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(3));
+             MAP_UNIT(3) |
+             DISABLE_TEX_STREAM_BUMP |
+             ENABLE_TEX_STREAM_COORD_SET |
+             TEX_STREAM_COORD_SET(3) |
+             ENABLE_TEX_STREAM_MAP_IDX | TEX_STREAM_MAP_IDX(3));
 
    OUT_BATCH(_3DSTATE_MAP_COORD_TRANSFORM);
    OUT_BATCH(DISABLE_TEX_TRANSFORM | TEXTURE_SET(0));
@@ -324,21 +343,13 @@ static void i830_emit_invarient_state( intelContextPtr intel )
    OUT_BATCH(DISABLE_TEX_TRANSFORM | TEXTURE_SET(3));
 
    OUT_BATCH(_3DSTATE_RASTER_RULES_CMD |
-	     ENABLE_POINT_RASTER_RULE |
-	     OGL_POINT_RASTER_RULE |
-	     ENABLE_LINE_STRIP_PROVOKE_VRTX |
-	     ENABLE_TRI_FAN_PROVOKE_VRTX |
-	     ENABLE_TRI_STRIP_PROVOKE_VRTX |
-	     LINE_STRIP_PROVOKE_VRTX(1) |
-	     TRI_FAN_PROVOKE_VRTX(2) | 
-	     TRI_STRIP_PROVOKE_VRTX(2));
-
-   OUT_BATCH(_3DSTATE_SCISSOR_ENABLE_CMD | 
-	     DISABLE_SCISSOR_RECT);
-
-   OUT_BATCH(_3DSTATE_SCISSOR_RECT_0_CMD);
-   OUT_BATCH(0);
-   OUT_BATCH(0);
+             ENABLE_POINT_RASTER_RULE |
+             OGL_POINT_RASTER_RULE |
+             ENABLE_LINE_STRIP_PROVOKE_VRTX |
+             ENABLE_TRI_FAN_PROVOKE_VRTX |
+             ENABLE_TRI_STRIP_PROVOKE_VRTX |
+             LINE_STRIP_PROVOKE_VRTX(1) |
+             TRI_FAN_PROVOKE_VRTX(2) | TRI_STRIP_PROVOKE_VRTX(2));
 
    OUT_BATCH(_3DSTATE_VERTEX_TRANSFORM);
    OUT_BATCH(DISABLE_VIEWPORT_TRANSFORM | DISABLE_PERSPECTIVE_DIVIDE);
@@ -349,7 +360,7 @@ static void i830_emit_invarient_state( intelContextPtr intel )
 
 
    OUT_BATCH(_3DSTATE_COLOR_FACTOR_CMD);
-   OUT_BATCH(0x80808080);	/* .5 required in alpha for GL_DOT3_RGBA_EXT */
+   OUT_BATCH(0x80808080);       /* .5 required in alpha for GL_DOT3_RGBA_EXT */
 
    ADVANCE_BATCH();
 }
@@ -358,33 +369,39 @@ static void i830_emit_invarient_state( intelContextPtr intel )
 #define emit( intel, state, size )			\
 do {							\
    int k;						\
-   BEGIN_BATCH( size / sizeof(GLuint));			\
-   for (k = 0 ; k < size / sizeof(GLuint) ; k++)	\
+   BEGIN_BATCH(size / sizeof(GLuint), 0);		\
+   for (k = 0 ; k < size / sizeof(GLuint) ; k++) {	\
+      if (0) _mesa_printf("  0x%08x\n", state[k]);		\
       OUT_BATCH(state[k]);				\
+   }							\
    ADVANCE_BATCH();					\
-} while (0);
+} while (0)
 
-static GLuint get_state_size( struct i830_hw_state *state )
+static GLuint
+get_state_size(struct i830_hw_state *state)
 {
    GLuint dirty = state->active & ~state->emitted;
    GLuint sz = 0;
    GLuint i;
 
-   if (dirty & I830_UPLOAD_CTX) 
+   if (dirty & I830_UPLOAD_INVARIENT)
+      sz += 40 * sizeof(int);
+
+   if (dirty & I830_UPLOAD_CTX)
       sz += sizeof(state->Ctx);
 
-   if (dirty & I830_UPLOAD_BUFFERS) 
+   if (dirty & I830_UPLOAD_BUFFERS)
       sz += sizeof(state->Buffer);
 
-   if (dirty & I830_UPLOAD_STIPPLE) 
+   if (dirty & I830_UPLOAD_STIPPLE)
       sz += sizeof(state->Stipple);
 
    for (i = 0; i < I830_TEX_UNITS; i++) {
-      if ((dirty & I830_UPLOAD_TEX(i)))  
-	 sz += sizeof(state->Tex[i]); 
+      if ((dirty & I830_UPLOAD_TEX(i)))
+         sz += sizeof(state->Tex[i]);
 
-      if (dirty & I830_UPLOAD_TEXBLEND(i)) 
-	 sz += state->TexBlendWordsUsed[i] * 4;
+      if (dirty & I830_UPLOAD_TEXBLEND(i))
+         sz += state->TexBlendWordsUsed[i] * 4;
    }
 
    return sz;
@@ -393,135 +410,197 @@ static GLuint get_state_size( struct i830_hw_state *state )
 
 /* Push the state into the sarea and/or texture memory.
  */
-static void i830_emit_state( intelContextPtr intel )
+static void
+i830_emit_state(struct intel_context *intel)
 {
-   i830ContextPtr i830 = I830_CONTEXT(intel);
+   struct i830_context *i830 = i830_context(&intel->ctx);
    struct i830_hw_state *state = i830->current;
    int i;
-   GLuint dirty = state->active & ~state->emitted;
-   GLuint counter = intel->batch.counter;
+   GLuint dirty;
    BATCH_LOCALS;
 
-   if (intel->batch.space < get_state_size(state)) {
-      intelFlushBatch(intel, GL_TRUE);
-      dirty = state->active & ~state->emitted;
-      counter = intel->batch.counter;
+   /* We don't hold the lock at this point, so want to make sure that
+    * there won't be a buffer wrap.  
+    *
+    * It might be better to talk about explicit places where
+    * scheduling is allowed, rather than assume that it is whenever a
+    * batchbuffer fills up.
+    */
+   intel_batchbuffer_require_space(intel->batch, get_state_size(state), 0);
+
+   /* Do this here as we may have flushed the batchbuffer above,
+    * causing more state to be dirty!
+    */
+   dirty = state->active & ~state->emitted;
+
+   if (dirty & I830_UPLOAD_INVARIENT) {
+      DBG("I830_UPLOAD_INVARIENT:\n");
+      i830_emit_invarient_state(intel);
    }
 
    if (dirty & I830_UPLOAD_CTX) {
-      if (VERBOSE) fprintf(stderr, "I830_UPLOAD_CTX:\n"); 
-      emit( i830, state->Ctx, sizeof(state->Ctx) );
+      DBG("I830_UPLOAD_CTX:\n");
+      emit(i830, state->Ctx, sizeof(state->Ctx));
+
    }
 
    if (dirty & I830_UPLOAD_BUFFERS) {
-      if (VERBOSE) fprintf(stderr, "I830_UPLOAD_BUFFERS:\n"); 
-      emit( i830, state->Buffer, sizeof(state->Buffer) );
+      DBG("I830_UPLOAD_BUFFERS:\n");
+      BEGIN_BATCH(I830_DEST_SETUP_SIZE + 2, 0);
+      OUT_BATCH(state->Buffer[I830_DESTREG_CBUFADDR0]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_CBUFADDR1]);
+      OUT_RELOC(state->draw_region->buffer,
+                DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
+
+      if (state->depth_region) {
+         OUT_BATCH(state->Buffer[I830_DESTREG_DBUFADDR0]);
+         OUT_BATCH(state->Buffer[I830_DESTREG_DBUFADDR1]);
+         OUT_RELOC(state->depth_region->buffer,
+                   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                   DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
+      }
+
+      OUT_BATCH(state->Buffer[I830_DESTREG_DV0]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_DV1]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SENABLE]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SR0]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SR1]);
+      OUT_BATCH(state->Buffer[I830_DESTREG_SR2]);
+      ADVANCE_BATCH();
    }
 
    if (dirty & I830_UPLOAD_STIPPLE) {
-      if (VERBOSE) fprintf(stderr, "I830_UPLOAD_STIPPLE:\n"); 
-      emit( i830, state->Stipple, sizeof(state->Stipple) );
+      DBG("I830_UPLOAD_STIPPLE:\n");
+      emit(i830, state->Stipple, sizeof(state->Stipple));
    }
 
    for (i = 0; i < I830_TEX_UNITS; i++) {
-      if ((dirty & I830_UPLOAD_TEX(i))) { 
- 	 if (VERBOSE) fprintf(stderr, "I830_UPLOAD_TEX(%d):\n", i); 
-	 emit( i830, state->Tex[i], sizeof(state->Tex[i])); 
-      } 
+      if ((dirty & I830_UPLOAD_TEX(i))) {
+         DBG("I830_UPLOAD_TEX(%d):\n", i);
+
+         BEGIN_BATCH(I830_TEX_SETUP_SIZE + 1, 0);
+         OUT_BATCH(state->Tex[i][I830_TEXREG_TM0LI]);
+
+         if (state->tex_buffer[i]) {
+            OUT_RELOC(state->tex_buffer[i],
+                      DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+                      DRM_BO_MASK_MEM | DRM_BO_FLAG_READ,
+                      state->tex_offset[i] | TM0S0_USE_FENCE);
+         }
+         else {
+            assert(i == 0);
+            assert(state == &i830->meta);
+            OUT_BATCH(0);
+         }
+
+         OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S1]);
+         OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S2]);
+         OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S3]);
+         OUT_BATCH(state->Tex[i][I830_TEXREG_TM0S4]);
+         OUT_BATCH(state->Tex[i][I830_TEXREG_MCS]);
+         OUT_BATCH(state->Tex[i][I830_TEXREG_CUBE]);
+      }
 
       if (dirty & I830_UPLOAD_TEXBLEND(i)) {
-	 if (VERBOSE) fprintf(stderr, "I830_UPLOAD_TEXBLEND(%d):\n", i); 
-	 emit( i830, state->TexBlend[i], 
-	       state->TexBlendWordsUsed[i] * 4 );
+         DBG("I830_UPLOAD_TEXBLEND(%d): %d words\n", i,
+             state->TexBlendWordsUsed[i]);
+         emit(i830, state->TexBlend[i], state->TexBlendWordsUsed[i] * 4);
       }
    }
 
    state->emitted |= dirty;
-   intel->batch.last_emit_state = counter;
-   assert(counter == intel->batch.counter);
 }
 
-static void i830_destroy_context( intelContextPtr intel )
+static void
+i830_destroy_context(struct intel_context *intel)
 {
    _tnl_free_vertices(&intel->ctx);
 }
 
 static void
-i830_set_color_region(intelContextPtr intel, const intelRegion *region)
+i830_set_draw_region(struct intel_context *intel,
+                     struct intel_region *draw_region,
+                     struct intel_region *depth_region)
 {
-   i830ContextPtr i830 = I830_CONTEXT(intel);
-   I830_STATECHANGE( i830, I830_UPLOAD_BUFFERS );
+   struct i830_context *i830 = i830_context(&intel->ctx);
+
+   intel_region_release(intel->intelScreen, &i830->state.draw_region);
+   intel_region_release(intel->intelScreen, &i830->state.depth_region);
+   intel_region_reference(&i830->state.draw_region, draw_region);
+   intel_region_reference(&i830->state.depth_region, depth_region);
+
+   /* XXX FBO: Need code from i915_set_draw_region() */
+
+   I830_STATECHANGE(i830, I830_UPLOAD_BUFFERS);
+   I830_STATECHANGE(i830, I830_UPLOAD_BUFFERS);
    i830->state.Buffer[I830_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i830->state.Buffer[I830_DESTREG_CBUFADDR2] = region->offset;
-}
-
-
-static void
-i830_set_z_region(intelContextPtr intel, const intelRegion *region)
-{
-   i830ContextPtr i830 = I830_CONTEXT(intel);
-   I830_STATECHANGE( i830, I830_UPLOAD_BUFFERS );
+      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(draw_region->pitch) |
+       BUF_3D_USE_FENCE);
    i830->state.Buffer[I830_DESTREG_DBUFADDR1] =
-      (BUF_3D_ID_DEPTH | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i830->state.Buffer[I830_DESTREG_DBUFADDR2] = region->offset;
+      (BUF_3D_ID_DEPTH | BUF_3D_PITCH(depth_region->pitch) |
+       BUF_3D_USE_FENCE);
 }
 
-
+#if 0
 static void
 i830_update_color_z_regions(intelContextPtr intel,
-                            const intelRegion *colorRegion,
-                            const intelRegion *depthRegion)
+                            const intelRegion * colorRegion,
+                            const intelRegion * depthRegion)
 {
    i830ContextPtr i830 = I830_CONTEXT(intel);
 
    i830->state.Buffer[I830_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(colorRegion->pitch) | BUF_3D_USE_FENCE);
+      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(colorRegion->pitch) |
+       BUF_3D_USE_FENCE);
    i830->state.Buffer[I830_DESTREG_CBUFADDR2] = colorRegion->offset;
 
    i830->state.Buffer[I830_DESTREG_DBUFADDR1] =
       (BUF_3D_ID_DEPTH | BUF_3D_PITCH(depthRegion->pitch) | BUF_3D_USE_FENCE);
    i830->state.Buffer[I830_DESTREG_DBUFADDR2] = depthRegion->offset;
 }
+#endif
 
 
 /* This isn't really handled at the moment.
  */
-static void i830_lost_hardware( intelContextPtr intel )
+static void
+i830_lost_hardware(struct intel_context *intel)
 {
-   I830_CONTEXT(intel)->state.emitted = 0;
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   i830->state.emitted = 0;
 }
 
 
 
-static void i830_emit_flush( intelContextPtr intel )
+static GLuint
+i830_flush_cmd(void)
 {
-   BATCH_LOCALS;
-
-   BEGIN_BATCH(2);
-   OUT_BATCH( MI_FLUSH | FLUSH_MAP_CACHE ); 
-   OUT_BATCH( 0 );
-   ADVANCE_BATCH();
+   return MI_FLUSH | FLUSH_MAP_CACHE;
 }
 
 
-
-
-void i830InitVtbl( i830ContextPtr i830 )
+static void 
+i830_assert_not_dirty( struct intel_context *intel )
+{
+   struct i830_context *i830 = i830_context(&intel->ctx);
+   struct i830_hw_state *state = i830->current;
+   GLuint dirty = state->active & ~state->emitted;
+   assert(!dirty);
+}
+
+
+void
+i830InitVtbl(struct i830_context *i830)
 {
-   i830->intel.vtbl.alloc_tex_obj = i830AllocTexObj;
    i830->intel.vtbl.check_vertex_size = i830_check_vertex_size;
-   i830->intel.vtbl.clear_with_tris = i830ClearWithTris;
-   i830->intel.vtbl.rotate_window = i830RotateWindow;
    i830->intel.vtbl.destroy = i830_destroy_context;
-   i830->intel.vtbl.emit_invarient_state = i830_emit_invarient_state;
    i830->intel.vtbl.emit_state = i830_emit_state;
    i830->intel.vtbl.lost_hardware = i830_lost_hardware;
    i830->intel.vtbl.reduced_primitive_state = i830_reduced_primitive_state;
-   i830->intel.vtbl.set_color_region = i830_set_color_region;
-   i830->intel.vtbl.set_z_region = i830_set_z_region;
-   i830->intel.vtbl.update_color_z_regions = i830_update_color_z_regions;
+   i830->intel.vtbl.set_draw_region = i830_set_draw_region;
    i830->intel.vtbl.update_texture_state = i830UpdateTextureState;
-   i830->intel.vtbl.emit_flush = i830_emit_flush;
+   i830->intel.vtbl.flush_cmd = i830_flush_cmd;
    i830->intel.vtbl.render_start = i830_render_start;
+   i830->intel.vtbl.assert_not_dirty = i830_assert_not_dirty;
 }

src/mesa/drivers/dri/i915/i915_context.c

@@ -41,77 +41,88 @@
 #include "utils.h"
 #include "i915_reg.h"
 
+#include "intel_regions.h"
+#include "intel_batchbuffer.h"
+
 /***************************************
  * Mesa's Driver Functions
  ***************************************/
 
-static const struct dri_extension i915_extensions[] =
-{
-    { "GL_ARB_depth_texture",              NULL },
-    { "GL_ARB_fragment_program",           NULL },
-    { "GL_ARB_shadow",                     NULL },
-    { "GL_EXT_shadow_funcs",               NULL },
-    /* ARB extn won't work if not enabled */
-    { "GL_SGIX_depth_texture",             NULL },
-    { NULL,                                NULL }
+static const struct dri_extension i915_extensions[] = {
+   {"GL_ARB_depth_texture", NULL},
+   {"GL_ARB_fragment_program", NULL},
+   {"GL_ARB_shadow", NULL},
+   {"GL_ARB_texture_env_crossbar", NULL},
+   {"GL_ARB_texture_non_power_of_two", NULL},
+   {"GL_EXT_shadow_funcs", NULL},
+   /* ARB extn won't work if not enabled */
+   {"GL_SGIX_depth_texture", NULL},
+   {NULL, NULL}
 };
 
 /* Override intel default.
  */
-static void i915InvalidateState( GLcontext *ctx, GLuint new_state )
+static void
+i915InvalidateState(GLcontext * ctx, GLuint new_state)
 {
-   _swrast_InvalidateState( ctx, new_state );
-   _swsetup_InvalidateState( ctx, new_state );
-   _ac_InvalidateState( ctx, new_state );
-   _tnl_InvalidateState( ctx, new_state );
-   _tnl_invalidate_vertex_state( ctx, new_state );
-   INTEL_CONTEXT(ctx)->NewGLState |= new_state;
+   _swrast_InvalidateState(ctx, new_state);
+   _swsetup_InvalidateState(ctx, new_state);
+   _ac_InvalidateState(ctx, new_state);
+   _tnl_InvalidateState(ctx, new_state);
+   _tnl_invalidate_vertex_state(ctx, new_state);
+   intel_context(ctx)->NewGLState |= new_state;
 
    /* Todo: gather state values under which tracked parameters become
     * invalidated, add callbacks for things like
     * ProgramLocalParameters, etc.
     */
    {
-      struct i915_fragment_program *p = 
-	 (struct i915_fragment_program *)ctx->FragmentProgram._Current;
+      struct i915_fragment_program *p =
+         (struct i915_fragment_program *) ctx->FragmentProgram._Current;
       if (p && p->nr_params)
-	 p->params_uptodate = 0;
+         p->params_uptodate = 0;
    }
 
-   if (new_state & (_NEW_FOG|_NEW_HINT|_NEW_PROGRAM))
+   if (new_state & (_NEW_FOG | _NEW_HINT | _NEW_PROGRAM))
       i915_update_fog(ctx);
 }
 
 
-static void i915InitDriverFunctions( struct dd_function_table *functions )
+static void
+i915InitDriverFunctions(struct dd_function_table *functions)
 {
-   intelInitDriverFunctions( functions );
-   i915InitStateFunctions( functions );
-   i915InitTextureFuncs( functions );
-   i915InitFragProgFuncs( functions );
+   intelInitDriverFunctions(functions);
+   i915InitStateFunctions(functions);
+   i915InitTextureFuncs(functions);
+   i915InitFragProgFuncs(functions);
    functions->UpdateState = i915InvalidateState;
 }
 
 
 
-GLboolean i915CreateContext( const __GLcontextModes *mesaVis,
-			    __DRIcontextPrivate *driContextPriv,
-			    void *sharedContextPrivate)
+GLboolean
+i915CreateContext(const __GLcontextModes * mesaVis,
+                  __DRIcontextPrivate * driContextPriv,
+                  void *sharedContextPrivate)
 {
    struct dd_function_table functions;
-   i915ContextPtr i915 = (i915ContextPtr) CALLOC_STRUCT(i915_context);
-   intelContextPtr intel = &i915->intel;
+   struct i915_context *i915 =
+      (struct i915_context *) CALLOC_STRUCT(i915_context);
+   struct intel_context *intel = &i915->intel;
    GLcontext *ctx = &intel->ctx;
-   GLuint i;
 
-   if (!i915) return GL_FALSE;
+   if (!i915)
+      return GL_FALSE;
 
-   i915InitVtbl( i915 );
+   _mesa_printf("\ntexmem-0-3 branch\n\n");
 
-   i915InitDriverFunctions( &functions );
+   i915InitVtbl(i915);
+   i915InitMetaFuncs(i915);
 
-   if (!intelInitContext( intel, mesaVis, driContextPriv,
-			  sharedContextPrivate, &functions )) {
+   i915InitDriverFunctions(&functions);
+
+   if (!intelInitContext(intel, mesaVis, driContextPriv,
+                         sharedContextPrivate, &functions)) {
       FREE(i915);
       return GL_FALSE;
    }
@@ -120,63 +131,44 @@ GLboolean i915CreateContext( const __GLcontextModes *mesaVis,
    ctx->Const.MaxTextureImageUnits = I915_TEX_UNITS;
    ctx->Const.MaxTextureCoordUnits = I915_TEX_UNITS;
 
-   intel->nr_heaps = 1;
-   intel->texture_heaps[0] = 
-      driCreateTextureHeap( 0, intel,
-			    intel->intelScreen->tex.size,
-			    12,
-			    I830_NR_TEX_REGIONS,
-			    intel->sarea->texList,
-			    (unsigned *) & intel->sarea->texAge,
-			    & intel->swapped,
-			    sizeof( struct i915_texture_object ),
-			    (destroy_texture_object_t *)intelDestroyTexObj );
 
-   /* FIXME: driCalculateMaxTextureLevels assumes that mipmaps are
-    * tightly packed, but they're not in Intel graphics
-    * hardware.
+   /* Advertise the full hardware capabilities.  The new memory
+    * manager should cope much better with overload situations:
     */
+   ctx->Const.MaxTextureLevels = 12;
+   ctx->Const.Max3DTextureLevels = 9;
+   ctx->Const.MaxCubeTextureLevels = 12;
+   ctx->Const.MaxTextureRectSize = (1 << 11);
    ctx->Const.MaxTextureUnits = I915_TEX_UNITS;
-   i = driQueryOptioni( &intel->intelScreen->optionCache, "allow_large_textures");
-   driCalculateMaxTextureLevels( intel->texture_heaps,
-				 intel->nr_heaps,
-				 &intel->ctx.Const,
-				 4,
-				 11, /* max 2D texture size is 2048x2048 */
-				 8,  /* 3D texture */
-				 11, /* cube texture. */
-				 11, /* rect texture */
-				 12,
-				 GL_FALSE,
-				 i );
 
    /* GL_ARB_fragment_program limits - don't think Mesa actually
     * validates programs against these, and in any case one ARB
     * instruction can translate to more than one HW instruction, so
     * we'll still have to check and fallback each time.
     */
-   
    ctx->Const.FragmentProgram.MaxNativeTemps = I915_MAX_TEMPORARY;
-   ctx->Const.FragmentProgram.MaxNativeAttribs = 11; /* 8 tex, 2 color, fog */
+   ctx->Const.FragmentProgram.MaxNativeAttribs = 11;    /* 8 tex, 2 color, fog */
    ctx->Const.FragmentProgram.MaxNativeParameters = I915_MAX_CONSTANT;
    ctx->Const.FragmentProgram.MaxNativeAluInstructions = I915_MAX_ALU_INSN;
    ctx->Const.FragmentProgram.MaxNativeTexInstructions = I915_MAX_TEX_INSN;
-   ctx->Const.FragmentProgram.MaxNativeInstructions = (I915_MAX_ALU_INSN + 
-						I915_MAX_TEX_INSN);
-   ctx->Const.FragmentProgram.MaxNativeTexIndirections = I915_MAX_TEX_INDIRECT;
+   ctx->Const.FragmentProgram.MaxNativeInstructions = (I915_MAX_ALU_INSN +
+                                                       I915_MAX_TEX_INSN);
+   ctx->Const.FragmentProgram.MaxNativeTexIndirections =
+      I915_MAX_TEX_INDIRECT;
    ctx->Const.FragmentProgram.MaxNativeAddressRegs = 0; /* I don't think we have one */
 
+   ctx->_MaintainTexEnvProgram = 1;
+   ctx->_UseTexEnvProgram = 1;
 
-   driInitExtensions( ctx, i915_extensions, GL_FALSE );
+   driInitExtensions(ctx, i915_extensions, GL_FALSE);
 
 
-   _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 
-		       36 * sizeof(GLfloat) );
+   _tnl_init_vertices(ctx, ctx->Const.MaxArrayLockSize + 12,
+                      36 * sizeof(GLfloat));
 
    intel->verts = TNL_CONTEXT(ctx)->clipspace.vertex_buf;
 
-   i915InitState( i915 );
+   i915InitState(i915);
 
    return GL_TRUE;
 }
-

src/mesa/drivers/dri/i915/i915_context.h

@@ -45,6 +45,8 @@
 #define I915_UPLOAD_PROGRAM          0x8
 #define I915_UPLOAD_CONSTANTS        0x10
 #define I915_UPLOAD_FOG              0x20
+#define I915_UPLOAD_INVARIENT        0x40
+#define I915_UPLOAD_DEFAULTS         0x80
 #define I915_UPLOAD_TEX(i)           (0x00010000<<(i))
 #define I915_UPLOAD_TEX_ALL          (0x00ff0000)
 #define I915_UPLOAD_TEX_0_SHIFT      16
@@ -54,10 +56,8 @@
  */
 #define I915_DESTREG_CBUFADDR0 0
 #define I915_DESTREG_CBUFADDR1 1
-#define I915_DESTREG_CBUFADDR2 2
 #define I915_DESTREG_DBUFADDR0 3
 #define I915_DESTREG_DBUFADDR1 4
-#define I915_DESTREG_DBUFADDR2 5
 #define I915_DESTREG_DV0 6
 #define I915_DESTREG_DV1 7
 #define I915_DESTREG_SENABLE 8
@@ -88,7 +88,6 @@
 #define I915_STPREG_ST1        1
 #define I915_STP_SETUP_SIZE    2
 
-#define I915_TEXREG_MS2        0
 #define I915_TEXREG_MS3        1
 #define I915_TEXREG_MS4        2
 #define I915_TEXREG_SS2        3
@@ -96,6 +95,15 @@
 #define I915_TEXREG_SS4        5
 #define I915_TEX_SETUP_SIZE    6
 
+#define I915_DEFREG_C0    0
+#define I915_DEFREG_C1    1
+#define I915_DEFREG_S0    2
+#define I915_DEFREG_S1    3
+#define I915_DEFREG_Z0    4
+#define I915_DEFREG_Z1    5
+#define I915_DEF_SETUP_SIZE    6
+
+
 #define I915_MAX_CONSTANT      32
 #define I915_CONSTANT_SIZE     (2+(4*I915_MAX_CONSTANT))
 
@@ -106,13 +114,14 @@
 /* Hardware version of a parsed fragment program.  "Derived" from the
  * mesa fragment_program struct.
  */
-struct i915_fragment_program {
+struct i915_fragment_program
+{
    struct gl_fragment_program FragProg;
 
    GLboolean translated;
    GLboolean params_uptodate;
    GLboolean on_hardware;
-   GLboolean error;		/* If program is malformed for any reason. */
+   GLboolean error;             /* If program is malformed for any reason. */
 
    GLuint nr_tex_indirect;
    GLuint nr_tex_insn;
@@ -134,22 +143,22 @@ struct i915_fragment_program {
    GLuint constant_flags[I915_MAX_CONSTANT];
    GLuint nr_constants;
 
-   GLuint *csr;			/* Cursor, points into program.
-				 */
+   GLuint *csr;                 /* Cursor, points into program.
+                                 */
 
-   GLuint *decl;		/* Cursor, points into declarations.
-				 */
-   
-   GLuint decl_s;		/* flags for which s regs need to be decl'd */
-   GLuint decl_t;		/* flags for which t regs need to be decl'd */
+   GLuint *decl;                /* Cursor, points into declarations.
+                                 */
 
-   GLuint temp_flag;		/* Tracks temporary regs which are in
-				 * use.
-				 */
+   GLuint decl_s;               /* flags for which s regs need to be decl'd */
+   GLuint decl_t;               /* flags for which t regs need to be decl'd */
 
-   GLuint utemp_flag;		/* Tracks TYPE_U temporary regs which are in
-				 * use.
-				 */
+   GLuint temp_flag;            /* Tracks temporary regs which are in
+                                 * use.
+                                 */
+
+   GLuint utemp_flag;           /* Tracks TYPE_U temporary regs which are in
+                                 * use.
+                                 */
 
 
 
@@ -158,26 +167,25 @@ struct i915_fragment_program {
    GLuint wpos_tex;
    GLboolean depth_written;
 
-   struct { 
-      GLuint reg;		/* Hardware constant idx */
-      const GLfloat *values; 	/* Pointer to tracked values */
+   struct
+   {
+      GLuint reg;               /* Hardware constant idx */
+      const GLfloat *values;    /* Pointer to tracked values */
    } param[I915_MAX_CONSTANT];
    GLuint nr_params;
-      
-
 
 
    /* Helpers for i915_texprog.c:
     */
-   GLuint src_texture;		/* Reg containing sampled texture color,
-				 * else UREG_BAD.
-				 */
+   GLuint src_texture;          /* Reg containing sampled texture color,
+                                 * else UREG_BAD.
+                                 */
 
-   GLuint src_previous;		/* Reg containing color from previous 
-				 * stage.  May need to be decl'd.
-				 */
+   GLuint src_previous;         /* Reg containing color from previous 
+                                 * stage.  May need to be decl'd.
+                                 */
 
-   GLuint last_tex_stage;	/* Number of last enabled texture unit */
+   GLuint last_tex_stage;       /* Number of last enabled texture unit */
 
    struct vertex_buffer *VB;
 };
@@ -187,41 +195,53 @@ struct i915_fragment_program {
 
 
 
-struct i915_texture_object
-{
-   struct intel_texture_object intel;
-   GLenum lastTarget;
-   GLboolean refs_border_color;
-   GLuint Setup[I915_TEX_SETUP_SIZE];
-};
 
 #define I915_TEX_UNITS 8
 
 
-struct i915_hw_state {
+struct i915_hw_state
+{
    GLuint Ctx[I915_CTX_SETUP_SIZE];
    GLuint Buffer[I915_DEST_SETUP_SIZE];
    GLuint Stipple[I915_STP_SETUP_SIZE];
    GLuint Fog[I915_FOG_SETUP_SIZE];
+   GLuint Defaults[I915_DEF_SETUP_SIZE];
    GLuint Tex[I915_TEX_UNITS][I915_TEX_SETUP_SIZE];
    GLuint Constant[I915_CONSTANT_SIZE];
    GLuint ConstantSize;
    GLuint Program[I915_PROGRAM_SIZE];
    GLuint ProgramSize;
-   GLuint active;		/* I915_UPLOAD_* */
-   GLuint emitted;		/* I915_UPLOAD_* */
+
+   /* Region pointers for relocation: 
+    */
+   struct intel_region *draw_region;
+   struct intel_region *depth_region;
+/*    struct intel_region *tex_region[I915_TEX_UNITS]; */
+
+   /* Regions aren't actually that appropriate here as the memory may
+    * be from a PBO or FBO.  Just use the buffer id.  Will have to do
+    * this for draw and depth for FBO's...
+    */
+   struct _DriBufferObject *tex_buffer[I915_TEX_UNITS];
+   GLuint tex_offset[I915_TEX_UNITS];
+
+
+   GLuint active;               /* I915_UPLOAD_* */
+   GLuint emitted;              /* I915_UPLOAD_* */
 };
 
 #define I915_FOG_PIXEL  2
 #define I915_FOG_VERTEX 1
 #define I915_FOG_NONE   0
 
-struct i915_context 
+struct i915_context
 {
    struct intel_context intel;
 
    GLuint last_ReallyEnabled;
    GLuint vertex_fog;
+   GLuint lodbias_ss2[MAX_TEXTURE_UNITS];
+
 
    struct i915_fragment_program tex_program;
    struct i915_fragment_program *current_program;
@@ -230,24 +250,14 @@ struct i915_context
 };
 
 
-typedef struct i915_context *i915ContextPtr;
-typedef struct i915_texture_object *i915TextureObjectPtr;
-
-#define I915_CONTEXT(ctx)	((i915ContextPtr)(ctx))
-
-
-
 #define I915_STATECHANGE(i915, flag)					\
 do {									\
-   if (0) fprintf(stderr, "I915_STATECHANGE %x in %s\n", flag, __FUNCTION__);	\
    INTEL_FIREVERTICES( &(i915)->intel );					\
    (i915)->state.emitted &= ~(flag);					\
 } while (0)
 
 #define I915_ACTIVESTATE(i915, flag, mode)			\
 do {								\
-   if (0) fprintf(stderr, "I915_ACTIVESTATE %x %d in %s\n",	\
-		  flag, mode, __FUNCTION__);			\
    INTEL_FIREVERTICES( &(i915)->intel );				\
    if (mode)							\
       (i915)->state.active |= (flag);				\
@@ -259,7 +269,13 @@ do {								\
 /*======================================================================
  * i915_vtbl.c
  */
-extern void i915InitVtbl( i915ContextPtr i915 );
+extern void i915InitVtbl(struct i915_context *i915);
+
+extern void
+i915_state_draw_region(struct intel_context *intel,
+                       struct i915_hw_state *state,
+                       struct intel_region *color_region,
+                       struct intel_region *depth_region);
 
 
 
@@ -288,70 +304,64 @@ do {									\
 /*======================================================================
  * i915_context.c
  */
-extern GLboolean i915CreateContext( const __GLcontextModes *mesaVis,
-				    __DRIcontextPrivate *driContextPriv,
-				    void *sharedContextPrivate);
+extern GLboolean i915CreateContext(const __GLcontextModes * mesaVis,
+                                   __DRIcontextPrivate * driContextPriv,
+                                   void *sharedContextPrivate);
 
 
 /*======================================================================
  * i915_texprog.c
  */
-extern void i915ValidateTextureProgram( i915ContextPtr i915 );
+extern void i915ValidateTextureProgram(struct i915_context *i915);
 
 
 /*======================================================================
  * i915_debug.c
  */
-extern void i915_disassemble_program( const GLuint *program, GLuint sz );
-extern void i915_print_ureg( const char *msg, GLuint ureg );
+extern void i915_disassemble_program(const GLuint * program, GLuint sz);
+extern void i915_print_ureg(const char *msg, GLuint ureg);
 
 
 /*======================================================================
  * i915_state.c
  */
-extern void i915InitStateFunctions( struct dd_function_table *functions );
-extern void i915InitState( i915ContextPtr i915 );
-extern void i915_update_fog(GLcontext *ctxx);
+extern void i915InitStateFunctions(struct dd_function_table *functions);
+extern void i915InitState(struct i915_context *i915);
+extern void i915_update_fog(GLcontext * ctx);
 
 
 /*======================================================================
  * i915_tex.c
  */
-extern void i915UpdateTextureState( intelContextPtr intel );
-extern void i915InitTextureFuncs( struct dd_function_table *functions );
-extern intelTextureObjectPtr i915AllocTexObj( struct gl_texture_object *texObj );
+extern void i915UpdateTextureState(struct intel_context *intel);
+extern void i915InitTextureFuncs(struct dd_function_table *functions);
 
 /*======================================================================
  * i915_metaops.c
  */
-extern GLboolean
-i915TryTextureReadPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *pack,
-			  GLvoid *pixels );
+void i915InitMetaFuncs(struct i915_context *i915);
 
-extern GLboolean
-i915TryTextureDrawPixels( GLcontext *ctx,
-			  GLint x, GLint y, GLsizei width, GLsizei height,
-			  GLenum format, GLenum type,
-			  const struct gl_pixelstore_attrib *unpack,
-			  const GLvoid *pixels );
-
-extern void 
-i915ClearWithTris( intelContextPtr intel, GLbitfield mask,
-		   GLboolean all, GLint cx, GLint cy, GLint cw, GLint ch);
-
-
-extern void
-i915RotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
-                 GLuint srcBuf);
 
 /*======================================================================
  * i915_fragprog.c
  */
-extern void i915ValidateFragmentProgram( i915ContextPtr i915 );
-extern void i915InitFragProgFuncs( struct dd_function_table *functions );
-	
-#endif
+extern void i915ValidateFragmentProgram(struct i915_context *i915);
+extern void i915InitFragProgFuncs(struct dd_function_table *functions);
 
+/*======================================================================
+ * Inline conversion functions.  These are better-typed than the
+ * macros used previously:
+ */
+static INLINE struct i915_context *
+i915_context(GLcontext * ctx)
+{
+   return (struct i915_context *) ctx;
+}
+
+
+
+#define I915_CONTEXT(ctx)	i915_context(ctx)
+
+
+
+#endif

src/mesa/drivers/dri/i915/i915_debug.c

@@ -67,27 +67,27 @@ static const char *opcodes[0x20] = {
 
 
 static const int args[0x20] = {
-   0,				/* 0 nop */
-   2,				/* 1 add */
-   1,				/* 2 mov */
-   2,				/* 3 m ul */
-   3, 				/* 4 mad */
-   3,				/* 5 dp2add */
-   2,				/* 6 dp3 */
-   2,				/* 7 dp4 */
-   1,				/* 8 frc */
-   1,				/* 9 rcp */
-   1,				/* a rsq */
-   1,				/* b exp */
-   1,				/* c log */
-   3,				/* d cmp */
-   2,				/* e min */
-   2,				/* f max */
-   1,				/* 10 flr */
-   1,				/* 11 mod */
-   1,				/* 12 trc */
-   2,				/* 13 sge */
-   2,				/* 14 slt */
+   0,                           /* 0 nop */
+   2,                           /* 1 add */
+   1,                           /* 2 mov */
+   2,                           /* 3 m ul */
+   3,                           /* 4 mad */
+   3,                           /* 5 dp2add */
+   2,                           /* 6 dp3 */
+   2,                           /* 7 dp4 */
+   1,                           /* 8 frc */
+   1,                           /* 9 rcp */
+   1,                           /* a rsq */
+   1,                           /* b exp */
+   1,                           /* c log */
+   3,                           /* d cmp */
+   2,                           /* e min */
+   2,                           /* f max */
+   1,                           /* 10 flr */
+   1,                           /* 11 mod */
+   1,                           /* 12 trc */
+   2,                           /* 13 sge */
+   2,                           /* 14 slt */
    1,
    1,
    1,
@@ -113,26 +113,35 @@ static const char *regname[0x8] = {
    "UNKNOWN",
 };
 
-static void print_reg_type_nr( GLuint type, GLuint nr )
+static void
+print_reg_type_nr(GLuint type, GLuint nr)
 {
    switch (type) {
    case REG_TYPE_T:
       switch (nr) {
-      case T_DIFFUSE: fprintf(stderr, "T_DIFFUSE"); return;
-      case T_SPECULAR: fprintf(stderr, "T_SPECULAR"); return;
-      case T_FOG_W: fprintf(stderr, "T_FOG_W"); return;
-      default: fprintf(stderr, "T_TEX%d", nr); return;
+      case T_DIFFUSE:
+         fprintf(stderr, "T_DIFFUSE");
+         return;
+      case T_SPECULAR:
+         fprintf(stderr, "T_SPECULAR");
+         return;
+      case T_FOG_W:
+         fprintf(stderr, "T_FOG_W");
+         return;
+      default:
+         fprintf(stderr, "T_TEX%d", nr);
+         return;
       }
    case REG_TYPE_OC:
       if (nr == 0) {
-	 fprintf(stderr, "oC");
-	 return;
+         fprintf(stderr, "oC");
+         return;
       }
       break;
    case REG_TYPE_OD:
       if (nr == 0) {
-	 fprintf(stderr, "oD");
-	 return;
+         fprintf(stderr, "oD");
+         return;
       }
       break;
    default:
@@ -151,7 +160,8 @@ static void print_reg_type_nr( GLuint type, GLuint nr )
 		      (SRC_W << A2_SRC2_CHANNEL_W_SHIFT))
 
 
-static void print_reg_neg_swizzle( GLuint reg )
+static void
+print_reg_neg_swizzle(GLuint reg)
 {
    int i;
 
@@ -161,50 +171,71 @@ static void print_reg_neg_swizzle( GLuint reg )
 
    fprintf(stderr, ".");
 
-   for (i = 3 ; i >= 0; i--) {
-      if (reg & (1<<((i*4)+3))) 
-	 fprintf(stderr, "-");
-	 
-      switch ((reg>>(i*4)) & 0x7) {
-      case 0: fprintf(stderr, "x"); break;
-      case 1: fprintf(stderr, "y"); break;
-      case 2: fprintf(stderr, "z"); break;
-      case 3: fprintf(stderr, "w"); break;
-      case 4: fprintf(stderr, "0"); break;
-      case 5: fprintf(stderr, "1"); break;
-      default: fprintf(stderr, "?"); break;
+   for (i = 3; i >= 0; i--) {
+      if (reg & (1 << ((i * 4) + 3)))
+         fprintf(stderr, "-");
+
+      switch ((reg >> (i * 4)) & 0x7) {
+      case 0:
+         fprintf(stderr, "x");
+         break;
+      case 1:
+         fprintf(stderr, "y");
+         break;
+      case 2:
+         fprintf(stderr, "z");
+         break;
+      case 3:
+         fprintf(stderr, "w");
+         break;
+      case 4:
+         fprintf(stderr, "0");
+         break;
+      case 5:
+         fprintf(stderr, "1");
+         break;
+      default:
+         fprintf(stderr, "?");
+         break;
       }
    }
 }
 
 
-static void print_src_reg( GLuint dword )
+static void
+print_src_reg(GLuint dword)
 {
    GLuint nr = (dword >> A2_SRC2_NR_SHIFT) & REG_NR_MASK;
    GLuint type = (dword >> A2_SRC2_TYPE_SHIFT) & REG_TYPE_MASK;
-   print_reg_type_nr( type, nr );
-   print_reg_neg_swizzle( dword );
+   print_reg_type_nr(type, nr);
+   print_reg_neg_swizzle(dword);
 }
 
-void i915_print_ureg( const char *msg, GLuint ureg )
+void
+i915_print_ureg(const char *msg, GLuint ureg)
 {
    fprintf(stderr, "%s: ", msg);
-   print_src_reg( ureg >> 8 );
+   print_src_reg(ureg >> 8);
    fprintf(stderr, "\n");
 }
 
-static void print_dest_reg( GLuint dword )
+static void
+print_dest_reg(GLuint dword)
 {
    GLuint nr = (dword >> A0_DEST_NR_SHIFT) & REG_NR_MASK;
    GLuint type = (dword >> A0_DEST_TYPE_SHIFT) & REG_TYPE_MASK;
-   print_reg_type_nr( type, nr );
+   print_reg_type_nr(type, nr);
    if ((dword & A0_DEST_CHANNEL_ALL) == A0_DEST_CHANNEL_ALL)
       return;
    fprintf(stderr, ".");
-   if (dword & A0_DEST_CHANNEL_X) fprintf(stderr, "x");
-   if (dword & A0_DEST_CHANNEL_Y) fprintf(stderr, "y");
-   if (dword & A0_DEST_CHANNEL_Z) fprintf(stderr, "z");
-   if (dword & A0_DEST_CHANNEL_W) fprintf(stderr, "w");
+   if (dword & A0_DEST_CHANNEL_X)
+      fprintf(stderr, "x");
+   if (dword & A0_DEST_CHANNEL_Y)
+      fprintf(stderr, "y");
+   if (dword & A0_DEST_CHANNEL_Z)
+      fprintf(stderr, "z");
+   if (dword & A0_DEST_CHANNEL_W)
+      fprintf(stderr, "w");
 }
 
 
@@ -213,14 +244,15 @@ static void print_dest_reg( GLuint dword )
 #define GET_SRC2_REG(r)      (r)
 
 
-static void print_arith_op( GLuint opcode, const GLuint *program )
+static void
+print_arith_op(GLuint opcode, const GLuint * program)
 {
    if (opcode != A0_NOP) {
       print_dest_reg(program[0]);
       if (program[0] & A0_DEST_SATURATE)
-	 fprintf(stderr, " = SATURATE ");
+         fprintf(stderr, " = SATURATE ");
       else
-	 fprintf(stderr, " = ");
+         fprintf(stderr, " = ");
    }
 
    fprintf(stderr, "%s ", opcodes[opcode]);
@@ -233,7 +265,7 @@ static void print_arith_op( GLuint opcode, const GLuint *program )
 
    fprintf(stderr, ", ");
    print_src_reg(GET_SRC1_REG(program[1], program[2]));
-   if (args[opcode] == 2) { 
+   if (args[opcode] == 2) {
       fprintf(stderr, "\n");
       return;
    }
@@ -245,22 +277,24 @@ static void print_arith_op( GLuint opcode, const GLuint *program )
 }
 
 
-static void print_tex_op( GLuint opcode, const GLuint *program )
+static void
+print_tex_op(GLuint opcode, const GLuint * program)
 {
    print_dest_reg(program[0] | A0_DEST_CHANNEL_ALL);
    fprintf(stderr, " = ");
 
    fprintf(stderr, "%s ", opcodes[opcode]);
 
-   fprintf(stderr, "S[%d],", 
-	   program[0] & T0_SAMPLER_NR_MASK);
+   fprintf(stderr, "S[%d],", program[0] & T0_SAMPLER_NR_MASK);
 
-   print_reg_type_nr( (program[1]>>T1_ADDRESS_REG_TYPE_SHIFT) & REG_TYPE_MASK,
-		      (program[1]>>T1_ADDRESS_REG_NR_SHIFT) & REG_NR_MASK );
+   print_reg_type_nr((program[1] >> T1_ADDRESS_REG_TYPE_SHIFT) &
+                     REG_TYPE_MASK,
+                     (program[1] >> T1_ADDRESS_REG_NR_SHIFT) & REG_NR_MASK);
    fprintf(stderr, "\n");
 }
 
-static void print_dcl_op( GLuint opcode, const GLuint *program )
+static void
+print_dcl_op(GLuint opcode, const GLuint * program)
 {
    fprintf(stderr, "%s ", opcodes[opcode]);
    print_dest_reg(program[0] | A0_DEST_CHANNEL_ALL);
@@ -268,31 +302,32 @@ static void print_dcl_op( GLuint opcode, const GLuint *program )
 }
 
 
-void i915_disassemble_program( const GLuint *program, GLuint sz )
+void
+i915_disassemble_program(const GLuint * program, GLuint sz)
 {
    GLuint size = program[0] & 0x1ff;
    GLint i;
-   
+
    fprintf(stderr, "BEGIN\n");
 
-   if (size+2 != sz) {
+   if (size + 2 != sz) {
       fprintf(stderr, "%s: program size mismatch %d/%d\n", __FUNCTION__,
-	      size+2, sz);
+              size + 2, sz);
       exit(1);
    }
 
-   program ++;
-   for (i = 1 ; i < sz ; i+=3, program+=3) {
-      GLuint opcode = program[0] & (0x1f<<24);
+   program++;
+   for (i = 1; i < sz; i += 3, program += 3) {
+      GLuint opcode = program[0] & (0x1f << 24);
 
       if ((GLint) opcode >= A0_NOP && opcode <= A0_SLT)
-	 print_arith_op(opcode >> 24, program);
+         print_arith_op(opcode >> 24, program);
       else if (opcode >= T0_TEXLD && opcode <= T0_TEXKILL)
-	 print_tex_op(opcode >> 24, program);
+         print_tex_op(opcode >> 24, program);
       else if (opcode == D0_DCL)
-	 print_dcl_op(opcode >> 24, program);
-      else 
-	 fprintf(stderr, "Unknown opcode 0x%x\n", opcode);
+         print_dcl_op(opcode >> 24, program);
+      else
+         fprintf(stderr, "Unknown opcode 0x%x\n", opcode);
    }
 
    fprintf(stderr, "END\n\n");

src/mesa/drivers/dri/i915/i915_metaops.c

@@ -34,127 +34,170 @@
 #include "intel_screen.h"
 #include "intel_batchbuffer.h"
 #include "intel_ioctl.h"
+#include "intel_regions.h"
 #include "intel_rotate.h"
 
 #include "i915_context.h"
 #include "i915_reg.h"
 
-/* A large amount of state doesn't need to be uploaded.
+/* We touch almost everything:
  */
-#define ACTIVE (I915_UPLOAD_PROGRAM | 		\
-		I915_UPLOAD_STIPPLE |		\
+#define ACTIVE (I915_UPLOAD_INVARIENT | 	\
 		I915_UPLOAD_CTX |		\
 		I915_UPLOAD_BUFFERS |		\
-		I915_UPLOAD_TEX(0))		
+		I915_UPLOAD_STIPPLE |		\
+                I915_UPLOAD_PROGRAM | 		\
+                I915_UPLOAD_FOG | 		\
+		I915_UPLOAD_TEX(0))
 
-#define SET_STATE( i915, STATE )			\
+#define SET_STATE( i915, STATE )		\
 do {						\
    i915->current->emitted &= ~ACTIVE;		\
-   i915->current = &i915->STATE;			\
+   i915->current = &i915->STATE;		\
    i915->current->emitted &= ~ACTIVE;		\
 } while (0)
 
-/* Operations where the 3D engine is decoupled temporarily from the
- * current GL state and used for other purposes than simply rendering
- * incoming triangles.
- */
-static void set_initial_state( i915ContextPtr i915 )
-{
-   memcpy(&i915->meta, &i915->initial, sizeof(i915->meta) );
-   i915->meta.active = ACTIVE;
-   i915->meta.emitted = 0;
-}
 
-
-static void set_no_depth_stencil_write( i915ContextPtr i915 )
+static void
+meta_no_stencil_write(struct intel_context *intel)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
    /* ctx->Driver.Enable( ctx, GL_STENCIL_TEST, GL_FALSE )
     */
-   i915->meta.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_TEST_ENABLE | 
-				       S5_STENCIL_WRITE_ENABLE);
-
-   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
-    */
-   i915->meta.Ctx[I915_CTXREG_LIS6] &= ~(S6_DEPTH_TEST_ENABLE |
-				       S6_DEPTH_WRITE_ENABLE);
+   i915->meta.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_TEST_ENABLE |
+                                         S5_STENCIL_WRITE_ENABLE);
 
    i915->meta.emitted &= ~I915_UPLOAD_CTX;
 }
 
+static void
+meta_no_depth_write(struct intel_context *intel)
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
+   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
+    */
+   i915->meta.Ctx[I915_CTXREG_LIS6] &= ~(S6_DEPTH_TEST_ENABLE |
+                                         S6_DEPTH_WRITE_ENABLE);
+
+   i915->meta.emitted &= ~I915_UPLOAD_CTX;
+}
+
+static void
+meta_depth_replace(struct intel_context *intel)
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
+   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_TRUE )
+    * ctx->Driver.DepthMask( ctx, GL_TRUE )
+    */
+   i915->meta.Ctx[I915_CTXREG_LIS6] |= (S6_DEPTH_TEST_ENABLE |
+                                        S6_DEPTH_WRITE_ENABLE);
+
+   /* ctx->Driver.DepthFunc( ctx, GL_REPLACE )
+    */
+   i915->meta.Ctx[I915_CTXREG_LIS6] &= ~S6_DEPTH_TEST_FUNC_MASK;
+   i915->meta.Ctx[I915_CTXREG_LIS6] |=
+      COMPAREFUNC_ALWAYS << S6_DEPTH_TEST_FUNC_SHIFT;
+
+   i915->meta.emitted &= ~I915_UPLOAD_CTX;
+}
+
+
 /* Set stencil unit to replace always with the reference value.
  */
-static void set_stencil_replace( i915ContextPtr i915,
-				 GLuint s_mask,
-				 GLuint s_clear)
+static void
+meta_stencil_replace(struct intel_context *intel,
+                     GLuint s_mask, GLuint s_clear)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
    GLuint op = STENCILOP_REPLACE;
    GLuint func = COMPAREFUNC_ALWAYS;
 
    /* ctx->Driver.Enable( ctx, GL_STENCIL_TEST, GL_TRUE )
     */
-   i915->meta.Ctx[I915_CTXREG_LIS5] |= (S5_STENCIL_TEST_ENABLE | 
-				      S5_STENCIL_WRITE_ENABLE);
-
-
-   /* ctx->Driver.Enable( ctx, GL_DEPTH_TEST, GL_FALSE )
-    */
-   i915->meta.Ctx[I915_CTXREG_LIS6] &= ~(S6_DEPTH_TEST_ENABLE |
-				       S6_DEPTH_WRITE_ENABLE);
-
+   i915->meta.Ctx[I915_CTXREG_LIS5] |= (S5_STENCIL_TEST_ENABLE |
+                                        S5_STENCIL_WRITE_ENABLE);
 
    /* ctx->Driver.StencilMask( ctx, s_mask )
     */
    i915->meta.Ctx[I915_CTXREG_STATE4] &= ~MODE4_ENABLE_STENCIL_WRITE_MASK;
 
    i915->meta.Ctx[I915_CTXREG_STATE4] |= (ENABLE_STENCIL_WRITE_MASK |
-					STENCIL_WRITE_MASK(s_mask));
-
+                                          STENCIL_WRITE_MASK(s_mask));
 
    /* ctx->Driver.StencilOp( ctx, GL_REPLACE, GL_REPLACE, GL_REPLACE )
     */
    i915->meta.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_FAIL_MASK |
-				       S5_STENCIL_PASS_Z_FAIL_MASK |
-				       S5_STENCIL_PASS_Z_PASS_MASK);
+                                         S5_STENCIL_PASS_Z_FAIL_MASK |
+                                         S5_STENCIL_PASS_Z_PASS_MASK);
 
    i915->meta.Ctx[I915_CTXREG_LIS5] |= ((op << S5_STENCIL_FAIL_SHIFT) |
-				      (op << S5_STENCIL_PASS_Z_FAIL_SHIFT) |
-				      (op << S5_STENCIL_PASS_Z_PASS_SHIFT));
+                                        (op << S5_STENCIL_PASS_Z_FAIL_SHIFT) |
+                                        (op << S5_STENCIL_PASS_Z_PASS_SHIFT));
 
 
    /* ctx->Driver.StencilFunc( ctx, GL_ALWAYS, s_ref, ~0 )
     */
    i915->meta.Ctx[I915_CTXREG_STATE4] &= ~MODE4_ENABLE_STENCIL_TEST_MASK;
    i915->meta.Ctx[I915_CTXREG_STATE4] |= (ENABLE_STENCIL_TEST_MASK |
-					STENCIL_TEST_MASK(0xff));
+                                          STENCIL_TEST_MASK(0xff));
 
    i915->meta.Ctx[I915_CTXREG_LIS5] &= ~(S5_STENCIL_REF_MASK |
-				       S5_STENCIL_TEST_FUNC_MASK);
-					
-   i915->meta.Ctx[I915_CTXREG_LIS5] |= ((s_clear << S5_STENCIL_REF_SHIFT) |  
-				      (func << S5_STENCIL_TEST_FUNC_SHIFT)); 
+                                         S5_STENCIL_TEST_FUNC_MASK);
+
+   i915->meta.Ctx[I915_CTXREG_LIS5] |= ((s_clear << S5_STENCIL_REF_SHIFT) |
+                                        (func << S5_STENCIL_TEST_FUNC_SHIFT));
 
 
    i915->meta.emitted &= ~I915_UPLOAD_CTX;
 }
 
 
-static void set_color_mask( i915ContextPtr i915, GLboolean state )
+static void
+meta_color_mask(struct intel_context *intel, GLboolean state)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
    const GLuint mask = (S5_WRITEDISABLE_RED |
-			S5_WRITEDISABLE_GREEN |
-			S5_WRITEDISABLE_BLUE |
-			S5_WRITEDISABLE_ALPHA);
+                        S5_WRITEDISABLE_GREEN |
+                        S5_WRITEDISABLE_BLUE | S5_WRITEDISABLE_ALPHA);
 
    /* Copy colormask state from "regular" hw context.
     */
    if (state) {
       i915->meta.Ctx[I915_CTXREG_LIS5] &= ~mask;
-      i915->meta.Ctx[I915_CTXREG_LIS5] |= 
-	 (i915->state.Ctx[I915_CTXREG_LIS5] & mask);
+      i915->meta.Ctx[I915_CTXREG_LIS5] |=
+         (i915->state.Ctx[I915_CTXREG_LIS5] & mask);
    }
-   else 
+   else
       i915->meta.Ctx[I915_CTXREG_LIS5] |= mask;
-      
+
+   i915->meta.emitted &= ~I915_UPLOAD_CTX;
+}
+
+
+
+static void
+meta_import_pixel_state(struct intel_context *intel)
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   memcpy(i915->meta.Fog, i915->state.Fog, I915_FOG_SETUP_SIZE * 4);
+
+   i915->meta.Ctx[I915_CTXREG_LIS5] = i915->state.Ctx[I915_CTXREG_LIS5];
+   i915->meta.Ctx[I915_CTXREG_LIS6] = i915->state.Ctx[I915_CTXREG_LIS6];
+   i915->meta.Ctx[I915_CTXREG_STATE4] = i915->state.Ctx[I915_CTXREG_STATE4];
+   i915->meta.Ctx[I915_CTXREG_BLENDCOLOR1] =
+      i915->state.Ctx[I915_CTXREG_BLENDCOLOR1];
+   i915->meta.Ctx[I915_CTXREG_IAB] = i915->state.Ctx[I915_CTXREG_IAB];
+
+   i915->meta.Buffer[I915_DESTREG_SENABLE] =
+      i915->state.Buffer[I915_DESTREG_SENABLE];
+   i915->meta.Buffer[I915_DESTREG_SR1] = i915->state.Buffer[I915_DESTREG_SR1];
+   i915->meta.Buffer[I915_DESTREG_SR2] = i915->state.Buffer[I915_DESTREG_SR2];
+
+   i915->meta.emitted &= ~I915_UPLOAD_FOG;
+   i915->meta.emitted &= ~I915_UPLOAD_BUFFERS;
    i915->meta.emitted &= ~I915_UPLOAD_CTX;
 }
 
@@ -211,69 +254,64 @@ static void set_color_mask( i915ContextPtr i915, GLboolean state )
 
 
 
-static void set_no_texture( i915ContextPtr i915 )
+static void
+meta_no_texture(struct intel_context *intel)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
    static const GLuint prog[] = {
       _3DSTATE_PIXEL_SHADER_PROGRAM,
 
       /* Declare incoming diffuse color:
        */
-      (D0_DCL |
-       D0_DECL_REG( REG_T_DIFFUSE ) |
-       D0_CHANNEL_ALL),
+      (D0_DCL | D0_DECL_REG(REG_T_DIFFUSE) | D0_CHANNEL_ALL),
       D1_MBZ,
       D2_MBZ,
 
       /* output-color = mov(t_diffuse)
        */
       (A0_MOV |
-       A0_DEST_REG( REG_OC ) |
-       A0_DEST_CHANNEL_ALL |
-       A0_SRC0_REG( REG_T_DIFFUSE )),
+       A0_DEST_REG(REG_OC) |
+       A0_DEST_CHANNEL_ALL | A0_SRC0_REG(REG_T_DIFFUSE)),
       (A1_SRC0_XYZW),
       0,
    };
 
-   
-   memcpy( i915->meta.Program, prog, sizeof(prog) );
+
+   memcpy(i915->meta.Program, prog, sizeof(prog));
    i915->meta.ProgramSize = sizeof(prog) / sizeof(*prog);
    i915->meta.Program[0] |= i915->meta.ProgramSize - 2;
    i915->meta.emitted &= ~I915_UPLOAD_PROGRAM;
 }
 
-
-static void enable_texture_blend_replace( i915ContextPtr i915 )
+static void
+meta_texture_blend_replace(struct intel_context *intel)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
+
    static const GLuint prog[] = {
       _3DSTATE_PIXEL_SHADER_PROGRAM,
 
       /* Declare the sampler:
        */
-      (D0_DCL |
-       D0_DECL_REG( REG_S(0) ) |
-       D0_SAMPLE_TYPE_2D |
-       D0_CHANNEL_NONE),
+      (D0_DCL | D0_DECL_REG(REG_S(0)) | D0_SAMPLE_TYPE_2D | D0_CHANNEL_NONE),
       D1_MBZ,
       D2_MBZ,
 
       /* Declare the interpolated texture coordinate:
        */
-      (D0_DCL |
-       D0_DECL_REG( REG_T_TEX(0) ) |
-       D0_CHANNEL_ALL),
+      (D0_DCL | D0_DECL_REG(REG_T_TEX(0)) | D0_CHANNEL_ALL),
       D1_MBZ,
       D2_MBZ,
 
       /* output-color = texld(sample0, texcoord0) 
        */
-      (T0_TEXLD | 
-       T0_DEST_REG( REG_OC ) |
-       T0_SAMPLER( 0 )),
+      (T0_TEXLD | T0_DEST_REG(REG_OC) | T0_SAMPLER(0)),
       T1_ADDRESS_REG(REG_TYPE_T, 0),
       T2_MBZ
    };
 
-   memcpy( i915->meta.Program, prog, sizeof(prog) );
+   memcpy(i915->meta.Program, prog, sizeof(prog));
    i915->meta.ProgramSize = sizeof(prog) / sizeof(*prog);
    i915->meta.Program[0] |= i915->meta.ProgramSize - 2;
    i915->meta.emitted &= ~I915_UPLOAD_PROGRAM;
@@ -286,415 +324,186 @@ static void enable_texture_blend_replace( i915ContextPtr i915 )
 /* Set up an arbitary piece of memory as a rectangular texture
  * (including the front or back buffer).
  */
-static void set_tex_rect_source( i915ContextPtr i915,
-				 GLuint offset,
-				 GLuint width, 
-				 GLuint height,
-				 GLuint pitch, /* in bytes! */
-				 GLuint textureFormat )
+static GLboolean
+meta_tex_rect_source(struct intel_context *intel,
+                     struct _DriBufferObject *buffer,
+                     GLuint offset,
+                     GLuint pitch, GLuint height, GLenum format, GLenum type)
 {
+   struct i915_context *i915 = i915_context(&intel->ctx);
    GLuint unit = 0;
    GLint numLevels = 1;
    GLuint *state = i915->meta.Tex[0];
+   GLuint textureFormat;
+   GLuint cpp;
 
-#if 0
-   printf("TexRect source offset 0x%x  pitch %d\n", offset, pitch);
-#endif
+   /* A full implementation of this would do the upload through
+    * glTexImage2d, and get all the conversion operations at that
+    * point.  We are restricted, but still at least have access to the
+    * fragment program swizzle.
+    */
+   switch (format) {
+   case GL_BGRA:
+      switch (type) {
+      case GL_UNSIGNED_INT_8_8_8_8_REV:
+      case GL_UNSIGNED_BYTE:
+         textureFormat = (MAPSURF_32BIT | MT_32BIT_ARGB8888);
+         cpp = 4;
+         break;
+      default:
+         return GL_FALSE;
+      }
+      break;
+   case GL_RGBA:
+      switch (type) {
+      case GL_UNSIGNED_INT_8_8_8_8_REV:
+      case GL_UNSIGNED_BYTE:
+         textureFormat = (MAPSURF_32BIT | MT_32BIT_ABGR8888);
+         cpp = 4;
+         break;
+      default:
+         return GL_FALSE;
+      }
+      break;
+   case GL_BGR:
+      switch (type) {
+      case GL_UNSIGNED_SHORT_5_6_5_REV:
+         textureFormat = (MAPSURF_16BIT | MT_16BIT_RGB565);
+         cpp = 2;
+         break;
+      default:
+         return GL_FALSE;
+      }
+      break;
+   case GL_RGB:
+      switch (type) {
+      case GL_UNSIGNED_SHORT_5_6_5:
+         textureFormat = (MAPSURF_16BIT | MT_16BIT_RGB565);
+         cpp = 2;
+         break;
+      default:
+         return GL_FALSE;
+      }
+      break;
 
-/*    fprintf(stderr, "%s: offset: %x w: %d h: %d pitch %d format %x\n", */
-/* 	   __FUNCTION__, offset, width, height, pitch, textureFormat ); */
+   default:
+      return GL_FALSE;
+   }
+
+
+   if ((pitch * cpp) & 3) {
+      _mesa_printf("%s: texture is not dword pitch\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+/*    intel_region_release(intel, &i915->meta.tex_region[0]); */
+/*    intel_region_reference(&i915->meta.tex_region[0], region); */
+   i915->meta.tex_buffer[0] = buffer;
+   i915->meta.tex_offset[0] = offset;
 
-   state[I915_TEXREG_MS2] = offset;
    state[I915_TEXREG_MS3] = (((height - 1) << MS3_HEIGHT_SHIFT) |
-			    ((width - 1) << MS3_WIDTH_SHIFT) |
-			    textureFormat |
-			    MS3_USE_FENCE_REGS);
+                             ((pitch - 1) << MS3_WIDTH_SHIFT) |
+                             textureFormat | MS3_USE_FENCE_REGS);
 
-   state[I915_TEXREG_MS4] = ((((pitch / 4) - 1) << MS4_PITCH_SHIFT) | 
-			    ((((numLevels-1) * 4)) << MS4_MAX_LOD_SHIFT));
+   state[I915_TEXREG_MS4] = (((((pitch * cpp) / 4) - 1) << MS4_PITCH_SHIFT) |
+                             MS4_CUBE_FACE_ENA_MASK |
+                             ((((numLevels - 1) * 4)) << MS4_MAX_LOD_SHIFT));
 
    state[I915_TEXREG_SS2] = ((FILTER_NEAREST << SS2_MIN_FILTER_SHIFT) |
-			    (MIPFILTER_NONE << SS2_MIP_FILTER_SHIFT) |
-			    (FILTER_NEAREST << SS2_MAG_FILTER_SHIFT));
+                             (MIPFILTER_NONE << SS2_MIP_FILTER_SHIFT) |
+                             (FILTER_NEAREST << SS2_MAG_FILTER_SHIFT));
+
    state[I915_TEXREG_SS3] = ((TEXCOORDMODE_WRAP << SS3_TCX_ADDR_MODE_SHIFT) |
-			    (TEXCOORDMODE_WRAP << SS3_TCY_ADDR_MODE_SHIFT) |
-			    (TEXCOORDMODE_WRAP << SS3_TCZ_ADDR_MODE_SHIFT) |
-			    (unit<<SS3_TEXTUREMAP_INDEX_SHIFT));
+                             (TEXCOORDMODE_WRAP << SS3_TCY_ADDR_MODE_SHIFT) |
+                             (TEXCOORDMODE_WRAP << SS3_TCZ_ADDR_MODE_SHIFT) |
+                             (unit << SS3_TEXTUREMAP_INDEX_SHIFT));
 
    state[I915_TEXREG_SS4] = 0;
 
    i915->meta.emitted &= ~I915_UPLOAD_TEX(0);
+   return GL_TRUE;
 }
 
 
-/* Select between front and back draw buffers.
+/**
+ * Set the color and depth drawing region for meta ops.
  */
-static void set_draw_region( i915ContextPtr i915, const intelRegion *region )
+static void
+meta_draw_region(struct intel_context *intel,
+                 struct intel_region *color_region,
+                 struct intel_region *depth_region)
 {
-#if 0
-   printf("Rotate into region: offset 0x%x  pitch %d\n",
-          region->offset, region->pitch);
-#endif
-   i915->meta.Buffer[I915_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i915->meta.Buffer[I915_DESTREG_CBUFADDR2] = region->offset;
-   i915->meta.emitted &= ~I915_UPLOAD_BUFFERS;
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   i915_state_draw_region(intel, &i915->meta, color_region, depth_region);
 }
 
 
-#if 0
-/* Setup an arbitary draw format, useful for targeting texture or agp
- * memory.
- */
-static void set_draw_format( i915ContextPtr i915,
-			     GLuint format,
-			     GLuint depth_format)
+static void
+set_vertex_format(struct intel_context *intel)
 {
-   i915->meta.Buffer[I915_DESTREG_DV1] = (DSTORG_HORT_BIAS(0x8) | /* .5 */
-					DSTORG_VERT_BIAS(0x8) | /* .5 */
-					format |
-					LOD_PRECLAMP_OGL |
-					TEX_DEFAULT_COLOR_OGL |
-					depth_format);
+   struct i915_context *i915 = i915_context(&intel->ctx);
 
-   i915->meta.emitted &= ~I915_UPLOAD_BUFFERS;
-/*    fprintf(stderr, "%s: DV1: %x\n",  */
-/* 	   __FUNCTION__, i915->meta.Buffer[I915_DESTREG_DV1]); */
-}
-#endif
-
-static void set_vertex_format( i915ContextPtr i915 )
-{
-   i915->meta.Ctx[I915_CTXREG_LIS2] = 
+   i915->meta.Ctx[I915_CTXREG_LIS2] =
       (S2_TEXCOORD_FMT(0, TEXCOORDFMT_2D) |
-       S2_TEXCOORD_FMT(1, TEXCOORDFMT_NOT_PRESENT) | 
+       S2_TEXCOORD_FMT(1, TEXCOORDFMT_NOT_PRESENT) |
        S2_TEXCOORD_FMT(2, TEXCOORDFMT_NOT_PRESENT) |
        S2_TEXCOORD_FMT(3, TEXCOORDFMT_NOT_PRESENT) |
        S2_TEXCOORD_FMT(4, TEXCOORDFMT_NOT_PRESENT) |
-       S2_TEXCOORD_FMT(5, TEXCOORDFMT_NOT_PRESENT) | 
+       S2_TEXCOORD_FMT(5, TEXCOORDFMT_NOT_PRESENT) |
        S2_TEXCOORD_FMT(6, TEXCOORDFMT_NOT_PRESENT) |
        S2_TEXCOORD_FMT(7, TEXCOORDFMT_NOT_PRESENT));
 
    i915->meta.Ctx[I915_CTXREG_LIS4] &= ~S4_VFMT_MASK;
 
-   i915->meta.Ctx[I915_CTXREG_LIS4] |= 
-      (S4_VFMT_COLOR |
-       S4_VFMT_SPEC_FOG |
-       S4_VFMT_XYZW);
+   i915->meta.Ctx[I915_CTXREG_LIS4] |= (S4_VFMT_COLOR | S4_VFMT_XYZ);
 
    i915->meta.emitted &= ~I915_UPLOAD_CTX;
-
 }
 
 
-static void draw_quad(i915ContextPtr i915, 
-		      GLfloat x0, GLfloat x1,
-		      GLfloat y0, GLfloat y1, 
-		      GLubyte red, GLubyte green,
-		      GLubyte blue, GLubyte alpha,
-		      GLfloat s0, GLfloat s1,
-		      GLfloat t0, GLfloat t1 )
-{
-   GLuint vertex_size = 8;
-   GLuint *vb = intelEmitInlinePrimitiveLocked( &i915->intel, 
-						PRIM3D_TRIFAN, 
-						4 * vertex_size,
-						vertex_size );
-   intelVertex tmp;
-   int i;
 
-   if (0)
-      fprintf(stderr, "%s: %f,%f-%f,%f 0x%x%x%x%x %f,%f-%f,%f\n",
-	      __FUNCTION__,
-	      x0,y0,x1,y1,red,green,blue,alpha,s0,t0,s1,t1);
-
-
-   /* initial vertex, left bottom */
-   tmp.v.x = x0;
-   tmp.v.y = y0;
-   tmp.v.z = 1.0;
-   tmp.v.w = 1.0; 
-   tmp.v.color.red = red;
-   tmp.v.color.green = green;
-   tmp.v.color.blue = blue;
-   tmp.v.color.alpha = alpha;
-   tmp.v.specular.red = 0;
-   tmp.v.specular.green = 0;
-   tmp.v.specular.blue = 0;
-   tmp.v.specular.alpha = 0;
-   tmp.v.u0 = s0;
-   tmp.v.v0 = t0;
-
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* right bottom */
-   vb += vertex_size;
-   tmp.v.x = x1;
-   tmp.v.u0 = s1;
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* right top */
-   vb += vertex_size;
-   tmp.v.y = y1;
-   tmp.v.v0 = t1;
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-
-   /* left top */
-   vb += vertex_size;
-   tmp.v.x = x0;
-   tmp.v.u0 = s0;
-   for (i = 0 ; i < vertex_size ; i++)
-      vb[i] = tmp.ui[i];
-}
-
-
-static void draw_poly(i915ContextPtr i915, 
-		      GLubyte red, GLubyte green, GLubyte blue, GLubyte alpha,
-                      GLuint numVerts,
-                      /*const*/ GLfloat verts[][2],
-                      /*const*/ GLfloat texcoords[][2])
-{
-   GLuint vertex_size = 8;
-   GLuint *vb = intelEmitInlinePrimitiveLocked( &i915->intel, 
-						PRIM3D_TRIFAN, 
-						numVerts * vertex_size,
-						vertex_size );
-   intelVertex tmp;
-   int i, k;
-
-   /* initial constant vertex fields */
-   tmp.v.z = 1.0;
-   tmp.v.w = 1.0; 
-   tmp.v.color.red = red;
-   tmp.v.color.green = green;
-   tmp.v.color.blue = blue;
-   tmp.v.color.alpha = alpha;
-   tmp.v.specular.red = 0;
-   tmp.v.specular.green = 0;
-   tmp.v.specular.blue = 0;
-   tmp.v.specular.alpha = 0;
-
-   for (k = 0; k < numVerts; k++) {
-      tmp.v.x = verts[k][0];
-      tmp.v.y = verts[k][1];
-      tmp.v.u0 = texcoords[k][0];
-      tmp.v.v0 = texcoords[k][1];
-
-      for (i = 0 ; i < vertex_size ; i++)
-         vb[i] = tmp.ui[i];
-
-      vb += vertex_size;
-   }
-}
-
-
-void 
-i915ClearWithTris(intelContextPtr intel, GLbitfield mask,
-		  GLboolean all,
-		  GLint cx, GLint cy, GLint cw, GLint ch)
-{
-   i915ContextPtr i915 = I915_CONTEXT( intel );
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   intelScreenPrivate *screen = intel->intelScreen;
-   int x0, y0, x1, y1;
-
-   SET_STATE( i915, meta ); 
-   set_initial_state( i915 ); 
-   set_no_texture( i915 ); 
-   set_vertex_format( i915 ); 
-
-   LOCK_HARDWARE(intel);
-
-   if (!all) {
-      x0 = cx;
-      y0 = cy;
-      x1 = x0 + cw;
-      y1 = y0 + ch;
-   } else {
-      x0 = 0;
-      y0 = 0;
-      x1 = x0 + dPriv->w;
-      y1 = y0 + dPriv->h;
-   }
-
-   /* Don't do any clipping to screen - these are window coordinates.
-    * The active cliprects will be applied as for any other geometry.
-    */
-
-   if (mask & BUFFER_BIT_FRONT_LEFT) { 
-      set_no_depth_stencil_write( i915 );
-      set_color_mask( i915, GL_TRUE );
-      set_draw_region( i915, &screen->front );
-
-      draw_quad(i915, x0, x1, y0, y1,
-		intel->clear_red, intel->clear_green, 
- 		intel->clear_blue, intel->clear_alpha, 
-		0, 0, 0, 0);
-   }
-
-   if (mask & BUFFER_BIT_BACK_LEFT) {
-      set_no_depth_stencil_write( i915 );
-      set_color_mask( i915, GL_TRUE );
-      set_draw_region( i915, &screen->back );
-
-      draw_quad(i915, x0, x1, y0, y1,
-		intel->clear_red, intel->clear_green,
-		intel->clear_blue, intel->clear_alpha,
-		0, 0, 0, 0);
-   }
-
-   if (mask & BUFFER_BIT_STENCIL) {
-      set_stencil_replace( i915, 
-			   intel->ctx.Stencil.WriteMask[0], 
-			   intel->ctx.Stencil.Clear);
-      
-      set_color_mask( i915, GL_FALSE );
-      set_draw_region( i915, &screen->front ); /* could be either? */
-
-      draw_quad( i915, x0, x1, y0, y1, 0, 0, 0, 0, 0, 0, 0, 0 );
-   }
-
-   UNLOCK_HARDWARE(intel);
-
-   SET_STATE( i915, state );
-}
-
-
-/**
- * Copy the window contents named by dPriv to the rotated (or reflected)
- * color buffer.
- * srcBuf is BUFFER_BIT_FRONT_LEFT or BUFFER_BIT_BACK_LEFT to indicate the source.
+/* Operations where the 3D engine is decoupled temporarily from the
+ * current GL state and used for other purposes than simply rendering
+ * incoming triangles.
  */
-void
-i915RotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
-                 GLuint srcBuf)
+static void
+install_meta_state(struct intel_context *intel)
 {
-   i915ContextPtr i915 = I915_CONTEXT( intel );
-   intelScreenPrivate *screen = intel->intelScreen;
-   const GLuint cpp = screen->cpp;
-   drm_clip_rect_t fullRect;
-   GLuint textureFormat, srcOffset, srcPitch;
-   const drm_clip_rect_t *clipRects;
-   int numClipRects;
-   int i;
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   memcpy(&i915->meta, &i915->initial, sizeof(i915->meta));
+   i915->meta.active = ACTIVE;
+   i915->meta.emitted = 0;
 
-   int xOrig, yOrig;
-   int origNumClipRects;
-   drm_clip_rect_t *origRects;
-
-   /*
-    * set up hardware state
-    */
-   intelFlush( &intel->ctx );
-
-   SET_STATE( i915, meta ); 
-   set_initial_state( i915 ); 
-   set_no_texture( i915 ); 
-   set_vertex_format( i915 ); 
-   set_no_depth_stencil_write( i915 );
-   set_color_mask( i915, GL_TRUE );
-
-   LOCK_HARDWARE(intel);
-
-   /* save current drawing origin and cliprects (restored at end) */
-   xOrig = intel->drawX;
-   yOrig = intel->drawY;
-   origNumClipRects = intel->numClipRects;
-   origRects = intel->pClipRects;
-
-   if (!intel->numClipRects)
-      goto done;
-
-   /*
-    * set drawing origin, cliprects for full-screen access to rotated screen
-    */
-   fullRect.x1 = 0;
-   fullRect.y1 = 0;
-   fullRect.x2 = screen->rotatedWidth;
-   fullRect.y2 = screen->rotatedHeight;
-   intel->drawX = 0;
-   intel->drawY = 0;
-   intel->numClipRects = 1;
-   intel->pClipRects = &fullRect;
-
-   set_draw_region( i915, &screen->rotated );
-
-   if (cpp == 4)
-      textureFormat = MAPSURF_32BIT | MT_32BIT_ARGB8888;
-   else
-      textureFormat = MAPSURF_16BIT | MT_16BIT_RGB565;
-
-   if (srcBuf == BUFFER_BIT_FRONT_LEFT) {
-      srcPitch = screen->front.pitch;   /* in bytes */
-      srcOffset = screen->front.offset; /* bytes */
-      clipRects = dPriv->pClipRects;
-      numClipRects = dPriv->numClipRects;
-   }
-   else {
-      srcPitch = screen->back.pitch;   /* in bytes */
-      srcOffset = screen->back.offset; /* bytes */
-      clipRects = dPriv->pBackClipRects;
-      numClipRects = dPriv->numBackClipRects;
-   }
-
-   /* set the whole screen up as a texture to avoid alignment issues */
-   set_tex_rect_source(i915,
-                       srcOffset,
-                       screen->width,
-		       screen->height,
-                       srcPitch,
-                       textureFormat);
-
-   enable_texture_blend_replace(i915);
-
-   /*
-    * loop over the source window's cliprects
-    */
-   for (i = 0; i < numClipRects; i++) {
-      int srcX0 = clipRects[i].x1;
-      int srcY0 = clipRects[i].y1;
-      int srcX1 = clipRects[i].x2;
-      int srcY1 = clipRects[i].y2;
-      GLfloat verts[4][2], tex[4][2];
-      int j;
-
-      /* build vertices for four corners of clip rect */
-      verts[0][0] = srcX0;  verts[0][1] = srcY0;
-      verts[1][0] = srcX1;  verts[1][1] = srcY0;
-      verts[2][0] = srcX1;  verts[2][1] = srcY1;
-      verts[3][0] = srcX0;  verts[3][1] = srcY1;
-
-      /* .. and texcoords */
-      tex[0][0] = srcX0;  tex[0][1] = srcY0;
-      tex[1][0] = srcX1;  tex[1][1] = srcY0;
-      tex[2][0] = srcX1;  tex[2][1] = srcY1;
-      tex[3][0] = srcX0;  tex[3][1] = srcY1;
-
-      /* transform coords to rotated screen coords */
-      for (j = 0; j < 4; j++) {
-         matrix23TransformCoordf(&screen->rotMatrix,
-                                 &verts[j][0], &verts[j][1]);
-      }
-
-      /* draw polygon to map source image to dest region */
-      draw_poly(i915, 255, 255, 255, 255, 4, verts, tex);
-
-   } /* cliprect loop */
-
-   intelFlushBatchLocked( intel, GL_FALSE, GL_FALSE, GL_FALSE );
-
- done:
-   /* restore original drawing origin and cliprects */
-   intel->drawX = xOrig;
-   intel->drawY = yOrig;
-   intel->numClipRects = origNumClipRects;
-   intel->pClipRects = origRects;
-
-   UNLOCK_HARDWARE(intel);
-
-   SET_STATE( i915, state );
+   SET_STATE(i915, meta);
+   set_vertex_format(intel);
+   meta_no_texture(intel);
 }
 
+static void
+leave_meta_state(struct intel_context *intel)
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   intel_region_release(intel->intelScreen, &i915->meta.draw_region);
+   intel_region_release(intel->intelScreen, &i915->meta.depth_region);
+/*    intel_region_release(intel, &i915->meta.tex_region[0]); */
+   SET_STATE(i915, state);
+}
+
+
+
+void
+i915InitMetaFuncs(struct i915_context *i915)
+{
+   i915->intel.vtbl.install_meta_state = install_meta_state;
+   i915->intel.vtbl.leave_meta_state = leave_meta_state;
+   i915->intel.vtbl.meta_no_depth_write = meta_no_depth_write;
+   i915->intel.vtbl.meta_no_stencil_write = meta_no_stencil_write;
+   i915->intel.vtbl.meta_stencil_replace = meta_stencil_replace;
+   i915->intel.vtbl.meta_depth_replace = meta_depth_replace;
+   i915->intel.vtbl.meta_color_mask = meta_color_mask;
+   i915->intel.vtbl.meta_no_texture = meta_no_texture;
+   i915->intel.vtbl.meta_texture_blend_replace = meta_texture_blend_replace;
+   i915->intel.vtbl.meta_tex_rect_source = meta_tex_rect_source;
+   i915->intel.vtbl.meta_draw_region = meta_draw_region;
+   i915->intel.vtbl.meta_import_pixel_state = meta_import_pixel_state;
+}

src/mesa/drivers/dri/i915/i915_program.c

@@ -72,58 +72,62 @@
 
 #define I915_CONSTFLAG_PARAM 0x1f
 
-GLuint i915_get_temp( struct i915_fragment_program *p )
+GLuint
+i915_get_temp(struct i915_fragment_program *p)
 {
-   int bit = ffs( ~p->temp_flag );
+   int bit = ffs(~p->temp_flag);
    if (!bit) {
       fprintf(stderr, "%s: out of temporaries\n", __FILE__);
       exit(1);
    }
 
-   p->temp_flag |= 1<<(bit-1);
-   return UREG(REG_TYPE_R, (bit-1));
+   p->temp_flag |= 1 << (bit - 1);
+   return UREG(REG_TYPE_R, (bit - 1));
 }
 
 
-GLuint i915_get_utemp( struct i915_fragment_program *p )
+GLuint
+i915_get_utemp(struct i915_fragment_program * p)
 {
-   int bit = ffs( ~p->utemp_flag );
+   int bit = ffs(~p->utemp_flag);
    if (!bit) {
       fprintf(stderr, "%s: out of temporaries\n", __FILE__);
       exit(1);
    }
 
-   p->utemp_flag |= 1<<(bit-1);
-   return UREG(REG_TYPE_U, (bit-1));
+   p->utemp_flag |= 1 << (bit - 1);
+   return UREG(REG_TYPE_U, (bit - 1));
 }
 
-void i915_release_utemps( struct i915_fragment_program *p )
+void
+i915_release_utemps(struct i915_fragment_program *p)
 {
    p->utemp_flag = ~0x7;
 }
 
 
-GLuint i915_emit_decl( struct i915_fragment_program *p,
-		      GLuint type, GLuint nr, GLuint d0_flags )
+GLuint
+i915_emit_decl(struct i915_fragment_program *p,
+               GLuint type, GLuint nr, GLuint d0_flags)
 {
    GLuint reg = UREG(type, nr);
 
    if (type == REG_TYPE_T) {
-      if (p->decl_t & (1<<nr))
-	 return reg;
+      if (p->decl_t & (1 << nr))
+         return reg;
 
-      p->decl_t |= (1<<nr);
+      p->decl_t |= (1 << nr);
    }
    else if (type == REG_TYPE_S) {
-      if (p->decl_s & (1<<nr))
-	 return reg;
+      if (p->decl_s & (1 << nr))
+         return reg;
 
-      p->decl_s |= (1<<nr);
+      p->decl_s |= (1 << nr);
    }
-   else 
+   else
       return reg;
 
-   *(p->decl++) = (D0_DCL | D0_DEST( reg ) | d0_flags);
+   *(p->decl++) = (D0_DCL | D0_DEST(reg) | d0_flags);
    *(p->decl++) = D1_MBZ;
    *(p->decl++) = D2_MBZ;
 
@@ -131,24 +135,26 @@ GLuint i915_emit_decl( struct i915_fragment_program *p,
    return reg;
 }
 
-GLuint i915_emit_arith( struct i915_fragment_program *p,
-		       GLuint op,
-		       GLuint dest,
-		       GLuint mask,
-		       GLuint saturate,
-		       GLuint src0,
-		       GLuint src1,
-		       GLuint src2 )
+GLuint
+i915_emit_arith(struct i915_fragment_program * p,
+                GLuint op,
+                GLuint dest,
+                GLuint mask,
+                GLuint saturate, GLuint src0, GLuint src1, GLuint src2)
 {
    GLuint c[3];
    GLuint nr_const = 0;
 
    assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
-   assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
+   dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest));
+   assert(dest);
 
-   if (GET_UREG_TYPE(src0) == REG_TYPE_CONST) c[nr_const++] = 0;
-   if (GET_UREG_TYPE(src1) == REG_TYPE_CONST) c[nr_const++] = 1;
-   if (GET_UREG_TYPE(src2) == REG_TYPE_CONST) c[nr_const++] = 2;
+   if (GET_UREG_TYPE(src0) == REG_TYPE_CONST)
+      c[nr_const++] = 0;
+   if (GET_UREG_TYPE(src1) == REG_TYPE_CONST)
+      c[nr_const++] = 1;
+   if (GET_UREG_TYPE(src2) == REG_TYPE_CONST)
+      c[nr_const++] = 2;
 
    /* Recursively call this function to MOV additional const values
     * into temporary registers.  Use utemp registers for this -
@@ -164,81 +170,97 @@ GLuint i915_emit_arith( struct i915_fragment_program *p,
       old_utemp_flag = p->utemp_flag;
 
       first = GET_UREG_NR(s[c[0]]);
-      for (i = 1 ; i < nr_const ; i++) {
-	 if (GET_UREG_NR(s[c[i]]) != first) {
-	    GLuint tmp = i915_get_utemp(p);
+      for (i = 1; i < nr_const; i++) {
+         if (GET_UREG_NR(s[c[i]]) != first) {
+            GLuint tmp = i915_get_utemp(p);
 
-	    i915_emit_arith( p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
-			    s[c[i]], 0, 0 );
-	    s[c[i]] = tmp;
-	 }
+            i915_emit_arith(p, A0_MOV, tmp, A0_DEST_CHANNEL_ALL, 0,
+                            s[c[i]], 0, 0);
+            s[c[i]] = tmp;
+         }
       }
 
       src0 = s[0];
       src1 = s[1];
       src2 = s[2];
-      p->utemp_flag = old_utemp_flag; /* restore */
+      p->utemp_flag = old_utemp_flag;   /* restore */
    }
 
-   *(p->csr++) = (op | 
-		  A0_DEST( dest ) |
-		  mask | 
-		  saturate |
-		  A0_SRC0( src0 ));
-   *(p->csr++) = (A1_SRC0( src0 ) |
-		  A1_SRC1( src1 ));
-   *(p->csr++) = (A2_SRC1( src1 ) |
-		  A2_SRC2( src2 ));
+   *(p->csr++) = (op | A0_DEST(dest) | mask | saturate | A0_SRC0(src0));
+   *(p->csr++) = (A1_SRC0(src0) | A1_SRC1(src1));
+   *(p->csr++) = (A2_SRC1(src1) | A2_SRC2(src2));
 
    p->nr_alu_insn++;
    return dest;
 }
 
 GLuint i915_emit_texld( struct i915_fragment_program *p,
-			  GLuint dest,
-			  GLuint destmask,
-			  GLuint sampler,
-			  GLuint coord,
-			  GLuint op )
+			GLuint dest,
+			GLuint destmask,
+			GLuint sampler,
+			GLuint coord,
+			GLuint op )
 {
-   assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
-   assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
-
-   if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
-      p->nr_tex_indirect++;
+   if (coord != UREG(GET_UREG_TYPE(coord), GET_UREG_NR(coord))) {
+      /* No real way to work around this in the general case - need to
+       * allocate and declare a new temporary register (a utemp won't
+       * do).  Will fallback for now.
+       */
+      i915_program_error(p, "Can't (yet) swizzle TEX arguments");
+      return 0;
    }
 
-   *(p->csr++) = (op | 
-		  T0_DEST( dest ) |
-		  destmask |
-		  T0_SAMPLER( sampler ));
+   /* Don't worry about saturate as we only support  
+    */
+   if (destmask != A0_DEST_CHANNEL_ALL) {
+      GLuint tmp = i915_get_utemp(p);
+      i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, sampler, coord, op );
+      i915_emit_arith( p, A0_MOV, dest, destmask, 0, tmp, 0, 0 );
+      return dest;
+   }
+   else {
+      assert(GET_UREG_TYPE(dest) != REG_TYPE_CONST);
+      assert(dest = UREG(GET_UREG_TYPE(dest), GET_UREG_NR(dest)));
 
-   *(p->csr++) = T1_ADDRESS_REG( coord );
-   *(p->csr++) = T2_MBZ;
+      if (GET_UREG_TYPE(coord) != REG_TYPE_T) {
+	 p->nr_tex_indirect++;
+      }
 
-   p->nr_tex_insn++;
-   return dest;
+      *(p->csr++) = (op | 
+		     T0_DEST( dest ) |
+		     T0_SAMPLER( sampler ));
+
+      *(p->csr++) = T1_ADDRESS_REG( coord );
+      *(p->csr++) = T2_MBZ;
+
+      p->nr_tex_insn++;
+      return dest;
+   }
 }
 
 
-GLuint i915_emit_const1f( struct i915_fragment_program *p, GLfloat c0 )
+GLuint
+i915_emit_const1f(struct i915_fragment_program * p, GLfloat c0)
 {
    GLint reg, idx;
 
-   if (c0 == 0.0) return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
-   if (c0 == 1.0) return swizzle(UREG(REG_TYPE_R, 0), ONE,  ONE,  ONE,  ONE );
+   if (c0 == 0.0)
+      return swizzle(UREG(REG_TYPE_R, 0), ZERO, ZERO, ZERO, ZERO);
+   if (c0 == 1.0)
+      return swizzle(UREG(REG_TYPE_R, 0), ONE, ONE, ONE, ONE);
 
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
       if (p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
-	 continue;
+         continue;
       for (idx = 0; idx < 4; idx++) {
-	 if (!(p->constant_flags[reg] & (1<<idx)) ||
-	     p->constant[reg][idx] == c0) {
-	    p->constant[reg][idx] = c0;
-	    p->constant_flags[reg] |= 1<<idx;
-	    if (reg+1 > p->nr_constants) p->nr_constants = reg+1;
-	    return swizzle(UREG(REG_TYPE_CONST, reg),idx,ZERO,ZERO,ONE);
-	 }
+         if (!(p->constant_flags[reg] & (1 << idx)) ||
+             p->constant[reg][idx] == c0) {
+            p->constant[reg][idx] = c0;
+            p->constant_flags[reg] |= 1 << idx;
+            if (reg + 1 > p->nr_constants)
+               p->nr_constants = reg + 1;
+            return swizzle(UREG(REG_TYPE_CONST, reg), idx, ZERO, ZERO, ONE);
+         }
       }
    }
 
@@ -247,29 +269,35 @@ GLuint i915_emit_const1f( struct i915_fragment_program *p, GLfloat c0 )
    return 0;
 }
 
-GLuint i915_emit_const2f( struct i915_fragment_program *p, 
-			 GLfloat c0, GLfloat c1 )
+GLuint
+i915_emit_const2f(struct i915_fragment_program * p, GLfloat c0, GLfloat c1)
 {
    GLint reg, idx;
 
-   if (c0 == 0.0) return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
-   if (c0 == 1.0) return swizzle(i915_emit_const1f(p, c1), ONE,  X, Z, W); 
+   if (c0 == 0.0)
+      return swizzle(i915_emit_const1f(p, c1), ZERO, X, Z, W);
+   if (c0 == 1.0)
+      return swizzle(i915_emit_const1f(p, c1), ONE, X, Z, W);
 
-   if (c1 == 0.0) return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
-   if (c1 == 1.0) return swizzle(i915_emit_const1f(p, c0), X, ONE,  Z, W);
+   if (c1 == 0.0)
+      return swizzle(i915_emit_const1f(p, c0), X, ZERO, Z, W);
+   if (c1 == 1.0)
+      return swizzle(i915_emit_const1f(p, c0), X, ONE, Z, W);
 
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
       if (p->constant_flags[reg] == 0xf ||
-	  p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
-	 continue;
+          p->constant_flags[reg] == I915_CONSTFLAG_PARAM)
+         continue;
       for (idx = 0; idx < 3; idx++) {
-	 if (!(p->constant_flags[reg] & (3<<idx))) {
-	    p->constant[reg][idx] = c0;
-	    p->constant[reg][idx+1] = c1;
-	    p->constant_flags[reg] |= 3<<idx;
-	    if (reg+1 > p->nr_constants) p->nr_constants = reg+1;
-	    return swizzle(UREG(REG_TYPE_CONST, reg),idx,idx+1,ZERO,ONE);
-	 }
+         if (!(p->constant_flags[reg] & (3 << idx))) {
+            p->constant[reg][idx] = c0;
+            p->constant[reg][idx + 1] = c1;
+            p->constant_flags[reg] |= 3 << idx;
+            if (reg + 1 > p->nr_constants)
+               p->nr_constants = reg + 1;
+            return swizzle(UREG(REG_TYPE_CONST, reg), idx, idx + 1, ZERO,
+                           ONE);
+         }
       }
    }
 
@@ -280,27 +308,28 @@ GLuint i915_emit_const2f( struct i915_fragment_program *p,
 
 
 
-GLuint i915_emit_const4f( struct i915_fragment_program *p, 
-			 GLfloat c0, GLfloat c1, GLfloat c2, GLfloat c3 )
+GLuint
+i915_emit_const4f(struct i915_fragment_program * p,
+                  GLfloat c0, GLfloat c1, GLfloat c2, GLfloat c3)
 {
    GLint reg;
 
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
       if (p->constant_flags[reg] == 0xf &&
-	  p->constant[reg][0] == c0 &&
-	  p->constant[reg][1] == c1 &&
-	  p->constant[reg][2] == c2 &&
-	  p->constant[reg][3] == c3) {
-	 return UREG(REG_TYPE_CONST, reg);
+          p->constant[reg][0] == c0 &&
+          p->constant[reg][1] == c1 &&
+          p->constant[reg][2] == c2 && p->constant[reg][3] == c3) {
+         return UREG(REG_TYPE_CONST, reg);
       }
       else if (p->constant_flags[reg] == 0) {
-	 p->constant[reg][0] = c0;
-	 p->constant[reg][1] = c1;
-	 p->constant[reg][2] = c2;
-	 p->constant[reg][3] = c3;
-	 p->constant_flags[reg] = 0xf;
-	 if (reg+1 > p->nr_constants) p->nr_constants = reg+1;
-	 return UREG(REG_TYPE_CONST, reg);
+         p->constant[reg][0] = c0;
+         p->constant[reg][1] = c1;
+         p->constant[reg][2] = c2;
+         p->constant[reg][3] = c3;
+         p->constant_flags[reg] = 0xf;
+         if (reg + 1 > p->nr_constants)
+            p->nr_constants = reg + 1;
+         return UREG(REG_TYPE_CONST, reg);
       }
    }
 
@@ -310,34 +339,36 @@ GLuint i915_emit_const4f( struct i915_fragment_program *p,
 }
 
 
-GLuint i915_emit_const4fv( struct i915_fragment_program *p, const GLfloat *c )
+GLuint
+i915_emit_const4fv(struct i915_fragment_program * p, const GLfloat * c)
 {
-   return i915_emit_const4f( p, c[0], c[1], c[2], c[3] );
+   return i915_emit_const4f(p, c[0], c[1], c[2], c[3]);
 }
 
 
-GLuint i915_emit_param4fv( struct i915_fragment_program *p, 
-			  const GLfloat *values )
+GLuint
+i915_emit_param4fv(struct i915_fragment_program * p, const GLfloat * values)
 {
    GLint reg, i;
 
    for (i = 0; i < p->nr_params; i++) {
       if (p->param[i].values == values)
-	 return UREG(REG_TYPE_CONST, p->param[i].reg);
+         return UREG(REG_TYPE_CONST, p->param[i].reg);
    }
 
 
    for (reg = 0; reg < I915_MAX_CONSTANT; reg++) {
       if (p->constant_flags[reg] == 0) {
-	 p->constant_flags[reg] = I915_CONSTFLAG_PARAM;
-	 i = p->nr_params++;
+         p->constant_flags[reg] = I915_CONSTFLAG_PARAM;
+         i = p->nr_params++;
 
-	 p->param[i].values = values;
-	 p->param[i].reg = reg;
-	 p->params_uptodate = 0;
+         p->param[i].values = values;
+         p->param[i].reg = reg;
+         p->params_uptodate = 0;
 
-	 if (reg+1 > p->nr_constants) p->nr_constants = reg+1;
-	 return UREG(REG_TYPE_CONST, reg);
+         if (reg + 1 > p->nr_constants)
+            p->nr_constants = reg + 1;
+         return UREG(REG_TYPE_CONST, reg);
       }
    }
 
@@ -349,7 +380,8 @@ GLuint i915_emit_param4fv( struct i915_fragment_program *p,
 
 
 
-void i915_program_error( struct i915_fragment_program *p, const char *msg )
+void
+i915_program_error(struct i915_fragment_program *p, const char *msg)
 {
    /* XXX we shouldn't print anything to stdout, record GL error or
     * call _mesa_problem()
@@ -358,23 +390,24 @@ void i915_program_error( struct i915_fragment_program *p, const char *msg )
    p->error = 1;
 }
 
-void i915_init_program( i915ContextPtr i915, struct i915_fragment_program *p )
+void
+i915_init_program(struct i915_context *i915, struct i915_fragment_program *p)
 {
    GLcontext *ctx = &i915->intel.ctx;
-   TNLcontext *tnl = TNL_CONTEXT( ctx );
-   
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
    p->translated = 0;
    p->params_uptodate = 0;
    p->on_hardware = 0;
    p->error = 0;
 
-   p->nr_tex_indirect = 1;	/* correct? */
+   p->nr_tex_indirect = 1;      /* correct? */
    p->nr_tex_insn = 0;
    p->nr_alu_insn = 0;
    p->nr_decl_insn = 0;
 
-   p->ctx = ctx;  
-   memset( p->constant_flags, 0, sizeof(p->constant_flags) );
+   p->ctx = ctx;
+   memset(p->constant_flags, 0, sizeof(p->constant_flags));
 
    p->nr_constants = 0;
    p->csr = p->program;
@@ -396,12 +429,13 @@ void i915_init_program( i915ContextPtr i915, struct i915_fragment_program *p )
 }
 
 
-void i915_fini_program( struct i915_fragment_program *p )
+void
+i915_fini_program(struct i915_fragment_program *p)
 {
    GLuint program_size = p->csr - p->program;
    GLuint decl_size = p->decl - p->declarations;
-   
-   if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT) 
+
+   if (p->nr_tex_indirect > I915_MAX_TEX_INDIRECT)
       i915_program_error(p, "Exceeded max nr indirect texture lookups");
 
    if (p->nr_tex_insn > I915_MAX_TEX_INSN)
@@ -431,22 +465,24 @@ void i915_fini_program( struct i915_fragment_program *p )
    p->declarations[0] |= program_size + decl_size - 2;
 }
 
-void i915_upload_program( i915ContextPtr i915, struct i915_fragment_program *p )
+void
+i915_upload_program(struct i915_context *i915,
+                    struct i915_fragment_program *p)
 {
    GLuint program_size = p->csr - p->program;
    GLuint decl_size = p->decl - p->declarations;
 
-   FALLBACK( &i915->intel, I915_FALLBACK_PROGRAM, p->error );
+   FALLBACK(&i915->intel, I915_FALLBACK_PROGRAM, p->error);
 
    /* Could just go straight to the batchbuffer from here:
     */
    if (i915->state.ProgramSize != (program_size + decl_size) ||
-       memcmp(i915->state.Program + decl_size, p->program, 
-	      program_size*sizeof(int)) != 0) {
-      I915_STATECHANGE( i915, I915_UPLOAD_PROGRAM );
-      memcpy(i915->state.Program, p->declarations, decl_size*sizeof(int));
+       memcmp(i915->state.Program + decl_size, p->program,
+              program_size * sizeof(int)) != 0) {
+      I915_STATECHANGE(i915, I915_UPLOAD_PROGRAM);
+      memcpy(i915->state.Program, p->declarations, decl_size * sizeof(int));
       memcpy(i915->state.Program + decl_size, p->program,
-	     program_size*sizeof(int));
+             program_size * sizeof(int));
       i915->state.ProgramSize = decl_size + program_size;
    }
 
@@ -455,30 +491,28 @@ void i915_upload_program( i915ContextPtr i915, struct i915_fragment_program *p )
     */
    if (p->nr_constants) {
       GLuint nr = p->nr_constants;
-      
-      I915_ACTIVESTATE( i915, I915_UPLOAD_CONSTANTS, 1 );
-      I915_STATECHANGE( i915, I915_UPLOAD_CONSTANTS );
+
+      I915_ACTIVESTATE(i915, I915_UPLOAD_CONSTANTS, 1);
+      I915_STATECHANGE(i915, I915_UPLOAD_CONSTANTS);
 
       i915->state.Constant[0] = _3DSTATE_PIXEL_SHADER_CONSTANTS | ((nr) * 4);
-      i915->state.Constant[1] = (1<<(nr-1)) | ((1<<(nr-1))-1);
-      
-      memcpy(&i915->state.Constant[2], p->constant, 4*sizeof(int)*(nr));
+      i915->state.Constant[1] = (1 << (nr - 1)) | ((1 << (nr - 1)) - 1);
+
+      memcpy(&i915->state.Constant[2], p->constant, 4 * sizeof(int) * (nr));
       i915->state.ConstantSize = 2 + (nr) * 4;
 
       if (0) {
-	 GLuint i;
-	 for (i = 0; i < nr; i++) {
-	    fprintf(stderr, "const[%d]: %f %f %f %f\n", i, 
-		    p->constant[i][0],
-		    p->constant[i][1],
-		    p->constant[i][2],
-		    p->constant[i][3]);
-	 }
+         GLuint i;
+         for (i = 0; i < nr; i++) {
+            fprintf(stderr, "const[%d]: %f %f %f %f\n", i,
+                    p->constant[i][0],
+                    p->constant[i][1], p->constant[i][2], p->constant[i][3]);
+         }
       }
    }
    else {
-      I915_ACTIVESTATE( i915, I915_UPLOAD_CONSTANTS, 0 );
-   }  
+      I915_ACTIVESTATE(i915, I915_UPLOAD_CONSTANTS, 0);
+   }
 
    p->on_hardware = 1;
 }

src/mesa/drivers/dri/i915/i915_program.h

@@ -48,11 +48,11 @@
 #define UREG_CHANNEL_W_NEGATE_SHIFT   11
 #define UREG_CHANNEL_W_SHIFT          8
 #define UREG_CHANNEL_ZERO_NEGATE_MBZ  5
-#define UREG_CHANNEL_ZERO_SHIFT       4      
+#define UREG_CHANNEL_ZERO_SHIFT       4
 #define UREG_CHANNEL_ONE_NEGATE_MBZ   1
-#define UREG_CHANNEL_ONE_SHIFT        0      
+#define UREG_CHANNEL_ONE_SHIFT        0
 
-#define UREG_BAD          0xffffffff /* not a valid ureg */
+#define UREG_BAD          0xffffffff    /* not a valid ureg */
 
 #define X    SRC_X
 #define Y    SRC_Y
@@ -84,78 +84,75 @@
 
 /* One neat thing about the UREG representation:  
  */
-static __inline int swizzle( int reg, int x, int y, int z, int w )
+static INLINE int
+swizzle(int reg, int x, int y, int z, int w)
 {
    return ((reg & ~UREG_XYZW_CHANNEL_MASK) |
-	   CHANNEL_SRC( GET_CHANNEL_SRC( reg, x ), 0 ) |
-	   CHANNEL_SRC( GET_CHANNEL_SRC( reg, y ), 1 ) |
-	   CHANNEL_SRC( GET_CHANNEL_SRC( reg, z ), 2 ) |
-	   CHANNEL_SRC( GET_CHANNEL_SRC( reg, w ), 3 ));
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, x), 0) |
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, y), 1) |
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, z), 2) |
+           CHANNEL_SRC(GET_CHANNEL_SRC(reg, w), 3));
 }
 
 /* Another neat thing about the UREG representation:  
  */
-static __inline int negate( int reg, int x, int y, int z, int w )
+static INLINE int
+negate(int reg, int x, int y, int z, int w)
 {
-   return reg ^ (((x&1)<<UREG_CHANNEL_X_NEGATE_SHIFT)|
-		 ((y&1)<<UREG_CHANNEL_Y_NEGATE_SHIFT)|
-		 ((z&1)<<UREG_CHANNEL_Z_NEGATE_SHIFT)|
-		 ((w&1)<<UREG_CHANNEL_W_NEGATE_SHIFT));
+   return reg ^ (((x & 1) << UREG_CHANNEL_X_NEGATE_SHIFT) |
+                 ((y & 1) << UREG_CHANNEL_Y_NEGATE_SHIFT) |
+                 ((z & 1) << UREG_CHANNEL_Z_NEGATE_SHIFT) |
+                 ((w & 1) << UREG_CHANNEL_W_NEGATE_SHIFT));
 }
 
 
-extern GLuint i915_get_temp( struct i915_fragment_program *p );
-extern GLuint i915_get_utemp( struct i915_fragment_program *p );
-extern void i915_release_utemps( struct i915_fragment_program *p );
+extern GLuint i915_get_temp(struct i915_fragment_program *p);
+extern GLuint i915_get_utemp(struct i915_fragment_program *p);
+extern void i915_release_utemps(struct i915_fragment_program *p);
 
 
-extern GLuint i915_emit_texld( struct i915_fragment_program *p,
-			      GLuint dest,
-			      GLuint destmask,
-			      GLuint sampler,
-			      GLuint coord,
-			      GLuint op );
+extern GLuint i915_emit_texld(struct i915_fragment_program *p,
+                              GLuint dest,
+                              GLuint destmask,
+                              GLuint sampler, GLuint coord, GLuint op);
 
-extern GLuint i915_emit_arith( struct i915_fragment_program *p,
-			      GLuint op,
-			      GLuint dest,
-			      GLuint mask,
-			      GLuint saturate,
-			      GLuint src0,
-			      GLuint src1,
-			      GLuint src2 );
+extern GLuint i915_emit_arith(struct i915_fragment_program *p,
+                              GLuint op,
+                              GLuint dest,
+                              GLuint mask,
+                              GLuint saturate,
+                              GLuint src0, GLuint src1, GLuint src2);
 
-extern GLuint i915_emit_decl( struct i915_fragment_program *p,
-			     GLuint type, GLuint nr, GLuint d0_flags );
+extern GLuint i915_emit_decl(struct i915_fragment_program *p,
+                             GLuint type, GLuint nr, GLuint d0_flags);
 
 
-extern GLuint i915_emit_const1f( struct i915_fragment_program *p, 
-				GLfloat c0 );
+extern GLuint i915_emit_const1f(struct i915_fragment_program *p, GLfloat c0);
 
-extern GLuint i915_emit_const2f( struct i915_fragment_program *p, 
-				GLfloat c0, GLfloat c1 );
+extern GLuint i915_emit_const2f(struct i915_fragment_program *p,
+                                GLfloat c0, GLfloat c1);
 
-extern GLuint i915_emit_const4fv( struct i915_fragment_program *p,
-				 const GLfloat *c );
+extern GLuint i915_emit_const4fv(struct i915_fragment_program *p,
+                                 const GLfloat * c);
 
-extern GLuint i915_emit_const4f( struct i915_fragment_program *p, 
-				GLfloat c0, GLfloat c1, 
-				GLfloat c2, GLfloat c3 );
+extern GLuint i915_emit_const4f(struct i915_fragment_program *p,
+                                GLfloat c0, GLfloat c1,
+                                GLfloat c2, GLfloat c3);
 
 
-extern GLuint i915_emit_param4fv( struct i915_fragment_program *p, 
-				 const GLfloat *values );
+extern GLuint i915_emit_param4fv(struct i915_fragment_program *p,
+                                 const GLfloat * values);
 
-extern void i915_program_error( struct i915_fragment_program *p,
-                                const char *msg );
+extern void i915_program_error(struct i915_fragment_program *p,
+                               const char *msg);
 
-extern void i915_init_program( i915ContextPtr i915,
-			      struct i915_fragment_program *p );
+extern void i915_init_program(struct i915_context *i915,
+                              struct i915_fragment_program *p);
 
-extern void i915_upload_program( i915ContextPtr i915, 
-				struct i915_fragment_program *p );
+extern void i915_upload_program(struct i915_context *i915,
+                                struct i915_fragment_program *p);
 
-extern void i915_fini_program( struct i915_fragment_program *p );
+extern void i915_fini_program(struct i915_fragment_program *p);
 
 
 

src/mesa/drivers/dri/i915/i915_reg.h

@@ -425,8 +425,10 @@
 #define S7_DEPTH_OFFSET_CONST_MASK     ~0
 
 /* 3DSTATE_MAP_DEINTERLACER_PARAMETERS */
-/* 3DSTATE_MAP_PALETTE_LOAD_32, p206 */
 
+/* 3DSTATE_MAP_PALETTE_LOAD_32, p206 */
+#define _3DSTATE_MAP_PALETTE_LOAD_32    (CMD_3D|(0x1d<<24)|(0x8f<<16))
+/* subsequent dwords up to length (max 16) are ARGB8888 color values */
 
 /* _3DSTATE_MODES_4, p218 */
 #define _3DSTATE_MODES_4_CMD		(CMD_3D|(0x0d<<24))
@@ -435,7 +437,7 @@
 #define LOGICOP_MASK			(0xf<<18)
 #define MODE4_ENABLE_STENCIL_TEST_MASK	((1<<17)|(0xff00))
 #define ENABLE_STENCIL_TEST_MASK	(1<<17)
-#define STENCIL_TEST_MASK(x)		((x)<<8)
+#define STENCIL_TEST_MASK(x)		(((x)&0xff)<<8)
 #define MODE4_ENABLE_STENCIL_WRITE_MASK	((1<<16)|(0x00ff))
 #define ENABLE_STENCIL_WRITE_MASK	(1<<16)
 #define STENCIL_WRITE_MASK(x)		((x)&0xff)
@@ -458,7 +460,7 @@
 
 
 #define I915_MAX_TEX_INDIRECT 4
-#define I915_MAX_TEX_INSN     32     
+#define I915_MAX_TEX_INSN     32
 #define I915_MAX_ALU_INSN     64
 #define I915_MAX_DECL_INSN    27
 #define I915_MAX_TEMPORARY    16
@@ -470,33 +472,33 @@
  */
 #define _3DSTATE_PIXEL_SHADER_PROGRAM    (CMD_3D|(0x1d<<24)|(0x5<<16))
 
-#define REG_TYPE_R                 0 /* temporary regs, no need to
-				      * dcl, must be written before
-				      * read -- Preserved between
-				      * phases. 
-				      */
-#define REG_TYPE_T                 1 /* Interpolated values, must be
-				      * dcl'ed before use.
-				      *
-				      * 0..7: texture coord,
-				      * 8: diffuse spec,
-				      * 9: specular color,
-				      * 10: fog parameter in w.
-				      */
-#define REG_TYPE_CONST             2 /* Restriction: only one const
-				      * can be referenced per
-				      * instruction, though it may be
-				      * selected for multiple inputs.
-				      * Constants not initialized
-				      * default to zero.
-				      */
-#define REG_TYPE_S                 3 /* sampler */
-#define REG_TYPE_OC                4 /* output color (rgba) */
-#define REG_TYPE_OD                5 /* output depth (w), xyz are
-				      * temporaries.  If not written,
-				      * interpolated depth is used?
-				      */
-#define REG_TYPE_U                 6 /* unpreserved temporaries */
+#define REG_TYPE_R                 0    /* temporary regs, no need to
+                                         * dcl, must be written before
+                                         * read -- Preserved between
+                                         * phases. 
+                                         */
+#define REG_TYPE_T                 1    /* Interpolated values, must be
+                                         * dcl'ed before use.
+                                         *
+                                         * 0..7: texture coord,
+                                         * 8: diffuse spec,
+                                         * 9: specular color,
+                                         * 10: fog parameter in w.
+                                         */
+#define REG_TYPE_CONST             2    /* Restriction: only one const
+                                         * can be referenced per
+                                         * instruction, though it may be
+                                         * selected for multiple inputs.
+                                         * Constants not initialized
+                                         * default to zero.
+                                         */
+#define REG_TYPE_S                 3    /* sampler */
+#define REG_TYPE_OC                4    /* output color (rgba) */
+#define REG_TYPE_OD                5    /* output depth (w), xyz are
+                                         * temporaries.  If not written,
+                                         * interpolated depth is used?
+                                         */
+#define REG_TYPE_U                 6    /* unpreserved temporaries */
 #define REG_TYPE_MASK              0x7
 #define REG_NR_MASK                0xf
 
@@ -513,34 +515,34 @@
 #define T_TEX7     7
 #define T_DIFFUSE  8
 #define T_SPECULAR 9
-#define T_FOG_W    10		/* interpolated fog is in W coord */
+#define T_FOG_W    10           /* interpolated fog is in W coord */
 
 /* Arithmetic instructions */
 
 /* .replicate_swizzle == selection and replication of a particular
  * scalar channel, ie., .xxxx, .yyyy, .zzzz or .wwww 
  */
-#define A0_NOP    (0x0<<24)		/* no operation */
-#define A0_ADD    (0x1<<24)		/* dst = src0 + src1 */
-#define A0_MOV    (0x2<<24)		/* dst = src0 */
-#define A0_MUL    (0x3<<24)		/* dst = src0 * src1 */
-#define A0_MAD    (0x4<<24)		/* dst = src0 * src1 + src2 */
-#define A0_DP2ADD (0x5<<24)		/* dst.xyzw = src0.xy dot src1.xy + src2.replicate_swizzle */
-#define A0_DP3    (0x6<<24)		/* dst.xyzw = src0.xyz dot src1.xyz */
-#define A0_DP4    (0x7<<24)		/* dst.xyzw = src0.xyzw dot src1.xyzw */
-#define A0_FRC    (0x8<<24)		/* dst = src0 - floor(src0) */
-#define A0_RCP    (0x9<<24)		/* dst.xyzw = 1/(src0.replicate_swizzle) */
-#define A0_RSQ    (0xa<<24)		/* dst.xyzw = 1/(sqrt(abs(src0.replicate_swizzle))) */
-#define A0_EXP    (0xb<<24)		/* dst.xyzw = exp2(src0.replicate_swizzle) */
-#define A0_LOG    (0xc<<24)		/* dst.xyzw = log2(abs(src0.replicate_swizzle)) */
-#define A0_CMP    (0xd<<24)		/* dst = (src0 >= 0.0) ? src1 : src2 */
-#define A0_MIN    (0xe<<24)		/* dst = (src0 < src1) ? src0 : src1 */
-#define A0_MAX    (0xf<<24)		/* dst = (src0 >= src1) ? src0 : src1 */
-#define A0_FLR    (0x10<<24)		/* dst = floor(src0) */
-#define A0_MOD    (0x11<<24)		/* dst = src0 fmod 1.0 */
-#define A0_TRC    (0x12<<24)		/* dst = int(src0) */
-#define A0_SGE    (0x13<<24)		/* dst = src0 >= src1 ? 1.0 : 0.0 */
-#define A0_SLT    (0x14<<24)		/* dst = src0 < src1 ? 1.0 : 0.0 */
+#define A0_NOP    (0x0<<24)     /* no operation */
+#define A0_ADD    (0x1<<24)     /* dst = src0 + src1 */
+#define A0_MOV    (0x2<<24)     /* dst = src0 */
+#define A0_MUL    (0x3<<24)     /* dst = src0 * src1 */
+#define A0_MAD    (0x4<<24)     /* dst = src0 * src1 + src2 */
+#define A0_DP2ADD (0x5<<24)     /* dst.xyzw = src0.xy dot src1.xy + src2.replicate_swizzle */
+#define A0_DP3    (0x6<<24)     /* dst.xyzw = src0.xyz dot src1.xyz */
+#define A0_DP4    (0x7<<24)     /* dst.xyzw = src0.xyzw dot src1.xyzw */
+#define A0_FRC    (0x8<<24)     /* dst = src0 - floor(src0) */
+#define A0_RCP    (0x9<<24)     /* dst.xyzw = 1/(src0.replicate_swizzle) */
+#define A0_RSQ    (0xa<<24)     /* dst.xyzw = 1/(sqrt(abs(src0.replicate_swizzle))) */
+#define A0_EXP    (0xb<<24)     /* dst.xyzw = exp2(src0.replicate_swizzle) */
+#define A0_LOG    (0xc<<24)     /* dst.xyzw = log2(abs(src0.replicate_swizzle)) */
+#define A0_CMP    (0xd<<24)     /* dst = (src0 >= 0.0) ? src1 : src2 */
+#define A0_MIN    (0xe<<24)     /* dst = (src0 < src1) ? src0 : src1 */
+#define A0_MAX    (0xf<<24)     /* dst = (src0 >= src1) ? src0 : src1 */
+#define A0_FLR    (0x10<<24)    /* dst = floor(src0) */
+#define A0_MOD    (0x11<<24)    /* dst = src0 fmod 1.0 */
+#define A0_TRC    (0x12<<24)    /* dst = int(src0) */
+#define A0_SGE    (0x13<<24)    /* dst = src0 >= src1 ? 1.0 : 0.0 */
+#define A0_SLT    (0x14<<24)    /* dst = src0 < src1 ? 1.0 : 0.0 */
 #define A0_DEST_SATURATE                 (1<<22)
 #define A0_DEST_TYPE_SHIFT                19
 /* Allow: R, OC, OD, U */
@@ -599,23 +601,23 @@
 
 
 /* Texture instructions */
-#define T0_TEXLD     (0x15<<24)	/* Sample texture using predeclared
-				 * sampler and address, and output
-				 * filtered texel data to destination
-				 * register */
-#define T0_TEXLDP    (0x16<<24)	/* Same as texld but performs a
-				 * perspective divide of the texture
-				 * coordinate .xyz values by .w before
-				 * sampling. */
-#define T0_TEXLDB    (0x17<<24)	/* Same as texld but biases the
-				 * computed LOD by w.  Only S4.6 two's
-				 * comp is used.  This implies that a
-				 * float to fixed conversion is
-				 * done. */
-#define T0_TEXKILL   (0x18<<24)	/* Does not perform a sampling
-				 * operation.  Simply kills the pixel
-				 * if any channel of the address
-				 * register is < 0.0. */
+#define T0_TEXLD     (0x15<<24) /* Sample texture using predeclared
+                                 * sampler and address, and output
+                                 * filtered texel data to destination
+                                 * register */
+#define T0_TEXLDP    (0x16<<24) /* Same as texld but performs a
+                                 * perspective divide of the texture
+                                 * coordinate .xyz values by .w before
+                                 * sampling. */
+#define T0_TEXLDB    (0x17<<24) /* Same as texld but biases the
+                                 * computed LOD by w.  Only S4.6 two's
+                                 * comp is used.  This implies that a
+                                 * float to fixed conversion is
+                                 * done. */
+#define T0_TEXKILL   (0x18<<24) /* Does not perform a sampling
+                                 * operation.  Simply kills the pixel
+                                 * if any channel of the address
+                                 * register is < 0.0. */
 #define T0_DEST_TYPE_SHIFT                19
 /* Allow: R, OC, OD, U */
 /* Note: U (unpreserved) regs do not retain their values between
@@ -627,18 +629,18 @@
  */
 #define T0_DEST_NR_SHIFT                 14
 /* Allow R: 0..15, OC,OD: 0..0, U: 0..2 */
-#define T0_SAMPLER_NR_SHIFT              0 /* This field ignored for TEXKILL */
+#define T0_SAMPLER_NR_SHIFT              0      /* This field ignored for TEXKILL */
 #define T0_SAMPLER_NR_MASK               (0xf<<0)
 
-#define T1_ADDRESS_REG_TYPE_SHIFT        24 /* Reg to use as texture coord */
+#define T1_ADDRESS_REG_TYPE_SHIFT        24     /* Reg to use as texture coord */
 /* Allow R, T, OC, OD -- R, OC, OD are 'dependent' reads, new program phase */
 #define T1_ADDRESS_REG_NR_SHIFT          17
 #define T2_MBZ                           0
 
 /* Declaration instructions */
-#define D0_DCL       (0x19<<24)	/* Declare a t (interpolated attrib)
-				 * register or an s (sampler)
-				 * register. */
+#define D0_DCL       (0x19<<24) /* Declare a t (interpolated attrib)
+                                 * register or an s (sampler)
+                                 * register. */
 #define D0_SAMPLE_TYPE_SHIFT              22
 #define D0_SAMPLE_TYPE_2D                 (0x0<<22)
 #define D0_SAMPLE_TYPE_CUBE               (0x1<<22)
@@ -695,12 +697,12 @@
 #define    MAPSURF_4BIT_INDEXED		   (7<<7)
 #define MS3_MT_FORMAT_MASK         (0x7 << 3)
 #define MS3_MT_FORMAT_SHIFT        3
-#define    MT_4BIT_IDX_ARGB8888	           (7<<3) /* SURFACE_4BIT_INDEXED */
-#define    MT_8BIT_I8		           (0<<3) /* SURFACE_8BIT */
+#define    MT_4BIT_IDX_ARGB8888	           (7<<3)       /* SURFACE_4BIT_INDEXED */
+#define    MT_8BIT_I8		           (0<<3)       /* SURFACE_8BIT */
 #define    MT_8BIT_L8		           (1<<3)
 #define    MT_8BIT_A8		           (4<<3)
 #define    MT_8BIT_MONO8	           (5<<3)
-#define    MT_16BIT_RGB565 		   (0<<3) /* SURFACE_16BIT */
+#define    MT_16BIT_RGB565 		   (0<<3)       /* SURFACE_16BIT */
 #define    MT_16BIT_ARGB1555		   (1<<3)
 #define    MT_16BIT_ARGB4444		   (2<<3)
 #define    MT_16BIT_AY88		   (3<<3)
@@ -709,7 +711,7 @@
 #define    MT_16BIT_I16	                   (7<<3)
 #define    MT_16BIT_L16	                   (8<<3)
 #define    MT_16BIT_A16	                   (9<<3)
-#define    MT_32BIT_ARGB8888		   (0<<3) /* SURFACE_32BIT */
+#define    MT_32BIT_ARGB8888		   (0<<3)       /* SURFACE_32BIT */
 #define    MT_32BIT_ABGR8888		   (1<<3)
 #define    MT_32BIT_XRGB8888		   (2<<3)
 #define    MT_32BIT_XBGR8888		   (3<<3)
@@ -725,11 +727,11 @@
 #define    MT_32BIT_xI824	           (0xD<<3)
 #define    MT_32BIT_xA824	           (0xE<<3)
 #define    MT_32BIT_xL824	           (0xF<<3)
-#define    MT_422_YCRCB_SWAPY	           (0<<3) /* SURFACE_422 */
+#define    MT_422_YCRCB_SWAPY	           (0<<3)       /* SURFACE_422 */
 #define    MT_422_YCRCB_NORMAL	           (1<<3)
 #define    MT_422_YCRCB_SWAPUV	           (2<<3)
 #define    MT_422_YCRCB_SWAPUVY	           (3<<3)
-#define    MT_COMPRESS_DXT1		   (0<<3) /* SURFACE_COMPRESSED */
+#define    MT_COMPRESS_DXT1		   (0<<3)       /* SURFACE_COMPRESSED */
 #define    MT_COMPRESS_DXT2_3	           (1<<3)
 #define    MT_COMPRESS_DXT4_5	           (2<<3)
 #define    MT_COMPRESS_FXT1		   (3<<3)
@@ -751,7 +753,7 @@
 #define MS4_MIP_LAYOUT_LEGACY           (0<<8)
 #define MS4_MIP_LAYOUT_BELOW_LPT        (0<<8)
 #define MS4_MIP_LAYOUT_RIGHT_LPT        (1<<8)
-#define MS4_VOLUME_DEPTH_SHIFT          0    
+#define MS4_VOLUME_DEPTH_SHIFT          0
 #define MS4_VOLUME_DEPTH_MASK           (0xff<<0)
 
 /* p244 */
@@ -779,7 +781,7 @@
 #define   FILTER_4X4_1    	3
 #define   FILTER_4X4_2    	4
 #define   FILTER_4X4_FLAT 	5
-#define   FILTER_6X5_MONO   	6 /* XXX - check */
+#define   FILTER_6X5_MONO   	6       /* XXX - check */
 #define SS2_MIN_FILTER_SHIFT          14
 #define SS2_MIN_FILTER_MASK           (0x7<<14)
 #define SS2_LOD_BIAS_SHIFT            5
@@ -826,10 +828,14 @@
 #define ST1_ENABLE               (1<<16)
 #define ST1_MASK                 (0xffff)
 
+#define _3DSTATE_DEFAULT_Z          ((0x3<<29)|(0x1d<<24)|(0x98<<16))
+#define _3DSTATE_DEFAULT_DIFFUSE    ((0x3<<29)|(0x1d<<24)|(0x99<<16))
+#define _3DSTATE_DEFAULT_SPECULAR   ((0x3<<29)|(0x1d<<24)|(0x9a<<16))
 
-#define MI_FLUSH           ((0<<29)|(4<<23))
-#define FLUSH_MAP_CACHE    (1<<0)
-#define FLUSH_RENDER_CACHE (1<<1)
+
+#define MI_FLUSH                   ((0<<29)|(4<<23))
+#define FLUSH_MAP_CACHE            (1<<0)
+#define INHIBIT_FLUSH_RENDER_CACHE (1<<2)
 
 
 #endif

src/mesa/drivers/dri/i915/i915_tex.c

@@ -45,79 +45,14 @@
 
 
 
-
-
-
-/**
- * Allocate space for and load the mesa images into the texture memory block.
- * This will happen before drawing with a new texture, or drawing with a
- * texture after it was swapped out or teximaged again.
- */
-
-intelTextureObjectPtr i915AllocTexObj( struct gl_texture_object *texObj )
+static void
+i915TexEnv(GLcontext * ctx, GLenum target,
+           GLenum pname, const GLfloat * param)
 {
-   i915TextureObjectPtr t = CALLOC_STRUCT( i915_texture_object );
-   if ( !t ) 
-      return NULL;
-
-   texObj->DriverData = t;
-   t->intel.base.tObj = texObj;
-   t->intel.dirty = I915_UPLOAD_TEX_ALL;
-   make_empty_list( &t->intel.base );
-   return &t->intel;
-}
-
-
-static void i915TexParameter( GLcontext *ctx, GLenum target,
-			     struct gl_texture_object *tObj,
-			     GLenum pname, const GLfloat *params )
-{
-   i915TextureObjectPtr t = (i915TextureObjectPtr) tObj->DriverData;
- 
-   switch (pname) {
-   case GL_TEXTURE_MIN_FILTER:
-   case GL_TEXTURE_MAG_FILTER:
-   case GL_TEXTURE_MAX_ANISOTROPY_EXT:
-   case GL_TEXTURE_WRAP_S:
-   case GL_TEXTURE_WRAP_T:
-   case GL_TEXTURE_WRAP_R:
-   case GL_TEXTURE_BORDER_COLOR:
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-
-   case GL_TEXTURE_COMPARE_MODE:
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-   case GL_TEXTURE_COMPARE_FUNC:
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-
-   case GL_TEXTURE_BASE_LEVEL:
-   case GL_TEXTURE_MAX_LEVEL:
-   case GL_TEXTURE_MIN_LOD:
-   case GL_TEXTURE_MAX_LOD:
-      /* The i915 and its successors can do a lot of this without
-       * reloading the textures.  A project for someone?
-       */
-      intelFlush( ctx );
-      driSwapOutTextureObject( (driTextureObject *) t );
-      t->intel.dirty = I915_UPLOAD_TEX_ALL;
-      break;
-
-   default:
-      return;
-   }
-}
-
-
-static void i915TexEnv( GLcontext *ctx, GLenum target, 
-			GLenum pname, const GLfloat *param )
-{
-   i915ContextPtr i915 = I915_CONTEXT( ctx );
-   GLuint unit = ctx->Texture.CurrentUnit;
+   struct i915_context *i915 = I915_CONTEXT(ctx);
 
    switch (pname) {
-   case GL_TEXTURE_ENV_COLOR: 	/* Should be a tracked param */
+   case GL_TEXTURE_ENV_COLOR:  /* Should be a tracked param */
    case GL_TEXTURE_ENV_MODE:
    case GL_COMBINE_RGB:
    case GL_COMBINE_ALPHA:
@@ -135,19 +70,21 @@ static void i915TexEnv( GLcontext *ctx, GLenum target,
    case GL_OPERAND2_ALPHA:
    case GL_RGB_SCALE:
    case GL_ALPHA_SCALE:
-      i915->tex_program.translated = 0; 
+      i915->tex_program.translated = 0;
       break;
 
-   case GL_TEXTURE_LOD_BIAS: {
-      int b = (int) ((*param) * 16.0);
-      if (b > 255) b = 255;
-      if (b < -256) b = -256;
-      I915_STATECHANGE(i915, I915_UPLOAD_TEX(unit));
-      i915->state.Tex[unit][I915_TEXREG_SS2] &= ~SS2_LOD_BIAS_MASK;
-      i915->state.Tex[unit][I915_TEXREG_SS2] |= 
-	 ((b << SS2_LOD_BIAS_SHIFT) & SS2_LOD_BIAS_MASK);
-      break;
-   }
+   case GL_TEXTURE_LOD_BIAS:{
+         GLuint unit = ctx->Texture.CurrentUnit;
+         GLint b = (int) ((*param) * 16.0);
+         if (b > 255)
+            b = 255;
+         if (b < -256)
+            b = -256;
+         I915_STATECHANGE(i915, I915_UPLOAD_TEX(unit));
+         i915->lodbias_ss2[unit] =
+            ((b << SS2_LOD_BIAS_SHIFT) & SS2_LOD_BIAS_MASK);
+         break;
+      }
 
    default:
       break;
@@ -155,33 +92,22 @@ static void i915TexEnv( GLcontext *ctx, GLenum target,
 }
 
 
-static void i915BindTexture( GLcontext *ctx, GLenum target,
-			    struct gl_texture_object *texObj )
+static void
+i915BindTexture(GLcontext * ctx, GLenum target,
+                struct gl_texture_object *texobj)
 {
-   i915TextureObjectPtr tex;
-   
-   if (!texObj->DriverData)
-      i915AllocTexObj( texObj );
-   
-   tex = (i915TextureObjectPtr)texObj->DriverData;
-
-   if (tex->lastTarget != texObj->Target) {
-      tex->intel.dirty = I915_UPLOAD_TEX_ALL;
-      tex->lastTarget = texObj->Target;
-   }
-
    /* Need this if image format changes between bound textures.
     * Could try and shortcircuit by checking for differences in
     * state between incoming and outgoing textures:
     */
-   I915_CONTEXT(ctx)->tex_program.translated = 0; 
+   I915_CONTEXT(ctx)->tex_program.translated = 0;
 }
 
 
 
-void i915InitTextureFuncs( struct dd_function_table *functions )
+void
+i915InitTextureFuncs(struct dd_function_table *functions)
 {
    functions->BindTexture = i915BindTexture;
    functions->TexEnv = i915TexEnv;
-   functions->TexParameter = i915TexParameter;
 }

src/mesa/drivers/dri/i915/i915_tex_layout.c

@@ -0,0 +1,380 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Code to layout images in a mipmap tree for i915 and i945
+ * respectively.
+ */
+
+#include "intel_mipmap_tree.h"
+#include "macros.h"
+#include "intel_context.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+static GLint initial_offsets[6][2] = { {0, 0},
+{0, 2},
+{1, 0},
+{1, 2},
+{1, 1},
+{1, 3}
+};
+
+
+static GLint step_offsets[6][2] = { {0, 2},
+{0, 2},
+{-1, 2},
+{-1, 2},
+{-1, 1},
+{-1, 1}
+};
+
+static GLuint
+minify(GLuint d)
+{
+   return MAX2(1, d >> 1);
+}
+
+GLboolean
+i915_miptree_layout(struct intel_mipmap_tree * mt)
+{
+   GLint level;
+
+   switch (mt->target) {
+   case GL_TEXTURE_CUBE_MAP:{
+         const GLuint dim = mt->width0;
+         GLuint face;
+
+         /* double pitch for cube layouts */
+         mt->pitch = ((dim * mt->cpp * 2 + 3) & ~3) / mt->cpp;
+         mt->total_height = dim * 4;
+
+         for (level = mt->first_level; level <= mt->last_level; level++)
+            intel_miptree_set_level_info(mt, level, 6,
+                                         0, 0,
+                                         mt->pitch, mt->total_height, 1);
+
+         for (face = 0; face < 6; face++) {
+            GLuint x = initial_offsets[face][0] * dim;
+            GLuint y = initial_offsets[face][1] * dim;
+            GLuint d = dim;
+
+            for (level = mt->first_level; level <= mt->last_level; level++) {
+               intel_miptree_set_image_offset(mt, level, face, x, y);
+
+               if (d == 0)
+                  _mesa_printf("cube mipmap %d/%d (%d..%d) is 0x0\n",
+                               face, level, mt->first_level, mt->last_level);
+
+               d >>= 1;
+               x += step_offsets[face][0] * d;
+               y += step_offsets[face][1] * d;
+            }
+         }
+         break;
+      }
+   case GL_TEXTURE_3D:{
+         GLuint width = mt->width0;
+         GLuint height = mt->height0;
+         GLuint depth = mt->depth0;
+         GLuint stack_height = 0;
+
+         /* Calculate the size of a single slice. 
+          */
+         mt->pitch = ((mt->width0 * mt->cpp + 3) & ~3) / mt->cpp;
+
+         /* XXX: hardware expects/requires 9 levels at minimum.
+          */
+         for (level = mt->first_level; level <= MAX2(8, mt->last_level);
+              level++) {
+            intel_miptree_set_level_info(mt, level, 1, 0, mt->total_height,
+                                         width, height, depth);
+
+
+            stack_height += MAX2(2, height);
+
+            width = minify(width);
+            height = minify(height);
+            depth = minify(depth);
+         }
+
+         /* Fixup depth image_offsets: 
+          */
+         depth = mt->depth0;
+         for (level = mt->first_level; level <= mt->last_level; level++) {
+            GLuint i;
+            for (i = 0; i < depth; i++) 
+               intel_miptree_set_image_offset(mt, level, i,
+                                              0, i * stack_height);
+
+            depth = minify(depth);
+         }
+
+
+         /* Multiply slice size by texture depth for total size.  It's
+          * remarkable how wasteful of memory the i915 texture layouts
+          * are.  They are largely fixed in the i945.
+          */
+         mt->total_height = stack_height * mt->depth0;
+         break;
+      }
+
+   default:{
+         GLuint width = mt->width0;
+         GLuint height = mt->height0;
+	 GLuint img_height;
+
+         mt->pitch = ((mt->width0 * mt->cpp + 3) & ~3) / mt->cpp;
+         mt->total_height = 0;
+
+         for (level = mt->first_level; level <= mt->last_level; level++) {
+            intel_miptree_set_level_info(mt, level, 1,
+                                         0, mt->total_height,
+                                         width, height, 1);
+
+            if (mt->compressed)
+               img_height = MAX2(1, height / 4);
+            else
+               img_height = MAX2(2, height);
+
+	    mt->total_height += img_height;
+	    mt->total_height += 1;
+	    mt->total_height &= ~1;
+
+            width = minify(width);
+            height = minify(height);
+         }
+         break;
+      }
+   }
+   DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__,
+       mt->pitch,
+       mt->total_height, mt->cpp, mt->pitch * mt->total_height * mt->cpp);
+
+   return GL_TRUE;
+}
+
+
+GLboolean
+i945_miptree_layout(struct intel_mipmap_tree * mt)
+{
+   GLint level;
+
+   switch (mt->target) {
+   case GL_TEXTURE_CUBE_MAP:{
+         const GLuint dim = mt->width0;
+         GLuint face;
+
+         /* Depending on the size of the largest images, pitch can be
+          * determined either by the old-style packing of cubemap faces,
+          * or the final row of 4x4, 2x2 and 1x1 faces below this. 
+          */
+         if (dim > 32)
+            mt->pitch = ((dim * mt->cpp * 2 + 3) & ~3) / mt->cpp;
+         else
+            mt->pitch = 14 * 8;
+
+         mt->total_height = dim * 4 + 4;
+
+         /* Set all the levels to effectively occupy the whole rectangular region. 
+          */
+         for (level = mt->first_level; level <= mt->last_level; level++)
+            intel_miptree_set_level_info(mt, level, 6,
+                                         0, 0,
+                                         mt->pitch, mt->total_height, 1);
+
+
+
+         for (face = 0; face < 6; face++) {
+            GLuint x = initial_offsets[face][0] * dim;
+            GLuint y = initial_offsets[face][1] * dim;
+            GLuint d = dim;
+
+            if (dim == 4 && face >= 4) {
+               y = mt->total_height - 4;
+               x = (face - 4) * 8;
+            }
+            else if (dim < 4) {
+               y = mt->total_height - 4;
+               x = face * 8;
+            }
+
+            for (level = mt->first_level; level <= mt->last_level; level++) {
+               intel_miptree_set_image_offset(mt, level, face, x, y);
+
+               d >>= 1;
+
+               switch (d) {
+               case 4:
+                  switch (face) {
+                  case FACE_POS_X:
+                  case FACE_NEG_X:
+                     x += step_offsets[face][0] * d;
+                     y += step_offsets[face][1] * d;
+                     break;
+                  case FACE_POS_Y:
+                  case FACE_NEG_Y:
+                     y += 12;
+                     x -= 8;
+                     break;
+                  case FACE_POS_Z:
+                  case FACE_NEG_Z:
+                     y = mt->total_height - 4;
+                     x = (face - 4) * 8;
+                     break;
+                  }
+
+               case 2:
+                  y = mt->total_height - 4;
+                  x = 16 + face * 8;
+                  break;
+
+               case 1:
+                  x += 48;
+                  break;
+
+               default:
+                  x += step_offsets[face][0] * d;
+                  y += step_offsets[face][1] * d;
+                  break;
+               }
+            }
+         }
+         break;
+      }
+   case GL_TEXTURE_3D:{
+         GLuint width = mt->width0;
+         GLuint height = mt->height0;
+         GLuint depth = mt->depth0;
+         GLuint pack_x_pitch, pack_x_nr;
+         GLuint pack_y_pitch;
+         GLuint level;
+
+         mt->pitch = ((mt->width0 * mt->cpp + 3) & ~3) / mt->cpp;
+         mt->total_height = 0;
+
+         pack_y_pitch = MAX2(mt->height0, 2);
+         pack_x_pitch = mt->pitch;
+         pack_x_nr = 1;
+
+         for (level = mt->first_level; level <= mt->last_level; level++) {
+            GLuint nr_images = mt->target == GL_TEXTURE_3D ? depth : 6;
+            GLint x = 0;
+            GLint y = 0;
+            GLint q, j;
+
+            intel_miptree_set_level_info(mt, level, nr_images,
+                                         0, mt->total_height,
+                                         width, height, depth);
+
+            for (q = 0; q < nr_images;) {
+               for (j = 0; j < pack_x_nr && q < nr_images; j++, q++) {
+                  intel_miptree_set_image_offset(mt, level, q, x, y);
+                  x += pack_x_pitch;
+               }
+
+               x = 0;
+               y += pack_y_pitch;
+            }
+
+
+            mt->total_height += y;
+
+            if (pack_x_pitch > 4) {
+               pack_x_pitch >>= 1;
+               pack_x_nr <<= 1;
+               assert(pack_x_pitch * pack_x_nr <= mt->pitch);
+            }
+
+            if (pack_y_pitch > 2) {
+               pack_y_pitch >>= 1;
+            }
+
+            width = minify(width);
+            height = minify(height);
+            depth = minify(depth);
+         }
+         break;
+      }
+
+   case GL_TEXTURE_1D:
+   case GL_TEXTURE_2D:
+   case GL_TEXTURE_RECTANGLE_ARB:{
+         GLuint x = 0;
+         GLuint y = 0;
+         GLuint width = mt->width0;
+         GLuint height = mt->height0;
+	 GLint align_h = 2;
+
+         mt->pitch = ((mt->width0 * mt->cpp + 3) & ~3) / mt->cpp;
+         mt->total_height = 0;
+
+         for (level = mt->first_level; level <= mt->last_level; level++) {
+	    GLuint img_height;
+
+            intel_miptree_set_level_info(mt, level, 1,
+                                         x, y, 
+					 width, 
+					 mt->compressed ? height/4 : height, 1);
+
+
+	    if (mt->compressed)
+               img_height = MAX2(1, height / 4);
+            else
+               img_height = MAX2(align_h, height);
+
+            /* LPT change: step right after second mipmap.
+             */
+            if (level == mt->first_level + 1) {
+               x += mt->pitch / 2;
+	       x = (x + 3) & ~3;
+	    }
+            else {
+	       y += img_height;
+	       y += align_h - 1;
+	       y &= ~(align_h - 1);
+	    }
+
+            /* Because the images are packed better, the final offset
+             * might not be the maximal one:
+             */
+            mt->total_height = MAX2(mt->total_height, y);
+
+            width = minify(width);
+            height = minify(height);
+         }
+         break;
+      }
+   default:
+      _mesa_problem(NULL, "Unexpected tex target in i945_miptree_layout()");
+   }
+
+   DBG("%s: %dx%dx%d - sz 0x%x\n", __FUNCTION__,
+       mt->pitch,
+       mt->total_height, mt->cpp, mt->pitch * mt->total_height * mt->cpp);
+
+   return GL_TRUE;
+}

src/mesa/drivers/dri/i915/i915_texprog.c

@@ -1,671 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
-#include <strings.h>
-
-#include "glheader.h"
-#include "macros.h"
-#include "enums.h"
-
-#include "tnl/t_context.h"
-#include "intel_batchbuffer.h"
-
-#include "i915_reg.h"
-#include "i915_context.h"
-#include "i915_program.h"
-
-static GLuint translate_tex_src_bit( struct i915_fragment_program *p,
-				     GLubyte bit )
-{
-   switch (bit) {
-   case TEXTURE_1D_BIT:   return D0_SAMPLE_TYPE_2D;
-   case TEXTURE_2D_BIT:   return D0_SAMPLE_TYPE_2D;
-   case TEXTURE_RECT_BIT: return D0_SAMPLE_TYPE_2D;
-   case TEXTURE_3D_BIT:   return D0_SAMPLE_TYPE_VOLUME;
-   case TEXTURE_CUBE_BIT: return D0_SAMPLE_TYPE_CUBE;
-   default: i915_program_error(p, "TexSrcBit"); return 0;
-   }
-}
-
-static GLuint get_source( struct i915_fragment_program *p, 
-			  GLenum src, GLuint unit )
-{
-   switch (src) {
-   case GL_TEXTURE: 
-      if (p->src_texture == UREG_BAD) {
-
-	 /* TODO: Use D0_CHANNEL_XY where possible.
-	  */
-	 GLuint dim = translate_tex_src_bit( p, p->ctx->Texture.Unit[unit]._ReallyEnabled);
-	 GLuint sampler = i915_emit_decl(p, REG_TYPE_S, unit, dim);
-	 GLuint texcoord = i915_emit_decl(p, REG_TYPE_T, unit, D0_CHANNEL_ALL);
-	 GLuint tmp = i915_get_temp( p );
-	 GLuint op = T0_TEXLD;
-
-	 if (p->VB->TexCoordPtr[unit]->size == 4)
-	    op = T0_TEXLDP;
-
-	 p->src_texture = i915_emit_texld( p, tmp, A0_DEST_CHANNEL_ALL, 
-					  sampler, texcoord, op );
-      }
-
-      return p->src_texture;
-
-      /* Crossbar: */
-   case GL_TEXTURE0:
-   case GL_TEXTURE1:
-   case GL_TEXTURE2:
-   case GL_TEXTURE3:
-   case GL_TEXTURE4:
-   case GL_TEXTURE5:
-   case GL_TEXTURE6:
-   case GL_TEXTURE7: {
-      return UREG_BAD;
-   }
-
-   case GL_CONSTANT:
-      return i915_emit_const4fv( p, p->ctx->Texture.Unit[unit].EnvColor );
-   case GL_PRIMARY_COLOR:
-      return i915_emit_decl(p, REG_TYPE_T, T_DIFFUSE, D0_CHANNEL_ALL);
-   case GL_PREVIOUS:
-   default: 
-      i915_emit_decl(p, 
-		GET_UREG_TYPE(p->src_previous),
-		GET_UREG_NR(p->src_previous), D0_CHANNEL_ALL); 
-      return p->src_previous;
-   }
-}
-			
-
-static GLuint emit_combine_source( struct i915_fragment_program *p, 
-				   GLuint mask,
-				   GLuint unit,
-				   GLenum source, 
-				   GLenum operand )
-{
-   GLuint arg, src;
-
-   src = get_source(p, source, unit);
-
-   switch (operand) {
-   case GL_ONE_MINUS_SRC_COLOR: 
-      /* Get unused tmp,
-       * Emit tmp = 1.0 + arg.-x-y-z-w
-       */
-      arg = i915_get_temp( p );
-      return i915_emit_arith( p, A0_ADD, arg, mask, 0,
-		  swizzle(src, ONE, ONE, ONE, ONE ),
-		  negate(src, 1,1,1,1), 0);
-
-   case GL_SRC_ALPHA: 
-      if (mask == A0_DEST_CHANNEL_W)
-	 return src;
-      else
-	 return swizzle( src, W, W, W, W );
-   case GL_ONE_MINUS_SRC_ALPHA: 
-      /* Get unused tmp,
-       * Emit tmp = 1.0 + arg.-w-w-w-w
-       */
-      arg = i915_get_temp( p );
-      return i915_emit_arith( p, A0_ADD, arg, mask, 0,
-			 swizzle(src, ONE, ONE, ONE, ONE ),
-			 negate( swizzle(src,W,W,W,W), 1,1,1,1), 0);
-   case GL_SRC_COLOR: 
-   default:
-      return src;
-   }
-}
-
-
-
-static int nr_args( GLenum mode )
-{
-   switch (mode) {
-   case GL_REPLACE: return 1; 
-   case GL_MODULATE: return 2;
-   case GL_ADD: return 2;
-   case GL_ADD_SIGNED: return 2;
-   case GL_INTERPOLATE:	return 3;
-   case GL_SUBTRACT: return 2;
-   case GL_DOT3_RGB_EXT: return 2;
-   case GL_DOT3_RGBA_EXT: return 2;
-   case GL_DOT3_RGB: return 2;
-   case GL_DOT3_RGBA: return 2;
-   default: return 0;
-   }
-}
-
-
-static GLboolean args_match( struct gl_texture_unit *texUnit )
-{
-   int i, nr = nr_args(texUnit->Combine.ModeRGB);
-
-   for (i = 0 ; i < nr ; i++) {
-      if (texUnit->Combine.SourceA[i] != texUnit->Combine.SourceRGB[i]) 
-	 return GL_FALSE;
-
-      switch(texUnit->Combine.OperandA[i]) {
-      case GL_SRC_ALPHA: 
-	 switch(texUnit->Combine.OperandRGB[i]) {
-	 case GL_SRC_COLOR: 
-	 case GL_SRC_ALPHA: 
-	    break;
-	 default:
-	    return GL_FALSE;
-	 }
-	 break;
-      case GL_ONE_MINUS_SRC_ALPHA: 
-	 switch(texUnit->Combine.OperandRGB[i]) {
-	 case GL_ONE_MINUS_SRC_COLOR: 
-	 case GL_ONE_MINUS_SRC_ALPHA: 
-	    break;
-	 default:
-	    return GL_FALSE;
-	 }
-	 break;
-      default: 
-	 return GL_FALSE;	/* impossible */
-      }
-   }
-
-   return GL_TRUE;
-}
-
-
-static GLuint emit_combine( struct i915_fragment_program *p,
-			    GLuint dest,
-			    GLuint mask,
-			    GLuint saturate,
-			    GLuint unit,
-			    GLenum mode,
-			    const GLenum *source,
-			    const GLenum *operand)
-{
-   int tmp, src[3], nr = nr_args(mode);
-   int i;
-
-   for (i = 0; i < nr; i++)
-      src[i] = emit_combine_source( p, mask, unit, source[i], operand[i] );
-
-   switch (mode) {
-   case GL_REPLACE: 
-      if (mask == A0_DEST_CHANNEL_ALL && !saturate)
-	 return src[0];
-      else
-	 return i915_emit_arith( p, A0_MOV, dest, mask, saturate, src[0], 0, 0 );
-   case GL_MODULATE: 
-      return i915_emit_arith( p, A0_MUL, dest, mask, saturate,
-			     src[0], src[1], 0 );
-   case GL_ADD: 
-      return i915_emit_arith( p, A0_ADD, dest, mask, saturate, 
-			     src[0], src[1], 0 );
-   case GL_ADD_SIGNED:
-      /* tmp = arg0 + arg1
-       * result = tmp + -.5
-       */
-      tmp = i915_emit_const1f(p, .5);
-      tmp = negate(swizzle(tmp,X,X,X,X),1,1,1,1);
-      i915_emit_arith( p, A0_ADD, dest, mask, 0, src[0], src[1], 0 );
-      i915_emit_arith( p, A0_ADD, dest, mask, saturate, dest, tmp, 0 );
-      return dest;
-   case GL_INTERPOLATE:		/* TWO INSTRUCTIONS */
-      /* Arg0 * (Arg2) + Arg1 * (1-Arg2)
-       *
-       * Arg0*Arg2 + Arg1 - Arg1Arg2 
-       *
-       * tmp = Arg0*Arg2 + Arg1, 
-       * result = (-Arg1)Arg2 + tmp 
-       */
-      tmp = i915_get_temp( p );
-      i915_emit_arith( p, A0_MAD, tmp, mask, 0, src[0], src[2], src[1] );
-      i915_emit_arith( p, A0_MAD, dest, mask, saturate, 
-		      negate(src[1], 1,1,1,1), src[2], tmp );
-      return dest;
-   case GL_SUBTRACT: 
-      /* negate src[1] */
-      return i915_emit_arith( p, A0_ADD, dest, mask, saturate, src[0],
-			 negate(src[1],1,1,1,1), 0 );
-
-   case GL_DOT3_RGBA:
-   case GL_DOT3_RGBA_EXT: 
-   case GL_DOT3_RGB_EXT:
-   case GL_DOT3_RGB: {
-      GLuint tmp0 = i915_get_temp( p );
-      GLuint tmp1 = i915_get_temp( p );
-      GLuint neg1 = negate(swizzle(i915_emit_const1f(p, 1),X,X,X,X), 1,1,1,1);
-      GLuint two = swizzle(i915_emit_const1f(p, 2),X,X,X,X);
-      i915_emit_arith( p, A0_MAD, tmp0, A0_DEST_CHANNEL_ALL, 0, 
-		      two, src[0], neg1);
-      if (src[0] == src[1])
-	 tmp1 = tmp0;
-      else
-	 i915_emit_arith( p, A0_MAD, tmp1, A0_DEST_CHANNEL_ALL, 0, 
-			 two, src[1], neg1);
-      i915_emit_arith( p, A0_DP3, dest, mask, saturate, tmp0, tmp1, 0);
-      return dest;
-   }
-
-   default: 
-      return src[0];
-   }
-}
-
-static GLuint get_dest( struct i915_fragment_program *p, int unit )
-{
-   if (p->ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
-      return i915_get_temp( p );
-   else if (unit != p->last_tex_stage)
-      return i915_get_temp( p );
-   else
-      return UREG(REG_TYPE_OC, 0);
-}
-      
-
-
-static GLuint emit_texenv( struct i915_fragment_program *p, int unit )
-{
-   struct gl_texture_unit *texUnit = &p->ctx->Texture.Unit[unit];
-   GLenum envMode = texUnit->EnvMode;
-   struct gl_texture_object *tObj = texUnit->_Current;
-   GLenum format = tObj->Image[0][tObj->BaseLevel]->_BaseFormat;
-   GLuint saturate = unit < p->last_tex_stage ? A0_DEST_SATURATE : 0;
-
-   switch(envMode) {
-   case GL_BLEND: {
-      const int cf = get_source(p, GL_PREVIOUS, unit);
-      const int cc = get_source(p, GL_CONSTANT, unit);
-      const int cs = get_source(p, GL_TEXTURE, unit);
-      const int out = get_dest(p, unit);
-
-      if (format == GL_INTENSITY) {
-	 /* cv = cf(1 - cs) + cc.cs
-	  * cv = cf - cf.cs + cc.cs
-	  */
-	 /* u[2] = MAD( -cf * cs + cf )
-	  * cv   = MAD( cc * cs + u[2] )
-	  */
-	 
-	 i915_emit_arith( p, A0_MAD, out, A0_DEST_CHANNEL_ALL, 0, 
-			 negate(cf,1,1,1,1), cs, cf );
-
-	 i915_emit_arith( p, A0_MAD, out, A0_DEST_CHANNEL_ALL, saturate, 
-			 cc, cs, out );
-
-	 return out;
-      } else {
-	 /* cv = cf(1 - cs) + cc.cs
-	  * cv = cf - cf.cs + cc.cs
-	  * av =      af.as
-	  */
-	 /* u[2] = MAD( cf.-x-y-zw * cs.xyzw + cf.xyz0 )
-	  * oC   = MAD( cc.xyz0 * cs.xyz0 + u[2].xyzw )
-	  */
-	 i915_emit_arith( p, A0_MAD, out, A0_DEST_CHANNEL_ALL, 0,
-			 negate(cf,1,1,1,0),  
-			 cs,
-			 swizzle(cf,X,Y,Z,ZERO) );
-
-
-	 i915_emit_arith( p, A0_MAD, out, A0_DEST_CHANNEL_ALL, saturate,
-			 swizzle(cc,X,Y,Z,ZERO),  
-			 swizzle(cs,X,Y,Z,ZERO),
-			 out );
-
-	 return out;
-      }
-   }
-
-   case GL_DECAL: {
-      if (format == GL_RGB ||
-	  format == GL_RGBA) {
-	 int cf = get_source( p, GL_PREVIOUS, unit );
-	 int cs = get_source( p, GL_TEXTURE, unit );
-	 int out = get_dest(p, unit);
-	 
-	 /* cv = cf(1-as) + cs.as
-	  * cv = cf.(-as) + cf + cs.as
-	  * av = af
-	  */ 
-	 
-	 /* u[2] = mad( cf.xyzw * cs.-w-w-w1 + cf.xyz0 )
-	  * oc = mad( cs.xyz0 * cs.www0 + u[2].xyzw )
-	  */
-	 i915_emit_arith( p, A0_MAD, out, A0_DEST_CHANNEL_ALL, 0,
-			 cf,  
-			 negate(swizzle(cs,W,W,W,ONE),1,1,1,0),
-			 swizzle(cf,X,Y,Z,ZERO) );
-	 
-	 i915_emit_arith( p, A0_MAD, out, A0_DEST_CHANNEL_ALL, saturate,
-			 swizzle(cs,X,Y,Z,ZERO),  
-			 swizzle(cs,W,W,W,ZERO),
-			 out );
-	 return out;
-      }
-      else {
-	 return get_source( p, GL_PREVIOUS, unit );
-      }
-   }
-
-   case GL_REPLACE: {
-      const int cs = get_source( p, GL_TEXTURE, unit );	/* saturated */
-      switch (format) {
-      case GL_ALPHA: {
-	 const int cf = get_source( p, GL_PREVIOUS, unit ); /* saturated */
-	 i915_emit_arith( p, A0_MOV, cs, A0_DEST_CHANNEL_XYZ, 0, cf, 0, 0 );
-	 return cs;
-      }
-      case GL_RGB:
-      case GL_LUMINANCE: {
-	 const int cf = get_source( p, GL_PREVIOUS, unit ); /* saturated */
-	 i915_emit_arith( p, A0_MOV, cs, A0_DEST_CHANNEL_W, 0, cf, 0, 0 );
-	 return cs;
-      }
-      default:
-	 return cs;
-      }
-   }
-
-   case GL_MODULATE: {
-      const int cf = get_source( p, GL_PREVIOUS, unit );
-      const int cs = get_source( p, GL_TEXTURE, unit );
-      const int out = get_dest(p, unit);
-      switch (format) {
-      case GL_ALPHA: 
-	 i915_emit_arith( p, A0_MUL, out, A0_DEST_CHANNEL_ALL, saturate,
-			 swizzle(cs, ONE, ONE, ONE, W), cf, 0 );
-	 break;
-      default:
-	 i915_emit_arith( p, A0_MUL, out, A0_DEST_CHANNEL_ALL, saturate, 
-			 cs, cf, 0 );
-	 break;
-      }
-      return out;
-   }
-   case GL_ADD: {
-      int cf = get_source( p, GL_PREVIOUS, unit );
-      int cs = get_source( p, GL_TEXTURE, unit );
-      const int out = get_dest( p, unit );
-
-      if (format == GL_INTENSITY) {
-	 /* output-color.rgba = add( incoming, u[1] )
-	  */
-	 i915_emit_arith( p, A0_ADD, out, A0_DEST_CHANNEL_ALL, saturate, 
-			 cs, cf, 0 );
-	 return out;
-      }
-      else {
-	 /* cv.xyz = cf.xyz + cs.xyz
-	  * cv.w   = cf.w * cs.w
-	  *
-	  * cv.xyzw = MAD( cf.111w * cs.xyzw + cf.xyz0 )
-	  */
- 	 i915_emit_arith( p, A0_MAD, out, A0_DEST_CHANNEL_ALL, saturate,
-			 swizzle(cf,ONE,ONE,ONE,W), 
-			 cs,  
-			 swizzle(cf,X,Y,Z,ZERO) ); 
-	 return out;
-      }
-      break;
-   }
-   case GL_COMBINE: {
-      GLuint rgb_shift, alpha_shift, out, shift;
-      GLuint dest = get_dest(p, unit);
-
-      /* The EXT version of the DOT3 extension does not support the
-       * scale factor, but the ARB version (and the version in OpenGL
-       * 1.3) does.
-       */
-      switch (texUnit->Combine.ModeRGB) {
-      case GL_DOT3_RGB_EXT:
-	 alpha_shift = texUnit->Combine.ScaleShiftA;
-	 rgb_shift = 0;
-	 break;
-
-      case GL_DOT3_RGBA_EXT:
-	 alpha_shift = 0;
-	 rgb_shift = 0;
-	 break;
-
-      default:
-	 rgb_shift = texUnit->Combine.ScaleShiftRGB;
-	 alpha_shift = texUnit->Combine.ScaleShiftA;
-	 break;
-      }
-
-
-      /* Emit the RGB and A combine ops
-       */
-      if (texUnit->Combine.ModeRGB == texUnit->Combine.ModeA && 
-	  args_match( texUnit )) {
-	 out = emit_combine( p, dest, A0_DEST_CHANNEL_ALL, saturate,
-			     unit,
-			     texUnit->Combine.ModeRGB,
-			     texUnit->Combine.SourceRGB,
-			     texUnit->Combine.OperandRGB );
-      }
-      else if (texUnit->Combine.ModeRGB == GL_DOT3_RGBA_EXT ||
-	       texUnit->Combine.ModeRGB == GL_DOT3_RGBA) {
-
-	 out = emit_combine( p, dest, A0_DEST_CHANNEL_ALL, saturate,
-			     unit,
-			     texUnit->Combine.ModeRGB,
-			     texUnit->Combine.SourceRGB,
-			     texUnit->Combine.OperandRGB );
-      }
-      else {
-	 /* Need to do something to stop from re-emitting identical
-	  * argument calculations here:
-	  */
-	 out = emit_combine( p, dest, A0_DEST_CHANNEL_XYZ, saturate,
-			     unit,
-			     texUnit->Combine.ModeRGB,
-			     texUnit->Combine.SourceRGB,
-			     texUnit->Combine.OperandRGB );
-	 out = emit_combine( p, dest, A0_DEST_CHANNEL_W, saturate,
-			     unit,
-			     texUnit->Combine.ModeA,
-			     texUnit->Combine.SourceA,
-			     texUnit->Combine.OperandA );
-      }
-
-      /* Deal with the final shift:
-       */
-      if (alpha_shift || rgb_shift) {
-	 if (rgb_shift == alpha_shift) {
-	    shift = i915_emit_const1f(p, 1<<rgb_shift);
-	    shift = swizzle(shift,X,X,X,X);
-	 }
-	 else {
-	    shift = i915_emit_const2f(p, 1<<rgb_shift, 1<<alpha_shift);
-	    shift = swizzle(shift,X,X,X,Y);
-	 }
-	 return i915_emit_arith( p, A0_MUL, dest, A0_DEST_CHANNEL_ALL, 
-				saturate, out, shift, 0 );
-      }
-
-      return out;
-   }
-
-   default:
-      return get_source(p, GL_PREVIOUS, 0);
-   }
-}
-
-static void emit_program_fini( struct i915_fragment_program *p )
-{
-   int cf = get_source( p, GL_PREVIOUS, 0 );
-   int out = UREG( REG_TYPE_OC, 0 );
-
-   if (p->ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
-      /* Emit specular add.
-       */
-      GLuint s = i915_emit_decl(p, REG_TYPE_T, T_SPECULAR, D0_CHANNEL_ALL);
-      i915_emit_arith( p, A0_ADD, out, A0_DEST_CHANNEL_ALL, 0, cf, 
-		  swizzle(s, X,Y,Z,ZERO), 0 );
-   }
-   else if (cf != out) {
-      /* Will wind up in here if no texture enabled or a couple of
-       * other scenarios (GL_REPLACE for instance).
-       */
-      i915_emit_arith( p, A0_MOV, out, A0_DEST_CHANNEL_ALL, 0, cf, 0, 0 );
-   }
-}
-
-
-static void i915EmitTextureProgram( i915ContextPtr i915 )
-{
-   GLcontext *ctx = &i915->intel.ctx;
-   struct i915_fragment_program *p = &i915->tex_program;
-   GLuint unit;
-
-   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
-
-   i915_init_program( i915, p );
-
-   if (ctx->Texture._EnabledUnits) {
-      for (unit = 0 ; unit < ctx->Const.MaxTextureUnits ; unit++)
-	 if (ctx->Texture.Unit[unit]._ReallyEnabled) {
-	    p->last_tex_stage = unit;
-	 }
-
-      for (unit = 0 ; unit < ctx->Const.MaxTextureUnits; unit++)
-	 if (ctx->Texture.Unit[unit]._ReallyEnabled) {
-	    p->src_previous = emit_texenv( p, unit );
-	    p->src_texture = UREG_BAD;
-	    p->temp_flag = 0xffff000;
-	    p->temp_flag |= 1 << GET_UREG_NR(p->src_previous);
-	 }
-   }
-
-   emit_program_fini( p );
-
-   i915_fini_program( p );
-   i915_upload_program( i915, p );
-
-   p->translated = 1;
-}
-
-
-void i915ValidateTextureProgram( i915ContextPtr i915 )
-{
-   intelContextPtr intel = &i915->intel;
-   GLcontext *ctx = &intel->ctx;
-   TNLcontext *tnl = TNL_CONTEXT(ctx);
-   struct vertex_buffer *VB = &tnl->vb;
-   DECLARE_RENDERINPUTS(index_bitset);
-   int i, offset;
-   GLuint s4 = i915->state.Ctx[I915_CTXREG_LIS4] & ~S4_VFMT_MASK;
-   GLuint s2 = S2_TEXCOORD_NONE;
-
-   RENDERINPUTS_COPY( index_bitset, tnl->render_inputs_bitset );
-
-   /* Important:
-    */
-   VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
-   intel->vertex_attr_count = 0;
-   intel->coloroffset = 0;
-   intel->specoffset = 0;
-   offset = 0;
-
-   if (i915->vertex_fog == I915_FOG_PIXEL) {
-      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16 );
-      RENDERINPUTS_CLEAR( index_bitset, _TNL_ATTRIB_FOG );
-   }
-   else if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, S4_VFMT_XYZW, 16 );
-   }
-   else {
-      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, S4_VFMT_XYZ, 12 );
-   }
-
-   /* How undefined is undefined? */
-   if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_POINTSIZE )) {
-      EMIT_ATTR( _TNL_ATTRIB_POINTSIZE, EMIT_1F, S4_VFMT_POINT_WIDTH, 4 );
-   }
-      
-   intel->coloroffset = offset / 4;
-   EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, S4_VFMT_COLOR, 4 );
-            
-   if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 ) ||
-       RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_FOG )) {
-      if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_COLOR1 )) {
-	 intel->specoffset = offset / 4;
-	 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, S4_VFMT_SPEC_FOG, 3 );
-      } else 
-	 EMIT_PAD( 3 );
-      
-      if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_FOG ))
-	 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, S4_VFMT_SPEC_FOG, 1 );
-      else
-	 EMIT_PAD( 1 );
-   }
-
-   if (RENDERINPUTS_TEST_RANGE( index_bitset, _TNL_FIRST_TEX, _TNL_LAST_TEX )) {
-      for (i = 0; i < 8; i++) {
-	 if (RENDERINPUTS_TEST( index_bitset, _TNL_ATTRIB_TEX(i) )) {
-	    int sz = VB->TexCoordPtr[i]->size;
-	    
-	    s2 &= ~S2_TEXCOORD_FMT(i, S2_TEXCOORD_FMT0_MASK);
-	    s2 |= S2_TEXCOORD_FMT(i, SZ_TO_HW(sz));
-
-	    EMIT_ATTR( _TNL_ATTRIB_TEX0+i, EMIT_SZ(sz), 0, sz * 4 );
-	 }
-      }
-   }
-
-   /* Only need to change the vertex emit code if there has been a
-    * statechange to a new hardware vertex format:
-    */
-   if (s2 != i915->state.Ctx[I915_CTXREG_LIS2] ||
-       s4 != i915->state.Ctx[I915_CTXREG_LIS4]) {
-    
-      I915_STATECHANGE( i915, I915_UPLOAD_CTX );
-
-      i915->tex_program.translated = 0;
-
-      /* Must do this *after* statechange, so as not to affect
-       * buffered vertices reliant on the old state:
-       */
-      intel->vertex_size = _tnl_install_attrs( ctx, 
-					       intel->vertex_attrs, 
-					       intel->vertex_attr_count,
-					       intel->ViewportMatrix.m, 0 ); 
-
-      intel->vertex_size >>= 2;
-
-      i915->state.Ctx[I915_CTXREG_LIS2] = s2;
-      i915->state.Ctx[I915_CTXREG_LIS4] = s4;
-
-      assert(intel->vtbl.check_vertex_size( intel, intel->vertex_size ));
-   }
-
-   if (!i915->tex_program.translated ||
-       i915->last_ReallyEnabled != ctx->Texture._EnabledUnits) {
-      i915EmitTextureProgram( i915 );      
-      i915->last_ReallyEnabled = ctx->Texture._EnabledUnits;
-   }
-}

src/mesa/drivers/dri/i915/i915_vtbl.c

@@ -37,112 +37,138 @@
 #include "tnl/t_vertex.h"
 
 #include "intel_batchbuffer.h"
+#include "intel_tex.h"
+#include "intel_regions.h"
 
 #include "i915_reg.h"
 #include "i915_context.h"
 
-static void i915_render_start( intelContextPtr intel )
+static void
+i915_render_start(struct intel_context *intel)
 {
-   GLcontext *ctx = &intel->ctx;
-   i915ContextPtr i915 = I915_CONTEXT(intel);
+   struct i915_context *i915 = i915_context(&intel->ctx);
 
-   if (ctx->FragmentProgram._Active) 
-      i915ValidateFragmentProgram( i915 );
-   else 
-      i915ValidateTextureProgram( i915 );
+   i915ValidateFragmentProgram(i915);
 }
 
 
-static void i915_reduced_primitive_state( intelContextPtr intel,
-					  GLenum rprim )
+static void
+i915_reduced_primitive_state(struct intel_context *intel, GLenum rprim)
 {
-    i915ContextPtr i915 = I915_CONTEXT(intel);
-    GLuint st1 = i915->state.Stipple[I915_STPREG_ST1];
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   GLuint st1 = i915->state.Stipple[I915_STPREG_ST1];
 
-    st1 &= ~ST1_ENABLE;
+   st1 &= ~ST1_ENABLE;
 
-    switch (rprim) {
-    case GL_TRIANGLES:
-       if (intel->ctx.Polygon.StippleFlag &&
-	   intel->hw_stipple)
-	  st1 |= ST1_ENABLE;
-       break;
-    case GL_LINES:
-    case GL_POINTS:
-    default:
-       break;
-    }
+   switch (rprim) {
+   case GL_TRIANGLES:
+      if (intel->ctx.Polygon.StippleFlag && intel->hw_stipple)
+         st1 |= ST1_ENABLE;
+      break;
+   case GL_LINES:
+   case GL_POINTS:
+   default:
+      break;
+   }
 
-    i915->intel.reduced_primitive = rprim;
+   i915->intel.reduced_primitive = rprim;
 
-    if (st1 != i915->state.Stipple[I915_STPREG_ST1]) {
-       I915_STATECHANGE(i915, I915_UPLOAD_STIPPLE);
-       i915->state.Stipple[I915_STPREG_ST1] = st1;
-    }
+   if (st1 != i915->state.Stipple[I915_STPREG_ST1]) {
+      INTEL_FIREVERTICES(intel);
+
+      I915_STATECHANGE(i915, I915_UPLOAD_STIPPLE);
+      i915->state.Stipple[I915_STPREG_ST1] = st1;
+   }
 }
 
 
 /* Pull apart the vertex format registers and figure out how large a
  * vertex is supposed to be. 
  */
-static GLboolean i915_check_vertex_size( intelContextPtr intel,
-					 GLuint expected )
+static GLboolean
+i915_check_vertex_size(struct intel_context *intel, GLuint expected)
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
+   struct i915_context *i915 = i915_context(&intel->ctx);
    int lis2 = i915->current->Ctx[I915_CTXREG_LIS2];
    int lis4 = i915->current->Ctx[I915_CTXREG_LIS4];
    int i, sz = 0;
 
    switch (lis4 & S4_VFMT_XYZW_MASK) {
-   case S4_VFMT_XY: sz = 2; break;
-   case S4_VFMT_XYZ: sz = 3; break;
-   case S4_VFMT_XYW: sz = 3; break;
-   case S4_VFMT_XYZW: sz = 4; break;
-   default: 
+   case S4_VFMT_XY:
+      sz = 2;
+      break;
+   case S4_VFMT_XYZ:
+      sz = 3;
+      break;
+   case S4_VFMT_XYW:
+      sz = 3;
+      break;
+   case S4_VFMT_XYZW:
+      sz = 4;
+      break;
+   default:
       fprintf(stderr, "no xyzw specified\n");
       return 0;
    }
 
-   if (lis4 & S4_VFMT_SPEC_FOG) sz++;
-   if (lis4 & S4_VFMT_COLOR) sz++;
-   if (lis4 & S4_VFMT_DEPTH_OFFSET) sz++;
-   if (lis4 & S4_VFMT_POINT_WIDTH) sz++;
-   if (lis4 & S4_VFMT_FOG_PARAM) sz++;
-	
-   for (i = 0 ; i < 8 ; i++) { 
+   if (lis4 & S4_VFMT_SPEC_FOG)
+      sz++;
+   if (lis4 & S4_VFMT_COLOR)
+      sz++;
+   if (lis4 & S4_VFMT_DEPTH_OFFSET)
+      sz++;
+   if (lis4 & S4_VFMT_POINT_WIDTH)
+      sz++;
+   if (lis4 & S4_VFMT_FOG_PARAM)
+      sz++;
+
+   for (i = 0; i < 8; i++) {
       switch (lis2 & S2_TEXCOORD_FMT0_MASK) {
-      case TEXCOORDFMT_2D: sz += 2; break;
-      case TEXCOORDFMT_3D: sz += 3; break;
-      case TEXCOORDFMT_4D: sz += 4; break;
-      case TEXCOORDFMT_1D: sz += 1; break;
-      case TEXCOORDFMT_2D_16: sz += 1; break;
-      case TEXCOORDFMT_4D_16: sz += 2; break;
-      case TEXCOORDFMT_NOT_PRESENT: break;
+      case TEXCOORDFMT_2D:
+         sz += 2;
+         break;
+      case TEXCOORDFMT_3D:
+         sz += 3;
+         break;
+      case TEXCOORDFMT_4D:
+         sz += 4;
+         break;
+      case TEXCOORDFMT_1D:
+         sz += 1;
+         break;
+      case TEXCOORDFMT_2D_16:
+         sz += 1;
+         break;
+      case TEXCOORDFMT_4D_16:
+         sz += 2;
+         break;
+      case TEXCOORDFMT_NOT_PRESENT:
+         break;
       default:
-	 fprintf(stderr, "bad texcoord fmt %d\n", i);
-	 return GL_FALSE;
+         fprintf(stderr, "bad texcoord fmt %d\n", i);
+         return GL_FALSE;
       }
       lis2 >>= S2_TEXCOORD_FMT1_SHIFT;
    }
-	
-   if (sz != expected) 
+
+   if (sz != expected)
       fprintf(stderr, "vertex size mismatch %d/%d\n", sz, expected);
-   
+
    return sz == expected;
 }
 
 
-static void i915_emit_invarient_state( intelContextPtr intel )
+static void
+i915_emit_invarient_state(struct intel_context *intel)
 {
    BATCH_LOCALS;
 
-   BEGIN_BATCH( 200 );
+   BEGIN_BATCH(200, 0);
 
    OUT_BATCH(_3DSTATE_AA_CMD |
-	     AA_LINE_ECAAR_WIDTH_ENABLE |
-	     AA_LINE_ECAAR_WIDTH_1_0 |
-	     AA_LINE_REGION_WIDTH_ENABLE |
-	     AA_LINE_REGION_WIDTH_1_0);
+             AA_LINE_ECAAR_WIDTH_ENABLE |
+             AA_LINE_ECAAR_WIDTH_1_0 |
+             AA_LINE_REGION_WIDTH_ENABLE | AA_LINE_REGION_WIDTH_1_0);
 
    OUT_BATCH(_3DSTATE_DFLT_DIFFUSE_CMD);
    OUT_BATCH(0);
@@ -155,35 +181,27 @@ static void i915_emit_invarient_state( intelContextPtr intel )
 
    /* Don't support texture crossbar yet */
    OUT_BATCH(_3DSTATE_COORD_SET_BINDINGS |
-	     CSB_TCB(0, 0) |
-	     CSB_TCB(1, 1) |
-	     CSB_TCB(2, 2) |
-	     CSB_TCB(3, 3) |
-	     CSB_TCB(4, 4) |
-	     CSB_TCB(5, 5) |
-	     CSB_TCB(6, 6) |
-	     CSB_TCB(7, 7));
+             CSB_TCB(0, 0) |
+             CSB_TCB(1, 1) |
+             CSB_TCB(2, 2) |
+             CSB_TCB(3, 3) |
+             CSB_TCB(4, 4) | CSB_TCB(5, 5) | CSB_TCB(6, 6) | CSB_TCB(7, 7));
 
    OUT_BATCH(_3DSTATE_RASTER_RULES_CMD |
-	     ENABLE_POINT_RASTER_RULE |
-	     OGL_POINT_RASTER_RULE |
-	     ENABLE_LINE_STRIP_PROVOKE_VRTX |
-	     ENABLE_TRI_FAN_PROVOKE_VRTX |
-	     LINE_STRIP_PROVOKE_VRTX(1) |
-	     TRI_FAN_PROVOKE_VRTX(2) | 
-	     ENABLE_TEXKILL_3D_4D |
-	     TEXKILL_4D);
+             ENABLE_POINT_RASTER_RULE |
+             OGL_POINT_RASTER_RULE |
+             ENABLE_LINE_STRIP_PROVOKE_VRTX |
+             ENABLE_TRI_FAN_PROVOKE_VRTX |
+             LINE_STRIP_PROVOKE_VRTX(1) |
+             TRI_FAN_PROVOKE_VRTX(2) | ENABLE_TEXKILL_3D_4D | TEXKILL_4D);
 
    /* Need to initialize this to zero.
     */
-   OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | 
-	     I1_LOAD_S(3) |
-	     (1));
+   OUT_BATCH(_3DSTATE_LOAD_STATE_IMMEDIATE_1 | I1_LOAD_S(3) | (1));
    OUT_BATCH(0);
- 
+
    /* XXX: Use this */
-   OUT_BATCH(_3DSTATE_SCISSOR_ENABLE_CMD | 
-	     DISABLE_SCISSOR_RECT);
+   OUT_BATCH(_3DSTATE_SCISSOR_ENABLE_CMD | DISABLE_SCISSOR_RECT);
 
    OUT_BATCH(_3DSTATE_SCISSOR_RECT_0_CMD);
    OUT_BATCH(0);
@@ -191,29 +209,22 @@ static void i915_emit_invarient_state( intelContextPtr intel )
 
    OUT_BATCH(_3DSTATE_DEPTH_SUBRECT_DISABLE);
 
-   OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0); /* disable indirect state */
+   OUT_BATCH(_3DSTATE_LOAD_INDIRECT | 0);       /* disable indirect state */
    OUT_BATCH(0);
 
 
    /* Don't support twosided stencil yet */
-   OUT_BATCH(_3DSTATE_BACKFACE_STENCIL_OPS |
-	     BFO_ENABLE_STENCIL_TWO_SIDE |
-	     0 );
-   
+   OUT_BATCH(_3DSTATE_BACKFACE_STENCIL_OPS | BFO_ENABLE_STENCIL_TWO_SIDE | 0);
+
    ADVANCE_BATCH();
 }
 
 
-#define emit( intel, state, size )			\
-do {							\
-   int k;						\
-   BEGIN_BATCH( (size) / sizeof(GLuint));		\
-   for (k = 0 ; k < (size) / sizeof(GLuint) ; k++)	\
-      OUT_BATCH((state)[k]);				\
-   ADVANCE_BATCH();					\
-} while (0);
+#define emit(intel, state, size )		     \
+   intel_batchbuffer_data(intel->batch, state, size, 0 )
 
-static GLuint get_dirty( struct i915_hw_state *state )
+static GLuint
+get_dirty(struct i915_hw_state *state)
 {
    GLuint dirty;
 
@@ -224,12 +235,12 @@ static GLuint get_dirty( struct i915_hw_state *state )
    if (dirty & I915_UPLOAD_TEX_ALL)
       state->emitted &= ~I915_UPLOAD_TEX_ALL;
    dirty = state->active & ~state->emitted;
-
    return dirty;
 }
 
 
-static GLuint get_state_size( struct i915_hw_state *state )
+static GLuint
+get_state_size(struct i915_hw_state *state)
 {
    GLuint dirty = get_dirty(state);
    GLuint i;
@@ -238,28 +249,28 @@ static GLuint get_state_size( struct i915_hw_state *state )
    if (dirty & I915_UPLOAD_CTX)
       sz += sizeof(state->Ctx);
 
-   if (dirty & I915_UPLOAD_BUFFERS) 
+   if (dirty & I915_UPLOAD_BUFFERS)
       sz += sizeof(state->Buffer);
 
    if (dirty & I915_UPLOAD_STIPPLE)
       sz += sizeof(state->Stipple);
 
-   if (dirty & I915_UPLOAD_FOG) 
+   if (dirty & I915_UPLOAD_FOG)
       sz += sizeof(state->Fog);
 
    if (dirty & I915_UPLOAD_TEX_ALL) {
       int nr = 0;
-      for (i = 0; i < I915_TEX_UNITS; i++) 
-	 if (dirty & I915_UPLOAD_TEX(i)) 
-	    nr++;
+      for (i = 0; i < I915_TEX_UNITS; i++)
+         if (dirty & I915_UPLOAD_TEX(i))
+            nr++;
 
-      sz += (2+nr*3) * sizeof(GLuint) * 2;
+      sz += (2 + nr * 3) * sizeof(GLuint) * 2;
    }
 
-   if (dirty & I915_UPLOAD_CONSTANTS) 
+   if (dirty & I915_UPLOAD_CONSTANTS)
       sz += state->ConstantSize * sizeof(GLuint);
 
-   if (dirty & I915_UPLOAD_PROGRAM) 
+   if (dirty & I915_UPLOAD_PROGRAM)
       sz += state->ProgramSize * sizeof(GLuint);
 
    return sz;
@@ -268,42 +279,83 @@ static GLuint get_state_size( struct i915_hw_state *state )
 
 /* Push the state into the sarea and/or texture memory.
  */
-static void i915_emit_state( intelContextPtr intel )
+static void
+i915_emit_state(struct intel_context *intel)
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
+   struct i915_context *i915 = i915_context(&intel->ctx);
    struct i915_hw_state *state = i915->current;
    int i;
-   GLuint dirty = get_dirty(state);
-   GLuint counter = intel->batch.counter;
+   GLuint dirty;
    BATCH_LOCALS;
 
-   if (intel->batch.space < get_state_size(state)) {
-      intelFlushBatch(intel, GL_TRUE);
-      dirty = get_dirty(state);
-      counter = intel->batch.counter;
-   }
+   /* We don't hold the lock at this point, so want to make sure that
+    * there won't be a buffer wrap.  
+    *
+    * It might be better to talk about explicit places where
+    * scheduling is allowed, rather than assume that it is whenever a
+    * batchbuffer fills up.
+    */
+   intel_batchbuffer_require_space(intel->batch, get_state_size(state), 0);
 
-   if (VERBOSE) 
+   /* Do this here as we may have flushed the batchbuffer above,
+    * causing more state to be dirty!
+    */
+   dirty = get_dirty(state);
+
+   if (INTEL_DEBUG & DEBUG_STATE)
       fprintf(stderr, "%s dirty: %x\n", __FUNCTION__, dirty);
 
+   if (dirty & I915_UPLOAD_INVARIENT) {
+      if (INTEL_DEBUG & DEBUG_STATE)
+         fprintf(stderr, "I915_UPLOAD_INVARIENT:\n");
+      i915_emit_invarient_state(intel);
+   }
+
    if (dirty & I915_UPLOAD_CTX) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_CTX:\n"); 
-      emit( i915, state->Ctx, sizeof(state->Ctx) );
+      if (INTEL_DEBUG & DEBUG_STATE)
+         fprintf(stderr, "I915_UPLOAD_CTX:\n");
+      emit(intel, state->Ctx, sizeof(state->Ctx));
    }
 
    if (dirty & I915_UPLOAD_BUFFERS) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_BUFFERS:\n"); 
-      emit( i915, state->Buffer, sizeof(state->Buffer) );
+      if (INTEL_DEBUG & DEBUG_STATE)
+         fprintf(stderr, "I915_UPLOAD_BUFFERS:\n");
+      BEGIN_BATCH(I915_DEST_SETUP_SIZE + 2, 0);
+      OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR0]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_CBUFADDR1]);
+      OUT_RELOC(state->draw_region->buffer,
+                DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE,
+                state->draw_region->draw_offset);
+
+      if (state->depth_region) {
+         OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR0]);
+         OUT_BATCH(state->Buffer[I915_DESTREG_DBUFADDR1]);
+         OUT_RELOC(state->depth_region->buffer,
+                   DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                   DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE,
+                   state->depth_region->draw_offset);
+      }
+
+      OUT_BATCH(state->Buffer[I915_DESTREG_DV0]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_DV1]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SENABLE]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SR0]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SR1]);
+      OUT_BATCH(state->Buffer[I915_DESTREG_SR2]);
+      ADVANCE_BATCH();
    }
 
    if (dirty & I915_UPLOAD_STIPPLE) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_STIPPLE:\n"); 
-      emit( i915, state->Stipple, sizeof(state->Stipple) );
+      if (INTEL_DEBUG & DEBUG_STATE)
+         fprintf(stderr, "I915_UPLOAD_STIPPLE:\n");
+      emit(intel, state->Stipple, sizeof(state->Stipple));
    }
 
    if (dirty & I915_UPLOAD_FOG) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_FOG:\n"); 
-      emit( i915, state->Fog, sizeof(state->Fog) );
+      if (INTEL_DEBUG & DEBUG_STATE)
+         fprintf(stderr, "I915_UPLOAD_FOG:\n");
+      emit(intel, state->Fog, sizeof(state->Fog));
    }
 
    /* Combine all the dirty texture state into a single command to
@@ -312,142 +364,186 @@ static void i915_emit_state( intelContextPtr intel )
    if (dirty & I915_UPLOAD_TEX_ALL) {
       int nr = 0;
 
-      for (i = 0; i < I915_TEX_UNITS; i++) 
-	 if (dirty & I915_UPLOAD_TEX(i)) 
-	    nr++;
+      for (i = 0; i < I915_TEX_UNITS; i++)
+         if (dirty & I915_UPLOAD_TEX(i))
+            nr++;
 
-      BEGIN_BATCH(2+nr*3);
-      OUT_BATCH(_3DSTATE_MAP_STATE | (3*nr));
+      BEGIN_BATCH(2 + nr * 3, 0);
+      OUT_BATCH(_3DSTATE_MAP_STATE | (3 * nr));
       OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
-      for (i = 0 ; i < I915_TEX_UNITS ; i++)
-	 if (dirty & I915_UPLOAD_TEX(i)) {
-	    OUT_BATCH(state->Tex[i][I915_TEXREG_MS2]);
-	    OUT_BATCH(state->Tex[i][I915_TEXREG_MS3]);
-	    OUT_BATCH(state->Tex[i][I915_TEXREG_MS4]);
-	 }
+      for (i = 0; i < I915_TEX_UNITS; i++)
+         if (dirty & I915_UPLOAD_TEX(i)) {
+
+            if (state->tex_buffer[i]) {
+               OUT_RELOC(state->tex_buffer[i],
+                         DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+                         DRM_BO_MASK_MEM | DRM_BO_FLAG_READ,
+                         state->tex_offset[i]);
+            }
+            else {
+               assert(i == 0);
+               assert(state == &i915->meta);
+               OUT_BATCH(0);
+            }
+
+            OUT_BATCH(state->Tex[i][I915_TEXREG_MS3]);
+            OUT_BATCH(state->Tex[i][I915_TEXREG_MS4]);
+         }
       ADVANCE_BATCH();
 
-      BEGIN_BATCH(2+nr*3);
-      OUT_BATCH(_3DSTATE_SAMPLER_STATE | (3*nr));
+      BEGIN_BATCH(2 + nr * 3, 0);
+      OUT_BATCH(_3DSTATE_SAMPLER_STATE | (3 * nr));
       OUT_BATCH((dirty & I915_UPLOAD_TEX_ALL) >> I915_UPLOAD_TEX_0_SHIFT);
-      for (i = 0 ; i < I915_TEX_UNITS ; i++)
-	 if (dirty & I915_UPLOAD_TEX(i)) {
-	    OUT_BATCH(state->Tex[i][I915_TEXREG_SS2]);
-	    OUT_BATCH(state->Tex[i][I915_TEXREG_SS3]);
-	    OUT_BATCH(state->Tex[i][I915_TEXREG_SS4]);
-	 }
+      for (i = 0; i < I915_TEX_UNITS; i++)
+         if (dirty & I915_UPLOAD_TEX(i)) {
+            OUT_BATCH(state->Tex[i][I915_TEXREG_SS2]);
+            OUT_BATCH(state->Tex[i][I915_TEXREG_SS3]);
+            OUT_BATCH(state->Tex[i][I915_TEXREG_SS4]);
+         }
       ADVANCE_BATCH();
    }
 
    if (dirty & I915_UPLOAD_CONSTANTS) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_CONSTANTS:\n"); 
-      emit( i915, state->Constant, state->ConstantSize * sizeof(GLuint) );
+      if (INTEL_DEBUG & DEBUG_STATE)
+         fprintf(stderr, "I915_UPLOAD_CONSTANTS:\n");
+      emit(intel, state->Constant, state->ConstantSize * sizeof(GLuint));
    }
 
    if (dirty & I915_UPLOAD_PROGRAM) {
-      if (VERBOSE) fprintf(stderr, "I915_UPLOAD_PROGRAM:\n"); 
+      if (INTEL_DEBUG & DEBUG_STATE)
+         fprintf(stderr, "I915_UPLOAD_PROGRAM:\n");
 
-      assert((state->Program[0] & 0x1ff)+2 == state->ProgramSize);
-      
-      emit( i915, state->Program, state->ProgramSize * sizeof(GLuint) );
-      if (VERBOSE)
-	 i915_disassemble_program( state->Program, state->ProgramSize );
+      assert((state->Program[0] & 0x1ff) + 2 == state->ProgramSize);
+
+      emit(intel, state->Program, state->ProgramSize * sizeof(GLuint));
+      if (INTEL_DEBUG & DEBUG_STATE)
+         i915_disassemble_program(state->Program, state->ProgramSize);
    }
 
    state->emitted |= dirty;
-   intel->batch.last_emit_state = counter;
-   assert(counter == intel->batch.counter);
 }
 
-static void i915_destroy_context( intelContextPtr intel )
+static void
+i915_destroy_context(struct intel_context *intel)
 {
    _tnl_free_vertices(&intel->ctx);
 }
 
 
 /**
- * Set the color buffer drawing region.
+ * Set the drawing regions for the color and depth/stencil buffers.
+ * This involves setting the pitch, cpp and buffer ID/location.
+ * Also set pixel format for color and Z rendering
+ * Used for setting both regular and meta state.
  */
+void
+i915_state_draw_region(struct intel_context *intel,
+                       struct i915_hw_state *state,
+                       struct intel_region *color_region,
+                       struct intel_region *depth_region)
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   GLuint value;
+
+   ASSERT(state == &i915->state || state == &i915->meta);
+
+   if (state->draw_region != color_region) {
+      intel_region_release(intel->intelScreen, &state->draw_region);
+      intel_region_reference(&state->draw_region, color_region);
+   }
+   if (state->depth_region != depth_region) {
+      intel_region_release(intel->intelScreen, &state->depth_region);
+      intel_region_reference(&state->depth_region, depth_region);
+   }
+
+   /*
+    * Set stride/cpp values
+    */
+   if (color_region) {
+      state->Buffer[I915_DESTREG_CBUFADDR0] = _3DSTATE_BUF_INFO_CMD;
+      state->Buffer[I915_DESTREG_CBUFADDR1] =
+         (BUF_3D_ID_COLOR_BACK |
+          BUF_3D_PITCH(color_region->pitch * color_region->cpp) |
+          BUF_3D_USE_FENCE);
+   }
+
+   if (depth_region) {
+      state->Buffer[I915_DESTREG_DBUFADDR0] = _3DSTATE_BUF_INFO_CMD;
+      state->Buffer[I915_DESTREG_DBUFADDR1] =
+         (BUF_3D_ID_DEPTH |
+          BUF_3D_PITCH(depth_region->pitch * depth_region->cpp) |
+          BUF_3D_USE_FENCE);
+   }
+
+   /*
+    * Compute/set I915_DESTREG_DV1 value
+    */
+   value = (DSTORG_HORT_BIAS(0x8) |     /* .5 */
+            DSTORG_VERT_BIAS(0x8) |     /* .5 */
+            LOD_PRECLAMP_OGL | TEX_DEFAULT_COLOR_OGL);
+   if (color_region && color_region->cpp == 4) {
+      value |= DV_PF_8888;
+   }
+   else {
+      value |= (DITHER_FULL_ALWAYS | DV_PF_565);
+   }
+   if (depth_region && depth_region->cpp == 4) {
+      value |= DEPTH_FRMT_24_FIXED_8_OTHER;
+   }
+   else {
+      value |= DEPTH_FRMT_16_FIXED;
+   }
+   state->Buffer[I915_DESTREG_DV1] = value;
+
+   I915_STATECHANGE(i915, I915_UPLOAD_BUFFERS);
+}
+
+
 static void
-i915_set_color_region( intelContextPtr intel, const intelRegion *region)
+i915_set_draw_region(struct intel_context *intel,
+                     struct intel_region *color_region,
+                     struct intel_region *depth_region)
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
-   I915_STATECHANGE( i915, I915_UPLOAD_BUFFERS );
-   i915->state.Buffer[I915_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_CBUFADDR2] = region->offset;
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   i915_state_draw_region(intel, &i915->state, color_region, depth_region);
 }
 
 
-/**
- * specify the z-buffer/stencil region
- */
+
 static void
-i915_set_z_region( intelContextPtr intel, const intelRegion *region)
+i915_lost_hardware(struct intel_context *intel)
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
-   I915_STATECHANGE( i915, I915_UPLOAD_BUFFERS );
-   i915->state.Buffer[I915_DESTREG_DBUFADDR1] =
-      (BUF_3D_ID_DEPTH | BUF_3D_PITCH(region->pitch) | BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_DBUFADDR2] = region->offset;
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   i915->state.emitted = 0;
+}
+
+static GLuint
+i915_flush_cmd(void)
+{
+   return MI_FLUSH | FLUSH_MAP_CACHE;
+}
+
+static void 
+i915_assert_not_dirty( struct intel_context *intel )
+{
+   struct i915_context *i915 = i915_context(&intel->ctx);
+   struct i915_hw_state *state = i915->current;
+   GLuint dirty = get_dirty(state);
+   assert(!dirty);
 }
 
 
-/**
- * Set both the color and Z/stencil drawing regions.
- * Similar to two previous functions, but don't use I915_STATECHANGE()
- */
-static void
-i915_update_color_z_regions(intelContextPtr intel,
-                            const intelRegion *colorRegion,
-                            const intelRegion *depthRegion)
+void
+i915InitVtbl(struct i915_context *i915)
 {
-   i915ContextPtr i915 = I915_CONTEXT(intel);
-
-   i915->state.Buffer[I915_DESTREG_CBUFADDR1] =
-      (BUF_3D_ID_COLOR_BACK | BUF_3D_PITCH(colorRegion->pitch) | BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_CBUFADDR2] = colorRegion->offset;
-
-   i915->state.Buffer[I915_DESTREG_DBUFADDR1] =
-      (BUF_3D_ID_DEPTH |
-       BUF_3D_PITCH(depthRegion->pitch) |  /* pitch in bytes */
-       BUF_3D_USE_FENCE);
-   i915->state.Buffer[I915_DESTREG_DBUFADDR2] = depthRegion->offset;
-}
-
-
-static void i915_lost_hardware( intelContextPtr intel )
-{
-   I915_CONTEXT(intel)->state.emitted = 0;
-}
-
-static void i915_emit_flush( intelContextPtr intel )
-{
-   BATCH_LOCALS;
-
-   BEGIN_BATCH(2);
-   OUT_BATCH( MI_FLUSH | FLUSH_MAP_CACHE | FLUSH_RENDER_CACHE ); 
-   OUT_BATCH( 0 );
-   ADVANCE_BATCH();
-}
-
-
-void i915InitVtbl( i915ContextPtr i915 )
-{
-   i915->intel.vtbl.alloc_tex_obj = i915AllocTexObj;
    i915->intel.vtbl.check_vertex_size = i915_check_vertex_size;
-   i915->intel.vtbl.clear_with_tris = i915ClearWithTris;
-   i915->intel.vtbl.rotate_window = i915RotateWindow;
    i915->intel.vtbl.destroy = i915_destroy_context;
-   i915->intel.vtbl.emit_invarient_state = i915_emit_invarient_state;
    i915->intel.vtbl.emit_state = i915_emit_state;
    i915->intel.vtbl.lost_hardware = i915_lost_hardware;
    i915->intel.vtbl.reduced_primitive_state = i915_reduced_primitive_state;
    i915->intel.vtbl.render_start = i915_render_start;
-   i915->intel.vtbl.set_color_region = i915_set_color_region;
-   i915->intel.vtbl.set_z_region = i915_set_z_region;
-   i915->intel.vtbl.update_color_z_regions = i915_update_color_z_regions;
+   i915->intel.vtbl.set_draw_region = i915_set_draw_region;
    i915->intel.vtbl.update_texture_state = i915UpdateTextureState;
-   i915->intel.vtbl.emit_flush = i915_emit_flush;
+   i915->intel.vtbl.flush_cmd = i915_flush_cmd;
+   i915->intel.vtbl.assert_not_dirty = i915_assert_not_dirty;
 }
-

src/mesa/drivers/dri/i915/intel_batchbuffer.h

@@ -1,126 +1,123 @@
-/**************************************************************************
- * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
- * All Rights Reserved.
- * 
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the
- * "Software"), to deal in the Software without restriction, including
- * without limitation the rights to use, copy, modify, merge, publish,
- * distribute, sub license, and/or sell copies of the Software, and to
- * permit persons to whom the Software is furnished to do so, subject to
- * the following conditions:
- * 
- * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
- * of the Software.
- * 
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
- * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
- * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
- * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
- * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- * 
- **************************************************************************/
-
 #ifndef INTEL_BATCHBUFFER_H
 #define INTEL_BATCHBUFFER_H
 
-#include "intel_context.h"
-#include "intel_ioctl.h"
+#include "mtypes.h"
+#include "dri_bufmgr.h"
 
+struct intel_context;
 
-#define BATCH_LOCALS	GLubyte *batch_ptr;
+#define BATCH_SZ 4096
+#define BATCH_RESERVED 16
 
-/* #define VERBOSE 0 */
-#ifndef VERBOSE
-extern int VERBOSE;
-#endif
+#define MAX_RELOCS 100
 
+#define INTEL_BATCH_NO_CLIPRECTS 0x1
+#define INTEL_BATCH_CLIPRECTS    0x2
 
-#define BEGIN_BATCH(n)							\
-do {									\
-   if (VERBOSE) fprintf(stderr, 					\
-			"BEGIN_BATCH(%ld) in %s, %d dwords free\n",	\
-			((unsigned long)n), __FUNCTION__,		\
-			intel->batch.space/4);				\
-   if (intel->batch.space < (n)*4)					\
-      intelFlushBatch(intel, GL_TRUE);					\
-   if (intel->batch.space == intel->batch.size)	intel->batch.func = __FUNCTION__;			\
-   batch_ptr = intel->batch.ptr;					\
-} while (0)
-
-#define OUT_BATCH(n)					\
-do {							\
-   *(GLuint *)batch_ptr = (n);				\
-   if (VERBOSE) fprintf(stderr, " -- %08x at %s/%d\n", (n), __FILE__, __LINE__);	\
-   batch_ptr += 4;					\
-} while (0)
-
-#define ADVANCE_BATCH()						\
-do {								\
-   if (VERBOSE) fprintf(stderr, "ADVANCE_BATCH()\n");		\
-   intel->batch.space -= (batch_ptr - intel->batch.ptr);	\
-   intel->batch.ptr = batch_ptr;				\
-   assert(intel->batch.space >= 0);				\
-} while(0)
-
-extern void intelInitBatchBuffer( GLcontext *ctx );
-extern void intelDestroyBatchBuffer( GLcontext *ctx );
-
-extern void intelStartInlinePrimitive( intelContextPtr intel, GLuint prim );
-extern void intelWrapInlinePrimitive( intelContextPtr intel );
-extern void intelRestartInlinePrimitive( intelContextPtr intel );
-extern GLuint *intelEmitInlinePrimitiveLocked(intelContextPtr intel, 
-					      int primitive, int dwords,
-					      int vertex_size);
-extern void intelCopyBuffer( const __DRIdrawablePrivate *dpriv,
-			     const drm_clip_rect_t	*rect);
-extern void intelClearWithBlit(GLcontext *ctx, GLbitfield mask, GLboolean all,
-			     GLint cx1, GLint cy1, GLint cw, GLint ch);
-
-extern void intelEmitCopyBlitLocked( intelContextPtr intel,
-				     GLuint cpp,
-				     GLshort src_pitch,
-				     GLuint  src_offset,
-				     GLshort dst_pitch,
-				     GLuint  dst_offset,
-				     GLshort srcx, GLshort srcy,
-				     GLshort dstx, GLshort dsty,
-				     GLshort w, GLshort h );
-
-extern void intelEmitFillBlitLocked( intelContextPtr intel,
-				     GLuint cpp,
-				     GLshort dst_pitch,
-				     GLuint dst_offset,
-				     GLshort x, GLshort y, 
-				     GLshort w, GLshort h,
-				     GLuint color );
-
-
-
-
-static __inline GLuint *intelExtendInlinePrimitive( intelContextPtr intel, 
-						GLuint dwords )
+struct buffer_reloc
 {
-   GLuint sz = dwords * sizeof(GLuint);
-   GLuint *ptr;
+   struct _DriBufferObject *buf;
+   GLuint offset;
+   GLuint delta;                /* not needed? */
+};
 
-   if (intel->batch.space < sz) {
-      intelWrapInlinePrimitive( intel );
-/*       assert(intel->batch.space >= sz); */
-   }
+struct intel_batchbuffer
+{
+   struct bufmgr *bm;
+   struct intel_context *intel;
 
-/*    assert(intel->prim.primitive != ~0); */
-   ptr = (GLuint *)intel->batch.ptr;
-   intel->batch.ptr += sz;
-   intel->batch.space -= sz;
+   struct _DriBufferObject *buffer;
+   struct _DriFenceObject *last_fence;
+   GLuint flags;
 
-   return ptr;
+   drmBOList list;
+   GLuint list_count;
+   GLubyte *map;
+   GLubyte *ptr;
+
+   struct buffer_reloc reloc[MAX_RELOCS];
+   GLuint nr_relocs;
+};
+
+struct intel_batchbuffer *intel_batchbuffer_alloc(struct intel_context
+                                                  *intel);
+
+void intel_batchbuffer_free(struct intel_batchbuffer *batch);
+
+
+void intel_batchbuffer_finish(struct intel_batchbuffer *batch);
+
+struct _DriFenceObject *intel_batchbuffer_flush(struct intel_batchbuffer
+                                                *batch);
+
+void intel_batchbuffer_reset(struct intel_batchbuffer *batch);
+
+
+/* Unlike bmBufferData, this currently requires the buffer be mapped.
+ * Consider it a convenience function wrapping multple
+ * intel_buffer_dword() calls.
+ */
+void intel_batchbuffer_data(struct intel_batchbuffer *batch,
+                            const void *data, GLuint bytes, GLuint flags);
+
+void intel_batchbuffer_release_space(struct intel_batchbuffer *batch,
+                                     GLuint bytes);
+
+GLboolean intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
+                                       struct _DriBufferObject *buffer,
+                                       GLuint flags,
+                                       GLuint mask, GLuint offset);
+
+/* Inline functions - might actually be better off with these
+ * non-inlined.  Certainly better off switching all command packets to
+ * be passed as structs rather than dwords, but that's a little bit of
+ * work...
+ */
+static INLINE GLuint
+intel_batchbuffer_space(struct intel_batchbuffer *batch)
+{
+   return (BATCH_SZ - BATCH_RESERVED) - (batch->ptr - batch->map);
 }
 
 
+static INLINE void
+intel_batchbuffer_emit_dword(struct intel_batchbuffer *batch, GLuint dword)
+{
+   assert(batch->map);
+   assert(intel_batchbuffer_space(batch) >= 4);
+   *(GLuint *) (batch->ptr) = dword;
+   batch->ptr += 4;
+}
+
+static INLINE void
+intel_batchbuffer_require_space(struct intel_batchbuffer *batch,
+                                GLuint sz, GLuint flags)
+{
+   assert(sz < BATCH_SZ - 8);
+   if (intel_batchbuffer_space(batch) < sz ||
+       (batch->flags != 0 && flags != 0 && batch->flags != flags))
+      intel_batchbuffer_flush(batch);
+
+   batch->flags |= flags;
+}
+
+/* Here are the crusty old macros, to be removed:
+ */
+#define BATCH_LOCALS
+
+#define BEGIN_BATCH(n, flags) do {				\
+   assert(!intel->prim.flush);					\
+   intel_batchbuffer_require_space(intel->batch, (n)*4, flags);	\
+} while (0)
+
+#define OUT_BATCH(d)  intel_batchbuffer_emit_dword(intel->batch, d)
+
+#define OUT_RELOC(buf,flags,mask,delta) do { 				\
+   assert((delta) >= 0);							\
+   intel_batchbuffer_emit_reloc(intel->batch, buf, flags, mask, delta);	\
+} while (0)
+
+#define ADVANCE_BATCH() do { } while(0)
+
 
 #endif

src/mesa/drivers/dri/i915/intel_batchpool.c

@@ -0,0 +1,416 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Thomas Hellstr<74>m <thomas-at-tungstengraphics-dot-com>
+ */
+
+#include <xf86drm.h>
+#include <stdlib.h>
+#include <errno.h>
+#include "imports.h"
+#include "glthread.h"
+#include "dri_bufpool.h"
+#include "dri_bufmgr.h"
+#include "intel_screen.h"
+
+typedef struct
+{
+   drmMMListHead head;
+   struct _BPool *parent;
+   struct _DriFenceObject *fence;
+   unsigned long start;
+   int unfenced;
+   int mapped;
+} BBuf;
+
+typedef struct _BPool
+{
+   _glthread_Mutex mutex;
+   unsigned long bufSize;
+   unsigned poolSize;
+   unsigned numFree;
+   unsigned numTot;
+   unsigned numDelayed;
+   unsigned checkDelayed;
+   drmMMListHead free;
+   drmMMListHead delayed;
+   drmMMListHead head;
+   drmBO kernelBO;
+   void *virtual;
+   BBuf *bufs;
+} BPool;
+
+
+static BPool *
+createBPool(int fd, unsigned long bufSize, unsigned numBufs, unsigned flags,
+            unsigned checkDelayed)
+{
+   BPool *p = (BPool *) malloc(sizeof(*p));
+   BBuf *buf;
+   int i;
+
+   if (!p)
+      return NULL;
+
+   p->bufs = (BBuf *) malloc(numBufs * sizeof(*p->bufs));
+   if (!p->bufs) {
+      free(p);
+      return NULL;
+   }
+
+   DRMINITLISTHEAD(&p->free);
+   DRMINITLISTHEAD(&p->head);
+   DRMINITLISTHEAD(&p->delayed);
+
+   p->numTot = numBufs;
+   p->numFree = numBufs;
+   p->bufSize = bufSize;
+   p->numDelayed = 0;
+   p->checkDelayed = checkDelayed;
+
+   _glthread_INIT_MUTEX(p->mutex);
+
+   if (drmBOCreate(fd, NULL, 0, numBufs * bufSize, NULL, drm_bo_type_dc,
+                   flags, 0, &p->kernelBO)) {
+      free(p->bufs);
+      free(p);
+      return NULL;
+   }
+   if (drmBOMap(fd, &p->kernelBO, DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0,
+                &p->virtual)) {
+      drmBODestroy(fd, &p->kernelBO);
+      free(p->bufs);
+      free(p);
+      return NULL;
+   }
+
+   /*
+    * We unmap the buffer so that we can validate it later. Note that this is
+    * just a synchronizing operation. The buffer will have a virtual mapping
+    * until it is destroyed.
+    */
+
+   drmBOUnmap(fd, &p->kernelBO);
+
+   buf = p->bufs;
+   for (i = 0; i < numBufs; ++i) {
+      buf->parent = p;
+      buf->fence = NULL;
+      buf->start = i * bufSize;
+      buf->mapped = 0;
+      buf->unfenced = 0;
+      DRMLISTADDTAIL(&buf->head, &p->free);
+      buf++;
+   }
+
+   return p;
+}
+
+
+static void
+pool_checkFree(BPool * p, int wait)
+{
+   drmMMListHead *list, *prev;
+   BBuf *buf;
+   int signaled = 0;
+   int i;
+
+   list = p->delayed.next;
+
+   if (p->numDelayed > 3) {
+      for (i = 0; i < p->numDelayed; i += 3) {
+         list = list->next;
+      }
+   }
+
+   prev = list->prev;
+   for (; list != &p->delayed; list = prev, prev = list->prev) {
+
+      buf = DRMLISTENTRY(BBuf, list, head);
+
+      if (!signaled) {
+         if (wait) {
+            driFenceFinish(buf->fence, DRM_FENCE_TYPE_EXE, 1);
+            signaled = 1;
+         }
+         else {
+            signaled = driFenceSignaled(buf->fence, DRM_FENCE_TYPE_EXE);
+         }
+      }
+
+      if (!signaled)
+         break;
+
+      driFenceUnReference(buf->fence);
+      buf->fence = NULL;
+      DRMLISTDEL(list);
+      p->numDelayed--;
+      DRMLISTADD(list, &p->free);
+      p->numFree++;
+   }
+}
+
+static void *
+pool_create(struct _DriBufferPool *pool,
+            unsigned long size, unsigned flags, unsigned hint,
+            unsigned alignment)
+{
+   BPool *p = (BPool *) pool->data;
+
+   drmMMListHead *item;
+
+   if (alignment && (alignment != 4096))
+      return NULL;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+
+   if (p->numFree == 0)
+      pool_checkFree(p, GL_TRUE);
+
+   if (p->numFree == 0) {
+      fprintf(stderr, "Out of fixed size buffer objects\n");
+      BM_CKFATAL(-ENOMEM);
+   }
+
+   item = p->free.next;
+
+   if (item == &p->free) {
+      fprintf(stderr, "Fixed size buffer pool corruption\n");
+   }
+
+   DRMLISTDEL(item);
+   --p->numFree;
+
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return (void *) DRMLISTENTRY(BBuf, item, head);
+}
+
+
+static int
+pool_destroy(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+
+   if (buf->fence) {
+      DRMLISTADDTAIL(&buf->head, &p->delayed);
+      p->numDelayed++;
+   }
+   else {
+      buf->unfenced = 0;
+      DRMLISTADD(&buf->head, &p->free);
+      p->numFree++;
+   }
+
+   if ((p->numDelayed % p->checkDelayed) == 0)
+      pool_checkFree(p, 0);
+
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return 0;
+}
+
+
+static int
+pool_map(struct _DriBufferPool *pool, void *private, unsigned flags,
+         int hint, void **virtual)
+{
+
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+
+   /*
+    * Currently Mesa doesn't have any condition variables to resolve this
+    * cleanly in a multithreading environment.
+    * We bail out instead.
+    */
+
+   if (buf->mapped) {
+      fprintf(stderr, "Trying to map already mapped buffer object\n");
+      BM_CKFATAL(-EINVAL);
+   }
+
+#if 0
+   if (buf->unfenced && !(hint & DRM_BO_HINT_ALLOW_UNFENCED_MAP)) {
+      fprintf(stderr, "Trying to map an unfenced buffer object 0x%08x"
+              " 0x%08x %d\n", hint, flags, buf->start);
+      BM_CKFATAL(-EINVAL);
+   }
+
+#endif
+
+   if (buf->fence) {
+      _glthread_UNLOCK_MUTEX(p->mutex);
+      return -EBUSY;
+   }
+
+   buf->mapped = GL_TRUE;
+   *virtual = (unsigned char *) p->virtual + buf->start;
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return 0;
+}
+
+static int
+pool_waitIdle(struct _DriBufferPool *pool, void *private, int lazy)
+{
+   BBuf *buf = (BBuf *) private;
+   driFenceFinish(buf->fence, 0, lazy);
+   return 0;
+}
+
+static int
+pool_unmap(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+
+   buf->mapped = 0;
+   return 0;
+}
+
+static unsigned long
+pool_offset(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   return p->kernelBO.offset + buf->start;
+}
+
+static unsigned
+pool_flags(struct _DriBufferPool *pool, void *private)
+{
+   BPool *p = (BPool *) pool->data;
+
+   return p->kernelBO.flags;
+}
+
+static unsigned long
+pool_size(struct _DriBufferPool *pool, void *private)
+{
+   BPool *p = (BPool *) pool->data;
+
+   return p->bufSize;
+}
+
+
+static int
+pool_fence(struct _DriBufferPool *pool, void *private,
+           struct _DriFenceObject *fence)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   _glthread_LOCK_MUTEX(p->mutex);
+   if (buf->fence) {
+      driFenceUnReference(buf->fence);
+   }
+   buf->fence = fence;
+   buf->unfenced = 0;
+   driFenceReference(buf->fence);
+   _glthread_UNLOCK_MUTEX(p->mutex);
+
+   return 0;
+}
+
+static drmBO *
+pool_kernel(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+
+   return &p->kernelBO;
+}
+
+static int
+pool_validate(struct _DriBufferPool *pool, void *private)
+{
+   BBuf *buf = (BBuf *) private;
+   BPool *p = buf->parent;
+   _glthread_LOCK_MUTEX(p->mutex);
+   buf->unfenced = GL_TRUE;
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   return 0;
+}
+
+static void
+pool_takedown(struct _DriBufferPool *pool)
+{
+   BPool *p = (BPool *) pool->data;
+
+   /*
+    * Wait on outstanding fences. 
+    */
+
+   _glthread_LOCK_MUTEX(p->mutex);
+   while ((p->numFree < p->numTot) && p->numDelayed) {
+      _glthread_UNLOCK_MUTEX(p->mutex);
+      sched_yield();
+      pool_checkFree(p, GL_TRUE);
+      _glthread_LOCK_MUTEX(p->mutex);
+   }
+
+   drmBODestroy(pool->fd, &p->kernelBO);
+   free(p->bufs);
+   _glthread_UNLOCK_MUTEX(p->mutex);
+   free(p);
+   free(pool);
+}
+
+
+struct _DriBufferPool *
+driBatchPoolInit(int fd, unsigned flags,
+                 unsigned long bufSize,
+                 unsigned numBufs, unsigned checkDelayed)
+{
+   struct _DriBufferPool *pool;
+
+   pool = (struct _DriBufferPool *) malloc(sizeof(*pool));
+   if (!pool)
+      return NULL;
+
+   pool->data = createBPool(fd, bufSize, numBufs, flags, checkDelayed);
+
+   pool->fd = fd;
+   pool->map = &pool_map;
+   pool->unmap = &pool_unmap;
+   pool->destroy = &pool_destroy;
+   pool->offset = &pool_offset;
+   pool->flags = &pool_flags;
+   pool->size = &pool_size;
+   pool->create = &pool_create;
+   pool->fence = &pool_fence;
+   pool->kernel = &pool_kernel;
+   pool->validate = &pool_validate;
+   pool->waitIdle = &pool_waitIdle;
+   pool->setstatic = NULL;
+   pool->takeDown = &pool_takedown;
+   return pool;
+}

src/mesa/drivers/dri/i915/intel_blit.c

@@ -0,0 +1,516 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include <stdio.h>
+#include <errno.h>
+
+#include "mtypes.h"
+#include "context.h"
+#include "enums.h"
+
+#include "intel_batchbuffer.h"
+#include "intel_blit.h"
+#include "intel_buffers.h"
+#include "intel_context.h"
+#include "intel_fbo.h"
+#include "intel_reg.h"
+#include "intel_regions.h"
+#include "vblank.h"
+
+#define FILE_DEBUG_FLAG DEBUG_BLIT
+
+/**
+ * Copy the back color buffer to the front color buffer. 
+ * Used for SwapBuffers().
+ */
+void
+intelCopyBuffer(const __DRIdrawablePrivate * dPriv,
+                const drm_clip_rect_t * rect)
+{
+
+   struct intel_context *intel;
+   GLboolean missed_target;
+   int64_t ust;
+
+   DBG("%s\n", __FUNCTION__);
+
+   assert(dPriv);
+
+   intel = intelScreenContext(dPriv->driScreenPriv->private);
+   if (!intel)
+      return;
+
+   if (intel->last_swap_fence) {
+      driFenceFinish(intel->last_swap_fence, DRM_FENCE_TYPE_EXE, GL_TRUE);
+      driFenceUnReference(intel->last_swap_fence);
+      intel->last_swap_fence = NULL;
+   }
+   intel->last_swap_fence = intel->first_swap_fence;
+   intel->first_swap_fence = NULL;
+
+   if (!rect) {
+      driWaitForVBlank(dPriv, &intel->vbl_seq, intel->vblank_flags,
+                       &missed_target);
+   }
+
+
+   /* The LOCK_HARDWARE is required for the cliprects.  Buffer offsets
+    * should work regardless.
+    */
+   LOCK_HARDWARE(intel);
+
+   if (intel->driDrawable && intel->driDrawable->numClipRects) {
+      const intelScreenPrivate *intelScreen = intel->intelScreen;
+      struct gl_framebuffer *fb
+         = (struct gl_framebuffer *) dPriv->driverPrivate;
+      const struct intel_region *frontRegion
+         = intel_get_rb_region(fb, BUFFER_FRONT_LEFT);
+      const struct intel_region *backRegion
+         = intel_get_rb_region(fb, BUFFER_BACK_LEFT);
+      const int nbox = dPriv->numClipRects;
+      const drm_clip_rect_t *pbox = dPriv->pClipRects;
+      const int pitch = frontRegion->pitch;
+      const int cpp = frontRegion->cpp;
+      int BR13, CMD;
+      int i;
+
+      ASSERT(fb);
+      ASSERT(fb->Name == 0);    /* Not a user-created FBO */
+      ASSERT(frontRegion);
+      ASSERT(backRegion);
+      ASSERT(frontRegion->pitch == backRegion->pitch);
+      ASSERT(frontRegion->cpp == backRegion->cpp);
+
+      if (cpp == 2) {
+         BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24);
+         CMD = XY_SRC_COPY_BLT_CMD;
+      }
+      else {
+         BR13 = (pitch * cpp) | (0xCC << 16) | (1 << 24) | (1 << 25);
+         CMD = (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
+                XY_SRC_COPY_BLT_WRITE_RGB);
+      }
+
+      for (i = 0; i < nbox; i++, pbox++) {
+         drm_clip_rect_t box;
+
+         if (pbox->x1 > pbox->x2 ||
+             pbox->y1 > pbox->y2 ||
+             pbox->x2 > intelScreen->width || pbox->y2 > intelScreen->height)
+            continue;
+
+         box = *pbox;
+
+         if (rect) {
+            if (rect->x1 > box.x1)
+               box.x1 = rect->x1;
+            if (rect->y1 > box.y1)
+               box.y1 = rect->y1;
+            if (rect->x2 < box.x2)
+               box.x2 = rect->x2;
+            if (rect->y2 < box.y2)
+               box.y2 = rect->y2;
+
+            if (box.x1 > box.x2 || box.y1 > box.y2)
+               continue;
+         }
+
+         BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+         OUT_BATCH(CMD);
+         OUT_BATCH(BR13);
+         OUT_BATCH((pbox->y1 << 16) | pbox->x1);
+         OUT_BATCH((pbox->y2 << 16) | pbox->x2);
+
+         if (intel->sarea->pf_current_page == 0)
+            OUT_RELOC(frontRegion->buffer,
+                      DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                      DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
+         else
+            OUT_RELOC(backRegion->buffer,
+                      DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                      DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, 0);
+         OUT_BATCH((pbox->y1 << 16) | pbox->x1);
+         OUT_BATCH(BR13 & 0xffff);
+
+         if (intel->sarea->pf_current_page == 0)
+            OUT_RELOC(backRegion->buffer,
+                      DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+                      DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
+         else
+            OUT_RELOC(frontRegion->buffer,
+                      DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+                      DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, 0);
+
+         ADVANCE_BATCH();
+      }
+
+      if (intel->first_swap_fence)
+         driFenceUnReference(intel->first_swap_fence);
+      intel->first_swap_fence = intel_batchbuffer_flush(intel->batch);
+      driFenceReference(intel->first_swap_fence);
+   }
+
+   UNLOCK_HARDWARE(intel);
+
+   if (!rect) {
+      intel->swap_count++;
+      (*dri_interface->getUST) (&ust);
+      if (missed_target) {
+         intel->swap_missed_count++;
+         intel->swap_missed_ust = ust - intel->swap_ust;
+      }
+
+      intel->swap_ust = ust;
+   }
+}
+
+
+
+
+void
+intelEmitFillBlit(struct intel_context *intel,
+                  GLuint cpp,
+                  GLshort dst_pitch,
+                  struct _DriBufferObject *dst_buffer,
+                  GLuint dst_offset,
+                  GLshort x, GLshort y, GLshort w, GLshort h, GLuint color)
+{
+   GLuint BR13, CMD;
+   BATCH_LOCALS;
+
+   dst_pitch *= cpp;
+
+   switch (cpp) {
+   case 1:
+   case 2:
+   case 3:
+      BR13 = dst_pitch | (0xF0 << 16) | (1 << 24);
+      CMD = XY_COLOR_BLT_CMD;
+      break;
+   case 4:
+      BR13 = dst_pitch | (0xF0 << 16) | (1 << 24) | (1 << 25);
+      CMD = (XY_COLOR_BLT_CMD | XY_COLOR_BLT_WRITE_ALPHA |
+             XY_COLOR_BLT_WRITE_RGB);
+      break;
+   default:
+      return;
+   }
+
+   DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+       __FUNCTION__, dst_buffer, dst_pitch, dst_offset, x, y, w, h);
+
+
+   BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
+   OUT_BATCH(CMD);
+   OUT_BATCH(BR13);
+   OUT_BATCH((y << 16) | x);
+   OUT_BATCH(((y + h) << 16) | (x + w));
+   OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+             DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, dst_offset);
+   OUT_BATCH(color);
+   ADVANCE_BATCH();
+}
+
+
+/* Copy BitBlt
+ */
+void
+intelEmitCopyBlit(struct intel_context *intel,
+                  GLuint cpp,
+                  GLshort src_pitch,
+                  struct _DriBufferObject *src_buffer,
+                  GLuint src_offset,
+                  GLshort dst_pitch,
+                  struct _DriBufferObject *dst_buffer,
+                  GLuint dst_offset,
+                  GLshort src_x, GLshort src_y,
+                  GLshort dst_x, GLshort dst_y, GLshort w, GLshort h)
+{
+   GLuint CMD, BR13;
+   int dst_y2 = dst_y + h;
+   int dst_x2 = dst_x + w;
+   BATCH_LOCALS;
+
+
+   DBG("%s src:buf(%p)/%d+%d %d,%d dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+       __FUNCTION__,
+       src_buffer, src_pitch, src_offset, src_x, src_y,
+       dst_buffer, dst_pitch, dst_offset, dst_x, dst_y, w, h);
+
+   src_pitch *= cpp;
+   dst_pitch *= cpp;
+
+   switch (cpp) {
+   case 1:
+   case 2:
+   case 3:
+      BR13 = (((GLint) dst_pitch) & 0xffff) | (0xCC << 16) | (1 << 24);
+      CMD = XY_SRC_COPY_BLT_CMD;
+      break;
+   case 4:
+      BR13 =
+         (((GLint) dst_pitch) & 0xffff) | (0xCC << 16) | (1 << 24) | (1 <<
+                                                                      25);
+      CMD =
+         (XY_SRC_COPY_BLT_CMD | XY_SRC_COPY_BLT_WRITE_ALPHA |
+          XY_SRC_COPY_BLT_WRITE_RGB);
+      break;
+   default:
+      return;
+   }
+
+   if (dst_y2 < dst_y || dst_x2 < dst_x) {
+      return;
+   }
+
+   /* Initial y values don't seem to work with negative pitches.  If
+    * we adjust the offsets manually (below), it seems to work fine.
+    *
+    * On the other hand, if we always adjust, the hardware doesn't
+    * know which blit directions to use, so overlapping copypixels get
+    * the wrong result.
+    */
+   if (dst_pitch > 0 && src_pitch > 0) {
+      BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+      OUT_BATCH(CMD);
+      OUT_BATCH(BR13);
+      OUT_BATCH((dst_y << 16) | dst_x);
+      OUT_BATCH((dst_y2 << 16) | dst_x2);
+      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE, dst_offset);
+      OUT_BATCH((src_y << 16) | src_x);
+      OUT_BATCH(((GLint) src_pitch & 0xffff));
+      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+                DRM_BO_MASK_MEM | DRM_BO_FLAG_READ, src_offset);
+      ADVANCE_BATCH();
+   }
+   else {
+      BEGIN_BATCH(8, INTEL_BATCH_NO_CLIPRECTS);
+      OUT_BATCH(CMD);
+      OUT_BATCH(BR13);
+      OUT_BATCH((0 << 16) | dst_x);
+      OUT_BATCH((h << 16) | dst_x2);
+      OUT_RELOC(dst_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE,
+                dst_offset + dst_y * dst_pitch);
+      OUT_BATCH((0 << 16) | src_x);
+      OUT_BATCH(((GLint) src_pitch & 0xffff));
+      OUT_RELOC(src_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_READ,
+                DRM_BO_MASK_MEM | DRM_BO_FLAG_READ,
+                src_offset + src_y * src_pitch);
+      ADVANCE_BATCH();
+   }
+}
+
+
+/**
+ * Use blitting to clear the renderbuffers named by 'flags'.
+ * Note: we can't use the ctx->DrawBuffer->_ColorDrawBufferMask field
+ * since that might include software renderbuffers or renderbuffers
+ * which we're clearing with triangles.
+ * \param mask  bitmask of BUFFER_BIT_* values indicating buffers to clear
+ */
+void
+intelClearWithBlit(GLcontext * ctx, GLbitfield mask, GLboolean all,
+                   GLint cx, GLint cy, GLint cw, GLint ch)
+{
+   struct intel_context *intel = intel_context(ctx);
+   GLuint clear_depth;
+   GLbitfield skipBuffers = 0;
+   BATCH_LOCALS;
+
+   DBG("%s %x\n", __FUNCTION__, mask);
+
+   /*
+    * Compute values for clearing the buffers.
+    */
+   clear_depth = 0;
+   if (mask & BUFFER_BIT_DEPTH) {
+      clear_depth = (GLuint) (ctx->DrawBuffer->_DepthMax * ctx->Depth.Clear);
+   }
+   if (mask & BUFFER_BIT_STENCIL) {
+      clear_depth |= (ctx->Stencil.Clear & 0xff) << 24;
+   }
+
+   /* If clearing both depth and stencil, skip BUFFER_BIT_STENCIL in
+    * the loop below.
+    */
+   if ((mask & BUFFER_BIT_DEPTH) && (mask & BUFFER_BIT_STENCIL)) {
+      skipBuffers = BUFFER_BIT_STENCIL;
+   }
+
+   /* XXX Move this flush/lock into the following conditional? */
+   intelFlush(&intel->ctx);
+   LOCK_HARDWARE(intel);
+
+   if (intel->numClipRects) {
+      drm_clip_rect_t clear;
+      int i;
+
+      /* Refresh the cx/y/w/h values as they may have been invalidated
+       * by a new window position or size picked up when we did
+       * LOCK_HARDWARE above.  The values passed by mesa are not
+       * reliable.
+       */
+      {
+         cx = ctx->DrawBuffer->_Xmin;
+         cy = ctx->DrawBuffer->_Ymin;
+         ch = ctx->DrawBuffer->_Ymax - ctx->DrawBuffer->_Ymin;
+         cw = ctx->DrawBuffer->_Xmax - ctx->DrawBuffer->_Xmin;
+      }
+
+      if (intel->ctx.DrawBuffer->Name == 0) {
+         /* clearing a window */
+
+         /* flip top to bottom */
+         clear.x1 = cx + intel->drawX;
+         clear.y1 = intel->driDrawable->y + intel->driDrawable->h - cy - ch;
+         clear.x2 = clear.x1 + cw;
+         clear.y2 = clear.y1 + ch;
+
+         /* adjust for page flipping */
+         if (intel->sarea->pf_current_page == 1) {
+            const GLuint tmp = mask;
+            mask &= ~(BUFFER_BIT_FRONT_LEFT | BUFFER_BIT_BACK_LEFT);
+            if (tmp & BUFFER_BIT_FRONT_LEFT)
+               mask |= BUFFER_BIT_BACK_LEFT;
+            if (tmp & BUFFER_BIT_BACK_LEFT)
+               mask |= BUFFER_BIT_FRONT_LEFT;
+         }
+      }
+      else {
+         /* clearing FBO */
+         ASSERT(intel->numClipRects == 1);
+         ASSERT(intel->pClipRects == &intel->fboRect);
+         clear.x1 = cx;
+         clear.y1 = intel->ctx.DrawBuffer->Height - cy - ch;
+         clear.x2 = clear.y1 + cw;
+         clear.y2 = clear.y1 + ch;
+         /* no change to mask */
+      }
+
+      for (i = 0; i < intel->numClipRects; i++) {
+         const drm_clip_rect_t *box = &intel->pClipRects[i];
+         drm_clip_rect_t b;
+         GLuint buf;
+         GLuint clearMask = mask;       /* use copy, since we modify it below */
+
+         if (!all) {
+            intel_intersect_cliprects(&b, &clear, box);
+         }
+         else {
+            b = *box;
+         }
+
+         if (0)
+            _mesa_printf("clear %d,%d..%d,%d, mask %x\n",
+                         b.x1, b.y1, b.x2, b.y2, mask);
+
+         /* Loop over all renderbuffers */
+         for (buf = 0; buf < BUFFER_COUNT && clearMask; buf++) {
+            const GLbitfield bufBit = 1 << buf;
+            if ((clearMask & bufBit) && !(bufBit & skipBuffers)) {
+               /* OK, clear this renderbuffer */
+               const struct intel_renderbuffer *irb
+                  = intel_renderbuffer(ctx->DrawBuffer->
+                                       Attachment[buf].Renderbuffer);
+               struct _DriBufferObject *write_buffer =
+                  intel_region_buffer(intel->intelScreen, irb->region,
+                                      all ? INTEL_WRITE_FULL :
+                                      INTEL_WRITE_PART);
+
+               GLuint clearVal;
+               GLint pitch, cpp;
+               GLuint BR13, CMD;
+
+               ASSERT(irb);
+               ASSERT(irb->region);
+
+               pitch = irb->region->pitch;
+               cpp = irb->region->cpp;
+
+               DBG("%s dst:buf(%p)/%d+%d %d,%d sz:%dx%d\n",
+                   __FUNCTION__,
+                   irb->region->buffer, (pitch * cpp),
+                   irb->region->draw_offset,
+                   b.x1, b.y1, b.x2 - b.x1, b.y2 - b.y1);
+
+
+               /* Setup the blit command */
+               if (cpp == 4) {
+                  BR13 = (0xF0 << 16) | (pitch * cpp) | (1 << 24) | (1 << 25);
+                  if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
+                     CMD = XY_COLOR_BLT_CMD;
+                     if (clearMask & BUFFER_BIT_DEPTH)
+                        CMD |= XY_COLOR_BLT_WRITE_RGB;
+                     if (clearMask & BUFFER_BIT_STENCIL)
+                        CMD |= XY_COLOR_BLT_WRITE_ALPHA;
+                  }
+                  else {
+                     /* clearing RGBA */
+                     CMD = (XY_COLOR_BLT_CMD |
+                            XY_COLOR_BLT_WRITE_ALPHA |
+                            XY_COLOR_BLT_WRITE_RGB);
+                  }
+               }
+               else {
+                  ASSERT(cpp == 2 || cpp == 0);
+                  BR13 = (0xF0 << 16) | (pitch * cpp) | (1 << 24);
+                  CMD = XY_COLOR_BLT_CMD;
+               }
+
+               if (buf == BUFFER_DEPTH || buf == BUFFER_STENCIL) {
+                  clearVal = clear_depth;
+               }
+               else {
+                  clearVal = (cpp == 4)
+                     ? intel->ClearColor8888 : intel->ClearColor565;
+               }
+               /*
+                  _mesa_debug(ctx, "hardware blit clear buf %d rb id %d\n",
+                  buf, irb->Base.Name);
+                */
+               BEGIN_BATCH(6, INTEL_BATCH_NO_CLIPRECTS);
+               OUT_BATCH(CMD);
+               OUT_BATCH(BR13);
+               OUT_BATCH((b.y1 << 16) | b.x1);
+               OUT_BATCH((b.y2 << 16) | b.x2);
+               OUT_RELOC(write_buffer, DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_WRITE,
+                         DRM_BO_MASK_MEM | DRM_BO_FLAG_WRITE,
+                         irb->region->draw_offset);
+               OUT_BATCH(clearVal);
+               ADVANCE_BATCH();
+               clearMask &= ~bufBit;    /* turn off bit, for faster loop exit */
+            }
+         }
+      }
+      intel_batchbuffer_flush(intel->batch);
+   }
+
+   UNLOCK_HARDWARE(intel);
+}

src/mesa/drivers/dri/i915/intel_blit.h

@@ -0,0 +1,63 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_BLIT_H
+#define INTEL_BLIT_H
+
+#include "intel_context.h"
+#include "intel_ioctl.h"
+#include "dri_bufmgr.h"
+
+extern void intelCopyBuffer(const __DRIdrawablePrivate * dpriv,
+                            const drm_clip_rect_t * rect);
+
+extern void intelClearWithBlit(GLcontext * ctx, GLbitfield mask,
+                               GLboolean all, GLint cx1, GLint cy1, GLint cw,
+                               GLint ch);
+
+extern void intelEmitCopyBlit(struct intel_context *intel,
+                              GLuint cpp,
+                              GLshort src_pitch,
+                              struct _DriBufferObject *src_buffer,
+                              GLuint src_offset,
+                              GLshort dst_pitch,
+                              struct _DriBufferObject *dst_buffer,
+                              GLuint dst_offset,
+                              GLshort srcx, GLshort srcy,
+                              GLshort dstx, GLshort dsty,
+                              GLshort w, GLshort h);
+
+extern void intelEmitFillBlit(struct intel_context *intel,
+                              GLuint cpp,
+                              GLshort dst_pitch,
+                              struct _DriBufferObject *dst_buffer,
+                              GLuint dst_offset,
+                              GLshort x, GLshort y,
+                              GLshort w, GLshort h, GLuint color);
+
+
+#endif

src/mesa/drivers/dri/i915/intel_buffer_objects.c

@@ -0,0 +1,250 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "imports.h"
+#include "mtypes.h"
+#include "bufferobj.h"
+
+#include "intel_context.h"
+#include "intel_buffer_objects.h"
+#include "intel_regions.h"
+#include "dri_bufmgr.h"
+
+/**
+ * There is some duplication between mesa's bufferobjects and our
+ * bufmgr buffers.  Both have an integer handle and a hashtable to
+ * lookup an opaque structure.  It would be nice if the handles and
+ * internal structure where somehow shared.
+ */
+static struct gl_buffer_object *
+intel_bufferobj_alloc(GLcontext * ctx, GLuint name, GLenum target)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *obj = CALLOC_STRUCT(intel_buffer_object);
+
+   _mesa_initialize_buffer_object(&obj->Base, name, target);
+
+   driGenBuffers(intel->intelScreen->regionPool,
+                 "bufferobj", 1, &obj->buffer, 64, 0, 0);
+
+   return &obj->Base;
+}
+
+
+/* Break the COW tie to the region.  The region gets to keep the data.
+ */
+void
+intel_bufferobj_release_region(struct intel_context *intel,
+                               struct intel_buffer_object *intel_obj)
+{
+   assert(intel_obj->region->buffer == intel_obj->buffer);
+   intel_obj->region->pbo = NULL;
+   intel_obj->region = NULL;
+   driBOUnReference(intel_obj->buffer);
+   intel_obj->buffer = NULL;
+
+   /* This leads to a large number of buffer deletion/creation events.
+    * Currently the drm doesn't like that:
+    */
+   driGenBuffers(intel->intelScreen->regionPool,
+                 "buffer object", 1, &intel_obj->buffer, 64, 0, 0);
+   driBOData(intel_obj->buffer, intel_obj->Base.Size, NULL, 0);
+}
+
+/* Break the COW tie to the region.  Both the pbo and the region end
+ * up with a copy of the data.
+ */
+void
+intel_bufferobj_cow(struct intel_context *intel,
+                    struct intel_buffer_object *intel_obj)
+{
+   assert(intel_obj->region);
+   intel_region_cow(intel->intelScreen, intel_obj->region);
+}
+
+
+/**
+ * Deallocate/free a vertex/pixel buffer object.
+ * Called via glDeleteBuffersARB().
+ */
+static void
+intel_bufferobj_free(GLcontext * ctx, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+
+   if (intel_obj->region) {
+      intel_bufferobj_release_region(intel, intel_obj);
+   }
+   else if (intel_obj->buffer) {
+      driDeleteBuffers(1, &intel_obj->buffer);
+   }
+
+   _mesa_free(intel_obj);
+}
+
+
+
+/**
+ * Allocate space for and store data in a buffer object.  Any data that was
+ * previously stored in the buffer object is lost.  If data is NULL,
+ * memory will be allocated, but no copy will occur.
+ * Called via glBufferDataARB().
+ */
+static void
+intel_bufferobj_data(GLcontext * ctx,
+                     GLenum target,
+                     GLsizeiptrARB size,
+                     const GLvoid * data,
+                     GLenum usage, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   intel_obj->Base.Size = size;
+   intel_obj->Base.Usage = usage;
+
+   if (intel_obj->region)
+      intel_bufferobj_release_region(intel, intel_obj);
+
+   driBOData(intel_obj->buffer, size, data, 0);
+}
+
+
+/**
+ * Replace data in a subrange of buffer object.  If the data range
+ * specified by size + offset extends beyond the end of the buffer or
+ * if data is NULL, no copy is performed.
+ * Called via glBufferSubDataARB().
+ */
+static void
+intel_bufferobj_subdata(GLcontext * ctx,
+                        GLenum target,
+                        GLintptrARB offset,
+                        GLsizeiptrARB size,
+                        const GLvoid * data, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+
+   if (intel_obj->region)
+      intel_bufferobj_cow(intel, intel_obj);
+
+   driBOSubData(intel_obj->buffer, offset, size, data);
+}
+
+
+/**
+ * Called via glGetBufferSubDataARB().
+ */
+static void
+intel_bufferobj_get_subdata(GLcontext * ctx,
+                            GLenum target,
+                            GLintptrARB offset,
+                            GLsizeiptrARB size,
+                            GLvoid * data, struct gl_buffer_object *obj)
+{
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+   driBOGetSubData(intel_obj->buffer, offset, size, data);
+}
+
+
+
+/**
+ * Called via glMapBufferARB().
+ */
+static void *
+intel_bufferobj_map(GLcontext * ctx,
+                    GLenum target,
+                    GLenum access, struct gl_buffer_object *obj)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   /* XXX: Translate access to flags arg below:
+    */
+   assert(intel_obj);
+
+   if (intel_obj->region)
+      intel_bufferobj_cow(intel, intel_obj);
+
+   obj->Pointer = driBOMap(intel_obj->buffer,
+                           DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0);
+   return obj->Pointer;
+}
+
+
+/**
+ * Called via glMapBufferARB().
+ */
+static GLboolean
+intel_bufferobj_unmap(GLcontext * ctx,
+                      GLenum target, struct gl_buffer_object *obj)
+{
+   struct intel_buffer_object *intel_obj = intel_buffer_object(obj);
+
+   assert(intel_obj);
+   assert(obj->Pointer);
+   driBOUnmap(intel_obj->buffer);
+   obj->Pointer = NULL;
+   return GL_TRUE;
+}
+
+struct _DriBufferObject *
+intel_bufferobj_buffer(struct intel_context *intel,
+                       struct intel_buffer_object *intel_obj, GLuint flag)
+{
+   if (intel_obj->region) {
+      if (flag == INTEL_WRITE_PART)
+         intel_bufferobj_cow(intel, intel_obj);
+      else if (flag == INTEL_WRITE_FULL)
+         intel_bufferobj_release_region(intel, intel_obj);
+   }
+
+   return intel_obj->buffer;
+}
+
+void
+intel_bufferobj_init(struct intel_context *intel)
+{
+   GLcontext *ctx = &intel->ctx;
+
+   ctx->Driver.NewBufferObject = intel_bufferobj_alloc;
+   ctx->Driver.DeleteBuffer = intel_bufferobj_free;
+   ctx->Driver.BufferData = intel_bufferobj_data;
+   ctx->Driver.BufferSubData = intel_bufferobj_subdata;
+   ctx->Driver.GetBufferSubData = intel_bufferobj_get_subdata;
+   ctx->Driver.MapBuffer = intel_bufferobj_map;
+   ctx->Driver.UnmapBuffer = intel_bufferobj_unmap;
+}

src/mesa/drivers/dri/i915/intel_buffer_objects.h

@@ -0,0 +1,86 @@
+ /**************************************************************************
+ * 
+ * Copyright 2005 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_BUFFEROBJ_H
+#define INTEL_BUFFEROBJ_H
+
+#include "mtypes.h"
+
+struct intel_context;
+struct intel_region;
+struct gl_buffer_object;
+
+
+/**
+ * Intel vertex/pixel buffer object, derived from Mesa's gl_buffer_object.
+ */
+struct intel_buffer_object
+{
+   struct gl_buffer_object Base;
+   struct _DriBufferObject *buffer;     /* the low-level buffer manager's buffer handle */
+
+   struct intel_region *region; /* Is there a zero-copy texture
+                                   associated with this (pixel)
+                                   buffer object? */
+};
+
+
+/* Get the bm buffer associated with a GL bufferobject:
+ */
+struct _DriBufferObject *intel_bufferobj_buffer(struct intel_context *intel,
+                                                struct intel_buffer_object
+                                                *obj, GLuint flag);
+
+/* Hook the bufferobject implementation into mesa: 
+ */
+void intel_bufferobj_init(struct intel_context *intel);
+
+
+
+/* Are the obj->Name tests necessary?  Unfortunately yes, mesa
+ * allocates a couple of gl_buffer_object structs statically, and
+ * the Name == 0 test is the only way to identify them and avoid
+ * casting them erroneously to our structs.
+ */
+static INLINE struct intel_buffer_object *
+intel_buffer_object(struct gl_buffer_object *obj)
+{
+   if (obj->Name)
+      return (struct intel_buffer_object *) obj;
+   else
+      return NULL;
+}
+
+/* Helpers for zerocopy image uploads.  See also intel_regions.h:
+ */
+void intel_bufferobj_cow(struct intel_context *intel,
+                         struct intel_buffer_object *intel_obj);
+void intel_bufferobj_release_region(struct intel_context *intel,
+                                    struct intel_buffer_object *intel_obj);
+
+
+#endif

src/mesa/drivers/dri/i915/intel_buffers.c

@@ -0,0 +1,968 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_blit.h"
+#include "intel_buffers.h"
+#include "intel_depthstencil.h"
+#include "intel_fbo.h"
+#include "intel_tris.h"
+#include "intel_regions.h"
+#include "intel_batchbuffer.h"
+#include "context.h"
+#include "framebuffer.h"
+#include "swrast/swrast.h"
+
+
+/**
+ * XXX move this into a new dri/common/cliprects.c file.
+ */
+GLboolean
+intel_intersect_cliprects(drm_clip_rect_t * dst,
+                          const drm_clip_rect_t * a,
+                          const drm_clip_rect_t * b)
+{
+   GLint bx = b->x1;
+   GLint by = b->y1;
+   GLint bw = b->x2 - bx;
+   GLint bh = b->y2 - by;
+
+   if (bx < a->x1)
+      bw -= a->x1 - bx, bx = a->x1;
+   if (by < a->y1)
+      bh -= a->y1 - by, by = a->y1;
+   if (bx + bw > a->x2)
+      bw = a->x2 - bx;
+   if (by + bh > a->y2)
+      bh = a->y2 - by;
+   if (bw <= 0)
+      return GL_FALSE;
+   if (bh <= 0)
+      return GL_FALSE;
+
+   dst->x1 = bx;
+   dst->y1 = by;
+   dst->x2 = bx + bw;
+   dst->y2 = by + bh;
+
+   return GL_TRUE;
+}
+
+/**
+ * Return pointer to current color drawing region, or NULL.
+ */
+struct intel_region *
+intel_drawbuf_region(struct intel_context *intel)
+{
+   struct intel_renderbuffer *irbColor =
+      intel_renderbuffer(intel->ctx.DrawBuffer->_ColorDrawBuffers[0][0]);
+   if (irbColor)
+      return irbColor->region;
+   else
+      return NULL;
+}
+
+/**
+ * Return pointer to current color reading region, or NULL.
+ */
+struct intel_region *
+intel_readbuf_region(struct intel_context *intel)
+{
+   struct intel_renderbuffer *irb
+      = intel_renderbuffer(intel->ctx.ReadBuffer->_ColorReadBuffer);
+   if (irb)
+      return irb->region;
+   else
+      return NULL;
+}
+
+
+
+static void
+intelBufferSize(GLframebuffer * buffer, GLuint * width, GLuint * height)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
+   /* Need to lock to make sure the driDrawable is uptodate.  This
+    * information is used to resize Mesa's software buffers, so it has
+    * to be correct.
+    */
+   /* XXX This isn't 100% correct, the given buffer might not be
+    * bound to the current context!
+    */
+   LOCK_HARDWARE(intel);
+   if (intel->driDrawable) {
+      *width = intel->driDrawable->w;
+      *height = intel->driDrawable->h;
+   }
+   else {
+      *width = 0;
+      *height = 0;
+   }
+   UNLOCK_HARDWARE(intel);
+}
+
+
+
+/**
+ * Update the following fields for rendering to a user-created FBO:
+ *   intel->numClipRects
+ *   intel->pClipRects
+ *   intel->drawX
+ *   intel->drawY
+ */
+static void
+intelSetRenderbufferClipRects(struct intel_context *intel)
+{
+   assert(intel->ctx.DrawBuffer->Width > 0);
+   assert(intel->ctx.DrawBuffer->Height > 0);
+   intel->fboRect.x1 = 0;
+   intel->fboRect.y1 = 0;
+   intel->fboRect.x2 = intel->ctx.DrawBuffer->Width;
+   intel->fboRect.y2 = intel->ctx.DrawBuffer->Height;
+   intel->numClipRects = 1;
+   intel->pClipRects = &intel->fboRect;
+   intel->drawX = 0;
+   intel->drawY = 0;
+}
+
+
+/**
+ * As above, but for rendering to front buffer of a window.
+ * \sa intelSetRenderbufferClipRects
+ */
+static void
+intelSetFrontClipRects(struct intel_context *intel)
+{
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+
+   if (!dPriv)
+      return;
+
+   intel->numClipRects = dPriv->numClipRects;
+   intel->pClipRects = dPriv->pClipRects;
+   intel->drawX = dPriv->x;
+   intel->drawY = dPriv->y;
+}
+
+
+/**
+ * As above, but for rendering to back buffer of a window.
+ */
+static void
+intelSetBackClipRects(struct intel_context *intel)
+{
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+
+   if (!dPriv)
+      return;
+
+   if (intel->sarea->pf_enabled == 0 && dPriv->numBackClipRects == 0) {
+      /* use the front clip rects */
+      intel->numClipRects = dPriv->numClipRects;
+      intel->pClipRects = dPriv->pClipRects;
+      intel->drawX = dPriv->x;
+      intel->drawY = dPriv->y;
+   }
+   else {
+      /* use the back clip rects */
+      intel->numClipRects = dPriv->numBackClipRects;
+      intel->pClipRects = dPriv->pBackClipRects;
+      intel->drawX = dPriv->backX;
+      intel->drawY = dPriv->backY;
+
+      if (dPriv->numBackClipRects == 1 &&
+          dPriv->x == dPriv->backX && dPriv->y == dPriv->backY) {
+
+         /* Repeat the calculation of the back cliprect dimensions here
+          * as early versions of dri.a in the Xserver are incorrect.  Try
+          * very hard not to restrict future versions of dri.a which
+          * might eg. allocate truly private back buffers.
+          */
+         int x1, y1;
+         int x2, y2;
+
+         x1 = dPriv->x;
+         y1 = dPriv->y;
+         x2 = dPriv->x + dPriv->w;
+         y2 = dPriv->y + dPriv->h;
+
+         if (x1 < 0)
+            x1 = 0;
+         if (y1 < 0)
+            y1 = 0;
+         if (x2 > intel->intelScreen->width)
+            x2 = intel->intelScreen->width;
+         if (y2 > intel->intelScreen->height)
+            y2 = intel->intelScreen->height;
+
+         if (x1 == dPriv->pBackClipRects[0].x1 &&
+             y1 == dPriv->pBackClipRects[0].y1) {
+
+            dPriv->pBackClipRects[0].x2 = x2;
+            dPriv->pBackClipRects[0].y2 = y2;
+         }
+      }
+   }
+}
+
+
+/**
+ * This will be called whenever the currently bound window is moved/resized.
+ * XXX: actually, it seems to NOT be called when the window is only moved (BP).
+ */
+void
+intelWindowMoved(struct intel_context *intel)
+{
+   GLcontext *ctx = &intel->ctx;
+
+   if (!intel->ctx.DrawBuffer) {
+      /* when would this happen? -BP */
+      intelSetFrontClipRects(intel);
+   }
+   else if (intel->ctx.DrawBuffer->Name != 0) {
+      /* drawing to user-created FBO - do nothing */
+      /* Cliprects would be set from intelDrawBuffer() */
+   }
+   else {
+      /* drawing to a window */
+      switch (intel->ctx.DrawBuffer->_ColorDrawBufferMask[0]) {
+      case BUFFER_BIT_FRONT_LEFT:
+         intelSetFrontClipRects(intel);
+         break;
+      case BUFFER_BIT_BACK_LEFT:
+         intelSetBackClipRects(intel);
+         break;
+      default:
+         /* glDrawBuffer(GL_NONE or GL_FRONT_AND_BACK): software fallback */
+         intelSetFrontClipRects(intel);
+      }
+   }
+
+   /* this update Mesa's notion of window size */
+   if (ctx->WinSysDrawBuffer) {
+      _mesa_resize_framebuffer(ctx, ctx->WinSysDrawBuffer,
+                               intel->driDrawable->w, intel->driDrawable->h);
+   }
+
+   /* Update hardware scissor */
+   ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y,
+                       ctx->Scissor.Width, ctx->Scissor.Height);
+}
+
+
+
+/* A true meta version of this would be very simple and additionally
+ * machine independent.  Maybe we'll get there one day.
+ */
+static void
+intelClearWithTris(struct intel_context *intel,
+                   GLbitfield mask,
+                   GLboolean all, GLint cx, GLint cy, GLint cw, GLint ch)
+{
+   GLcontext *ctx = &intel->ctx;
+   drm_clip_rect_t clear;
+
+   if (INTEL_DEBUG & DEBUG_BLIT)
+      _mesa_printf("%s 0x%x\n", __FUNCTION__, mask);
+
+   LOCK_HARDWARE(intel);
+
+   /* XXX FBO: was: intel->driDrawable->numClipRects */
+   if (intel->numClipRects) {
+      GLuint buf;
+
+      intel->vtbl.install_meta_state(intel);
+
+
+      /* Refresh the cx/y/w/h values as they may have been invalidated
+       * by a new window position or size picked up when we did
+       * LOCK_HARDWARE above.  The values passed by mesa are not
+       * reliable.
+       */
+      {
+         cx = ctx->DrawBuffer->_Xmin;
+         cy = ctx->DrawBuffer->_Ymin;
+         ch = ctx->DrawBuffer->_Ymax - ctx->DrawBuffer->_Ymin;
+         cw = ctx->DrawBuffer->_Xmax - ctx->DrawBuffer->_Xmin;
+      }
+
+      /* note: regardless of 'all', cx, cy, cw, ch are now correct */
+      clear.x1 = cx;
+      clear.y1 = cy;
+      clear.x2 = cx + cw;
+      clear.y2 = cy + ch;
+
+      /* Back and stencil cliprects are the same.  Try and do both
+       * buffers at once:
+       */
+      if (mask &
+          (BUFFER_BIT_BACK_LEFT | BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH)) {
+         struct intel_region *backRegion =
+            intel_get_rb_region(ctx->DrawBuffer, BUFFER_BACK_LEFT);
+         struct intel_region *depthRegion =
+            intel_get_rb_region(ctx->DrawBuffer, BUFFER_DEPTH);
+         const GLuint clearColor = (backRegion && backRegion->cpp == 4)
+            ? intel->ClearColor8888 : intel->ClearColor565;
+
+         intel->vtbl.meta_draw_region(intel, backRegion, depthRegion);
+
+         if (mask & BUFFER_BIT_BACK_LEFT)
+            intel->vtbl.meta_color_mask(intel, GL_TRUE);
+         else
+            intel->vtbl.meta_color_mask(intel, GL_FALSE);
+
+         if (mask & BUFFER_BIT_STENCIL)
+            intel->vtbl.meta_stencil_replace(intel,
+                                             intel->ctx.Stencil.WriteMask[0],
+                                             intel->ctx.Stencil.Clear);
+         else
+            intel->vtbl.meta_no_stencil_write(intel);
+
+         if (mask & BUFFER_BIT_DEPTH)
+            intel->vtbl.meta_depth_replace(intel);
+         else
+            intel->vtbl.meta_no_depth_write(intel);
+
+         /* XXX: Using INTEL_BATCH_NO_CLIPRECTS here is dangerous as the
+          * drawing origin may not be correctly emitted.
+          */
+         intel_meta_draw_quad(intel, clear.x1, clear.x2, clear.y1, clear.y2, intel->ctx.Depth.Clear, clearColor, 0, 0, 0, 0);   /* texcoords */
+
+         mask &=
+            ~(BUFFER_BIT_BACK_LEFT | BUFFER_BIT_STENCIL | BUFFER_BIT_DEPTH);
+      }
+
+      /* clear the remaining (color) renderbuffers */
+      for (buf = 0; buf < BUFFER_COUNT && mask; buf++) {
+         const GLuint bufBit = 1 << buf;
+         if (mask & bufBit) {
+            struct intel_renderbuffer *irbColor =
+               intel_renderbuffer(ctx->DrawBuffer->
+                                  Attachment[buf].Renderbuffer);
+            GLuint color = (irbColor->region->cpp == 4)
+               ? intel->ClearColor8888 : intel->ClearColor565;
+
+            ASSERT(irbColor);
+
+            intel->vtbl.meta_no_depth_write(intel);
+            intel->vtbl.meta_no_stencil_write(intel);
+            intel->vtbl.meta_color_mask(intel, GL_TRUE);
+            intel->vtbl.meta_draw_region(intel, irbColor->region, NULL);
+
+            /* XXX: Using INTEL_BATCH_NO_CLIPRECTS here is dangerous as the
+             * drawing origin may not be correctly emitted.
+             */
+            intel_meta_draw_quad(intel, clear.x1, clear.x2, clear.y1, clear.y2, 0,      /* depth clear val */
+                                 color, 0, 0, 0, 0);    /* texcoords */
+
+            mask &= ~bufBit;
+         }
+      }
+
+      intel->vtbl.leave_meta_state(intel);
+      intel_batchbuffer_flush(intel->batch);
+   }
+   UNLOCK_HARDWARE(intel);
+}
+
+
+
+
+/**
+ * Copy the window contents named by dPriv to the rotated (or reflected)
+ * color buffer.
+ * srcBuf is BUFFER_BIT_FRONT_LEFT or BUFFER_BIT_BACK_LEFT to indicate the source.
+ */
+void
+intelRotateWindow(struct intel_context *intel,
+                  __DRIdrawablePrivate * dPriv, GLuint srcBuf)
+{
+   intelScreenPrivate *screen = intel->intelScreen;
+   drm_clip_rect_t fullRect;
+   struct intel_region *src;
+   const drm_clip_rect_t *clipRects;
+   int numClipRects;
+   int i;
+   GLenum format, type;
+
+   int xOrig, yOrig;
+   int origNumClipRects;
+   drm_clip_rect_t *origRects;
+
+   /*
+    * set up hardware state
+    */
+   intelFlush(&intel->ctx);
+
+   LOCK_HARDWARE(intel);
+
+   if (!intel->numClipRects) {
+      UNLOCK_HARDWARE(intel);
+      return;
+   }
+
+   intel->vtbl.install_meta_state(intel);
+
+   intel->vtbl.meta_no_depth_write(intel);
+   intel->vtbl.meta_no_stencil_write(intel);
+   intel->vtbl.meta_color_mask(intel, GL_FALSE);
+
+
+   /* save current drawing origin and cliprects (restored at end) */
+   xOrig = intel->drawX;
+   yOrig = intel->drawY;
+   origNumClipRects = intel->numClipRects;
+   origRects = intel->pClipRects;
+
+   /*
+    * set drawing origin, cliprects for full-screen access to rotated screen
+    */
+   fullRect.x1 = 0;
+   fullRect.y1 = 0;
+   fullRect.x2 = screen->rotatedWidth;
+   fullRect.y2 = screen->rotatedHeight;
+   intel->drawX = 0;
+   intel->drawY = 0;
+   intel->numClipRects = 1;
+   intel->pClipRects = &fullRect;
+
+   intel->vtbl.meta_draw_region(intel, screen->rotated_region, NULL);    /* ? */
+
+   if (srcBuf == BUFFER_BIT_FRONT_LEFT) {
+      src = intel->intelScreen->front_region;
+      clipRects = dPriv->pClipRects;
+      numClipRects = dPriv->numClipRects;
+   }
+   else {
+      src = intel->intelScreen->back_region;
+      clipRects = dPriv->pBackClipRects;
+      numClipRects = dPriv->numBackClipRects;
+   }
+
+   if (src->cpp == 4) {
+      format = GL_BGRA;
+      type = GL_UNSIGNED_BYTE;
+   }
+   else {
+      format = GL_BGR;
+      type = GL_UNSIGNED_SHORT_5_6_5_REV;
+   }
+
+   /* set the whole screen up as a texture to avoid alignment issues */
+   intel->vtbl.meta_tex_rect_source(intel,
+                                    src->buffer,
+                                    screen->width,
+                                    screen->height, src->pitch, format, type);
+
+   intel->vtbl.meta_texture_blend_replace(intel);
+
+   /*
+    * loop over the source window's cliprects
+    */
+   for (i = 0; i < numClipRects; i++) {
+      int srcX0 = clipRects[i].x1;
+      int srcY0 = clipRects[i].y1;
+      int srcX1 = clipRects[i].x2;
+      int srcY1 = clipRects[i].y2;
+      GLfloat verts[4][2], tex[4][2];
+      int j;
+
+      /* build vertices for four corners of clip rect */
+      verts[0][0] = srcX0;
+      verts[0][1] = srcY0;
+      verts[1][0] = srcX1;
+      verts[1][1] = srcY0;
+      verts[2][0] = srcX1;
+      verts[2][1] = srcY1;
+      verts[3][0] = srcX0;
+      verts[3][1] = srcY1;
+
+      /* .. and texcoords */
+      tex[0][0] = srcX0;
+      tex[0][1] = srcY0;
+      tex[1][0] = srcX1;
+      tex[1][1] = srcY0;
+      tex[2][0] = srcX1;
+      tex[2][1] = srcY1;
+      tex[3][0] = srcX0;
+      tex[3][1] = srcY1;
+
+      /* transform coords to rotated screen coords */
+
+      for (j = 0; j < 4; j++) {
+         matrix23TransformCoordf(&screen->rotMatrix,
+                                 &verts[j][0], &verts[j][1]);
+      }
+
+      /* draw polygon to map source image to dest region */
+      intel_meta_draw_poly(intel, 4, verts, 0, 0, tex);
+
+   }                            /* cliprect loop */
+
+   intel->vtbl.leave_meta_state(intel);
+   intel_batchbuffer_flush(intel->batch);
+
+   /* restore original drawing origin and cliprects */
+   intel->drawX = xOrig;
+   intel->drawY = yOrig;
+   intel->numClipRects = origNumClipRects;
+   intel->pClipRects = origRects;
+
+   UNLOCK_HARDWARE(intel);
+}
+
+
+/**
+ * Called by ctx->Driver.Clear.
+ */
+static void
+intelClear(GLcontext * ctx,
+           GLbitfield mask,
+           GLboolean all, GLint cx, GLint cy, GLint cw, GLint ch)
+{
+   struct intel_context *intel = intel_context(ctx);
+   const GLuint colorMask = *((GLuint *) & ctx->Color.ColorMask);
+   GLbitfield tri_mask = 0;
+   GLbitfield blit_mask = 0;
+   GLbitfield swrast_mask = 0;
+   GLuint i;
+
+   if (0)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   /* HW color buffers (front, back, aux, generic FBO, etc) */
+   if (colorMask == ~0) {
+      /* clear all R,G,B,A */
+      /* XXX FBO: need to check if colorbuffers are software RBOs! */
+      blit_mask |= (mask & BUFFER_BITS_COLOR);
+   }
+   else {
+      /* glColorMask in effect */
+      tri_mask |= (mask & BUFFER_BITS_COLOR);
+   }
+
+   /* HW stencil */
+   if (mask & BUFFER_BIT_STENCIL) {
+      const struct intel_region *stencilRegion
+         = intel_get_rb_region(ctx->DrawBuffer, BUFFER_STENCIL);
+      if (stencilRegion) {
+         /* have hw stencil */
+         if ((ctx->Stencil.WriteMask[0] & 0xff) != 0xff) {
+            /* not clearing all stencil bits, so use triangle clearing */
+            tri_mask |= BUFFER_BIT_STENCIL;
+         }
+         else {
+            /* clearing all stencil bits, use blitting */
+            blit_mask |= BUFFER_BIT_STENCIL;
+         }
+      }
+   }
+
+   /* HW depth */
+   if (mask & BUFFER_BIT_DEPTH) {
+      /* clear depth with whatever method is used for stencil (see above) */
+      if (tri_mask & BUFFER_BIT_STENCIL)
+         tri_mask |= BUFFER_BIT_DEPTH;
+      else
+         blit_mask |= BUFFER_BIT_DEPTH;
+   }
+
+   /* SW fallback clearing */
+   swrast_mask = mask & ~tri_mask & ~blit_mask;
+
+   for (i = 0; i < BUFFER_COUNT; i++) {
+      GLuint bufBit = 1 << i;
+      if ((blit_mask | tri_mask) & bufBit) {
+         if (!ctx->DrawBuffer->Attachment[i].Renderbuffer->ClassID) {
+            blit_mask &= ~bufBit;
+            tri_mask &= ~bufBit;
+            swrast_mask |= bufBit;
+         }
+      }
+   }
+
+
+   intelFlush(ctx);             /* XXX intelClearWithBlit also does this */
+
+   if (blit_mask)
+      intelClearWithBlit(ctx, blit_mask, all, cx, cy, cw, ch);
+
+   if (tri_mask)
+      intelClearWithTris(intel, tri_mask, all, cx, cy, cw, ch);
+
+   if (swrast_mask)
+      _swrast_Clear(ctx, swrast_mask, all, cx, cy, cw, ch);
+}
+
+
+
+/* Flip the front & back buffers
+ */
+static void
+intelPageFlip(const __DRIdrawablePrivate * dPriv)
+{
+#if 0
+   struct intel_context *intel;
+   int tmp, ret;
+
+   if (INTEL_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   assert(dPriv);
+   assert(dPriv->driContextPriv);
+   assert(dPriv->driContextPriv->driverPrivate);
+
+   intel = (struct intel_context *) dPriv->driContextPriv->driverPrivate;
+
+   intelFlush(&intel->ctx);
+   LOCK_HARDWARE(intel);
+
+   if (dPriv->pClipRects) {
+      *(drm_clip_rect_t *) intel->sarea->boxes = dPriv->pClipRects[0];
+      intel->sarea->nbox = 1;
+   }
+
+   ret = drmCommandNone(intel->driFd, DRM_I830_FLIP);
+   if (ret) {
+      fprintf(stderr, "%s: %d\n", __FUNCTION__, ret);
+      UNLOCK_HARDWARE(intel);
+      exit(1);
+   }
+
+   tmp = intel->sarea->last_enqueue;
+   intelRefillBatchLocked(intel);
+   UNLOCK_HARDWARE(intel);
+
+
+   intelSetDrawBuffer(&intel->ctx, intel->ctx.Color.DriverDrawBuffer);
+#endif
+}
+
+#if 0
+void
+intelSwapBuffers(__DRIdrawablePrivate * dPriv)
+{
+   if (dPriv->driverPrivate) {
+      const struct gl_framebuffer *fb
+         = (struct gl_framebuffer *) dPriv->driverPrivate;
+      if (fb->Visual.doubleBufferMode) {
+         GET_CURRENT_CONTEXT(ctx);
+         if (ctx && ctx->DrawBuffer == fb) {
+            _mesa_notifySwapBuffers(ctx);       /* flush pending rendering */
+         }
+         if (0 /*intel->doPageFlip */ ) {       /* doPageFlip is never set !!! */
+            intelPageFlip(dPriv);
+         }
+         else {
+            intelCopyBuffer(dPriv);
+         }
+      }
+   }
+   else {
+      _mesa_problem(NULL,
+                    "dPriv has no gl_framebuffer pointer in intelSwapBuffers");
+   }
+}
+#else
+/* Trunk version:
+ */
+void
+intelSwapBuffers(__DRIdrawablePrivate * dPriv)
+{
+   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+      GET_CURRENT_CONTEXT(ctx);
+      struct intel_context *intel;
+
+      if (ctx == NULL)
+	 return;
+
+      intel = intel_context(ctx);
+
+      if (ctx->Visual.doubleBufferMode) {
+         intelScreenPrivate *screen = intel->intelScreen;
+         _mesa_notifySwapBuffers(ctx);  /* flush pending rendering comands */
+         if (0 /*intel->doPageFlip */ ) {       /* doPageFlip is never set !!! */
+            intelPageFlip(dPriv);
+         }
+         else {
+            intelCopyBuffer(dPriv, NULL);
+         }
+         if (screen->current_rotation != 0) {
+            intelRotateWindow(intel, dPriv, BUFFER_BIT_FRONT_LEFT);
+         }
+      }
+   }
+   else {
+      /* XXX this shouldn't be an error but we can't handle it for now */
+      fprintf(stderr, "%s: drawable has no context!\n", __FUNCTION__);
+   }
+}
+#endif
+
+void
+intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h)
+{
+   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+      struct intel_context *intel =
+         (struct intel_context *) dPriv->driContextPriv->driverPrivate;
+      GLcontext *ctx = &intel->ctx;
+
+      if (ctx->Visual.doubleBufferMode) {
+         drm_clip_rect_t rect;
+         rect.x1 = x + dPriv->x;
+         rect.y1 = (dPriv->h - y - h) + dPriv->y;
+         rect.x2 = rect.x1 + w;
+         rect.y2 = rect.y1 + h;
+         _mesa_notifySwapBuffers(ctx);  /* flush pending rendering comands */
+         intelCopyBuffer(dPriv, &rect);
+      }
+   }
+   else {
+      /* XXX this shouldn't be an error but we can't handle it for now */
+      fprintf(stderr, "%s: drawable has no context!\n", __FUNCTION__);
+   }
+}
+
+
+/**
+ * Update the hardware state for drawing into a window or framebuffer object.
+ *
+ * Called by glDrawBuffer, glBindFramebufferEXT, MakeCurrent, and other
+ * places within the driver.
+ *
+ * Basically, this needs to be called any time the current framebuffer
+ * changes, the renderbuffers change, or we need to draw into different
+ * color buffers.
+ */
+void
+intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_region *colorRegion, *depthRegion = NULL;
+   struct intel_renderbuffer *irbDepth = NULL, *irbStencil = NULL;
+   int front = 0;               /* drawing to front color buffer? */
+
+   if (!fb) {
+      /* this can happen during the initial context initialization */
+      return;
+   }
+
+   /* Do this here, note core Mesa, since this function is called from
+    * many places within the driver.
+    */
+   if (ctx->NewState & (_NEW_BUFFERS | _NEW_COLOR | _NEW_PIXEL)) {
+      /* this updates the DrawBuffer->_NumColorDrawBuffers fields, etc */
+      _mesa_update_framebuffer(ctx);
+      /* this updates the DrawBuffer's Width/Height if it's a FBO */
+      _mesa_update_draw_buffer_bounds(ctx);
+   }
+
+   if (fb->_Status != GL_FRAMEBUFFER_COMPLETE_EXT) {
+      /* this may occur when we're called by glBindFrameBuffer() during
+       * the process of someone setting up renderbuffers, etc.
+       */
+      /*_mesa_debug(ctx, "DrawBuffer: incomplete user FBO\n");*/
+      return;
+   }
+
+   if (fb->Name)
+      intel_validate_paired_depth_stencil(ctx, fb);
+
+   /*
+    * How many color buffers are we drawing into?
+    */
+   if (fb->_NumColorDrawBuffers[0] != 1
+#if 0
+       /* XXX FBO temporary - always use software rendering */
+       || 1
+#endif
+      ) {
+      /* writing to 0 or 2 or 4 color buffers */
+      /*_mesa_debug(ctx, "Software rendering\n");*/
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE);
+      front = 1;                /* might not have back color buffer */
+   }
+   else {
+      /* draw to exactly one color buffer */
+      /*_mesa_debug(ctx, "Hardware rendering\n");*/
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE);
+      if (fb->_ColorDrawBufferMask[0] == BUFFER_BIT_FRONT_LEFT) {
+         front = 1;
+      }
+   }
+
+   /*
+    * Get the intel_renderbuffer for the colorbuffer we're drawing into.
+    * And set up cliprects.
+    */
+   if (fb->Name == 0) {
+      /* drawing to window system buffer */
+      if (intel->sarea->pf_current_page == 1) {
+         /* page flipped back/front */
+         front ^= 1;
+      }
+      if (front) {
+         intelSetFrontClipRects(intel);
+         colorRegion = intel_get_rb_region(fb, BUFFER_FRONT_LEFT);
+      }
+      else {
+         intelSetBackClipRects(intel);
+         colorRegion = intel_get_rb_region(fb, BUFFER_BACK_LEFT);
+      }
+   }
+   else {
+      /* drawing to user-created FBO */
+      struct intel_renderbuffer *irb;
+      intelSetRenderbufferClipRects(intel);
+      irb = intel_renderbuffer(fb->_ColorDrawBuffers[0][0]);
+      colorRegion = (irb && irb->region) ? irb->region : NULL;
+   }
+
+   /* Update culling direction which changes depending on the
+    * orientation of the buffer:
+    */
+   if (ctx->Driver.FrontFace)
+      ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace);
+   else
+      ctx->NewState |= _NEW_POLYGON;
+
+   if (!colorRegion) {
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE);
+   }
+   else {
+      FALLBACK(intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE);
+   }
+
+   /***
+    *** Get depth buffer region and check if we need a software fallback.
+    *** Note that the depth buffer is usually a DEPTH_STENCIL buffer.
+    ***/
+   if (fb->_DepthBuffer && fb->_DepthBuffer->Wrapped) {
+      irbDepth = intel_renderbuffer(fb->_DepthBuffer->Wrapped);
+      if (irbDepth->region) {
+         FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_FALSE);
+         depthRegion = irbDepth->region;
+      }
+      else {
+         FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_TRUE);
+         depthRegion = NULL;
+      }
+   }
+   else {
+      /* not using depth buffer */
+      FALLBACK(intel, INTEL_FALLBACK_DEPTH_BUFFER, GL_FALSE);
+      depthRegion = NULL;
+   }
+
+   /***
+    *** Stencil buffer
+    *** This can only be hardware accelerated if we're using a
+    *** combined DEPTH_STENCIL buffer (for now anyway).
+    ***/
+   if (fb->_StencilBuffer && fb->_StencilBuffer->Wrapped) {
+      irbStencil = intel_renderbuffer(fb->_StencilBuffer->Wrapped);
+      if (irbStencil && irbStencil->region) {
+         ASSERT(irbStencil->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+         FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_FALSE);
+         /* need to re-compute stencil hw state */
+         ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
+         if (!depthRegion)
+            depthRegion = irbStencil->region;
+      }
+      else {
+         FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_TRUE);
+      }
+   }
+   else {
+      /* XXX FBO: instead of FALSE, pass ctx->Stencil.Enabled ??? */
+      FALLBACK(intel, INTEL_FALLBACK_STENCIL_BUFFER, GL_FALSE);
+      /* need to re-compute stencil hw state */
+      ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
+   }
+
+
+   /**
+    ** Release old regions, reference new regions
+    **/
+#if 0                           /* XXX FBO: this seems to be redundant with i915_state_draw_region() */
+   if (intel->draw_region != colorRegion) {
+      intel_region_release(intel, &intel->draw_region);
+      intel_region_reference(&intel->draw_region, colorRegion);
+   }
+   if (intel->intelScreen->depth_region != depthRegion) {
+      intel_region_release(intel, &intel->intelScreen->depth_region);
+      intel_region_reference(&intel->intelScreen->depth_region, depthRegion);
+   }
+#endif
+
+   intel->vtbl.set_draw_region(intel, colorRegion, depthRegion);
+
+   /* update viewport since it depends on window size */
+   ctx->Driver.Viewport(ctx, ctx->Viewport.X, ctx->Viewport.Y,
+                        ctx->Viewport.Width, ctx->Viewport.Height);
+
+   /* Update hardware scissor */
+   ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y,
+                       ctx->Scissor.Width, ctx->Scissor.Height);
+}
+
+
+static void
+intelDrawBuffer(GLcontext * ctx, GLenum mode)
+{
+   intel_draw_buffer(ctx, ctx->DrawBuffer);
+}
+
+
+static void
+intelReadBuffer(GLcontext * ctx, GLenum mode)
+{
+   if (ctx->ReadBuffer == ctx->DrawBuffer) {
+      /* This will update FBO completeness status.
+       * A framebuffer will be incomplete if the GL_READ_BUFFER setting
+       * refers to a missing renderbuffer.  Calling glReadBuffer can set
+       * that straight and can make the drawing buffer complete.
+       */
+      intel_draw_buffer(ctx, ctx->DrawBuffer);
+   }
+   /* Generally, functions which read pixels (glReadPixels, glCopyPixels, etc)
+    * reference ctx->ReadBuffer and do appropriate state checks.
+    */
+}
+
+
+void
+intelInitBufferFuncs(struct dd_function_table *functions)
+{
+   functions->Clear = intelClear;
+   functions->GetBufferSize = intelBufferSize;
+   functions->ResizeBuffers = _mesa_resize_framebuffer;
+   functions->DrawBuffer = intelDrawBuffer;
+   functions->ReadBuffer = intelReadBuffer;
+}

src/mesa/drivers/dri/i915/intel_buffers.h

@@ -0,0 +1,56 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_BUFFERS_H
+#define INTEL_BUFFERS_H
+
+
+struct intel_context;
+
+
+extern GLboolean
+intel_intersect_cliprects(drm_clip_rect_t * dest,
+                          const drm_clip_rect_t * a,
+                          const drm_clip_rect_t * b);
+
+extern struct intel_region *intel_readbuf_region(struct intel_context *intel);
+
+extern struct intel_region *intel_drawbuf_region(struct intel_context *intel);
+
+extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);
+
+extern void intelWindowMoved(struct intel_context *intel);
+
+extern void intel_draw_buffer(GLcontext * ctx, struct gl_framebuffer *fb);
+
+extern void intelInitBufferFuncs(struct dd_function_table *functions);
+
+extern void
+intelRotateWindow(struct intel_context *intel,
+                  __DRIdrawablePrivate * dPriv, GLuint srcBuf);
+
+#endif /* INTEL_BUFFERS_H */

src/mesa/drivers/dri/i915/intel_context.h

@@ -48,157 +48,178 @@
 #define DV_PF_565  (2<<8)
 #define DV_PF_8888 (3<<8)
 
-#define INTEL_CONTEXT(ctx)	((intelContextPtr)(ctx))
+struct intel_region;
+struct intel_context;
+struct _DriBufferObject;
 
-typedef struct intel_context intelContext;
-typedef struct intel_context *intelContextPtr;
-typedef struct intel_texture_object *intelTextureObjectPtr;
-
-typedef void (*intel_tri_func)(intelContextPtr, intelVertex *, intelVertex *,
-							  intelVertex *);
-typedef void (*intel_line_func)(intelContextPtr, intelVertex *, intelVertex *);
-typedef void (*intel_point_func)(intelContextPtr, intelVertex *);
+typedef void (*intel_tri_func) (struct intel_context *, intelVertex *,
+                                intelVertex *, intelVertex *);
+typedef void (*intel_line_func) (struct intel_context *, intelVertex *,
+                                 intelVertex *);
+typedef void (*intel_point_func) (struct intel_context *, intelVertex *);
 
 #define INTEL_FALLBACK_DRAW_BUFFER	 0x1
 #define INTEL_FALLBACK_READ_BUFFER	 0x2
-#define INTEL_FALLBACK_USER		 0x4
-#define INTEL_FALLBACK_NO_BATCHBUFFER	 0x8
-#define INTEL_FALLBACK_NO_TEXMEM	 0x10
+#define INTEL_FALLBACK_DEPTH_BUFFER      0x4
+#define INTEL_FALLBACK_STENCIL_BUFFER    0x8
+#define INTEL_FALLBACK_USER		 0x10
 #define INTEL_FALLBACK_RENDERMODE	 0x20
 
-extern void intelFallback( intelContextPtr intel, GLuint bit, GLboolean mode );
+extern void intelFallback(struct intel_context *intel, GLuint bit,
+                          GLboolean mode);
 #define FALLBACK( intel, bit, mode ) intelFallback( intel, bit, mode )
 
 
-#define INTEL_TEX_MAXLEVELS 10
-
+#define INTEL_WRITE_PART  0x1
+#define INTEL_WRITE_FULL  0x2
+#define INTEL_READ        0x4
 
 struct intel_texture_object
 {
-   driTextureObject    base;	/* the parent class */
+   struct gl_texture_object base;       /* The "parent" object */
 
-   GLuint texelBytes;
-   GLuint age;
-   GLuint Pitch;
-   GLuint Height;
-   GLuint TextureOffset;
-   GLubyte *BufAddr;   
+   /* The mipmap tree must include at least these levels once
+    * validated:
+    */
+   GLuint firstLevel;
+   GLuint lastLevel;
 
-   GLuint min_level;
-   GLuint max_level;
-   GLuint depth_pitch;
+   /* Offset for firstLevel image:
+    */
+   GLuint textureOffset;
 
-   struct {
-      const struct gl_texture_image *image;
-      GLuint offset;       /* into BufAddr */
-      GLuint height;
-      GLuint internalFormat;
-   } image[6][INTEL_TEX_MAXLEVELS];
-
-   GLuint dirty;
-   GLuint firstLevel,lastLevel;
+   /* On validation any active images held in main memory or in other
+    * regions will be copied to this region and the old storage freed.
+    */
+   struct intel_mipmap_tree *mt;
 };
 
 
+
+struct intel_texture_image
+{
+   struct gl_texture_image base;
+
+   /* These aren't stored in gl_texture_image 
+    */
+   GLuint level;
+   GLuint face;
+
+   /* If intelImage->mt != NULL, image data is stored here.
+    * Else if intelImage->base.Data != NULL, image is stored there.
+    * Else there is no image data.
+    */
+   struct intel_mipmap_tree *mt;
+};
+
+
+#define INTEL_MAX_FIXUP 64
+
 struct intel_context
 {
-   GLcontext ctx;		/* the parent class */
+   GLcontext ctx;               /* the parent class */
 
-   struct {
-      void (*destroy)( intelContextPtr intel ); 
-      void (*emit_state)( intelContextPtr intel );
-      void (*emit_invarient_state)( intelContextPtr intel );
-      void (*lost_hardware)( intelContextPtr intel );
-      void (*update_texture_state)( intelContextPtr intel );
+   struct
+   {
+      void (*destroy) (struct intel_context * intel);
+      void (*emit_state) (struct intel_context * intel);
+      void (*lost_hardware) (struct intel_context * intel);
+      void (*update_texture_state) (struct intel_context * intel);
 
-      void (*render_start)( intelContextPtr intel );
-      void (*set_color_region)( intelContextPtr intel, const intelRegion *reg );
-      void (*set_z_region)( intelContextPtr intel, const intelRegion *reg );
-      void (*update_color_z_regions)(intelContextPtr intel,
-                                     const intelRegion *colorRegion,
-                                     const intelRegion *depthRegion);
-      void (*emit_flush)( intelContextPtr intel );
-      void (*reduced_primitive_state)( intelContextPtr intel, GLenum rprim );
+      void (*render_start) (struct intel_context * intel);
+      void (*set_draw_region) (struct intel_context * intel,
+                               struct intel_region * draw_region,
+                               struct intel_region * depth_region);
 
-      GLboolean (*check_vertex_size)( intelContextPtr intel, GLuint expected );
+        GLuint(*flush_cmd) (void);
 
-      void (*clear_with_tris)( intelContextPtr intel, GLbitfield mask,
-			       GLboolean all, 
-			       GLint cx, GLint cy, GLint cw, GLint ch);
+      void (*reduced_primitive_state) (struct intel_context * intel,
+                                       GLenum rprim);
 
-      void (*rotate_window)( intelContextPtr intel,
-                             __DRIdrawablePrivate *dPriv, GLuint srcBuf);
+        GLboolean(*check_vertex_size) (struct intel_context * intel,
+                                       GLuint expected);
 
-      intelTextureObjectPtr (*alloc_tex_obj)( struct gl_texture_object *tObj );
+
+      /* Metaops: 
+       */
+      void (*install_meta_state) (struct intel_context * intel);
+      void (*leave_meta_state) (struct intel_context * intel);
+
+      void (*meta_draw_region) (struct intel_context * intel,
+                                struct intel_region * draw_region,
+                                struct intel_region * depth_region);
+
+      void (*meta_color_mask) (struct intel_context * intel, GLboolean);
+
+      void (*meta_stencil_replace) (struct intel_context * intel,
+                                    GLuint mask, GLuint clear);
+
+      void (*meta_depth_replace) (struct intel_context * intel);
+
+      void (*meta_texture_blend_replace) (struct intel_context * intel);
+
+      void (*meta_no_stencil_write) (struct intel_context * intel);
+      void (*meta_no_depth_write) (struct intel_context * intel);
+      void (*meta_no_texture) (struct intel_context * intel);
+
+      void (*meta_import_pixel_state) (struct intel_context * intel);
+
+        GLboolean(*meta_tex_rect_source) (struct intel_context * intel,
+                                          struct _DriBufferObject * buffer,
+                                          GLuint offset,
+                                          GLuint pitch,
+                                          GLuint height,
+                                          GLenum format, GLenum type);
+      void (*rotate_window) (struct intel_context * intel,
+                             __DRIdrawablePrivate * dPriv, GLuint srcBuf);
+
+      void (*assert_not_dirty) (struct intel_context *intel);
 
    } vtbl;
 
-   GLint refcount;   
+   GLint refcount;
    GLuint Fallback;
    GLuint NewGLState;
-   
-   struct {
-      GLuint start_offset;
-      GLint size;
-      GLint space;
-      GLubyte *ptr;
-      GLuint counter;
-      GLuint last_emit_state;
-      GLboolean contains_geometry;
-      const char *func;
-      GLuint last_swap;
-   } batch;
-      
-   struct {
-      void *ptr;
-      GLint size;
-      GLuint offset;
-      GLuint active_buf;
-      GLuint irq_emitted;
-   } alloc;
 
-   struct {
+   struct _DriFenceObject *last_swap_fence;
+   struct _DriFenceObject *first_swap_fence;
+
+   struct intel_batchbuffer *batch;
+
+   struct
+   {
+      GLuint id;
       GLuint primitive;
-      GLubyte *start_ptr;      
-      void (*flush)( GLcontext * );
+      GLubyte *start_ptr;
+      void (*flush) (struct intel_context *);
    } prim;
 
    GLboolean locked;
+   char *prevLockFile;
+   int prevLockLine;
 
-   GLubyte clear_red;
-   GLubyte clear_green;
-   GLubyte clear_blue;
-   GLubyte clear_alpha;
-   GLuint ClearColor;
-   GLuint ClearDepth;
+   GLuint ClearColor565;
+   GLuint ClearColor8888;
 
+   /* Offsets of fields within the current vertex:
+    */
    GLuint coloroffset;
    GLuint specoffset;
-
-   /* Support for duplicating XYZW as WPOS parameter (crutch for I915).
-    */
    GLuint wpos_offset;
    GLuint wpos_size;
 
    struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
    GLuint vertex_attr_count;
 
-   GLfloat depth_scale;
-   GLfloat polygon_offset_scale; /* dependent on depth_scale, bpp */
-   GLuint depth_clear_mask;
-   GLuint stencil_clear_mask;
+   GLfloat polygon_offset_scale;        /* dependent on depth_scale, bpp */
 
-   GLboolean hw_stencil;
    GLboolean hw_stipple;
-   
-   /* Texture object bookkeeping
-    */
-   GLuint                nr_heaps;
-   driTexHeap          * texture_heaps[1];
-   driTextureObject      swapped;
-   GLuint                lastStamp;
+   GLboolean strict_conformance;
+
+   /* AGP memory buffer manager:
+    */
+   struct bufmgr *bm;
 
-   struct intel_texture_object *CurrentTexObj[MAX_TEXTURE_UNITS];
 
    /* State for intelvb.c and inteltris.c.
     */
@@ -207,8 +228,15 @@ struct intel_context
    GLenum render_primitive;
    GLenum reduced_primitive;
    GLuint vertex_size;
-   unsigned char *verts;	   /* points to tnl->clipspace.vertex_buf */
+   GLubyte *verts;              /* points to tnl->clipspace.vertex_buf */
 
+#if 0
+   struct intel_region *front_region;   /* XXX FBO: obsolete */
+   struct intel_region *rotated_region; /* XXX FBO: obsolete */
+   struct intel_region *back_region;    /* XXX FBO: obsolete */
+   struct intel_region *draw_region;    /* XXX FBO: rename to color_region */
+   struct intel_region *depth_region;   /**< currently bound depth/Z region */
+#endif
 
    /* Fallback rasterization functions 
     */
@@ -216,17 +244,13 @@ struct intel_context
    intel_line_func draw_line;
    intel_tri_func draw_tri;
 
-   /* Drawing buffer state
+   /* These refer to the current drawing buffer:
     */
-   intelRegion *drawRegion;  /* current drawing buffer */
-   intelRegion *readRegion;  /* current reading buffer */
-
-   int drawX;			/* origin of drawable in draw buffer */
-   int drawY;
-   GLuint numClipRects;		/* cliprects for that buffer */
+   int drawX, drawY;            /**< origin of drawing area within region */
+   GLuint numClipRects;         /**< cliprects for drawing */
    drm_clip_rect_t *pClipRects;
+   drm_clip_rect_t fboRect;     /**< cliprect for FBO rendering */
 
-   int dirtyAge;
    int perf_boxes;
 
    GLuint do_usleeps;
@@ -234,18 +258,16 @@ struct intel_context
    GLuint irqsEmitted;
    drm_i915_irq_wait_t iw;
 
-   GLboolean scissor;
-   drm_clip_rect_t draw_rect;
-   drm_clip_rect_t scissor_rect;
-
    drm_context_t hHWContext;
    drmLock *driHwLock;
    int driFd;
 
    __DRIdrawablePrivate *driDrawable;
    __DRIscreenPrivate *driScreen;
-   intelScreenPrivate *intelScreen; 
-   drmI830Sarea *sarea; 
+   intelScreenPrivate *intelScreen;
+   drmI830Sarea *sarea;
+
+   GLuint lastStamp;
 
    /**
     * Configuration cache
@@ -262,87 +284,15 @@ struct intel_context
 
    GLuint swap_count;
    GLuint swap_missed_count;
+
 };
 
-
-#define DEBUG_LOCKING	1
-
-#if DEBUG_LOCKING
-extern char *prevLockFile;
-extern int prevLockLine;
-
-#define DEBUG_LOCK()							\
-   do {									\
-      prevLockFile = (__FILE__);					\
-      prevLockLine = (__LINE__);					\
-   } while (0)
-
-#define DEBUG_RESET()							\
-   do {									\
-      prevLockFile = 0;							\
-      prevLockLine = 0;							\
-   } while (0)
-
-/* Slightly less broken way of detecting recursive locking in a
- * threaded environment.  The right way to do this would be to make
- * prevLockFile, prevLockLine thread-local.
- *
- * This technique instead checks to see if the same context is
- * requesting the lock twice -- this will not catch application
- * breakages where the same context is active in two different threads
- * at once, but it will catch driver breakages (recursive locking) in
- * threaded apps.
+/* These are functions now:
  */
-#define DEBUG_CHECK_LOCK()						\
-   do {									\
-      if ( *((volatile int *)intel->driHwLock) == 			\
-	   (DRM_LOCK_HELD | intel->hHWContext) ) {			\
-	 fprintf( stderr,						\
-		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
-		  prevLockFile, prevLockLine, __FILE__, __LINE__ );	\
-	 abort();							\
-      }									\
-   } while (0)
+void LOCK_HARDWARE( struct intel_context *intel );
+void UNLOCK_HARDWARE( struct intel_context *intel );
 
-#else
-
-#define DEBUG_LOCK()
-#define DEBUG_RESET()
-#define DEBUG_CHECK_LOCK()
-
-#endif
-
-
-
-
-/* Lock the hardware and validate our state.  
- */
-#define LOCK_HARDWARE( intel )				\
-do {							\
-    char __ret=0;					\
-    DEBUG_CHECK_LOCK();					\
-    assert(!(intel)->locked);				\
-    DRM_CAS((intel)->driHwLock, (intel)->hHWContext,	\
-        (DRM_LOCK_HELD|(intel)->hHWContext), __ret);	\
-    if (__ret)						\
-        intelGetLock( (intel), 0 );			\
-      DEBUG_LOCK();					\
-    (intel)->locked = 1;				\
-}while (0)
- 
-  
-  /* Unlock the hardware using the global current context 
-   */
-#define UNLOCK_HARDWARE(intel)						\
-do {									\
-   intel->locked = 0;							\
-   if (0) { 								\
-      intel->perf_boxes |= intel->sarea->perf_boxes;  			\
-      intel->sarea->perf_boxes = 0;					\
-   }									\
-   DRM_UNLOCK((intel)->driFd, (intel)->driHwLock, (intel)->hHWContext);	\
-   DEBUG_RESET();							\
-} while (0)
+extern char *__progname;
 
 
 #define SUBPIXEL_X 0.125
@@ -351,7 +301,7 @@ do {									\
 #define INTEL_FIREVERTICES(intel)		\
 do {						\
    if ((intel)->prim.flush)			\
-      (intel)->prim.flush(&(intel)->ctx);		\
+      (intel)->prim.flush(intel);		\
 } while (0)
 
 /* ================================================================
@@ -372,34 +322,26 @@ do {						\
   ((a<<24) | (r<<16) | (g<<8) | b)
 
 
-#define INTEL_PACKCOLOR(format, r,  g,  b, a)		\
-(format == DV_PF_555 ? INTEL_PACKCOLOR1555(r,g,b,a) :	\
- (format == DV_PF_565 ? INTEL_PACKCOLOR565(r,g,b) :	\
-  (format == DV_PF_8888 ? INTEL_PACKCOLOR8888(r,g,b,a) :	\
-   0)))
-
-
 
 /* ================================================================
  * From linux kernel i386 header files, copes with odd sizes better
  * than COPY_DWORDS would:
+ * XXX Put this in src/mesa/main/imports.h ???
  */
 #if defined(i386) || defined(__i386__)
-static __inline__ void * __memcpy(void * to, const void * from, size_t n)
+static INLINE void *
+__memcpy(void *to, const void *from, size_t n)
 {
    int d0, d1, d2;
-   __asm__ __volatile__(
-      "rep ; movsl\n\t"
-      "testb $2,%b4\n\t"
-      "je 1f\n\t"
-      "movsw\n"
-      "1:\ttestb $1,%b4\n\t"
-      "je 2f\n\t"
-      "movsb\n"
-      "2:"
-      : "=&c" (d0), "=&D" (d1), "=&S" (d2)
-      :"0" (n/4), "q" (n),"1" ((long) to),"2" ((long) from)
-      : "memory");
+   __asm__ __volatile__("rep ; movsl\n\t"
+                        "testb $2,%b4\n\t"
+                        "je 1f\n\t"
+                        "movsw\n"
+                        "1:\ttestb $1,%b4\n\t"
+                        "je 2f\n\t"
+                        "movsb\n" "2:":"=&c"(d0), "=&D"(d1), "=&S"(d2)
+                        :"0"(n / 4), "q"(n), "1"((long) to), "2"((long) from)
+                        :"memory");
    return (to);
 }
 #else
@@ -421,16 +363,18 @@ extern int INTEL_DEBUG;
 #define DEBUG_TEXTURE	0x1
 #define DEBUG_STATE	0x2
 #define DEBUG_IOCTL	0x4
-#define DEBUG_PRIMS	0x8
-#define DEBUG_VERTS	0x10
+#define DEBUG_BLIT	0x8
+#define DEBUG_MIPTREE   0x10
 #define DEBUG_FALLBACKS	0x20
 #define DEBUG_VERBOSE	0x40
-#define DEBUG_DRI       0x80
-#define DEBUG_DMA       0x100
-#define DEBUG_SANITY    0x200
-#define DEBUG_SYNC      0x400
-#define DEBUG_SLEEP     0x800
-#define DEBUG_PIXEL     0x1000
+#define DEBUG_BATCH     0x80
+#define DEBUG_PIXEL     0x100
+#define DEBUG_BUFMGR    0x200
+#define DEBUG_REGION    0x400
+#define DEBUG_FBO       0x800
+#define DEBUG_LOCK      0x1000
+
+#define DBG(...)  do { if (INTEL_DEBUG & FILE_DEBUG_FLAG) _mesa_printf(__VA_ARGS__); } while(0)
 
 
 #define PCI_CHIP_845_G			0x2562
@@ -447,27 +391,25 @@ extern int INTEL_DEBUG;
  * intel_context.c:
  */
 
-extern void intelInitDriverFunctions( struct dd_function_table *functions );
+extern GLboolean intelInitContext(struct intel_context *intel,
+                                  const __GLcontextModes * mesaVis,
+                                  __DRIcontextPrivate * driContextPriv,
+                                  void *sharedContextPrivate,
+                                  struct dd_function_table *functions);
 
-extern GLboolean intelInitContext( intelContextPtr intel, 
-				   const __GLcontextModes *mesaVis,
-				   __DRIcontextPrivate *driContextPriv,
-				   void *sharedContextPrivate,
-				   struct dd_function_table *functions );
+extern void intelGetLock(struct intel_context *intel, GLuint flags);
 
-extern void intelGetLock(intelContextPtr intel, GLuint flags);
-extern void intelSetBackClipRects(intelContextPtr intel);
-extern void intelSetFrontClipRects(intelContextPtr intel);
-extern void intelWindowMoved( intelContextPtr intel );
+extern void intelInitState(GLcontext * ctx);
+extern void intelFinish(GLcontext * ctx);
+extern void intelFlush(GLcontext * ctx);
 
-extern void intelInitState( GLcontext *ctx );
-extern const GLubyte *intelGetString( GLcontext *ctx, GLenum name );
+extern void intelInitDriverFunctions(struct dd_function_table *functions);
 
 
 /* ================================================================
  * intel_state.c:
  */
-extern void intelInitStateFuncs( struct dd_function_table *functions );
+extern void intelInitStateFuncs(struct dd_function_table *functions);
 
 #define COMPAREFUNC_ALWAYS		0
 #define COMPAREFUNC_NEVER		0x1
@@ -521,27 +463,39 @@ extern void intelInitStateFuncs( struct dd_function_table *functions );
 #define BLENDFACT_INV_CONST_ALPHA	0x0f
 #define BLENDFACT_MASK          	0x0f
 
-
-extern int intel_translate_compare_func( GLenum func );
-extern int intel_translate_stencil_op( GLenum op );
-extern int intel_translate_blend_factor( GLenum factor );
-extern int intel_translate_logic_op( GLenum opcode );
+#define MI_BATCH_BUFFER_END 	(0xA<<23)
 
 
-/* ================================================================
- * intel_ioctl.c:
+extern int intel_translate_compare_func(GLenum func);
+extern int intel_translate_stencil_op(GLenum op);
+extern int intel_translate_blend_factor(GLenum factor);
+extern int intel_translate_logic_op(GLenum opcode);
+
+
+/*======================================================================
+ * Inline conversion functions.  
+ * These are better-typed than the macros used previously:
  */
-extern void intel_dump_batchbuffer( long offset,
-				    int *ptr,
-				    int count );
+static INLINE struct intel_context *
+intel_context(GLcontext * ctx)
+{
+   return (struct intel_context *) ctx;
+}
 
+static INLINE struct intel_texture_object *
+intel_texture_object(struct gl_texture_object *obj)
+{
+   return (struct intel_texture_object *) obj;
+}
 
-/* ================================================================
- * intel_pixel.c:
- */	
-extern void intelInitPixelFuncs( struct dd_function_table *functions );
+static INLINE struct intel_texture_image *
+intel_texture_image(struct gl_texture_image *img)
+{
+   return (struct intel_texture_image *) img;
+}
 
+extern struct intel_renderbuffer *intel_renderbuffer(struct gl_renderbuffer
+                                                     *rb);
 
 
 #endif
-

src/mesa/drivers/dri/i915/intel_depthstencil.c

@@ -0,0 +1,282 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "glheader.h"
+#include "imports.h"
+#include "context.h"
+#include "depthstencil.h"
+#include "fbobject.h"
+#include "framebuffer.h"
+#include "hash.h"
+#include "mtypes.h"
+#include "renderbuffer.h"
+
+#include "intel_context.h"
+#include "intel_fbo.h"
+#include "intel_depthstencil.h"
+#include "intel_regions.h"
+
+
+/**
+ * The GL_EXT_framebuffer_object allows the user to create their own
+ * framebuffer objects consisting of color renderbuffers (0 or more),
+ * depth renderbuffers (0 or 1) and stencil renderbuffers (0 or 1).
+ *
+ * The spec considers depth and stencil renderbuffers to be totally independent
+ * buffers.  In reality, most graphics hardware today uses a combined
+ * depth+stencil buffer (one 32-bit pixel = 24 bits of Z + 8 bits of stencil).
+ *
+ * This causes difficulty because the user may create some number of depth
+ * renderbuffers and some number of stencil renderbuffers and bind them
+ * together in framebuffers in any combination.
+ *
+ * This code manages all that.
+ *
+ * 1. Depth renderbuffers are always allocated in hardware as 32bpp
+ *    GL_DEPTH24_STENCIL8 buffers.
+ *
+ * 2. Stencil renderbuffers are initially allocated in software as 8bpp
+ *    GL_STENCIL_INDEX8 buffers.
+ *
+ * 3. Depth and Stencil renderbuffers use the PairedStencil and PairedDepth
+ *    fields (respectively) to indicate if the buffer's currently paired
+ *    with another stencil or depth buffer (respectively).
+ *
+ * 4. When a depth and stencil buffer are initially both attached to the
+ *    current framebuffer, we merge the stencil buffer values into the
+ *    depth buffer (really a depth+stencil buffer).  The then hardware uses
+ *    the combined buffer.
+ *
+ * 5. Whenever a depth or stencil buffer is reallocated (with
+ *    glRenderbufferStorage) we undo the pairing and copy the stencil values
+ *    from the combined depth/stencil buffer back to the stencil-only buffer.
+ *
+ * 6. We also undo the pairing when we find a change in buffer bindings.
+ *
+ * 7. If a framebuffer is only using a depth renderbuffer (no stencil), we
+ *    just use the combined depth/stencil buffer and ignore the stencil values.
+ *
+ * 8. If a framebuffer is only using a stencil renderbuffer (no depth) we have
+ *    to promote the 8bpp software stencil buffer to a 32bpp hardware
+ *    depth+stencil buffer.
+ *
+ */
+
+
+
+static void
+map_regions(GLcontext * ctx,
+            struct intel_renderbuffer *depthRb,
+            struct intel_renderbuffer *stencilRb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   if (depthRb && depthRb->region) {
+      intel_region_map(intel->intelScreen, depthRb->region);
+      depthRb->pfMap = depthRb->region->map;
+      depthRb->pfPitch = depthRb->region->pitch;
+   }
+   if (stencilRb && stencilRb->region) {
+      intel_region_map(intel->intelScreen, stencilRb->region);
+      stencilRb->pfMap = stencilRb->region->map;
+      stencilRb->pfPitch = stencilRb->region->pitch;
+   }
+}
+
+static void
+unmap_regions(GLcontext * ctx,
+              struct intel_renderbuffer *depthRb,
+              struct intel_renderbuffer *stencilRb)
+{
+   struct intel_context *intel = intel_context(ctx);
+   if (depthRb && depthRb->region) {
+      intel_region_unmap(intel->intelScreen, depthRb->region);
+      depthRb->pfMap = NULL;
+      depthRb->pfPitch = 0;
+   }
+   if (stencilRb && stencilRb->region) {
+      intel_region_unmap(intel->intelScreen, stencilRb->region);
+      stencilRb->pfMap = NULL;
+      stencilRb->pfPitch = 0;
+   }
+}
+
+
+
+/**
+ * Undo the pairing/interleaving between depth and stencil buffers.
+ * irb should be a depth/stencil or stencil renderbuffer.
+ */
+void
+intel_unpair_depth_stencil(GLcontext * ctx, struct intel_renderbuffer *irb)
+{
+   if (irb->PairedStencil) {
+      /* irb is a depth/stencil buffer */
+      struct gl_renderbuffer *stencilRb;
+      struct intel_renderbuffer *stencilIrb;
+
+      ASSERT(irb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+
+      stencilRb = _mesa_lookup_renderbuffer(ctx, irb->PairedStencil);
+      stencilIrb = intel_renderbuffer(stencilRb);
+      if (stencilIrb) {
+         /* need to extract stencil values from the depth buffer */
+         ASSERT(stencilIrb->PairedDepth == irb->Base.Name);
+         map_regions(ctx, irb, stencilIrb);
+         _mesa_extract_stencil(ctx, &irb->Base, &stencilIrb->Base);
+         unmap_regions(ctx, irb, stencilIrb);
+         stencilIrb->PairedDepth = 0;
+      }
+      irb->PairedStencil = 0;
+   }
+   else if (irb->PairedDepth) {
+      /* irb is a stencil buffer */
+      struct gl_renderbuffer *depthRb;
+      struct intel_renderbuffer *depthIrb;
+
+      ASSERT(irb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT ||
+             irb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+
+      depthRb = _mesa_lookup_renderbuffer(ctx, irb->PairedDepth);
+      depthIrb = intel_renderbuffer(depthRb);
+      if (depthIrb) {
+         /* need to extract stencil values from the depth buffer */
+         ASSERT(depthIrb->PairedStencil == irb->Base.Name);
+         map_regions(ctx, depthIrb, irb);
+         _mesa_extract_stencil(ctx, &depthIrb->Base, &irb->Base);
+         unmap_regions(ctx, depthIrb, irb);
+         depthIrb->PairedStencil = 0;
+      }
+      irb->PairedDepth = 0;
+   }
+   else {
+      _mesa_problem(ctx, "Problem in undo_depth_stencil_pairing");
+   }
+
+   ASSERT(irb->PairedStencil == 0);
+   ASSERT(irb->PairedDepth == 0);
+}
+
+
+/**
+ * Examine the depth and stencil renderbuffers which are attached to the
+ * framebuffer.  If both depth and stencil are attached, make sure that the
+ * renderbuffers are 'paired' (combined).  If only depth or only stencil is
+ * attached, undo any previous pairing.
+ *
+ * Must be called if NewState & _NEW_BUFFER (when renderbuffer attachments
+ * change, for example).
+ */
+void
+intel_validate_paired_depth_stencil(GLcontext * ctx,
+                                    struct gl_framebuffer *fb)
+{
+   struct intel_renderbuffer *depthRb, *stencilRb;
+
+   depthRb = intel_get_renderbuffer(fb, BUFFER_DEPTH);
+   stencilRb = intel_get_renderbuffer(fb, BUFFER_STENCIL);
+
+   if (depthRb && stencilRb) {
+      if (depthRb == stencilRb) {
+         /* Using a user-created combined depth/stencil buffer.
+          * Nothing to do.
+          */
+         ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_STENCIL_EXT);
+         ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+      }
+      else {
+         /* Separate depth/stencil buffers, need to interleave now */
+         ASSERT(depthRb->Base._BaseFormat == GL_DEPTH_COMPONENT);
+         ASSERT(stencilRb->Base._BaseFormat == GL_STENCIL_INDEX);
+         /* may need to interleave depth/stencil now */
+         if (depthRb->PairedStencil == stencilRb->Base.Name) {
+            /* OK, the depth and stencil buffers are already interleaved */
+            ASSERT(stencilRb->PairedDepth == depthRb->Base.Name);
+         }
+         else {
+            /* need to setup new pairing/interleaving */
+            if (depthRb->PairedStencil) {
+               intel_unpair_depth_stencil(ctx, depthRb);
+            }
+            if (stencilRb->PairedDepth) {
+               intel_unpair_depth_stencil(ctx, stencilRb);
+            }
+
+            ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+            ASSERT(stencilRb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT ||
+                   stencilRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+
+            /* establish new pairing: interleave stencil into depth buffer */
+            map_regions(ctx, depthRb, stencilRb);
+            _mesa_insert_stencil(ctx, &depthRb->Base, &stencilRb->Base);
+            unmap_regions(ctx, depthRb, stencilRb);
+            depthRb->PairedStencil = stencilRb->Base.Name;
+            stencilRb->PairedDepth = depthRb->Base.Name;
+         }
+
+      }
+   }
+   else if (depthRb) {
+      /* Depth buffer but no stencil buffer.
+       * We'll use a GL_DEPTH24_STENCIL8 buffer and ignore the stencil bits.
+       */
+      /* can't assert this until storage is allocated:
+         ASSERT(depthRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+       */
+      /* intel_undo any previous pairing */
+      if (depthRb->PairedStencil) {
+         intel_unpair_depth_stencil(ctx, depthRb);
+      }
+   }
+   else if (stencilRb) {
+      /* Stencil buffer but no depth buffer.
+       * Since h/w doesn't typically support just 8bpp stencil w/out Z,
+       * we'll use a GL_DEPTH24_STENCIL8 buffer and ignore the depth bits.
+       */
+      /* undo any previous pairing */
+      if (stencilRb->PairedDepth) {
+         intel_unpair_depth_stencil(ctx, stencilRb);
+      }
+      if (stencilRb->Base._ActualFormat == GL_STENCIL_INDEX8_EXT) {
+         /* promote buffer to GL_DEPTH24_STENCIL8 for hw rendering */
+         _mesa_promote_stencil(ctx, &stencilRb->Base);
+         ASSERT(stencilRb->Base._ActualFormat == GL_DEPTH24_STENCIL8_EXT);
+      }
+   }
+
+   /* Finally, update the fb->_DepthBuffer and fb->_StencilBuffer fields */
+   _mesa_update_depth_buffer(ctx, fb, BUFFER_DEPTH);
+   if (depthRb && depthRb->PairedStencil)
+      _mesa_update_stencil_buffer(ctx, fb, BUFFER_DEPTH);
+   else
+      _mesa_update_stencil_buffer(ctx, fb, BUFFER_STENCIL);
+
+
+   /* The hardware should use fb->Attachment[BUFFER_DEPTH].Renderbuffer
+    * first, if present, then fb->Attachment[BUFFER_STENCIL].Renderbuffer
+    * if present.
+    */
+}

src/mesa/drivers/dri/i915/intel_depthstencil.h

@@ -0,0 +1,14 @@
+
+#ifndef INTEL_DEPTH_STENCIL_H
+#define INTEL_DEPTH_STENCIL_H
+
+
+extern void
+intel_unpair_depth_stencil(GLcontext * ctx, struct intel_renderbuffer *irb);
+
+extern void
+intel_validate_paired_depth_stencil(GLcontext * ctx,
+                                    struct gl_framebuffer *fb);
+
+
+#endif /* INTEL_DEPTH_STENCIL_H */

src/mesa/drivers/dri/i915/intel_fbo.c

@@ -0,0 +1,621 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "imports.h"
+#include "mtypes.h"
+#include "fbobject.h"
+#include "framebuffer.h"
+#include "renderbuffer.h"
+#include "context.h"
+#include "texformat.h"
+#include "texrender.h"
+
+#include "intel_context.h"
+#include "intel_buffers.h"
+#include "intel_depthstencil.h"
+#include "intel_fbo.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
+#include "intel_span.h"
+
+
+#define FILE_DEBUG_FLAG DEBUG_FBO
+
+#define INTEL_RB_CLASS 0x12345678
+
+
+/* XXX FBO: move this to intel_context.h (inlined) */
+/**
+ * Return a gl_renderbuffer ptr casted to intel_renderbuffer.
+ * NULL will be returned if the rb isn't really an intel_renderbuffer.
+ * This is determiend by checking the ClassID.
+ */
+struct intel_renderbuffer *
+intel_renderbuffer(struct gl_renderbuffer *rb)
+{
+   struct intel_renderbuffer *irb = (struct intel_renderbuffer *) rb;
+   if (irb && irb->Base.ClassID == INTEL_RB_CLASS) {
+      /*_mesa_warning(NULL, "Returning non-intel Rb\n");*/
+      return irb;
+   }
+   else
+      return NULL;
+}
+
+
+struct intel_renderbuffer *
+intel_get_renderbuffer(struct gl_framebuffer *fb, GLuint attIndex)
+{
+   return intel_renderbuffer(fb->Attachment[attIndex].Renderbuffer);
+}
+
+
+struct intel_region *
+intel_get_rb_region(struct gl_framebuffer *fb, GLuint attIndex)
+{
+   struct intel_renderbuffer *irb
+      = intel_renderbuffer(fb->Attachment[attIndex].Renderbuffer);
+   if (irb)
+      return irb->region;
+   else
+      return NULL;
+}
+
+
+
+/**
+ * Create a new framebuffer object.
+ */
+static struct gl_framebuffer *
+intel_new_framebuffer(GLcontext * ctx, GLuint name)
+{
+   /* there's no intel_framebuffer at this time, just use Mesa's class */
+   return _mesa_new_framebuffer(ctx, name);
+}
+
+
+static void
+intel_delete_renderbuffer(struct gl_renderbuffer *rb)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);
+
+   ASSERT(irb);
+
+   if (irb->PairedStencil || irb->PairedDepth) {
+      intel_unpair_depth_stencil(ctx, irb);
+   }
+
+   if (intel && irb->region) {
+      intel_region_release(intel->intelScreen, &irb->region);
+   }
+
+   _mesa_free(irb);
+}
+
+
+
+/**
+ * Return a pointer to a specific pixel in a renderbuffer.
+ */
+static void *
+intel_get_pointer(GLcontext * ctx, struct gl_renderbuffer *rb,
+                  GLint x, GLint y)
+{
+   /* By returning NULL we force all software rendering to go through
+    * the span routines.
+    */
+   return NULL;
+}
+
+
+
+/**
+ * Called via glRenderbufferStorageEXT() to set the format and allocate
+ * storage for a user-created renderbuffer.
+ */
+static GLboolean
+intel_alloc_renderbuffer_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                                 GLenum internalFormat,
+                                 GLuint width, GLuint height)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);
+   GLboolean softwareBuffer = GL_FALSE;
+   int cpp;
+
+   ASSERT(rb->Name != 0);
+
+   switch (internalFormat) {
+   case GL_R3_G3_B2:
+   case GL_RGB4:
+   case GL_RGB5:
+      rb->_ActualFormat = GL_RGB5;
+      rb->DataType = GL_UNSIGNED_BYTE;
+      rb->RedBits = 5;
+      rb->GreenBits = 6;
+      rb->BlueBits = 5;
+      cpp = 2;
+      break;
+   case GL_RGB:
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+   case GL_RGBA:
+   case GL_RGBA2:
+   case GL_RGBA4:
+   case GL_RGB5_A1:
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      rb->_ActualFormat = GL_RGBA8;
+      rb->DataType = GL_UNSIGNED_BYTE;
+      rb->RedBits = 8;
+      rb->GreenBits = 8;
+      rb->BlueBits = 8;
+      rb->AlphaBits = 8;
+      cpp = 4;
+      break;
+   case GL_STENCIL_INDEX:
+   case GL_STENCIL_INDEX1_EXT:
+   case GL_STENCIL_INDEX4_EXT:
+   case GL_STENCIL_INDEX8_EXT:
+   case GL_STENCIL_INDEX16_EXT:
+      /* alloc a depth+stencil buffer */
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->StencilBits = 8;
+      cpp = 4;
+      break;
+   case GL_DEPTH_COMPONENT16:
+      rb->_ActualFormat = GL_DEPTH_COMPONENT16;
+      rb->DataType = GL_UNSIGNED_SHORT;
+      rb->DepthBits = 16;
+      cpp = 2;
+      break;
+   case GL_DEPTH_COMPONENT:
+   case GL_DEPTH_COMPONENT24:
+   case GL_DEPTH_COMPONENT32:
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->DepthBits = 24;
+      cpp = 4;
+      break;
+   case GL_DEPTH_STENCIL_EXT:
+   case GL_DEPTH24_STENCIL8_EXT:
+      rb->_ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      rb->DataType = GL_UNSIGNED_INT_24_8_EXT;
+      rb->DepthBits = 24;
+      rb->StencilBits = 8;
+      cpp = 4;
+      break;
+   default:
+      _mesa_problem(ctx,
+                    "Unexpected format in intel_alloc_renderbuffer_storage");
+      return GL_FALSE;
+   }
+
+   intelFlush(ctx);
+
+   /* free old region */
+   if (irb->region) {
+      /*LOCK_HARDWARE(intel); */
+      intel_region_release(intel->intelScreen, &irb->region);
+      /*UNLOCK_HARDWARE(intel); */
+   }
+
+   /* allocate new memory region/renderbuffer */
+   if (softwareBuffer) {
+      return _mesa_soft_renderbuffer_storage(ctx, rb, internalFormat,
+                                             width, height);
+   }
+   else {
+      /* Choose a pitch to match hardware requirements:
+       */
+      GLuint pitch = ((cpp * width + 63) & ~63) / cpp;
+
+      /* alloc hardware renderbuffer */
+      DBG("Allocating %d x %d Intel RBO (pitch %d)\n", width,
+	  height, pitch);
+
+      irb->region = intel_region_alloc(intel->intelScreen, cpp, pitch, height);
+      if (!irb->region)
+         return GL_FALSE;       /* out of memory? */
+
+      ASSERT(irb->region->buffer);
+
+      rb->Width = width;
+      rb->Height = height;
+
+      /* This sets the Get/PutRow/Value functions */
+      intel_set_span_functions(&irb->Base);
+
+      return GL_TRUE;
+   }
+}
+
+
+
+/**
+ * Called for each hardware renderbuffer when a _window_ is resized.
+ * Just update fields.
+ * Not used for user-created renderbuffers!
+ */
+static GLboolean
+intel_alloc_window_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                           GLenum internalFormat, GLuint width, GLuint height)
+{
+   ASSERT(rb->Name == 0);
+   rb->Width = width;
+   rb->Height = height;
+   rb->_ActualFormat = internalFormat;
+   return GL_TRUE;
+}
+
+
+static GLboolean
+intel_nop_alloc_storage(GLcontext * ctx, struct gl_renderbuffer *rb,
+                        GLenum internalFormat, GLuint width, GLuint height)
+{
+   _mesa_problem(ctx, "intel_op_alloc_storage should never be called.");
+   return GL_FALSE;
+}
+
+
+
+/**
+ * Create a new intel_renderbuffer which corresponds to an on-screen window,
+ * not a user-created renderbuffer.
+ * \param width  the screen width
+ * \param height  the screen height
+ */
+struct intel_renderbuffer *
+intel_create_renderbuffer(GLenum intFormat, GLsizei width, GLsizei height,
+                          int offset, int pitch, int cpp, void *map)
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   struct intel_renderbuffer *irb;
+   const GLuint name = 0;
+
+   irb = CALLOC_STRUCT(intel_renderbuffer);
+   if (!irb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "creating renderbuffer");
+      return NULL;
+   }
+
+   _mesa_init_renderbuffer(&irb->Base, name);
+   irb->Base.ClassID = INTEL_RB_CLASS;
+
+   switch (intFormat) {
+   case GL_RGB5:
+      irb->Base._ActualFormat = GL_RGB5;
+      irb->Base._BaseFormat = GL_RGBA;
+      irb->Base.RedBits = 5;
+      irb->Base.GreenBits = 6;
+      irb->Base.BlueBits = 5;
+      irb->Base.DataType = GL_UNSIGNED_BYTE;
+      cpp = 2;
+      break;
+   case GL_RGBA8:
+      irb->Base._ActualFormat = GL_RGBA8;
+      irb->Base._BaseFormat = GL_RGBA;
+      irb->Base.RedBits = 8;
+      irb->Base.GreenBits = 8;
+      irb->Base.BlueBits = 8;
+      irb->Base.AlphaBits = 8;
+      irb->Base.DataType = GL_UNSIGNED_BYTE;
+      cpp = 4;
+      break;
+   case GL_STENCIL_INDEX8_EXT:
+      irb->Base._ActualFormat = GL_STENCIL_INDEX8_EXT;
+      irb->Base._BaseFormat = GL_STENCIL_INDEX;
+      irb->Base.StencilBits = 8;
+      irb->Base.DataType = GL_UNSIGNED_BYTE;
+      cpp = 1;
+      break;
+   case GL_DEPTH_COMPONENT16:
+      irb->Base._ActualFormat = GL_DEPTH_COMPONENT16;
+      irb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+      irb->Base.DepthBits = 16;
+      irb->Base.DataType = GL_UNSIGNED_SHORT;
+      cpp = 2;
+      break;
+   case GL_DEPTH_COMPONENT24:
+      irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      irb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+      irb->Base.DepthBits = 24;
+      irb->Base.DataType = GL_UNSIGNED_INT;
+      cpp = 4;
+      break;
+   case GL_DEPTH24_STENCIL8_EXT:
+      irb->Base._ActualFormat = GL_DEPTH24_STENCIL8_EXT;
+      irb->Base._BaseFormat = GL_DEPTH_STENCIL_EXT;
+      irb->Base.DepthBits = 24;
+      irb->Base.StencilBits = 8;
+      irb->Base.DataType = GL_UNSIGNED_INT_24_8_EXT;
+      cpp = 4;
+      break;
+   default:
+      _mesa_problem(NULL,
+                    "Unexpected intFormat in intel_create_renderbuffer");
+      return NULL;
+   }
+
+   irb->Base.InternalFormat = intFormat;
+
+   /* intel-specific methods */
+   irb->Base.Delete = intel_delete_renderbuffer;
+   irb->Base.AllocStorage = intel_alloc_window_storage;
+   irb->Base.GetPointer = intel_get_pointer;
+   /* This sets the Get/PutRow/Value functions */
+   intel_set_span_functions(&irb->Base);
+
+   irb->pfMap = map;
+   irb->pfPitch = pitch / cpp;	/* in pixels */
+
+#if 00
+   irb->region = intel_region_create_static(intel,
+                                            DRM_MM_TT,
+                                            offset, map, cpp, width, height);
+#endif
+
+   return irb;
+}
+
+
+/**
+ * Create a new renderbuffer object.
+ * Typically called via glBindRenderbufferEXT().
+ */
+static struct gl_renderbuffer *
+intel_new_renderbuffer(GLcontext * ctx, GLuint name)
+{
+   /*struct intel_context *intel = intel_context(ctx); */
+   struct intel_renderbuffer *irb;
+
+   irb = CALLOC_STRUCT(intel_renderbuffer);
+   if (!irb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "creating renderbuffer");
+      return NULL;
+   }
+
+   _mesa_init_renderbuffer(&irb->Base, name);
+   irb->Base.ClassID = INTEL_RB_CLASS;
+
+   /* intel-specific methods */
+   irb->Base.Delete = intel_delete_renderbuffer;
+   irb->Base.AllocStorage = intel_alloc_renderbuffer_storage;
+   irb->Base.GetPointer = intel_get_pointer;
+   /* span routines set in alloc_storage function */
+
+   return &irb->Base;
+}
+
+
+/**
+ * Called via glBindFramebufferEXT().
+ */
+static void
+intel_bind_framebuffer(GLcontext * ctx, GLenum target,
+                       struct gl_framebuffer *fb)
+{
+   if (target == GL_FRAMEBUFFER_EXT || target == GL_DRAW_FRAMEBUFFER_EXT) {
+      intel_draw_buffer(ctx, fb);
+      /* Integer depth range depends on depth buffer bits */
+      ctx->Driver.DepthRange(ctx, ctx->Viewport.Near, ctx->Viewport.Far);
+   }
+   else {
+      /* don't need to do anything if target == GL_READ_FRAMEBUFFER_EXT */
+   }
+}
+
+
+/**
+ * Called via glFramebufferRenderbufferEXT().
+ */
+static void
+intel_framebuffer_renderbuffer(GLcontext * ctx,
+                               struct gl_framebuffer *fb,
+                               GLenum attachment, struct gl_renderbuffer *rb)
+{
+   DBG("Intel FramebufferRenderbuffer %u %u\n", fb->Name, rb ? rb->Name : 0);
+
+   intelFlush(ctx);
+
+   _mesa_framebuffer_renderbuffer(ctx, fb, attachment, rb);
+   intel_draw_buffer(ctx, fb);
+}
+
+
+/**
+ * When glFramebufferTexture[123]D is called this function sets up the
+ * gl_renderbuffer wrapp around the texture image.
+ * This will have the region info needed for hardware rendering.
+ */
+static struct intel_renderbuffer *
+intel_wrap_texture(GLcontext * ctx, struct gl_texture_image *texImage)
+{
+   const GLuint name = ~0;      /* not significant, but distinct for debugging */
+   struct intel_renderbuffer *irb;
+
+   /* make an intel_renderbuffer to wrap the texture image */
+   irb = CALLOC_STRUCT(intel_renderbuffer);
+   if (!irb) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glFramebufferTexture");
+      return NULL;
+   }
+
+   _mesa_init_renderbuffer(&irb->Base, name);
+   irb->Base.ClassID = INTEL_RB_CLASS;
+
+   if (texImage->TexFormat == &_mesa_texformat_argb8888) {
+      irb->Base._ActualFormat = GL_RGBA8;
+      irb->Base._BaseFormat = GL_RGBA;
+      DBG("Render to RGBA8 texture OK\n");
+   }
+   else if (texImage->TexFormat == &_mesa_texformat_rgb565) {
+      irb->Base._ActualFormat = GL_RGB5;
+      irb->Base._BaseFormat = GL_RGB;
+      DBG("Render to RGB5 texture OK\n");
+   }
+   else if (texImage->TexFormat == &_mesa_texformat_z16) {
+      irb->Base._ActualFormat = GL_DEPTH_COMPONENT16;
+      irb->Base._BaseFormat = GL_DEPTH_COMPONENT;
+      DBG("Render to DEPTH16 texture OK\n");
+   }
+   else {
+      DBG("Render to texture BAD FORMAT %d\n",
+	  texImage->TexFormat->MesaFormat);
+      _mesa_free(irb);
+      return NULL;
+   }
+
+   irb->Base.InternalFormat = irb->Base._ActualFormat;
+   irb->Base.Width = texImage->Width;
+   irb->Base.Height = texImage->Height;
+   irb->Base.DataType = GL_UNSIGNED_BYTE;       /* FBO XXX fix */
+   irb->Base.RedBits = texImage->TexFormat->RedBits;
+   irb->Base.GreenBits = texImage->TexFormat->GreenBits;
+   irb->Base.BlueBits = texImage->TexFormat->BlueBits;
+   irb->Base.AlphaBits = texImage->TexFormat->AlphaBits;
+   irb->Base.DepthBits = texImage->TexFormat->DepthBits;
+
+   irb->Base.Delete = intel_delete_renderbuffer;
+   irb->Base.AllocStorage = intel_nop_alloc_storage;
+   intel_set_span_functions(&irb->Base);
+
+   irb->RenderToTexture = GL_TRUE;
+
+   return irb;
+}
+
+
+/**
+ * Called by glFramebufferTexture[123]DEXT() (and other places) to
+ * prepare for rendering into texture memory.  This might be called
+ * many times to choose different texture levels, cube faces, etc
+ * before intel_finish_render_texture() is ever called.
+ */
+static void
+intel_render_texture(GLcontext * ctx,
+                     struct gl_framebuffer *fb,
+                     struct gl_renderbuffer_attachment *att)
+{
+   struct gl_texture_image *newImage
+      = att->Texture->Image[att->CubeMapFace][att->TextureLevel];
+   struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer);
+   struct intel_texture_image *intel_image;
+   GLuint imageOffset;
+
+   (void) fb;
+
+   ASSERT(newImage);
+
+   if (!irb) {
+      irb = intel_wrap_texture(ctx, newImage);
+      if (irb) {
+         /* bind the wrapper to the attachment point */
+         att->Renderbuffer = &irb->Base;
+      }
+      else {
+         /* fallback to software rendering */
+         _mesa_render_texture(ctx, fb, att);
+         return;
+      }
+   }
+
+   DBG("Begin render texture tex=%u w=%d h=%d refcount=%d\n",
+       att->Texture->Name, newImage->Width, newImage->Height,
+       irb->Base.RefCount);
+
+   /* point the renderbufer's region to the texture image region */
+   intel_image = intel_texture_image(newImage);
+   if (irb->region != intel_image->mt->region)
+      intel_region_reference(&irb->region, intel_image->mt->region);
+
+   /* compute offset of the particular 2D image within the texture region */
+   imageOffset = intel_miptree_image_offset(intel_image->mt,
+                                            att->CubeMapFace,
+                                            att->TextureLevel);
+
+   if (att->Texture->Target == GL_TEXTURE_3D) {
+      const GLuint *offsets = intel_miptree_depth_offsets(intel_image->mt,
+                                                          att->TextureLevel);
+      imageOffset += offsets[att->Zoffset];
+   }
+
+   /* store that offset in the region */
+   intel_image->mt->region->draw_offset = imageOffset;
+
+   /* update drawing region, etc */
+   intel_draw_buffer(ctx, fb);
+}
+
+
+/**
+ * Called by Mesa when rendering to a texture is done.
+ */
+static void
+intel_finish_render_texture(GLcontext * ctx,
+                            struct gl_renderbuffer_attachment *att)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_renderbuffer *irb = intel_renderbuffer(att->Renderbuffer);
+
+   DBG("End render texture (tid %u) tex %u\n", _glthread_GetID(), att->Texture->Name);
+
+   if (irb) {
+      /* just release the region */
+      intel_region_release(intel->intelScreen, &irb->region);
+   }
+   else if (att->Renderbuffer) {
+      /* software fallback */
+      _mesa_finish_render_texture(ctx, att);
+      /* XXX FBO: Need to unmap the buffer (or in intelSpanRenderStart???) */
+   }
+}
+
+
+/**
+ * Do one-time context initializations related to GL_EXT_framebuffer_object.
+ * Hook in device driver functions.
+ */
+void
+intel_fbo_init(struct intel_context *intel)
+{
+   intel->ctx.Driver.NewFramebuffer = intel_new_framebuffer;
+   intel->ctx.Driver.NewRenderbuffer = intel_new_renderbuffer;
+   intel->ctx.Driver.BindFramebuffer = intel_bind_framebuffer;
+   intel->ctx.Driver.FramebufferRenderbuffer = intel_framebuffer_renderbuffer;
+   intel->ctx.Driver.RenderTexture = intel_render_texture;
+   intel->ctx.Driver.FinishRenderTexture = intel_finish_render_texture;
+}

src/mesa/drivers/dri/i915/intel_fbo.h

@@ -0,0 +1,80 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_FBO_H
+#define INTEL_FBO_H
+
+
+struct intel_context;
+struct intel_region;
+
+
+/**
+ * Intel renderbuffer, derived from gl_renderbuffer.
+ * Note: The PairedDepth and PairedStencil fields use renderbuffer IDs,
+ * not pointers because in some circumstances a deleted renderbuffer could
+ * result in a dangling pointer here.
+ */
+struct intel_renderbuffer
+{
+   struct gl_renderbuffer Base;
+   struct intel_region *region;
+   void *pfMap;                 /* possibly paged flipped map pointer */
+   GLuint pfPitch;              /* possibly paged flipped pitch */
+   GLboolean RenderToTexture;   /* RTT? */
+
+   GLuint PairedDepth;   /**< only used if this is a depth renderbuffer */
+   GLuint PairedStencil; /**< only used if this is a stencil renderbuffer */
+};
+
+
+extern struct intel_renderbuffer *intel_create_renderbuffer(GLenum intFormat,
+                                                            GLsizei width,
+                                                            GLsizei height,
+                                                            int offset,
+                                                            int pitch,
+                                                            int cpp,
+                                                            void *map);
+
+
+extern void intel_fbo_init(struct intel_context *intel);
+
+
+/* XXX make inline or macro */
+extern struct intel_renderbuffer *intel_get_renderbuffer(struct gl_framebuffer
+                                                         *fb,
+                                                         GLuint attIndex);
+
+
+/* XXX make inline or macro */
+extern struct intel_region *intel_get_rb_region(struct gl_framebuffer *fb,
+                                                GLuint attIndex);
+
+
+
+
+#endif /* INTEL_FBO_H */

src/mesa/drivers/dri/i915/intel_ioctl.c

@@ -38,623 +38,101 @@
 #include "intel_context.h"
 #include "intel_ioctl.h"
 #include "intel_batchbuffer.h"
+#include "intel_blit.h"
+#include "intel_regions.h"
 #include "drm.h"
 
-u_int32_t intelGetLastFrame (intelContextPtr intel) 
-{
-   int ret;
-   u_int32_t frame;
-   drm_i915_getparam_t gp;
-   
-   gp.param = I915_PARAM_LAST_DISPATCH;
-   gp.value = (int *)&frame;
-   ret = drmCommandWriteRead( intel->driFd, DRM_I915_GETPARAM,
-			      &gp, sizeof(gp) );
-   return frame;
-}
+#define FILE_DEBUG_FLAG DEBUG_IOCTL
 
-int intelEmitIrqLocked( intelContextPtr intel )
+int
+intelEmitIrqLocked(struct intel_context *intel)
 {
    drmI830IrqEmit ie;
    int ret, seq;
-      
-   assert(((*(int *)intel->driHwLock) & ~DRM_LOCK_CONT) == 
-	  (DRM_LOCK_HELD|intel->hHWContext));
+
+   assert(((*(int *) intel->driHwLock) & ~DRM_LOCK_CONT) ==
+          (DRM_LOCK_HELD | intel->hHWContext));
 
    ie.irq_seq = &seq;
-	 
-   ret = drmCommandWriteRead( intel->driFd, DRM_I830_IRQ_EMIT, 
-			      &ie, sizeof(ie) );
-   if ( ret ) {
-      fprintf( stderr, "%s: drmI830IrqEmit: %d\n", __FUNCTION__, ret );
+
+   ret = drmCommandWriteRead(intel->driFd, DRM_I830_IRQ_EMIT,
+                             &ie, sizeof(ie));
+   if (ret) {
+      fprintf(stderr, "%s: drmI830IrqEmit: %d\n", __FUNCTION__, ret);
       exit(1);
    }
-   
-   if (0)
-      fprintf(stderr, "%s -->  %d\n", __FUNCTION__, seq );
+
+   DBG("%s -->  %d\n", __FUNCTION__, seq);
 
    return seq;
 }
 
-void intelWaitIrq( intelContextPtr intel, int seq )
+void
+intelWaitIrq(struct intel_context *intel, int seq)
 {
    int ret;
-      
-   if (0)
-      fprintf(stderr, "%s %d\n", __FUNCTION__, seq );
+
+   DBG("%s %d\n", __FUNCTION__, seq);
 
    intel->iw.irq_seq = seq;
-	 
+
    do {
-     ret = drmCommandWrite( intel->driFd, DRM_I830_IRQ_WAIT, &intel->iw, sizeof(intel->iw) );
+      ret =
+         drmCommandWrite(intel->driFd, DRM_I830_IRQ_WAIT, &intel->iw,
+                         sizeof(intel->iw));
    } while (ret == -EAGAIN || ret == -EINTR);
 
-   if ( ret ) {
-      fprintf( stderr, "%s: drmI830IrqWait: %d\n", __FUNCTION__, ret );
-      if (0)
-	 intel_dump_batchbuffer( intel->alloc.offset,
-				 intel->alloc.ptr,
-				 intel->alloc.size );
+   if (ret) {
+      fprintf(stderr, "%s: drmI830IrqWait: %d\n", __FUNCTION__, ret);
       exit(1);
    }
 }
 
 
-
-static void age_intel( intelContextPtr intel, int age )
-{
-   GLuint i;
-
-   for (i = 0 ; i < MAX_TEXTURE_UNITS ; i++)
-      if (intel->CurrentTexObj[i]) 
-	 intel->CurrentTexObj[i]->age = age;
-}
-
-void intel_dump_batchbuffer( long offset,
-			     int *ptr,
-			     int count )
-{
-   int i;
-   fprintf(stderr, "\n\n\nSTART BATCH (%d dwords):\n", count);
-   for (i = 0; i < count/4; i += 4) 
-      fprintf(stderr, "\t0x%x: 0x%08x 0x%08x 0x%08x 0x%08x\n", 
-	      (unsigned int)offset + i*4, ptr[i], ptr[i+1], ptr[i+2], ptr[i+3]);
-   fprintf(stderr, "END BATCH\n\n\n");
-}
-
-void intelRefillBatchLocked( intelContextPtr intel, GLboolean allow_unlock )
-{
-   GLuint last_irq = intel->alloc.irq_emitted;
-   GLuint half = intel->alloc.size / 2;
-   GLuint buf = (intel->alloc.active_buf ^= 1);
-
-   intel->alloc.irq_emitted = intelEmitIrqLocked( intel );
-
-   if (last_irq) {
-      if (allow_unlock) UNLOCK_HARDWARE( intel ); 
-      intelWaitIrq( intel, last_irq );
-      if (allow_unlock) LOCK_HARDWARE( intel ); 
-   }
-
-   if (0)
-      fprintf(stderr, "%s: now using half %d\n", __FUNCTION__, buf);
-
-   intel->batch.start_offset = intel->alloc.offset + buf * half;
-   intel->batch.ptr = (unsigned char *)intel->alloc.ptr + buf * half;
-   intel->batch.size = half - 8;
-   intel->batch.space = half - 8;
-   assert(intel->batch.space >= 0);
-}
-
-#define MI_BATCH_BUFFER_END 	(0xA<<23)
-
-
-void intelFlushBatchLocked( intelContextPtr intel, 
-			    GLboolean ignore_cliprects,
-			    GLboolean refill,
-			    GLboolean allow_unlock)
+void
+intel_batch_ioctl(struct intel_context *intel,
+                  GLuint start_offset,
+                  GLuint used,
+                  GLboolean ignore_cliprects, GLboolean allow_unlock)
 {
    drmI830BatchBuffer batch;
 
    assert(intel->locked);
+   assert(used);
 
-   if (0)
-      fprintf(stderr, "%s used %d of %d offset %x..%x refill %d (started in %s)\n",
-	      __FUNCTION__, 
-	      (intel->batch.size - intel->batch.space), 
-	      intel->batch.size,
-	      intel->batch.start_offset,
-	      intel->batch.start_offset + 
-	      (intel->batch.size - intel->batch.space), 
-	      refill,
-	      intel->batch.func);
+   DBG("%s used %d offset %x..%x ignore_cliprects %d\n",
+       __FUNCTION__,
+       used, start_offset, start_offset + used, ignore_cliprects);
 
    /* Throw away non-effective packets.  Won't work once we have
     * hardware contexts which would preserve statechanges beyond a
     * single buffer.
     */
-   if (intel->numClipRects == 0 && !ignore_cliprects) {
-      
-      /* Without this yeild, an application with no cliprects can hog
-       * the hardware.  Without unlocking, the effect is much worse -
-       * effectively a lock-out of other contexts.
-       */
-      if (allow_unlock) {
-	 UNLOCK_HARDWARE( intel );
-	 sched_yield();
-	 LOCK_HARDWARE( intel );
-      }
-
-      /* Note that any state thought to have been emitted actually
-       * hasn't:
-       */
-      intel->batch.ptr -= (intel->batch.size - intel->batch.space);
-      intel->batch.space = intel->batch.size;
-      intel->vtbl.lost_hardware( intel ); 
-   }
-
-   if (intel->batch.space != intel->batch.size) {
-
-      if (intel->sarea->ctxOwner != intel->hHWContext) {
-	 intel->perf_boxes |= I830_BOX_LOST_CONTEXT;
-	 intel->sarea->ctxOwner = intel->hHWContext;
-      }
-
-      batch.start = intel->batch.start_offset;
-      batch.used = intel->batch.size - intel->batch.space;
-      batch.cliprects = intel->pClipRects;
-      batch.num_cliprects = ignore_cliprects ? 0 : intel->numClipRects;
-      batch.DR1 = 0;
-      batch.DR4 = ((((GLuint)intel->drawX) & 0xffff) | 
-		   (((GLuint)intel->drawY) << 16));
-      
-      if (intel->alloc.offset) {
-	 if ((batch.used & 0x4) == 0) {
-	    ((int *)intel->batch.ptr)[0] = 0;
-	    ((int *)intel->batch.ptr)[1] = MI_BATCH_BUFFER_END;
-	    batch.used += 0x8;
-	    intel->batch.ptr += 0x8;
-	 }
-	 else {
-	    ((int *)intel->batch.ptr)[0] = MI_BATCH_BUFFER_END;
-	    batch.used += 0x4;
-	    intel->batch.ptr += 0x4;
-	 }      
-      }
-
-      if (0)
- 	 intel_dump_batchbuffer( batch.start,
-				 (int *)(intel->batch.ptr - batch.used),
-				 batch.used );
-
-      intel->batch.start_offset += batch.used;
-      intel->batch.size -= batch.used;
-
-      if (intel->batch.size < 8) {
-	 refill = GL_TRUE;
-	 intel->batch.space = intel->batch.size = 0;
-      }
-      else {
-	 intel->batch.size -= 8;
-	 intel->batch.space = intel->batch.size;
-      }
 
 
-      assert(intel->batch.space >= 0);
-      assert(batch.start >= intel->alloc.offset);
-      assert(batch.start < intel->alloc.offset + intel->alloc.size);
-      assert(batch.start + batch.used > intel->alloc.offset);
-      assert(batch.start + batch.used <= 
-	     intel->alloc.offset + intel->alloc.size);
 
+   batch.start = start_offset;
+   batch.used = used;
+   batch.cliprects = intel->pClipRects;
+   batch.num_cliprects = ignore_cliprects ? 0 : intel->numClipRects;
+   batch.DR1 = 0;
+   batch.DR4 = ((((GLuint) intel->drawX) & 0xffff) |
+                (((GLuint) intel->drawY) << 16));
 
-      if (intel->alloc.offset) {
-	 if (drmCommandWrite (intel->driFd, DRM_I830_BATCHBUFFER, &batch, 
-			      sizeof(batch))) {
-	    fprintf(stderr, "DRM_I830_BATCHBUFFER: %d\n",  -errno);
-	    UNLOCK_HARDWARE(intel);
-	    exit(1);
-	 }
-      } else {
-	 drmI830CmdBuffer cmd;
-	 cmd.buf = (char *)intel->alloc.ptr + batch.start;
-	 cmd.sz = batch.used;
-	 cmd.DR1 = batch.DR1;
-	 cmd.DR4 = batch.DR4;
-	 cmd.num_cliprects = batch.num_cliprects;
-	 cmd.cliprects = batch.cliprects;
-	 
-	 if (drmCommandWrite (intel->driFd, DRM_I830_CMDBUFFER, &cmd, 
-			      sizeof(cmd))) {
-	    fprintf(stderr, "DRM_I830_CMDBUFFER: %d\n",  -errno);
-	    UNLOCK_HARDWARE(intel);
-	    exit(1);
-	 }
-      }	 
+   DBG("%s: 0x%x..0x%x DR4: %x cliprects: %d\n",
+       __FUNCTION__,
+       batch.start,
+       batch.start + batch.used * 4, batch.DR4, batch.num_cliprects);
 
-      
-      age_intel(intel, intel->sarea->last_enqueue);
-
-      /* FIXME: use hardware contexts to avoid 'losing' hardware after
-       * each buffer flush.
-       */
-      if (intel->batch.contains_geometry) 
-	 assert(intel->batch.last_emit_state == intel->batch.counter);
-
-      intel->batch.counter++;
-      intel->batch.contains_geometry = 0;
-      intel->batch.func = 0;
-      intel->vtbl.lost_hardware( intel );
-   }
-
-   if (refill)
-      intelRefillBatchLocked( intel, allow_unlock );
-}
-
-void intelFlushBatch( intelContextPtr intel, GLboolean refill )
-{
-   if (intel->locked) {
-      intelFlushBatchLocked( intel, GL_FALSE, refill, GL_FALSE );
-   } 
-   else {
-      LOCK_HARDWARE(intel);
-      intelFlushBatchLocked( intel, GL_FALSE, refill, GL_TRUE );
+   if (drmCommandWrite(intel->driFd, DRM_I830_BATCHBUFFER, &batch,
+                       sizeof(batch))) {
+      fprintf(stderr, "DRM_I830_BATCHBUFFER: %d\n", -errno);
       UNLOCK_HARDWARE(intel);
-   }
-}
-
-
-void intelWaitForIdle( intelContextPtr intel )
-{   
-   if (0)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   intel->vtbl.emit_flush( intel );
-   intelFlushBatch( intel, GL_TRUE );
-
-   /* Use an irq to wait for dma idle -- Need to track lost contexts
-    * to shortcircuit consecutive calls to this function:
-    */
-   intelWaitIrq( intel, intel->alloc.irq_emitted );
-   intel->alloc.irq_emitted = 0;
-}
-
-
-/**
- * Check if we need to rotate/warp the front color buffer to the
- * rotated screen.  We generally need to do this when we get a glFlush
- * or glFinish after drawing to the front color buffer.
- */
-static void
-intelCheckFrontRotate(GLcontext *ctx)
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   if (intel->ctx.DrawBuffer->_ColorDrawBufferMask[0] == BUFFER_BIT_FRONT_LEFT) {
-      intelScreenPrivate *screen = intel->intelScreen;
-      if (screen->current_rotation != 0) {
-         __DRIdrawablePrivate *dPriv = intel->driDrawable;
-         intelRotateWindow(intel, dPriv, BUFFER_BIT_FRONT_LEFT);
-      }
-   }
-}
-
-
-/**
- * NOT directly called via glFlush.
- */
-void intelFlush( GLcontext *ctx )
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-
-   if (intel->Fallback)
-      _swrast_flush( ctx );
-
-   INTEL_FIREVERTICES( intel );
-
-   if (intel->batch.size != intel->batch.space)
-      intelFlushBatch( intel, GL_FALSE );
-}
-
-
-/**
- * Called via glFlush.
- */
-void intelglFlush( GLcontext *ctx )
-{
-   intelFlush(ctx);
-   intelCheckFrontRotate(ctx);
-}
-
-
-void intelFinish( GLcontext *ctx  ) 
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   intelFlush( ctx );
-   intelWaitForIdle( intel );
-   intelCheckFrontRotate(ctx);
-}
-
-
-void intelClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
-		GLint cx, GLint cy, GLint cw, GLint ch)
-{
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   const GLuint colorMask = *((GLuint *) &ctx->Color.ColorMask);
-   GLbitfield tri_mask = 0;
-   GLbitfield blit_mask = 0;
-   GLbitfield swrast_mask = 0;
-
-   if (0)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* Take care of cliprects, which are handled differently for
-    * clears, etc.
-    */
-   intelFlush( &intel->ctx );
-
-   if (mask & BUFFER_BIT_FRONT_LEFT) {
-      if (colorMask == ~0) {
-	 blit_mask |= BUFFER_BIT_FRONT_LEFT;
-      } 
-      else {
-	 tri_mask |= BUFFER_BIT_FRONT_LEFT;
-      }
-   }
-
-   if (mask & BUFFER_BIT_BACK_LEFT) {
-      if (colorMask == ~0) {
-	 blit_mask |= BUFFER_BIT_BACK_LEFT;
-      } 
-      else {
-	 tri_mask |= BUFFER_BIT_BACK_LEFT;
-      }
-   }
-
-   if (mask & BUFFER_BIT_DEPTH) {
-      blit_mask |= BUFFER_BIT_DEPTH;
-   }
-
-   if (mask & BUFFER_BIT_STENCIL) {
-      if (!intel->hw_stencil) {
-	 swrast_mask |= BUFFER_BIT_STENCIL;
-      }
-      else if ((ctx->Stencil.WriteMask[0] & 0xff) != 0xff) {
-	 tri_mask |= BUFFER_BIT_STENCIL;
-      } 
-      else {
-	 blit_mask |= BUFFER_BIT_STENCIL;
-      }
-   }
-
-   swrast_mask |= (mask & BUFFER_BIT_ACCUM);
-
-   if (blit_mask) 
-      intelClearWithBlit( ctx, blit_mask, all, cx, cy, cw, ch );
-
-   if (tri_mask) 
-      intel->vtbl.clear_with_tris( intel, tri_mask, all, cx, cy, cw, ch);
-
-   if (swrast_mask)
-      _swrast_Clear( ctx, swrast_mask, all, cx, cy, cw, ch );
-}
-
-
-void
-intelRotateWindow(intelContextPtr intel, __DRIdrawablePrivate *dPriv,
-                  GLuint srcBuffer)
-{
-   if (intel->vtbl.rotate_window) {
-      intel->vtbl.rotate_window(intel, dPriv, srcBuffer);
-   }
-}
-
-
-void *intelAllocateAGP( intelContextPtr intel, GLsizei size )
-{
-   int region_offset;
-   drmI830MemAlloc alloc;
-   int ret;
-
-   if (0)
-      fprintf(stderr, "%s: %d bytes\n", __FUNCTION__, size);
-
-   alloc.region = I830_MEM_REGION_AGP;
-   alloc.alignment = 0;
-   alloc.size = size;
-   alloc.region_offset = &region_offset;
-
-   LOCK_HARDWARE(intel);
-
-   /* Make sure the global heap is initialized
-    */
-   if (intel->texture_heaps[0])
-      driAgeTextures( intel->texture_heaps[0] );
-
-
-   ret = drmCommandWriteRead( intel->driFd,
-			      DRM_I830_ALLOC,
-			      &alloc, sizeof(alloc));
-   
-   if (ret) {
-      fprintf(stderr, "%s: DRM_I830_ALLOC ret %d\n", __FUNCTION__, ret);
-      UNLOCK_HARDWARE(intel);
-      return NULL;
-   }
-   
-   if (0)
-      fprintf(stderr, "%s: allocated %d bytes\n", __FUNCTION__, size);
-
-   /* Need to propogate this information (agp memory in use) to our
-    * local texture lru.  The kernel has already updated the global
-    * lru.  An alternative would have been to allocate memory the
-    * usual way and then notify the kernel to pin the allocation.
-    */
-   if (intel->texture_heaps[0])
-      driAgeTextures( intel->texture_heaps[0] );
-
-   UNLOCK_HARDWARE(intel);   
-
-   return (void *)((char *)intel->intelScreen->tex.map + region_offset);
-}
-
-void intelFreeAGP( intelContextPtr intel, void *pointer )
-{
-   int region_offset;
-   drmI830MemFree memfree;
-   int ret;
-
-   region_offset = (char *)pointer - (char *)intel->intelScreen->tex.map;
-
-   if (region_offset < 0 || 
-       region_offset > intel->intelScreen->tex.size) {
-      fprintf(stderr, "offset %d outside range 0..%d\n", region_offset,
-	      intel->intelScreen->tex.size);
-      return;
-   }
-
-   memfree.region = I830_MEM_REGION_AGP;
-   memfree.region_offset = region_offset;
-   
-   ret = drmCommandWrite( intel->driFd,
-			  DRM_I830_FREE,
-			  &memfree, sizeof(memfree));
-   
-   if (ret) 
-      fprintf(stderr, "%s: DRM_I830_FREE ret %d\n", __FUNCTION__, ret);
-}
-
-/* This version of AllocateMemoryMESA allocates only agp memory, and
- * only does so after the point at which the driver has been
- * initialized.
- *
- * Theoretically a valid context isn't required.  However, in this
- * implementation, it is, as I'm using the hardware lock to protect
- * the kernel data structures, and the current context to get the
- * device fd.
- */
-void *intelAllocateMemoryMESA(__DRInativeDisplay *dpy, int scrn,
-			      GLsizei size, GLfloat readfreq,
-			      GLfloat writefreq, GLfloat priority)
-{
-   GET_CURRENT_CONTEXT(ctx);
-
-   if (INTEL_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s sz %d %f/%f/%f\n", __FUNCTION__, size, readfreq, 
-	      writefreq, priority);
-
-   if (getenv("INTEL_NO_ALLOC"))
-      return NULL;
-   
-   if (!ctx || INTEL_CONTEXT(ctx) == 0) 
-      return NULL;
-   
-   return intelAllocateAGP( INTEL_CONTEXT(ctx), size );
-}
-
-
-/* Called via glXFreeMemoryMESA() */
-void intelFreeMemoryMESA(__DRInativeDisplay *dpy, int scrn, GLvoid *pointer)
-{
-   GET_CURRENT_CONTEXT(ctx);
-   if (INTEL_DEBUG & DEBUG_IOCTL) 
-      fprintf(stderr, "%s %p\n", __FUNCTION__, pointer);
-
-   if (!ctx || INTEL_CONTEXT(ctx) == 0) {
-      fprintf(stderr, "%s: no context\n", __FUNCTION__);
-      return;
-   }
-
-   intelFreeAGP( INTEL_CONTEXT(ctx), pointer );
-}
-
-/* Called via glXGetMemoryOffsetMESA() 
- *
- * Returns offset of pointer from the start of agp aperture.
- */
-GLuint intelGetMemoryOffsetMESA(__DRInativeDisplay *dpy, int scrn, 
-				const GLvoid *pointer)
-{
-   GET_CURRENT_CONTEXT(ctx);
-   intelContextPtr intel;
-
-   if (!ctx || !(intel = INTEL_CONTEXT(ctx)) ) {
-      fprintf(stderr, "%s: no context\n", __FUNCTION__);
-      return ~0;
-   }
-
-   if (!intelIsAgpMemory( intel, pointer, 0 ))
-      return ~0;
-
-   return intelAgpOffsetFromVirtual( intel, pointer );
-}
-
-
-GLboolean intelIsAgpMemory( intelContextPtr intel, const GLvoid *pointer,
-			   GLint size )
-{
-   int offset = (char *)pointer - (char *)intel->intelScreen->tex.map;
-   int valid = (size >= 0 &&
-		offset >= 0 &&
-		offset + size < intel->intelScreen->tex.size);
-
-   if (INTEL_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "intelIsAgpMemory( %p ) : %d\n", pointer, valid );
-   
-   return valid;
-}
-
-
-GLuint intelAgpOffsetFromVirtual( intelContextPtr intel, const GLvoid *pointer )
-{
-   int offset = (char *)pointer - (char *)intel->intelScreen->tex.map;
-
-   if (offset < 0 || offset > intel->intelScreen->tex.size)
-      return ~0;
-   else
-      return intel->intelScreen->tex.offset + offset;
-}
-
-
-
-
-
-/* Flip the front & back buffes
- */
-void intelPageFlip( const __DRIdrawablePrivate *dPriv )
-{
-#if 0
-   intelContextPtr intel;
-   int tmp, ret;
-
-   if (INTEL_DEBUG & DEBUG_IOCTL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   intel = (intelContextPtr) dPriv->driContextPriv->driverPrivate;
-
-   intelFlush( &intel->ctx );
-   LOCK_HARDWARE( intel );
-
-   if (dPriv->pClipRects) {
-      *(drm_clip_rect_t *)intel->sarea->boxes = dPriv->pClipRects[0];
-      intel->sarea->nbox = 1;
-   }
-
-   ret = drmCommandNone(intel->driFd, DRM_I830_FLIP); 
-   if (ret) {
-      fprintf(stderr, "%s: %d\n", __FUNCTION__, ret);
-      UNLOCK_HARDWARE( intel );
       exit(1);
    }
 
-   tmp = intel->sarea->last_enqueue;
-   intelRefillBatchLocked( intel );
-   UNLOCK_HARDWARE( intel );
-
-
-   intelSetDrawBuffer( &intel->ctx, intel->ctx.Color.DriverDrawBuffer );
-#endif
+   /* FIXME: use hardware contexts to avoid 'losing' hardware after
+    * each buffer flush.
+    */
+   intel->vtbl.lost_hardware(intel);
 }

src/mesa/drivers/dri/i915/intel_ioctl.h

@@ -30,44 +30,11 @@
 
 #include "intel_context.h"
 
-extern void intelWaitAgeLocked( intelContextPtr intel, int age, GLboolean unlock );
+void intelWaitIrq(struct intel_context *intel, int seq);
+int intelEmitIrqLocked(struct intel_context *intel);
 
-extern void intelClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
-		       GLint cx, GLint cy, GLint cw, GLint ch);
-
-extern void intelPageFlip( const __DRIdrawablePrivate *dpriv );
-
-extern void intelRotateWindow(intelContextPtr intel,
-                              __DRIdrawablePrivate *dPriv, GLuint srcBuffer);
-
-extern void intelWaitForIdle( intelContextPtr intel );
-extern void intelFlushBatch( intelContextPtr intel, GLboolean refill );
-extern void intelFlushBatchLocked( intelContextPtr intel,
-				   GLboolean ignore_cliprects,
-				   GLboolean refill,
-				   GLboolean allow_unlock);
-extern void intelRefillBatchLocked( intelContextPtr intel, GLboolean allow_unlock );
-extern void intelFinish( GLcontext *ctx );
-extern void intelFlush( GLcontext *ctx );
-extern void intelglFlush( GLcontext *ctx );
-
-extern void *intelAllocateAGP( intelContextPtr intel, GLsizei size );
-extern void intelFreeAGP( intelContextPtr intel, void *pointer );
-
-extern void *intelAllocateMemoryMESA( __DRInativeDisplay *dpy, int scrn, 
-				      GLsizei size, GLfloat readfreq,
-				      GLfloat writefreq, GLfloat priority );
-
-extern void intelFreeMemoryMESA( __DRInativeDisplay *dpy, int scrn, 
-				 GLvoid *pointer );
-
-extern GLuint intelGetMemoryOffsetMESA( __DRInativeDisplay *dpy, int scrn, const GLvoid *pointer );
-extern GLboolean intelIsAgpMemory( intelContextPtr intel, const GLvoid *pointer,
-				  GLint size );
-
-extern GLuint intelAgpOffsetFromVirtual( intelContextPtr intel, const GLvoid *p );
-
-extern void intelWaitIrq( intelContextPtr intel, int seq );
-extern u_int32_t intelGetLastFrame (intelContextPtr intel);
-extern int intelEmitIrqLocked( intelContextPtr intel );
+void intel_batch_ioctl(struct intel_context *intel,
+                       GLuint start_offset,
+                       GLuint used,
+                       GLboolean ignore_cliprects, GLboolean allow_unlock);
 #endif

src/mesa/drivers/dri/i915/intel_mipmap_tree.c

@@ -0,0 +1,341 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "intel_context.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
+#include "enums.h"
+
+#define FILE_DEBUG_FLAG DEBUG_MIPTREE
+
+static GLenum
+target_to_target(GLenum target)
+{
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB:
+      return GL_TEXTURE_CUBE_MAP_ARB;
+   default:
+      return target;
+   }
+}
+
+struct intel_mipmap_tree *
+intel_miptree_create(struct intel_context *intel,
+                     GLenum target,
+                     GLenum internal_format,
+                     GLuint first_level,
+                     GLuint last_level,
+                     GLuint width0,
+                     GLuint height0,
+                     GLuint depth0, GLuint cpp, GLboolean compressed)
+{
+   GLboolean ok;
+   struct intel_mipmap_tree *mt = calloc(sizeof(*mt), 1);
+
+   DBG("%s target %s format %s level %d..%d\n", __FUNCTION__,
+       _mesa_lookup_enum_by_nr(target),
+       _mesa_lookup_enum_by_nr(internal_format), first_level, last_level);
+
+   mt->target = target_to_target(target);
+   mt->internal_format = internal_format;
+   mt->first_level = first_level;
+   mt->last_level = last_level;
+   mt->width0 = width0;
+   mt->height0 = height0;
+   mt->depth0 = depth0;
+   mt->cpp = compressed ? 2 : cpp;
+   mt->compressed = compressed;
+   mt->refcount = 0; /*Allow for unused miptrees */
+
+   switch (intel->intelScreen->deviceID) {
+   case PCI_CHIP_I945_G:
+   case PCI_CHIP_I945_GM:
+      ok = i945_miptree_layout(mt);
+      break;
+   case PCI_CHIP_I915_G:
+   case PCI_CHIP_I915_GM:
+   case PCI_CHIP_I830_M:
+   case PCI_CHIP_I855_GM:
+   case PCI_CHIP_I865_G:
+   default:
+      /* All the i830 chips and the i915 use this layout:
+       */
+      ok = i915_miptree_layout(mt);
+      break;
+   }
+
+   if (ok)
+      mt->region = intel_region_alloc(intel->intelScreen,
+                                      mt->cpp, mt->pitch, mt->total_height);
+
+   if (!mt->region) {
+      free(mt);
+      return NULL;
+   }
+
+   return mt;
+}
+
+
+void
+intel_miptree_reference(struct intel_mipmap_tree **dst,
+                        struct intel_mipmap_tree *src)
+{
+   src->refcount++;
+   *dst = src;
+   DBG("%s %p refcount now %d\n", __FUNCTION__, src, src->refcount);
+}
+
+void
+intel_miptree_release(struct intel_context *intel,
+                      struct intel_mipmap_tree **mt)
+{
+   if (!*mt)
+      return;
+
+   DBG("%s %p refcount will be %d\n", __FUNCTION__, *mt, (*mt)->refcount - 1);
+   if (--(*mt)->refcount <= 0) {
+      GLuint i;
+
+      DBG("%s deleting %p\n", __FUNCTION__, *mt);
+
+      intel_region_release(intel->intelScreen, &((*mt)->region));
+
+      for (i = 0; i < MAX_TEXTURE_LEVELS; i++)
+         if ((*mt)->level[i].image_offset)
+            free((*mt)->level[i].image_offset);
+
+      free(*mt);
+   }
+   *mt = NULL;
+}
+
+
+
+
+/* Can the image be pulled into a unified mipmap tree.  This mirrors
+ * the completeness test in a lot of ways.
+ *
+ * Not sure whether I want to pass gl_texture_image here.
+ */
+GLboolean
+intel_miptree_match_image(struct intel_mipmap_tree *mt,
+                          struct gl_texture_image *image,
+                          GLuint face, GLuint level)
+{
+   /* Images with borders are never pulled into mipmap trees. 
+    */
+   if (image->Border) 
+      return GL_FALSE;
+
+   if (image->InternalFormat != mt->internal_format ||
+       image->IsCompressed != mt->compressed)
+      return GL_FALSE;
+
+   /* Test image dimensions against the base level image adjusted for
+    * minification.  This will also catch images not present in the
+    * tree, changed targets, etc.
+    */
+   if (image->Width != mt->level[level].width ||
+       image->Height != mt->level[level].height ||
+       image->Depth != mt->level[level].depth)
+      return GL_FALSE;
+
+   return GL_TRUE;
+}
+
+
+void
+intel_miptree_set_level_info(struct intel_mipmap_tree *mt,
+                             GLuint level,
+                             GLuint nr_images,
+                             GLuint x, GLuint y, GLuint w, GLuint h, GLuint d)
+{
+
+   mt->level[level].width = w;
+   mt->level[level].height = h;
+   mt->level[level].depth = d;
+   mt->level[level].level_offset = (x + y * mt->pitch) * mt->cpp;
+   mt->level[level].nr_images = nr_images;
+
+   DBG("%s level %d size: %d,%d,%d offset %d,%d (0x%x)\n", __FUNCTION__,
+       level, w, h, d, x, y, mt->level[level].level_offset);
+
+   /* Not sure when this would happen, but anyway: 
+    */
+   if (mt->level[level].image_offset) {
+      free(mt->level[level].image_offset);
+      mt->level[level].image_offset = NULL;
+   }
+
+   assert(nr_images);
+
+   mt->level[level].image_offset = malloc(nr_images * sizeof(GLuint));
+   mt->level[level].image_offset[0] = 0;
+}
+
+
+
+void
+intel_miptree_set_image_offset(struct intel_mipmap_tree *mt,
+                               GLuint level, GLuint img, GLuint x, GLuint y)
+{
+   if (img == 0 && level == 0)
+      assert(x == 0 && y == 0);
+
+   assert(img < mt->level[level].nr_images);
+
+   mt->level[level].image_offset[img] = (x + y * mt->pitch);
+
+   DBG("%s level %d img %d pos %d,%d image_offset %x\n",
+       __FUNCTION__, level, img, x, y, mt->level[level].image_offset[img]);
+}
+
+
+/* Although we use the image_offset[] array to store relative offsets
+ * to cube faces, Mesa doesn't know anything about this and expects
+ * each cube face to be treated as a separate image.
+ *
+ * These functions present that view to mesa:
+ */
+const GLuint *
+intel_miptree_depth_offsets(struct intel_mipmap_tree *mt, GLuint level)
+{
+   static const GLuint zero = 0;
+
+   if (mt->target != GL_TEXTURE_3D || mt->level[level].nr_images == 1)
+      return &zero;
+   else
+      return mt->level[level].image_offset;
+}
+
+
+GLuint
+intel_miptree_image_offset(struct intel_mipmap_tree * mt,
+                           GLuint face, GLuint level)
+{
+   if (mt->target == GL_TEXTURE_CUBE_MAP_ARB)
+      return (mt->level[level].level_offset +
+              mt->level[level].image_offset[face] * mt->cpp);
+   else
+      return mt->level[level].level_offset;
+}
+
+
+
+/**
+ * Map a teximage in a mipmap tree.
+ * \param row_stride  returns row stride in bytes
+ * \param image_stride  returns image stride in bytes (for 3D textures).
+ * \return address of mapping
+ */
+GLubyte *
+intel_miptree_image_map(struct intel_context * intel,
+                        struct intel_mipmap_tree * mt,
+                        GLuint face,
+                        GLuint level,
+                        GLuint * row_stride, GLuint * image_offsets)
+{
+   DBG("%s \n", __FUNCTION__);
+
+   if (row_stride)
+      *row_stride = mt->pitch * mt->cpp;
+
+   if (image_offsets)
+      memcpy(image_offsets, mt->level[level].image_offset,
+             mt->level[level].depth * sizeof(GLuint));
+
+   return (intel_region_map(intel->intelScreen, mt->region) +
+           intel_miptree_image_offset(mt, face, level));
+}
+
+void
+intel_miptree_image_unmap(struct intel_context *intel,
+                          struct intel_mipmap_tree *mt)
+{
+   DBG("%s\n", __FUNCTION__);
+   intel_region_unmap(intel->intelScreen, mt->region);
+}
+
+
+
+/* Upload data for a particular image.
+ */
+void
+intel_miptree_image_data(struct intel_context *intel,
+                         struct intel_mipmap_tree *dst,
+                         GLuint face,
+                         GLuint level,
+                         void *src,
+                         GLuint src_row_pitch, GLuint src_image_pitch)
+{
+   GLuint depth = dst->level[level].depth;
+   GLuint dst_offset = intel_miptree_image_offset(dst, face, level);
+   const GLuint *dst_depth_offset = intel_miptree_depth_offsets(dst, level);
+   GLuint i;
+
+   DBG("%s\n", __FUNCTION__);
+   for (i = 0; i < depth; i++) {
+      intel_region_data(intel->intelScreen, dst->region, dst_offset + dst_depth_offset[i], 0, 0, src, src_row_pitch, 0, 0,   /* source x,y */
+                        dst->level[level].width, dst->level[level].height);
+
+      src += src_image_pitch;
+   }
+}
+
+/* Copy mipmap image between trees
+ */
+void
+intel_miptree_image_copy(struct intel_context *intel,
+                         struct intel_mipmap_tree *dst,
+                         GLuint face, GLuint level,
+                         struct intel_mipmap_tree *src)
+{
+   GLuint width = src->level[level].width;
+   GLuint height = src->level[level].height;
+   GLuint depth = src->level[level].depth;
+   GLuint dst_offset = intel_miptree_image_offset(dst, face, level);
+   GLuint src_offset = intel_miptree_image_offset(src, face, level);
+   const GLuint *dst_depth_offset = intel_miptree_depth_offsets(dst, level);
+   const GLuint *src_depth_offset = intel_miptree_depth_offsets(src, level);
+   GLuint i;
+
+   for (i = 0; i < depth; i++) {
+      intel_region_copy(intel->intelScreen,
+                        dst->region, dst_offset + dst_depth_offset[i],
+                        0,
+                        0,
+                        src->region, src_offset + src_depth_offset[i],
+                        0, 0, width, height);
+   }
+
+}

src/mesa/drivers/dri/i915/intel_mipmap_tree.h

@@ -0,0 +1,198 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_MIPMAP_TREE_H
+#define INTEL_MIPMAP_TREE_H
+
+#include "intel_regions.h"
+
+/* A layer on top of the intel_regions code which adds:
+ *
+ * - Code to size and layout a region to hold a set of mipmaps.
+ * - Query to determine if a new image fits in an existing tree.
+ * - More refcounting 
+ *     - maybe able to remove refcounting from intel_region?
+ * - ?
+ *
+ * The fixed mipmap layout of intel hardware where one offset
+ * specifies the position of all images in a mipmap hierachy
+ * complicates the implementation of GL texture image commands,
+ * compared to hardware where each image is specified with an
+ * independent offset.
+ *
+ * In an ideal world, each texture object would be associated with a
+ * single bufmgr buffer or 2d intel_region, and all the images within
+ * the texture object would slot into the tree as they arrive.  The
+ * reality can be a little messier, as images can arrive from the user
+ * with sizes that don't fit in the existing tree, or in an order
+ * where the tree layout cannot be guessed immediately.  
+ * 
+ * This structure encodes an idealized mipmap tree.  The GL image
+ * commands build these where possible, otherwise store the images in
+ * temporary system buffers.
+ */
+
+
+/**
+ * Describes the location of each texture image within a texture region.
+ */
+struct intel_mipmap_level
+{
+   GLuint level_offset;
+   GLuint width;
+   GLuint height;
+   GLuint depth;
+   GLuint nr_images;
+
+   /* Explicitly store the offset of each image for each cube face or
+    * depth value.  Pretty much have to accept that hardware formats
+    * are going to be so diverse that there is no unified way to
+    * compute the offsets of depth/cube images within a mipmap level,
+    * so have to store them as a lookup table:
+    */
+   GLuint *image_offset;
+};
+
+struct intel_mipmap_tree
+{
+   /* Effectively the key:
+    */
+   GLenum target;
+   GLenum internal_format;
+
+   GLuint first_level;
+   GLuint last_level;
+
+   GLuint width0, height0, depth0; /**< Level zero image dimensions */
+   GLuint cpp;
+   GLboolean compressed;
+
+   /* Derived from the above:
+    */
+   GLuint pitch;
+   GLuint depth_pitch;          /* per-image on i945? */
+   GLuint total_height;
+
+   /* Includes image offset tables:
+    */
+   struct intel_mipmap_level level[MAX_TEXTURE_LEVELS];
+
+   /* The data is held here:
+    */
+   struct intel_region *region;
+
+   /* These are also refcounted:
+    */
+   GLuint refcount;
+};
+
+
+
+struct intel_mipmap_tree *intel_miptree_create(struct intel_context *intel,
+                                               GLenum target,
+                                               GLenum internal_format,
+                                               GLuint first_level,
+                                               GLuint last_level,
+                                               GLuint width0,
+                                               GLuint height0,
+                                               GLuint depth0,
+                                               GLuint cpp,
+                                               GLboolean compressed);
+
+void intel_miptree_reference(struct intel_mipmap_tree **dst,
+                             struct intel_mipmap_tree *src);
+
+void intel_miptree_release(struct intel_context *intel,
+                           struct intel_mipmap_tree **mt);
+
+/* Check if an image fits an existing mipmap tree layout
+ */
+GLboolean intel_miptree_match_image(struct intel_mipmap_tree *mt,
+                                    struct gl_texture_image *image,
+                                    GLuint face, GLuint level);
+
+/* Return a pointer to an image within a tree.  Return image stride as
+ * well.
+ */
+GLubyte *intel_miptree_image_map(struct intel_context *intel,
+                                 struct intel_mipmap_tree *mt,
+                                 GLuint face,
+                                 GLuint level,
+                                 GLuint * row_stride, GLuint * image_stride);
+
+void intel_miptree_image_unmap(struct intel_context *intel,
+                               struct intel_mipmap_tree *mt);
+
+
+/* Return the linear offset of an image relative to the start of the
+ * tree:
+ */
+GLuint intel_miptree_image_offset(struct intel_mipmap_tree *mt,
+                                  GLuint face, GLuint level);
+
+/* Return pointers to each 2d slice within an image.  Indexed by depth
+ * value.
+ */
+const GLuint *intel_miptree_depth_offsets(struct intel_mipmap_tree *mt,
+                                          GLuint level);
+
+
+void intel_miptree_set_level_info(struct intel_mipmap_tree *mt,
+                                  GLuint level,
+                                  GLuint nr_images,
+                                  GLuint x, GLuint y,
+                                  GLuint w, GLuint h, GLuint d);
+
+void intel_miptree_set_image_offset(struct intel_mipmap_tree *mt,
+                                    GLuint level,
+                                    GLuint img, GLuint x, GLuint y);
+
+
+/* Upload an image into a tree
+ */
+void intel_miptree_image_data(struct intel_context *intel,
+                              struct intel_mipmap_tree *dst,
+                              GLuint face,
+                              GLuint level,
+                              void *src,
+                              GLuint src_row_pitch, GLuint src_image_pitch);
+
+/* Copy an image between two trees
+ */
+void intel_miptree_image_copy(struct intel_context *intel,
+                              struct intel_mipmap_tree *dst,
+                              GLuint face, GLuint level,
+                              struct intel_mipmap_tree *src);
+
+/* i915_mipmap_tree.c:
+ */
+GLboolean i915_miptree_layout(struct intel_mipmap_tree *mt);
+GLboolean i945_miptree_layout(struct intel_mipmap_tree *mt);
+
+
+
+#endif

src/mesa/drivers/dri/i915/intel_pixel.c

@@ -1,6 +1,6 @@
 /**************************************************************************
  * 
- * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -12,7 +12,7 @@
  * the following conditions:
  * 
  * The above copyright notice and this permission notice (including the
- * next paragraph) shall be included in all copies or substantial portions
+ * next paragraph) shall be included in all copies or substantial portionsalloc
  * of the Software.
  * 
  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
@@ -25,487 +25,95 @@
  * 
  **************************************************************************/
 
-#include "glheader.h"
 #include "enums.h"
-#include "mtypes.h"
-#include "macros.h"
+#include "state.h"
 #include "swrast/swrast.h"
 
-#include "intel_screen.h"
 #include "intel_context.h"
-#include "intel_ioctl.h"
-#include "intel_batchbuffer.h"
-
-
-
-static GLboolean
-check_color( const GLcontext *ctx, GLenum type, GLenum format,
-	     const struct gl_pixelstore_attrib *packing,
-	     const void *pixels, GLint sz, GLint pitch )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   GLuint cpp = intel->intelScreen->cpp;
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (	(pitch & 63) ||
-	ctx->_ImageTransferState ||
-	packing->SwapBytes ||
-	packing->LsbFirst) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: failed 1\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   if ( type == GL_UNSIGNED_INT_8_8_8_8_REV && 
-	cpp == 4 && 
-	format == GL_BGRA ) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: passed 2\n", __FUNCTION__);
-      return GL_TRUE;
-   }
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s: failed\n", __FUNCTION__);
-
-   return GL_FALSE;
-}
-
-static GLboolean
-check_color_per_fragment_ops( const GLcontext *ctx )
-{
-   int result;
-   result = (!(     ctx->Color.AlphaEnabled || 
-		    ctx->Depth.Test ||
-		    ctx->Fog.Enabled ||
-		    ctx->Scissor.Enabled ||
-		    ctx->Stencil.Enabled ||
-		    !ctx->Color.ColorMask[0] ||
-		    !ctx->Color.ColorMask[1] ||
-		    !ctx->Color.ColorMask[2] ||
-		    !ctx->Color.ColorMask[3] ||
-		    ctx->Color.ColorLogicOpEnabled ||
-		    ctx->Texture._EnabledUnits
-           ) &&
-	   ctx->Current.RasterPosValid);
-   
-   return result;
-}
+#include "intel_pixel.h"
+#include "intel_regions.h"
 
 
 /**
- * Clip the given rectangle against the buffer's bounds (including scissor).
- * \param size returns the 
- * \return GL_TRUE if any pixels remain, GL_FALSE if totally clipped.
+ * Check if any fragment operations are in effect which might effect
+ * glDraw/CopyPixels.
+ */
+GLboolean
+intel_check_blit_fragment_ops(GLcontext * ctx)
+{
+   if (ctx->NewState)
+      _mesa_update_state(ctx);
+
+   /* XXX Note: Scissor could be done with the blitter:
+    */
+   return !(ctx->_ImageTransferState ||
+            ctx->Color.AlphaEnabled ||
+            ctx->Depth.Test ||
+            ctx->Fog.Enabled ||
+            ctx->Scissor.Enabled ||
+            ctx->Stencil.Enabled ||
+            !ctx->Color.ColorMask[0] ||
+            !ctx->Color.ColorMask[1] ||
+            !ctx->Color.ColorMask[2] ||
+            !ctx->Color.ColorMask[3] ||
+            ctx->Color.ColorLogicOpEnabled ||
+            ctx->Texture._EnabledUnits || ctx->FragmentProgram._Enabled);
+}
+
+
+GLboolean
+intel_check_meta_tex_fragment_ops(GLcontext * ctx)
+{
+   if (ctx->NewState)
+      _mesa_update_state(ctx);
+
+   /* Some of _ImageTransferState (scale, bias) could be done with
+    * fragment programs on i915.
+    */
+   return !(ctx->_ImageTransferState || ctx->Fog.Enabled ||     /* not done yet */
+            ctx->Texture._EnabledUnits || ctx->FragmentProgram._Enabled);
+}
+
+/* The intel_region struct doesn't really do enough to capture the
+ * format of the pixels in the region.  For now this code assumes that
+ * the region is a display surface and hence is either ARGB8888 or
+ * RGB565.
+ * XXX FBO: If we'd pass in the intel_renderbuffer instead of region, we'd
+ * know the buffer's pixel format.
  *
- * XXX Replace this with _mesa_clip_drawpixels() and _mesa_clip_readpixels()
- * from Mesa 6.4.  We shouldn't apply scissor for ReadPixels.
+ * \param format  as given to glDraw/ReadPixels
+ * \param type  as given to glDraw/ReadPixels
  */
-static GLboolean
-clip_pixelrect( const GLcontext *ctx,
-		const GLframebuffer *buffer,
-		GLint *x, GLint *y,
-		GLsizei *width, GLsizei *height)
+GLboolean
+intel_check_blit_format(struct intel_region * region,
+                        GLenum format, GLenum type)
 {
-   /* left clipping */
-   if (*x < buffer->_Xmin) {
-      *width -= (buffer->_Xmin - *x);
-      *x = buffer->_Xmin;
-   }
-
-   /* right clipping */
-   if (*x + *width > buffer->_Xmax)
-      *width -= (*x + *width - buffer->_Xmax - 1);
-
-   if (*width <= 0)
-      return GL_FALSE;
-
-   /* bottom clipping */
-   if (*y < buffer->_Ymin) {
-      *height -= (buffer->_Ymin - *y);
-      *y = buffer->_Ymin;
-   }
-
-   /* top clipping */
-   if (*y + *height > buffer->_Ymax)
-      *height -= (*y + *height - buffer->_Ymax - 1);
-
-   if (*height <= 0)
-      return GL_FALSE;
-
-   return GL_TRUE;
-}
-
-
-/**
- * Compute intersection of a clipping rectangle and pixel rectangle,
- * returning results in x/y/w/hOut vars.
- * \return GL_TRUE if there's intersection, GL_FALSE if disjoint.
- */
-static INLINE GLboolean
-intersect_region(const drm_clip_rect_t *box,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLint *xOut, GLint *yOut, GLint *wOut, GLint *hOut)
-{
-   GLint bx = box->x1;
-   GLint by = box->y1;
-   GLint bw = box->x2 - bx;
-   GLint bh = box->y2 - by;
-
-   if (bx < x) bw -= x - bx, bx = x;
-   if (by < y) bh -= y - by, by = y;
-   if (bx + bw > x + width) bw = x + width - bx;
-   if (by + bh > y + height) bh = y + height - by;
-
-   *xOut = bx;
-   *yOut = by;
-   *wOut = bw;
-   *hOut = bh;
-
-   if (bw <= 0) return GL_FALSE;
-   if (bh <= 0) return GL_FALSE;
-
-   return GL_TRUE;
-}
-
-
-
-static GLboolean
-intelTryReadPixels( GLcontext *ctx,
-		  GLint x, GLint y, GLsizei width, GLsizei height,
-		  GLenum format, GLenum type,
-		  const struct gl_pixelstore_attrib *pack,
-		  GLvoid *pixels )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   GLint size = 0; /* not really used */
-   GLint pitch = pack->RowLength ? pack->RowLength : width;
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   /* Only accelerate reading to agp buffers.
-    */
-   if ( !intelIsAgpMemory(intel, pixels, 
-			pitch * height * intel->intelScreen->cpp ) ) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: dest not agp\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   /* Need GL_PACK_INVERT_MESA to cope with upsidedown results from
-    * blitter:
-    */
-   if (!pack->Invert) {
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "%s: MESA_PACK_INVERT not set\n", __FUNCTION__);
-      return GL_FALSE;
-   }
-
-   if (!check_color(ctx, type, format, pack, pixels, size, pitch))
-      return GL_FALSE;
-
-   switch ( intel->intelScreen->cpp ) {
-   case 4:
-      break;
-   default:
-      return GL_FALSE;
-   }
-
-
-   /* Although the blits go on the command buffer, need to do this and
-    * fire with lock held to guarentee cliprects and drawing offset are
-    * correct.
-    *
-    * This is an unusual situation however, as the code which flushes
-    * a full command buffer expects to be called unlocked.  As a
-    * workaround, immediately flush the buffer on aquiring the lock.
-    */
-   intelFlush( &intel->ctx );
-   LOCK_HARDWARE( intel );
-   {
-      __DRIdrawablePrivate *dPriv = intel->driDrawable;
-      int nbox = dPriv->numClipRects;
-      int src_offset = intel->readRegion->offset;
-      int src_pitch = intel->intelScreen->front.pitch;
-      int dst_offset = intelAgpOffsetFromVirtual( intel, pixels);
-      drm_clip_rect_t *box = dPriv->pClipRects;
-      int i;
-
-      assert(dst_offset != ~0);  /* should have been caught above */
-
-      if (!clip_pixelrect(ctx, ctx->ReadBuffer, &x, &y, &width, &height)) {
-	 UNLOCK_HARDWARE( intel );
-	 if (INTEL_DEBUG & DEBUG_PIXEL)
-	    fprintf(stderr, "%s totally clipped -- nothing to do\n",
-		    __FUNCTION__);
-	 return GL_TRUE;
-      }
-
-      /* convert to screen coords (y=0=top) */
-      y = dPriv->h - y - height;
-      x += dPriv->x;
-      y += dPriv->y;
-
-      if (INTEL_DEBUG & DEBUG_PIXEL)
-	 fprintf(stderr, "readpixel blit src_pitch %d dst_pitch %d\n",
-		 src_pitch, pitch);
-
-      /* We don't really have to do window clipping for readpixels.
-       * The OpenGL spec says that pixels read from outside the
-       * visible window region (pixel ownership) have undefined value.
-       */
-      for (i = 0 ; i < nbox ; i++)
-      {
-         GLint bx, by, bw, bh;
-         if (intersect_region(box+i, x, y, width, height,
-                              &bx, &by, &bw, &bh)) {
-            intelEmitCopyBlitLocked( intel,
-                                     intel->intelScreen->cpp,
-                                     src_pitch, src_offset,
-                                     pitch, dst_offset,
-                                     bx, by,
-                                     bx - x, by - y,
-                                     bw, bh );
-         }
-      }
-   }
-   UNLOCK_HARDWARE( intel );
-   intelFinish( &intel->ctx );
-
-   return GL_TRUE;
-}
-
-static void
-intelReadPixels( GLcontext *ctx,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLenum format, GLenum type,
-		 const struct gl_pixelstore_attrib *pack,
-		 GLvoid *pixels )
-{
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (!intelTryReadPixels( ctx, x, y, width, height, format, type, pack, 
-                            pixels))
-      _swrast_ReadPixels( ctx, x, y, width, height, format, type, pack, 
-			  pixels);
-}
-
-
-
-
-static void do_draw_pix( GLcontext *ctx,
-			 GLint x, GLint y, GLsizei width, GLsizei height,
-			 GLint pitch,
-			 const void *pixels,
-			 GLuint dest )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;
-   drm_clip_rect_t *box = dPriv->pClipRects;
-   int nbox = dPriv->numClipRects;
-   int i;
-   int src_offset = intelAgpOffsetFromVirtual( intel, pixels);
-   int src_pitch = pitch;
-
-   assert(src_offset != ~0);  /* should be caught earlier */
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   intelFlush( &intel->ctx );
-   LOCK_HARDWARE( intel );
-   if (ctx->DrawBuffer)
-   {
-      y -= height;			/* cope with pixel zoom */
-   
-      if (!clip_pixelrect(ctx, ctx->DrawBuffer,
-			  &x, &y, &width, &height)) {
-	 UNLOCK_HARDWARE( intel );
-	 return;
-      }
-
-      y = dPriv->h - y - height; 	/* convert from gl to hardware coords */
-      x += dPriv->x;
-      y += dPriv->y;
-
-      for (i = 0 ; i < nbox ; i++ )
-      {
-	 GLint bx, by, bw, bh;
-	 if (intersect_region(box + i, x, y, width, height,
-			      &bx, &by, &bw, &bh)) {
-            intelEmitCopyBlitLocked( intel,
-                                     intel->intelScreen->cpp,
-                                     src_pitch, src_offset,
-                                     intel->intelScreen->front.pitch,
-                                     intel->drawRegion->offset,
-                                     bx - x, by - y,
-                                     bx, by,
-                                     bw, bh );
-         }
-      }
-   }
-   UNLOCK_HARDWARE( intel );
-   intelFinish( &intel->ctx );
-}
-
-
-
-static GLboolean
-intelTryDrawPixels( GLcontext *ctx,
-		  GLint x, GLint y, GLsizei width, GLsizei height,
-		  GLenum format, GLenum type,
-		  const struct gl_pixelstore_attrib *unpack,
-		  const GLvoid *pixels )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   GLint pitch = unpack->RowLength ? unpack->RowLength : width;
-   GLuint dest;
-   GLuint cpp = intel->intelScreen->cpp;
-   GLint size = width * pitch * cpp;
-
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   switch (format) {
-   case GL_RGB:
-   case GL_RGBA:
-   case GL_BGRA:
-      dest = intel->drawRegion->offset;
-
-      /* Planemask doesn't have full support in blits.
-       */
-      if (!ctx->Color.ColorMask[RCOMP] ||
-	  !ctx->Color.ColorMask[GCOMP] ||
-	  !ctx->Color.ColorMask[BCOMP] ||
-	  !ctx->Color.ColorMask[ACOMP]) {
-	 if (INTEL_DEBUG & DEBUG_PIXEL)
-	    fprintf(stderr, "%s: planemask\n", __FUNCTION__);
-	 return GL_FALSE;	
-      }
-
-      /* Can't do conversions on agp reads/draws. 
-       */
-      if ( !intelIsAgpMemory( intel, pixels, size ) ) {
-	 if (INTEL_DEBUG & DEBUG_PIXEL)
-	    fprintf(stderr, "%s: not agp memory\n", __FUNCTION__);
-	 return GL_FALSE;
-      }
-
-      if (!check_color(ctx, type, format, unpack, pixels, size, pitch)) {
-	 return GL_FALSE;
-      }
-      if (!check_color_per_fragment_ops(ctx)) {
-	 return GL_FALSE;
-      }
-
-      if (ctx->Pixel.ZoomX != 1.0F ||
-	  ctx->Pixel.ZoomY != -1.0F)
-	 return GL_FALSE;
-      break;
-
-   default:
-      return GL_FALSE;
-   }
-
-   if ( intelIsAgpMemory(intel, pixels, size) )
-   {
-      do_draw_pix( ctx, x, y, width, height, pitch, pixels, dest );
+   if (region->cpp == 4 &&
+       (type == GL_UNSIGNED_INT_8_8_8_8_REV ||
+        type == GL_UNSIGNED_BYTE) && format == GL_BGRA) {
       return GL_TRUE;
    }
-   else if (0)
-   {
-      /* Pixels is in regular memory -- get dma buffers and perform
-       * upload through them.  No point doing this for regular uploads
-       * but once we remove some of the restrictions above (colormask,
-       * pixelformat conversion, zoom?, etc), this could be a win.
-       */
+
+   if (region->cpp == 2 &&
+       type == GL_UNSIGNED_SHORT_5_6_5_REV && format == GL_BGR) {
+      return GL_TRUE;
    }
-   else
-      return GL_FALSE;
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      fprintf(stderr, "%s: bad format for blit (cpp %d, type %s format %s)\n",
+              __FUNCTION__, region->cpp,
+              _mesa_lookup_enum_by_nr(type), _mesa_lookup_enum_by_nr(format));
 
    return GL_FALSE;
 }
 
-static void
-intelDrawPixels( GLcontext *ctx,
-		 GLint x, GLint y, GLsizei width, GLsizei height,
-		 GLenum format, GLenum type,
-		 const struct gl_pixelstore_attrib *unpack,
-		 const GLvoid *pixels )
+
+void
+intelInitPixelFuncs(struct dd_function_table *functions)
 {
-   if (INTEL_DEBUG & DEBUG_PIXEL)
-      fprintf(stderr, "%s\n", __FUNCTION__);
-
-   if (!intelTryDrawPixels( ctx, x, y, width, height, format, type,
-                            unpack, pixels ))
-      _swrast_DrawPixels( ctx, x, y, width, height, format, type,
-			  unpack, pixels );
-}
-
-
-
-
-/**
- * Implement glCopyPixels for the front color buffer (or back buffer Pixmap)
- * for the color buffer.  Don't support zooming, pixel transfer, etc.
- * We do support copying from one window to another, ala glXMakeCurrentRead.
- */
-static void
-intelCopyPixels( GLcontext *ctx,
-		 GLint srcx, GLint srcy, GLsizei width, GLsizei height,
-		 GLint destx, GLint desty, GLenum type )
-{
-#if 0
-   const XMesaContext xmesa = XMESA_CONTEXT(ctx);
-   const SWcontext *swrast = SWRAST_CONTEXT( ctx );
-   XMesaDisplay *dpy = xmesa->xm_visual->display;
-   const XMesaDrawable drawBuffer = xmesa->xm_draw_buffer->buffer;
-   const XMesaDrawable readBuffer = xmesa->xm_read_buffer->buffer;
-   const XMesaGC gc = xmesa->xm_draw_buffer->gc;
-
-   ASSERT(dpy);
-   ASSERT(gc);
-
-   if (drawBuffer &&  /* buffer != 0 means it's a Window or Pixmap */
-       readBuffer &&
-       type == GL_COLOR &&
-       (swrast->_RasterMask & ~CLIP_BIT) == 0 && /* no blend, z-test, etc */
-       ctx->_ImageTransferState == 0 &&  /* no color tables, scale/bias, etc */
-       ctx->Pixel.ZoomX == 1.0 &&        /* no zooming */
-       ctx->Pixel.ZoomY == 1.0) {
-      /* Note: we don't do any special clipping work here.  We could,
-       * but X will do it for us.
-       */
-      srcy = FLIP(xmesa->xm_read_buffer, srcy) - height + 1;
-      desty = FLIP(xmesa->xm_draw_buffer, desty) - height + 1;
-      XCopyArea(dpy, readBuffer, drawBuffer, gc,
-                srcx, srcy, width, height, destx, desty);
-   }
-#else
-   _swrast_CopyPixels(ctx, srcx, srcy, width, height, destx, desty, type );
-#endif
-}
-
-
-
-
-void intelInitPixelFuncs( struct dd_function_table *functions )
-{
-   /* Pixel path fallbacks.
-    */
    functions->Accum = _swrast_Accum;
    functions->Bitmap = _swrast_Bitmap;
    functions->CopyPixels = intelCopyPixels;
-
-   if (!getenv("INTEL_NO_BLITS")) {
-      functions->ReadPixels = intelReadPixels;  
-      functions->DrawPixels = intelDrawPixels; 
-   }
-   else {
-      functions->ReadPixels = _swrast_ReadPixels;
-      functions->DrawPixels = _swrast_DrawPixels;
-   }
+   functions->ReadPixels = intelReadPixels;
+   functions->DrawPixels = intelDrawPixels;
 }

src/mesa/drivers/dri/i915/intel_pixel.h

@@ -0,0 +1,63 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_PIXEL_H
+#define INTEL_PIXEL_H
+
+#include "mtypes.h"
+
+void intelInitPixelFuncs(struct dd_function_table *functions);
+
+GLboolean intel_check_blit_fragment_ops(GLcontext * ctx);
+
+GLboolean intel_check_meta_tex_fragment_ops(GLcontext * ctx);
+
+GLboolean intel_check_blit_format(struct intel_region *region,
+                                  GLenum format, GLenum type);
+
+
+void intelReadPixels(GLcontext * ctx,
+                     GLint x, GLint y,
+                     GLsizei width, GLsizei height,
+                     GLenum format, GLenum type,
+                     const struct gl_pixelstore_attrib *pack,
+                     GLvoid * pixels);
+
+void intelDrawPixels(GLcontext * ctx,
+                     GLint x, GLint y,
+                     GLsizei width, GLsizei height,
+                     GLenum format,
+                     GLenum type,
+                     const struct gl_pixelstore_attrib *unpack,
+                     const GLvoid * pixels);
+
+void intelCopyPixels(GLcontext * ctx,
+                     GLint srcx, GLint srcy,
+                     GLsizei width, GLsizei height,
+                     GLint destx, GLint desty, GLenum type);
+
+#endif

src/mesa/drivers/dri/i915/intel_pixel_copy.c

@@ -0,0 +1,380 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "glheader.h"
+#include "enums.h"
+#include "image.h"
+#include "state.h"
+#include "mtypes.h"
+#include "macros.h"
+#include "swrast/swrast.h"
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_ioctl.h"
+#include "intel_batchbuffer.h"
+#include "intel_buffers.h"
+#include "intel_blit.h"
+#include "intel_regions.h"
+#include "intel_tris.h"
+#include "intel_pixel.h"
+
+#define FILE_DEBUG_FLAG DEBUG_PIXEL
+
+static struct intel_region *
+copypix_src_region(struct intel_context *intel, GLenum type)
+{
+   switch (type) {
+   case GL_COLOR:
+      return intel_readbuf_region(intel);
+   case GL_DEPTH:
+      /* Don't think this is really possible execpt at 16bpp, when we have no stencil.
+       */
+      if (intel->intelScreen->depth_region && 
+	  intel->intelScreen->depth_region->cpp == 2)
+         return intel->intelScreen->depth_region;
+   case GL_STENCIL:
+      /* Don't think this is really possible. 
+       */
+      break;
+   case GL_DEPTH_STENCIL_EXT:
+      /* Does it matter whether it is stencil/depth or depth/stencil?
+       */
+      return intel->intelScreen->depth_region;
+   default:
+      break;
+   }
+
+   return NULL;
+}
+
+
+/**
+ * Check if any fragment operations are in effect which might effect
+ * glCopyPixels.  Differs from intel_check_blit_fragment_ops in that
+ * we allow Scissor.
+ */
+static GLboolean
+intel_check_copypixel_blit_fragment_ops(GLcontext * ctx)
+{
+   if (ctx->NewState)
+      _mesa_update_state(ctx);
+
+   /* Could do logicop with the blitter: 
+    */
+   return !(ctx->_ImageTransferState ||
+            ctx->Color.AlphaEnabled ||
+            ctx->Depth.Test ||
+            ctx->Fog.Enabled ||
+            ctx->Stencil.Enabled ||
+            !ctx->Color.ColorMask[0] ||
+            !ctx->Color.ColorMask[1] ||
+            !ctx->Color.ColorMask[2] ||
+            !ctx->Color.ColorMask[3] ||
+            ctx->Color.ColorLogicOpEnabled ||
+            ctx->Texture._EnabledUnits ||
+	    ctx->FragmentProgram._Enabled);
+}
+
+/* Doesn't work for overlapping regions.  Could do a double copy or
+ * just fallback.
+ */
+static GLboolean
+do_texture_copypixels(GLcontext * ctx,
+                      GLint srcx, GLint srcy,
+                      GLsizei width, GLsizei height,
+                      GLint dstx, GLint dsty, GLenum type)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_region *dst = intel_drawbuf_region(intel);
+   struct intel_region *src = copypix_src_region(intel, type);
+   GLenum src_format;
+   GLenum src_type;
+
+   DBG("%s %d,%d %dx%d --> %d,%d\n", __FUNCTION__, 
+       srcx, srcy, width, height, dstx, dsty);
+
+   if (!src || !dst || type != GL_COLOR)
+      return GL_FALSE;
+
+   /* Can't handle overlapping regions.  Don't have sufficient control
+    * over rasterization to pull it off in-place.  Punt on these for
+    * now.
+    * 
+    * XXX: do a copy to a temporary. 
+    */
+   if (src->buffer == dst->buffer) {
+      drm_clip_rect_t srcbox;
+      drm_clip_rect_t dstbox;
+      drm_clip_rect_t tmp;
+
+      srcbox.x1 = srcx;
+      srcbox.y1 = srcy;
+      srcbox.x2 = srcx + width;
+      srcbox.y2 = srcy + height;
+
+      dstbox.x1 = dstx;
+      dstbox.y1 = dsty;
+      dstbox.x2 = dstx + width * ctx->Pixel.ZoomX;
+      dstbox.y2 = dsty + height * ctx->Pixel.ZoomY;
+
+      DBG("src %d,%d %d,%d\n", srcbox.x1, srcbox.y1, srcbox.x2, srcbox.y2);
+      DBG("dst %d,%d %d,%d (%dx%d) (%f,%f)\n", dstbox.x1, dstbox.y1, dstbox.x2, dstbox.y2,
+	  width, height, ctx->Pixel.ZoomX, ctx->Pixel.ZoomY);
+
+      if (intel_intersect_cliprects(&tmp, &srcbox, &dstbox)) {
+         DBG("%s: regions overlap\n", __FUNCTION__);
+         return GL_FALSE;
+      }
+   }
+
+   intelFlush(&intel->ctx);
+
+   intel->vtbl.install_meta_state(intel);
+
+   /* Is this true?  Also will need to turn depth testing on according
+    * to state:
+    */
+   intel->vtbl.meta_no_stencil_write(intel);
+   intel->vtbl.meta_no_depth_write(intel);
+
+   /* Set the 3d engine to draw into the destination region:
+    */
+   intel->vtbl.meta_draw_region(intel, dst, intel->intelScreen->depth_region);
+
+   intel->vtbl.meta_import_pixel_state(intel);
+
+   if (src->cpp == 2) {
+      src_format = GL_RGB;
+      src_type = GL_UNSIGNED_SHORT_5_6_5;
+   }
+   else {
+      src_format = GL_BGRA;
+      src_type = GL_UNSIGNED_BYTE;
+   }
+
+   /* Set the frontbuffer up as a large rectangular texture.
+    */
+   if (!intel->vtbl.meta_tex_rect_source(intel, src->buffer, 0,
+                                         src->pitch,
+                                         src->height, src_format, src_type)) {
+      intel->vtbl.leave_meta_state(intel);
+      return GL_FALSE;
+   }
+
+
+   intel->vtbl.meta_texture_blend_replace(intel);
+
+   LOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects) {
+      __DRIdrawablePrivate *dPriv = intel->driDrawable;
+
+
+      srcy = dPriv->h - srcy - height;  /* convert from gl to hardware coords */
+
+      srcx += dPriv->x;
+      srcy += dPriv->y;
+
+      /* Clip against the source region.  This is the only source
+       * clipping we do.  XXX: Just set the texcord wrap mode to clamp
+       * or similar.
+       *
+       */
+      if (0) {
+         GLint orig_x = srcx;
+         GLint orig_y = srcy;
+
+         if (!_mesa_clip_to_region(0, 0, src->pitch, src->height,
+                                   &srcx, &srcy, &width, &height))
+            goto out;
+
+         dstx += srcx - orig_x;
+         dsty += (srcy - orig_y) * ctx->Pixel.ZoomY;
+      }
+
+      /* Just use the regular cliprect mechanism...  Does this need to
+       * even hold the lock???
+       */
+      intel_meta_draw_quad(intel, 
+			   dstx, 
+			   dstx + width * ctx->Pixel.ZoomX, 
+			   dPriv->h - (dsty + height * ctx->Pixel.ZoomY), 
+			   dPriv->h - (dsty), 0,   /* XXX: what z value? */
+                           0x00ff00ff,
+                           srcx, srcx + width, srcy, srcy + height);
+
+    out:
+      intel->vtbl.leave_meta_state(intel);
+      intel_batchbuffer_flush(intel->batch);
+   }
+   UNLOCK_HARDWARE(intel);
+
+   DBG("%s: success\n", __FUNCTION__);
+   return GL_TRUE;
+}
+
+
+
+
+
+/**
+ * CopyPixels with the blitter.  Don't support zooming, pixel transfer, etc.
+ */
+static GLboolean
+do_blit_copypixels(GLcontext * ctx,
+                   GLint srcx, GLint srcy,
+                   GLsizei width, GLsizei height,
+                   GLint dstx, GLint dsty, GLenum type)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_region *dst = intel_drawbuf_region(intel);
+   struct intel_region *src = copypix_src_region(intel, type);
+
+   /* Copypixels can be more than a straight copy.  Ensure all the
+    * extra operations are disabled:
+    */
+   if (!intel_check_copypixel_blit_fragment_ops(ctx) ||
+       ctx->Pixel.ZoomX != 1.0F || ctx->Pixel.ZoomY != 1.0F)
+      return GL_FALSE;
+
+   if (!src || !dst)
+      return GL_FALSE;
+
+
+
+   intelFlush(&intel->ctx);
+
+   LOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects) {
+      __DRIdrawablePrivate *dPriv = intel->driDrawable;
+      drm_clip_rect_t *box = dPriv->pClipRects;
+      drm_clip_rect_t dest_rect;
+      GLint nbox = dPriv->numClipRects;
+      GLint delta_x = 0;
+      GLint delta_y = 0;
+      GLuint i;
+
+      /* Do scissoring in GL coordinates:
+       */
+      if (ctx->Scissor.Enabled)
+      {
+	 GLint x = ctx->Scissor.X;
+	 GLint y = ctx->Scissor.Y;
+	 GLuint w = ctx->Scissor.Width;
+	 GLuint h = ctx->Scissor.Height;
+	 GLint dx = dstx - srcx;
+         GLint dy = dsty - srcy;
+
+         if (!_mesa_clip_to_region(x, y, x+w, y+h, &dstx, &dsty, &width, &height))
+            goto out;
+	 
+         srcx = dstx - dx;
+         srcy = dsty - dy;
+      }
+
+      /* Convert from GL to hardware coordinates:
+       */
+      dsty = dPriv->h - dsty - height;  
+      srcy = dPriv->h - srcy - height;  
+      dstx += dPriv->x;
+      dsty += dPriv->y;
+      srcx += dPriv->x;
+      srcy += dPriv->y;
+
+      /* Clip against the source region.  This is the only source
+       * clipping we do.  Dst is clipped with cliprects below.
+       */
+      {
+         delta_x = srcx - dstx;
+         delta_y = srcy - dsty;
+
+         if (!_mesa_clip_to_region(0, 0, src->pitch, src->height,
+                                   &srcx, &srcy, &width, &height))
+            goto out;
+
+         dstx = srcx - delta_x;
+         dsty = srcy - delta_y;
+      }
+
+      dest_rect.x1 = dstx;
+      dest_rect.y1 = dsty;
+      dest_rect.x2 = dstx + width;
+      dest_rect.y2 = dsty + height;
+
+      /* Could do slightly more clipping: Eg, take the intersection of
+       * the existing set of cliprects and those cliprects translated
+       * by delta_x, delta_y:
+       * 
+       * This code will not overwrite other windows, but will
+       * introduce garbage when copying from obscured window regions.
+       */
+      for (i = 0; i < nbox; i++) {
+         drm_clip_rect_t rect;
+
+         if (!intel_intersect_cliprects(&rect, &dest_rect, &box[i]))
+            continue;
+
+
+         intelEmitCopyBlit(intel, dst->cpp, 
+			   src->pitch, src->buffer, 0, 
+			   dst->pitch, dst->buffer, 0, 
+			   rect.x1 + delta_x, rect.y1 + delta_y,       /* srcx, srcy */
+                           rect.x1, rect.y1,    /* dstx, dsty */
+                           rect.x2 - rect.x1, rect.y2 - rect.y1);
+      }
+
+    out:
+      intel_batchbuffer_flush(intel->batch);
+   }
+   UNLOCK_HARDWARE(intel);
+
+   DBG("%s: success\n", __FUNCTION__);
+   return GL_TRUE;
+}
+
+
+void
+intelCopyPixels(GLcontext * ctx,
+                GLint srcx, GLint srcy,
+                GLsizei width, GLsizei height,
+                GLint destx, GLint desty, GLenum type)
+{
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (do_blit_copypixels(ctx, srcx, srcy, width, height, destx, desty, type))
+      return;
+
+   if (do_texture_copypixels(ctx, srcx, srcy, width, height, destx, desty, type))
+      return;
+
+   DBG("fallback to _swrast_CopyPixels\n");
+
+   _swrast_CopyPixels(ctx, srcx, srcy, width, height, destx, desty, type);
+}

src/mesa/drivers/dri/i915/intel_pixel_draw.c

@@ -0,0 +1,365 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portionsalloc
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "glheader.h"
+#include "enums.h"
+#include "image.h"
+#include "mtypes.h"
+#include "macros.h"
+#include "bufferobj.h"
+#include "swrast/swrast.h"
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_ioctl.h"
+#include "intel_batchbuffer.h"
+#include "intel_blit.h"
+#include "intel_buffers.h"
+#include "intel_regions.h"
+#include "intel_pixel.h"
+#include "intel_buffer_objects.h"
+#include "intel_tris.h"
+
+
+
+static GLboolean
+do_texture_drawpixels(GLcontext * ctx,
+                      GLint x, GLint y,
+                      GLsizei width, GLsizei height,
+                      GLenum format, GLenum type,
+                      const struct gl_pixelstore_attrib *unpack,
+                      const GLvoid * pixels)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_region *dst = intel_drawbuf_region(intel);
+   struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
+   GLuint rowLength = unpack->RowLength ? unpack->RowLength : width;
+   GLuint src_offset;
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   intelFlush(&intel->ctx);
+   intel->vtbl.render_start(intel);
+   intel->vtbl.emit_state(intel);
+
+   if (!dst)
+      return GL_FALSE;
+
+   if (src) {
+      if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
+                                     format, type, pixels)) {
+         _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
+         return GL_TRUE;
+      }
+   }
+   else {
+      /* PBO only for now:
+       */
+/*       _mesa_printf("%s - not PBO\n", __FUNCTION__); */
+      return GL_FALSE;
+   }
+
+   /* There are a couple of things we can't do yet, one of which is
+    * set the correct state for pixel operations when GL texturing is
+    * enabled.  That's a pretty rare state and probably not worth the
+    * effort.  A completely device-independent version of this may do
+    * more.
+    *
+    * Similarly, we make no attempt to merge metaops processing with
+    * an enabled fragment program, though it would certainly be
+    * possible.
+    */
+   if (!intel_check_meta_tex_fragment_ops(ctx)) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - bad GL fragment state for metaops texture\n",
+                      __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   intel->vtbl.install_meta_state(intel);
+
+
+   /* Is this true?  Also will need to turn depth testing on according
+    * to state:
+    */
+   intel->vtbl.meta_no_stencil_write(intel);
+   intel->vtbl.meta_no_depth_write(intel);
+
+   /* Set the 3d engine to draw into the destination region:
+    */
+   intel->vtbl.meta_draw_region(intel, dst, intel->intelScreen->depth_region);
+
+   intel->vtbl.meta_import_pixel_state(intel);
+
+   src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
+                                             format, type, 0, 0, 0);
+
+
+   /* Setup the pbo up as a rectangular texture, if possible.
+    *
+    * TODO: This is almost always possible if the i915 fragment
+    * program is adjusted to correctly swizzle the sampled colors.
+    * The major exception is any 24bit texture, like RGB888, for which
+    * there is no hardware support.  
+    */
+   if (!intel->vtbl.meta_tex_rect_source(intel, src->buffer, src_offset,
+                                         rowLength, height, format, type)) {
+      intel->vtbl.leave_meta_state(intel);
+      return GL_FALSE;
+   }
+
+   intel->vtbl.meta_texture_blend_replace(intel);
+
+
+   LOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects) {
+      __DRIdrawablePrivate *dPriv = intel->driDrawable;
+      GLint srcx, srcy;
+      GLint dstx, dsty;
+
+      dstx = x;
+      dsty = dPriv->h - (y + height);
+
+      srcx = 0;                 /* skiprows/pixels already done */
+      srcy = 0;
+
+      if (0) {
+         const GLint orig_x = dstx;
+         const GLint orig_y = dsty;
+
+         if (!_mesa_clip_to_region(0, 0, dst->pitch, dst->height,
+                                   &dstx, &dsty, &width, &height))
+            goto out;
+
+         srcx += dstx - orig_x;
+         srcy += dsty - orig_y;
+      }
+
+
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("draw %d,%d %dx%d\n", dstx, dsty, width, height);
+
+      /* Must use the regular cliprect mechanism in order to get the
+       * drawing origin set correctly.  Otherwise scissor state is in
+       * incorrect coordinate space.  Does this even need to hold the
+       * lock???
+       */
+      intel_meta_draw_quad(intel,
+                           dstx, dstx + width * ctx->Pixel.ZoomX,
+                           dPriv->h - (y + height * ctx->Pixel.ZoomY),
+                           dPriv->h - (y),
+                           -ctx->Current.RasterPos[2] * .5,
+                           0x00ff00ff,
+                           srcx, srcx + width, srcy + height, srcy);
+    out:
+      intel->vtbl.leave_meta_state(intel);
+      intel_batchbuffer_flush(intel->batch);
+   }
+   UNLOCK_HARDWARE(intel);
+   return GL_TRUE;
+}
+
+
+
+
+
+/* Pros:  
+ *   - no waiting for idle before updating framebuffer.
+ *   
+ * Cons:
+ *   - if upload is by memcpy, this may actually be slower than fallback path.
+ *   - uploads the whole image even if destination is clipped
+ *   
+ * Need to benchmark.
+ *
+ * Given the questions about performance, implement for pbo's only.
+ * This path is definitely a win if the pbo is already in agp.  If it
+ * turns out otherwise, we can add the code necessary to upload client
+ * data to agp space before performing the blit.  (Though it may turn
+ * out to be better/simpler just to use the texture engine).
+ */
+static GLboolean
+do_blit_drawpixels(GLcontext * ctx,
+                   GLint x, GLint y,
+                   GLsizei width, GLsizei height,
+                   GLenum format, GLenum type,
+                   const struct gl_pixelstore_attrib *unpack,
+                   const GLvoid * pixels)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_region *dest = intel_drawbuf_region(intel);
+   struct intel_buffer_object *src = intel_buffer_object(unpack->BufferObj);
+   GLuint src_offset;
+   GLuint rowLength;
+   struct _DriFenceObject *fence = NULL;
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      _mesa_printf("%s\n", __FUNCTION__);
+
+
+   if (!dest) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - no dest\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   if (src) {
+      /* This validation should be done by core mesa:
+       */
+      if (!_mesa_validate_pbo_access(2, unpack, width, height, 1,
+                                     format, type, pixels)) {
+         _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
+         return GL_TRUE;
+      }
+   }
+   else {
+      /* PBO only for now:
+       */
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - not PBO\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   if (!intel_check_blit_format(dest, format, type)) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - bad format for blit\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   if (!intel_check_meta_tex_fragment_ops(ctx)) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - bad GL fragment state for meta tex\n",
+                      __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   if (ctx->Pixel.ZoomX != 1.0F) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - bad PixelZoomX for blit\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+
+   if (unpack->RowLength > 0)
+      rowLength = unpack->RowLength;
+   else
+      rowLength = width;
+
+   if (ctx->Pixel.ZoomY == -1.0F) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
+      return GL_FALSE;          /* later */
+      y -= height;
+   }
+   else if (ctx->Pixel.ZoomY == 1.0F) {
+      rowLength = -rowLength;
+   }
+   else {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - bad PixelZoomY for blit\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   src_offset = (GLuint) _mesa_image_address(2, unpack, pixels, width, height,
+                                             format, type, 0, 0, 0);
+
+   intelFlush(&intel->ctx);
+   LOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects) {
+      __DRIdrawablePrivate *dPriv = intel->driDrawable;
+      int nbox = dPriv->numClipRects;
+      drm_clip_rect_t *box = dPriv->pClipRects;
+      drm_clip_rect_t rect;
+      drm_clip_rect_t dest_rect;
+      struct _DriBufferObject *src_buffer =
+         intel_bufferobj_buffer(intel, src, INTEL_READ);
+      int i;
+
+      dest_rect.x1 = dPriv->x + x;
+      dest_rect.y1 = dPriv->y + dPriv->h - (y + height);
+      dest_rect.x2 = dest_rect.x1 + width;
+      dest_rect.y2 = dest_rect.y1 + height;
+
+      for (i = 0; i < nbox; i++) {
+         if (!intel_intersect_cliprects(&rect, &dest_rect, &box[i]))
+            continue;
+
+         intelEmitCopyBlit(intel,
+                           dest->cpp,
+                           rowLength,
+                           src_buffer, src_offset,
+                           dest->pitch,
+                           dest->buffer, 0,
+                           rect.x1 - dest_rect.x1,
+                           rect.y2 - dest_rect.y2,
+                           rect.x1,
+                           rect.y1, rect.x2 - rect.x1, rect.y2 - rect.y1);
+      }
+      fence = intel_batchbuffer_flush(intel->batch);
+      driFenceReference(fence);
+   }
+   UNLOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects)
+      driFenceFinish(fence, DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, GL_FALSE);
+
+   driFenceUnReference(fence);
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      _mesa_printf("%s - DONE\n", __FUNCTION__);
+
+   return GL_TRUE;
+}
+
+
+
+void
+intelDrawPixels(GLcontext * ctx,
+                GLint x, GLint y,
+                GLsizei width, GLsizei height,
+                GLenum format,
+                GLenum type,
+                const struct gl_pixelstore_attrib *unpack,
+                const GLvoid * pixels)
+{
+   if (do_blit_drawpixels(ctx, x, y, width, height, format, type,
+                          unpack, pixels))
+      return;
+
+   if (do_texture_drawpixels(ctx, x, y, width, height, format, type,
+                             unpack, pixels))
+      return;
+
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      _mesa_printf("%s: fallback to swrast\n", __FUNCTION__);
+
+   _swrast_DrawPixels(ctx, x, y, width, height, format, type, unpack, pixels);
+}

src/mesa/drivers/dri/i915/intel_pixel_read.c

@@ -0,0 +1,317 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "glheader.h"
+#include "enums.h"
+#include "mtypes.h"
+#include "macros.h"
+#include "image.h"
+#include "bufferobj.h"
+#include "swrast/swrast.h"
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_ioctl.h"
+#include "intel_batchbuffer.h"
+#include "intel_blit.h"
+#include "intel_buffers.h"
+#include "intel_regions.h"
+#include "intel_pixel.h"
+#include "intel_buffer_objects.h"
+
+/* For many applications, the new ability to pull the source buffers
+ * back out of the GTT and then do the packing/conversion operations
+ * in software will be as much of an improvement as trying to get the
+ * blitter and/or texture engine to do the work. 
+ *
+ * This step is gated on private backbuffers.
+ * 
+ * Obviously the frontbuffer can't be pulled back, so that is either
+ * an argument for blit/texture readpixels, or for blitting to a
+ * temporary and then pulling that back.
+ *
+ * When the destination is a pbo, however, it's not clear if it is
+ * ever going to be pulled to main memory (though the access param
+ * will be a good hint).  So it sounds like we do want to be able to
+ * choose between blit/texture implementation on the gpu and pullback
+ * and cpu-based copying.
+ *
+ * Unless you can magically turn client memory into a PBO for the
+ * duration of this call, there will be a cpu-based copying step in
+ * any case.
+ */
+
+
+static GLboolean
+do_texture_readpixels(GLcontext * ctx,
+                      GLint x, GLint y, GLsizei width, GLsizei height,
+                      GLenum format, GLenum type,
+                      const struct gl_pixelstore_attrib *pack,
+                      struct intel_region *dest_region)
+{
+#if 0
+   struct intel_context *intel = intel_context(ctx);
+   intelScreenPrivate *screen = intel->intelScreen;
+   GLint pitch = pack->RowLength ? pack->RowLength : width;
+   __DRIdrawablePrivate *dPriv = intel->driDrawable;
+   int textureFormat;
+   GLenum glTextureFormat;
+   int destFormat, depthFormat, destPitch;
+   drm_clip_rect_t tmp;
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+
+   if (ctx->_ImageTransferState ||
+       pack->SwapBytes || pack->LsbFirst || !pack->Invert) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         fprintf(stderr, "%s: check_color failed\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   intel->vtbl.meta_texrect_source(intel, intel_readbuf_region(intel));
+
+   if (!intel->vtbl.meta_render_dest(intel, dest_region, type, format)) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         fprintf(stderr, "%s: couldn't set dest %s/%s\n",
+                 __FUNCTION__,
+                 _mesa_lookup_enum_by_nr(type),
+                 _mesa_lookup_enum_by_nr(format));
+      return GL_FALSE;
+   }
+
+   LOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects) {
+      intel->vtbl.install_meta_state(intel);
+      intel->vtbl.meta_no_depth_write(intel);
+      intel->vtbl.meta_no_stencil_write(intel);
+
+      if (!driClipRectToFramebuffer(ctx->ReadBuffer, &x, &y, &width, &height)) {
+         UNLOCK_HARDWARE(intel);
+         SET_STATE(i830, state);
+         if (INTEL_DEBUG & DEBUG_PIXEL)
+            fprintf(stderr, "%s: cliprect failed\n", __FUNCTION__);
+         return GL_TRUE;
+      }
+
+      y = dPriv->h - y - height;
+      x += dPriv->x;
+      y += dPriv->y;
+
+
+      /* Set the frontbuffer up as a large rectangular texture.
+       */
+      intel->vtbl.meta_tex_rect_source(intel, src_region, textureFormat);
+
+
+      intel->vtbl.meta_texture_blend_replace(i830, glTextureFormat);
+
+
+      /* Set the 3d engine to draw into the destination region:
+       */
+
+      intel->vtbl.meta_draw_region(intel, dest_region);
+      intel->vtbl.meta_draw_format(intel, destFormat, depthFormat);     /* ?? */
+
+
+      /* Draw a single quad, no cliprects:
+       */
+      intel->vtbl.meta_disable_cliprects(intel);
+
+      intel->vtbl.draw_quad(intel,
+                            0, width, 0, height,
+                            0x00ff00ff, x, x + width, y, y + height);
+
+      intel->vtbl.leave_meta_state(intel);
+   }
+   UNLOCK_HARDWARE(intel);
+
+   intel_region_wait_fence(ctx, dest_region);   /* required by GL */
+   return GL_TRUE;
+#endif
+
+   return GL_FALSE;
+}
+
+
+
+
+static GLboolean
+do_blit_readpixels(GLcontext * ctx,
+                   GLint x, GLint y, GLsizei width, GLsizei height,
+                   GLenum format, GLenum type,
+                   const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_region *src = intel_readbuf_region(intel);
+   struct intel_buffer_object *dst = intel_buffer_object(pack->BufferObj);
+   GLuint dst_offset;
+   GLuint rowLength;
+   struct _DriFenceObject *fence = NULL;
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      _mesa_printf("%s\n", __FUNCTION__);
+
+   if (!src)
+      return GL_FALSE;
+
+   if (dst) {
+      /* XXX This validation should be done by core mesa:
+       */
+      if (!_mesa_validate_pbo_access(2, pack, width, height, 1,
+                                     format, type, pixels)) {
+         _mesa_error(ctx, GL_INVALID_OPERATION, "glDrawPixels");
+         return GL_TRUE;
+      }
+   }
+   else {
+      /* PBO only for now:
+       */
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - not PBO\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+
+   if (ctx->_ImageTransferState ||
+       !intel_check_blit_format(src, format, type)) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s - bad format for blit\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   if (pack->Alignment != 1 || pack->SwapBytes || pack->LsbFirst) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s: bad packing params\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   if (pack->RowLength > 0)
+      rowLength = pack->RowLength;
+   else
+      rowLength = width;
+
+   if (pack->Invert) {
+      if (INTEL_DEBUG & DEBUG_PIXEL)
+         _mesa_printf("%s: MESA_PACK_INVERT not done yet\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+   else {
+      rowLength = -rowLength;
+   }
+
+   /* XXX 64-bit cast? */
+   dst_offset = (GLuint) _mesa_image_address(2, pack, pixels, width, height,
+                                             format, type, 0, 0, 0);
+
+
+   /* Although the blits go on the command buffer, need to do this and
+    * fire with lock held to guarentee cliprects are correct.
+    */
+   intelFlush(&intel->ctx);
+   LOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects) {
+      GLboolean all = (width * height * src->cpp == dst->Base.Size &&
+                       x == 0 && dst_offset == 0);
+
+      struct _DriBufferObject *dst_buffer =
+         intel_bufferobj_buffer(intel, dst, all ? INTEL_WRITE_FULL :
+                                INTEL_WRITE_PART);
+      __DRIdrawablePrivate *dPriv = intel->driDrawable;
+      int nbox = dPriv->numClipRects;
+      drm_clip_rect_t *box = dPriv->pClipRects;
+      drm_clip_rect_t rect;
+      drm_clip_rect_t src_rect;
+      int i;
+
+      src_rect.x1 = dPriv->x + x;
+      src_rect.y1 = dPriv->y + dPriv->h - (y + height);
+      src_rect.x2 = src_rect.x1 + width;
+      src_rect.y2 = src_rect.y1 + height;
+
+
+
+      for (i = 0; i < nbox; i++) {
+         if (!intel_intersect_cliprects(&rect, &src_rect, &box[i]))
+            continue;
+
+         intelEmitCopyBlit(intel,
+                           src->cpp,
+                           src->pitch, src->buffer, 0,
+                           rowLength,
+                           dst_buffer, dst_offset,
+                           rect.x1,
+                           rect.y1,
+                           rect.x1 - src_rect.x1,
+                           rect.y2 - src_rect.y2,
+                           rect.x2 - rect.x1, rect.y2 - rect.y1);
+      }
+
+      fence = intel_batchbuffer_flush(intel->batch);
+      driFenceReference(fence);
+
+   }
+   UNLOCK_HARDWARE(intel);
+
+   if (intel->driDrawable->numClipRects)
+      driFenceFinish(fence, DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, 
+		     GL_FALSE);
+
+   driFenceUnReference(fence);
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      _mesa_printf("%s - DONE\n", __FUNCTION__);
+
+   return GL_TRUE;
+}
+
+void
+intelReadPixels(GLcontext * ctx,
+                GLint x, GLint y, GLsizei width, GLsizei height,
+                GLenum format, GLenum type,
+                const struct gl_pixelstore_attrib *pack, GLvoid * pixels)
+{
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   intelFlush(ctx);
+
+   if (do_blit_readpixels
+       (ctx, x, y, width, height, format, type, pack, pixels))
+      return;
+
+   if (do_texture_readpixels
+       (ctx, x, y, width, height, format, type, pack, pixels))
+      return;
+
+   if (INTEL_DEBUG & DEBUG_PIXEL)
+      _mesa_printf("%s: fallback to swrast\n", __FUNCTION__);
+
+   _swrast_ReadPixels(ctx, x, y, width, height, format, type, pack, pixels);
+}

src/mesa/drivers/dri/i915/intel_regions.c

@@ -0,0 +1,458 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Provide additional functionality on top of bufmgr buffers:
+ *   - 2d semantics and blit operations
+ *   - refcounting of buffers for multiple images in a buffer.
+ *   - refcounting of buffer mappings.
+ *   - some logic for moving the buffers to the best memory pools for
+ *     given operations.
+ *
+ * Most of this is to make it easier to implement the fixed-layout
+ * mipmap tree required by intel hardware in the face of GL's
+ * programming interface where each image can be specifed in random
+ * order and it isn't clear what layout the tree should have until the
+ * last moment.
+ */
+
+#include "intel_context.h"
+#include "intel_regions.h"
+#include "intel_blit.h"
+#include "intel_buffer_objects.h"
+#include "dri_bufmgr.h"
+#include "intel_batchbuffer.h"
+
+#define FILE_DEBUG_FLAG DEBUG_REGION
+
+void
+intel_region_idle(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   DBG("%s\n", __FUNCTION__);
+   if (region && region->buffer)
+      driBOWaitIdle(region->buffer, GL_FALSE);
+}
+
+/* XXX: Thread safety?
+ */
+GLubyte *
+intel_region_map(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   DBG("%s\n", __FUNCTION__);
+   if (!region->map_refcount++) {
+      if (region->pbo)
+         intel_region_cow(intelScreen, region);
+
+      region->map = driBOMap(region->buffer,
+                             DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0);
+   }
+
+   return region->map;
+}
+
+void
+intel_region_unmap(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   DBG("%s\n", __FUNCTION__);
+   if (!--region->map_refcount) {
+      driBOUnmap(region->buffer);
+      region->map = NULL;
+   }
+}
+
+struct intel_region *
+intel_region_alloc(intelScreenPrivate *intelScreen,
+                   GLuint cpp, GLuint pitch, GLuint height)
+{
+   struct intel_region *region = calloc(sizeof(*region), 1);
+
+   DBG("%s\n", __FUNCTION__);
+
+   region->cpp = cpp;
+   region->pitch = pitch;
+   region->height = height;     /* needed? */
+   region->refcount = 1;
+
+   driGenBuffers(intelScreen->regionPool,
+                 "region", 1, &region->buffer, 64, 0, 0);
+   driBOData(region->buffer, pitch * cpp * height, NULL, 0);
+   return region;
+}
+
+void
+intel_region_reference(struct intel_region **dst, struct intel_region *src)
+{
+   assert(*dst == NULL);
+   if (src) {
+      src->refcount++;
+      *dst = src;
+   }
+}
+
+void
+intel_region_release(intelScreenPrivate *intelScreen,
+                     struct intel_region **region)
+{
+   if (!*region)
+      return;
+
+   DBG("%s %d\n", __FUNCTION__, (*region)->refcount - 1);
+
+   ASSERT((*region)->refcount > 0);
+   (*region)->refcount--;
+
+   if ((*region)->refcount == 0) {
+      assert((*region)->map_refcount == 0);
+
+      if ((*region)->pbo)
+	 (*region)->pbo->region = NULL;
+      (*region)->pbo = NULL;
+      driBOUnReference((*region)->buffer);
+      free(*region);
+   }
+   *region = NULL;
+}
+
+
+struct intel_region *
+intel_region_create_static(intelScreenPrivate *intelScreen,
+                           GLuint mem_type,
+                           GLuint offset,
+                           void *virtual,
+                           GLuint cpp, GLuint pitch, GLuint height)
+{
+   struct intel_region *region = calloc(sizeof(*region), 1);
+   DBG("%s\n", __FUNCTION__);
+
+   region->cpp = cpp;
+   region->pitch = pitch;
+   region->height = height;     /* needed? */
+   region->refcount = 1;
+
+   /*
+    * We use a "shared" buffer type to indicate buffers created and
+    * shared by others.
+    */
+
+   driGenBuffers(intelScreen->staticPool, "static region", 1,
+                 &region->buffer, 64,
+                 DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_NO_EVICT |
+                 DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0);
+   driBOSetStatic(region->buffer, offset, pitch * cpp * height, virtual, 0);
+
+   return region;
+}
+
+
+
+void
+intel_region_update_static(intelScreenPrivate *intelScreen,
+			   struct intel_region *region,
+                           GLuint mem_type,
+                           GLuint offset,
+                           void *virtual,
+                           GLuint cpp, GLuint pitch, GLuint height)
+{
+   DBG("%s\n", __FUNCTION__);
+
+   region->cpp = cpp;
+   region->pitch = pitch;
+   region->height = height;     /* needed? */
+
+   /*
+    * We use a "shared" buffer type to indicate buffers created and
+    * shared by others.
+    */
+
+   driDeleteBuffers(1, &region->buffer);
+   driGenBuffers(intelScreen->staticPool, "static region", 1,
+                 &region->buffer, 64,
+                 DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_NO_EVICT |
+                 DRM_BO_FLAG_READ | DRM_BO_FLAG_WRITE, 0);
+   driBOSetStatic(region->buffer, offset, pitch * cpp * height, virtual, 0);
+
+}
+
+
+
+/*
+ * XXX Move this into core Mesa?
+ */
+static void
+_mesa_copy_rect(GLubyte * dst,
+                GLuint cpp,
+                GLuint dst_pitch,
+                GLuint dst_x,
+                GLuint dst_y,
+                GLuint width,
+                GLuint height,
+                GLubyte * src, GLuint src_pitch, GLuint src_x, GLuint src_y)
+{
+   GLuint i;
+
+   dst_pitch *= cpp;
+   src_pitch *= cpp;
+   dst += dst_x * cpp;
+   src += src_x * cpp;
+   dst += dst_y * dst_pitch;
+   src += src_y * dst_pitch;
+   width *= cpp;
+
+   if (width == dst_pitch && width == src_pitch)
+      memcpy(dst, src, height * width);
+   else {
+      for (i = 0; i < height; i++) {
+         memcpy(dst, src, width);
+         dst += dst_pitch;
+         src += src_pitch;
+      }
+   }
+}
+
+
+/* Upload data to a rectangular sub-region.  Lots of choices how to do this:
+ *
+ * - memcpy by span to current destination
+ * - upload data as new buffer and blit
+ *
+ * Currently always memcpy.
+ */
+void
+intel_region_data(intelScreenPrivate *intelScreen,
+                  struct intel_region *dst,
+                  GLuint dst_offset,
+                  GLuint dstx, GLuint dsty,
+                  void *src, GLuint src_pitch,
+                  GLuint srcx, GLuint srcy, GLuint width, GLuint height)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intel == NULL)
+      return;
+
+   if (dst->pbo) {
+      if (dstx == 0 &&
+          dsty == 0 && width == dst->pitch && height == dst->height)
+         intel_region_release_pbo(intelScreen, dst);
+      else
+         intel_region_cow(intelScreen, dst);
+   }
+
+
+   LOCK_HARDWARE(intel);
+
+   _mesa_copy_rect(intel_region_map(intelScreen, dst) + dst_offset,
+                   dst->cpp,
+                   dst->pitch,
+                   dstx, dsty, width, height, src, src_pitch, srcx, srcy);
+
+   intel_region_unmap(intelScreen, dst);
+
+   UNLOCK_HARDWARE(intel);
+
+}
+
+/* Copy rectangular sub-regions. Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+void
+intel_region_copy(intelScreenPrivate *intelScreen,
+                  struct intel_region *dst,
+                  GLuint dst_offset,
+                  GLuint dstx, GLuint dsty,
+                  struct intel_region *src,
+                  GLuint src_offset,
+                  GLuint srcx, GLuint srcy, GLuint width, GLuint height)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intel == NULL)
+      return;
+
+   if (dst->pbo) {
+      if (dstx == 0 &&
+          dsty == 0 && width == dst->pitch && height == dst->height)
+         intel_region_release_pbo(intelScreen, dst);
+      else
+         intel_region_cow(intelScreen, dst);
+   }
+
+   assert(src->cpp == dst->cpp);
+
+   intelEmitCopyBlit(intel,
+                     dst->cpp,
+                     src->pitch, src->buffer, src_offset,
+                     dst->pitch, dst->buffer, dst_offset,
+                     srcx, srcy, dstx, dsty, width, height);
+}
+
+/* Fill a rectangular sub-region.  Need better logic about when to
+ * push buffers into AGP - will currently do so whenever possible.
+ */
+void
+intel_region_fill(intelScreenPrivate *intelScreen,
+                  struct intel_region *dst,
+                  GLuint dst_offset,
+                  GLuint dstx, GLuint dsty,
+                  GLuint width, GLuint height, GLuint color)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intel == NULL)
+      return;   
+
+   if (dst->pbo) {
+      if (dstx == 0 &&
+          dsty == 0 && width == dst->pitch && height == dst->height)
+         intel_region_release_pbo(intelScreen, dst);
+      else
+         intel_region_cow(intelScreen, dst);
+   }
+
+   intelEmitFillBlit(intel,
+                     dst->cpp,
+                     dst->pitch, dst->buffer, dst_offset,
+                     dstx, dsty, width, height, color);
+}
+
+/* Attach to a pbo, discarding our data.  Effectively zero-copy upload
+ * the pbo's data.
+ */
+void
+intel_region_attach_pbo(intelScreenPrivate *intelScreen,
+                        struct intel_region *region,
+                        struct intel_buffer_object *pbo)
+{
+   if (region->pbo == pbo)
+      return;
+
+   /* If there is already a pbo attached, break the cow tie now.
+    * Don't call intel_region_release_pbo() as that would
+    * unnecessarily allocate a new buffer we would have to immediately
+    * discard.
+    */
+   if (region->pbo) {
+      region->pbo->region = NULL;
+      region->pbo = NULL;
+   }
+
+   if (region->buffer) {
+      driDeleteBuffers(1, &region->buffer);
+      region->buffer = NULL;
+   }
+
+   region->pbo = pbo;
+   region->pbo->region = region;
+   region->buffer = driBOReference(pbo->buffer);
+}
+
+
+/* Break the COW tie to the pbo.  The pbo gets to keep the data.
+ */
+void
+intel_region_release_pbo(intelScreenPrivate *intelScreen,
+                         struct intel_region *region)
+{
+   assert(region->buffer == region->pbo->buffer);
+   region->pbo->region = NULL;
+   region->pbo = NULL;
+   driBOUnReference(region->buffer);
+   region->buffer = NULL;
+
+   driGenBuffers(intelScreen->regionPool,
+                 "region", 1, &region->buffer, 64, 0, 0);
+   driBOData(region->buffer,
+             region->cpp * region->pitch * region->height, NULL, 0);
+}
+
+/* Break the COW tie to the pbo.  Both the pbo and the region end up
+ * with a copy of the data.
+ */
+void
+intel_region_cow(intelScreenPrivate *intelScreen, struct intel_region *region)
+{
+   struct intel_context *intel = intelScreenContext(intelScreen);
+   struct intel_buffer_object *pbo = region->pbo;
+
+   if (intel == NULL)
+      return;
+
+   intel_region_release_pbo(intelScreen, region);
+
+   assert(region->cpp * region->pitch * region->height == pbo->Base.Size);
+
+   DBG("%s (%d bytes)\n", __FUNCTION__, pbo->Base.Size);
+
+   /* Now blit from the texture buffer to the new buffer: 
+    */
+
+   intel_batchbuffer_flush(intel->batch);
+
+   if (!intel->locked) {
+      LOCK_HARDWARE(intel);
+      intelEmitCopyBlit(intel,
+			region->cpp,
+			region->pitch,
+			region->buffer, 0,
+			region->pitch,
+			pbo->buffer, 0,
+			0, 0, 0, 0, region->pitch, region->height);
+      
+      intel_batchbuffer_flush(intel->batch);
+      UNLOCK_HARDWARE(intel);
+   }
+   else {
+      intelEmitCopyBlit(intel,
+			region->cpp,
+			region->pitch,
+			region->buffer, 0,
+			region->pitch,
+			pbo->buffer, 0,
+			0, 0, 0, 0, region->pitch, region->height);
+      
+      intel_batchbuffer_flush(intel->batch);
+   }
+}
+
+struct _DriBufferObject *
+intel_region_buffer(intelScreenPrivate *intelScreen,
+                    struct intel_region *region, GLuint flag)
+{
+   if (region->pbo) {
+      if (flag == INTEL_WRITE_PART)
+         intel_region_cow(intelScreen, region);
+      else if (flag == INTEL_WRITE_FULL)
+         intel_region_release_pbo(intelScreen, region);
+   }
+
+   return region->buffer;
+}

src/mesa/drivers/dri/i915/intel_regions.h

@@ -0,0 +1,142 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef INTEL_REGIONS_H
+#define INTEL_REGIONS_H
+
+#include "mtypes.h"
+#include "intel_screen.h"
+
+struct intel_context;
+struct intel_buffer_object;
+
+/**
+ * A layer on top of the bufmgr buffers that adds a few useful things:
+ *
+ * - Refcounting for local buffer references.
+ * - Refcounting for buffer maps
+ * - Buffer dimensions - pitch and height.
+ * - Blitter commands for copying 2D regions between buffers. (really???)
+ */
+struct intel_region
+{
+   struct _DriBufferObject *buffer;   /**< buffer manager's buffer ID */
+   GLuint refcount; /**< Reference count for region */
+   GLuint cpp;      /**< bytes per pixel */
+   GLuint pitch;    /**< in pixels */
+   GLuint height;   /**< in pixels */
+   GLubyte *map;    /**< only non-NULL when region is actually mapped */
+   GLuint map_refcount;  /**< Reference count for mapping */
+
+   GLuint draw_offset; /**< Offset of drawing address within the region */
+
+   struct intel_buffer_object *pbo;     /* zero-copy uploads */
+};
+
+
+/* Allocate a refcounted region.  Pointers to regions should only be
+ * copied by calling intel_reference_region().
+ */
+struct intel_region *intel_region_alloc(intelScreenPrivate *intelScreen,
+                                        GLuint cpp,
+                                        GLuint pitch, GLuint height);
+
+void intel_region_reference(struct intel_region **dst,
+                            struct intel_region *src);
+
+void intel_region_release(intelScreenPrivate *intelScreen,
+                          struct intel_region **ib);
+
+extern struct intel_region 
+*intel_region_create_static(intelScreenPrivate *intelScreen,
+			    GLuint mem_type,
+			    GLuint offset,
+			    void *virtual,
+			    GLuint cpp,
+			    GLuint pitch, GLuint height);
+extern void 
+intel_region_update_static(intelScreenPrivate *intelScreen,
+			   struct intel_region *region,
+			   GLuint mem_type,
+			   GLuint offset,
+			   void *virtual,
+			   GLuint cpp, GLuint pitch, GLuint height);
+
+
+void intel_region_idle(intelScreenPrivate *intelScreen,
+		       struct intel_region *ib);
+
+/* Map/unmap regions.  This is refcounted also: 
+ */
+GLubyte *intel_region_map(intelScreenPrivate *intelScreen,
+                          struct intel_region *ib);
+
+void intel_region_unmap(intelScreenPrivate *intelScreen, struct intel_region *ib);
+
+
+/* Upload data to a rectangular sub-region
+ */
+void intel_region_data(intelScreenPrivate *intelScreen,
+                       struct intel_region *dest,
+                       GLuint dest_offset,
+                       GLuint destx, GLuint desty,
+                       void *src, GLuint src_stride,
+                       GLuint srcx, GLuint srcy, GLuint width, GLuint height);
+
+/* Copy rectangular sub-regions
+ */
+void intel_region_copy(intelScreenPrivate *intelScreen,
+                       struct intel_region *dest,
+                       GLuint dest_offset,
+                       GLuint destx, GLuint desty,
+                       struct intel_region *src,
+                       GLuint src_offset,
+                       GLuint srcx, GLuint srcy, GLuint width, GLuint height);
+
+/* Fill a rectangular sub-region
+ */
+void intel_region_fill(intelScreenPrivate *intelScreen,
+                       struct intel_region *dest,
+                       GLuint dest_offset,
+                       GLuint destx, GLuint desty,
+                       GLuint width, GLuint height, GLuint color);
+
+/* Helpers for zerocopy uploads, particularly texture image uploads:
+ */
+void intel_region_attach_pbo(intelScreenPrivate *intelScreen,
+                             struct intel_region *region,
+                             struct intel_buffer_object *pbo);
+void intel_region_release_pbo(intelScreenPrivate *intelScreen,
+                              struct intel_region *region);
+void intel_region_cow(intelScreenPrivate *intelScreen,
+                      struct intel_region *region);
+
+struct _DriBufferObject *intel_region_buffer(intelScreenPrivate *intelScreen,
+                                             struct intel_region *region,
+                                             GLuint flag);
+
+#endif

src/mesa/drivers/dri/i915/intel_render.c

@@ -51,14 +51,14 @@
  * dma buffers.  Use strip/fan hardware primitives where possible.
  * Try to simulate missing primitives with indexed vertices.
  */
-#define HAVE_POINTS      0  /* Has it, but can't use because subpixel has to
-			     * be adjusted for points on the INTEL/I845G
-			     */
+#define HAVE_POINTS      0      /* Has it, but can't use because subpixel has to
+                                 * be adjusted for points on the INTEL/I845G
+                                 */
 #define HAVE_LINES       1
 #define HAVE_LINE_STRIPS 1
 #define HAVE_TRIANGLES   1
 #define HAVE_TRI_STRIPS  1
-#define HAVE_TRI_STRIP_1 0  /* has it, template can't use it yet */
+#define HAVE_TRI_STRIP_1 0      /* has it, template can't use it yet */
 #define HAVE_TRI_FANS    1
 #define HAVE_POLYGONS    1
 #define HAVE_QUADS       0
@@ -66,7 +66,7 @@
 
 #define HAVE_ELTS        0
 
-static GLuint hw_prim[GL_POLYGON+1] = {
+static GLuint hw_prim[GL_POLYGON + 1] = {
    0,
    PRIM3D_LINELIST,
    PRIM3D_LINESTRIP,
@@ -79,7 +79,7 @@ static GLuint hw_prim[GL_POLYGON+1] = {
    PRIM3D_POLY
 };
 
-static const GLenum reduced_prim[GL_POLYGON+1] = {  
+static const GLenum reduced_prim[GL_POLYGON + 1] = {
    GL_POINTS,
    GL_LINES,
    GL_LINES,
@@ -92,58 +92,61 @@ static const GLenum reduced_prim[GL_POLYGON+1] = {
    GL_TRIANGLES
 };
 
-static const int scale_prim[GL_POLYGON+1] = {  
-   0,				/* fallback case */
+static const int scale_prim[GL_POLYGON + 1] = {
+   0,                           /* fallback case */
    1,
    2,
    2,
    1,
    3,
    3,
-   0,				/* fallback case */
-   0,				/* fallback case */
+   0,                           /* fallback case */
+   0,                           /* fallback case */
    3
 };
 
 
-static void intelDmaPrimitive( intelContextPtr intel, GLenum prim )
+static void
+intelDmaPrimitive(struct intel_context *intel, GLenum prim)
 {
-   if (0) fprintf(stderr, "%s %s\n", __FUNCTION__, _mesa_lookup_enum_by_nr(prim));
+   if (0)
+      fprintf(stderr, "%s %s\n", __FUNCTION__, _mesa_lookup_enum_by_nr(prim));
    INTEL_FIREVERTICES(intel);
-   intel->vtbl.reduced_primitive_state( intel, reduced_prim[prim] );
-   intelStartInlinePrimitive( intel, hw_prim[prim] );
+   intel->vtbl.reduced_primitive_state(intel, reduced_prim[prim]);
+   intelStartInlinePrimitive(intel, hw_prim[prim], INTEL_BATCH_CLIPRECTS);
 }
 
 
-#define LOCAL_VARS intelContextPtr intel = INTEL_CONTEXT(ctx)
+#define LOCAL_VARS struct intel_context *intel = intel_context(ctx)
 #define INIT( prim ) 				\
 do {						\
    intelDmaPrimitive( intel, prim );		\
 } while (0)
-#define FLUSH()  INTEL_FIREVERTICES( intel )
+
+#define FLUSH() INTEL_FIREVERTICES(intel)
 
 #define GET_SUBSEQUENT_VB_MAX_VERTS() \
-  (((intel->alloc.size / 2) - 1500) / (intel->vertex_size*4))
+  ((BATCH_SZ - 1500) / (intel->vertex_size*4))
 #define GET_CURRENT_VB_MAX_VERTS() GET_SUBSEQUENT_VB_MAX_VERTS()
 
 #define ALLOC_VERTS( nr ) \
    intelExtendInlinePrimitive( intel, (nr) * intel->vertex_size )
-  
+
 #define EMIT_VERTS( ctx, j, nr, buf ) \
-  _tnl_emit_vertices_to_buffer(ctx, j, (j)+(nr), buf )  
+  _tnl_emit_vertices_to_buffer(ctx, j, (j)+(nr), buf )
 
 #define TAG(x) intel_##x
 #include "tnl_dd/t_dd_dmatmp.h"
-  
-  
+
+
 /**********************************************************************/
 /*                          Render pipeline stage                     */
 /**********************************************************************/
 
 /* Heuristic to choose between the two render paths:  
  */
-static GLboolean choose_render( intelContextPtr intel,
-				struct vertex_buffer *VB )
+static GLboolean
+choose_render(struct intel_context *intel, struct vertex_buffer *VB)
 {
    int vertsz = intel->vertex_size;
    int cost_render = 0;
@@ -153,20 +156,20 @@ static GLboolean choose_render( intelContextPtr intel,
    int nr_rverts = 0;
    int rprim = intel->reduced_primitive;
    int i = 0;
-   
-   for (i = 0 ; i < VB->PrimitiveCount ; i++) {
+
+   for (i = 0; i < VB->PrimitiveCount; i++) {
       GLuint prim = VB->Primitive[i].mode;
       GLuint length = VB->Primitive[i].count;
 
       if (!length)
-	 continue;
+         continue;
 
       nr_prims++;
       nr_rverts += length * scale_prim[prim & PRIM_MODE_MASK];
 
       if (reduced_prim[prim & PRIM_MODE_MASK] != rprim) {
-	 nr_rprims++;
-	 rprim = reduced_prim[prim & PRIM_MODE_MASK];
+         nr_rprims++;
+         rprim = reduced_prim[prim & PRIM_MODE_MASK];
       }
    }
 
@@ -177,64 +180,63 @@ static GLboolean choose_render( intelContextPtr intel,
 
    /* One point for every 1024 dwords (4k) of dma:
     */
-   cost_render += (vertsz * i) / 1024; 
-   cost_fallback += (vertsz * nr_rverts) / 1024; 
+   cost_render += (vertsz * i) / 1024;
+   cost_fallback += (vertsz * nr_rverts) / 1024;
 
    if (0)
       fprintf(stderr, "cost render: %d fallback: %d\n",
-	      cost_render, cost_fallback);
+              cost_render, cost_fallback);
 
-   if (cost_render > cost_fallback) 
+   if (cost_render > cost_fallback)
       return GL_FALSE;
 
    return GL_TRUE;
 }
 
 
-static GLboolean intel_run_render( GLcontext *ctx, 
-				 struct tnl_pipeline_stage *stage )
+static GLboolean
+intel_run_render(GLcontext * ctx, struct tnl_pipeline_stage *stage)
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    GLuint i;
 
    /* Don't handle clipping or indexed vertices.
     */
-   if (intel->RenderIndex != 0 || 
-       !intel_validate_render( ctx, VB ) || 
-       !choose_render( intel, VB )) {
+   if (intel->RenderIndex != 0 ||
+       !intel_validate_render(ctx, VB) || !choose_render(intel, VB)) {
       return GL_TRUE;
    }
 
    tnl->clipspace.new_inputs |= VERT_BIT_POS;
 
-   tnl->Driver.Render.Start( ctx );
-   
-   for (i = 0 ; i < VB->PrimitiveCount ; i++)
-   {
+   tnl->Driver.Render.Start(ctx);
+
+   for (i = 0; i < VB->PrimitiveCount; i++) {
       GLuint prim = VB->Primitive[i].mode;
       GLuint start = VB->Primitive[i].start;
       GLuint length = VB->Primitive[i].count;
 
       if (!length)
-	 continue;
+         continue;
 
-      intel_render_tab_verts[prim & PRIM_MODE_MASK]( ctx, start, start + length,
-						     prim );
+      intel_render_tab_verts[prim & PRIM_MODE_MASK] (ctx, start,
+                                                     start + length, prim);
    }
-      
-   tnl->Driver.Render.Finish( ctx );
 
-   return GL_FALSE;     /* finished the pipe */
+   tnl->Driver.Render.Finish(ctx);
+
+   INTEL_FIREVERTICES(intel);
+
+   return GL_FALSE;             /* finished the pipe */
 }
 
-const struct tnl_pipeline_stage _intel_render_stage =
-{
+const struct tnl_pipeline_stage _intel_render_stage = {
    "intel render",
    NULL,
    NULL,
    NULL,
    NULL,
-   intel_run_render	/* run */
+   intel_run_render             /* run */
 };

src/mesa/drivers/dri/i915/intel_rotate.c

@@ -15,11 +15,14 @@
 
 void
 matrix23Set(struct matrix23 *m,
-            int m00, int m01, int m02,
-            int m10, int m11, int m12)
+            int m00, int m01, int m02, int m10, int m11, int m12)
 {
-   m->m00 = m00;   m->m01 = m01;   m->m02 = m02;
-   m->m10 = m10;   m->m11 = m11;   m->m12 = m12;
+   m->m00 = m00;
+   m->m01 = m01;
+   m->m02 = m02;
+   m->m10 = m10;
+   m->m11 = m11;
+   m->m12 = m12;
 }
 
 
@@ -66,9 +69,9 @@ matrix23TransformDistance(const struct matrix23 *m, int *xDist, int *yDist)
    *yDist = (y1 - y0) + (y2 - y0);
 
    if (*xDist < 0)
-       *xDist = -*xDist;
+      *xDist = -*xDist;
    if (*yDist < 0)
-       *yDist = -*yDist;
+      *yDist = -*yDist;
 }
 
 
@@ -76,7 +79,8 @@ matrix23TransformDistance(const struct matrix23 *m, int *xDist, int *yDist)
  * Transform the rect defined by (x, y, w, h) by m.
  */
 void
-matrix23TransformRect(const struct matrix23 *m, int *x, int *y, int *w, int *h)
+matrix23TransformRect(const struct matrix23 *m, int *x, int *y, int *w,
+                      int *h)
 {
    int x0 = *x, y0 = *y;
    int x1 = *x + *w, y1 = *y;
@@ -108,16 +112,16 @@ matrix23Rotate(struct matrix23 *m, int width, int height, int angle)
       matrix23Set(m, 1, 0, 0, 0, 1, 0);
       break;
    case 90:
-      matrix23Set(m, 0, 1, 0,  -1, 0, width);
+      matrix23Set(m, 0, 1, 0, -1, 0, width);
       break;
    case 180:
-      matrix23Set(m, -1, 0, width,  0, -1, height);
+      matrix23Set(m, -1, 0, width, 0, -1, height);
       break;
    case 270:
-      matrix23Set(m, 0, -1, height,  1, 0, 0);
+      matrix23Set(m, 0, -1, height, 1, 0, 0);
       break;
    default:
-      /*abort()*/;
+      /*abort() */ ;
    }
 }
 
@@ -129,16 +133,24 @@ void
 matrix23Flip(struct matrix23 *m, int width, int height, int xflip, int yflip)
 {
    if (xflip) {
-      m->m00 = -1;  m->m01 = 0;   m->m02 = width - 1;
+      m->m00 = -1;
+      m->m01 = 0;
+      m->m02 = width - 1;
    }
    else {
-      m->m00 = 1;   m->m01 = 0;   m->m02 = 0;
+      m->m00 = 1;
+      m->m01 = 0;
+      m->m02 = 0;
    }
    if (yflip) {
-      m->m10 = 0;   m->m11 = -1;  m->m12 = height - 1;
+      m->m10 = 0;
+      m->m11 = -1;
+      m->m12 = height - 1;
    }
    else {
-      m->m10 = 0;   m->m11 = 1;   m->m12 = 0;
+      m->m10 = 0;
+      m->m11 = 1;
+      m->m12 = 0;
    }
 }
 
@@ -169,14 +181,18 @@ main(int argc, char *argv[])
 {
    int width = 500, height = 400;
    int rot;
-   int fx = 0, fy = 0;  /* flip x and/or y ? */
+   int fx = 0, fy = 0;          /* flip x and/or y ? */
    int coords[4][2];
 
    /* four corner coords to test with */
-   coords[0][0] = 0;  coords[0][1] = 0;
-   coords[1][0] = width-1;  coords[1][1] = 0;
-   coords[2][0] = width-1;  coords[2][1] = height-1;
-   coords[3][0] = 0;  coords[3][1] = height-1;
+   coords[0][0] = 0;
+   coords[0][1] = 0;
+   coords[1][0] = width - 1;
+   coords[1][1] = 0;
+   coords[2][0] = width - 1;
+   coords[2][1] = height - 1;
+   coords[3][0] = 0;
+   coords[3][1] = height - 1;
 
 
    for (rot = 0; rot < 360; rot += 90) {

src/mesa/drivers/dri/i915/intel_rotate.h

@@ -11,11 +11,9 @@ struct matrix23
 
 extern void
 matrix23Set(struct matrix23 *m,
-            int m00, int m01, int m02,
-            int m10, int m11, int m12);
+            int m00, int m01, int m02, int m10, int m11, int m12);
 
-extern void
-matrix23TransformCoordi(const struct matrix23 *m, int *x, int *y);
+extern void matrix23TransformCoordi(const struct matrix23 *m, int *x, int *y);
 
 extern void
 matrix23TransformCoordf(const struct matrix23 *m, float *x, float *y);

src/mesa/drivers/dri/i915/intel_screen.c

@@ -38,46 +38,47 @@
 
 #include "intel_screen.h"
 
+#include "intel_buffers.h"
 #include "intel_tex.h"
 #include "intel_span.h"
 #include "intel_tris.h"
 #include "intel_ioctl.h"
+#include "intel_fbo.h"
 
 #include "i830_dri.h"
+#include "dri_bufpool.h"
+#include "intel_regions.h"
 
 PUBLIC const char __driConfigOptions[] =
-DRI_CONF_BEGIN
-    DRI_CONF_SECTION_PERFORMANCE
-       DRI_CONF_FTHROTTLE_MODE(DRI_CONF_FTHROTTLE_IRQS) 
-       DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0)
-    DRI_CONF_SECTION_END
-    DRI_CONF_SECTION_QUALITY
-       DRI_CONF_FORCE_S3TC_ENABLE(false)
-       DRI_CONF_ALLOW_LARGE_TEXTURES(1)
-      DRI_CONF_SECTION_END
-DRI_CONF_END;
-const GLuint __driNConfigOptions = 4;
+   DRI_CONF_BEGIN DRI_CONF_SECTION_PERFORMANCE
+   DRI_CONF_FTHROTTLE_MODE(DRI_CONF_FTHROTTLE_IRQS)
+   DRI_CONF_VBLANK_MODE(DRI_CONF_VBLANK_DEF_INTERVAL_0)
+   DRI_CONF_SECTION_END DRI_CONF_SECTION_QUALITY
+   DRI_CONF_FORCE_S3TC_ENABLE(false)
+   DRI_CONF_ALLOW_LARGE_TEXTURES(1)
+   DRI_CONF_SECTION_END DRI_CONF_END;
+     const GLuint __driNConfigOptions = 4;
 
 #ifdef USE_NEW_INTERFACE
-static PFNGLXCREATECONTEXTMODES create_context_modes = NULL;
-#endif /*USE_NEW_INTERFACE*/
+     static PFNGLXCREATECONTEXTMODES create_context_modes = NULL;
+#endif /*USE_NEW_INTERFACE */
 
-extern const struct dri_extension card_extensions[];
+     extern const struct dri_extension card_extensions[];
 
 /**
  * Map all the memory regions described by the screen.
  * \return GL_TRUE if success, GL_FALSE if error.
  */
 GLboolean
-intelMapScreenRegions(__DRIscreenPrivate *sPriv)
+intelMapScreenRegions(__DRIscreenPrivate * sPriv)
 {
-   intelScreenPrivate *intelScreen = (intelScreenPrivate *)sPriv->private;
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
 
    if (intelScreen->front.handle) {
       if (drmMap(sPriv->fd,
                  intelScreen->front.handle,
                  intelScreen->front.size,
-                 (drmAddress *)&intelScreen->front.map) != 0) {
+                 (drmAddress *) & intelScreen->front.map) != 0) {
          _mesa_problem(NULL, "drmMap(frontbuffer) failed!");
          return GL_FALSE;
       }
@@ -86,42 +87,140 @@ intelMapScreenRegions(__DRIscreenPrivate *sPriv)
       _mesa_warning(NULL, "no front buffer handle in intelMapScreenRegions!");
    }
 
+   _mesa_printf("Back 0x%08x ", intelScreen->back.handle);
    if (drmMap(sPriv->fd,
               intelScreen->back.handle,
               intelScreen->back.size,
-              (drmAddress *)&intelScreen->back.map) != 0) {
+              (drmAddress *) & intelScreen->back.map) != 0) {
       intelUnmapScreenRegions(intelScreen);
       return GL_FALSE;
    }
 
+   _mesa_printf("Depth 0x%08x ", intelScreen->depth.handle);
    if (drmMap(sPriv->fd,
               intelScreen->depth.handle,
               intelScreen->depth.size,
-              (drmAddress *)&intelScreen->depth.map) != 0) {
+              (drmAddress *) & intelScreen->depth.map) != 0) {
       intelUnmapScreenRegions(intelScreen);
       return GL_FALSE;
    }
 
+#if 0
+   _mesa_printf("TEX 0x%08x ", intelScreen->tex.handle);
    if (drmMap(sPriv->fd,
               intelScreen->tex.handle,
               intelScreen->tex.size,
-              (drmAddress *)&intelScreen->tex.map) != 0) {
+              (drmAddress *) & intelScreen->tex.map) != 0) {
       intelUnmapScreenRegions(intelScreen);
       return GL_FALSE;
    }
-
+#endif
    if (0)
       printf("Mappings:  front: %p  back: %p  depth: %p  tex: %p\n",
-          intelScreen->front.map,
-          intelScreen->back.map,
-          intelScreen->depth.map,
-          intelScreen->tex.map);
+             intelScreen->front.map,
+             intelScreen->back.map,
+             intelScreen->depth.map, intelScreen->tex.map);
    return GL_TRUE;
 }
 
 
+static struct intel_region *
+intel_recreate_static(intelScreenPrivate *intelScreen,
+		      struct intel_region *region,
+		      GLuint mem_type,
+		      GLuint offset,
+		      void *virtual,
+		      GLuint cpp, GLuint pitch, GLuint height)
+{
+  if (region) {
+    intel_region_update_static(intelScreen, region, mem_type, offset,
+			       virtual, cpp, pitch, height);
+  } else {
+    region = intel_region_create_static(intelScreen, mem_type, offset,
+					virtual, cpp, pitch, height);
+  }
+  return region;
+}
+    
+
+/* Create intel_region structs to describe the static front,back,depth
+ * buffers created by the xserver. 
+ *
+ * Although FBO's mean we now no longer use these as render targets in
+ * all circumstances, they won't go away until the back and depth
+ * buffers become private, and the front and rotated buffers will
+ * remain even then.
+ *
+ * Note that these don't allocate video memory, just describe
+ * allocations alread made by the X server.
+ */
+static void
+intel_recreate_static_regions(intelScreenPrivate *intelScreen)
+{
+   intelScreen->front_region =
+      intel_recreate_static(intelScreen,
+			    intelScreen->front_region,
+			    DRM_BO_FLAG_MEM_TT,
+			    intelScreen->front.offset,
+			    intelScreen->front.map,
+			    intelScreen->cpp,
+			    intelScreen->front.pitch / intelScreen->cpp,
+			    intelScreen->height);
+
+   intelScreen->rotated_region =
+      intel_recreate_static(intelScreen,
+			    intelScreen->rotated_region,
+			    DRM_BO_FLAG_MEM_TT,
+			    intelScreen->rotated.offset,
+			    intelScreen->rotated.map,
+			    intelScreen->cpp,
+			    intelScreen->rotated.pitch /
+			    intelScreen->cpp, intelScreen->height);
+
+
+   intelScreen->back_region =
+      intel_recreate_static(intelScreen,
+			    intelScreen->back_region,
+			    DRM_BO_FLAG_MEM_TT,
+			    intelScreen->back.offset,
+			    intelScreen->back.map,
+			    intelScreen->cpp,
+			    intelScreen->back.pitch / intelScreen->cpp,
+			    intelScreen->height);
+
+   /* Still assuming front.cpp == depth.cpp
+    */
+   intelScreen->depth_region =
+      intel_recreate_static(intelScreen,
+			    intelScreen->depth_region,
+			    DRM_BO_FLAG_MEM_TT,
+			    intelScreen->depth.offset,
+			    intelScreen->depth.map,
+			    intelScreen->cpp,
+			    intelScreen->depth.pitch / intelScreen->cpp,
+			    intelScreen->height);
+}
+
+/**
+ * Use the information in the sarea to update the screen parameters
+ * related to screen rotation. Needs to be called locked.
+ */
 void
-intelUnmapScreenRegions(intelScreenPrivate *intelScreen)
+intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea)
+{
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
+
+   intelUnmapScreenRegions(intelScreen);
+   intelUpdateScreenFromSAREA(intelScreen, sarea);
+   if (!intelMapScreenRegions(sPriv)) {
+      fprintf(stderr, "ERROR Remapping screen regions!!!\n");
+   }
+   intel_recreate_static_regions(intelScreen);
+}
+
+
+void
+intelUnmapScreenRegions(intelScreenPrivate * intelScreen)
 {
 #define REALLY_UNMAP 1
    if (intelScreen->front.map) {
@@ -154,9 +253,8 @@ intelUnmapScreenRegions(intelScreenPrivate *intelScreen)
 
 
 static void
-intelPrintDRIInfo(intelScreenPrivate *intelScreen,
-                  __DRIscreenPrivate *sPriv,
-                  I830DRIPtr gDRIPriv)
+intelPrintDRIInfo(intelScreenPrivate * intelScreen,
+                  __DRIscreenPrivate * sPriv, I830DRIPtr gDRIPriv)
 {
    fprintf(stderr, "*** Front size:   0x%x  offset: 0x%x  pitch: %d\n",
            intelScreen->front.size, intelScreen->front.offset,
@@ -177,9 +275,10 @@ intelPrintDRIInfo(intelScreenPrivate *intelScreen,
 
 
 static void
-intelPrintSAREA(const drmI830Sarea *sarea)
+intelPrintSAREA(const drmI830Sarea * sarea)
 {
-   fprintf(stderr, "SAREA: sarea width %d  height %d\n", sarea->width, sarea->height);
+   fprintf(stderr, "SAREA: sarea width %d  height %d\n", sarea->width,
+           sarea->height);
    fprintf(stderr, "SAREA: pitch: %d\n", sarea->pitch);
    fprintf(stderr,
            "SAREA: front offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
@@ -193,8 +292,7 @@ intelPrintSAREA(const drmI830Sarea *sarea)
            sarea->depth_offset, sarea->depth_size,
            (unsigned) sarea->depth_handle);
    fprintf(stderr, "SAREA: tex   offset: 0x%08x  size: 0x%x  handle: 0x%x\n",
-           sarea->tex_offset, sarea->tex_size,
-           (unsigned) sarea->tex_handle);
+           sarea->tex_offset, sarea->tex_size, (unsigned) sarea->tex_handle);
    fprintf(stderr, "SAREA: rotation: %d\n", sarea->rotation);
    fprintf(stderr,
            "SAREA: rotated offset: 0x%08x  size: 0x%x\n",
@@ -208,8 +306,8 @@ intelPrintSAREA(const drmI830Sarea *sarea)
  * information in the SAREA.  This function updates those parameters.
  */
 void
-intelUpdateScreenFromSAREA(intelScreenPrivate *intelScreen,
-                           drmI830Sarea *sarea)
+intelUpdateScreenFromSAREA(intelScreenPrivate * intelScreen,
+                           drmI830Sarea * sarea)
 {
    intelScreen->width = sarea->width;
    intelScreen->height = sarea->height;
@@ -223,7 +321,7 @@ intelUpdateScreenFromSAREA(intelScreenPrivate *intelScreen,
    intelScreen->back.pitch = sarea->pitch * intelScreen->cpp;
    intelScreen->back.handle = sarea->back_handle;
    intelScreen->back.size = sarea->back_size;
-			 
+
    intelScreen->depth.offset = sarea->depth_offset;
    intelScreen->depth.pitch = sarea->pitch * intelScreen->cpp;
    intelScreen->depth.handle = sarea->depth_handle;
@@ -248,58 +346,87 @@ intelUpdateScreenFromSAREA(intelScreenPrivate *intelScreen,
 }
 
 
-static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv)
+static GLboolean
+intelInitDriver(__DRIscreenPrivate * sPriv)
 {
    intelScreenPrivate *intelScreen;
-   I830DRIPtr         gDRIPriv = (I830DRIPtr)sPriv->pDevPriv;
+   I830DRIPtr gDRIPriv = (I830DRIPtr) sPriv->pDevPriv;
    drmI830Sarea *sarea;
    PFNGLXSCRENABLEEXTENSIONPROC glx_enable_extension =
-     (PFNGLXSCRENABLEEXTENSIONPROC) (*dri_interface->getProcAddress("glxEnableExtension"));
-   void * const psc = sPriv->psc->screenConfigs;
+      (PFNGLXSCRENABLEEXTENSIONPROC) (*dri_interface->
+                                      getProcAddress("glxEnableExtension"));
+   void *const psc = sPriv->psc->screenConfigs;
 
    if (sPriv->devPrivSize != sizeof(I830DRIRec)) {
-      fprintf(stderr,"\nERROR!  sizeof(I830DRIRec) does not match passed size from device driver\n");
+      fprintf(stderr,
+              "\nERROR!  sizeof(I830DRIRec) does not match passed size from device driver\n");
       return GL_FALSE;
    }
 
    /* Allocate the private area */
-   intelScreen = (intelScreenPrivate *)CALLOC(sizeof(intelScreenPrivate));
+   intelScreen = (intelScreenPrivate *) CALLOC(sizeof(intelScreenPrivate));
    if (!intelScreen) {
-      fprintf(stderr,"\nERROR!  Allocating private area failed\n");
+      fprintf(stderr, "\nERROR!  Allocating private area failed\n");
       return GL_FALSE;
    }
    /* parse information in __driConfigOptions */
-   driParseOptionInfo (&intelScreen->optionCache,
-		       __driConfigOptions, __driNConfigOptions);
+   driParseOptionInfo(&intelScreen->optionCache,
+                      __driConfigOptions, __driNConfigOptions);
 
    intelScreen->driScrnPriv = sPriv;
-   sPriv->private = (void *)intelScreen;
+   sPriv->private = (void *) intelScreen;
    intelScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
    sarea = (drmI830Sarea *)
-         (((GLubyte *)sPriv->pSAREA)+intelScreen->sarea_priv_offset);
+      (((GLubyte *) sPriv->pSAREA) + intelScreen->sarea_priv_offset);
 
    intelScreen->deviceID = gDRIPriv->deviceID;
    intelScreen->mem = gDRIPriv->mem;
    intelScreen->cpp = gDRIPriv->cpp;
 
    switch (gDRIPriv->bitsPerPixel) {
-   case 15: intelScreen->fbFormat = DV_PF_555; break;
-   case 16: intelScreen->fbFormat = DV_PF_565; break;
-   case 32: intelScreen->fbFormat = DV_PF_8888; break;
+   case 16:
+      intelScreen->fbFormat = DV_PF_565;
+      break;
+   case 32:
+      intelScreen->fbFormat = DV_PF_8888;
+      break;
+   default:
+      exit(1);
+      break;
    }
-			 
+
    intelUpdateScreenFromSAREA(intelScreen, sarea);
 
-   if (0)
-      intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv);
-
    if (!intelMapScreenRegions(sPriv)) {
-      fprintf(stderr,"\nERROR!  mapping regions\n");
+      fprintf(stderr, "\nERROR!  mapping regions\n");
       _mesa_free(intelScreen);
       sPriv->private = NULL;
       return GL_FALSE;
    }
 
+#if 0
+
+   /*
+    * FIXME: Remove this code and its references.
+    */
+
+   intelScreen->tex.offset = gDRIPriv->textureOffset;
+   intelScreen->logTextureGranularity = gDRIPriv->logTextureGranularity;
+   intelScreen->tex.handle = gDRIPriv->textures;
+   intelScreen->tex.size = gDRIPriv->textureSize;
+
+#else
+   intelScreen->tex.offset = 0;
+   intelScreen->logTextureGranularity = 0;
+   intelScreen->tex.handle = 0;
+   intelScreen->tex.size = 0;
+#endif
+
+   intelScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+
+   if (1)
+      intelPrintDRIInfo(intelScreen, sPriv, gDRIPriv);
+
    intelScreen->drmMinor = sPriv->drmMinor;
 
    /* Determine if IRQs are active? */
@@ -310,11 +437,11 @@ static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv)
       gp.param = I830_PARAM_IRQ_ACTIVE;
       gp.value = &intelScreen->irq_active;
 
-      ret = drmCommandWriteRead( sPriv->fd, DRM_I830_GETPARAM,
-				 &gp, sizeof(gp));
+      ret = drmCommandWriteRead(sPriv->fd, DRM_I830_GETPARAM,
+                                &gp, sizeof(gp));
       if (ret) {
-	 fprintf(stderr, "drmI830GetParam: %d\n", ret);
-	 return GL_FALSE;
+         fprintf(stderr, "drmI830GetParam: %d\n", ret);
+         return GL_FALSE;
       }
    }
 
@@ -326,125 +453,141 @@ static GLboolean intelInitDriver(__DRIscreenPrivate *sPriv)
       gp.param = I830_PARAM_ALLOW_BATCHBUFFER;
       gp.value = &intelScreen->allow_batchbuffer;
 
-      ret = drmCommandWriteRead( sPriv->fd, DRM_I830_GETPARAM,
-				 &gp, sizeof(gp));
+      ret = drmCommandWriteRead(sPriv->fd, DRM_I830_GETPARAM,
+                                &gp, sizeof(gp));
       if (ret) {
-	 fprintf(stderr, "drmI830GetParam: (%d) %d\n", gp.param, ret);
-	 return GL_FALSE;
+         fprintf(stderr, "drmI830GetParam: (%d) %d\n", gp.param, ret);
+         return GL_FALSE;
       }
    }
 
    if (glx_enable_extension != NULL) {
-      (*glx_enable_extension)( psc, "GLX_SGI_swap_control" );
-      (*glx_enable_extension)( psc, "GLX_SGI_video_sync" );
-      (*glx_enable_extension)( psc, "GLX_MESA_swap_control" );
-      (*glx_enable_extension)( psc, "GLX_MESA_swap_frame_usage" );
-      (*glx_enable_extension)( psc, "GLX_SGI_make_current_read" );
-      (*glx_enable_extension)( psc, "GLX_MESA_allocate_memory" );
-      (*glx_enable_extension)( psc, "GLX_MESA_copy_sub_buffer" );
+      (*glx_enable_extension) (psc, "GLX_SGI_swap_control");
+      (*glx_enable_extension) (psc, "GLX_SGI_video_sync");
+      (*glx_enable_extension) (psc, "GLX_MESA_swap_control");
+      (*glx_enable_extension) (psc, "GLX_MESA_swap_frame_usage");
+      (*glx_enable_extension) (psc, "GLX_SGI_make_current_read");
    }
-   
-   sPriv->psc->allocateMemory = (void *) intelAllocateMemoryMESA;
-   sPriv->psc->freeMemory     = (void *) intelFreeMemoryMESA;
-   sPriv->psc->memoryOffset   = (void *) intelGetMemoryOffsetMESA;
+
+   intelScreen->regionPool = driDRMPoolInit(sPriv->fd);
+   if (!intelScreen->regionPool)
+      return GL_FALSE;
+   intelScreen->staticPool = driDRMStaticPoolInit(sPriv->fd);
+   if (!intelScreen->staticPool)
+      return GL_FALSE;
+   intelScreen->texPool = intelScreen->regionPool;
+   intelScreen->batchPool = driBatchPoolInit(sPriv->fd,
+                                             DRM_BO_FLAG_EXE |
+                                             DRM_BO_FLAG_MEM_TT |
+                                             DRM_BO_FLAG_MEM_LOCAL,
+                                             4096, 100, 5);
+   intel_recreate_static_regions(intelScreen);
+
 
    return GL_TRUE;
 }
-		
-		
-static void intelDestroyScreen(__DRIscreenPrivate *sPriv)
+
+
+static void
+intelDestroyScreen(__DRIscreenPrivate * sPriv)
 {
-   intelScreenPrivate *intelScreen = (intelScreenPrivate *)sPriv->private;
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
 
    intelUnmapScreenRegions(intelScreen);
+
+   driPoolTakeDown(intelScreen->regionPool);
+   driPoolTakeDown(intelScreen->staticPool);
+   driPoolTakeDown(intelScreen->batchPool);
    FREE(intelScreen);
    sPriv->private = NULL;
 }
 
 
-static GLboolean intelCreateBuffer( __DRIscreenPrivate *driScrnPriv,
-				    __DRIdrawablePrivate *driDrawPriv,
-				    const __GLcontextModes *mesaVis,
-				    GLboolean isPixmap )
+/**
+ * This is called when we need to set up GL rendering to a new X window.
+ */
+static GLboolean
+intelCreateBuffer(__DRIscreenPrivate * driScrnPriv,
+                  __DRIdrawablePrivate * driDrawPriv,
+                  const __GLcontextModes * mesaVis, GLboolean isPixmap)
 {
    intelScreenPrivate *screen = (intelScreenPrivate *) driScrnPriv->private;
 
    if (isPixmap) {
-      return GL_FALSE; /* not implemented */
-   } else {
-      GLboolean swStencil = (mesaVis->stencilBits > 0 && 
-			     mesaVis->depthBits != 24);
+      return GL_FALSE;          /* not implemented */
+   }
+   else {
+      GLboolean swStencil = (mesaVis->stencilBits > 0 &&
+                             mesaVis->depthBits != 24);
+      GLenum rgbFormat = (mesaVis->redBits == 5 ? GL_RGB5 : GL_RGBA8);
 
       struct gl_framebuffer *fb = _mesa_create_framebuffer(mesaVis);
 
+      /* setup the hardware-based renderbuffers */
       {
-         driRenderbuffer *frontRb
-            = driNewRenderbuffer(GL_RGBA,
-                                 screen->front.map,
-                                 screen->cpp,
-                                 screen->front.offset, screen->front.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(frontRb, mesaVis);
+         struct intel_renderbuffer *frontRb
+            = intel_create_renderbuffer(rgbFormat,
+                                        screen->width, screen->height,
+                                        screen->front.offset,
+                                        screen->front.pitch,
+                                        screen->cpp,
+                                        screen->front.map);
+         intel_set_span_functions(&frontRb->Base);
          _mesa_add_renderbuffer(fb, BUFFER_FRONT_LEFT, &frontRb->Base);
       }
 
       if (mesaVis->doubleBufferMode) {
-         driRenderbuffer *backRb
-            = driNewRenderbuffer(GL_RGBA,
-                                 screen->back.map,
-                                 screen->cpp,
-                                 screen->back.offset, screen->back.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(backRb, mesaVis);
+         struct intel_renderbuffer *backRb
+            = intel_create_renderbuffer(rgbFormat,
+                                        screen->width, screen->height,
+                                        screen->back.offset,
+                                        screen->back.pitch,
+                                        screen->cpp,
+                                        screen->back.map);
+         intel_set_span_functions(&backRb->Base);
          _mesa_add_renderbuffer(fb, BUFFER_BACK_LEFT, &backRb->Base);
       }
 
-      if (mesaVis->depthBits == 16) {
-         driRenderbuffer *depthRb
-            = driNewRenderbuffer(GL_DEPTH_COMPONENT16,
-                                 screen->depth.map,
-                                 screen->cpp,
-                                 screen->depth.offset, screen->depth.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(depthRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
+      if (mesaVis->depthBits == 24 && mesaVis->stencilBits == 8) {
+         /* combined depth/stencil buffer */
+         struct intel_renderbuffer *depthStencilRb
+            = intel_create_renderbuffer(GL_DEPTH24_STENCIL8_EXT,
+                                        screen->width, screen->height,
+                                        screen->depth.offset,
+                                        screen->depth.pitch,
+                                        screen->cpp,    /* 4! */
+                                        screen->depth.map);
+         intel_set_span_functions(&depthStencilRb->Base);
+         /* note: bind RB to two attachment points */
+         _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthStencilRb->Base);
+         _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &depthStencilRb->Base);
       }
-      else if (mesaVis->depthBits == 24) {
-         driRenderbuffer *depthRb
-            = driNewRenderbuffer(GL_DEPTH_COMPONENT24,
-                                 screen->depth.map,
-                                 screen->cpp,
-                                 screen->depth.offset, screen->depth.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(depthRb, mesaVis);
+      else if (mesaVis->depthBits == 16) {
+         /* just 16-bit depth buffer, no hw stencil */
+         struct intel_renderbuffer *depthRb
+            = intel_create_renderbuffer(GL_DEPTH_COMPONENT16,
+                                        screen->width, screen->height,
+                                        screen->depth.offset,
+                                        screen->depth.pitch,
+                                        screen->cpp,    /* 2! */
+                                        screen->depth.map);
+         intel_set_span_functions(&depthRb->Base);
          _mesa_add_renderbuffer(fb, BUFFER_DEPTH, &depthRb->Base);
       }
 
-      if (mesaVis->stencilBits > 0 && !swStencil) {
-         driRenderbuffer *stencilRb
-            = driNewRenderbuffer(GL_STENCIL_INDEX8_EXT,
-                                 screen->depth.map,
-                                 screen->cpp,
-                                 screen->depth.offset, screen->depth.pitch,
-                                 driDrawPriv);
-         intelSetSpanFunctions(stencilRb, mesaVis);
-         _mesa_add_renderbuffer(fb, BUFFER_STENCIL, &stencilRb->Base);
-      }
-
-      _mesa_add_soft_renderbuffers(fb,
-                                   GL_FALSE, /* color */
-                                   GL_FALSE, /* depth */
-                                   swStencil,
-                                   mesaVis->accumRedBits > 0,
-                                   GL_FALSE, /* alpha */
-                                   GL_FALSE /* aux */);
+      /* now add any/all software-based renderbuffers we may need */
+      _mesa_add_soft_renderbuffers(fb, GL_FALSE,        /* never sw color */
+                                   GL_FALSE,    /* never sw depth */
+                                   swStencil, mesaVis->accumRedBits > 0, GL_FALSE,      /* never sw alpha */
+                                   GL_FALSE /* never sw aux */ );
       driDrawPriv->driverPrivate = (void *) fb;
 
       return (driDrawPriv->driverPrivate != NULL);
    }
 }
 
-static void intelDestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
+static void
+intelDestroyBuffer(__DRIdrawablePrivate * driDrawPriv)
 {
    _mesa_destroy_framebuffer((GLframebuffer *) (driDrawPriv->driverPrivate));
 }
@@ -454,13 +597,13 @@ static void intelDestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
  * Get information about previous buffer swaps.
  */
 static int
-intelGetSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo )
+intelGetSwapInfo(__DRIdrawablePrivate * dPriv, __DRIswapInfo * sInfo)
 {
-   intelContextPtr intel;
+   struct intel_context *intel;
 
-   if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL)
-	|| (dPriv->driContextPriv->driverPrivate == NULL)
-	|| (sInfo == NULL) ) {
+   if ((dPriv == NULL) || (dPriv->driContextPriv == NULL)
+       || (dPriv->driContextPriv->driverPrivate == NULL)
+       || (sInfo == NULL)) {
       return -1;
    }
 
@@ -470,8 +613,8 @@ intelGetSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo )
    sInfo->swap_missed_count = intel->swap_missed_count;
 
    sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0)
-       ? driCalculateSwapUsage( dPriv, 0, intel->swap_missed_ust )
-       : 0.0;
+      ? driCalculateSwapUsage(dPriv, 0, intel->swap_missed_ust)
+      : 0.0;
 
    return 0;
 }
@@ -481,39 +624,40 @@ intelGetSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo )
  * init-designated function to register chipids and createcontext
  * functions.
  */
-extern GLboolean i830CreateContext( const __GLcontextModes *mesaVis,
-				    __DRIcontextPrivate *driContextPriv,
-				    void *sharedContextPrivate);
+extern GLboolean i830CreateContext(const __GLcontextModes * mesaVis,
+                                   __DRIcontextPrivate * driContextPriv,
+                                   void *sharedContextPrivate);
 
-extern GLboolean i915CreateContext( const __GLcontextModes *mesaVis,
-				   __DRIcontextPrivate *driContextPriv,
-				   void *sharedContextPrivate);
+extern GLboolean i915CreateContext(const __GLcontextModes * mesaVis,
+                                   __DRIcontextPrivate * driContextPriv,
+                                   void *sharedContextPrivate);
 
 
 
 
-static GLboolean intelCreateContext( const __GLcontextModes *mesaVis,
-				   __DRIcontextPrivate *driContextPriv,
-				   void *sharedContextPrivate)
+static GLboolean
+intelCreateContext(const __GLcontextModes * mesaVis,
+                   __DRIcontextPrivate * driContextPriv,
+                   void *sharedContextPrivate)
 {
    __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
-   intelScreenPrivate *intelScreen = (intelScreenPrivate *)sPriv->private;
+   intelScreenPrivate *intelScreen = (intelScreenPrivate *) sPriv->private;
 
    switch (intelScreen->deviceID) {
+      /* Don't deal with i830 until texture work complete:
+       */
    case PCI_CHIP_845_G:
    case PCI_CHIP_I830_M:
    case PCI_CHIP_I855_GM:
    case PCI_CHIP_I865_G:
-      return i830CreateContext( mesaVis, driContextPriv, 
-				sharedContextPrivate );
+      return i830CreateContext(mesaVis, driContextPriv, sharedContextPrivate);
 
    case PCI_CHIP_I915_G:
    case PCI_CHIP_I915_GM:
    case PCI_CHIP_I945_G:
    case PCI_CHIP_I945_GM:
-      return i915CreateContext( mesaVis, driContextPriv, 
-			       sharedContextPrivate );
- 
+      return i915CreateContext(mesaVis, driContextPriv, sharedContextPrivate);
+
    default:
       fprintf(stderr, "Unrecognized deviceID %x\n", intelScreen->deviceID);
       return GL_FALSE;
@@ -522,30 +666,30 @@ static GLboolean intelCreateContext( const __GLcontextModes *mesaVis,
 
 
 static const struct __DriverAPIRec intelAPI = {
-   .InitDriver      = intelInitDriver,
-   .DestroyScreen   = intelDestroyScreen,
-   .CreateContext   = intelCreateContext,
-   .DestroyContext  = intelDestroyContext,
-   .CreateBuffer    = intelCreateBuffer,
-   .DestroyBuffer   = intelDestroyBuffer,
-   .SwapBuffers     = intelSwapBuffers,
-   .MakeCurrent     = intelMakeCurrent,
-   .UnbindContext   = intelUnbindContext,
-   .GetSwapInfo     = intelGetSwapInfo,
-   .GetMSC          = driGetMSC32,
-   .WaitForMSC      = driWaitForMSC32,
-   .WaitForSBC      = NULL,
-   .SwapBuffersMSC  = NULL,
-   .CopySubBuffer   = intelCopySubBuffer
+   .InitDriver = intelInitDriver,
+   .DestroyScreen = intelDestroyScreen,
+   .CreateContext = intelCreateContext,
+   .DestroyContext = intelDestroyContext,
+   .CreateBuffer = intelCreateBuffer,
+   .DestroyBuffer = intelDestroyBuffer,
+   .SwapBuffers = intelSwapBuffers,
+   .MakeCurrent = intelMakeCurrent,
+   .UnbindContext = intelUnbindContext,
+   .GetSwapInfo = intelGetSwapInfo,
+   .GetMSC = driGetMSC32,
+   .WaitForMSC = driWaitForMSC32,
+   .WaitForSBC = NULL,
+   .SwapBuffersMSC = NULL,
+   .CopySubBuffer = intelCopySubBuffer
 };
 
 
 static __GLcontextModes *
-intelFillInModes( unsigned pixel_bits, unsigned depth_bits,
-		 unsigned stencil_bits, GLboolean have_back_buffer )
+intelFillInModes(unsigned pixel_bits, unsigned depth_bits,
+                 unsigned stencil_bits, GLboolean have_back_buffer)
 {
-   __GLcontextModes * modes;
-   __GLcontextModes * m;
+   __GLcontextModes *modes;
+   __GLcontextModes *m;
    unsigned num_modes;
    unsigned depth_buffer_factor;
    unsigned back_buffer_factor;
@@ -576,43 +720,45 @@ intelFillInModes( unsigned pixel_bits, unsigned depth_bits,
    stencil_bits_array[2] = (stencil_bits == 0) ? 8 : stencil_bits;
 
    depth_buffer_factor = ((depth_bits != 0) || (stencil_bits != 0)) ? 3 : 1;
-   back_buffer_factor  = (have_back_buffer) ? 3 : 1;
+   back_buffer_factor = (have_back_buffer) ? 3 : 1;
 
    num_modes = depth_buffer_factor * back_buffer_factor * 4;
 
-    if ( pixel_bits == 16 ) {
-        fb_format = GL_RGB;
-        fb_type = GL_UNSIGNED_SHORT_5_6_5;
-    }
-    else {
-        fb_format = GL_BGRA;
-        fb_type = GL_UNSIGNED_INT_8_8_8_8_REV;
-    }
-
-   modes = (*dri_interface->createContextModes)( num_modes, sizeof( __GLcontextModes ) );
-   m = modes;
-   if ( ! driFillInModes( & m, fb_format, fb_type,
-			  depth_bits_array, stencil_bits_array, depth_buffer_factor,
-			  back_buffer_modes, back_buffer_factor,
-			  GLX_TRUE_COLOR ) ) {
-	fprintf( stderr, "[%s:%u] Error creating FBConfig!\n",
-		 __func__, __LINE__ );
-	return NULL;
+   if (pixel_bits == 16) {
+      fb_format = GL_RGB;
+      fb_type = GL_UNSIGNED_SHORT_5_6_5;
    }
-   if ( ! driFillInModes( & m, fb_format, fb_type,
-			  depth_bits_array, stencil_bits_array, depth_buffer_factor,
-			  back_buffer_modes, back_buffer_factor,
-			  GLX_DIRECT_COLOR ) ) {
-	fprintf( stderr, "[%s:%u] Error creating FBConfig!\n",
-		 __func__, __LINE__ );
-	return NULL;
+   else {
+      fb_format = GL_BGRA;
+      fb_type = GL_UNSIGNED_INT_8_8_8_8_REV;
+   }
+
+   modes =
+      (*dri_interface->createContextModes) (num_modes,
+                                            sizeof(__GLcontextModes));
+   m = modes;
+   if (!driFillInModes(&m, fb_format, fb_type,
+                       depth_bits_array, stencil_bits_array,
+                       depth_buffer_factor, back_buffer_modes,
+                       back_buffer_factor, GLX_TRUE_COLOR)) {
+      fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__,
+              __LINE__);
+      return NULL;
+   }
+   if (!driFillInModes(&m, fb_format, fb_type,
+                       depth_bits_array, stencil_bits_array,
+                       depth_buffer_factor, back_buffer_modes,
+                       back_buffer_factor, GLX_DIRECT_COLOR)) {
+      fprintf(stderr, "[%s:%u] Error creating FBConfig!\n", __func__,
+              __LINE__);
+      return NULL;
    }
 
    /* Mark the visual as slow if there are "fake" stencil bits.
     */
-   for ( m = modes ; m != NULL ; m = m->next ) {
-      if ( (m->stencilBits != 0) && (m->stencilBits != stencil_bits) ) {
-	 m->visualRating = GLX_SLOW_CONFIG;
+   for (m = modes; m != NULL; m = m->next) {
+      if ((m->stencilBits != 0) && (m->stencilBits != stencil_bits)) {
+         m->visualRating = GLX_SLOW_CONFIG;
       }
    }
 
@@ -630,43 +776,42 @@ intelFillInModes( unsigned pixel_bits, unsigned depth_bits,
  * \return A pointer to a \c __DRIscreenPrivate on success, or \c NULL on 
  *         failure.
  */
-PUBLIC
-void * __driCreateNewScreen_20050727( __DRInativeDisplay *dpy, int scrn, __DRIscreen *psc,
-			     const __GLcontextModes * modes,
-			     const __DRIversion * ddx_version,
-			     const __DRIversion * dri_version,
-			     const __DRIversion * drm_version,
-			     const __DRIframebuffer * frame_buffer,
-			     drmAddress pSAREA, int fd, 
-			     int internal_api_version,
-			     const __DRIinterfaceMethods * interface,
-			     __GLcontextModes ** driver_modes )
-			     
+PUBLIC void *
+__driCreateNewScreen_20050727(__DRInativeDisplay * dpy, int scrn,
+                              __DRIscreen * psc,
+                              const __GLcontextModes * modes,
+                              const __DRIversion * ddx_version,
+                              const __DRIversion * dri_version,
+                              const __DRIversion * drm_version,
+                              const __DRIframebuffer * frame_buffer,
+                              drmAddress pSAREA, int fd,
+                              int internal_api_version,
+                              const __DRIinterfaceMethods * interface,
+                              __GLcontextModes ** driver_modes)
 {
    __DRIscreenPrivate *psp;
    static const __DRIversion ddx_expected = { 1, 5, 0 };
    static const __DRIversion dri_expected = { 4, 0, 0 };
-   static const __DRIversion drm_expected = { 1, 4, 0 };
+   static const __DRIversion drm_expected = { 1, 6, 0 };
 
    dri_interface = interface;
 
-   if ( ! driCheckDriDdxDrmVersions2( "i915",
-				      dri_version, & dri_expected,
-				      ddx_version, & ddx_expected,
-				      drm_version, & drm_expected ) ) {
+   if (!driCheckDriDdxDrmVersions2("i915",
+                                   dri_version, &dri_expected,
+                                   ddx_version, &ddx_expected,
+                                   drm_version, &drm_expected)) {
       return NULL;
    }
 
    psp = __driUtilCreateNewScreen(dpy, scrn, psc, NULL,
-				  ddx_version, dri_version, drm_version,
-				  frame_buffer, pSAREA, fd,
-				  internal_api_version, &intelAPI);
-   if ( psp != NULL ) {
+                                  ddx_version, dri_version, drm_version,
+                                  frame_buffer, pSAREA, fd,
+                                  internal_api_version, &intelAPI);
+   if (psp != NULL) {
       I830DRIPtr dri_priv = (I830DRIPtr) psp->pDevPriv;
-      *driver_modes = intelFillInModes( dri_priv->cpp * 8,
-					(dri_priv->cpp == 2) ? 16 : 24,
-					(dri_priv->cpp == 2) ? 0  : 8,
-					1 );
+      *driver_modes = intelFillInModes(dri_priv->cpp * 8,
+                                       (dri_priv->cpp == 2) ? 16 : 24,
+                                       (dri_priv->cpp == 2) ? 0 : 8, 1);
 
       /* Calling driInitExtensions here, with a NULL context pointer, does not actually
        * enable the extensions.  It just makes sure that all the dispatch offsets for all
@@ -676,8 +821,24 @@ void * __driCreateNewScreen_20050727( __DRInativeDisplay *dpy, int scrn, __DRIsc
        *
        * Hello chicken.  Hello egg.  How are you two today?
        */
-      driInitExtensions( NULL, card_extensions, GL_FALSE );
+      driInitExtensions(NULL, card_extensions, GL_FALSE);
    }
 
    return (void *) psp;
 }
+
+struct intel_context *intelScreenContext(intelScreenPrivate *intelScreen)
+{
+  /*
+   * This should probably change to have the screen allocate a dummy
+   * context at screen creation. For now just use the current context.
+   */
+
+  GET_CURRENT_CONTEXT(ctx);
+  if (ctx == NULL) {
+     _mesa_problem(NULL, "No current context in intelScreenContext\n");
+     return NULL;
+  }
+  return intel_context(ctx);
+}
+

src/mesa/drivers/dri/i915/intel_screen.h

@@ -29,39 +29,48 @@
 #define _INTEL_INIT_H_
 
 #include <sys/time.h>
-#include "xmlconfig.h"
 #include "dri_util.h"
 #include "intel_rotate.h"
 #include "i830_common.h"
+#include "xmlconfig.h"
+#include "dri_bufpool.h"
 
-
-/* This roughly corresponds to a gl_renderbuffer (Mesa 6.4) */
-typedef struct {
+/* XXX: change name or eliminate to avoid conflict with "struct
+ * intel_region"!!!
+ */
+typedef struct
+{
    drm_handle_t handle;
-   drmSize size;        /* region size in bytes */
-   char *map;           /* memory map */
-   int offset;          /* from start of video mem, in bytes */
-   int pitch;           /* row stride, in bytes */
+   drmSize size;                /* region size in bytes */
+   char *map;                   /* memory map */
+   int offset;                  /* from start of video mem, in bytes */
+   int pitch;                   /* row stride, in bytes */
 } intelRegion;
 
-typedef struct 
+typedef struct
 {
    intelRegion front;
    intelRegion back;
    intelRegion rotated;
    intelRegion depth;
    intelRegion tex;
-   
+
+   struct intel_region *front_region;
+   struct intel_region *back_region;
+   struct intel_region *depth_region;
+   struct intel_region *rotated_region;
+
    int deviceID;
    int width;
    int height;
-   int mem;         /* unused */
-   
-   int cpp;         /* for front and back buffers */
-   int fbFormat;
+   int mem;                     /* unused */
+
+   int cpp;                     /* for front and back buffers */
+/*    int bitsPerPixel;   */
+   int fbFormat;                /* XXX FBO: this is obsolete - remove after i830 updates */
 
    int logTextureGranularity;
-   
+
    __DRIscreenPrivate *driScrnPriv;
    unsigned int sarea_priv_offset;
 
@@ -72,41 +81,51 @@ typedef struct
 
    struct matrix23 rotMatrix;
 
-   int current_rotation;  /* 0, 90, 180 or 270 */
+   int current_rotation;        /* 0, 90, 180 or 270 */
    int rotatedWidth, rotatedHeight;
 
    /**
    * Configuration cache with default values for all contexts
    */
    driOptionCache optionCache;
+   struct _DriBufferPool *batchPool;
+   struct _DriBufferPool *texPool;
+   struct _DriBufferPool *regionPool;
+   struct _DriBufferPool *staticPool;
 } intelScreenPrivate;
 
 
-extern GLboolean
-intelMapScreenRegions(__DRIscreenPrivate *sPriv);
+
+extern GLboolean intelMapScreenRegions(__DRIscreenPrivate * sPriv);
+
+extern void intelUnmapScreenRegions(intelScreenPrivate * intelScreen);
 
 extern void
-intelUnmapScreenRegions(intelScreenPrivate *intelScreen);
+intelUpdateScreenFromSAREA(intelScreenPrivate * intelScreen,
+                           drmI830Sarea * sarea);
 
-extern void
-intelUpdateScreenFromSAREA(intelScreenPrivate *intelScreen,
-                           drmI830Sarea *sarea);
+extern void intelDestroyContext(__DRIcontextPrivate * driContextPriv);
 
-extern void
-intelDestroyContext(__DRIcontextPrivate *driContextPriv);
+extern GLboolean intelUnbindContext(__DRIcontextPrivate * driContextPriv);
 
 extern GLboolean
-intelUnbindContext(__DRIcontextPrivate *driContextPriv);
+intelMakeCurrent(__DRIcontextPrivate * driContextPriv,
+                 __DRIdrawablePrivate * driDrawPriv,
+                 __DRIdrawablePrivate * driReadPriv);
 
-extern GLboolean
-intelMakeCurrent(__DRIcontextPrivate *driContextPriv,
-                 __DRIdrawablePrivate *driDrawPriv,
-                 __DRIdrawablePrivate *driReadPriv);
+extern void intelSwapBuffers(__DRIdrawablePrivate * dPriv);
 
 extern void
-intelSwapBuffers(__DRIdrawablePrivate *dPriv);
+intelCopySubBuffer(__DRIdrawablePrivate * dPriv, int x, int y, int w, int h);
+
+extern struct _DriBufferPool *driBatchPoolInit(int fd, unsigned flags,
+                                               unsigned long bufSize,
+                                               unsigned numBufs,
+                                               unsigned checkDelayed);
+
+extern struct intel_context *intelScreenContext(intelScreenPrivate *intelScreen);
 
 extern void
-intelCopySubBuffer( __DRIdrawablePrivate *dPriv, int x, int y, int w, int h );
+intelUpdateScreenRotation(__DRIscreenPrivate * sPriv, drmI830Sarea * sarea);
 
 #endif

src/mesa/drivers/dri/i915/intel_span.c

@@ -30,229 +30,380 @@
 #include "mtypes.h"
 #include "colormac.h"
 
+#include "intel_fbo.h"
 #include "intel_screen.h"
-
 #include "intel_span.h"
+#include "intel_regions.h"
 #include "intel_ioctl.h"
+#include "intel_tex.h"
+
 #include "swrast/swrast.h"
 
+/*
+  break intelWriteRGBASpan_ARGB8888
+*/
 
+#undef DBG
 #define DBG 0
 
-#define LOCAL_VARS						\
-   intelContextPtr intel = INTEL_CONTEXT(ctx);			\
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;		\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   GLuint pitch = drb->pitch;					\
-   GLuint height = dPriv->h;					\
-   char *buf = (char *) drb->Base.Data +			\
-			dPriv->x * drb->cpp +			\
-			dPriv->y * pitch;			\
-   GLushort p;							\
-   (void) buf; (void) p
+#define LOCAL_VARS							\
+   struct intel_context *intel = intel_context(ctx);			\
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);		\
+   const GLint yScale = irb->RenderToTexture ? 1 : -1;			\
+   const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1;	\
+   GLubyte *buf = (GLubyte *) irb->pfMap				\
+      + (intel->drawY * irb->pfPitch + intel->drawX) * irb->region->cpp;\
+   GLuint p;								\
+   assert(irb->pfMap);\
+   (void) p;
 
-#define LOCAL_DEPTH_VARS					\
-   intelContextPtr intel = INTEL_CONTEXT(ctx);			\
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;		\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   GLuint pitch = drb->pitch;					\
-   GLuint height = dPriv->h;					\
-   char *buf = (char *) drb->Base.Data +			\
-			dPriv->x * drb->cpp +			\
-			dPriv->y * pitch
+/* XXX FBO: this is identical to the macro in spantmp2.h except we get
+ * the cliprect info from the context, not the driDrawable.
+ * Move this into spantmp2.h someday.
+ */
+#define HW_CLIPLOOP()							\
+   do {									\
+      int _nc = intel->numClipRects;					\
+      while ( _nc-- ) {							\
+	 int minx = intel->pClipRects[_nc].x1 - intel->drawX;		\
+	 int miny = intel->pClipRects[_nc].y1 - intel->drawY;		\
+	 int maxx = intel->pClipRects[_nc].x2 - intel->drawX;		\
+	 int maxy = intel->pClipRects[_nc].y2 - intel->drawY;
 
-#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS 
 
-#define INIT_MONO_PIXEL(p,color)\
-	 p = INTEL_PACKCOLOR565(color[0],color[1],color[2])
 
-#define Y_FLIP(_y) (height - _y - 1)
+
+#define Y_FLIP(_y) ((_y) * yScale + yBias)
 
 #define HW_LOCK()
 
 #define HW_UNLOCK()
 
-/* 16 bit, 565 rgb color spanline and pixel functions
+/* 16 bit, RGB565 color spanline and pixel functions
  */
-#define WRITE_RGBA( _x, _y, r, g, b, a )				\
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = ( (((int)r & 0xf8) << 8) |	\
-		                             (((int)g & 0xfc) << 3) |	\
-		                             (((int)b & 0xf8) >> 3))
-#define WRITE_PIXEL( _x, _y, p )  \
-   *(GLushort *)(buf + _x*2 + _y*pitch) = p
+#define SPANTMP_PIXEL_FMT GL_RGB
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_SHORT_5_6_5
 
-#define READ_RGBA( rgba, _x, _y )				\
-do {								\
-   GLushort p = *(GLushort *)(buf + _x*2 + _y*pitch);		\
-   rgba[0] = (((p >> 11) & 0x1f) * 255) / 31;			\
-   rgba[1] = (((p >>  5) & 0x3f) * 255) / 63;			\
-   rgba[2] = (((p >>  0) & 0x1f) * 255) / 31;			\
-   rgba[3] = 255;						\
-} while(0)
+#define TAG(x)    intel##x##_RGB565
+#define TAG2(x,y) intel##x##_RGB565##y
+#define GET_PTR(X,Y) (buf + ((Y) * irb->pfPitch + (X)) * 2)
+#include "spantmp2.h"
 
-#define TAG(x) intel##x##_565
-#include "spantmp.h"
-
-/* 15 bit, 555 rgb color spanline and pixel functions
+/* 32 bit, ARGB8888 color spanline and pixel functions
  */
-#define WRITE_RGBA( _x, _y, r, g, b, a )			\
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = (((r & 0xf8) << 7) |	\
-		                            ((g & 0xf8) << 3) |	\
-                         		    ((b & 0xf8) >> 3))
+#define SPANTMP_PIXEL_FMT GL_BGRA
+#define SPANTMP_PIXEL_TYPE GL_UNSIGNED_INT_8_8_8_8_REV
 
-#define WRITE_PIXEL( _x, _y, p )  \
-   *(GLushort *)(buf + _x*2 + _y*pitch)  = p
+#define TAG(x)    intel##x##_ARGB8888
+#define TAG2(x,y) intel##x##_ARGB8888##y
+#define GET_PTR(X,Y) (buf + ((Y) * irb->pfPitch + (X)) * 4)
+#include "spantmp2.h"
 
-#define READ_RGBA( rgba, _x, _y )				\
-do {								\
-   GLushort p = *(GLushort *)(buf + _x*2 + _y*pitch);		\
-   rgba[0] = (p >> 7) & 0xf8;					\
-   rgba[1] = (p >> 3) & 0xf8;					\
-   rgba[2] = (p << 3) & 0xf8;					\
-   rgba[3] = 255;						\
-} while(0)
 
-#define TAG(x) intel##x##_555
-#include "spantmp.h"
+#define LOCAL_DEPTH_VARS						\
+   struct intel_context *intel = intel_context(ctx);			\
+   struct intel_renderbuffer *irb = intel_renderbuffer(rb);		\
+   const GLuint pitch = irb->pfPitch/***XXX region->pitch*/; /* in pixels */ \
+   const GLint yScale = irb->RenderToTexture ? 1 : -1;			\
+   const GLint yBias = irb->RenderToTexture ? 0 : irb->Base.Height - 1;	\
+   char *buf = (char *) irb->pfMap/*XXX use region->map*/ +             \
+      (intel->drawY * pitch + intel->drawX) * irb->region->cpp;
 
-/* 16 bit depthbuffer functions.
- */
+
+#define LOCAL_STENCIL_VARS LOCAL_DEPTH_VARS
+
+/**
+ ** 16-bit depthbuffer functions.
+ **/
 #define WRITE_DEPTH( _x, _y, d ) \
-   *(GLushort *)(buf + (_x)*2 + (_y)*pitch)  = d;
+   ((GLushort *)buf)[(_x) + (_y) * pitch] = d;
 
 #define READ_DEPTH( d, _x, _y )	\
-   d = *(GLushort *)(buf + (_x)*2 + (_y)*pitch);	 
+   d = ((GLushort *)buf)[(_x) + (_y) * pitch];
 
 
 #define TAG(x) intel##x##_z16
 #include "depthtmp.h"
 
 
-#undef LOCAL_VARS
-#define LOCAL_VARS						\
-   intelContextPtr intel = INTEL_CONTEXT(ctx);			\
-   __DRIdrawablePrivate *dPriv = intel->driDrawable;		\
-   driRenderbuffer *drb = (driRenderbuffer *) rb;		\
-   GLuint pitch = drb->pitch;					\
-   GLuint height = dPriv->h;					\
-   char *buf = (char *)drb->Base.Data +				\
-			dPriv->x * drb->cpp +			\
-			dPriv->y * pitch;			\
-   GLuint p;							\
-   (void) buf; (void) p
-
-#undef INIT_MONO_PIXEL
-#define INIT_MONO_PIXEL(p,color)\
-	 p = INTEL_PACKCOLOR8888(color[0],color[1],color[2],color[3])
-
-/* 32 bit, 8888 argb color spanline and pixel functions
- */
-#define WRITE_RGBA(_x, _y, r, g, b, a)			\
-    *(GLuint *)(buf + _x*4 + _y*pitch) = ((r << 16) |	\
-					  (g << 8)  |	\
-					  (b << 0)  |	\
-					  (a << 24) )
-
-#define WRITE_PIXEL(_x, _y, p)			\
-    *(GLuint *)(buf + _x*4 + _y*pitch) = p
-
-
-#define READ_RGBA(rgba, _x, _y)					\
-    do {							\
-	GLuint p = *(GLuint *)(buf + _x*4 + _y*pitch);		\
-	rgba[0] = (p >> 16) & 0xff;				\
-	rgba[1] = (p >> 8)  & 0xff;				\
-	rgba[2] = (p >> 0)  & 0xff;				\
-	rgba[3] = (p >> 24) & 0xff;				\
-    } while (0)
-
-#define TAG(x) intel##x##_8888
-#include "spantmp.h"
-
-
-/* 24/8 bit interleaved depth/stencil functions
- */
-#define WRITE_DEPTH( _x, _y, d ) {			\
-   GLuint tmp = *(GLuint *)(buf + (_x)*4 + (_y)*pitch);	\
-   tmp &= 0xff000000;					\
-   tmp |= (d) & 0xffffff;				\
-   *(GLuint *)(buf + (_x)*4 + (_y)*pitch) = tmp;		\
+/**
+ ** 24/8-bit interleaved depth/stencil functions
+ ** Note: we're actually reading back combined depth+stencil values.
+ ** The wrappers in main/depthstencil.c are used to extract the depth
+ ** and stencil values.
+ **/
+/* Change ZZZS -> SZZZ */
+#define WRITE_DEPTH( _x, _y, d ) {				\
+   GLuint tmp = ((d) >> 8) | ((d) << 24);			\
+   ((GLuint *)buf)[(_x) + (_y) * pitch] = tmp;			\
 }
 
-#define READ_DEPTH( d, _x, _y )		\
-   d = *(GLuint *)(buf + (_x)*4 + (_y)*pitch) & 0xffffff;
-
+/* Change SZZZ -> ZZZS */
+#define READ_DEPTH( d, _x, _y ) {				\
+   GLuint tmp = ((GLuint *)buf)[(_x) + (_y) * pitch];		\
+   d = (tmp << 8) | (tmp >> 24);				\
+}
 
 #define TAG(x) intel##x##_z24_s8
 #include "depthtmp.h"
 
-#define WRITE_STENCIL( _x, _y, d ) {			\
-   GLuint tmp = *(GLuint *)(buf + (_x)*4 + (_y)*pitch);	\
-   tmp &= 0xffffff;					\
-   tmp |= ((d)<<24);					\
-   *(GLuint *)(buf + (_x)*4 + (_y)*pitch) = tmp;		\
+
+/**
+ ** 8-bit stencil function (XXX FBO: This is obsolete)
+ **/
+#define WRITE_STENCIL( _x, _y, d ) {				\
+   GLuint tmp = ((GLuint *)buf)[(_x) + (_y) * pitch];		\
+   tmp &= 0xffffff;						\
+   tmp |= ((d) << 24);						\
+   ((GLuint *) buf)[(_x) + (_y) * pitch] = tmp;			\
 }
 
-#define READ_STENCIL( d, _x, _y )			\
-   d = *(GLuint *)(buf + (_x)*4 + (_y)*pitch) >> 24;
+#define READ_STENCIL( d, _x, _y )				\
+   d = ((GLuint *)buf)[(_x) + (_y) * pitch] >> 24;
 
 #define TAG(x) intel##x##_z24_s8
 #include "stenciltmp.h"
 
 
-/* Move locking out to get reasonable span performance.
+
+/**
+ * Map or unmap all the renderbuffers which we may need during
+ * software rendering.
+ * XXX in the future, we could probably convey extra information to
+ * reduce the number of mappings needed.  I.e. if doing a glReadPixels
+ * from the depth buffer, we really only need one mapping.
+ *
+ * XXX Rewrite this function someday.
+ * We can probably just loop over all the renderbuffer attachments,
+ * map/unmap all of them, and not worry about the _ColorDrawBuffers
+ * _ColorReadBuffer, _DepthBuffer or _StencilBuffer fields.
  */
-void intelSpanRenderStart( GLcontext *ctx )
+static void
+intel_map_unmap_buffers(struct intel_context *intel, GLboolean map)
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   GLcontext *ctx = &intel->ctx;
+   GLuint i, j;
+   struct intel_renderbuffer *irb;
 
-   intelFlush(&intel->ctx);
+   /* color draw buffers */
+   for (i = 0; i < ctx->Const.MaxDrawBuffers; i++) {
+      for (j = 0; j < ctx->DrawBuffer->_NumColorDrawBuffers[i]; j++) {
+         struct gl_renderbuffer *rb =
+            ctx->DrawBuffer->_ColorDrawBuffers[i][j];
+         irb = intel_renderbuffer(rb);
+         if (irb) {
+            /* this is a user-created intel_renderbuffer */
+            if (irb->region) {
+               if (map)
+                  intel_region_map(intel->intelScreen, irb->region);
+               else
+                  intel_region_unmap(intel->intelScreen, irb->region);
+            }
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+      }
+   }
+
+   /* check for render to textures */
+   for (i = 0; i < BUFFER_COUNT; i++) {
+      struct gl_renderbuffer_attachment *att =
+         ctx->DrawBuffer->Attachment + i;
+      struct gl_texture_object *tex = att->Texture;
+      if (tex) {
+         /* render to texture */
+         ASSERT(att->Renderbuffer);
+         if (map) {
+            struct gl_texture_image *texImg;
+            texImg = tex->Image[att->CubeMapFace][att->TextureLevel];
+            intel_tex_map_images(intel, intel_texture_object(tex));
+         }
+         else {
+            intel_tex_unmap_images(intel, intel_texture_object(tex));
+         }
+      }
+   }
+
+   /* color read buffers */
+   irb = intel_renderbuffer(ctx->ReadBuffer->_ColorReadBuffer);
+   if (irb && irb->region) {
+      if (map)
+         intel_region_map(intel->intelScreen, irb->region);
+      else
+         intel_region_unmap(intel->intelScreen, irb->region);
+      irb->pfMap = irb->region->map;
+      irb->pfPitch = irb->region->pitch;
+   }
+
+   /* Account for front/back color page flipping.
+    * The span routines use the pfMap and pfPitch fields which will
+    * swap the front/back region map/pitch if we're page flipped.
+    * Do this after mapping, above, so the map field is valid.
+    */
+#if 0
+   if (map && ctx->DrawBuffer->Name == 0) {
+      struct intel_renderbuffer *irbFront
+         = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_FRONT_LEFT);
+      struct intel_renderbuffer *irbBack
+         = intel_get_renderbuffer(ctx->DrawBuffer, BUFFER_BACK_LEFT);
+      if (irbBack) {
+         /* double buffered */
+         if (intel->sarea->pf_current_page == 0) {
+            irbFront->pfMap = irbFront->region->map;
+            irbFront->pfPitch = irbFront->region->pitch;
+            irbBack->pfMap = irbBack->region->map;
+            irbBack->pfPitch = irbBack->region->pitch;
+         }
+         else {
+            irbFront->pfMap = irbBack->region->map;
+            irbFront->pfPitch = irbBack->region->pitch;
+            irbBack->pfMap = irbFront->region->map;
+            irbBack->pfPitch = irbFront->region->pitch;
+         }
+      }
+   }
+#endif
+
+   /* depth buffer (Note wrapper!) */
+   if (ctx->DrawBuffer->_DepthBuffer) {
+      irb = intel_renderbuffer(ctx->DrawBuffer->_DepthBuffer->Wrapped);
+      if (irb && irb->region && irb->Base.Name != 0) {
+         if (map) {
+            intel_region_map(intel->intelScreen, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel->intelScreen, irb->region);
+            irb->pfMap = NULL;
+            irb->pfPitch = 0;
+         }
+      }
+   }
+
+   /* stencil buffer (Note wrapper!) */
+   if (ctx->DrawBuffer->_StencilBuffer) {
+      irb = intel_renderbuffer(ctx->DrawBuffer->_StencilBuffer->Wrapped);
+      if (irb && irb->region && irb->Base.Name != 0) {
+         if (map) {
+            intel_region_map(intel->intelScreen, irb->region);
+            irb->pfMap = irb->region->map;
+            irb->pfPitch = irb->region->pitch;
+         }
+         else {
+            intel_region_unmap(intel->intelScreen, irb->region);
+            irb->pfMap = NULL;
+            irb->pfPitch = 0;
+         }
+      }
+   }
+}
+
+
+
+/**
+ * Prepare for softare rendering.  Map current read/draw framebuffers'
+ * renderbuffes and all currently bound texture objects.
+ *
+ * Old note: Moved locking out to get reasonable span performance.
+ */
+void
+intelSpanRenderStart(GLcontext * ctx)
+{
+   struct intel_context *intel = intel_context(ctx);
+   GLuint i;
+
+   intelFinish(&intel->ctx);
    LOCK_HARDWARE(intel);
-   intelWaitForIdle(intel);
+
+#if 0
+   /* Just map the framebuffer and all textures.  Bufmgr code will
+    * take care of waiting on the necessary fences:
+    */
+   intel_region_map(intel->intelScreen, intel->front_region);
+   intel_region_map(intel->intelScreen, intel->back_region);
+   intel_region_map(intel->intelScreen, intel->intelScreen->depth_region);
+#endif
+
+   for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
+      if (ctx->Texture.Unit[i]._ReallyEnabled) {
+         struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
+         intel_tex_map_images(intel, intel_texture_object(texObj));
+      }
+   }
+
+   intel_map_unmap_buffers(intel, GL_TRUE);
 }
 
-void intelSpanRenderFinish( GLcontext *ctx )
+/**
+ * Called when done softare rendering.  Unmap the buffers we mapped in
+ * the above function.
+ */
+void
+intelSpanRenderFinish(GLcontext * ctx)
 {
-   intelContextPtr intel = INTEL_CONTEXT( ctx );
-   _swrast_flush( ctx );
-   UNLOCK_HARDWARE( intel );
+   struct intel_context *intel = intel_context(ctx);
+   GLuint i;
+
+   _swrast_flush(ctx);
+
+   /* Now unmap the framebuffer:
+    */
+#if 0
+   intel_region_unmap(intel, intel->front_region);
+   intel_region_unmap(intel, intel->back_region);
+   intel_region_unmap(intel, intel->intelScreen->depth_region);
+#endif
+
+   for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
+      if (ctx->Texture.Unit[i]._ReallyEnabled) {
+         struct gl_texture_object *texObj = ctx->Texture.Unit[i]._Current;
+         intel_tex_unmap_images(intel, intel_texture_object(texObj));
+      }
+   }
+
+   intel_map_unmap_buffers(intel, GL_FALSE);
+
+   UNLOCK_HARDWARE(intel);
 }
 
-void intelInitSpanFuncs( GLcontext *ctx )
+
+void
+intelInitSpanFuncs(GLcontext * ctx)
 {
    struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
    swdd->SpanRenderStart = intelSpanRenderStart;
-   swdd->SpanRenderFinish = intelSpanRenderFinish; 
+   swdd->SpanRenderFinish = intelSpanRenderFinish;
 }
 
 
 /**
- * Plug in the Get/Put routines for the given driRenderbuffer.
+ * Plug in appropriate span read/write functions for the given renderbuffer.
+ * These are used for the software fallbacks.
  */
 void
-intelSetSpanFunctions(driRenderbuffer *drb, const GLvisual *vis)
+intel_set_span_functions(struct gl_renderbuffer *rb)
 {
-   if (drb->Base.InternalFormat == GL_RGBA) {
-      if (vis->redBits == 5 && vis->greenBits == 5 && vis->blueBits == 5) {
-         intelInitPointers_555(&drb->Base);
-      }
-      else if (vis->redBits == 5 && vis->greenBits == 6 && vis->blueBits == 5) {
-         intelInitPointers_565(&drb->Base);
-      }
-      else {
-         assert(vis->redBits == 8);
-         assert(vis->greenBits == 8);
-         assert(vis->blueBits == 8);
-         intelInitPointers_8888(&drb->Base);
-      }
+   if (rb->_ActualFormat == GL_RGB5) {
+      /* 565 RGB */
+      intelInitPointers_RGB565(rb);
    }
-   else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT16) {
-      intelInitDepthPointers_z16(&drb->Base);
+   else if (rb->_ActualFormat == GL_RGBA8) {
+      /* 8888 RGBA */
+      intelInitPointers_ARGB8888(rb);
    }
-   else if (drb->Base.InternalFormat == GL_DEPTH_COMPONENT24) {
-      intelInitDepthPointers_z24_s8(&drb->Base);
+   else if (rb->_ActualFormat == GL_DEPTH_COMPONENT16) {
+      intelInitDepthPointers_z16(rb);
    }
-   else if (drb->Base.InternalFormat == GL_STENCIL_INDEX8_EXT) {
-      intelInitStencilPointers_z24_s8(&drb->Base);
+   else if (rb->_ActualFormat == GL_DEPTH_COMPONENT24 ||        /* XXX FBO remove */
+            rb->_ActualFormat == GL_DEPTH24_STENCIL8_EXT) {
+      intelInitDepthPointers_z24_s8(rb);
+   }
+   else if (rb->_ActualFormat == GL_STENCIL_INDEX8_EXT) {       /* XXX FBO remove */
+      intelInitStencilPointers_z24_s8(rb);
+   }
+   else {
+      _mesa_problem(NULL,
+                    "Unexpected _ActualFormat in intelSetSpanFunctions");
    }
 }

src/mesa/drivers/dri/i915/intel_span.h

@@ -28,14 +28,11 @@
 #ifndef _INTEL_SPAN_H
 #define _INTEL_SPAN_H
 
-#include "drirenderbuffer.h"
+extern void intelInitSpanFuncs(GLcontext * ctx);
 
-extern void intelInitSpanFuncs( GLcontext *ctx );
+extern void intelSpanRenderFinish(GLcontext * ctx);
+extern void intelSpanRenderStart(GLcontext * ctx);
 
-extern void intelSpanRenderFinish( GLcontext *ctx );
-extern void intelSpanRenderStart( GLcontext *ctx );
-
-extern void
-intelSetSpanFunctions(driRenderbuffer *rb, const GLvisual *vis);
+extern void intel_set_span_functions(struct gl_renderbuffer *rb);
 
 #endif

src/mesa/drivers/dri/i915/intel_state.c

@@ -30,252 +30,334 @@
 #include "context.h"
 #include "macros.h"
 #include "enums.h"
+#include "colormac.h"
 #include "dd.h"
 
 #include "intel_screen.h"
 #include "intel_context.h"
+#include "intel_fbo.h"
+#include "intel_regions.h"
 #include "swrast/swrast.h"
 
-int intel_translate_compare_func( GLenum func )
+int
+intel_translate_compare_func(GLenum func)
 {
-   switch(func) {
-   case GL_NEVER: 
-      return COMPAREFUNC_NEVER; 
-   case GL_LESS: 
-      return COMPAREFUNC_LESS; 
-   case GL_LEQUAL: 
-      return COMPAREFUNC_LEQUAL; 
-   case GL_GREATER: 
-      return COMPAREFUNC_GREATER; 
-   case GL_GEQUAL: 
-      return COMPAREFUNC_GEQUAL; 
-   case GL_NOTEQUAL: 
-      return COMPAREFUNC_NOTEQUAL; 
-   case GL_EQUAL: 
-      return COMPAREFUNC_EQUAL; 
-   case GL_ALWAYS: 
-      return COMPAREFUNC_ALWAYS; 
+   switch (func) {
+   case GL_NEVER:
+      return COMPAREFUNC_NEVER;
+   case GL_LESS:
+      return COMPAREFUNC_LESS;
+   case GL_LEQUAL:
+      return COMPAREFUNC_LEQUAL;
+   case GL_GREATER:
+      return COMPAREFUNC_GREATER;
+   case GL_GEQUAL:
+      return COMPAREFUNC_GEQUAL;
+   case GL_NOTEQUAL:
+      return COMPAREFUNC_NOTEQUAL;
+   case GL_EQUAL:
+      return COMPAREFUNC_EQUAL;
+   case GL_ALWAYS:
+      return COMPAREFUNC_ALWAYS;
    }
 
    fprintf(stderr, "Unknown value in %s: %x\n", __FUNCTION__, func);
-   return COMPAREFUNC_ALWAYS; 
+   return COMPAREFUNC_ALWAYS;
 }
 
-int intel_translate_stencil_op( GLenum op )
+int
+intel_translate_stencil_op(GLenum op)
 {
-   switch(op) {
-   case GL_KEEP: 
-      return STENCILOP_KEEP; 
-   case GL_ZERO: 
-      return STENCILOP_ZERO; 
-   case GL_REPLACE: 
-      return STENCILOP_REPLACE; 
-   case GL_INCR: 
+   switch (op) {
+   case GL_KEEP:
+      return STENCILOP_KEEP;
+   case GL_ZERO:
+      return STENCILOP_ZERO;
+   case GL_REPLACE:
+      return STENCILOP_REPLACE;
+   case GL_INCR:
       return STENCILOP_INCRSAT;
-   case GL_DECR: 
+   case GL_DECR:
       return STENCILOP_DECRSAT;
    case GL_INCR_WRAP:
-      return STENCILOP_INCR; 
+      return STENCILOP_INCR;
    case GL_DECR_WRAP:
-      return STENCILOP_DECR; 
-   case GL_INVERT: 
-      return STENCILOP_INVERT; 
-   default: 
+      return STENCILOP_DECR;
+   case GL_INVERT:
+      return STENCILOP_INVERT;
+   default:
       return STENCILOP_ZERO;
    }
 }
 
-int intel_translate_blend_factor( GLenum factor )
+int
+intel_translate_blend_factor(GLenum factor)
 {
-   switch(factor) {
-   case GL_ZERO: 
-      return BLENDFACT_ZERO; 
-   case GL_SRC_ALPHA: 
-      return BLENDFACT_SRC_ALPHA; 
-   case GL_ONE: 
-      return BLENDFACT_ONE; 
-   case GL_SRC_COLOR: 
-      return BLENDFACT_SRC_COLR; 
-   case GL_ONE_MINUS_SRC_COLOR: 
-      return BLENDFACT_INV_SRC_COLR; 
-   case GL_DST_COLOR: 
-      return BLENDFACT_DST_COLR; 
-   case GL_ONE_MINUS_DST_COLOR: 
-      return BLENDFACT_INV_DST_COLR; 
+   switch (factor) {
+   case GL_ZERO:
+      return BLENDFACT_ZERO;
+   case GL_SRC_ALPHA:
+      return BLENDFACT_SRC_ALPHA;
+   case GL_ONE:
+      return BLENDFACT_ONE;
+   case GL_SRC_COLOR:
+      return BLENDFACT_SRC_COLR;
+   case GL_ONE_MINUS_SRC_COLOR:
+      return BLENDFACT_INV_SRC_COLR;
+   case GL_DST_COLOR:
+      return BLENDFACT_DST_COLR;
+   case GL_ONE_MINUS_DST_COLOR:
+      return BLENDFACT_INV_DST_COLR;
    case GL_ONE_MINUS_SRC_ALPHA:
-      return BLENDFACT_INV_SRC_ALPHA; 
-   case GL_DST_ALPHA: 
-      return BLENDFACT_DST_ALPHA; 
+      return BLENDFACT_INV_SRC_ALPHA;
+   case GL_DST_ALPHA:
+      return BLENDFACT_DST_ALPHA;
    case GL_ONE_MINUS_DST_ALPHA:
-      return BLENDFACT_INV_DST_ALPHA; 
-   case GL_SRC_ALPHA_SATURATE: 
+      return BLENDFACT_INV_DST_ALPHA;
+   case GL_SRC_ALPHA_SATURATE:
       return BLENDFACT_SRC_ALPHA_SATURATE;
    case GL_CONSTANT_COLOR:
-      return BLENDFACT_CONST_COLOR; 
+      return BLENDFACT_CONST_COLOR;
    case GL_ONE_MINUS_CONSTANT_COLOR:
       return BLENDFACT_INV_CONST_COLOR;
    case GL_CONSTANT_ALPHA:
-      return BLENDFACT_CONST_ALPHA; 
+      return BLENDFACT_CONST_ALPHA;
    case GL_ONE_MINUS_CONSTANT_ALPHA:
       return BLENDFACT_INV_CONST_ALPHA;
    }
-   
+
    fprintf(stderr, "Unknown value in %s: %x\n", __FUNCTION__, factor);
    return BLENDFACT_ZERO;
 }
 
-int intel_translate_logic_op( GLenum opcode )
+int
+intel_translate_logic_op(GLenum opcode)
 {
-   switch(opcode) {
-   case GL_CLEAR: 
-      return LOGICOP_CLEAR; 
-   case GL_AND: 
-      return LOGICOP_AND; 
-   case GL_AND_REVERSE: 
-      return LOGICOP_AND_RVRSE; 
-   case GL_COPY: 
-      return LOGICOP_COPY; 
-   case GL_COPY_INVERTED: 
-      return LOGICOP_COPY_INV; 
-   case GL_AND_INVERTED: 
-      return LOGICOP_AND_INV; 
-   case GL_NOOP: 
-      return LOGICOP_NOOP; 
-   case GL_XOR: 
-      return LOGICOP_XOR; 
-   case GL_OR: 
-      return LOGICOP_OR; 
-   case GL_OR_INVERTED: 
-      return LOGICOP_OR_INV; 
-   case GL_NOR: 
-      return LOGICOP_NOR; 
-   case GL_EQUIV: 
-      return LOGICOP_EQUIV; 
-   case GL_INVERT: 
-      return LOGICOP_INV; 
-   case GL_OR_REVERSE: 
-      return LOGICOP_OR_RVRSE; 
-   case GL_NAND: 
-      return LOGICOP_NAND; 
-   case GL_SET: 
-      return LOGICOP_SET; 
+   switch (opcode) {
+   case GL_CLEAR:
+      return LOGICOP_CLEAR;
+   case GL_AND:
+      return LOGICOP_AND;
+   case GL_AND_REVERSE:
+      return LOGICOP_AND_RVRSE;
+   case GL_COPY:
+      return LOGICOP_COPY;
+   case GL_COPY_INVERTED:
+      return LOGICOP_COPY_INV;
+   case GL_AND_INVERTED:
+      return LOGICOP_AND_INV;
+   case GL_NOOP:
+      return LOGICOP_NOOP;
+   case GL_XOR:
+      return LOGICOP_XOR;
+   case GL_OR:
+      return LOGICOP_OR;
+   case GL_OR_INVERTED:
+      return LOGICOP_OR_INV;
+   case GL_NOR:
+      return LOGICOP_NOR;
+   case GL_EQUIV:
+      return LOGICOP_EQUIV;
+   case GL_INVERT:
+      return LOGICOP_INV;
+   case GL_OR_REVERSE:
+      return LOGICOP_OR_RVRSE;
+   case GL_NAND:
+      return LOGICOP_NAND;
+   case GL_SET:
+      return LOGICOP_SET;
    default:
       return LOGICOP_SET;
    }
 }
 
-static void intelDrawBuffer(GLcontext *ctx, GLenum mode )
+
+static void
+intelClearColor(GLcontext * ctx, const GLfloat color[4])
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   int front = 0;
- 
-   if (!ctx->DrawBuffer)
-      return;
+   struct intel_context *intel = intel_context(ctx);
+   GLubyte clear[4];
 
-   switch ( ctx->DrawBuffer->_ColorDrawBufferMask[0] ) {
-   case BUFFER_BIT_FRONT_LEFT:
-      front = 1;
-      FALLBACK( intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE );
-      break;
-   case BUFFER_BIT_BACK_LEFT:
-      front = 0;
-      FALLBACK( intel, INTEL_FALLBACK_DRAW_BUFFER, GL_FALSE );
-      break;
-   default:
-      FALLBACK( intel, INTEL_FALLBACK_DRAW_BUFFER, GL_TRUE );
-      return;
-   }
+   CLAMPED_FLOAT_TO_UBYTE(clear[0], color[0]);
+   CLAMPED_FLOAT_TO_UBYTE(clear[1], color[1]);
+   CLAMPED_FLOAT_TO_UBYTE(clear[2], color[2]);
+   CLAMPED_FLOAT_TO_UBYTE(clear[3], color[3]);
 
-   if ( intel->sarea->pf_current_page == 1 ) 
-      front ^= 1;
-   
-   intelSetFrontClipRects( intel );
-
-   if (front) {
-      intel->drawRegion = &intel->intelScreen->front;
-      intel->readRegion = &intel->intelScreen->front;
-   } else {
-      intel->drawRegion = &intel->intelScreen->back;
-      intel->readRegion = &intel->intelScreen->back;
-   }
-
-   intel->vtbl.set_color_region( intel, intel->drawRegion );
-}
-
-static void intelReadBuffer( GLcontext *ctx, GLenum mode )
-{
-   /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
+   /* compute both 32 and 16-bit clear values */
+   intel->ClearColor8888 = INTEL_PACKCOLOR8888(clear[0], clear[1],
+                                               clear[2], clear[3]);
+   intel->ClearColor565 = INTEL_PACKCOLOR565(clear[0], clear[1], clear[2]);
 }
 
 
-static void intelClearColor(GLcontext *ctx, const GLfloat color[4])
+/**
+ * Update the viewport transformation matrix.  Depends on:
+ *  - viewport pos/size
+ *  - depthrange
+ *  - window pos/size or FBO size
+ */
+static void
+intelCalcViewport(GLcontext * ctx)
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   intelScreenPrivate *screen = intel->intelScreen;
-
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_red, color[0]);
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_green, color[1]);
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_blue, color[2]);
-   CLAMPED_FLOAT_TO_UBYTE(intel->clear_alpha, color[3]);
-
-   intel->ClearColor = INTEL_PACKCOLOR(screen->fbFormat,
-				       intel->clear_red, 
-				       intel->clear_green, 
-				       intel->clear_blue, 
-				       intel->clear_alpha);
-}
-
-
-static void intelCalcViewport( GLcontext *ctx )
-{
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
+   struct intel_context *intel = intel_context(ctx);
    const GLfloat *v = ctx->Viewport._WindowMap.m;
+   const GLfloat depthScale = 1.0F / ctx->DrawBuffer->_DepthMaxF;
    GLfloat *m = intel->ViewportMatrix.m;
-   GLint h = 0;
+   GLfloat yScale, yBias;
 
-   if (intel->driDrawable) 
-      h = intel->driDrawable->h + SUBPIXEL_Y;
+   if (ctx->DrawBuffer->Name) {
+      /* User created FBO */
+      struct intel_renderbuffer *irb
+         = intel_renderbuffer(ctx->DrawBuffer->_ColorDrawBuffers[0][0]);
+      if (irb && !irb->RenderToTexture) {
+         /* y=0=top */
+         yScale = -1.0;
+         yBias = irb->Base.Height;
+      }
+      else {
+         /* y=0=bottom */
+         yScale = 1.0;
+         yBias = 0.0;
+      }
+   }
+   else {
+      /* window buffer, y=0=top */
+      yScale = -1.0;
+      yBias = (intel->driDrawable) ? intel->driDrawable->h : 0.0F;
+   }
 
-   /* See also intel_translate_vertex.  SUBPIXEL adjustments can be done
-    * via state vars, too.
-    */
-   m[MAT_SX] =   v[MAT_SX];
-   m[MAT_TX] =   v[MAT_TX] + SUBPIXEL_X;
-   m[MAT_SY] = - v[MAT_SY];
-   m[MAT_TY] = - v[MAT_TY] + h;
-   m[MAT_SZ] =   v[MAT_SZ] * intel->depth_scale;
-   m[MAT_TZ] =   v[MAT_TZ] * intel->depth_scale;
+   m[MAT_SX] = v[MAT_SX];
+   m[MAT_TX] = v[MAT_TX] + SUBPIXEL_X;
+
+   m[MAT_SY] = v[MAT_SY] * yScale;
+   m[MAT_TY] = v[MAT_TY] * yScale + yBias + SUBPIXEL_Y;
+
+   m[MAT_SZ] = v[MAT_SZ] * depthScale;
+   m[MAT_TZ] = v[MAT_TZ] * depthScale;
 }
 
-static void intelViewport( GLcontext *ctx,
-			  GLint x, GLint y,
-			  GLsizei width, GLsizei height )
+static void
+intelViewport(GLcontext * ctx,
+              GLint x, GLint y, GLsizei width, GLsizei height)
 {
-   intelCalcViewport( ctx );
+   intelCalcViewport(ctx);
 }
 
-static void intelDepthRange( GLcontext *ctx,
-			    GLclampd nearval, GLclampd farval )
+static void
+intelDepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval)
 {
-   intelCalcViewport( ctx );
+   intelCalcViewport(ctx);
 }
 
 /* Fallback to swrast for select and feedback.
  */
-static void intelRenderMode( GLcontext *ctx, GLenum mode )
+static void
+intelRenderMode(GLcontext * ctx, GLenum mode)
 {
-   intelContextPtr intel = INTEL_CONTEXT(ctx);
-   FALLBACK( intel, INTEL_FALLBACK_RENDERMODE, (mode != GL_RENDER) );
+   struct intel_context *intel = intel_context(ctx);
+   FALLBACK(intel, INTEL_FALLBACK_RENDERMODE, (mode != GL_RENDER));
 }
 
 
-void intelInitStateFuncs( struct dd_function_table *functions )
+void
+intelInitStateFuncs(struct dd_function_table *functions)
 {
-   functions->DrawBuffer = intelDrawBuffer;
-   functions->ReadBuffer = intelReadBuffer;
    functions->RenderMode = intelRenderMode;
    functions->Viewport = intelViewport;
    functions->DepthRange = intelDepthRange;
    functions->ClearColor = intelClearColor;
 }
 
+
+
+
+void
+intelInitState(GLcontext * ctx)
+{
+   /* Mesa should do this for us:
+    */
+   ctx->Driver.AlphaFunc(ctx, ctx->Color.AlphaFunc, ctx->Color.AlphaRef);
+
+   ctx->Driver.BlendColor(ctx, ctx->Color.BlendColor);
+
+   ctx->Driver.BlendEquationSeparate(ctx,
+                                     ctx->Color.BlendEquationRGB,
+                                     ctx->Color.BlendEquationA);
+
+   ctx->Driver.BlendFuncSeparate(ctx,
+                                 ctx->Color.BlendSrcRGB,
+                                 ctx->Color.BlendDstRGB,
+                                 ctx->Color.BlendSrcA, ctx->Color.BlendDstA);
+
+   ctx->Driver.ColorMask(ctx,
+                         ctx->Color.ColorMask[RCOMP],
+                         ctx->Color.ColorMask[GCOMP],
+                         ctx->Color.ColorMask[BCOMP],
+                         ctx->Color.ColorMask[ACOMP]);
+
+   ctx->Driver.CullFace(ctx, ctx->Polygon.CullFaceMode);
+   ctx->Driver.DepthFunc(ctx, ctx->Depth.Func);
+   ctx->Driver.DepthMask(ctx, ctx->Depth.Mask);
+
+   ctx->Driver.Enable(ctx, GL_ALPHA_TEST, ctx->Color.AlphaEnabled);
+   ctx->Driver.Enable(ctx, GL_BLEND, ctx->Color.BlendEnabled);
+   ctx->Driver.Enable(ctx, GL_COLOR_LOGIC_OP, ctx->Color.ColorLogicOpEnabled);
+   ctx->Driver.Enable(ctx, GL_COLOR_SUM, ctx->Fog.ColorSumEnabled);
+   ctx->Driver.Enable(ctx, GL_CULL_FACE, ctx->Polygon.CullFlag);
+   ctx->Driver.Enable(ctx, GL_DEPTH_TEST, ctx->Depth.Test);
+   ctx->Driver.Enable(ctx, GL_DITHER, ctx->Color.DitherFlag);
+   ctx->Driver.Enable(ctx, GL_FOG, ctx->Fog.Enabled);
+   ctx->Driver.Enable(ctx, GL_LIGHTING, ctx->Light.Enabled);
+   ctx->Driver.Enable(ctx, GL_LINE_SMOOTH, ctx->Line.SmoothFlag);
+   ctx->Driver.Enable(ctx, GL_POLYGON_STIPPLE, ctx->Polygon.StippleFlag);
+   ctx->Driver.Enable(ctx, GL_SCISSOR_TEST, ctx->Scissor.Enabled);
+   ctx->Driver.Enable(ctx, GL_STENCIL_TEST, ctx->Stencil.Enabled);
+   ctx->Driver.Enable(ctx, GL_TEXTURE_1D, GL_FALSE);
+   ctx->Driver.Enable(ctx, GL_TEXTURE_2D, GL_FALSE);
+   ctx->Driver.Enable(ctx, GL_TEXTURE_RECTANGLE_NV, GL_FALSE);
+   ctx->Driver.Enable(ctx, GL_TEXTURE_3D, GL_FALSE);
+   ctx->Driver.Enable(ctx, GL_TEXTURE_CUBE_MAP, GL_FALSE);
+
+   ctx->Driver.Fogfv(ctx, GL_FOG_COLOR, ctx->Fog.Color);
+   ctx->Driver.Fogfv(ctx, GL_FOG_MODE, 0);
+   ctx->Driver.Fogfv(ctx, GL_FOG_DENSITY, &ctx->Fog.Density);
+   ctx->Driver.Fogfv(ctx, GL_FOG_START, &ctx->Fog.Start);
+   ctx->Driver.Fogfv(ctx, GL_FOG_END, &ctx->Fog.End);
+
+   ctx->Driver.FrontFace(ctx, ctx->Polygon.FrontFace);
+
+   {
+      GLfloat f = (GLfloat) ctx->Light.Model.ColorControl;
+      ctx->Driver.LightModelfv(ctx, GL_LIGHT_MODEL_COLOR_CONTROL, &f);
+   }
+
+   ctx->Driver.LineWidth(ctx, ctx->Line.Width);
+   ctx->Driver.LogicOpcode(ctx, ctx->Color.LogicOp);
+   ctx->Driver.PointSize(ctx, ctx->Point.Size);
+   ctx->Driver.PolygonStipple(ctx, (const GLubyte *) ctx->PolygonStipple);
+   ctx->Driver.Scissor(ctx, ctx->Scissor.X, ctx->Scissor.Y,
+                       ctx->Scissor.Width, ctx->Scissor.Height);
+   ctx->Driver.ShadeModel(ctx, ctx->Light.ShadeModel);
+   ctx->Driver.StencilFuncSeparate(ctx, GL_FRONT,
+                                   ctx->Stencil.Function[0],
+                                   ctx->Stencil.Ref[0],
+                                   ctx->Stencil.ValueMask[0]);
+   ctx->Driver.StencilFuncSeparate(ctx, GL_BACK,
+                                   ctx->Stencil.Function[1],
+                                   ctx->Stencil.Ref[1],
+                                   ctx->Stencil.ValueMask[1]);
+   ctx->Driver.StencilMaskSeparate(ctx, GL_FRONT, ctx->Stencil.WriteMask[0]);
+   ctx->Driver.StencilMaskSeparate(ctx, GL_BACK, ctx->Stencil.WriteMask[1]);
+   ctx->Driver.StencilOpSeparate(ctx, GL_FRONT,
+                                 ctx->Stencil.FailFunc[0],
+                                 ctx->Stencil.ZFailFunc[0],
+                                 ctx->Stencil.ZPassFunc[0]);
+   ctx->Driver.StencilOpSeparate(ctx, GL_BACK,
+                                 ctx->Stencil.FailFunc[1],
+                                 ctx->Stencil.ZFailFunc[1],
+                                 ctx->Stencil.ZPassFunc[1]);
+
+
+   /* XXX this isn't really needed */
+   ctx->Driver.DrawBuffer(ctx, ctx->Color.DrawBuffer[0]);
+}

src/mesa/drivers/dri/i915/intel_tex.h

@@ -33,13 +33,102 @@
 #include "texmem.h"
 
 
-void intelInitTextureFuncs( struct dd_function_table *functions );
+void intelInitTextureFuncs(struct dd_function_table *functions);
 
-void intelDestroyTexObj( intelContextPtr intel, intelTextureObjectPtr t );
-int intelUploadTexImages( intelContextPtr intel, intelTextureObjectPtr t,
-			  GLuint face );
+const struct gl_texture_format *intelChooseTextureFormat(GLcontext * ctx,
+                                                         GLint internalFormat,
+                                                         GLenum format,
+                                                         GLenum type);
+
+
+void intelTexImage3D(GLcontext * ctx,
+                     GLenum target, GLint level,
+                     GLint internalFormat,
+                     GLint width, GLint height, GLint depth,
+                     GLint border,
+                     GLenum format, GLenum type, const void *pixels,
+                     const struct gl_pixelstore_attrib *packing,
+                     struct gl_texture_object *texObj,
+                     struct gl_texture_image *texImage);
+
+void intelTexSubImage3D(GLcontext * ctx,
+                        GLenum target,
+                        GLint level,
+                        GLint xoffset, GLint yoffset, GLint zoffset,
+                        GLsizei width, GLsizei height, GLsizei depth,
+                        GLenum format, GLenum type,
+                        const GLvoid * pixels,
+                        const struct gl_pixelstore_attrib *packing,
+                        struct gl_texture_object *texObj,
+                        struct gl_texture_image *texImage);
+
+void intelTexImage2D(GLcontext * ctx,
+                     GLenum target, GLint level,
+                     GLint internalFormat,
+                     GLint width, GLint height, GLint border,
+                     GLenum format, GLenum type, const void *pixels,
+                     const struct gl_pixelstore_attrib *packing,
+                     struct gl_texture_object *texObj,
+                     struct gl_texture_image *texImage);
+
+void intelTexSubImage2D(GLcontext * ctx,
+                        GLenum target,
+                        GLint level,
+                        GLint xoffset, GLint yoffset,
+                        GLsizei width, GLsizei height,
+                        GLenum format, GLenum type,
+                        const GLvoid * pixels,
+                        const struct gl_pixelstore_attrib *packing,
+                        struct gl_texture_object *texObj,
+                        struct gl_texture_image *texImage);
+
+void intelTexImage1D(GLcontext * ctx,
+                     GLenum target, GLint level,
+                     GLint internalFormat,
+                     GLint width, GLint border,
+                     GLenum format, GLenum type, const void *pixels,
+                     const struct gl_pixelstore_attrib *packing,
+                     struct gl_texture_object *texObj,
+                     struct gl_texture_image *texImage);
+
+void intelTexSubImage1D(GLcontext * ctx,
+                        GLenum target,
+                        GLint level,
+                        GLint xoffset,
+                        GLsizei width,
+                        GLenum format, GLenum type,
+                        const GLvoid * pixels,
+                        const struct gl_pixelstore_attrib *packing,
+                        struct gl_texture_object *texObj,
+                        struct gl_texture_image *texImage);
+
+void intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level,
+                         GLenum internalFormat,
+                         GLint x, GLint y, GLsizei width, GLint border);
+
+void intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level,
+                         GLenum internalFormat,
+                         GLint x, GLint y, GLsizei width, GLsizei height,
+                         GLint border);
+
+void intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level,
+                            GLint xoffset, GLint x, GLint y, GLsizei width);
+
+void intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
+                            GLint xoffset, GLint yoffset,
+                            GLint x, GLint y, GLsizei width, GLsizei height);
+
+void intelGetTexImage(GLcontext * ctx, GLenum target, GLint level,
+                      GLenum format, GLenum type, GLvoid * pixels,
+                      struct gl_texture_object *texObj,
+                      struct gl_texture_image *texImage);
+
+GLuint intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit);
+
+void intel_tex_map_images(struct intel_context *intel,
+                          struct intel_texture_object *intelObj);
+
+void intel_tex_unmap_images(struct intel_context *intel,
+                            struct intel_texture_object *intelObj);
 
-GLboolean 
-intel_driReinitTextureHeap( driTexHeap *heap,
-			    unsigned size );
 #endif

src/mesa/drivers/dri/i915/intel_tex_copy.c

@@ -0,0 +1,291 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "mtypes.h"
+#include "enums.h"
+#include "image.h"
+#include "teximage.h"
+#include "swrast/swrast.h"
+
+#include "intel_screen.h"
+#include "intel_context.h"
+#include "intel_batchbuffer.h"
+#include "intel_buffers.h"
+#include "intel_mipmap_tree.h"
+#include "intel_regions.h"
+#include "intel_fbo.h"
+#include "intel_tex.h"
+#include "intel_blit.h"
+#include "intel_pixel.h"
+
+
+/**
+ * Get the intel_region which is the source for any glCopyTex[Sub]Image call.
+ *
+ * Do the best we can using the blitter.  A future project is to use
+ * the texture engine and fragment programs for these copies.
+ */
+static const struct intel_region *
+get_teximage_source(struct intel_context *intel, GLenum internalFormat)
+{
+   struct intel_renderbuffer *irb;
+
+   if (0)
+      _mesa_printf("%s %s\n", __FUNCTION__,
+                   _mesa_lookup_enum_by_nr(internalFormat));
+
+   switch (internalFormat) {
+   case GL_DEPTH_COMPONENT:
+   case GL_DEPTH_COMPONENT16_ARB:
+      irb = intel_get_renderbuffer(intel->ctx.ReadBuffer, BUFFER_DEPTH);
+      if (irb && irb->region && irb->region->cpp == 2)
+         return irb->region;
+      return NULL;
+   case GL_DEPTH24_STENCIL8_EXT:
+   case GL_DEPTH_STENCIL_EXT:
+      irb = intel_get_renderbuffer(intel->ctx.ReadBuffer, BUFFER_DEPTH);
+      if (irb && irb->region && irb->region->cpp == 4)
+         return irb->region;
+      return NULL;
+   case GL_RGBA:
+      return intel_readbuf_region(intel);
+   case GL_RGB:
+      if (intel->intelScreen->cpp == 2)
+         return intel_readbuf_region(intel);
+      return NULL;
+   default:
+      return NULL;
+   }
+}
+
+
+static GLboolean
+do_copy_texsubimage(struct intel_context *intel,
+                    struct intel_texture_image *intelImage,
+                    GLenum internalFormat,
+                    GLint dstx, GLint dsty,
+                    GLint x, GLint y, GLsizei width, GLsizei height)
+{
+   GLcontext *ctx = &intel->ctx;
+   const struct intel_region *src =
+      get_teximage_source(intel, internalFormat);
+
+   if (!intelImage->mt || !src)
+      return GL_FALSE;
+
+   intelFlush(ctx);
+   LOCK_HARDWARE(intel);
+   {
+      GLuint image_offset = intel_miptree_image_offset(intelImage->mt,
+                                                       intelImage->face,
+                                                       intelImage->level);
+      const GLint orig_x = x;
+      const GLint orig_y = y;
+      const struct gl_framebuffer *fb = ctx->DrawBuffer;
+
+      if (_mesa_clip_to_region(fb->_Xmin, fb->_Ymin, fb->_Xmax, fb->_Ymax,
+                               &x, &y, &width, &height)) {
+         /* Update dst for clipped src.  Need to also clip the source rect.
+          */
+         dstx += x - orig_x;
+         dsty += y - orig_y;
+
+         if (ctx->ReadBuffer->Name == 0) {
+            /* reading from a window, adjust x, y */
+            __DRIdrawablePrivate *dPriv = intel->driDrawable;
+            GLuint window_y;
+            /* window_y = position of window on screen if y=0=bottom */
+            window_y = intel->intelScreen->height - (dPriv->y + dPriv->h);
+            y = window_y + y;
+            x += dPriv->x;
+         }
+         else {
+            /* reading from a FBO */
+            /* invert Y */
+            y = ctx->ReadBuffer->Height - y - 1;
+         }
+
+
+         /* A bit of fiddling to get the blitter to work with -ve
+          * pitches.  But we get a nice inverted blit this way, so it's
+          * worth it:
+          */
+         intelEmitCopyBlit(intel,
+                           intelImage->mt->cpp,
+                           -src->pitch,
+                           src->buffer,
+                           src->height * src->pitch * src->cpp,
+                           intelImage->mt->pitch,
+                           intelImage->mt->region->buffer,
+                           image_offset,
+                           x, y + height, dstx, dsty, width, height);
+
+         intel_batchbuffer_flush(intel->batch);
+      }
+   }
+
+
+   UNLOCK_HARDWARE(intel);
+
+#if 0
+   /* GL_SGIS_generate_mipmap -- this can be accelerated now.
+    * XXX Add a ctx->Driver.GenerateMipmaps() function?
+    */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      intel_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+#endif
+
+   return GL_TRUE;
+}
+
+
+
+
+
+void
+intelCopyTexImage1D(GLcontext * ctx, GLenum target, GLint level,
+                    GLenum internalFormat,
+                    GLint x, GLint y, GLsizei width, GLint border)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_object *texObj =
+      _mesa_select_tex_object(ctx, texUnit, target);
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texUnit, target, level);
+
+   if (border)
+      goto fail;
+
+   /* Setup or redefine the texture object, mipmap tree and texture
+    * image.  Don't populate yet.  
+    */
+   ctx->Driver.TexImage1D(ctx, target, level, internalFormat,
+                          width, border,
+                          GL_RGBA, CHAN_TYPE, NULL,
+                          &ctx->DefaultPacking, texObj, texImage);
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat, 0, 0, x, y, width, 1))
+      goto fail;
+
+   return;
+
+ fail:
+   _swrast_copy_teximage1d(ctx, target, level, internalFormat, x, y,
+                           width, border);
+}
+
+void
+intelCopyTexImage2D(GLcontext * ctx, GLenum target, GLint level,
+                    GLenum internalFormat,
+                    GLint x, GLint y, GLsizei width, GLsizei height,
+                    GLint border)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_object *texObj =
+      _mesa_select_tex_object(ctx, texUnit, target);
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texUnit, target, level);
+
+   if (border)
+      goto fail;
+
+   /* Setup or redefine the texture object, mipmap tree and texture
+    * image.  Don't populate yet.  
+    */
+   ctx->Driver.TexImage2D(ctx, target, level, internalFormat,
+                          width, height, border,
+                          GL_RGBA, CHAN_TYPE, NULL,
+                          &ctx->DefaultPacking, texObj, texImage);
+
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat, 0, 0, x, y, width, height))
+      goto fail;
+
+   return;
+
+ fail:
+   _swrast_copy_teximage2d(ctx, target, level, internalFormat, x, y,
+                           width, height, border);
+}
+
+
+void
+intelCopyTexSubImage1D(GLcontext * ctx, GLenum target, GLint level,
+                       GLint xoffset, GLint x, GLint y, GLsizei width)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texUnit, target, level);
+   GLenum internalFormat = texImage->InternalFormat;
+
+   /* XXX need to check <border> as in above function? */
+
+   /* Need to check texture is compatible with source format. 
+    */
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat, xoffset, 0, x, y, width, 1)) {
+      _swrast_copy_texsubimage1d(ctx, target, level, xoffset, x, y, width);
+   }
+}
+
+
+
+void
+intelCopyTexSubImage2D(GLcontext * ctx, GLenum target, GLint level,
+                       GLint xoffset, GLint yoffset,
+                       GLint x, GLint y, GLsizei width, GLsizei height)
+{
+   struct gl_texture_unit *texUnit =
+      &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+   struct gl_texture_image *texImage =
+      _mesa_select_tex_image(ctx, texUnit, target, level);
+   GLenum internalFormat = texImage->InternalFormat;
+
+
+   /* Need to check texture is compatible with source format. 
+    */
+
+   if (!do_copy_texsubimage(intel_context(ctx),
+                            intel_texture_image(texImage),
+                            internalFormat,
+                            xoffset, yoffset, x, y, width, height)) {
+      _swrast_copy_texsubimage2d(ctx, target, level,
+                                 xoffset, yoffset, x, y, width, height);
+   }
+}

src/mesa/drivers/dri/i915/intel_tex_format.c

@@ -0,0 +1,146 @@
+#include "intel_context.h"
+#include "intel_tex.h"
+#include "texformat.h"
+#include "enums.h"
+
+/* It works out that this function is fine for all the supported
+ * hardware.  However, there is still a need to map the formats onto
+ * hardware descriptors.
+ */
+/* Note that the i915 can actually support many more formats than
+ * these if we take the step of simply swizzling the colors
+ * immediately after sampling...
+ */
+const struct gl_texture_format *
+intelChooseTextureFormat(GLcontext * ctx, GLint internalFormat,
+                         GLenum format, GLenum type)
+{
+   struct intel_context *intel = intel_context(ctx);
+   const GLboolean do32bpt = (intel->intelScreen->cpp == 4);
+
+   switch (internalFormat) {
+   case 4:
+   case GL_RGBA:
+   case GL_COMPRESSED_RGBA:
+      if (format == GL_BGRA) {
+         if (type == GL_UNSIGNED_BYTE || type == GL_UNSIGNED_INT_8_8_8_8_REV) {
+            return &_mesa_texformat_argb8888;
+         }
+         else if (type == GL_UNSIGNED_SHORT_4_4_4_4_REV) {
+            return &_mesa_texformat_argb4444;
+         }
+         else if (type == GL_UNSIGNED_SHORT_1_5_5_5_REV) {
+            return &_mesa_texformat_argb1555;
+         }
+      }
+      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444;
+
+   case 3:
+   case GL_RGB:
+   case GL_COMPRESSED_RGB:
+      if (format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5) {
+         return &_mesa_texformat_rgb565;
+      }
+      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_rgb565;
+
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444;
+
+   case GL_RGBA4:
+   case GL_RGBA2:
+      return &_mesa_texformat_argb4444;
+
+   case GL_RGB5_A1:
+      return &_mesa_texformat_argb1555;
+
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      return &_mesa_texformat_argb8888;
+
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      return &_mesa_texformat_rgb565;
+
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+   case GL_COMPRESSED_ALPHA:
+      return &_mesa_texformat_a8;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+   case GL_COMPRESSED_LUMINANCE:
+      return &_mesa_texformat_l8;
+
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+   case GL_COMPRESSED_LUMINANCE_ALPHA:
+      return &_mesa_texformat_al88;
+
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+   case GL_COMPRESSED_INTENSITY:
+      return &_mesa_texformat_i8;
+
+   case GL_YCBCR_MESA:
+      if (type == GL_UNSIGNED_SHORT_8_8_MESA || type == GL_UNSIGNED_BYTE)
+         return &_mesa_texformat_ycbcr;
+      else
+         return &_mesa_texformat_ycbcr_rev;
+
+   case GL_COMPRESSED_RGB_FXT1_3DFX:
+      return &_mesa_texformat_rgb_fxt1;
+   case GL_COMPRESSED_RGBA_FXT1_3DFX:
+      return &_mesa_texformat_rgba_fxt1;
+
+   case GL_RGB_S3TC:
+   case GL_RGB4_S3TC:
+   case GL_COMPRESSED_RGB_S3TC_DXT1_EXT:
+      return &_mesa_texformat_rgb_dxt1;
+
+   case GL_COMPRESSED_RGBA_S3TC_DXT1_EXT:
+      return &_mesa_texformat_rgba_dxt1;
+
+   case GL_RGBA_S3TC:
+   case GL_RGBA4_S3TC:
+   case GL_COMPRESSED_RGBA_S3TC_DXT3_EXT:
+      return &_mesa_texformat_rgba_dxt3;
+
+   case GL_COMPRESSED_RGBA_S3TC_DXT5_EXT:
+      return &_mesa_texformat_rgba_dxt5;
+
+   case GL_DEPTH_COMPONENT:
+   case GL_DEPTH_COMPONENT16:
+   case GL_DEPTH_COMPONENT24:
+   case GL_DEPTH_COMPONENT32:
+      return &_mesa_texformat_z16;
+
+   default:
+      fprintf(stderr, "unexpected texture format %s in %s\n",
+              _mesa_lookup_enum_by_nr(internalFormat), __FUNCTION__);
+      return NULL;
+   }
+
+   return NULL;                 /* never get here */
+}

src/mesa/drivers/dri/i915/intel_tex_image.c

@@ -0,0 +1,626 @@
+
+#include <stdlib.h>
+#include <stdio.h>
+
+#include "glheader.h"
+#include "macros.h"
+#include "mtypes.h"
+#include "enums.h"
+#include "colortab.h"
+#include "convolve.h"
+#include "context.h"
+#include "simple_list.h"
+#include "texcompress.h"
+#include "texformat.h"
+#include "texobj.h"
+#include "texstore.h"
+
+#include "intel_context.h"
+#include "intel_mipmap_tree.h"
+#include "intel_buffer_objects.h"
+#include "intel_batchbuffer.h"
+#include "intel_tex.h"
+#include "intel_ioctl.h"
+#include "intel_blit.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+/* Functions to store texture images.  Where possible, mipmap_tree's
+ * will be created or further instantiated with image data, otherwise
+ * images will be stored in malloc'd memory.  A validation step is
+ * required to pull those images into a mipmap tree, or otherwise
+ * decide a fallback is required.
+ */
+
+
+static int
+logbase2(int n)
+{
+   GLint i = 1;
+   GLint log2 = 0;
+
+   while (n > i) {
+      i *= 2;
+      log2++;
+   }
+
+   return log2;
+}
+
+
+/* Otherwise, store it in memory if (Border != 0) or (any dimension ==
+ * 1).
+ *    
+ * Otherwise, if max_level >= level >= min_level, create tree with
+ * space for textures from min_level down to max_level.
+ *
+ * Otherwise, create tree with space for textures from (level
+ * 0)..(1x1).  Consider pruning this tree at a validation if the
+ * saving is worth it.
+ */
+static void
+guess_and_alloc_mipmap_tree(struct intel_context *intel,
+                            struct intel_texture_object *intelObj,
+                            struct intel_texture_image *intelImage)
+{
+   GLuint firstLevel;
+   GLuint lastLevel;
+   GLuint width = intelImage->base.Width;
+   GLuint height = intelImage->base.Height;
+   GLuint depth = intelImage->base.Depth;
+   GLuint l2width, l2height, l2depth;
+   GLuint i;
+
+   DBG("%s\n", __FUNCTION__);
+
+   if (intelImage->base.Border)
+      return;
+
+   if (intelImage->level > intelObj->base.BaseLevel &&
+       (intelImage->base.Width == 1 ||
+        (intelObj->base.Target != GL_TEXTURE_1D &&
+         intelImage->base.Height == 1) ||
+        (intelObj->base.Target == GL_TEXTURE_3D &&
+         intelImage->base.Depth == 1)))
+      return;
+
+   /* If this image disrespects BaseLevel, allocate from level zero.
+    * Usually BaseLevel == 0, so it's unlikely to happen.
+    */
+   if (intelImage->level < intelObj->base.BaseLevel)
+      firstLevel = 0;
+   else
+      firstLevel = intelObj->base.BaseLevel;
+
+
+   /* Figure out image dimensions at start level. 
+    */
+   for (i = intelImage->level; i > firstLevel; i--) {
+      width <<= 1;
+      if (height != 1)
+         height <<= 1;
+      if (depth != 1)
+         depth <<= 1;
+   }
+
+   /* Guess a reasonable value for lastLevel.  This is probably going
+    * to be wrong fairly often and might mean that we have to look at
+    * resizable buffers, or require that buffers implement lazy
+    * pagetable arrangements.
+    */
+   if ((intelObj->base.MinFilter == GL_NEAREST ||
+        intelObj->base.MinFilter == GL_LINEAR) &&
+       intelImage->level == firstLevel) {
+      lastLevel = firstLevel;
+   }
+   else {
+      l2width = logbase2(width);
+      l2height = logbase2(height);
+      l2depth = logbase2(depth);
+      lastLevel = firstLevel + MAX2(MAX2(l2width, l2height), l2depth);
+   }
+
+   assert(!intelObj->mt);
+   intelObj->mt = intel_miptree_create(intel,
+                                       intelObj->base.Target,
+                                       intelImage->base.InternalFormat,
+                                       firstLevel,
+                                       lastLevel,
+                                       width,
+                                       height,
+                                       depth,
+                                       intelImage->base.TexFormat->TexelBytes,
+                                       intelImage->base.IsCompressed);
+
+   DBG("%s - success\n", __FUNCTION__);
+}
+
+
+
+
+static GLuint
+target_to_face(GLenum target)
+{
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y_ARB:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z_ARB:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z_ARB:
+      return ((GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X);
+   default:
+      return 0;
+   }
+}
+
+/* There are actually quite a few combinations this will work for,
+ * more than what I've listed here.
+ */
+static GLboolean
+check_pbo_format(GLint internalFormat,
+                 GLenum format, GLenum type,
+                 const struct gl_texture_format *mesa_format)
+{
+   switch (internalFormat) {
+   case 4:
+   case GL_RGBA:
+      return (format == GL_BGRA &&
+              (type == GL_UNSIGNED_BYTE ||
+               type == GL_UNSIGNED_INT_8_8_8_8_REV) &&
+              mesa_format == &_mesa_texformat_argb8888);
+   case 3:
+   case GL_RGB:
+      return (format == GL_RGB &&
+              type == GL_UNSIGNED_SHORT_5_6_5 &&
+              mesa_format == &_mesa_texformat_rgb565);
+   case GL_YCBCR_MESA:
+      return (type == GL_UNSIGNED_SHORT_8_8_MESA || type == GL_UNSIGNED_BYTE);
+   default:
+      return GL_FALSE;
+   }
+}
+
+
+/* XXX: Do this for TexSubImage also:
+ */
+static GLboolean
+try_pbo_upload(struct intel_context *intel,
+               struct intel_texture_image *intelImage,
+               const struct gl_pixelstore_attrib *unpack,
+               GLint internalFormat,
+               GLint width, GLint height,
+               GLenum format, GLenum type, const void *pixels)
+{
+   struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
+   GLuint src_offset, src_stride;
+   GLuint dst_offset, dst_stride;
+
+   if (!pbo ||
+       intel->ctx._ImageTransferState ||
+       unpack->SkipPixels || unpack->SkipRows) {
+      _mesa_printf("%s: failure 1\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   src_offset = (GLuint) pixels;
+
+   if (unpack->RowLength > 0)
+      src_stride = unpack->RowLength;
+   else
+      src_stride = width;
+
+   dst_offset = intel_miptree_image_offset(intelImage->mt,
+                                           intelImage->face,
+                                           intelImage->level);
+
+   dst_stride = intelImage->mt->pitch;
+
+   intelFlush(&intel->ctx);
+   LOCK_HARDWARE(intel);
+   {
+      struct _DriBufferObject *src_buffer =
+         intel_bufferobj_buffer(intel, pbo, INTEL_READ);
+      struct _DriBufferObject *dst_buffer =
+         intel_region_buffer(intel->intelScreen, intelImage->mt->region,
+                             INTEL_WRITE_FULL);
+
+
+      intelEmitCopyBlit(intel,
+                        intelImage->mt->cpp,
+                        src_stride, src_buffer, src_offset,
+                        dst_stride, dst_buffer, dst_offset,
+                        0, 0, 0, 0, width, height);
+
+      intel_batchbuffer_flush(intel->batch);
+   }
+   UNLOCK_HARDWARE(intel);
+
+   return GL_TRUE;
+}
+
+
+
+static GLboolean
+try_pbo_zcopy(struct intel_context *intel,
+              struct intel_texture_image *intelImage,
+              const struct gl_pixelstore_attrib *unpack,
+              GLint internalFormat,
+              GLint width, GLint height,
+              GLenum format, GLenum type, const void *pixels)
+{
+   struct intel_buffer_object *pbo = intel_buffer_object(unpack->BufferObj);
+   GLuint src_offset, src_stride;
+   GLuint dst_offset, dst_stride;
+
+   if (!pbo ||
+       intel->ctx._ImageTransferState ||
+       unpack->SkipPixels || unpack->SkipRows) {
+      _mesa_printf("%s: failure 1\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   src_offset = (GLuint) pixels;
+
+   if (unpack->RowLength > 0)
+      src_stride = unpack->RowLength;
+   else
+      src_stride = width;
+
+   dst_offset = intel_miptree_image_offset(intelImage->mt,
+                                           intelImage->face,
+                                           intelImage->level);
+
+   dst_stride = intelImage->mt->pitch;
+
+   if (src_stride != dst_stride || dst_offset != 0 || src_offset != 0) {
+      _mesa_printf("%s: failure 2\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   intel_region_attach_pbo(intel->intelScreen, intelImage->mt->region, pbo);
+
+   return GL_TRUE;
+}
+
+
+
+
+
+
+static void
+intelTexImage(GLcontext * ctx,
+              GLint dims,
+              GLenum target, GLint level,
+              GLint internalFormat,
+              GLint width, GLint height, GLint depth,
+              GLint border,
+              GLenum format, GLenum type, const void *pixels,
+              const struct gl_pixelstore_attrib *unpack,
+              struct gl_texture_object *texObj,
+              struct gl_texture_image *texImage)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_object *intelObj = intel_texture_object(texObj);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
+   GLint postConvWidth = width;
+   GLint postConvHeight = height;
+   GLint texelBytes, sizeInBytes;
+   GLuint dstRowStride;
+
+
+   DBG("%s target %s level %d %dx%dx%d border %d\n", __FUNCTION__,
+       _mesa_lookup_enum_by_nr(target), level, width, height, depth, border);
+
+   intelFlush(ctx);
+
+   intelImage->face = target_to_face(target);
+   intelImage->level = level;
+
+   if (ctx->_ImageTransferState & IMAGE_CONVOLUTION_BIT) {
+      _mesa_adjust_image_for_convolution(ctx, dims, &postConvWidth,
+                                         &postConvHeight);
+   }
+
+   /* choose the texture format */
+   texImage->TexFormat = intelChooseTextureFormat(ctx, internalFormat,
+                                                  format, type);
+
+   assert(texImage->TexFormat);
+
+   switch (dims) {
+   case 1:
+      texImage->FetchTexelc = texImage->TexFormat->FetchTexel1D;
+      texImage->FetchTexelf = texImage->TexFormat->FetchTexel1Df;
+      break;
+   case 2:
+      texImage->FetchTexelc = texImage->TexFormat->FetchTexel2D;
+      texImage->FetchTexelf = texImage->TexFormat->FetchTexel2Df;
+      break;
+   case 3:
+      texImage->FetchTexelc = texImage->TexFormat->FetchTexel3D;
+      texImage->FetchTexelf = texImage->TexFormat->FetchTexel3Df;
+      break;
+   default:
+      assert(0);
+      break;
+   }
+
+   texelBytes = texImage->TexFormat->TexelBytes;
+
+
+   /* Minimum pitch of 32 bytes */
+   if (postConvWidth * texelBytes < 32) {
+      postConvWidth = 32 / texelBytes;
+      texImage->RowStride = postConvWidth;
+   }
+
+   assert(texImage->RowStride == postConvWidth);
+
+   /* Release the reference to a potentially orphaned buffer.   
+    * Release any old malloced memory.
+    */
+   if (intelImage->mt) {
+      intel_miptree_release(intel, &intelImage->mt);
+      assert(!texImage->Data);
+   }
+   else if (texImage->Data) {
+      free(texImage->Data);
+   }
+
+   /* If this is the only texture image in the tree, could call
+    * bmBufferData with NULL data to free the old block and avoid
+    * waiting on any outstanding fences.
+    */
+   if (intelObj->mt &&
+       intelObj->mt->first_level == level &&
+       intelObj->mt->last_level == level &&
+       intelObj->mt->target != GL_TEXTURE_CUBE_MAP_ARB &&
+       !intel_miptree_match_image(intelObj->mt, &intelImage->base,
+                                  intelImage->face, intelImage->level)) {
+
+      DBG("release it\n");
+      intel_miptree_release(intel, &intelObj->mt);
+      assert(!intelObj->mt);
+   }
+
+   if (!intelObj->mt) {
+      guess_and_alloc_mipmap_tree(intel, intelObj, intelImage);
+      if (!intelObj->mt) {
+	 DBG("guess_and_alloc_mipmap_tree: failed\n");
+      }
+   }
+
+
+   assert(!intelImage->mt);
+
+   if (intelObj->mt &&
+       intel_miptree_match_image(intelObj->mt, &intelImage->base,
+                                 intelImage->face, intelImage->level)) {
+
+      intel_miptree_reference(&intelImage->mt, intelObj->mt);
+      assert(intelImage->mt);
+   }
+
+   if (!intelImage->mt)
+      DBG("XXX: Image did not fit into tree - storing in local memory!\n");
+
+   /* PBO fastpaths:
+    */
+   if (dims <= 2 &&
+       intelImage->mt &&
+       intel_buffer_object(unpack->BufferObj) &&
+       check_pbo_format(internalFormat, format,
+                        type, intelImage->base.TexFormat)) {
+
+      DBG("trying pbo upload\n");
+
+      /* Attempt to texture directly from PBO data (zero copy upload).
+       *
+       * Currently disable as it can lead to worse as well as better
+       * performance (in particular when intel_region_cow() is
+       * required).
+       */
+      if (intelObj->mt == intelImage->mt &&
+          intelObj->mt->first_level == level &&
+          intelObj->mt->last_level == level) {
+
+         if (try_pbo_zcopy(intel, intelImage, unpack,
+                           internalFormat,
+                           width, height, format, type, pixels)) {
+
+            DBG("pbo zcopy upload succeeded\n");
+            return;
+         }
+      }
+
+
+      /* Otherwise, attempt to use the blitter for PBO image uploads.
+       */
+      if (try_pbo_upload(intel, intelImage, unpack,
+                         internalFormat,
+                         width, height, format, type, pixels)) {
+         DBG("pbo upload succeeded\n");
+         return;
+      }
+
+      DBG("pbo upload failed\n");
+   }
+
+
+
+   /* intelCopyTexImage calls this function with pixels == NULL, with
+    * the expectation that the mipmap tree will be set up but nothing
+    * more will be done.  This is where those calls return:
+    */
+   pixels = _mesa_validate_pbo_teximage(ctx, dims, width, height, 1,
+                                        format, type,
+                                        pixels, unpack, "glTexImage");
+   if (!pixels)
+      return;
+
+
+   if (intelImage->mt)
+      intel_region_idle(intel->intelScreen, intelImage->mt->region);
+
+   LOCK_HARDWARE(intel);
+
+   if (intelImage->mt) {
+      texImage->Data = intel_miptree_image_map(intel,
+                                               intelImage->mt,
+                                               intelImage->face,
+                                               intelImage->level,
+                                               &dstRowStride,
+                                               intelImage->base.ImageOffsets);
+   }
+   else {
+      /* Allocate regular memory and store the image there temporarily.   */
+      if (texImage->IsCompressed) {
+         sizeInBytes = texImage->CompressedSize;
+         dstRowStride =
+            _mesa_compressed_row_stride(texImage->InternalFormat, width);
+         assert(dims != 3);
+      }
+      else {
+         dstRowStride = postConvWidth * texelBytes;
+         sizeInBytes = depth * dstRowStride * postConvHeight;
+      }
+
+      texImage->Data = malloc(sizeInBytes);
+   }
+
+   DBG("Upload image %dx%dx%d row_len %x "
+       "pitch %x\n",
+       width, height, depth, width * texelBytes, dstRowStride);
+
+   /* Copy data.  Would like to know when it's ok for us to eg. use
+    * the blitter to copy.  Or, use the hardware to do the format
+    * conversion and copy:
+    */
+   if (!texImage->TexFormat->StoreImage(ctx, dims, 
+					texImage->_BaseFormat, 
+					texImage->TexFormat, 
+					texImage->Data, 0, 0, 0, /* dstX/Y/Zoffset */
+                                        dstRowStride,
+                                        texImage->ImageOffsets,
+                                        width, height, depth,
+                                        format, type, pixels, unpack)) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage");
+   }
+
+   _mesa_unmap_teximage_pbo(ctx, unpack);
+
+   if (intelImage->mt) {
+      intel_miptree_image_unmap(intel, intelImage->mt);
+      texImage->Data = NULL;
+   }
+
+   UNLOCK_HARDWARE(intel);
+
+#if 0
+   /* GL_SGIS_generate_mipmap -- this can be accelerated now.
+    */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      intel_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+#endif
+}
+
+void
+intelTexImage3D(GLcontext * ctx,
+                GLenum target, GLint level,
+                GLint internalFormat,
+                GLint width, GLint height, GLint depth,
+                GLint border,
+                GLenum format, GLenum type, const void *pixels,
+                const struct gl_pixelstore_attrib *unpack,
+                struct gl_texture_object *texObj,
+                struct gl_texture_image *texImage)
+{
+   intelTexImage(ctx, 3, target, level,
+                 internalFormat, width, height, depth, border,
+                 format, type, pixels, unpack, texObj, texImage);
+}
+
+
+void
+intelTexImage2D(GLcontext * ctx,
+                GLenum target, GLint level,
+                GLint internalFormat,
+                GLint width, GLint height, GLint border,
+                GLenum format, GLenum type, const void *pixels,
+                const struct gl_pixelstore_attrib *unpack,
+                struct gl_texture_object *texObj,
+                struct gl_texture_image *texImage)
+{
+   intelTexImage(ctx, 2, target, level,
+                 internalFormat, width, height, 1, border,
+                 format, type, pixels, unpack, texObj, texImage);
+}
+
+void
+intelTexImage1D(GLcontext * ctx,
+                GLenum target, GLint level,
+                GLint internalFormat,
+                GLint width, GLint border,
+                GLenum format, GLenum type, const void *pixels,
+                const struct gl_pixelstore_attrib *unpack,
+                struct gl_texture_object *texObj,
+                struct gl_texture_image *texImage)
+{
+   intelTexImage(ctx, 1, target, level,
+                 internalFormat, width, 1, 1, border,
+                 format, type, pixels, unpack, texObj, texImage);
+}
+
+
+
+/**
+ * Need to map texture image into memory before copying image data,
+ * then unmap it.
+ */
+void
+intelGetTexImage(GLcontext * ctx, GLenum target, GLint level,
+                 GLenum format, GLenum type, GLvoid * pixels,
+                 struct gl_texture_object *texObj,
+                 struct gl_texture_image *texImage)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
+
+   /* Map */
+   if (intelImage->mt) {
+      /* Image is stored in hardware format in a buffer managed by the
+       * kernel.  Need to explicitly map and unmap it.
+       */
+      intelImage->base.Data =
+         intel_miptree_image_map(intel,
+                                 intelImage->mt,
+                                 intelImage->face,
+                                 intelImage->level,
+                                 &intelImage->base.RowStride,
+                                 intelImage->base.ImageOffsets);
+   }
+   else {
+      /* Otherwise, the image should actually be stored in
+       * intelImage->base.Data.  This is pretty confusing for
+       * everybody, I'd much prefer to separate the two functions of
+       * texImage->Data - storage for texture images in main memory
+       * and access (ie mappings) of images.  In other words, we'd
+       * create a new texImage->Map field and leave Data simply for
+       * storage.
+       */
+      assert(intelImage->base.Data);
+   }
+
+   _mesa_get_teximage(ctx, target, level, format, type, pixels,
+                      texObj, texImage);
+
+   /* Unmap */
+   if (intelImage->mt) {
+      intel_miptree_image_unmap(intel, intelImage->mt);
+      intelImage->base.Data = NULL;
+   }
+}

src/mesa/drivers/dri/i915/intel_tex_subimage.c

@@ -0,0 +1,183 @@
+
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "mtypes.h"
+#include "texobj.h"
+#include "texstore.h"
+#include "enums.h"
+
+#include "intel_context.h"
+#include "intel_tex.h"
+#include "intel_mipmap_tree.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+static void
+intelTexSubimage(GLcontext * ctx,
+                 GLint dims,
+                 GLenum target, GLint level,
+                 GLint xoffset, GLint yoffset, GLint zoffset,
+                 GLint width, GLint height, GLint depth,
+                 GLenum format, GLenum type, const void *pixels,
+                 const struct gl_pixelstore_attrib *packing,
+                 struct gl_texture_object *texObj,
+                 struct gl_texture_image *texImage)
+{
+   struct intel_context *intel = intel_context(ctx);
+   struct intel_texture_image *intelImage = intel_texture_image(texImage);
+   GLuint dstImageStride;
+   GLuint dstRowStride;
+
+   DBG("%s target %s level %d offset %d,%d %dx%d\n", __FUNCTION__,
+       _mesa_lookup_enum_by_nr(target),
+       level, xoffset, yoffset, width, height);
+
+   intelFlush(ctx);
+
+   pixels =
+      _mesa_validate_pbo_teximage(ctx, dims, width, height, depth, format,
+                                  type, pixels, packing, "glTexSubImage2D");
+   if (!pixels)
+      return;
+
+   if (intelImage->mt)
+      intel_region_idle(intel->intelScreen, intelImage->mt->region);
+
+   LOCK_HARDWARE(intel);
+
+   /* Map buffer if necessary.  Need to lock to prevent other contexts
+    * from uploading the buffer under us.
+    */
+   if (intelImage->mt) 
+      texImage->Data = intel_miptree_image_map(intel,
+                                               intelImage->mt,
+                                               intelImage->face,
+                                               intelImage->level,
+                                               &dstRowStride,
+                                               &dstImageStride);
+
+   assert(dstRowStride);
+
+   if (!texImage->TexFormat->StoreImage(ctx, dims, texImage->_BaseFormat,
+                                        texImage->TexFormat,
+                                        texImage->Data,
+                                        xoffset, yoffset, zoffset,
+                                        dstRowStride,
+                                        texImage->ImageOffsets,
+                                        width, height, depth,
+                                        format, type, pixels, packing)) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "intelTexSubImage");
+   }
+
+#if 0
+   /* GL_SGIS_generate_mipmap */
+   if (level == texObj->BaseLevel && texObj->GenerateMipmap) {
+      _mesa_generate_mipmap(ctx, target,
+                            &ctx->Texture.Unit[ctx->Texture.CurrentUnit],
+                            texObj);
+   }
+#endif
+
+   _mesa_unmap_teximage_pbo(ctx, packing);
+
+   if (intelImage->mt) {
+      intel_miptree_image_unmap(intel, intelImage->mt);
+      texImage->Data = NULL;
+   }
+
+   UNLOCK_HARDWARE(intel);
+}
+
+
+
+
+
+void
+intelTexSubImage3D(GLcontext * ctx,
+                   GLenum target,
+                   GLint level,
+                   GLint xoffset, GLint yoffset, GLint zoffset,
+                   GLsizei width, GLsizei height, GLsizei depth,
+                   GLenum format, GLenum type,
+                   const GLvoid * pixels,
+                   const struct gl_pixelstore_attrib *packing,
+                   struct gl_texture_object *texObj,
+                   struct gl_texture_image *texImage)
+{
+
+   intelTexSubimage(ctx, 3,
+                    target, level,
+                    xoffset, yoffset, zoffset,
+                    width, height, depth,
+                    format, type, pixels, packing, texObj, texImage);
+
+}
+
+
+
+void
+intelTexSubImage2D(GLcontext * ctx,
+                   GLenum target,
+                   GLint level,
+                   GLint xoffset, GLint yoffset,
+                   GLsizei width, GLsizei height,
+                   GLenum format, GLenum type,
+                   const GLvoid * pixels,
+                   const struct gl_pixelstore_attrib *packing,
+                   struct gl_texture_object *texObj,
+                   struct gl_texture_image *texImage)
+{
+
+   intelTexSubimage(ctx, 2,
+                    target, level,
+                    xoffset, yoffset, 0,
+                    width, height, 1,
+                    format, type, pixels, packing, texObj, texImage);
+
+}
+
+
+void
+intelTexSubImage1D(GLcontext * ctx,
+                   GLenum target,
+                   GLint level,
+                   GLint xoffset,
+                   GLsizei width,
+                   GLenum format, GLenum type,
+                   const GLvoid * pixels,
+                   const struct gl_pixelstore_attrib *packing,
+                   struct gl_texture_object *texObj,
+                   struct gl_texture_image *texImage)
+{
+   intelTexSubimage(ctx, 1,
+                    target, level,
+                    xoffset, 0, 0,
+                    width, 1, 1,
+                    format, type, pixels, packing, texObj, texImage);
+
+}

src/mesa/drivers/dri/i915/intel_tex_validate.c

@@ -0,0 +1,253 @@
+#include "mtypes.h"
+#include "macros.h"
+
+#include "intel_context.h"
+#include "intel_mipmap_tree.h"
+#include "intel_tex.h"
+
+#define FILE_DEBUG_FLAG DEBUG_TEXTURE
+
+/**
+ * Compute which mipmap levels that really need to be sent to the hardware.
+ * This depends on the base image size, GL_TEXTURE_MIN_LOD,
+ * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
+ */
+static void
+intel_calculate_first_last_level(struct intel_texture_object *intelObj)
+{
+   struct gl_texture_object *tObj = &intelObj->base;
+   const struct gl_texture_image *const baseImage =
+      tObj->Image[0][tObj->BaseLevel];
+
+   /* These must be signed values.  MinLod and MaxLod can be negative numbers,
+    * and having firstLevel and lastLevel as signed prevents the need for
+    * extra sign checks.
+    */
+   int firstLevel;
+   int lastLevel;
+
+   /* Yes, this looks overly complicated, but it's all needed.
+    */
+   switch (tObj->Target) {
+   case GL_TEXTURE_1D:
+   case GL_TEXTURE_2D:
+   case GL_TEXTURE_3D:
+   case GL_TEXTURE_CUBE_MAP:
+      if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) {
+         /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
+          */
+         firstLevel = lastLevel = tObj->BaseLevel;
+      }
+      else {
+         firstLevel = tObj->BaseLevel + (GLint) (tObj->MinLod + 0.5);
+         firstLevel = MAX2(firstLevel, tObj->BaseLevel);
+         lastLevel = tObj->BaseLevel + (GLint) (tObj->MaxLod + 0.5);
+         lastLevel = MAX2(lastLevel, tObj->BaseLevel);
+         lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2);
+         lastLevel = MIN2(lastLevel, tObj->MaxLevel);
+         lastLevel = MAX2(firstLevel, lastLevel);       /* need at least one level */
+      }
+      break;
+   case GL_TEXTURE_RECTANGLE_NV:
+   case GL_TEXTURE_4D_SGIS:
+      firstLevel = lastLevel = 0;
+      break;
+   default:
+      return;
+   }
+
+   /* save these values */
+   intelObj->firstLevel = firstLevel;
+   intelObj->lastLevel = lastLevel;
+}
+
+static void
+copy_image_data_to_tree(struct intel_context *intel,
+                        struct intel_texture_object *intelObj,
+                        struct intel_texture_image *intelImage)
+{
+   if (intelImage->mt) {
+      /* Copy potentially with the blitter:
+       */
+      intel_miptree_image_copy(intel,
+                               intelObj->mt,
+                               intelImage->face,
+                               intelImage->level, intelImage->mt);
+
+      intel_miptree_release(intel, &intelImage->mt);
+   }
+   else {
+      assert(intelImage->base.Data != NULL);
+
+      /* More straightforward upload.  
+       */
+      intel_miptree_image_data(intel,
+                               intelObj->mt,
+                               intelImage->face,
+                               intelImage->level,
+                               intelImage->base.Data,
+                               intelImage->base.RowStride,
+                               intelImage->base.RowStride *
+                               intelImage->base.Height);
+
+      free(intelImage->base.Data);
+      intelImage->base.Data = NULL;
+   }
+
+   intel_miptree_reference(&intelImage->mt, intelObj->mt);
+}
+
+
+/*  
+ */
+GLuint
+intel_finalize_mipmap_tree(struct intel_context *intel, GLuint unit)
+{
+   struct gl_texture_object *tObj = intel->ctx.Texture.Unit[unit]._Current;
+   struct intel_texture_object *intelObj = intel_texture_object(tObj);
+
+   GLuint face, i;
+   GLuint nr_faces = 0;
+   struct intel_texture_image *firstImage;
+
+   /* We know/require this is true by now: 
+    */
+   assert(intelObj->base.Complete);
+
+   /* What levels must the tree include at a minimum?
+    */
+   intel_calculate_first_last_level(intelObj);
+   firstImage =
+      intel_texture_image(intelObj->base.Image[0][intelObj->firstLevel]);
+
+   /* Fallback case:
+    */
+   if (firstImage->base.Border) {
+      if (intelObj->mt) {
+         intel_miptree_release(intel, &intelObj->mt);
+      }
+      return GL_FALSE;
+   }
+
+
+   /* If both firstImage and intelObj have a tree which can contain
+    * all active images, favour firstImage.  Note that because of the
+    * completeness requirement, we know that the image dimensions
+    * will match.
+    */
+   if (firstImage->mt &&
+       firstImage->mt != intelObj->mt &&
+       firstImage->mt->first_level <= intelObj->firstLevel &&
+       firstImage->mt->last_level >= intelObj->lastLevel) {
+
+      if (intelObj->mt)
+         intel_miptree_release(intel, &intelObj->mt);
+
+      intel_miptree_reference(&intelObj->mt, firstImage->mt);
+   }
+
+   /* Check tree can hold all active levels.  Check tree matches
+    * target, imageFormat, etc.
+    * 
+    * XXX: For some layouts (eg i945?), the test might have to be
+    * first_level == firstLevel, as the tree isn't valid except at the
+    * original start level.  Hope to get around this by
+    * programming minLod, maxLod, baseLevel into the hardware and
+    * leaving the tree alone.
+    */
+   if (intelObj->mt &&
+       ((intelObj->mt->first_level > intelObj->firstLevel) ||
+        (intelObj->mt->last_level < intelObj->lastLevel) ||
+        (intelObj->mt->internal_format != firstImage->base.InternalFormat))) {
+      intel_miptree_release(intel, &intelObj->mt);
+   }
+
+
+   /* May need to create a new tree:
+    */
+   if (!intelObj->mt) {
+      intelObj->mt = intel_miptree_create(intel,
+                                          intelObj->base.Target,
+                                          firstImage->base.InternalFormat,
+                                          intelObj->firstLevel,
+                                          intelObj->lastLevel,
+                                          firstImage->base.Width,
+                                          firstImage->base.Height,
+                                          firstImage->base.Depth,
+                                          firstImage->base.TexFormat->
+                                          TexelBytes,
+                                          firstImage->base.IsCompressed);
+   }
+
+   /* Pull in any images not in the object's tree:
+    */
+   nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+   for (face = 0; face < nr_faces; face++) {
+      for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) {
+         struct intel_texture_image *intelImage =
+            intel_texture_image(intelObj->base.Image[face][i]);
+
+         /* Need to import images in main memory or held in other trees.
+          */
+         if (intelObj->mt != intelImage->mt) {
+            copy_image_data_to_tree(intel, intelObj, intelImage);
+         }
+      }
+   }
+
+   return GL_TRUE;
+}
+
+
+
+void
+intel_tex_map_images(struct intel_context *intel,
+                     struct intel_texture_object *intelObj)
+{
+   GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+   GLuint face, i;
+
+   DBG("%s\n", __FUNCTION__);
+
+   for (face = 0; face < nr_faces; face++) {
+      for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) {
+         struct intel_texture_image *intelImage =
+            intel_texture_image(intelObj->base.Image[face][i]);
+
+         if (intelImage->mt) {
+            intelImage->base.Data =
+               intel_miptree_image_map(intel,
+                                       intelImage->mt,
+                                       intelImage->face,
+                                       intelImage->level,
+                                       &intelImage->base.RowStride,
+                                       intelImage->base.ImageOffsets);
+            /* convert stride to texels, not bytes */
+            intelImage->base.RowStride /= intelImage->mt->cpp;
+/*             intelImage->base.ImageStride /= intelImage->mt->cpp; */
+         }
+      }
+   }
+}
+
+
+
+void
+intel_tex_unmap_images(struct intel_context *intel,
+                       struct intel_texture_object *intelObj)
+{
+   GLuint nr_faces = (intelObj->base.Target == GL_TEXTURE_CUBE_MAP) ? 6 : 1;
+   GLuint face, i;
+
+   for (face = 0; face < nr_faces; face++) {
+      for (i = intelObj->firstLevel; i <= intelObj->lastLevel; i++) {
+         struct intel_texture_image *intelImage =
+            intel_texture_image(intelObj->base.Image[face][i]);
+
+         if (intelImage->mt) {
+            intel_miptree_image_unmap(intel, intelImage->mt);
+            intelImage->base.Data = NULL;
+         }
+      }
+   }
+}

src/mesa/drivers/dri/i915/intel_texmem.c

@@ -1,72 +0,0 @@
-#include "texmem.h"
-#include "simple_list.h"
-#include "imports.h"
-#include "macros.h"
-
-#include "intel_tex.h"
-
-static GLuint
-driLog2( GLuint n )
-{
-   GLuint log2;
-
-   for ( log2 = 1 ; n > 1 ; log2++ ) {
-      n >>= 1;
-   }
-
-   return log2;
-}
-
-static void calculate_heap_size( driTexHeap * heap, unsigned size, 
-				 unsigned nr_regions, unsigned alignmentShift )
-{
-   unsigned     l;
-
-   l = driLog2( (size - 1) / nr_regions );
-   if ( l < alignmentShift )
-   {
-      l = alignmentShift;
-   }
-
-   heap->logGranularity = l;
-   heap->size = size & ~((1L << l) - 1);
-}
-
-
-GLboolean 
-intel_driReinitTextureHeap( driTexHeap *heap,
-			    unsigned size )
-{
-   driTextureObject *t, *tmp;
-
-   /* Kick out everything:
-    */
-   foreach_s ( t, tmp, & heap->texture_objects ) {
-      if ( t->tObj != NULL ) {
-	 driSwapOutTextureObject( t );
-      }
-      else {
-	 driDestroyTextureObject( t );
-      }
-   }
-   
-   /* Destroy the memory manager:
-    */
-   mmDestroy( heap->memory_heap );
-      
-   /* Recreate the memory manager:
-    */
-   calculate_heap_size(heap, size, heap->nrRegions, heap->alignmentShift);
-   heap->memory_heap = mmInit( 0, heap->size );
-   if ( heap->memory_heap == NULL ) {
-      fprintf(stderr, "driReinitTextureHeap: couldn't recreate memory heap\n");
-      FREE( heap );
-      return GL_FALSE;
-   }
-
-   make_empty_list( & heap->texture_objects );
-
-   return GL_TRUE;
-}
-
-

src/mesa/drivers/dri/i915/intel_tris.h

@@ -30,6 +30,8 @@
 
 #include "mtypes.h"
 
+
+
 #define _INTEL_NEW_RENDERSTATE (_DD_NEW_LINE_STIPPLE |		\
 			       _DD_NEW_TRI_UNFILLED |		\
 			       _DD_NEW_TRI_LIGHT_TWOSIDE |	\
@@ -38,9 +40,30 @@
 			       _NEW_PROGRAM |		\
 			       _NEW_POLYGONSTIPPLE)
 
-extern void intelInitTriFuncs( GLcontext *ctx );
+extern void intelInitTriFuncs(GLcontext * ctx);
+
+extern void intelChooseRenderState(GLcontext * ctx);
+
+extern void intelStartInlinePrimitive(struct intel_context *intel,
+                                      GLuint prim, GLuint flags);
+extern void intelWrapInlinePrimitive(struct intel_context *intel);
+
+GLuint *intelExtendInlinePrimitive(struct intel_context *intel,
+                                   GLuint dwords);
+
+
+void intel_meta_draw_quad(struct intel_context *intel,
+                          GLfloat x0, GLfloat x1,
+                          GLfloat y0, GLfloat y1,
+                          GLfloat z,
+                          GLuint color,
+                          GLfloat s0, GLfloat s1, GLfloat t0, GLfloat t1);
+
+void intel_meta_draw_poly(struct intel_context *intel,
+                          GLuint n,
+                          GLfloat xy[][2],
+                          GLfloat z, GLuint color, GLfloat tex[][2]);
+
 
-extern void intelPrintRenderState( const char *msg, GLuint state );
-extern void intelChooseRenderState( GLcontext *ctx );
 
 #endif

src/mesa/main/attrib.c

@@ -276,6 +276,8 @@ _mesa_PushAttrib(GLbitfield mask)
       struct gl_pixel_attrib *attr;
       attr = MALLOC_STRUCT( gl_pixel_attrib );
       MEMCPY( attr, &ctx->Pixel, sizeof(struct gl_pixel_attrib) );
+      /* push the Read FBO's ReadBuffer state, not ctx->Pixel.ReadBuffer */
+      attr->ReadBuffer = ctx->ReadBuffer->ColorReadBuffer;
       newnode = new_attrib_node( GL_PIXEL_MODE_BIT );
       newnode->data = attr;
       newnode->next = head;

src/mesa/main/buffers.c

@@ -591,11 +591,9 @@ _mesa_ReadBuffer(GLenum buffer)
  * \note This function should only be called through the GL API, not
  * from device drivers (as was done in the past).
  */
-void GLAPIENTRY
-_mesa_ResizeBuffersMESA( void )
-{
-   GET_CURRENT_CONTEXT(ctx);
 
+void _mesa_resizebuffers( GLcontext *ctx )
+{
    ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH( ctx );
 
    if (MESA_VERBOSE & VERBOSE_API)
@@ -637,6 +635,14 @@ _mesa_ResizeBuffersMESA( void )
    ctx->NewState |= _NEW_BUFFERS;  /* to update scissor / window bounds */
 }
 
+void GLAPIENTRY
+_mesa_ResizeBuffersMESA( void )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   _mesa_resizebuffers( ctx );
+}
+
 
 /*
  * XXX move somewhere else someday?

src/mesa/main/buffers.h

@@ -78,4 +78,6 @@ extern void
 _mesa_set_scissor(GLcontext *ctx, 
                   GLint x, GLint y, GLsizei width, GLsizei height);
 
+extern void _mesa_resizebuffers( GLcontext *ctx );
+
 #endif

src/mesa/main/context.c

@@ -704,9 +704,7 @@ alloc_shared_state( GLcontext *ctx )
 
    ss->ArrayObjects = _mesa_NewHashTable();
 
-#if FEATURE_ARB_shader_objects
    ss->GL2Objects = _mesa_NewHashTable ();
-#endif
 
    ss->Default1D = (*ctx->Driver.NewTextureObject)(ctx, 0, GL_TEXTURE_1D);
    if (!ss->Default1D)
@@ -777,10 +775,8 @@ alloc_shared_state( GLcontext *ctx )
    if (ss->ArrayObjects)
       _mesa_DeleteHashTable (ss->ArrayObjects);
 
-#if FEATURE_ARB_shader_objects
    if (ss->GL2Objects)
       _mesa_DeleteHashTable (ss->GL2Objects);
-#endif
 
 #if FEATURE_EXT_framebuffer_object
    if (ss->FrameBuffers)
@@ -933,9 +929,7 @@ free_shared_state( GLcontext *ctx, struct gl_shared_state *ss )
    _mesa_HashDeleteAll(ss->ArrayObjects, delete_arrayobj_cb, ctx);
    _mesa_DeleteHashTable(ss->ArrayObjects);
 
-#if FEATURE_ARB_shader_objects
    _mesa_DeleteHashTable(ss->GL2Objects);
-#endif
 
 #if FEATURE_EXT_framebuffer_object
    _mesa_DeleteHashTable(ss->FrameBuffers);
@@ -1713,6 +1707,8 @@ _mesa_make_current( GLcontext *newCtx, GLframebuffer *drawBuffer,
          if (readBuffer != drawBuffer && !readBuffer->Initialized) {
             initialize_framebuffer_size(newCtx, readBuffer);
          }
+
+	 _mesa_resizebuffers(newCtx);
 #endif
          if (newCtx->FirstTimeCurrent) {
             /* set initial viewport and scissor size now */

src/mesa/main/fbobject.c

@@ -570,6 +570,11 @@ _mesa_BindRenderbufferEXT(GLenum target, GLuint renderbuffer)
    }
 
    FLUSH_VERTICES(ctx, _NEW_BUFFERS);
+   /* The above doesn't fully flush the drivers in the way that a
+    * glFlush does, but that is required here:
+    */
+   ctx->Driver.Flush(ctx);
+
 
    if (renderbuffer) {
       newRb = _mesa_lookup_renderbuffer(ctx, renderbuffer);
@@ -972,7 +977,9 @@ _mesa_BindFramebufferEXT(GLenum target, GLuint framebuffer)
    }
 
    FLUSH_VERTICES(ctx, _NEW_BUFFERS);
-
+   if (ctx->Driver.Flush) {  
+      ctx->Driver.Flush(ctx);
+   }
    if (framebuffer) {
       /* Binding a user-created framebuffer object */
       newFb = _mesa_lookup_framebuffer(ctx, framebuffer);
@@ -1047,6 +1054,10 @@ _mesa_DeleteFramebuffersEXT(GLsizei n, const GLuint *framebuffers)
 
    ASSERT_OUTSIDE_BEGIN_END(ctx);
    FLUSH_VERTICES(ctx, _NEW_BUFFERS);
+   /* The above doesn't fully flush the drivers in the way that a
+    * glFlush does, but that is required here:
+    */
+   ctx->Driver.Flush(ctx);
 
    for (i = 0; i < n; i++) {
       if (framebuffers[i] > 0) {
@@ -1262,6 +1273,10 @@ framebuffer_texture(GLuint dims, GLenum target, GLenum attachment,
    }
 
    FLUSH_VERTICES(ctx, _NEW_BUFFERS);
+   /* The above doesn't fully flush the drivers in the way that a
+    * glFlush does, but that is required here:
+    */
+   ctx->Driver.Flush(ctx);
 
    _glthread_LOCK_MUTEX(fb->Mutex);
    if (texObj) {
@@ -1380,6 +1395,10 @@ _mesa_FramebufferRenderbufferEXT(GLenum target, GLenum attachment,
    }
 
    FLUSH_VERTICES(ctx, _NEW_BUFFERS);
+   /* The above doesn't fully flush the drivers in the way that a
+    * glFlush does, but that is required here:
+    */
+   ctx->Driver.Flush(ctx);
 
    assert(ctx->Driver.FramebufferRenderbuffer);
    ctx->Driver.FramebufferRenderbuffer(ctx, fb, attachment, rb);
@@ -1443,6 +1462,10 @@ _mesa_GetFramebufferAttachmentParameterivEXT(GLenum target, GLenum attachment,
    }
 
    FLUSH_VERTICES(ctx, _NEW_BUFFERS);
+   /* The above doesn't fully flush the drivers in the way that a
+    * glFlush does, but that is required here:
+    */
+   ctx->Driver.Flush(ctx);
 
    switch (pname) {
    case GL_FRAMEBUFFER_ATTACHMENT_OBJECT_TYPE_EXT:

src/mesa/main/imports.h

@@ -604,6 +604,17 @@ do {                                                                    \
 #endif
 
 
+/**
+ * Return 1 if this is a little endian machine, 0 if big endian.
+ */
+static INLINE GLboolean
+_mesa_little_endian(void)
+{
+   const GLuint ui = 1; /* intentionally not static */
+   return *((const GLubyte *) &ui);
+}
+
+
 
 /**********************************************************************
  * Functions

src/mesa/main/teximage.c

@@ -658,7 +658,7 @@ _mesa_free_texture_image_data(GLcontext *ctx,
 void
 _mesa_delete_texture_image( GLcontext *ctx, struct gl_texture_image *texImage )
 {
-   if (texImage->Data) {
+   if (texImage->Data || ctx->Driver.FreeTexImageData) {
       ctx->Driver.FreeTexImageData( ctx, texImage );
    }
    ASSERT(texImage->Data == NULL);

src/mesa/main/texstore.c

@@ -63,11 +63,21 @@
 #include "texformat.h"
 #include "teximage.h"
 #include "texstore.h"
+#include "enums.h"
 
 
-static const GLint ZERO = 4, ONE = 5;
+enum {
+   ZERO = 4, 
+   ONE = 5
+};
 
-static GLboolean can_swizzle(GLenum logicalBaseFormat)
+
+/**
+ * Return GL_TRUE if the given image format is one that be converted
+ * to another format by swizzling.
+ */
+static GLboolean
+can_swizzle(GLenum logicalBaseFormat)
 {
    switch (logicalBaseFormat) {
    case GL_RGBA:
@@ -76,6 +86,12 @@ static GLboolean can_swizzle(GLenum logicalBaseFormat)
    case GL_INTENSITY:
    case GL_ALPHA:
    case GL_LUMINANCE:
+   case GL_RED:
+   case GL_GREEN:
+   case GL_BLUE:
+   case GL_BGR:
+   case GL_BGRA:
+   case GL_ABGR_EXT:
       return GL_TRUE;
    default:
       return GL_FALSE;
@@ -83,89 +99,172 @@ static GLboolean can_swizzle(GLenum logicalBaseFormat)
 }
 
 
+
+enum {
+   IDX_LUMINANCE = 0,
+   IDX_ALPHA,
+   IDX_INTENSITY,
+   IDX_LUMINANCE_ALPHA,
+   IDX_RGB,
+   IDX_RGBA,
+   IDX_RED,
+   IDX_GREEN,
+   IDX_BLUE,
+   IDX_BGR,
+   IDX_BGRA,
+   IDX_ABGR,
+   MAX_IDX
+};
+
+#define MAP1(x)       MAP4(x, ZERO, ZERO, ZERO)
+#define MAP2(x,y)     MAP4(x, y, ZERO, ZERO)
+#define MAP3(x,y,z)   MAP4(x, y, z, ZERO)
+#define MAP4(x,y,z,w) { x, y, z, w, ZERO, ONE }
+
+
+static const struct {
+   GLubyte format_idx;
+   GLubyte to_rgba[6];
+   GLubyte from_rgba[6];
+} mappings[MAX_IDX] = 
+{
+   {
+      IDX_LUMINANCE,
+      MAP4(0,0,0,ONE),
+      MAP1(0)
+   },
+
+   {
+      IDX_ALPHA,
+      MAP4(ZERO, ZERO, ZERO, 0),
+      MAP1(3)
+   },
+
+   {
+      IDX_INTENSITY,
+      MAP4(0, 0, 0, 0),
+      MAP1(0),
+   },
+
+   {
+      IDX_LUMINANCE_ALPHA,
+      MAP4(0,0,0,1),
+      MAP2(0,3)
+   },
+
+   {
+      IDX_RGB,
+      MAP4(0,1,2,ONE),
+      MAP3(0,1,2)
+   },
+
+   {
+      IDX_RGBA,
+      MAP4(0,1,2,3),
+      MAP4(0,1,2,3),
+   },
+
+
+   {
+      IDX_RED,
+      MAP4(0, ZERO, ZERO, ONE),
+      MAP1(0),
+   },
+
+   {
+      IDX_GREEN,
+      MAP4(ZERO, 0, ZERO, ONE),
+      MAP1(1),
+   },
+
+   {
+      IDX_BLUE,
+      MAP4(ZERO, ZERO, 0, ONE),
+      MAP1(2),
+   },
+
+   {
+      IDX_BGR,
+      MAP4(2,1,0,ONE),
+      MAP3(2,1,0)
+   },
+
+   {
+      IDX_BGRA,
+      MAP4(2,1,0,3),
+      MAP4(2,1,0,3)
+   },
+
+   {
+      IDX_ABGR,
+      MAP4(3,2,1,0),
+      MAP4(3,2,1,0)
+   },
+};
+
+
+
+/**
+ * Convert a GL image format enum to an IDX_* value (see above).
+ */
+static int
+get_map_idx(GLenum value)
+{
+   switch (value) {
+   case GL_LUMINANCE: return IDX_LUMINANCE;
+   case GL_ALPHA: return IDX_ALPHA;
+   case GL_INTENSITY: return IDX_INTENSITY;
+   case GL_LUMINANCE_ALPHA: return IDX_LUMINANCE_ALPHA;
+   case GL_RGB: return IDX_RGB;
+   case GL_RGBA: return IDX_RGBA;
+   case GL_RED: return IDX_RED;
+   case GL_GREEN: return IDX_GREEN;
+   case GL_BLUE: return IDX_BLUE;
+   case GL_BGR: return IDX_BGR;
+   case GL_BGRA: return IDX_BGRA;
+   case GL_ABGR_EXT: return IDX_ABGR;
+   default:
+      _mesa_problem(NULL, "Unexpected inFormat");
+      return 0;
+   }
+}   
+
+
 /**
  * When promoting texture formats (see below) we need to compute the
  * mapping of dest components back to source components.
  * This function does that.
- * \param logicalBaseFormat  the logical format of the texture
- * \param textureBaseFormat  the final texture format
- * \return map[4]  the four mapping values
+ * \param inFormat  the incoming format of the texture
+ * \param outFormat  the final texture format
+ * \return map[6]  a full 6-component map
  */
 static void
-compute_component_mapping(GLenum logicalBaseFormat, GLenum textureBaseFormat,
-                          GLubyte map[6])
+compute_component_mapping(GLenum inFormat, GLenum outFormat, 
+			  GLubyte *map)
 {
-   map[ZERO] = ZERO;
-   map[ONE] = ONE;
+   const int inFmt = get_map_idx(inFormat);
+   const int outFmt = get_map_idx(outFormat);
+   const GLubyte *in2rgba = mappings[inFmt].to_rgba;
+   const GLubyte *rgba2out = mappings[outFmt].from_rgba;
+   int i;
+   
+   for (i = 0; i < 4; i++)
+      map[i] = in2rgba[rgba2out[i]];
 
-   /* compute mapping from dest components back to src components */
-   switch (textureBaseFormat) {
-   case GL_RGB:
-   case GL_RGBA:
-      switch (logicalBaseFormat) {
-      case GL_LUMINANCE:
-         map[0] = map[1] = map[2] = 0;
-         if (textureBaseFormat == GL_RGBA)
-            map[3] = ONE;
-         break;
-      case GL_ALPHA:
-         ASSERT(textureBaseFormat == GL_RGBA);
-         map[0] = map[1] = map[2] = ZERO;
-         map[3] = 0;
-         break;
-      case GL_INTENSITY:
-         map[0] = map[1] = map[2] = 0;
-         if (textureBaseFormat == GL_RGBA)
-            map[3] = 0;
-         break;
-      case GL_LUMINANCE_ALPHA:
-         ASSERT(textureBaseFormat == GL_RGBA);
-         map[0] = map[1] = map[2] = 0;
-         map[3] = 1;
-         break;
-      case GL_RGB:
-         ASSERT(textureBaseFormat == GL_RGBA);
-         map[0] = 0;
-         map[1] = 1;
-         map[2] = 2;
-         map[3] = ONE;
-         break;
-      case GL_RGBA:
-         ASSERT(textureBaseFormat == GL_RGBA);
-         map[0] = 0;
-         map[1] = 1;
-         map[2] = 2;
-         map[3] = 3;
-         break;
-      default:
-         _mesa_problem(NULL, "Unexpected logicalBaseFormat");
-         map[0] = map[1] = map[2] = map[3] = 0;
-      }
-      break;
-   case GL_LUMINANCE_ALPHA:
-      switch (logicalBaseFormat) {
-      case GL_LUMINANCE:
-         map[0] = 0;
-         map[1] = ONE;
-         break;
-      case GL_ALPHA:
-         map[0] = ZERO;
-         map[1] = 0;
-         break;
-      case GL_INTENSITY:
-         map[0] = 0;
-         map[1] = 0;
-         break;
-      default:
-         _mesa_problem(NULL, "Unexpected logicalBaseFormat");
-         map[0] = map[1] = 0;
-      }
-      break;
-   default:
-      _mesa_problem(NULL, "Unexpected logicalBaseFormat");
-      map[0] = map[1] = 0;
-      break;
-   }   
+   map[ZERO] = ZERO;
+   map[ONE] = ONE;   
+
+/*
+   _mesa_printf("from %x/%s to %x/%s map %d %d %d %d %d %d\n",
+		inFormat, _mesa_lookup_enum_by_nr(inFormat),
+		outFormat, _mesa_lookup_enum_by_nr(outFormat),
+		map[0], 
+		map[1], 
+		map[2], 
+		map[3], 
+		map[4], 
+		map[5]); 
+*/
 }
 
 
@@ -552,14 +651,16 @@ _mesa_make_temp_chan_image(GLcontext *ctx, GLuint dims,
  * \param dstComponents  number of color components in destination pixels
  * \param src  source pixels
  * \param srcComponents  number of color components in source pixels
- * \param map  the swizzle mapping
+ * \param map  the swizzle mapping.  map[X] says where to find the X component
+ *             in the source image's pixels.  For example, if the source image
+ *             is GL_BGRA and X = red, map[0] yields 2.
  * \param count  number of pixels to copy/swizzle.
  */
 static void
 swizzle_copy(GLubyte *dst, GLuint dstComponents, const GLubyte *src, 
              GLuint srcComponents, const GLubyte *map, GLuint count)
 {
-   GLubyte tmp[8];
+   GLubyte tmp[6];
    GLuint i;
 
    tmp[ZERO] = 0x0;
@@ -596,10 +697,60 @@ swizzle_copy(GLubyte *dst, GLuint dstComponents, const GLubyte *src,
 	 dst += 2;
       }
       break;
+   case 1:
+      for (i = 0; i < count; i++) {
+ 	 COPY_4UBV(tmp, src); 
+	 src += srcComponents;      
+	 dst[0] = tmp[map[0]];
+	 dst += 1;
+      }
+      break;
    }
 }
 
 
+static const GLubyte map_identity[6] = { 0, 1, 2, 3, ZERO, ONE };
+static const GLubyte map_3210[6] = { 3, 2, 1, 0, ZERO, ONE };
+
+/* Deal with the _REV input types:
+ */
+static const GLubyte *
+type_mapping( GLenum srcType )
+{
+   switch (srcType) {
+   case GL_UNSIGNED_BYTE:
+      return map_identity;
+   case GL_UNSIGNED_INT_8_8_8_8:
+      return map_3210;
+   case GL_UNSIGNED_INT_8_8_8_8_REV:
+      return map_identity;
+   default:
+      return NULL;
+   }
+}
+
+/* Mapping required if input type is 
+ */
+static const GLubyte *
+byteswap_mapping( GLboolean swapBytes,
+		  GLenum srcType )
+{
+   if (!swapBytes) 
+      return map_identity;
+
+   switch (srcType) {
+   case GL_UNSIGNED_BYTE:
+      return map_identity;
+   case GL_UNSIGNED_INT_8_8_8_8:
+   case GL_UNSIGNED_INT_8_8_8_8_REV:
+      return map_3210;
+   default:
+      return NULL;
+   }
+}
+
+
+
 /**
  * Transfer a GLubyte texture image with component swizzling.
  */
@@ -607,7 +758,12 @@ static void
 _mesa_swizzle_ubyte_image(GLcontext *ctx, 
 			  GLuint dimensions,
 			  GLenum srcFormat,
-			  const GLubyte *dstmap, GLint dstComponents,
+			  GLenum srcType,
+
+			  GLenum baseInternalFormat,
+
+			  const GLubyte *rgba2dst,
+			  GLuint dstComponents,
 
 			  GLvoid *dstAddr,
 			  GLint dstXoffset, GLint dstYoffset, GLint dstZoffset,
@@ -619,9 +775,9 @@ _mesa_swizzle_ubyte_image(GLcontext *ctx,
 			  const struct gl_pixelstore_attrib *srcPacking )
 {
    GLint srcComponents = _mesa_components_in_format(srcFormat);
-   GLubyte srcmap[6], map[4];
+   const GLubyte *srctype2ubyte, *swap;
+   GLubyte map[4], src2base[6], base2rgba[6];
    GLint i;
-
    const GLint srcRowStride =
       _mesa_image_row_stride(srcPacking, srcWidth,
                              srcFormat, GL_UNSIGNED_BYTE);
@@ -635,10 +791,20 @@ _mesa_swizzle_ubyte_image(GLcontext *ctx,
 
    (void) ctx;
 
-   compute_component_mapping(srcFormat, GL_RGBA, srcmap);
+   /* Translate from src->baseInternal->GL_RGBA->dst.  This will
+    * correctly deal with RGBA->RGB->RGBA conversions where the final
+    * A value must be 0xff regardless of the incoming alpha values.
+    */
+   compute_component_mapping(srcFormat, baseInternalFormat, src2base);
+   compute_component_mapping(baseInternalFormat, GL_RGBA, base2rgba);
+   swap = byteswap_mapping(srcPacking->SwapBytes, srcType);
+   srctype2ubyte = type_mapping(srcType);
+
 
    for (i = 0; i < 4; i++)
-      map[i] = srcmap[dstmap[i]];
+      map[i] = srctype2ubyte[swap[src2base[base2rgba[rgba2dst[i]]]]];
+
+/*    _mesa_printf("map %d %d %d %d\n", map[0], map[1], map[2], map[3]);  */
 
    if (srcRowStride == srcWidth * srcComponents &&
        dimensions < 3) {
@@ -817,6 +983,59 @@ _mesa_texstore_rgba(TEXSTORE_PARAMS)
          }
       }
    }
+   else if (!ctx->_ImageTransferState &&
+	    _mesa_little_endian() &&
+	    CHAN_TYPE == GL_UNSIGNED_BYTE &&
+	    (srcType == GL_UNSIGNED_BYTE ||
+	     srcType == GL_UNSIGNED_INT_8_8_8_8 ||
+	     srcType == GL_UNSIGNED_INT_8_8_8_8_REV) &&
+	    can_swizzle(baseInternalFormat) &&
+	    can_swizzle(srcFormat)) {
+
+      const GLubyte *dstmap;
+      GLuint components;
+
+      /* dstmap - how to swizzle from RGBA to dst format:
+       */
+      if (dstFormat == &_mesa_texformat_rgba) {
+	 dstmap = mappings[IDX_RGBA].from_rgba;
+	 components = 4;
+      }
+      else if (dstFormat == &_mesa_texformat_rgb) {
+	 dstmap = mappings[IDX_RGB].from_rgba;
+	 components = 3;
+      }
+      else if (dstFormat == &_mesa_texformat_alpha) {
+	 dstmap = mappings[IDX_ALPHA].from_rgba;
+	 components = 1;
+      }
+      else if (dstFormat == &_mesa_texformat_luminance) {
+	 dstmap = mappings[IDX_LUMINANCE].from_rgba;
+	 components = 1;
+      }
+      else if (dstFormat == &_mesa_texformat_luminance_alpha) {
+	 dstmap = mappings[IDX_LUMINANCE_ALPHA].from_rgba;
+	 components = 2;
+      }
+      else if (dstFormat == &_mesa_texformat_intensity) {
+	 dstmap = mappings[IDX_INTENSITY].from_rgba;
+	 components = 1;
+      }
+      else {
+         _mesa_problem(ctx, "Unexpected dstFormat in _mesa_texstore_rgba");
+         return GL_FALSE;
+      }
+
+      _mesa_swizzle_ubyte_image(ctx, dims,
+				srcFormat,
+				srcType,
+				baseInternalFormat,
+				dstmap, components,
+				dstAddr, dstXoffset, dstYoffset, dstZoffset,
+				dstRowStride, dstImageOffsets,
+				srcWidth, srcHeight, srcDepth, srcAddr,
+				srcPacking);      
+   }
    else {
       /* general path */
       const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
@@ -1050,13 +1269,14 @@ _mesa_texstore_rgb565(TEXSTORE_PARAMS)
 }
 
 
+/**
+ * Store a texture in MESA_FORMAT_RGBA8888 or MESA_FORMAT_RGBA8888_REV.
+ */
 GLboolean
 _mesa_texstore_rgba8888(TEXSTORE_PARAMS)
 {
-   const GLuint ui = 1;
-   const GLubyte littleEndian = *((const GLubyte *) &ui);
+   const GLboolean littleEndian = _mesa_little_endian();
 
-   (void)littleEndian;
    ASSERT(dstFormat == &_mesa_texformat_rgba8888 ||
           dstFormat == &_mesa_texformat_rgba8888_rev);
    ASSERT(dstFormat->TexelBytes == 4);
@@ -1066,7 +1286,25 @@ _mesa_texstore_rgba8888(TEXSTORE_PARAMS)
        dstFormat == &_mesa_texformat_rgba8888 &&
        baseInternalFormat == GL_RGBA &&
       ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
-       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV))) {
+       (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && !littleEndian) ||
+       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
+       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && littleEndian))) {
+       /* simple memcpy path */
+      memcpy_texture(ctx, dims,
+                     dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
+                     dstRowStride,
+                     dstImageOffsets,
+                     srcWidth, srcHeight, srcDepth, srcFormat, srcType,
+                     srcAddr, srcPacking);
+   }
+   else if (!ctx->_ImageTransferState &&
+       !srcPacking->SwapBytes &&
+       dstFormat == &_mesa_texformat_rgba8888_rev &&
+       baseInternalFormat == GL_RGBA &&
+      ((srcFormat == GL_RGBA && srcType == GL_UNSIGNED_INT_8_8_8_8_REV) ||
+       (srcFormat == GL_RGBA && srcType == GL_UNSIGNED_BYTE && littleEndian) ||
+       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_INT_8_8_8_8) ||
+       (srcFormat == GL_ABGR_EXT && srcType == GL_UNSIGNED_BYTE && !littleEndian))) {
       /* simple memcpy path */
       memcpy_texture(ctx, dims,
                      dstFormat, dstAddr, dstXoffset, dstYoffset, dstZoffset,
@@ -1075,40 +1313,41 @@ _mesa_texstore_rgba8888(TEXSTORE_PARAMS)
                      srcWidth, srcHeight, srcDepth, srcFormat, srcType,
                      srcAddr, srcPacking);
    }
-#if 0
-   /* broken? */
    else if (!ctx->_ImageTransferState &&
-	    !srcPacking->SwapBytes &&
-	    srcType == GL_UNSIGNED_BYTE && 
-	    dstFormat == &_mesa_texformat_rgba8888 &&
-	    littleEndian &&
-	    /* Three texture formats involved: srcFormat,
-	     * baseInternalFormat and destFormat (GL_RGBA). Only two
-	     * may differ. _mesa_swizzle_ubyte_image can't handle two
-	     * propagations at once correctly. */
-	    (srcFormat == baseInternalFormat ||
-	     baseInternalFormat == GL_RGBA) &&
+	    littleEndian && 
+	    (srcType == GL_UNSIGNED_BYTE ||
+	     srcType == GL_UNSIGNED_INT_8_8_8_8 ||
+	     srcType == GL_UNSIGNED_INT_8_8_8_8_REV) &&
+	    can_swizzle(baseInternalFormat) &&
 	    can_swizzle(srcFormat)) {
+
       GLubyte dstmap[4];
 
-      /* dstmap - how to swizzle from GL_RGBA to dst format:
-       *
-       * FIXME - add !litteEndian and _rev varients:
+      /* dstmap - how to swizzle from RGBA to dst format:
        */
-      dstmap[3] = 0;
-      dstmap[2] = 1;
-      dstmap[1] = 2;
-      dstmap[0] = 3;
+      if (dstFormat == &_mesa_texformat_rgba8888) {
+	 dstmap[3] = 0;
+	 dstmap[2] = 1;
+	 dstmap[1] = 2;
+	 dstmap[0] = 3;
+      }
+      else {
+	 dstmap[3] = 3;
+	 dstmap[2] = 2;
+	 dstmap[1] = 1;
+	 dstmap[0] = 0;
+      }
       
       _mesa_swizzle_ubyte_image(ctx, dims,
 				srcFormat,
+				srcType,
+				baseInternalFormat,
 				dstmap, 4,
 				dstAddr, dstXoffset, dstYoffset, dstZoffset,
-				dstRowStride, dstImageStride,
+				dstRowStride, dstImageOffsets,
 				srcWidth, srcHeight, srcDepth, srcAddr,
 				srcPacking);      
    }
-#endif
    else {
       /* general path */
       const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
@@ -1159,8 +1398,7 @@ _mesa_texstore_rgba8888(TEXSTORE_PARAMS)
 GLboolean
 _mesa_texstore_argb8888(TEXSTORE_PARAMS)
 {
-   const GLuint ui = 1;
-   const GLubyte littleEndian = *((const GLubyte *) &ui);
+   const GLboolean littleEndian = _mesa_little_endian();
 
    ASSERT(dstFormat == &_mesa_texformat_argb8888 ||
           dstFormat == &_mesa_texformat_argb8888_rev);
@@ -1200,6 +1438,8 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS)
             !srcPacking->SwapBytes &&
 	    dstFormat == &_mesa_texformat_argb8888 &&
             srcFormat == GL_RGB &&
+	    (baseInternalFormat == GL_RGBA ||
+	     baseInternalFormat == GL_RGB) &&
             srcType == GL_UNSIGNED_BYTE) {
 
       int img, row, col;
@@ -1228,6 +1468,7 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS)
             !srcPacking->SwapBytes &&
 	    dstFormat == &_mesa_texformat_argb8888 &&
             srcFormat == GL_RGBA &&
+	    baseInternalFormat == GL_RGBA &&
             (srcType == GL_UNSIGNED_BYTE && littleEndian)) {
       GLint img, row, col;
       /* For some reason, streaming copies to write-combined regions
@@ -1262,6 +1503,7 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS)
             !srcPacking->SwapBytes &&
 	    dstFormat == &_mesa_texformat_argb8888 &&
             srcFormat == GL_RGBA &&
+	    baseInternalFormat == GL_RGBA &&
             srcType == GL_UNSIGNED_BYTE) {
 
       GLint img, row, col;
@@ -1287,29 +1529,36 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS)
       }
    }
    else if (!ctx->_ImageTransferState &&
-	    !srcPacking->SwapBytes &&
-	    dstFormat == &_mesa_texformat_argb8888 &&
-	    srcType == GL_UNSIGNED_BYTE && 
 	    littleEndian &&
-	    /* Three texture formats involved: srcFormat,
-	     * baseInternalFormat and destFormat (GL_RGBA). Only two
-	     * may differ. _mesa_swizzle_ubyte_image can't handle two
-	     * propagations at once correctly. */
-	    (srcFormat == baseInternalFormat ||
-	     baseInternalFormat == GL_RGBA) &&
+	    (srcType == GL_UNSIGNED_BYTE ||
+	     srcType == GL_UNSIGNED_INT_8_8_8_8 ||
+	     srcType == GL_UNSIGNED_INT_8_8_8_8_REV) &&
+	    can_swizzle(baseInternalFormat) &&	   
 	    can_swizzle(srcFormat)) {
 
       GLubyte dstmap[4];
 
-      /* dstmap - how to swizzle from GL_RGBA to dst format:
+      /* dstmap - how to swizzle from RGBA to dst format:
        */
-      dstmap[3] = 3;		/* alpha */
-      dstmap[2] = 0;		/* red */
-      dstmap[1] = 1;		/* green */
-      dstmap[0] = 2;		/* blue */
+      if (dstFormat == &_mesa_texformat_argb8888) {
+	 dstmap[3] = 3;		/* alpha */
+	 dstmap[2] = 0;		/* red */
+	 dstmap[1] = 1;		/* green */
+	 dstmap[0] = 2;		/* blue */
+      }
+      else {
+	 assert(dstFormat == &_mesa_texformat_argb8888_rev);
+	 dstmap[3] = 2;
+	 dstmap[2] = 1;
+	 dstmap[1] = 0;
+	 dstmap[0] = 3;
+      }
  
       _mesa_swizzle_ubyte_image(ctx, dims,
 				srcFormat,
+				srcType,
+
+				baseInternalFormat,
 				dstmap, 4,
 				dstAddr, dstXoffset, dstYoffset, dstZoffset,
 				dstRowStride,
@@ -1367,8 +1616,7 @@ _mesa_texstore_argb8888(TEXSTORE_PARAMS)
 GLboolean
 _mesa_texstore_rgb888(TEXSTORE_PARAMS)
 {
-   const GLuint ui = 1;
-   const GLubyte littleEndian = *((const GLubyte *) &ui);
+   const GLboolean littleEndian = _mesa_little_endian();
 
    ASSERT(dstFormat == &_mesa_texformat_rgb888);
    ASSERT(dstFormat->TexelBytes == 3);
@@ -1413,6 +1661,31 @@ _mesa_texstore_rgb888(TEXSTORE_PARAMS)
          }
       }
    }
+   else if (!ctx->_ImageTransferState &&
+	    littleEndian &&
+	    srcType == GL_UNSIGNED_BYTE &&
+	    can_swizzle(baseInternalFormat) &&
+	    can_swizzle(srcFormat)) {
+
+      GLubyte dstmap[4];
+
+      /* dstmap - how to swizzle from RGBA to dst format:
+       */
+      dstmap[0] = 2;
+      dstmap[1] = 1;
+      dstmap[2] = 0;
+      dstmap[3] = ONE;		/* ? */
+      
+      _mesa_swizzle_ubyte_image(ctx, dims,
+				srcFormat,
+				srcType,
+				baseInternalFormat,
+				dstmap, 3,
+				dstAddr, dstXoffset, dstYoffset, dstZoffset,
+				dstRowStride, dstImageOffsets,
+				srcWidth, srcHeight, srcDepth, srcAddr,
+				srcPacking);      
+   }
    else {
       /* general path */
       const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
@@ -1469,8 +1742,7 @@ _mesa_texstore_rgb888(TEXSTORE_PARAMS)
 GLboolean
 _mesa_texstore_bgr888(TEXSTORE_PARAMS)
 {
-   const GLuint ui = 1;
-   const GLubyte littleEndian = *((const GLubyte *) &ui);
+   const GLboolean littleEndian = _mesa_little_endian();
 
    ASSERT(dstFormat == &_mesa_texformat_bgr888);
    ASSERT(dstFormat->TexelBytes == 3);
@@ -1515,6 +1787,31 @@ _mesa_texstore_bgr888(TEXSTORE_PARAMS)
          }
       }
    }
+   else if (!ctx->_ImageTransferState &&
+	    littleEndian &&
+	    srcType == GL_UNSIGNED_BYTE &&
+	    can_swizzle(baseInternalFormat) &&
+	    can_swizzle(srcFormat)) {
+
+      GLubyte dstmap[4];
+
+      /* dstmap - how to swizzle from RGBA to dst format:
+       */
+      dstmap[0] = 0;
+      dstmap[1] = 1;
+      dstmap[2] = 2;
+      dstmap[3] = ONE;		/* ? */
+      
+      _mesa_swizzle_ubyte_image(ctx, dims,
+				srcFormat,
+				srcType,
+				baseInternalFormat,
+				dstmap, 3,
+				dstAddr, dstXoffset, dstYoffset, dstZoffset,
+				dstRowStride, dstImageOffsets,
+				srcWidth, srcHeight, srcDepth, srcAddr,
+				srcPacking);      
+   }   
    else {
       /* general path */
       const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
@@ -1689,8 +1986,7 @@ _mesa_texstore_argb1555(TEXSTORE_PARAMS)
 GLboolean
 _mesa_texstore_al88(TEXSTORE_PARAMS)
 {
-   const GLuint ui = 1;
-   const GLubyte littleEndian = *((const GLubyte *) &ui);
+   const GLboolean littleEndian = _mesa_little_endian();
 
    ASSERT(dstFormat == &_mesa_texformat_al88 ||
           dstFormat == &_mesa_texformat_al88_rev);
@@ -1711,6 +2007,37 @@ _mesa_texstore_al88(TEXSTORE_PARAMS)
                      srcWidth, srcHeight, srcDepth, srcFormat, srcType,
                      srcAddr, srcPacking);
    }
+   else if (!ctx->_ImageTransferState &&
+	    littleEndian &&
+	    srcType == GL_UNSIGNED_BYTE &&
+	    can_swizzle(baseInternalFormat) &&
+	    can_swizzle(srcFormat)) {
+
+      GLubyte dstmap[4];
+
+      /* dstmap - how to swizzle from RGBA to dst format:
+       */
+      if (dstFormat == &_mesa_texformat_al88) {
+	 dstmap[0] = 0;
+	 dstmap[1] = 3;
+      }
+      else {
+	 dstmap[0] = 3;
+	 dstmap[1] = 0;
+      }
+      dstmap[2] = ZERO;		/* ? */
+      dstmap[3] = ONE;		/* ? */
+      
+      _mesa_swizzle_ubyte_image(ctx, dims,
+				srcFormat,
+				srcType,
+				baseInternalFormat,
+				dstmap, 2,
+				dstAddr, dstXoffset, dstYoffset, dstZoffset,
+				dstRowStride, dstImageOffsets,
+				srcWidth, srcHeight, srcDepth, srcAddr,
+				srcPacking);      
+   }   
    else {
       /* general path */
       const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
@@ -1831,6 +2158,36 @@ _mesa_texstore_a8(TEXSTORE_PARAMS)
                      srcWidth, srcHeight, srcDepth, srcFormat, srcType,
                      srcAddr, srcPacking);
    }
+   else if (!ctx->_ImageTransferState &&
+	    _mesa_little_endian() &&
+	    srcType == GL_UNSIGNED_BYTE &&
+	    can_swizzle(baseInternalFormat) &&
+	    can_swizzle(srcFormat)) {
+
+      GLubyte dstmap[4];
+
+      /* dstmap - how to swizzle from RGBA to dst format:
+       */
+      if (dstFormat == &_mesa_texformat_a8) {
+	 dstmap[0] = 3;
+      }
+      else {
+	 dstmap[0] = 0;
+      }
+      dstmap[1] = ZERO;		/* ? */
+      dstmap[2] = ZERO;		/* ? */
+      dstmap[3] = ONE;		/* ? */
+      
+      _mesa_swizzle_ubyte_image(ctx, dims,
+				srcFormat,
+				srcType,
+				baseInternalFormat,
+				dstmap, 1,
+				dstAddr, dstXoffset, dstYoffset, dstZoffset,
+				dstRowStride, dstImageOffsets,
+				srcWidth, srcHeight, srcDepth, srcAddr,
+				srcPacking);      
+   }   
    else {
       /* general path */
       const GLchan *tempImage = _mesa_make_temp_chan_image(ctx, dims,
@@ -1912,8 +2269,7 @@ _mesa_texstore_ci8(TEXSTORE_PARAMS)
 GLboolean
 _mesa_texstore_ycbcr(TEXSTORE_PARAMS)
 {
-   const GLuint ui = 1;
-   const GLubyte littleEndian = *((const GLubyte *) &ui);
+   const GLboolean littleEndian = _mesa_little_endian();
    (void) ctx; (void) dims; (void) baseInternalFormat;
 
    ASSERT((dstFormat == &_mesa_texformat_ycbcr) ||
@@ -2164,6 +2520,113 @@ _mesa_texstore_rgba_float16(TEXSTORE_PARAMS)
 }
 
 
+#if FEATURE_EXT_texture_sRGB
+GLboolean
+_mesa_texstore_srgb8(TEXSTORE_PARAMS)
+{
+   const GLboolean littleEndian = _mesa_little_endian();
+   const struct gl_texture_format *newDstFormat;
+   StoreTexImageFunc store;
+   GLboolean k;
+
+   ASSERT(dstFormat == &_mesa_texformat_srgb8);
+
+   /* reuse normal rgb texstore code */
+   if (littleEndian) {
+      newDstFormat = &_mesa_texformat_bgr888;
+      store = _mesa_texstore_bgr888;
+   }
+   else {
+      newDstFormat = &_mesa_texformat_rgb888;
+      store = _mesa_texstore_rgb888;
+   }
+
+   k = store(ctx, dims, baseInternalFormat,
+             newDstFormat, dstAddr,
+             dstXoffset, dstYoffset, dstZoffset,
+             dstRowStride, dstImageOffsets,
+             srcWidth, srcHeight, srcDepth,
+             srcFormat, srcType,
+             srcAddr, srcPacking);
+   return k;
+}
+
+
+GLboolean
+_mesa_texstore_srgba8(TEXSTORE_PARAMS)
+{
+   const GLboolean littleEndian = _mesa_little_endian();
+   const struct gl_texture_format *newDstFormat;
+   GLboolean k;
+
+   ASSERT(dstFormat == &_mesa_texformat_srgba8);
+
+   /* reuse normal rgba texstore code */
+   if (littleEndian)
+      newDstFormat = &_mesa_texformat_rgba8888_rev;
+   else
+      newDstFormat = &_mesa_texformat_rgba8888;
+
+   k = _mesa_texstore_rgba8888(ctx, dims, baseInternalFormat,
+                               newDstFormat, dstAddr,
+                               dstXoffset, dstYoffset, dstZoffset,
+                               dstRowStride, dstImageOffsets,
+                               srcWidth, srcHeight, srcDepth,
+                               srcFormat, srcType,
+                               srcAddr, srcPacking);
+   return k;
+}
+
+
+GLboolean
+_mesa_texstore_sl8(TEXSTORE_PARAMS)
+{
+   const struct gl_texture_format *newDstFormat;
+   GLboolean k;
+
+   ASSERT(dstFormat == &_mesa_texformat_sl8);
+
+   newDstFormat = &_mesa_texformat_l8;
+
+   /* _mesa_textore_a8 handles luminance8 too */
+   k = _mesa_texstore_a8(ctx, dims, baseInternalFormat,
+                         newDstFormat, dstAddr,
+                         dstXoffset, dstYoffset, dstZoffset,
+                         dstRowStride, dstImageOffsets,
+                         srcWidth, srcHeight, srcDepth,
+                         srcFormat, srcType,
+                         srcAddr, srcPacking);
+   return k;
+}
+
+
+GLboolean
+_mesa_texstore_sla8(TEXSTORE_PARAMS)
+{
+   const GLboolean littleEndian = _mesa_little_endian();
+   const struct gl_texture_format *newDstFormat;
+   GLboolean k;
+
+   ASSERT(dstFormat == &_mesa_texformat_sla8);
+
+   /* reuse normal luminance/alpha texstore code */
+   if (littleEndian)
+      newDstFormat = &_mesa_texformat_al88;
+   else
+      newDstFormat = &_mesa_texformat_al88_rev;
+
+   k = _mesa_texstore_al88(ctx, dims, baseInternalFormat,
+                           newDstFormat, dstAddr,
+                           dstXoffset, dstYoffset, dstZoffset,
+                           dstRowStride, dstImageOffsets,
+                           srcWidth, srcHeight, srcDepth,
+                           srcFormat, srcType,
+                           srcAddr, srcPacking);
+   return k;
+}
+
+#endif /* FEATURE_EXT_texture_sRGB */
+
 
 /**
  * Check if an unpack PBO is active prior to fetching a texture image.

src/mesa/swrast/s_context.c

@@ -98,7 +98,7 @@ _swrast_update_rasterflags( GLcontext *ctx )
       rasterMask |= MULTI_DRAW_BIT; /* all color index bits disabled */
    }
 
-   if (ctx->FragmentProgram._Active) {
+   if (ctx->FragmentProgram._Enabled) {
       rasterMask |= FRAGPROG_BIT;
    }
 
@@ -206,7 +206,7 @@ _swrast_update_fog_state( GLcontext *ctx )
 
    /* determine if fog is needed, and if so, which fog mode */
    swrast->_FogEnabled = GL_FALSE;
-   if (ctx->FragmentProgram._Active) {
+   if (ctx->FragmentProgram._Enabled) {
       if (ctx->FragmentProgram._Current->Base.Target==GL_FRAGMENT_PROGRAM_ARB) {
          const struct gl_fragment_program *fp
             = ctx->FragmentProgram._Current;
@@ -230,7 +230,7 @@ _swrast_update_fog_state( GLcontext *ctx )
 static void
 _swrast_update_fragment_program( GLcontext *ctx )
 {
-   if (ctx->FragmentProgram._Active) {
+   if (ctx->FragmentProgram._Enabled) {
       const struct gl_fragment_program *fp = ctx->FragmentProgram._Current;
       _mesa_load_state_parameters(ctx, fp->Base.Parameters);
    }
@@ -299,7 +299,7 @@ _swrast_validate_triangle( GLcontext *ctx,
 
    if (ctx->Texture._EnabledUnits == 0
        && NEED_SECONDARY_COLOR(ctx)
-       && !ctx->FragmentProgram._Active) {
+       && !ctx->FragmentProgram._Enabled) {
       /* separate specular color, but no texture */
       swrast->SpecTriangle = swrast->Triangle;
       swrast->Triangle = _swrast_add_spec_terms_triangle;
@@ -322,7 +322,7 @@ _swrast_validate_line( GLcontext *ctx, const SWvertex *v0, const SWvertex *v1 )
 
    if (ctx->Texture._EnabledUnits == 0
        && NEED_SECONDARY_COLOR(ctx)
-       && !ctx->FragmentProgram._Active) {
+       && !ctx->FragmentProgram._Enabled) {
       swrast->SpecLine = swrast->Line;
       swrast->Line = _swrast_add_spec_terms_line;
    }
@@ -345,7 +345,7 @@ _swrast_validate_point( GLcontext *ctx, const SWvertex *v0 )
 
    if (ctx->Texture._EnabledUnits == 0
        && NEED_SECONDARY_COLOR(ctx)
-       && !ctx->FragmentProgram._Active) {
+       && !ctx->FragmentProgram._Enabled) {
       swrast->SpecPoint = swrast->Point;
       swrast->Point = _swrast_add_spec_terms_point;
    }

src/mesa/swrast/s_span.c

@@ -129,7 +129,7 @@ _swrast_span_default_texcoords( GLcontext *ctx, struct sw_span *span )
    GLuint i;
    for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
       const GLfloat *tc = ctx->Current.RasterTexCoords[i];
-      if (ctx->FragmentProgram._Active || ctx->ATIFragmentShader._Enabled) {
+      if (ctx->FragmentProgram._Enabled || ctx->ATIFragmentShader._Enabled) {
          COPY_4V(span->tex[i], tc);
       }
       else if (tc[3] > 0.0F) {
@@ -410,7 +410,7 @@ interpolate_texcoords(GLcontext *ctx, struct sw_span *span)
             if (obj) {
                const struct gl_texture_image *img = obj->Image[0][obj->BaseLevel];
                needLambda = (obj->MinFilter != obj->MagFilter)
-                  || ctx->FragmentProgram._Active;
+                  || ctx->FragmentProgram._Enabled;
                texW = img->WidthScale;
                texH = img->HeightScale;
             }
@@ -435,7 +435,7 @@ interpolate_texcoords(GLcontext *ctx, struct sw_span *span)
                GLfloat r = span->tex[u][2];
                GLfloat q = span->tex[u][3];
                GLuint i;
-               if (ctx->FragmentProgram._Active || ctx->ATIFragmentShader._Enabled ||
+               if (ctx->FragmentProgram._Enabled || ctx->ATIFragmentShader._Enabled ||
                    ctx->ShaderObjects._FragmentShaderPresent) {
                   /* do perspective correction but don't divide s, t, r by q */
                   const GLfloat dwdx = span->dwdx;
@@ -487,7 +487,7 @@ interpolate_texcoords(GLcontext *ctx, struct sw_span *span)
                GLfloat r = span->tex[u][2];
                GLfloat q = span->tex[u][3];
                GLuint i;
-               if (ctx->FragmentProgram._Active || ctx->ATIFragmentShader._Enabled ||
+               if (ctx->FragmentProgram._Enabled || ctx->ATIFragmentShader._Enabled ||
                    ctx->ShaderObjects._FragmentShaderPresent) {
                   /* do perspective correction but don't divide s, t, r by q */
                   const GLfloat dwdx = span->dwdx;
@@ -546,7 +546,7 @@ interpolate_texcoords(GLcontext *ctx, struct sw_span *span)
       if (obj) {
          const struct gl_texture_image *img = obj->Image[0][obj->BaseLevel];
          needLambda = (obj->MinFilter != obj->MagFilter)
-            || ctx->FragmentProgram._Active;
+            || ctx->FragmentProgram._Enabled;
          texW = (GLfloat) img->WidthScale;
          texH = (GLfloat) img->HeightScale;
       }
@@ -571,7 +571,7 @@ interpolate_texcoords(GLcontext *ctx, struct sw_span *span)
          GLfloat r = span->tex[0][2];
          GLfloat q = span->tex[0][3];
          GLuint i;
-         if (ctx->FragmentProgram._Active || ctx->ATIFragmentShader._Enabled ||
+         if (ctx->FragmentProgram._Enabled || ctx->ATIFragmentShader._Enabled ||
              ctx->ShaderObjects._FragmentShaderPresent) {
             /* do perspective correction but don't divide s, t, r by q */
             const GLfloat dwdx = span->dwdx;
@@ -623,7 +623,7 @@ interpolate_texcoords(GLcontext *ctx, struct sw_span *span)
          GLfloat r = span->tex[0][2];
          GLfloat q = span->tex[0][3];
          GLuint i;
-         if (ctx->FragmentProgram._Active || ctx->ATIFragmentShader._Enabled ||
+         if (ctx->FragmentProgram._Enabled || ctx->ATIFragmentShader._Enabled ||
              ctx->ShaderObjects._FragmentShaderPresent) {
             /* do perspective correction but don't divide s, t, r by q */
             const GLfloat dwdx = span->dwdx;
@@ -1120,7 +1120,7 @@ _swrast_write_rgba_span( GLcontext *ctx, struct sw_span *span)
    const GLbitfield origInterpMask = span->interpMask;
    const GLbitfield origArrayMask = span->arrayMask;
    const GLboolean deferredTexture = !(ctx->Color.AlphaEnabled ||
-                                       ctx->FragmentProgram._Active ||
+                                       ctx->FragmentProgram._Enabled ||
                                        ctx->ShaderObjects._FragmentShaderPresent);
 
    ASSERT(span->primitive == GL_POINT  ||  span->primitive == GL_LINE ||
@@ -1197,15 +1197,12 @@ _swrast_write_rgba_span( GLcontext *ctx, struct sw_span *span)
          interpolate_fog(ctx, span);
 
       /* Compute fragment colors with fragment program or texture lookups */
-#if FEATURE_ARB_fragment_shader
       if (ctx->ShaderObjects._FragmentShaderPresent) {
          if (span->interpMask & SPAN_Z)
             _swrast_span_interpolate_z (ctx, span);
          _swrast_exec_arbshader (ctx, span);
       }
-      else
-#endif
-      if (ctx->FragmentProgram._Active) {
+      else if (ctx->FragmentProgram._Enabled) {
          /* frag prog may need Z values */
          if (span->interpMask & SPAN_Z)
             _swrast_span_interpolate_z(ctx, span);
@@ -1284,15 +1281,12 @@ _swrast_write_rgba_span( GLcontext *ctx, struct sw_span *span)
       if (span->interpMask & SPAN_FOG)
          interpolate_fog(ctx, span);
 
-#if FEATURE_ARB_fragment_shader
       if (ctx->ShaderObjects._FragmentShaderPresent) {
          if (span->interpMask & SPAN_Z)
             _swrast_span_interpolate_z (ctx, span);
          _swrast_exec_arbshader (ctx, span);
       }
-      else
-#endif
-      if (ctx->FragmentProgram._Active)
+      else if (ctx->FragmentProgram._Enabled)
          _swrast_exec_fragment_program( ctx, span );
       else if (ctx->ATIFragmentShader._Enabled)
          _swrast_exec_fragment_shader( ctx, span );

src/mesa/swrast/s_triangle.c

@@ -1072,7 +1072,7 @@ _swrast_choose_triangle( GLcontext *ctx )
          }
       }
 
-      if (ctx->Texture._EnabledCoordUnits || ctx->FragmentProgram._Active ||
+      if (ctx->Texture._EnabledCoordUnits || ctx->FragmentProgram._Enabled ||
           ctx->ATIFragmentShader._Enabled || ctx->ShaderObjects._FragmentShaderPresent) {
          /* Ugh, we do a _lot_ of tests to pick the best textured tri func */
          const struct gl_texture_object *texObj2D;
@@ -1088,7 +1088,7 @@ _swrast_choose_triangle( GLcontext *ctx )
 
          /* First see if we can use an optimized 2-D texture function */
          if (ctx->Texture._EnabledCoordUnits == 0x1
-             && !ctx->FragmentProgram._Active
+             && !ctx->FragmentProgram._Enabled
              && !ctx->ATIFragmentShader._Enabled
              && !ctx->ShaderObjects._FragmentShaderPresent
              && ctx->Texture.Unit[0]._ReallyEnabled == TEXTURE_2D_BIT