graw: Add struct pipe_surface forward declaration.

Fixes this GCC warning.
graw.h:93: warning: 'struct pipe_surface' declared inside parameter list
graw.h:93: warning: its scope is only this definition or declaration,
which is probably not what you want

Makefile

@@ -180,7 +180,7 @@ ultrix-gcc:
 
 # Rules for making release tarballs
 
-VERSION=7.9-devel
+VERSION=7.10-devel
 DIRECTORY = Mesa-$(VERSION)
 LIB_NAME = MesaLib-$(VERSION)
 GLUT_NAME = MesaGLUT-$(VERSION)
@@ -329,6 +329,8 @@ GALLIUM_FILES = \
 	$(DIRECTORY)/src/gallium/Makefile.template			\
 	$(DIRECTORY)/src/gallium/SConscript				\
 	$(DIRECTORY)/src/gallium/targets/Makefile.dri			\
+	$(DIRECTORY)/src/gallium/targets/Makefile.xorg			\
+	$(DIRECTORY)/src/gallium/targets/SConscript.dri			\
 	$(DIRECTORY)/src/gallium/*/Makefile				\
 	$(DIRECTORY)/src/gallium/*/SConscript				\
 	$(DIRECTORY)/src/gallium/*/*/Makefile				\
@@ -356,6 +358,7 @@ DRI_FILES = \
 	$(DIRECTORY)/src/mesa/drivers/dri/common/xmlpool/*.[ch]		\
 	$(DIRECTORY)/src/mesa/drivers/dri/common/xmlpool/*.po		\
 	$(DIRECTORY)/src/mesa/drivers/dri/*/*.[chS]			\
+	$(DIRECTORY)/src/mesa/drivers/dri/*/*.cpp			\
 	$(DIRECTORY)/src/mesa/drivers/dri/*/*/*.[chS]			\
 	$(DIRECTORY)/src/mesa/drivers/dri/*/Makefile			\
 	$(DIRECTORY)/src/mesa/drivers/dri/*/*/Makefile			\

SConstruct

@@ -3,14 +3,14 @@
 #
 # For example, invoke scons as 
 #
-#   scons debug=1 dri=0 machine=x86
+#   scons build=debug llvm=yes machine=x86
 #
 # to set configuration variables. Or you can write those options to a file
 # named config.py:
 #
 #   # config.py
-#   debug=1
-#   dri=0
+#   build='debug'
+#   llvm=True
 #   machine='x86'
 # 
 # Invoke
@@ -30,54 +30,8 @@ import common
 #######################################################################
 # Configuration options
 
-default_statetrackers = 'mesa'
-default_targets = 'graw-null'
-
-if common.default_platform in ('linux', 'freebsd', 'darwin'):
-	default_drivers = 'softpipe,galahad,failover,svga,i915,i965,trace,identity,llvmpipe'
-	default_winsys = 'xlib'
-elif common.default_platform in ('winddk',):
-	default_drivers = 'softpipe,svga,i915,i965,trace,identity'
-	default_winsys = 'all'
-elif common.default_platform in ('embedded',):
-	default_drivers = 'softpipe,llvmpipe'
-	default_winsys = 'xlib'
-else:
-	default_drivers = 'all'
-	default_winsys = 'all'
-
 opts = Variables('config.py')
 common.AddOptions(opts)
-opts.Add(ListVariable('statetrackers', 'state trackers to build', default_statetrackers,
-                     ['mesa', 'python', 'xorg', 'egl']))
-opts.Add(ListVariable('drivers', 'pipe drivers to build', default_drivers,
-                     ['softpipe', 'galahad', 'failover', 'svga', 'i915', 'i965', 'trace', 'r300', 'r600', 'identity', 'llvmpipe', 'nouveau', 'nv50', 'nvfx']))
-opts.Add(ListVariable('winsys', 'winsys drivers to build', default_winsys,
-                     ['xlib', 'vmware', 'i915', 'i965', 'gdi', 'radeon', 'r600', 'graw-xlib']))
-
-opts.Add(ListVariable('targets', 'driver targets to build', default_targets,
-		      ['dri-i915',
-		       'dri-i965',
-		       'dri-nouveau',
-		       'dri-radeong',
-		       'dri-swrast',
-		       'dri-vmwgfx',
-		       'egl-i915',
-		       'egl-i965',
-		       'egl-nouveau',
-		       'egl-radeon',
-		       'egl-swrast',
-		       'egl-vmwgfx',
-		       'graw-xlib',
-		       'graw-null',
-		       'libgl-gdi',
-		       'libgl-xlib',
-		       'xorg-i915',
-		       'xorg-i965',
-		       'xorg-nouveau',
-		       'xorg-radeon',
-		       'xorg-vmwgfx']))
-
 opts.Add(EnumVariable('MSVS_VERSION', 'MS Visual C++ version', None, allowed_values=('7.1', '8.0', '9.0')))
 
 env = Environment(
@@ -87,61 +41,26 @@ env = Environment(
 	ENV = os.environ,
 )
 
-if os.environ.has_key('CC'):
-	env['CC'] = os.environ['CC']
-if os.environ.has_key('CFLAGS'):
-	env['CCFLAGS'] += SCons.Util.CLVar(os.environ['CFLAGS'])
-if os.environ.has_key('CXX'):
-	env['CXX'] = os.environ['CXX']
-if os.environ.has_key('CXXFLAGS'):
-	env['CXXFLAGS'] += SCons.Util.CLVar(os.environ['CXXFLAGS'])
-if os.environ.has_key('LDFLAGS'):
-	env['LINKFLAGS'] += SCons.Util.CLVar(os.environ['LDFLAGS'])
+# Backwards compatability with old target configuration variable
+try:
+    targets = ARGUMENTS['targets']
+except KeyError:
+    pass
+else:
+    targets = targets.split(',')
+    print 'scons: warning: targets option is deprecated; pass the targets on their own such as'
+    print
+    print '  scons %s' % ' '.join(targets)
+    print 
+    COMMAND_LINE_TARGETS.append(targets)
+
 
 Help(opts.GenerateHelpText(env))
 
-# replicate options values in local variables
-debug = env['debug']
-dri = env['dri']
-machine = env['machine']
-platform = env['platform']
-
-# derived options
-x86 = machine == 'x86'
-ppc = machine == 'ppc'
-gcc = platform in ('linux', 'freebsd', 'darwin', 'embedded')
-msvc = platform in ('windows', 'winddk')
-
-Export([
-	'debug', 
-	'x86', 
-	'ppc', 
-	'dri', 
-	'platform',
-	'gcc',
-	'msvc',
-])
-
 
 #######################################################################
 # Environment setup
 
-# Always build trace, rbug, identity, softpipe, and llvmpipe (where possible)
-if 'trace' not in env['drivers']:
-    env['drivers'].append('trace')
-if 'rbug' not in env['drivers']:
-    env['drivers'].append('rbug')
-if 'galahad' not in env['drivers']:
-    env['drivers'].append('galahad')
-if 'identity' not in env['drivers']:
-    env['drivers'].append('identity')
-if 'softpipe' not in env['drivers']:
-    env['drivers'].append('softpipe')
-if env['llvm'] and 'llvmpipe' not in env['drivers']:
-    env['drivers'].append('llvmpipe')
-if 'sw' not in env['drivers']:
-    env['drivers'].append('sw')
-
 # Includes
 env.Prepend(CPPPATH = [
 	'#/include',
@@ -157,7 +76,7 @@ if env['msvc']:
     env.Append(CPPPATH = ['#include/c99'])
 
 # Embedded
-if platform == 'embedded':
+if env['platform'] == 'embedded':
 	env.Append(CPPDEFINES = [
 		'_POSIX_SOURCE',
 		('_POSIX_C_SOURCE', '199309L'), 
@@ -174,7 +93,7 @@ if platform == 'embedded':
 	])
 
 # Posix
-if platform in ('posix', 'linux', 'freebsd', 'darwin'):
+if env['platform'] in ('posix', 'linux', 'freebsd', 'darwin'):
 	env.Append(CPPDEFINES = [
 		'_POSIX_SOURCE',
 		('_POSIX_C_SOURCE', '199309L'), 
@@ -184,9 +103,9 @@ if platform in ('posix', 'linux', 'freebsd', 'darwin'):
 		'PTHREADS',
 		'HAVE_POSIX_MEMALIGN',
 	])
-	if gcc:
+	if env['gcc']:
 		env.Append(CFLAGS = ['-fvisibility=hidden'])
-	if platform == 'darwin':
+	if env['platform'] == 'darwin':
 		env.Append(CPPDEFINES = ['_DARWIN_C_SOURCE'])
 	env.Append(LIBS = [
 		'm',
@@ -208,9 +127,7 @@ Export('env')
 
 SConscript(
 	'src/SConscript',
-	variant_dir = env['build'],
+	variant_dir = env['build_dir'],
 	duplicate = 0 # http://www.scons.org/doc/0.97/HTML/scons-user/x2261.html
 )
 
-env.Default('src')
-

common.py

@@ -8,6 +8,8 @@ import subprocess
 import sys
 import platform as _platform
 
+import SCons.Script.SConscript
+
 
 #######################################################################
 # Defaults
@@ -20,6 +22,15 @@ _platform_map = {
 default_platform = sys.platform
 default_platform = _platform_map.get(default_platform, default_platform)
 
+# Search sys.argv[] for a "platform=foo" argument since we don't have
+# an 'env' variable at this point.
+if 'platform' in SCons.Script.ARGUMENTS:
+    selected_platform = SCons.Script.ARGUMENTS['platform']
+else:
+    selected_platform = default_platform
+
+cross_compiling = selected_platform != default_platform
+
 _machine_map = {
 	'x86': 'x86',
 	'i386': 'x86',
@@ -37,38 +48,26 @@ if 'PROCESSOR_ARCHITECTURE' in os.environ:
 else:
 	default_machine = _platform.machine()
 default_machine = _machine_map.get(default_machine, 'generic')
+default_toolchain = 'default'
+
+if selected_platform == 'windows' and cross_compiling:
+    default_machine = 'x86'
+    default_toolchain = 'crossmingw'
 
 
 # find default_llvm value
 if 'LLVM' in os.environ:
     default_llvm = 'yes'
 else:
-    # Search sys.argv[] for a "platform=foo" argument since we don't have
-    # an 'env' variable at this point.
-    platform = default_platform
-    pattern = re.compile("(platform=)(.*)")
-    for arg in sys.argv:
-        m = pattern.match(arg)
-        if m:
-            platform = m.group(2)
-
     default_llvm = 'no'
     try:
-        if platform != 'windows' and subprocess.call(['llvm-config', '--version'], stdout=subprocess.PIPE) == 0:
+        if selected_platform != 'windows' and \
+           subprocess.call(['llvm-config', '--version'], stdout=subprocess.PIPE) == 0:
             default_llvm = 'yes'
     except:
         pass
 
 
-# find default_dri value
-if default_platform in ('linux', 'freebsd'):
-	default_dri = 'yes'
-elif default_platform in ('winddk', 'windows', 'wince', 'darwin'):
-	default_dri = 'no'
-else:
-	default_dri = 'no'
-
-
 #######################################################################
 # Common options
 
@@ -81,13 +80,14 @@ def AddOptions(opts):
 		from SCons.Variables.EnumVariable import EnumVariable as EnumOption
 	except ImportError:
 		from SCons.Options.EnumOption import EnumOption
-	opts.Add(BoolOption('debug', 'debug build', 'yes'))
-	opts.Add(BoolOption('profile', 'profile build', 'no'))
+	opts.Add(EnumOption('build', 'build type', 'debug',
+	                  allowed_values=('debug', 'checked', 'profile', 'release')))
 	opts.Add(BoolOption('quiet', 'quiet command lines', 'yes'))
 	opts.Add(EnumOption('machine', 'use machine-specific assembly code', default_machine,
 											 allowed_values=('generic', 'ppc', 'x86', 'x86_64')))
 	opts.Add(EnumOption('platform', 'target platform', default_platform,
 											 allowed_values=('linux', 'cell', 'windows', 'winddk', 'wince', 'darwin', 'embedded', 'cygwin', 'sunos5', 'freebsd8')))
-	opts.Add('toolchain', 'compiler toolchain', 'default')
+	opts.Add('toolchain', 'compiler toolchain', default_toolchain)
 	opts.Add(BoolOption('llvm', 'use LLVM', default_llvm))
-	opts.Add(BoolOption('dri', 'build DRI drivers', default_dri))
+	opts.Add(BoolOption('debug', 'DEPRECATED: debug build', 'yes'))
+	opts.Add(BoolOption('profile', 'DEPRECATED: profile build', 'no'))

configs/autoconf.in

@@ -53,7 +53,7 @@ MKDEP_OPTIONS = @MKDEP_OPTIONS@
 INSTALL = @INSTALL@
 
 # Python and flags (generally only needed by the developers)
-PYTHON2 = python
+PYTHON2 = @PYTHON2@
 PYTHON_FLAGS = -t -O -O
 
 # Library names (base name)

configs/default

@@ -9,7 +9,7 @@ CONFIG_NAME = default
 
 # Version info
 MESA_MAJOR=7
-MESA_MINOR=9
+MESA_MINOR=10
 MESA_TINY=0
 MESA_VERSION = $(MESA_MAJOR).$(MESA_MINOR).$(MESA_TINY)
 

configs/linux-dri

@@ -58,7 +58,7 @@ EGL_DRIVERS_DIRS = glx
 
 DRIVER_DIRS = dri
 GALLIUM_WINSYS_DIRS = sw sw/xlib drm/vmware drm/intel drm/i965
-GALLIUM_TARGET_DIRS = egl-swrast
+GALLIUM_TARGET_DIRS = 
 GALLIUM_STATE_TRACKERS_DIRS = egl
 
 DRI_DIRS = i810 i915 i965 mach64 mga r128 r200 r300 radeon \

configure.ac

@@ -30,9 +30,14 @@ AC_PROG_CPP
 AC_PROG_CC
 AC_PROG_CXX
 AC_CHECK_PROGS([MAKE], [gmake make])
+AC_CHECK_PROGS([PYTHON2], [python2 python])
 AC_PATH_PROG([MKDEP], [makedepend])
 AC_PATH_PROG([SED], [sed])
 
+if test "x$MKDEP" = "x"; then
+    AC_MSG_ERROR([makedepend is required to build Mesa])
+fi
+
 dnl Our fallback install-sh is a symlink to minstall. Use the existing
 dnl configuration in that case.
 AC_PROG_INSTALL
@@ -460,6 +465,71 @@ if test "x$enable_selinux" = "xyes"; then
     DEFINES="$DEFINES -DMESA_SELINUX"
 fi
 
+dnl Determine which APIs to support
+AC_ARG_ENABLE([opengl],
+    [AS_HELP_STRING([--disable-opengl],
+        [disable support for standard OpenGL API @<:@default=no@:>@])],
+    [enable_opengl="$enableval"],
+    [enable_opengl=yes])
+AC_ARG_ENABLE([gles1],
+    [AS_HELP_STRING([--enable-gles1],
+        [enable support for OpenGL ES 1.x API @<:@default=no@:>@])],
+    [enable_gles1="$enableval"],
+    [enable_gles1=no])
+AC_ARG_ENABLE([gles2],
+    [AS_HELP_STRING([--enable-gles2],
+        [enable support for OpenGL ES 2.x API @<:@default=no@:>@])],
+    [enable_gles2="$enableval"],
+    [enable_gles2=no])
+AC_ARG_ENABLE([gles-overlay],
+    [AS_HELP_STRING([--enable-gles-overlay],
+        [build separate OpenGL ES only libraries @<:@default=no@:>@])],
+    [enable_gles_overlay="$enableval"],
+    [enable_gles_overlay=no])
+
+AC_ARG_ENABLE([openvg],
+    [AS_HELP_STRING([--enable-openvg],
+        [enable support for OpenVG API @<:@default=no@:>@])],
+    [enable_openvg="$enableval"],
+    [enable_openvg=no])
+
+dnl smooth the transition; should be removed eventually
+if test "x$enable_openvg" = xno; then
+    case "x$with_state_trackers" in
+    x*vega*)
+        AC_MSG_WARN([vega state tracker is enabled without --enable-openvg])
+        enable_openvg=yes
+        ;;
+    esac
+fi
+
+if test "x$enable_opengl" = xno -a \
+        "x$enable_gles1" = xno -a \
+        "x$enable_gles2" = xno -a \
+        "x$enable_gles_overlay" = xno -a \
+        "x$enable_openvg" = xno; then
+    AC_MSG_ERROR([at least one API should be enabled])
+fi
+
+API_DEFINES=""
+GLES_OVERLAY=0
+if test "x$enable_opengl" = xno; then
+    API_DEFINES="$API_DEFINES -DFEATURE_GL=0"
+else
+    API_DEFINES="$API_DEFINES -DFEATURE_GL=1"
+fi
+if test "x$enable_gles1" = xyes; then
+    API_DEFINES="$API_DEFINES -DFEATURE_ES1=1"
+fi
+if test "x$enable_gles2" = xyes; then
+    API_DEFINES="$API_DEFINES -DFEATURE_ES2=1"
+fi
+if test "x$enable_gles_overlay" = xyes; then
+    GLES_OVERLAY=1
+fi
+AC_SUBST([API_DEFINES])
+AC_SUBST([GLES_OVERLAY])
+
 dnl
 dnl Driver configuration. Options are xlib, dri and osmesa right now.
 dnl More later: fbdev, ...
@@ -479,6 +549,10 @@ linux*)
     ;;
 esac
 
+if test "x$enable_opengl" = xno; then
+    default_driver="no"
+fi
+
 AC_ARG_WITH([driver],
     [AS_HELP_STRING([--with-driver=DRIVER],
         [driver for Mesa: xlib,dri,osmesa @<:@default=dri when available, or xlib@:>@])],
@@ -487,6 +561,11 @@ AC_ARG_WITH([driver],
 dnl Check for valid option
 case "x$mesa_driver" in
 xxlib|xdri|xosmesa)
+    if test "x$enable_opengl" = xno; then
+        AC_MSG_ERROR([Driver '$mesa_driver' requires OpenGL enabled])
+    fi
+    ;;
+xno)
     ;;
 *)
     AC_MSG_ERROR([Driver '$mesa_driver' is not a valid option])
@@ -502,7 +581,7 @@ dnl Driver specific build directories
 dnl
 
 dnl this variable will be prepended to SRC_DIRS and is not exported
-CORE_DIRS="mapi/glapi glsl mesa"
+CORE_DIRS=""
 
 SRC_DIRS=""
 GLU_DIRS="sgi"
@@ -512,6 +591,30 @@ GALLIUM_WINSYS_DIRS="sw"
 GALLIUM_DRIVERS_DIRS="softpipe failover galahad trace rbug identity"
 GALLIUM_STATE_TRACKERS_DIRS=""
 
+# build glapi if OpenGL is enabled
+if test "x$enable_opengl" = xyes; then
+    CORE_DIRS="$CORE_DIRS mapi/glapi"
+fi
+
+# build es1api and es2api if OpenGL ES is enabled
+case "x$enable_gles1$enable_gles2$enable_gles_overlay" in
+x*yes*)
+    CORE_DIRS="$CORE_DIRS mapi/es1api mapi/es2api"
+    ;;
+esac
+
+# build vgapi if OpenVG is enabled
+if test "x$enable_openvg" = xyes; then
+    CORE_DIRS="$CORE_DIRS mapi/vgapi"
+fi
+
+# build glsl and mesa if OpenGL or OpenGL ES is enabled
+case "x$enable_opengl$enable_gles1$enable_gles2$enable_gles_overlay" in
+x*yes*)
+    CORE_DIRS="$CORE_DIRS glsl mesa"
+    ;;
+esac
+
 case "$mesa_driver" in
 xlib)
     DRIVER_DIRS="x11"
@@ -526,6 +629,9 @@ dri)
 osmesa)
     DRIVER_DIRS="osmesa"
     ;;
+no)
+    DRIVER_DRIS=""
+    ;;
 esac
 AC_SUBST([SRC_DIRS])
 AC_SUBST([GLU_DIRS])
@@ -618,7 +724,7 @@ xlib)
         GL_LIB_DEPS=""
     fi
     ;;
-dri)
+dri|no) # these checks are still desired when there is no mesa_driver
     # DRI must be shared, I think
     if test "$enable_static" = yes; then
         AC_MSG_ERROR([Can't use static libraries for DRI drivers])
@@ -692,6 +798,11 @@ AC_SUBST([GLESv2_PC_LIB_PRIV])
 
 AC_SUBST([HAVE_XF86VIDMODE])
 
+PKG_CHECK_MODULES([LIBDRM_RADEON],
+		  [libdrm_radeon libdrm >= $LIBDRM_RADEON_REQUIRED],
+		  HAVE_LIBDRM_RADEON=yes,
+		  HAVE_LIBDRM_RADEON=no)
+
 dnl
 dnl More X11 setup
 dnl
@@ -738,51 +849,6 @@ if test "x$with_dri_drivers" = x; then
     with_dri_drivers=no
 fi
 
-dnl Determine which APIs to support
-AC_ARG_ENABLE([opengl],
-    [AS_HELP_STRING([--disable-opengl],
-        [disable support for standard OpenGL API @<:@default=no@:>@])],
-    [enable_opengl="$enableval"],
-    [enable_opengl=yes])
-AC_ARG_ENABLE([gles1],
-    [AS_HELP_STRING([--enable-gles1],
-        [enable support for OpenGL ES 1.x API @<:@default=no@:>@])],
-    [enable_gles1="$enableval"],
-    [enable_gles1=no])
-AC_ARG_ENABLE([gles2],
-    [AS_HELP_STRING([--enable-gles2],
-        [enable support for OpenGL ES 2.x API @<:@default=no@:>@])],
-    [enable_gles2="$enableval"],
-    [enable_gles2=no])
-AC_ARG_ENABLE([gles-overlay],
-    [AS_HELP_STRING([--enable-gles-overlay],
-        [build separate OpenGL ES only libraries @<:@default=no@:>@])],
-    [enable_gles_overlay="$enableval"],
-    [enable_gles_overlay=no])
-
-API_DEFINES=""
-GLES_OVERLAY=0
-if test "x$enable_opengl" = xno; then
-    API_DEFINES="$API_DEFINES -DFEATURE_GL=0"
-else
-    API_DEFINES="$API_DEFINES -DFEATURE_GL=1"
-fi
-if test "x$enable_gles1" = xyes; then
-    API_DEFINES="$API_DEFINES -DFEATURE_ES1=1"
-fi
-if test "x$enable_gles2" = xyes; then
-    API_DEFINES="$API_DEFINES -DFEATURE_ES2=1"
-fi
-if test "x$enable_gles_overlay" = xyes -o \
-    "x$enable_gles1" = xyes -o "x$enable_gles2" = xyes; then
-    CORE_DIRS="mapi/es1api mapi/es2api $CORE_DIRS"
-    if test "x$enable_gles_overlay" = xyes; then
-        GLES_OVERLAY=1
-    fi
-fi
-AC_SUBST([API_DEFINES])
-AC_SUBST([GLES_OVERLAY])
-
 dnl If $with_dri_drivers is yes, directories will be added through
 dnl platform checks
 DRI_DIRS=""
@@ -803,7 +869,7 @@ yes)
 esac
 
 dnl Set DRI_DIRS, DEFINES and LIB_DEPS
-if test "$mesa_driver" = dri; then
+if test "$mesa_driver" = dri -o "$mesa_driver" = no; then
     # Use TLS in GLX?
     if test "x$GLX_USE_TLS" = xyes; then
         DEFINES="$DEFINES -DGLX_USE_TLS -DPTHREADS"
@@ -881,19 +947,21 @@ if test "$mesa_driver" = dri; then
     DRI_DIRS=`echo "$DRI_DIRS" | $SED 's/  */ /g'`
 
     # Check for expat
-    EXPAT_INCLUDES=""
-    EXPAT_LIB=-lexpat
-    AC_ARG_WITH([expat],
-        [AS_HELP_STRING([--with-expat=DIR],
-            [expat install directory])],[
-        EXPAT_INCLUDES="-I$withval/include"
-        CPPFLAGS="$CPPFLAGS $EXPAT_INCLUDES"
-        LDFLAGS="$LDFLAGS -L$withval/$LIB_DIR"
-        EXPAT_LIB="-L$withval/$LIB_DIR -lexpat"
-        ])
-    AC_CHECK_HEADER([expat.h],[],[AC_MSG_ERROR([Expat required for DRI.])])
-    AC_CHECK_LIB([expat],[XML_ParserCreate],[],
-        [AC_MSG_ERROR([Expat required for DRI.])])
+    if test "$mesa_driver" = dri; then
+        EXPAT_INCLUDES=""
+        EXPAT_LIB=-lexpat
+        AC_ARG_WITH([expat],
+            [AS_HELP_STRING([--with-expat=DIR],
+                [expat install directory])],[
+            EXPAT_INCLUDES="-I$withval/include"
+            CPPFLAGS="$CPPFLAGS $EXPAT_INCLUDES"
+            LDFLAGS="$LDFLAGS -L$withval/$LIB_DIR"
+            EXPAT_LIB="-L$withval/$LIB_DIR -lexpat"
+            ])
+        AC_CHECK_HEADER([expat.h],[],[AC_MSG_ERROR([Expat required for DRI.])])
+        AC_CHECK_LIB([expat],[XML_ParserCreate],[],
+            [AC_MSG_ERROR([Expat required for DRI.])])
+    fi
 
     # put all the necessary libs together
     DRI_LIB_DEPS="$SELINUX_LIBS $LIBDRM_LIBS $EXPAT_LIB -lm -lpthread $DLOPEN_LIBS $TALLOC_LIBS"
@@ -910,12 +978,7 @@ esac
 
 case $DRI_DIRS in
 *radeon*|*r200*|*r300*|*r600*)
-    PKG_CHECK_MODULES([LIBDRM_RADEON],
-		      [libdrm_radeon libdrm >= $LIBDRM_RADEON_REQUIRED],
-		      HAVE_LIBDRM_RADEON=yes,
-		      HAVE_LIBDRM_RADEON=no)
-
-    if test "$HAVE_LIBDRM_RADEON" = yes; then
+    if test "x$HAVE_LIBDRM_RADEON" = xyes; then
 	RADEON_CFLAGS="-DHAVE_LIBDRM_RADEON=1 $LIBDRM_RADEON_CFLAGS"
 	RADEON_LDFLAGS=$LIBDRM_RADEON_LIBS
     fi
@@ -939,6 +1002,9 @@ AC_ARG_ENABLE([gl-osmesa],
     [gl_osmesa="$enableval"],
     [gl_osmesa="$default_gl_osmesa"])
 if test "x$gl_osmesa" = xyes; then
+    if test "x$enable_opengl" = xno; then
+        AC_MSG_ERROR([OpenGL is not available for OSMesa driver])
+    fi
     if test "$mesa_driver" = osmesa; then
         AC_MSG_ERROR([libGL is not available for OSMesa driver])
     else
@@ -995,13 +1061,21 @@ AC_ARG_ENABLE([egl],
         [disable EGL library @<:@default=enabled@:>@])],
     [enable_egl="$enableval"],
     [enable_egl=yes])
+if test "x$enable_egl" = xno; then
+    if test "x$mesa_driver" = xno; then
+        AC_MSG_ERROR([cannot disable EGL when there is no mesa driver])
+    fi
+    if test "x$enable_openvg" = xyes; then
+        AC_MSG_ERROR([cannot enable OpenVG without EGL])
+    fi
+fi
 if test "x$enable_egl" = xyes; then
     SRC_DIRS="$SRC_DIRS egl"
     EGL_LIB_DEPS="$DLOPEN_LIBS -lpthread"
     EGL_DRIVERS_DIRS=""
     if test "$enable_static" != yes; then
         # build egl_glx when libGL is built
-        if test "$mesa_driver" != osmesa; then
+        if test "$mesa_driver" = xlib -o "$mesa_driver" = dri; then
             EGL_DRIVERS_DIRS="glx"
         fi
 
@@ -1035,6 +1109,12 @@ AC_ARG_ENABLE([glu],
         [enable OpenGL Utility library @<:@default=enabled@:>@])],
     [enable_glu="$enableval"],
     [enable_glu=yes])
+
+if test "x$enable_glu" = xyes -a "x$mesa_driver" = xno; then
+    AC_MSG_NOTICE([Disabling GLU since there is no OpenGL driver])
+    enable_glu=no
+fi
+
 if test "x$enable_glu" = xyes; then
     SRC_DIRS="$SRC_DIRS glu"
 
@@ -1084,9 +1164,13 @@ AC_ARG_ENABLE([glw],
     [enable_glw="$enableval"],
     [enable_glw=yes])
 dnl Don't build GLw on osmesa
-if test "x$enable_glw" = xyes && test "$mesa_driver" = osmesa; then
-    AC_MSG_WARN([Disabling GLw since the driver is OSMesa])
-    enable_glw=no
+if test "x$enable_glw" = xyes; then
+    case "$mesa_driver" in
+    osmesa|no)
+        AC_MSG_NOTICE([Disabling GLw since there is no OpenGL driver])
+        enable_glw=no
+        ;;
+    esac
 fi
 AC_ARG_ENABLE([motif],
     [AS_HELP_STRING([--enable-motif],
@@ -1160,16 +1244,20 @@ AC_ARG_ENABLE([glut],
     [enable_glut="$enableval"],
     [enable_glut="$default_glut"])
 
+dnl Don't build glut on osmesa
+if test "x$enable_glut" = xyes; then
+    case "$mesa_driver" in
+    osmesa|no)
+        AC_MSG_NOTICE([Disabling glut since there is no OpenGL driver])
+        enable_glut=no
+        ;;
+    esac
+fi
 dnl Can't build glut if GLU not available
 if test "x$enable_glu$enable_glut" = xnoyes; then
     AC_MSG_WARN([Disabling glut since GLU is disabled])
     enable_glut=no
 fi
-dnl Don't build glut on osmesa
-if test "x$enable_glut" = xyes && test "$mesa_driver" = osmesa; then
-    AC_MSG_WARN([Disabling glut since the driver is OSMesa])
-    enable_glut=no
-fi
 
 if test "x$enable_glut" = xyes; then
     SRC_DIRS="$SRC_DIRS glut/glx"
@@ -1234,6 +1322,9 @@ AC_ARG_ENABLE([gallium],
         [build gallium @<:@default=enabled@:>@])],
     [enable_gallium="$enableval"],
     [enable_gallium=yes])
+if test "x$enable_gallium" = xno -a "x$enable_openvg" = xyes; then
+    AC_MSG_ERROR([cannot enable OpenVG without Gallium])
+fi
 if test "x$enable_gallium" = xyes; then
     SRC_DIRS="$SRC_DIRS gallium gallium/winsys gallium/targets"
     AC_CHECK_HEADER([udis86.h], [HAS_UDIS86="yes"],
@@ -1246,12 +1337,6 @@ AC_SUBST([LLVM_LIBS])
 AC_SUBST([LLVM_LDFLAGS])
 AC_SUBST([LLVM_VERSION])
 
-VG_LIB_DEPS=""
-EGL_CLIENT_APIS='$(GL_LIB)'
-if test "x$enable_gles_overlay" = xyes; then
-    EGL_CLIENT_APIS="$EGL_CLIENT_APIS "'$(GLESv1_CM_LIB) $(GLESv2_LIB)'
-fi
-
 dnl
 dnl Gallium state trackers configuration
 dnl
@@ -1267,6 +1352,8 @@ no)
     GALLIUM_STATE_TRACKERS_DIRS=""
     ;;
 yes)
+    st_egl="no"
+
     # look at what else is built
     case "$mesa_driver" in
     xlib)
@@ -1275,16 +1362,27 @@ yes)
     dri)
         GALLIUM_STATE_TRACKERS_DIRS="dri"
         HAVE_ST_DRI="yes"
-        if test "x$enable_egl" = xyes; then
-            GALLIUM_STATE_TRACKERS_DIRS="$GALLIUM_STATE_TRACKERS_DIRS egl"
-            HAVE_ST_EGL="yes"
-        fi
+        st_egl="yes"
         # Have only tested st/xorg on 1.6.0 servers
         PKG_CHECK_MODULES(XORG, [xorg-server >= 1.6.0 libdrm >= $LIBDRM_XORG_REQUIRED libkms >= $LIBKMS_XORG_REQUIRED],
             HAVE_ST_XORG="yes"; GALLIUM_STATE_TRACKERS_DIRS="$GALLIUM_STATE_TRACKERS_DIRS xorg",
             HAVE_ST_XORG="no")
         ;;
+    no)
+        st_egl="yes"
     esac
+
+    if test "x$enable_egl" = xyes; then
+        if test "$enable_openvg" = yes; then
+            GALLIUM_STATE_TRACKERS_DIRS="$GALLIUM_STATE_TRACKERS_DIRS vega"
+            st_egl="yes"
+        fi
+
+        if test "$st_egl" = yes; then
+            GALLIUM_STATE_TRACKERS_DIRS="$GALLIUM_STATE_TRACKERS_DIRS egl"
+            HAVE_ST_EGL="yes"
+        fi
+    fi
     ;;
 *)
     # verify the requested state tracker exist
@@ -1310,22 +1408,10 @@ yes)
             PKG_CHECK_MODULES([LIBKMS_XORG], [libkms >= $LIBKMS_XORG_REQUIRED])
             HAVE_ST_XORG="yes"
             ;;
-        es)
-            AC_MSG_WARN([state tracker 'es' has been replaced by --enable-gles-overlay])
-
-            if test "x$enable_gles_overlay" != xyes; then
-                if test "x$enable_gles1" != xyes -a "x$enable_gles2" != xyes; then
-                    CORE_DIRS="mapi/es1api mapi/es2api $CORE_DIRS"
-                fi
-                GLES_OVERLAY=1
-                EGL_CLIENT_APIS="$EGL_CLIENT_APIS "'$(GLESv1_CM_LIB) $(GLESv2_LIB)'
-            fi
-	    tracker=""
-            ;;
         vega)
-            CORE_DIRS="$CORE_DIRS mapi/vgapi"
-            VG_LIB_DEPS="$VG_LIB_DEPS -lpthread"
-            EGL_CLIENT_APIS="$EGL_CLIENT_APIS "'$(VG_LIB)'
+            if test "x$enable_openvg" != xyes; then
+                AC_MSG_ERROR([cannot build vega state tracker without --enable-openvg])
+            fi
             ;;
         esac
 
@@ -1343,6 +1429,23 @@ yes)
     ;;
 esac
 
+
+EGL_CLIENT_APIS=""
+VG_LIB_DEPS=""
+
+case "x$enable_opengl$enable_gles1$enable_gles2" in
+x*yes*)
+    EGL_CLIENT_APIS="$EGL_CLIENT_APIS "'$(GL_LIB)'
+    ;;
+esac
+if test "x$enable_gles_overlay" = xyes; then
+    EGL_CLIENT_APIS="$EGL_CLIENT_APIS "'$(GLESv1_CM_LIB) $(GLESv2_LIB)'
+fi
+if test "x$enable_openvg" = xyes; then
+    EGL_CLIENT_APIS="$EGL_CLIENT_APIS "'$(VG_LIB)'
+    VG_LIB_DEPS="$VG_LIB_DEPS -lpthread"
+fi
+
 AC_SUBST([VG_LIB_DEPS])
 AC_SUBST([EGL_CLIENT_APIS])
 
@@ -1363,7 +1466,7 @@ fi
 AC_ARG_WITH([egl-platforms],
     [AS_HELP_STRING([--with-egl-platforms@<:@=DIRS...@:>@],
         [comma delimited native platforms libEGL supports, e.g.
-        "x11,kms" @<:@default=auto@:>@])],
+        "x11,drm" @<:@default=auto@:>@])],
     [with_egl_platforms="$withval"],
     [with_egl_platforms=yes])
 AC_ARG_WITH([egl-displays],
@@ -1376,6 +1479,9 @@ case "$with_egl_platforms" in
 yes)
     if test "x$enable_egl" = xyes && test "x$mesa_driver" != xosmesa; then
         EGL_PLATFORMS="x11"
+        if test "$mesa_driver" = dri; then
+            EGL_PLATFORMS="$EGL_PLATFORMS drm"
+        fi
     fi
     ;;
 *)
@@ -1518,18 +1624,28 @@ elif test "x$enable_gallium_i965" = xauto; then
 fi
 
 dnl
-dnl Gallium Radeon configuration
+dnl Gallium Radeon r300g configuration
 dnl
 AC_ARG_ENABLE([gallium-radeon],
     [AS_HELP_STRING([--enable-gallium-radeon],
         [build gallium radeon @<:@default=disabled@:>@])],
     [enable_gallium_radeon="$enableval"],
     [enable_gallium_radeon=auto])
+if test "x$enable_gallium_radeon" = xauto; then
+    if test "x$HAVE_LIBDRM_RADEON" = xyes; then
+	GALLIUM_DRIVERS_DIRS="$GALLIUM_DRIVERS_DIRS r300"
+	gallium_check_st "radeon/drm" "dri-r300"
+    else
+	AC_MSG_WARN([libdrm_radeon is missing, not building gallium-radeon (r300)])
+    fi
+fi
 if test "x$enable_gallium_radeon" = xyes; then
-    GALLIUM_DRIVERS_DIRS="$GALLIUM_DRIVERS_DIRS r300"
-    gallium_check_st "radeon/drm" "dri-r300" "xorg-radeon"
-elif test "x$enable_gallium_radeon" = xauto; then
-    GALLIUM_DRIVERS_DIRS="$GALLIUM_DRIVERS_DIRS r300"
+    if test "x$HAVE_LIBDRM_RADEON" = xyes; then
+	GALLIUM_DRIVERS_DIRS="$GALLIUM_DRIVERS_DIRS r300"
+	gallium_check_st "radeon/drm" "dri-r300" "xorg-radeon"
+    else
+	AC_MSG_ERROR([libdrm_radeon is missing, cannot build gallium-radeon (r300)])
+    fi
 fi
 
 dnl
@@ -1541,8 +1657,12 @@ AC_ARG_ENABLE([gallium-r600],
     [enable_gallium_r600="$enableval"],
     [enable_gallium_r600=auto])
 if test "x$enable_gallium_r600" = xyes; then
-    GALLIUM_DRIVERS_DIRS="$GALLIUM_DRIVERS_DIRS r600"
-    gallium_check_st "r600/drm" "dri-r600"
+    if test "x$HAVE_LIBDRM_RADEON" = xyes; then
+	GALLIUM_DRIVERS_DIRS="$GALLIUM_DRIVERS_DIRS r600"
+	gallium_check_st "r600/drm" "dri-r600"
+    else
+	AC_MSG_ERROR([libdrm_radeon is missing, cannot build gallium-r600])
+    fi
 fi
 
 dnl
@@ -1601,25 +1721,56 @@ echo "        exec_prefix:     $exec_prefix"
 echo "        libdir:          $libdir"
 echo "        includedir:      $includedir"
 
+dnl API info
+echo ""
+echo "        OpenGL:          $enable_opengl (ES1: $enable_gles1 ES2: $enable_gles2)"
+echo "        GLES overlay:    $enable_gles_overlay"
+echo "        OpenVG:          $enable_openvg"
+
 dnl Driver info
 echo ""
 echo "        Driver:          $mesa_driver"
-if echo "$DRIVER_DIRS" | grep 'osmesa' >/dev/null 2>&1; then
-    echo "        OSMesa:          lib$OSMESA_LIB"
-else
-    echo "        OSMesa:          no"
+if test "$mesa_driver" != no; then
+    if echo "$DRIVER_DIRS" | grep 'osmesa' >/dev/null 2>&1; then
+        echo "        OSMesa:          lib$OSMESA_LIB"
+    else
+        echo "        OSMesa:          no"
+    fi
+    if test "$mesa_driver" = dri; then
+        # cleanup the drivers var
+        dri_dirs=`echo $DRI_DIRS | $SED 's/^ *//;s/  */ /;s/ *$//'`
+        if test "x$DRI_DIRS" = x; then
+            echo "        DRI drivers:     no"
+        else
+            echo "        DRI drivers:     $dri_dirs"
+        fi
+        echo "        DRI driver dir:  $DRI_DRIVER_INSTALL_DIR"
+        echo "        Use XCB:         $enable_xcb"
+    fi
 fi
-if test "$mesa_driver" = dri; then
-    # cleanup the drivers var
-    dri_dirs=`echo $DRI_DIRS | $SED 's/^ *//;s/  */ /;s/ *$//'`
-if test "x$DRI_DIRS" = x; then
-    echo "        DRI drivers:     no"
-else
-    echo "        DRI drivers:     $dri_dirs"
+echo ""
+echo "        GLU:             $enable_glu"
+echo "        GLw:             $enable_glw (Motif: $enable_motif)"
+echo "        glut:            $enable_glut"
+
+dnl EGL
+echo ""
+echo "        EGL:             $enable_egl"
+if test "$enable_egl" = yes; then
+    echo "        EGL platforms:   $EGL_PLATFORMS"
+
+    egl_drivers=""
+    for d in $EGL_DRIVERS_DIRS; do
+        egl_drivers="$egl_drivers egl_$d"
+    done
+
+    if test "$enable_gallium" = yes -a "$HAVE_ST_EGL" = yes; then
+        echo "        EGL drivers:    ${egl_drivers} egl_gallium"
+        echo "        EGL Gallium STs:$EGL_CLIENT_APIS"
+    else
+        echo "        EGL drivers:    $egl_drivers"
+    fi
 fi
-    echo "        DRI driver dir:  $DRI_DRIVER_INSTALL_DIR"
-fi
-echo "        Use XCB:         $enable_xcb"
 
 echo ""
 if test "x$MESA_LLVM" = x1; then
@@ -1638,9 +1789,6 @@ if echo "$SRC_DIRS" | grep 'gallium' >/dev/null 2>&1; then
     echo "        Winsys dirs:     $GALLIUM_WINSYS_DIRS"
     echo "        Driver dirs:     $GALLIUM_DRIVERS_DIRS"
     echo "        Trackers dirs:   $GALLIUM_STATE_TRACKERS_DIRS"
-    if test "x$HAVE_ST_EGL" = xyes; then
-        echo "        EGL client APIs: $EGL_CLIENT_APIS"
-    fi
 else
     echo "        Gallium:         no"
 fi
@@ -1649,15 +1797,6 @@ dnl Libraries
 echo ""
 echo "        Shared libs:     $enable_shared"
 echo "        Static libs:     $enable_static"
-if test "$enable_egl" = yes; then
-    echo "        EGL:             $EGL_DRIVERS_DIRS"
-    echo "        EGL platforms:   $EGL_PLATFORMS"
-else
-    echo "        EGL:             no"
-fi
-echo "        GLU:             $enable_glu"
-echo "        GLw:             $enable_glw (Motif: $enable_motif)"
-echo "        glut:            $enable_glut"
 
 dnl Compiler options
 # cleanup the CFLAGS/CXXFLAGS/DEFINES vars
@@ -1670,6 +1809,8 @@ echo ""
 echo "        CFLAGS:          $cflags"
 echo "        CXXFLAGS:        $cxxflags"
 echo "        Macros:          $defines"
+echo ""
+echo "        PYTHON2:         $PYTHON2"
 
 echo ""
 echo "        Run '${MAKE-make}' to build Mesa"

docs/GL3.txt

@@ -25,7 +25,7 @@ Non-normalized Integer texture/framebuffer formats    not started
 Packed depth/stencil formats                          DONE
 Per-buffer blend and masks (GL_EXT_draw_buffers2)     DONE
 GL_EXT_texture_compression_rgtc                       not started
-Red and red/green texture formats                     Ian?
+Red and red/green texture formats                     DONE (swrast, i965, gallium)
 Transform feedback (GL_EXT_transform_feedback)        ~50% done
    glBindFragDataLocation, glGetFragDataLocation,
    glBindBufferRange, glBindBufferBase commands
@@ -42,9 +42,9 @@ GL 3.1:
 GLSL 1.30 and 1.40                                    not started
 Instanced drawing (GL_ARB_draw_instanced)             ~50% done
 Buffer copying (GL_ARB_copy_buffer)                   DONE
-Primitive restart (GL_NV_primitive_restart)           not started
+Primitive restart (GL_NV_primitive_restart)           DONE (gallium)
 16 vertex texture image units                         not started
-Texture buffer objs (GL_ARB_textur_buffer_object)     not started
+Texture buffer objs (GL_ARB_texture_buffer_object)    not started
 Rectangular textures (GL_ARB_texture_rectangle)       DONE
 Uniform buffer objs (GL_ARB_uniform_buffer_object)    not started
 Signed normalized texture formats                     ~50% done
@@ -69,7 +69,7 @@ GL 3.3:
 
 GLSL 3.30                                             not started
 GL_ARB_blend_func_extended                            not started
-GL_ARB_explicit_attrib_location                       not started
+GL_ARB_explicit_attrib_location                       DONE (swrast, i915, i965)
 GL_ARB_occlusion_query2                               not started
 GL_ARB_sampler_objects                                not started
 GL_ARB_texture_rgb10_a2ui                             not started

docs/devinfo.html

@@ -145,7 +145,7 @@ Make sure the values in src/mesa/main/version.h are correct.
 </p>
 
 <p>
-Update the docs/news.html file and docs/download.html files.
+Update docs/news.html.
 </p>
 
 <p>
@@ -208,10 +208,11 @@ sftp USERNAME,mesa3d@web.sourceforge.net
 
 <p>
 Make an announcement on the mailing lists:
-<em>m</em><em>e</em><em>s</em><em>a</em><em>3</em><em>d</em><em>-</em><em>d</em><em>e</em><em>v</em><em>@</em><em>l</em><em>i</em><em>s</em><em>t</em><em>s</em><em>.</em><em>s</em><em>f</em><em>.</em><em>n</em><em>e</em><em>t</em>,
-<em>m</em><em>e</em><em>s</em><em>a</em><em>3</em><em>d</em><em>-</em><em>u</em><em>s</em><em>e</em><em>r</em><em>s</em><em>@</em><em>l</em><em>i</em><em>s</em><em>t</em><em>s</em><em>.</em><em>s</em><em>f</em><em>.</em><em>n</em><em>e</em><em>t</em>
+
+<em>m</em><em>e</em><em>s</em><em>a</em><em>-</em><em>d</em><em>e</em><em>v</em><em>@</em><em>l</em><em>i</em><em>s</em><em>t</em><em>s</em><em>.</em><em>f</em><em>r</em><em>e</em><em>e</em><em>d</em><em>e</em><em>s</em><em>k</em><em>t</em><em>o</em><em>p</em><em>.</em><em>o</em><em>r</em><em>g</em>,
+<em>m</em><em>e</em><em>s</em><em>a</em><em>-</em><em>u</em><em>s</em><em>e</em><em>r</em><em>s</em><em>@</em><em>l</em><em>i</em><em>s</em><em>t</em><em>s</em><em>.</em><em>f</em><em>r</em><em>e</em><em>e</em><em>d</em><em>e</em><em>s</em><em>k</em><em>t</em><em>o</em><em>p</em><em>.</em><em>o</em><em>r</em><em>g</em>
 and
-<em>m</em><em>e</em><em>s</em><em>a</em><em>3</em><em>d</em><em>-</em><em>a</em><em>n</em><em>n</em><em>o</em><em>u</em><em>n</em><em>c</em><em>e</em><em>@</em><em>l</em><em>i</em><em>s</em><em>t</em><em>s</em><em>.</em><em>s</em><em>f</em><em>.</em><em>n</em><em>e</em><em>t</em>
+<em>m</em><em>e</em><em>s</em><em>a</em><em>-</em><em>a</em><em>n</em><em>n</em><em>o</em><em>u</em><em>n</em><em>c</em><em>e</em><em>@</em><em>l</em><em>i</em><em>s</em><em>t</em><em>s</em><em>.</em><em>f</em><em>r</em><em>e</em><em>e</em><em>d</em><em>e</em><em>s</em><em>k</em><em>t</em><em>o</em><em>p</em><em>.</em><em>o</em><em>r</em><em>g</em>
 </p>
 
 

docs/egl.html

@@ -32,14 +32,14 @@ cards.</p>
 the Gallium driver for your hardware.  For example</p>
 
 <pre>
-  $ ./configure --enable-gles-overlay --with-state-trackers=egl,vega --enable-gallium-intel
+  $ ./configure --enable-gles-overlay --enable-openvg --enable-gallium-intel
 </pre>
 
 <p>The main library and OpenGL is enabled by default.  The first option enables
-<a href="opengles.html">OpenGL ES 1.x and 2.x</a>.  The <code>egl</code> state
-tracker is needed by a number of EGL drivers.  EGL drivers will be covered
-later.  The <a href="openvg.html">vega state tracker</a> provides OpenVG
-1.x.</p>
+<a href="opengles.html">OpenGL ES 1.x and 2.x</a>.  The second option enables
+<a href="openvg.html">OpenVG</a>.
+</p>
+
 </li>
 
 <li>Build and install Mesa as usual.</li>
@@ -72,31 +72,22 @@ drivers will be installed to <code>${libdir}/egl</code>.</p>
 <li><code>--with-egl-platforms</code>
 
 <p>List the platforms (window systems) to support.  Its argument is a comma
-seprated string such as <code>--with-egl-platforms=x11,kms</code>.  It decides
+seprated string such as <code>--with-egl-platforms=x11,drm</code>.  It decides
 the platforms a driver may support.  The first listed platform is also used by
 the main library to decide the native platform: the platform the EGL native
 types such as <code>EGLNativeDisplayType</code> or
 <code>EGLNativeWindowType</code> defined for.</p>
 
-<p>The available platforms are <code>x11</code>, <code>kms</code>,
+<p>The available platforms are <code>x11</code>, <code>drm</code>,
 <code>fbdev</code>, and <code>gdi</code>.  The <code>gdi</code> platform can
 only be built with SCons.</p>
 
 </li>
 
-<li><code>--with-state-trackers</code>
-
-<p>The argument is a comma separated string.  It is usually used to specify the
-rendering APIs, such as OpenVG, to build.  But it is also used to specify
-<code>egl</code> state tracker that <code>egl_gallium</code> depends on.</p>
-
-</li>
-
 <li><code>--enable-gles-overlay</code>
 
-<p>OpenGL and OpenGL ES are not controlled by
-<code>--with-state-trackers</code>.  OpenGL is always built.  To build OpenGL
-ES, this option must be explicitly given.</p>
+<p>OpenGL is built by default.  To build OpenGL ES, this option must be
+explicitly given.</p>
 
 </li>
 
@@ -108,6 +99,12 @@ one big library that supports multiple APIs.</p>
 
 </li>
 
+<li><code>--enable-openvg</code>
+
+<p>OpenVG must be explicitly enabled by this option.</p>
+
+</li>
+
 </ul>
 
 <h2>Use EGL</h2>
@@ -139,10 +136,6 @@ binaries.</p>
 specified EGL driver to be loaded.  It comes in handy when one wants to test a
 specific driver.  This variable is ignored for setuid/setgid binaries.</p>
 
-<p><code>egl_gallium</code> dynamically loads hardware drivers and client API
-modules found in <code>EGL_DRIVERS_PATH</code>.  Thus, specifying this variable
-alone is not sufficient for <code>egl_gallium</code> for uninstalled build.</p>
-
 </li>
 
 <li><code>EGL_PLATFORM</code>
@@ -177,7 +170,7 @@ variable to true forces the use of software rendering.</p>
 
 <p>This driver is based on Gallium3D.  It supports all rendering APIs and
 hardwares supported by Gallium3D.  It is the only driver that supports OpenVG.
-The supported platforms are X11, KMS, FBDEV, and GDI.</p>
+The supported platforms are X11, DRM, FBDEV, and GDI.</p>
 
 </li>
 
@@ -193,16 +186,8 @@ is not available in GLX or GLX extensions.</p>
 
 <p>This driver supports the X Window System as its window system.  It functions
 as a DRI2 driver loader.  Unlike <code>egl_glx</code>, it has no dependency on
-<code>libGL</code>.  It talks to the X server directly using DRI2 protocol.</p>
-
-</li>
-<li><code>egl_dri</code>
-
-<p>This driver lacks maintenance and does <em>not</em> build.  It is similiar
-to <code>egl_dri2</code> in that it functions as a DRI(1) driver loader.  But
-unlike <code>egl_dri2</code>, it supports Linux framebuffer devices as its
-window system and supports EGL_MESA_screen_surface extension.  As DRI1 drivers
-are phasing out, it might eventually be replaced by <code>egl_dri2</code>.</p>
+<code>libGL</code>.  It talks to the X server directly using (XCB-)DRI2
+protocol.</p>
 
 </li>
 </ul>
@@ -295,7 +280,6 @@ should as well lock the display before using it.
 
 <ul>
 <li>Pass the conformance tests</li>
-<li>Reference counting in main library?</li>
 <li>Mixed use of OpenGL, OpenGL ES 1.1, and OpenGL ES 2.0 is supported.  But
 which one of <code>libGL.so</code>, <code>libGLESv1_CM.so</code>, and
 <code>libGLESv2.so</code> should an application link to?  Bad things may happen

docs/fbdev-dri.html

@@ -9,6 +9,9 @@
 <center><h1>Mesa fbdev/DRI Drivers</h1></center>
 <br>
 
+<h1><center>NOTE: this information is obsolete and will be removed at
+a future date</center></h1>
+
 <h1>1. Introduction</h1>
 
 <p>
@@ -22,7 +25,7 @@ Contributors to this project include Jon Smirl, Keith Whitwell and Dave Airlie.
 
 <p>
 Applications in the fbdev/DRI environment use
-the <a href="http://www.nabble.com/file/p15480666/MiniGXL.html"> MiniGLX</a> interface to choose pixel
+the MiniGLX interface to choose pixel
 formats, create rendering contexts, etc.  It's a subset of the GLX and
 Xlib interfaces allowing some degree of application portability between
 the X and X-less environments.
@@ -315,8 +318,7 @@ It means that the sample_server process is not running.
 <h1>5.0 Programming Information</h1>
 
 <p>
-OpenGL/Mesa is interfaced to fbdev via the <a href="http://www.nabble.com/file/p15480666/MiniGLX.html">MiniGLX</a>
-interface.
+OpenGL/Mesa is interfaced to fbdev via the MiniGLX interface.
 MiniGLX is a subset of Xlib and GLX API functions which provides just
 enough functionality to setup OpenGL rendering and respond to simple
 input events.
@@ -332,7 +334,7 @@ This allows some degree of flexibility for software development and testing.
 However, the MiniGLX API is not binary-compatible with full Xlib/GLX.
 Some of the structures are different and some macros/functions work
 differently.
-See the <code>GL/miniglx.h</code> header file for details.
+See the GL/miniglx.h header file for details.
 </p>
 
 

docs/news.html

@@ -11,6 +11,22 @@
 <H1>News</H1>
 
 
+<h2>October 4, 2010</h2>
+
+<p>
+<a href="relnotes-7.9.html">Mesa 7.9</a> (final) is released.  This is a new
+development release.
+</p>
+
+
+<h2>September 27, 2010</h2>
+
+<p>
+<a href="relnotes-7.9.html">Mesa 7.9.0-rc1</a> is released.  This is a
+release candidate for the 7.9 development release.
+</p>
+
+
 <h2>June 16, 2010</h2>
 
 <p>
@@ -1277,7 +1293,6 @@ grateful.
 <p>
 </p><h2>March 18, 1999</h2>
 <p>The new webpages are now online. Enjoy, and let me know if you find any errors.
-For an eye-candy free version you can use <a href="http://www.mesa3d.org/txt/">http://www.mesa3d.org/txt/</a>.</p>
 <p>
 </p><h2>February 16, 1999</h2>
 <p><a href="http://www.sgi.com/">SGI</a> releases its <a href="http://www.sgi.com/software/opensource/glx/">GLX

docs/openvg.html

@@ -26,36 +26,27 @@ Please refer to <a href="egl.html">Mesa EGL</a> for more information about EGL.
 
 <h2>Building the library</h2>
 <ol>
-<li>Build Mesa3D with Gallium3D. Any build that builds Gallium3D libraries, EGL, and Gallium EGL drivers will suffice</li>
-<li>cd src/gallium/state_trackers/vega; make</li>
-<li>The last step will build libOpenVG library. You can add the libdir to LD_LIBRARY_PATH or install libOpenVG</li>
+<li>Run <code>configure</code> with <code>--enable-openvg</code>.  If you do
+not need OpenGL, you can add <code>--disable-opengl</code> to save the
+compilation time.</li>
+
+<li>Build and install Mesa as usual.</li>
 </ol>
 
 <h3>Sample build</h3>
 A sample build looks as follows:
 <pre>
-  $ ./configure --with-state-trackers=egl,vega --enable-gallium-intel
+  $ ./configure --disable-opengl --enable-openvg
   $ make
   $ make install
 </pre>
 
+<p>It will install <code>libOpenVG.so</code>, <code>libEGL.so</code>, and one
+or more EGL drivers.</p>
+
 <h2>OpenVG Demos</h2>
 
-<p>
-To build the OpenVG demos:
-</p>
-<pre>
-  cd progs/openvg
-  make
-</pre>
-<p>
-To run a demo:
-</p>
-<pre>
-  cd openvg/demos
-  ./lion
-</pre>
-
+<p>OpenVG demos can be found in mesa/demos repository.</p>
 
 </body>
 </html>

docs/relnotes-7.10.html

@@ -0,0 +1,56 @@
+<HTML>
+
+<TITLE>Mesa Release Notes</TITLE>
+
+<head><link rel="stylesheet" type="text/css" href="mesa.css"></head>
+
+<BODY>
+
+<body bgcolor="#eeeeee">
+
+<H1>Mesa 7.10 Release Notes / tbd</H1>
+
+<p>
+Mesa 7.10 is a new development release.
+People who are concerned with stability and reliability should stick
+with a previous release or wait for Mesa 7.10.1.
+</p>
+<p>
+Mesa 7.10 implements the OpenGL 2.1 API, but the version reported by
+glGetString(GL_VERSION) depends on the particular driver being used.
+Some drivers don't support all the features required in OpenGL 2.1.
+</p>
+<p>
+See the <a href="install.html">Compiling/Installing page</a> for prerequisites
+for DRI hardware acceleration.
+</p>
+
+
+<h2>MD5 checksums</h2>
+<pre>
+tbd
+</pre>
+
+
+<h2>New features</h2>
+<ul>
+<li>GL_ARB_explicit_attrib_location extension (Intel and software drivers).
+<li>GL_ARB_texture_rg (Intel, software drivers, gallium drivers).
+<li>GL_EXT_separate_shader_objects extension (Intel and software drivers).
+<li>GL_NV_primitive_restart extension (Gallium softpipe, llvmpipe).
+</ul>
+
+
+<h2>Bug fixes</h2>
+<ul>
+<li>tbd</li>
+</ul>
+
+
+<h2>Changes</h2>
+<ul>
+<li>tbd</li>
+</ul>
+
+</body>
+</html>

docs/relnotes-7.8.2.html

@@ -26,7 +26,15 @@ for DRI hardware acceleration.
 
 <h2>MD5 checksums</h2>
 <pre>
-tbd
+c89b63d253605ed40e8ac370d25a833c  MesaLib-7.8.2.tar.gz
+6be2d343a0089bfd395ce02aaf8adb57  MesaLib-7.8.2.tar.bz2
+a04ad3b06ac5ff3969a003fa7bbf7d5b  MesaLib-7.8.2.zip
+7c213f92efeb471f0331670d5079d4c0  MesaDemos-7.8.2.tar.gz
+757d9e2e06f48b1a52848be9b0307ced  MesaDemos-7.8.2.tar.bz2
+8d0e5cfe68b8ebf90265d350ae2c48b1  MesaDemos-7.8.2.zip
+b74482e3f44f35ed395c4aada4fd8240  MesaGLUT-7.8.2.tar.gz
+a471807b65e49c325808ba4551be93ed  MesaGLUT-7.8.2.tar.bz2
+9f190268c42be582ef66e47365ee61e3  MesaGLUT-7.8.2.zip
 </pre>
 
 
@@ -44,10 +52,95 @@ tbd
 <ul>
 <li>Fixed Gallium glDrawPixels(GL_DEPTH_COMPONENT).
 <li>Fixed Gallium Cell driver to buildable, runable state
+<li>Fixed bad error checking for glFramebufferRenderbuffer(attachment=GL_DEPTH_STENCIL_ATTACHMENT).
+<li>Fixed incorrect Z coordinate handling in "meta" glDraw/CopyPixels.
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=23670">Bug
+    #23670</a>.</li>
+
+<li>Assorted i965 driver fixes.
+    Including but not limited to:
+  <ul>
+    <li>Fix scissoring when width or height is
+    0.  <a href="https://bugs.freedesktop.org/show_bug.cgi?id=27643">Bug
+    #27643</a>.
+    <li>Fix bit allocation for number of color regions for
+    ARB_draw_buffers.</li>
+    <li>Set the correct provoking vertex for clipped first-mode
+    trifans.  <a href="https://bugs.freedesktop.org/show_bug.cgi?id=24470">Bug
+    #24470</a>.</li>
+    <li>Use <code>R16G16B16A16_FLOAT</code> for 3-component half-float.</li>
+    <li>Fix assertion for surface tile offset usage on Ironlake.</li>
+    <li>Fix cube map layouts on Ironlake.</li>
+    <li>When an RB gets a new region, clear the old from the state
+    cache.  <a href="https://bugs.freedesktop.org/show_bug.cgi?id=24119">Bug
+    #24119</a>.</li>
+    <li>Reject shaders with uninlined function calls instead of hanging.</li>
+  </ul>
+</li>
+
+<li>Assorted i915 driver fixes.  Including but not limited to:
+  <ul>
+    <li>Fixed texture LOD clamping in i915 driver.
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=24846">Bug
+    #24846</a>.</li>
+    <li>Fix off-by-one for drawing rectangle.
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=27408">Bug
+    #27408</a>.</li>
+  </ul>
+</li>
+
+<li>Fixed hangs in etracer on 830 and 845
+    chipsets.  <a href="https://bugs.freedesktop.org/show_bug.cgi?id=26557">Bug
+    #26557</a>.</li>
+<li>Fixed tiling of small textures on all Intel drivers.</li>
+<li>Fixed crash in Savage driver when using <code>_mesa_CopyTexImage2D</code>.
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=27652">Bug
+    #27652</a>.</li>
+
+<li>Assorted GLX fixes.  Including but not limited to:
+  <ul>
+    <li>Fixed <code>__glXInitializeVisualConfigFromTags</code>'s handling of
+        unrecognized fbconfig tags.</li>
+    <li>Fixed regression with <code>GLX_USE_GL</code>.
+    <li>Fixed config chooser logic for 'mask' matching.</li>
+    <li>Report swap events correctly in direct rendered case (DRI2)</li>
+    <li>Fixed build with dri2proto which doesn't define
+        <code>X_DRI2SwapInterval</code>.</li>
+    <li>Get <code>GLX_SCREEN</code> first in <code>__glXQueryContextInfo</code>.
+        <a href="https://bugs.freedesktop.org/show_bug.cgi?id=14245">Bug
+	#14245</a>.</li>
+  </ul>
+</li>
+
+<li>Assorted GLSL fixes.  Including but not limited to:
+  <ul>
+    <li>Change variable declared assertion into conditional in GLSL
+        compiler.  <a href="https://bugs.freedesktop.org/show_bug.cgi?id=27921">Bug
+        #27921</a>.</li>
+    <li>Fix instruction indexing
+        bugs.  <a href="https://bugs.freedesktop.org/show_bug.cgi?id=27566">Bug
+        #27566</a>.</li>
+    <li>Updated uniform location / offset encoding to be more like
+        other implementations.</li>
+    <li>Don't overwrite a driver's shader infolog with generic failure
+        message.</li>
+  </ul>
+</li>
+
+<li>Fixed OSMesa build for 16 and 32-bit color channel depth.
+<li>Fixed OSMesa build with hidden symbol visibility. libOSMesa no longer links to libGL.
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=28305">Bug
+    #28305</a>.
+<li>Fixed handling of multiple render targets in fixed-function
+    texture envrionmnent programs.</li>
+<li>Fixed conversion errors in <code>signed_rgba8888[rev]</code> texel
+    fetch.</li>
+<li>Don't set srcLevel on <code>GL_TEXTURE_RECTANGLE_ARB</code> targets.</li>
+<li>Various build fixes for OpenBSD.</li>
+<li>Various build fixes for OS X.</li>
+<li>Various build fixes for GCC 3.3.</li>
 </ul>
 
 
-<h2>Changes</h2>
-<p>None.</p>
 </body>
 </html>

docs/relnotes-7.8.3.html

@@ -0,0 +1,89 @@
+<HTML>
+
+<TITLE>Mesa Release Notes</TITLE>
+
+<head><link rel="stylesheet" type="text/css" href="mesa.css"></head>
+
+<BODY>
+
+<body bgcolor="#eeeeee">
+
+<H1>Mesa 7.8.3 Release Notes / (date tbd)</H1>
+
+<p>
+Mesa 7.8.3 is a bug fix release which fixes bugs found since the 7.8.2 release.
+</p>
+<p>
+Mesa 7.8.3 implements the OpenGL 2.1 API, but the version reported by
+glGetString(GL_VERSION) depends on the particular driver being used.
+Some drivers don't support all the features required in OpenGL 2.1.
+</p>
+<p>
+See the <a href="install.html">Compiling/Installing page</a> for prerequisites
+for DRI hardware acceleration.
+</p>
+
+
+<h2>MD5 checksums</h2>
+<pre>
+x  MesaLib-7.8.3.tar.gz
+x  MesaLib-7.8.3.tar.bz2
+x  MesaLib-7.8.3.zip
+x  MesaDemos-7.8.3.tar.gz
+x  MesaDemos-7.8.3.tar.bz2
+x  MesaDemos-7.8.3.zip
+x  MesaGLUT-7.8.3.tar.gz
+x  MesaGLUT-7.8.3.tar.bz2
+x  MesaGLUT-7.8.3.zip
+</pre>
+
+
+<h2>New features</h2>
+<p>None.</p>
+
+
+<h2>Changes</h2>
+<ul>
+<li>The radeon driver should use less memory when searching for a valid mip
+image.</li>
+</ul>
+
+
+<h2>Bug fixes</h2>
+<ul>
+<li>Fix unsupported FB with D24S8 (bug 
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=23670">29116</a>)</li>
+<li>Fix ReadPixels crash when reading depth/stencil from an FBO</li>
+<li>Fixed a bug rendering to 16-bit buffers using swrast.</li>
+<li>Fixed a state tracker/TGSI bug that caused crashes when using Windows'
+    memory debugging features.</li>
+<li>Fixed an issue rendering to 32-bit channels with swrast (bug
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=29487">29487</a>)</li>
+<li>GLSL: fix indirect <TT>gl_TextureMatrix</TT> addressing (bug
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=28967">28967</a>)</li>
+<li>GLSL: fix for bug
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=27216">27216</a></li>
+<li>GLSL: fix zw fragcoord entries in some cases (bug
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=29183">29183</a>)</li>
+<li>Fix texture env generation in some cases (bug
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=28169">28169</a>)</li>
+<li>osmesa: a fix for calling <TT>OSMesaMakeCurrent</TT> twice was applied (bug 
+    <a href="https://bugs.freedesktop.org/show_bug.cgi?id=10966">10966</a></li>
+<li>A bug was fixed which could cause Mesa to ignore the
+    <TT>MESA_EXTENSION_OVERRIDE</TT> environment variable.</li>
+<li>A bug related to specular highlights on backfaces was fixed.</li>
+<li>A radeon-specific issue with <TT>glCopyTex(Sub)Image</TT> was
+    corrected.</li>
+<li>radeon/wine: flush command stream in more cases, fixing wine d3d9
+    tests.</li>
+<li>r600: fix sin+cos normalization.</li>
+<li>r600: (properly) ignore <TT>GL_COORD_REPLACE</TT> when point sprites are
+    disabled.</li>
+<li>radeon: avoid flushing when the context is not current.</li>
+<li>r300c: a bug affecting unaligned BOs was fixed.</li>
+<li>r300c: a hardlock caused by ARB_half_float_vertex incorrectly advertised on some chipsets.</li>
+</ul>
+
+
+</body>
+</html>

docs/relnotes-7.9.html

@@ -8,7 +8,7 @@
 
 <body bgcolor="#eeeeee">
 
-<H1>Mesa 7.9 Release Notes / date TBD</H1>
+<H1>Mesa 7.9 Release Notes / October 4, 2010</H1>
 
 <p>
 Mesa 7.9 is a new development release.
@@ -28,7 +28,12 @@ for DRI hardware acceleration.
 
 <h2>MD5 checksums</h2>
 <pre>
-tbd
+ed65ab425b25895c7f473d0a5e6e64f8  MesaLib-7.9.tar.gz
+82c740c49d572baa6da2b1a1eee90bca  MesaLib-7.9.tar.bz2
+cd2b6ecec759b0457475e94bbb38fedb  MesaLib-7.9.zip
+7b54af9fb9b1f6a1a65db2520f50848f  MesaGLUT-7.9.tar.gz
+20d07419d1929f833fdb36bced290ad5  MesaGLUT-7.9.tar.bz2
+62a7edecd7c92675cd6029b05217eb0a  MesaGLUT-7.9.zip
 </pre>
 
 
@@ -37,16 +42,85 @@ tbd
 <li>New, improved GLSL compiler written by Intel.
     See the <a href="shading.html"> Shading Language</a> page for
     more information.
-<li>GL_EXT_timer_query extension (i965 driver only)
+<li>New, very experimental Gallium driver for R600-R700 Radeons.
+<li>Support for AMD Evergreen-based Radeons (HD 5xxx)
+<li>GL_EXT_timer_query extension (i965 driver and softpipe only)
 <li>GL_EXT_framebuffer_multisample extension (intel drivers, MAX_SAMPLES = 1)
 <li>GL_ARB_texture_swizzle extension (alias of GL_EXT_texture_swizzle)
-<li>GL_ARB_draw_elements_base_vertex, GL_ARB_fragment_program_shadow
-    and GL_EXT_draw_buffers2 in Gallium drivers
+<li>GL_ARB_draw_elements_base_vertex, GL_ARB_fragment_program_shadow,
+    GL_ARB_window_pos, GL_EXT_gpu_program_parameters,
+    GL_ATI_texture_env_combine3, GL_MESA_pack_invert, and GL_OES_EGL_image
+    extensions in Gallium drivers
+<li>GL_ARB_depth_clamp and GL_NV_depth_clamp extensions (in nv50 and r600
+    Gallium drivers)
+<li>GL_ARB_half_float_vertex extension (in nvfx, r300, r600, softpipe,
+    and llvmpipe Gallium drivers)
+<li>GL_EXT_draw_buffers2 (in nv50, r600, softpipe, and llvmpipe Gallium
+    drivers)
+<li>GL_EXT_texture_swizzle (in nvfx, r300, r600, softpipe, and llvmpipe
+    Gallium drivers)
+<li>GL_ATI_texture_mirror_once (in nvfx, nv50, r300, r600, softpipe, and
+    llvmpipe Gallium drivers)
+<li>GL_NV_conditional_render (in r300 Gallium driver)
+<li>Initial "signs of life" support for Sandybridge hardware in i965 DRI
+    driver.
 </ul>
 
 
 <h2>Bug fixes</h2>
+<p>This list is likely incomplete.</p>
 <ul>
+<li>Massive improvements to the Gallium driver for R300-R500 Radeons; this
+    driver is now considered stable for use as a DRI (OpenGL) driver.
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=10908">Bug 10908</a> - GLSL: gl_FogParamaters gl_Fog built-in uniform not functioning</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=13753">Bug 13753</a> - Numerous bugs in GLSL uniform handling</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=16854">Bug 16854</a> - GLSL function call at global scope causes SEGV</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=16856">Bug 16856</a> - GLSL indexing of unsized array results in assertion failure</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=18659">Bug 18659</a> - Crash in shader/slang/slang_codegen.c _slang_gen_function_call_name()</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=19089">Bug 19089</a> - [GLSL] glsl1/shadow2D() cases fail</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=22622">Bug 22622</a> - [GM965 GLSL] noise*() cause GPU lockup</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=23743">Bug 23743</a> - For loop from 0 to 0 not optimized out</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=24553">Bug 24553</a> - shader compilation times explode when using more () pairs</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25664">Bug 25664</a> - [GLSL] re-declaring an empty array fails to compile</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25769">Bug 25769</a> - [GLSL] "float" can be implicitly converted to "int"</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25808">Bug 25808</a> - [GLSL] const variable is modified successfully</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25826">Bug 25826</a> - [GLSL] declaring an unsized array then re-declaring with a size fails</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25827">Bug 25827</a> - [GLSL] vector constructor accepts too many arguments successfully</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25829">Bug 25829</a> - [GLSL] allowing non-void function without returning value</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25830">Bug 25830</a> - [GLSL] allowing non-constant-expression as const declaration initializer</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25877">Bug 25877</a> - [GLSL 1.10] implicit conversion from "int" to "float" should not be allowed</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25878">Bug 25878</a> - [GLSL] sampler is converted to int successfully</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25994">Bug 25994</a> - [GM45][GLSL] 'return' statement in vertex shader unsupported</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=25999">Bug 25999</a> - [GLSL] embedded structure constructor fails to compile</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=26000">Bug 26000</a> - [GLSL] allowing different parameter qualifier between the function definition and declaration</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=26001">Bug 26001</a> - [GLSL 1.10] constructing matrix from matrix succeeds</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=26224">Bug 26224</a> - [GLSL] Cannot get location of a uniform struct member</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=26990">Bug 26990</a> - [GLSL] variable declaration in "while" fails to compile</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27004">Bug 27004</a> - [GLSL] allowing macro redefinition</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27060">Bug 27060</a> - [965] piglit glsl-fs-raytrace failure due to lack of function calls.</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27216">Bug 27216</a> - Assignment with a function call in an if statement causes an assertion failure</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27261">Bug 27261</a> - GLSL Compiler fails on the following vertex shader</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27265">Bug 27265</a> - GLSL Compiler doesnt link the attached vertex shader</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27388">Bug 27388</a> - [i965] piglit glsl-vs-arrays failure</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27403">Bug 27403</a> - GLSL struct causing "Invalid src register file ..." error</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=27914">Bug 27914</a> - GLSL compiler uses MUL+ADD where it could use MAD</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28055">Bug 28055</a> - glsl-texcoord-array fails GLSL compilation</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28374">Bug 28374</a> - SIGSEGV shader/slang/slang_typeinfo.c:534</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28748">Bug 28748</a> - [i965] uninlined function calls support</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28833">Bug 28833</a> - piglit/shaders/glsl-texcoord-array fail</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28834">Bug 28834</a> - Add support for system fpclassify to GL_OES_query_matrix function for OpenBSD / NetBSD</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28837">Bug 28837</a> - varying vec4 index support</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28845">Bug 28845</a> - The GLU tesselator code has some warnings</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28889">Bug 28889</a> - [regression] wine game crash</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28894">Bug 28894</a> - slang build fails if absolute path contains spaces</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28913">Bug 28913</a> - [GLSL] allowing two version statements</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28931">Bug 28931</a> - Floating Point Exception in Warzone2100 Trunk version</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28966">Bug 28966</a> - [r300g] Dynamic branching 3 demo does not run</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=28967">Bug 28967</a> - slang/slang_emit.c:350: storage_to_src_reg: Assertion `index &gt;= 0' failed.</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=29013">Bug 29013</a> - [r300g] translate_rgb_op: unknown opcode ILLEGAL OPCODE</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=29020">Bug 29020</a> - [r300g] Wine d3d9 tests hardlock</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=29910">Bug 29910</a> - Mesa advertises bogus GL_ARB_shading_language_120</li>
+<li><a href="https://bugs.freedesktop.org/show_bug.cgi?id=30196">Bug 30196</a> - [GLSL] gl_TextureMatrix{Inverse,Transpose,InverseTranspose} unsupported</li>
 </ul>
 
 

docs/relnotes.html

@@ -14,6 +14,7 @@ The release notes summarize what's new or changed in each Mesa release.
 
 <UL>
 <LI><A HREF="relnotes-7.9.html">7.9 release notes</A>
+<LI><A HREF="relnotes-7.8.3.html">7.8.3 release notes</A>
 <LI><A HREF="relnotes-7.8.2.html">7.8.2 release notes</A>
 <LI><A HREF="relnotes-7.8.1.html">7.8.1 release notes</A>
 <LI><A HREF="relnotes-7.8.html">7.8 release notes</A>

include/GL/internal/dri_interface.h

@@ -805,7 +805,7 @@ typedef struct __DRIimageExtensionRec __DRIimageExtension;
 struct __DRIimageExtensionRec {
     __DRIextension base;
 
-    __DRIimage *(*createImageFromName)(__DRIcontext *context,
+    __DRIimage *(*createImageFromName)(__DRIscreen *screen,
 				       int width, int height, int format,
 				       int name, int pitch,
 				       void *loaderPrivate);
@@ -841,7 +841,7 @@ typedef struct __DRIimageLookupExtensionRec __DRIimageLookupExtension;
 struct __DRIimageLookupExtensionRec {
     __DRIextension base;
 
-    __DRIimage *(*lookupEGLImage)(__DRIcontext *context, void *image,
+    __DRIimage *(*lookupEGLImage)(__DRIscreen *screen, void *image,
 				  void *loaderPrivate);
 };
 

scons/crossmingw.py

@@ -54,11 +54,13 @@ prefixes32 = SCons.Util.Split("""
     i586-mingw32msvc-
     i686-mingw32msvc-
     i686-pc-mingw32-
+    i686-w64-mingw32-
 """)
 prefixes64 = SCons.Util.Split("""
     amd64-mingw32-
     amd64-mingw32msvc-
     amd64-pc-mingw32-
+    x86_64-w64-mingw32-
 """)
 
 def find(env):

scons/gallium.py

@@ -49,30 +49,35 @@ def symlink(target, source, env):
     os.symlink(os.path.basename(source), target)
 
 def install(env, source, subdir):
-    target_dir = os.path.join(env.Dir('#.').srcnode().abspath, env['build'], subdir)
-    env.Install(target_dir, source)
+    target_dir = os.path.join(env.Dir('#.').srcnode().abspath, env['build_dir'], subdir)
+    return env.Install(target_dir, source)
 
 def install_program(env, source):
-    install(env, source, 'bin')
+    return install(env, source, 'bin')
 
 def install_shared_library(env, sources, version = ()):
-    install_dir = os.path.join(env.Dir('#.').srcnode().abspath, env['build'])
+    targets = []
+    install_dir = os.path.join(env.Dir('#.').srcnode().abspath, env['build_dir'])
     version = tuple(map(str, version))
     if env['SHLIBSUFFIX'] == '.dll':
         dlls = env.FindIxes(sources, 'SHLIBPREFIX', 'SHLIBSUFFIX')
-        install(env, dlls, 'bin')
+        targets += install(env, dlls, 'bin')
         libs = env.FindIxes(sources, 'LIBPREFIX', 'LIBSUFFIX')
-        install(env, libs, 'lib')
+        targets += install(env, libs, 'lib')
     else:
         for source in sources:
             target_dir =  os.path.join(install_dir, 'lib')
             target_name = '.'.join((str(source),) + version)
             last = env.InstallAs(os.path.join(target_dir, target_name), source)
+            targets += last
             while len(version):
                 version = version[:-1]
                 target_name = '.'.join((str(source),) + version)
                 action = SCons.Action.Action(symlink, "$TARGET -> $SOURCE")
                 last = env.Command(os.path.join(target_dir, target_name), last, action) 
+                targets += last
+    return targets
+
 
 def createInstallMethods(env):
     env.AddMethod(install_program, 'InstallProgram')
@@ -98,6 +103,41 @@ def num_jobs():
     return 1
 
 
+def pkg_config_modules(env, name, modules):
+    '''Simple wrapper for pkg-config.'''
+
+    env[name] = False
+
+    if env['platform'] == 'windows':
+        return
+
+    if not env.Detect('pkg-config'):
+        return
+
+    if subprocess.call(["pkg-config", "--exists", ' '.join(modules)]) != 0:
+        return
+
+    # Put -I and -L flags directly into the environment, as these don't affect
+    # the compilation of targets that do not use them
+    try:
+        env.ParseConfig('pkg-config --cflags-only-I --libs-only-L ' + ' '.join(modules))
+    except OSError:
+        return
+
+    # Other flags may affect the compilation of unrelated targets, so store
+    # them with a prefix, (e.g., XXX_CFLAGS, XXX_LIBS, etc)
+    try:
+        flags = env.ParseFlags('!pkg-config --cflags-only-other --libs-only-l --libs-only-other ' + ' '.join(modules))
+    except OSError:
+        return
+    prefix = name.upper() + '_'
+    for flag_name, flag_value in flags.iteritems():
+        env[prefix + flag_name] = flag_value
+
+    env[name] = True
+
+
+
 def generate(env):
     """Common environment generation code"""
 
@@ -110,27 +150,32 @@ def generate(env):
             env['toolchain'] = 'wcesdk'
     env.Tool(env['toolchain'])
 
-    if env['platform'] == 'embedded':
-        # Allow overriding compiler from environment
-        if os.environ.has_key('CC'):
-            env['CC'] = os.environ['CC']
-            # Update CCVERSION to match
-            pipe = SCons.Action._subproc(env, [env['CC'], '--version'],
-                                         stdin = 'devnull',
-                                         stderr = 'devnull',
-                                         stdout = subprocess.PIPE)
-            if pipe.wait() == 0:
-                line = pipe.stdout.readline()
-                match = re.search(r'[0-9]+(\.[0-9]+)+', line)
-                if match:
-                    env['CCVERSION'] = match.group(0)
-            
+    # Allow override compiler and specify additional flags from environment
+    if os.environ.has_key('CC'):
+        env['CC'] = os.environ['CC']
+        # Update CCVERSION to match
+        pipe = SCons.Action._subproc(env, [env['CC'], '--version'],
+                                     stdin = 'devnull',
+                                     stderr = 'devnull',
+                                     stdout = subprocess.PIPE)
+        if pipe.wait() == 0:
+            line = pipe.stdout.readline()
+            match = re.search(r'[0-9]+(\.[0-9]+)+', line)
+            if match:
+                env['CCVERSION'] = match.group(0)
+    if os.environ.has_key('CFLAGS'):
+        env['CCFLAGS'] += SCons.Util.CLVar(os.environ['CFLAGS'])
+    if os.environ.has_key('CXX'):
+        env['CXX'] = os.environ['CXX']
+    if os.environ.has_key('CXXFLAGS'):
+        env['CXXFLAGS'] += SCons.Util.CLVar(os.environ['CXXFLAGS'])
+    if os.environ.has_key('LDFLAGS'):
+        env['LINKFLAGS'] += SCons.Util.CLVar(os.environ['LDFLAGS'])
 
     env['gcc'] = 'gcc' in os.path.basename(env['CC']).split('-')
     env['msvc'] = env['CC'] == 'cl'
 
     # shortcuts
-    debug = env['debug']
     machine = env['machine']
     platform = env['platform']
     x86 = env['machine'] == 'x86'
@@ -138,20 +183,48 @@ def generate(env):
     gcc = env['gcc']
     msvc = env['msvc']
 
+    # Backwards compatability with the debug= profile= options
+    if env['build'] == 'debug':
+        if not env['debug']:
+            print 'scons: warning: debug option is deprecated and will be removed eventually; use instead'
+            print
+            print ' scons build=release'
+            print
+            env['build'] = 'release'
+        if env['profile']:
+            print 'scons: warning: profile option is deprecated and will be removed eventually; use instead'
+            print
+            print ' scons build=profile'
+            print
+            env['build'] = 'profile'
+    if False:
+        # Enforce SConscripts to use the new build variable
+        env.popitem('debug')
+        env.popitem('profile')
+    else:
+        # Backwards portability with older sconscripts
+        if env['build'] in ('debug', 'checked'):
+            env['debug'] = True
+            env['profile'] = False
+        if env['build'] == 'profile':
+            env['debug'] = False
+            env['profile'] = True
+        if env['build'] == 'release':
+            env['debug'] = False
+            env['profile'] = False
+
     # Put build output in a separate dir, which depends on the current
     # configuration. See also http://www.scons.org/wiki/AdvancedBuildExample
     build_topdir = 'build'
     build_subdir = env['platform']
     if env['machine'] != 'generic':
         build_subdir += '-' + env['machine']
-    if env['debug']:
-        build_subdir += "-debug"
-    if env['profile']:
-        build_subdir += "-profile"
+    if env['build'] != 'release':
+        build_subdir += '-' +  env['build']
     build_dir = os.path.join(build_topdir, build_subdir)
     # Place the .sconsign file in the build dir too, to avoid issues with
     # different scons versions building the same source file
-    env['build'] = build_dir
+    env['build_dir'] = build_dir
     env.SConsignFile(os.path.join(build_dir, '.sconsign'))
     if 'SCONS_CACHE_DIR' in os.environ:
         print 'scons: Using build cache in %s.' % (os.environ['SCONS_CACHE_DIR'],)
@@ -163,13 +236,16 @@ def generate(env):
     if env.GetOption('num_jobs') <= 1:
         env.SetOption('num_jobs', num_jobs())
 
+    env.Decider('MD5-timestamp')
+    env.SetOption('max_drift', 60)
+
     # C preprocessor options
     cppdefines = []
-    if debug:
+    if env['build'] in ('debug', 'checked'):
         cppdefines += ['DEBUG']
     else:
         cppdefines += ['NDEBUG']
-    if env['profile']:
+    if env['build'] == 'profile':
         cppdefines += ['PROFILE']
     if platform == 'windows':
         cppdefines += [
@@ -190,7 +266,7 @@ def generate(env):
                 '_SCL_SECURE_NO_WARNINGS',
                 '_SCL_SECURE_NO_DEPRECATE',
             ]
-        if debug:
+        if env['build'] in ('debug', 'checked'):
             cppdefines += ['_DEBUG']
     if env['toolchain'] == 'winddk':
         # Mimic WINDDK's builtin flags. See also:
@@ -217,7 +293,7 @@ def generate(env):
             ('__BUILDMACHINE__', 'WinDDK'),
             ('FPO', '0'),
         ]
-        if debug:
+        if env['build'] in ('debug', 'checked'):
             cppdefines += [('DBG', 1)]
     if platform == 'wince':
         cppdefines += [
@@ -253,15 +329,16 @@ def generate(env):
     ccflags = [] # C & C++
     if gcc:
         ccversion = env['CCVERSION']
-        if debug:
-            ccflags += ['-O0', '-g3']
+        if env['build'] == 'debug':
+            ccflags += ['-O0']
         elif ccversion.startswith('4.2.'):
             # gcc 4.2.x optimizer is broken
             print "warning: gcc 4.2.x optimizer is broken -- disabling optimizations"
-            ccflags += ['-O0', '-g3']
+            ccflags += ['-O0']
         else:
-            ccflags += ['-O3', '-g3']
-        if env['profile']:
+            ccflags += ['-O3']
+        ccflags += ['-g3']
+        if env['build'] in ('checked', 'profile'):
             # See http://code.google.com/p/jrfonseca/wiki/Gprof2Dot#Which_options_should_I_pass_to_gcc_when_compiling_for_profiling?
             ccflags += [
                 '-fno-omit-frame-pointer',
@@ -320,7 +397,7 @@ def generate(env):
         # See also:
         # - http://msdn.microsoft.com/en-us/library/19z1t1wy.aspx
         # - cl /?
-        if debug:
+        if env['build'] == 'debug':
             ccflags += [
               '/Od', # disable optimizations
               '/Oi', # enable intrinsic functions
@@ -389,7 +466,7 @@ def generate(env):
     if env['platform'] == 'windows' and msvc:
         # Choose the appropriate MSVC CRT
         # http://msdn.microsoft.com/en-us/library/2kzt1wy3.aspx
-        if env['debug']:
+        if env['build'] in ('debug', 'checked'):
             env.Append(CCFLAGS = ['/MTd'])
             env.Append(SHCCFLAGS = ['/LDd'])
         else:
@@ -421,7 +498,7 @@ def generate(env):
         else:
             env['_LIBFLAGS'] = '-Wl,--start-group ' + env['_LIBFLAGS'] + ' -Wl,--end-group'
     if msvc:
-        if not env['debug']:
+        if env['build'] != 'debug':
             # enable Link-time Code Generation
             linkflags += ['/LTCG']
             env.Append(ARFLAGS = ['/LTCG'])
@@ -460,7 +537,7 @@ def generate(env):
 
             '/entry:DrvEnableDriver',
         ]
-        if env['debug'] or env['profile']:
+        if env['build'] != 'release':
             linkflags += [
                 '/MAP', # http://msdn.microsoft.com/en-us/library/k7xkk3e2.aspx
             ]
@@ -477,9 +554,19 @@ def generate(env):
     # Default libs
     env.Append(LIBS = [])
 
-    # Load LLVM
+    # Load tools
     if env['llvm']:
         env.Tool('llvm')
+        env.Tool('udis86')
+    
+    pkg_config_modules(env, 'x11', ['x11', 'xext'])
+    pkg_config_modules(env, 'drm', ['libdrm'])
+    pkg_config_modules(env, 'drm_intel', ['libdrm_intel'])
+    pkg_config_modules(env, 'drm_radeon', ['libdrm_radeon'])
+    pkg_config_modules(env, 'xorg', ['xorg-server'])
+    pkg_config_modules(env, 'kms', ['libkms'])
+
+    env['dri'] = env['x11'] and env['drm']
 
     # Custom builders and methods
     env.Tool('custom')

scons/llvm.py

@@ -38,6 +38,8 @@ import SCons.Util
 
 
 def generate(env):
+    env['llvm'] = False
+
     try:
         llvm_dir = os.environ['LLVM']
     except KeyError:
@@ -64,13 +66,13 @@ def generate(env):
         # XXX: There is no llvm-config on Windows, so assume a standard layout
         if llvm_dir is None:
             print 'scons: LLVM environment variable must be specified when building for windows'
-            env.Exit(1)
+            return
 
         # Try to determine the LLVM version from llvm/Config/config.h
         llvm_config = os.path.join(llvm_dir, 'include/llvm/Config/config.h')
         if not os.path.exists(llvm_config):
             print 'scons: could not find %s' % llvm_config
-            env.Exit(1)
+            return
         llvm_version_re = re.compile(r'^#define PACKAGE_VERSION "([^"]*)"')
         llvm_version = None
         for line in open(llvm_config, 'rt'):
@@ -81,7 +83,7 @@ def generate(env):
                 break
         if llvm_version is None:
             print 'scons: could not determine the LLVM version from %s' % llvm_config
-            env.Exit(1)
+            return
 
         env.Prepend(CPPPATH = [os.path.join(llvm_dir, 'include')])
         env.AppendUnique(CPPDEFINES = [
@@ -124,7 +126,7 @@ def generate(env):
             # Some of the LLVM C headers use the inline keyword without
             # defining it.
             env.Append(CPPDEFINES = [('inline', '__inline')])
-            if env['debug']:
+            if env['build'] in ('debug', 'checked'):
                 # LLVM libraries are static, build with /MT, and they
                 # automatically link agains LIBCMT. When we're doing a
                 # debug build we'll be linking against LIBCMTD, so disable
@@ -133,7 +135,7 @@ def generate(env):
     else:
         if not env.Detect('llvm-config'):
             print 'scons: llvm-config script not found' % llvm_version
-            env.Exit(1)
+            return
 
         llvm_version = env.backtick('llvm-config --version').rstrip()
         llvm_version = distutils.version.LooseVersion(llvm_version)
@@ -144,11 +146,12 @@ def generate(env):
             env.ParseConfig('llvm-config --ldflags')
         except OSError:
             print 'scons: llvm-config version %s failed' % llvm_version
-            env.Exit(1)
+            return
         else:
             env['LINK'] = env['CXX']
 
     assert llvm_version is not None
+    env['llvm'] = True
 
     print 'scons: Found LLVM version %s' % llvm_version
     env['LLVM_VERSION'] = llvm_version

scons/udis86.py

@@ -31,8 +31,10 @@ def generate(env):
     conf = env.Configure()
 
     if conf.CheckHeader('udis86.h'): # and conf.CheckLib('udis86'):
-        env.Append(CPPDEFINES = [('HAVE_UDIS86', '1')])
+        env['UDIS86'] = True
         env.Prepend(LIBS = ['udis86'])
+    else:
+        env['UDIS86'] = False
 
     conf.Finish()
 

scons/wcesdk.py

@@ -122,7 +122,7 @@ def get_wce600_paths(env):
     host_cpu = os.environ.get('_HOSTCPUTYPE', 'i386')
     target_cpu = os.environ.get('_TGTCPU', 'x86')
 
-    if env['debug']:
+    if env['build'] == 'debug':
         build = 'debug'
     else:
         build = 'retail'

src/SConscript

@@ -1,19 +1,17 @@
 Import('*')
 
-if 'egl' in env['statetrackers']:
-    SConscript('mapi/vgapi/SConscript')
+SConscript('mapi/vgapi/SConscript')
+
+if env['platform'] == 'windows':
     SConscript('egl/main/SConscript')
+    SConscript('talloc/SConscript')
 
-if 'mesa' in env['statetrackers']:
-    if platform == 'windows':
-        SConscript('talloc/SConscript')
+SConscript('glsl/SConscript')
+SConscript('mapi/glapi/SConscript')
+SConscript('mesa/SConscript')
 
-    SConscript('glsl/SConscript')
-    SConscript('mapi/glapi/SConscript')
-    SConscript('mesa/SConscript')
-
-    if platform != 'embedded':
-        SConscript('glut/glx/SConscript')
+if env['platform'] != 'embedded':
+    SConscript('glut/glx/SConscript')
 
 SConscript('gallium/SConscript')
 

src/egl/drivers/Makefile.template

@@ -24,8 +24,8 @@ $(EGL_DRIVER_PATH): $(EGL_DRIVER)
 
 $(EGL_DRIVER): $(EGL_OBJECTS) Makefile $(TOP)/src/egl/drivers/Makefile.template
 	@$(MKLIB) -o $(EGL_DRIVER) -noprefix \
-		-linker '$(CC)' -ldflags '$(LDFLAGS)' \
-		-L$(TOP)/$(LIB_DIR) $(MKLIB_OPTIONS) \
+		-linker '$(CC)' -ldflags '-L$(TOP)/$(LIB_DIR) $(LDFLAGS)' \
+		$(MKLIB_OPTIONS) \
 		$(EGL_OBJECTS) $(EGL_LIBS) -l$(EGL_LIB)
 
 .c.o:

src/egl/drivers/dri2/egl_dri2.c

@@ -248,21 +248,20 @@ dri2_add_config(_EGLDisplay *disp, const __DRIconfig *dri_config, int id,
    if (double_buffer)
       return NULL;
 
-   if (depth > 0 && depth != _eglGetConfigKey(&base, EGL_BUFFER_SIZE))
+   if (depth > 0 && depth != base.BufferSize)
       return NULL;
 
-   _eglSetConfigKey(&base, EGL_NATIVE_RENDERABLE, EGL_TRUE);
+   base.NativeRenderable = EGL_TRUE;
 
-   _eglSetConfigKey(&base, EGL_SURFACE_TYPE, surface_type);
+   base.SurfaceType = surface_type;
    if (surface_type & (EGL_PIXMAP_BIT | EGL_PBUFFER_BIT)) {
-      _eglSetConfigKey(&base, EGL_BIND_TO_TEXTURE_RGB, bind_to_texture_rgb);
-      if (_eglGetConfigKey(&base, EGL_ALPHA_SIZE) > 0)
-	 _eglSetConfigKey(&base,
-			  EGL_BIND_TO_TEXTURE_RGBA, bind_to_texture_rgba);
+      base.BindToTextureRGB = bind_to_texture_rgb;
+      if (base.AlphaSize > 0)
+         base.BindToTextureRGBA = bind_to_texture_rgba;
    }
 
-   _eglSetConfigKey(&base, EGL_RENDERABLE_TYPE, disp->ClientAPIsMask);
-   _eglSetConfigKey(&base, EGL_CONFORMANT, disp->ClientAPIsMask);
+   base.RenderableType = disp->ClientAPIsMask;
+   base.Conformant = disp->ClientAPIsMask;
 
    if (!_eglValidateConfig(&base, EGL_FALSE)) {
       _eglLog(_EGL_DEBUG, "DRI2: failed to validate config %d", id);
@@ -273,7 +272,7 @@ dri2_add_config(_EGLDisplay *disp, const __DRIconfig *dri_config, int id,
    if (conf != NULL) {
       memcpy(&conf->base, &base, sizeof base);
       conf->dri_config = dri_config;
-      _eglAddConfig(disp, &conf->base);
+      _eglLinkConfig(&conf->base);
    }
 
    return conf;
@@ -292,7 +291,7 @@ dri2_process_buffers(struct dri2_egl_surface *dri2_surf,
    struct dri2_egl_display *dri2_dpy =
       dri2_egl_display(dri2_surf->base.Resource.Display);
    xcb_rectangle_t rectangle;
-   int i;
+   unsigned i;
 
    dri2_surf->buffer_count = count;
    dri2_surf->have_fake_front = 0;
@@ -339,6 +338,8 @@ dri2_get_buffers(__DRIdrawable * driDrawable,
    xcb_dri2_get_buffers_reply_t *reply;
    xcb_dri2_get_buffers_cookie_t cookie;
 
+   (void) driDrawable;
+
    cookie = xcb_dri2_get_buffers_unchecked (dri2_dpy->conn,
 					    dri2_surf->drawable,
 					    count, count, attachments);
@@ -360,23 +361,28 @@ dri2_get_buffers(__DRIdrawable * driDrawable,
 static void
 dri2_flush_front_buffer(__DRIdrawable * driDrawable, void *loaderPrivate)
 {
+   (void) driDrawable;
+
    /* FIXME: Does EGL support front buffer rendering at all? */
 
 #if 0
    struct dri2_egl_surface *dri2_surf = loaderPrivate;
 
    dri2WaitGL(dri2_surf);
+#else
+   (void) loaderPrivate;
 #endif
 }
 
 static __DRIimage *
-dri2_lookup_egl_image(__DRIcontext *context, void *image, void *data)
+dri2_lookup_egl_image(__DRIscreen *screen, void *image, void *data)
 {
-   struct dri2_egl_context *dri2_ctx = data;
-   _EGLDisplay *disp = dri2_ctx->base.Resource.Display;
+   _EGLDisplay *disp = data;
    struct dri2_egl_image *dri2_img;
    _EGLImage *img;
 
+   (void) screen;
+
    img = _eglLookupImage(image, disp);
    if (img == NULL) {
       _eglError(EGL_BAD_PARAMETER, "dri2_lookup_egl_image");
@@ -407,6 +413,8 @@ dri2_get_buffers_with_format(__DRIdrawable * driDrawable,
    xcb_dri2_get_buffers_with_format_cookie_t cookie;
    xcb_dri2_attach_format_t *format_attachments;
 
+   (void) driDrawable;
+
    format_attachments = (xcb_dri2_attach_format_t *) attachments;
    cookie = xcb_dri2_get_buffers_with_format_unchecked (dri2_dpy->conn,
 							dri2_surf->drawable,
@@ -440,14 +448,14 @@ struct dri2_extension_match {
 static struct dri2_extension_match dri2_driver_extensions[] = {
    { __DRI_CORE, 1, offsetof(struct dri2_egl_display, core) },
    { __DRI_DRI2, 1, offsetof(struct dri2_egl_display, dri2) },
-   { NULL }
+   { NULL, 0, 0 }
 };
 
 static struct dri2_extension_match dri2_core_extensions[] = {
    { __DRI2_FLUSH, 1, offsetof(struct dri2_egl_display, flush) },
    { __DRI_TEX_BUFFER, 2, offsetof(struct dri2_egl_display, tex_buffer) },
    { __DRI_IMAGE, 1, offsetof(struct dri2_egl_display, image) },
-   { NULL }
+   { NULL, 0, 0 }
 };
 
 static EGLBoolean
@@ -728,7 +736,7 @@ dri2_create_screen(_EGLDisplay *disp)
    dri2_dpy = disp->DriverData;
    dri2_dpy->dri_screen =
       dri2_dpy->dri2->createNewScreen(0, dri2_dpy->fd, dri2_dpy->extensions,
-				      &dri2_dpy->driver_configs, dri2_dpy);
+				      &dri2_dpy->driver_configs, disp);
 
    if (dri2_dpy->dri_screen == NULL) {
       _eglLog(_EGL_WARNING, "DRI2: failed to create dri screen");
@@ -772,6 +780,8 @@ dri2_initialize_x11(_EGLDriver *drv, _EGLDisplay *disp,
 {
    struct dri2_egl_display *dri2_dpy;
 
+   (void) drv;
+
    dri2_dpy = malloc(sizeof *dri2_dpy);
    if (!dri2_dpy)
       return _eglError(EGL_BAD_ALLOC, "eglInitialize");
@@ -1075,6 +1085,8 @@ dri2_create_context(_EGLDriver *drv, _EGLDisplay *disp, _EGLConfig *conf,
    const __DRIconfig *dri_config;
    int api;
 
+   (void) drv;
+
    dri2_ctx = malloc(sizeof *dri2_ctx);
    if (!dri2_ctx) {
       _eglError(EGL_BAD_ALLOC, "eglCreateContext");
@@ -1146,7 +1158,9 @@ dri2_destroy_surface(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *surf)
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
    struct dri2_egl_surface *dri2_surf = dri2_egl_surface(surf);
 
-   if (_eglIsSurfaceBound(surf))
+   (void) drv;
+
+   if (!_eglPutSurface(surf))
       return EGL_TRUE;
 
    (*dri2_dpy->core->destroyDrawable)(dri2_surf->dri_drawable);
@@ -1173,15 +1187,17 @@ dri2_make_current(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
    struct dri2_egl_surface *dri2_dsurf = dri2_egl_surface(dsurf);
    struct dri2_egl_surface *dri2_rsurf = dri2_egl_surface(rsurf);
    struct dri2_egl_context *dri2_ctx = dri2_egl_context(ctx);
+   _EGLContext *old_ctx;
+   _EGLSurface *old_dsurf, *old_rsurf;
    __DRIdrawable *ddraw, *rdraw;
    __DRIcontext *cctx;
 
-   /* bind the new context and return the "orphaned" one */
-   if (!_eglBindContext(&ctx, &dsurf, &rsurf))
+   /* make new bindings */
+   if (!_eglBindContext(ctx, dsurf, rsurf, &old_ctx, &old_dsurf, &old_rsurf))
       return EGL_FALSE;
 
    /* flush before context switch */
-   if (ctx && dri2_drv->glFlush)
+   if (old_ctx && dri2_drv->glFlush)
       dri2_drv->glFlush();
 
    ddraw = (dri2_dsurf) ? dri2_dsurf->dri_drawable : NULL;
@@ -1190,16 +1206,29 @@ dri2_make_current(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
 
    if ((cctx == NULL && ddraw == NULL && rdraw == NULL) ||
        dri2_dpy->core->bindContext(cctx, ddraw, rdraw)) {
-      if (dsurf && !_eglIsSurfaceLinked(dsurf))
-	 dri2_destroy_surface(drv, disp, dsurf);
-      if (rsurf && rsurf != dsurf && !_eglIsSurfaceLinked(dsurf))
-	 dri2_destroy_surface(drv, disp, rsurf);
-      if (ctx != NULL && !_eglIsContextLinked(ctx))
-	 dri2_dpy->core->unbindContext(dri2_egl_context(ctx)->dri_context);
+      dri2_destroy_surface(drv, disp, old_dsurf);
+      dri2_destroy_surface(drv, disp, old_rsurf);
+      if (old_ctx) {
+	 dri2_dpy->core->unbindContext(dri2_egl_context(old_ctx)->dri_context);
+         /* no destroy? */
+         _eglPutContext(old_ctx);
+      }
 
       return EGL_TRUE;
    } else {
-      _eglBindContext(&ctx, &dsurf, &rsurf);
+      /* undo the previous _eglBindContext */
+      _eglBindContext(old_ctx, old_dsurf, old_rsurf, &ctx, &dsurf, &rsurf);
+      assert(&dri2_ctx->base == ctx &&
+             &dri2_dsurf->base == dsurf &&
+             &dri2_rsurf->base == rsurf);
+
+      _eglPutSurface(dsurf);
+      _eglPutSurface(rsurf);
+      _eglPutContext(ctx);
+
+      _eglPutSurface(old_dsurf);
+      _eglPutSurface(old_rsurf);
+      _eglPutContext(old_ctx);
 
       return EGL_FALSE;
    }
@@ -1221,6 +1250,8 @@ dri2_create_surface(_EGLDriver *drv, _EGLDisplay *disp, EGLint type,
    xcb_screen_iterator_t s;
    xcb_generic_error_t *error;
 
+   (void) drv;
+
    dri2_surf = malloc(sizeof *dri2_surf);
    if (!dri2_surf) {
       _eglError(EGL_BAD_ALLOC, "dri2_create_surface");
@@ -1234,8 +1265,7 @@ dri2_create_surface(_EGLDriver *drv, _EGLDisplay *disp, EGLint type,
    if (type == EGL_PBUFFER_BIT) {
       dri2_surf->drawable = xcb_generate_id(dri2_dpy->conn);
       s = xcb_setup_roots_iterator(xcb_get_setup(dri2_dpy->conn));
-      xcb_create_pixmap(dri2_dpy->conn,
-			_eglGetConfigKey(conf, EGL_BUFFER_SIZE),
+      xcb_create_pixmap(dri2_dpy->conn, conf->BufferSize,
 			dri2_surf->drawable, s.data->root,
 			dri2_surf->base.Width, dri2_surf->base.Height);
    } else {
@@ -1369,7 +1399,7 @@ dri2_swap_buffers_region(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *draw,
    xcb_rectangle_t rectangles[16];
    int i;
 
-   if (numRects > ARRAY_SIZE(rectangles))
+   if (numRects > (int)ARRAY_SIZE(rectangles))
       return dri2_copy_region(drv, disp, draw, dri2_surf->region);
 
    /* FIXME: Invert y here? */
@@ -1394,6 +1424,8 @@ dri2_swap_buffers_region(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *draw,
 static _EGLProc
 dri2_get_proc_address(_EGLDriver *drv, const char *procname)
 {
+   (void) drv;
+
    /* FIXME: Do we need to support lookup of EGL symbols too? */
 
    return (_EGLProc) _glapi_get_proc_address(procname);
@@ -1405,6 +1437,8 @@ dri2_wait_client(_EGLDriver *drv, _EGLDisplay *disp, _EGLContext *ctx)
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
    struct dri2_egl_surface *dri2_surf = dri2_egl_surface(ctx->DrawSurface);
 
+   (void) drv;
+
    /* FIXME: If EGL allows frontbuffer rendering for window surfaces,
     * we need to copy fake to real here.*/
 
@@ -1416,6 +1450,9 @@ dri2_wait_client(_EGLDriver *drv, _EGLDisplay *disp, _EGLContext *ctx)
 static EGLBoolean
 dri2_wait_native(_EGLDriver *drv, _EGLDisplay *disp, EGLint engine)
 {
+   (void) drv;
+   (void) disp;
+
    if (engine != EGL_CORE_NATIVE_ENGINE)
       return _eglError(EGL_BAD_PARAMETER, "eglWaitNative");
    /* glXWaitX(); */
@@ -1438,6 +1475,8 @@ dri2_copy_buffers(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *surf,
    struct dri2_egl_surface *dri2_surf = dri2_egl_surface(surf);
    xcb_gcontext_t gc;
 
+   (void) drv;
+
    (*dri2_dpy->flush->flush)(dri2_surf->dri_drawable);
 
    gc = xcb_generate_id(dri2_dpy->conn);
@@ -1501,6 +1540,11 @@ static EGLBoolean
 dri2_release_tex_image(_EGLDriver *drv,
 		       _EGLDisplay *disp, _EGLSurface *surf, EGLint buffer)
 {
+   (void) drv;
+   (void) disp;
+   (void) surf;
+   (void) buffer;
+
    return EGL_TRUE;
 }
 
@@ -1509,7 +1553,6 @@ dri2_create_image_khr_pixmap(_EGLDisplay *disp, _EGLContext *ctx,
 			     EGLClientBuffer buffer, const EGLint *attr_list)
 {
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
-   struct dri2_egl_context *dri2_ctx = dri2_egl_context(ctx);
    struct dri2_egl_image *dri2_img;
    unsigned int attachments[1];
    xcb_drawable_t drawable;
@@ -1521,6 +1564,8 @@ dri2_create_image_khr_pixmap(_EGLDisplay *disp, _EGLContext *ctx,
    xcb_generic_error_t *error;
    int stride, format;
 
+   (void) ctx;
+
    drawable = (xcb_drawable_t) buffer;
    xcb_dri2_create_drawable (dri2_dpy->conn, drawable);
    attachments[0] = XCB_DRI2_ATTACHMENT_BUFFER_FRONT_LEFT;
@@ -1569,7 +1614,7 @@ dri2_create_image_khr_pixmap(_EGLDisplay *disp, _EGLContext *ctx,
       return EGL_NO_IMAGE_KHR;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list)) {
+   if (!_eglInitImage(&dri2_img->base, disp)) {
       free(buffers_reply);
       free(geometry_reply);
       return EGL_NO_IMAGE_KHR;
@@ -1577,7 +1622,7 @@ dri2_create_image_khr_pixmap(_EGLDisplay *disp, _EGLContext *ctx,
 
    stride = buffers[0].pitch / buffers[0].cpp;
    dri2_img->dri_image =
-      dri2_dpy->image->createImageFromName(dri2_ctx->dri_context,
+      dri2_dpy->image->createImageFromName(dri2_dpy->dri_screen,
 					   buffers_reply->width,
 					   buffers_reply->height,
 					   format,
@@ -1612,7 +1657,7 @@ dri2_create_image_khr_renderbuffer(_EGLDisplay *disp, _EGLContext *ctx,
       return EGL_NO_IMAGE_KHR;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list))
+   if (!_eglInitImage(&dri2_img->base, disp))
       return EGL_NO_IMAGE_KHR;
 
    dri2_img->dri_image = 
@@ -1628,56 +1673,29 @@ dri2_create_image_mesa_drm_buffer(_EGLDisplay *disp, _EGLContext *ctx,
 				  EGLClientBuffer buffer, const EGLint *attr_list)
 {
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
-   struct dri2_egl_context *dri2_ctx = dri2_egl_context(ctx);
    struct dri2_egl_image *dri2_img;
-   EGLint width, height, format, name, stride, pitch, i, err;
+   EGLint format, name, pitch, err;
+   _EGLImageAttribs attrs;
+
+   (void) ctx;
 
    name = (EGLint) buffer;
 
-   err = EGL_SUCCESS;
-   width = 0;
-   height = 0;
-   format = 0;
-   stride = 0;
+   err = _eglParseImageAttribList(&attrs, disp, attr_list);
+   if (err != EGL_SUCCESS)
+      return NULL;
 
-   for (i = 0; attr_list[i] != EGL_NONE; i++) {
-      EGLint attr = attr_list[i++];
-      EGLint val = attr_list[i];
-
-      switch (attr) {
-      case EGL_WIDTH:
-	 width = val;
-         break;
-      case EGL_HEIGHT:
-	 height = val;
-         break;
-      case EGL_DRM_BUFFER_FORMAT_MESA:
-	 format = val;
-         break;
-      case EGL_DRM_BUFFER_STRIDE_MESA:
-	 stride = val;
-         break;
-      default:
-         err = EGL_BAD_ATTRIBUTE;
-         break;
-      }
-
-      if (err != EGL_SUCCESS) {
-         _eglLog(_EGL_WARNING, "bad image attribute 0x%04x", attr);
-	 return NULL;
-      }
-   }
-
-   if (width <= 0 || height <= 0 || stride <= 0) {
+   if (attrs.Width <= 0 || attrs.Height <= 0 ||
+       attrs.DRMBufferStrideMESA <= 0) {
       _eglError(EGL_BAD_PARAMETER,
 		"bad width, height or stride");
       return NULL;
    }
 
-   switch (format) {
+   switch (attrs.DRMBufferFormatMESA) {
    case EGL_DRM_BUFFER_FORMAT_ARGB32_MESA:
       format = __DRI_IMAGE_FORMAT_ARGB8888;
-      pitch = stride;
+      pitch = attrs.DRMBufferStrideMESA;
       break;
    default:
       _eglError(EGL_BAD_PARAMETER,
@@ -1691,15 +1709,15 @@ dri2_create_image_mesa_drm_buffer(_EGLDisplay *disp, _EGLContext *ctx,
       return NULL;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list)) {
+   if (!_eglInitImage(&dri2_img->base, disp)) {
       free(dri2_img);
       return NULL;
    }
 
    dri2_img->dri_image =
-      dri2_dpy->image->createImageFromName(dri2_ctx->dri_context,
-					   width,
-					   height,
+      dri2_dpy->image->createImageFromName(dri2_dpy->dri_screen,
+					   attrs.Width,
+					   attrs.Height,
 					   format,
 					   name,
 					   pitch,
@@ -1718,6 +1736,8 @@ dri2_create_image_khr(_EGLDriver *drv, _EGLDisplay *disp,
 		      _EGLContext *ctx, EGLenum target,
 		      EGLClientBuffer buffer, const EGLint *attr_list)
 {
+   (void) drv;
+
    switch (target) {
    case EGL_NATIVE_PIXMAP_KHR:
       return dri2_create_image_khr_pixmap(disp, ctx, buffer, attr_list);
@@ -1737,6 +1757,8 @@ dri2_destroy_image_khr(_EGLDriver *drv, _EGLDisplay *disp, _EGLImage *image)
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
    struct dri2_egl_image *dri2_img = dri2_egl_image(image);
 
+   (void) drv;
+
    dri2_dpy->image->destroyImage(dri2_img->dri_image);
    free(dri2_img);
 
@@ -1749,10 +1771,13 @@ dri2_create_drm_image_mesa(_EGLDriver *drv, _EGLDisplay *disp,
 {
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
    struct dri2_egl_image *dri2_img;
-   int width, height, format, i;
-   unsigned int use, dri_use, valid_mask;
+   _EGLImageAttribs attrs;
+   unsigned int dri_use, valid_mask;
+   int format;
    EGLint err = EGL_SUCCESS;
 
+   (void) drv;
+
    dri2_img = malloc(sizeof *dri2_img);
    if (!dri2_img) {
       _eglError(EGL_BAD_ALLOC, "dri2_create_image_khr");
@@ -1764,74 +1789,50 @@ dri2_create_drm_image_mesa(_EGLDriver *drv, _EGLDisplay *disp,
       goto cleanup_img;
    }
 
-   if (!_eglInitImage(&dri2_img->base, disp, attr_list)) {
+   if (!_eglInitImage(&dri2_img->base, disp)) {
       err = EGL_BAD_PARAMETER;
       goto cleanup_img;
    }
 
-   width = 0;
-   height = 0;
-   format = 0;
-   use = 0;
-   for (i = 0; attr_list[i] != EGL_NONE; i++) {
-      EGLint attr = attr_list[i++];
-      EGLint val = attr_list[i];
+   err = _eglParseImageAttribList(&attrs, disp, attr_list);
+   if (err != EGL_SUCCESS)
+      goto cleanup_img;
 
-      switch (attr) {
-      case EGL_WIDTH:
-	 width = val;
-         break;
-      case EGL_HEIGHT:
-	 height = val;
-         break;
-      case EGL_DRM_BUFFER_FORMAT_MESA:
-	 format = val;
-         break;
-      case EGL_DRM_BUFFER_USE_MESA:
-	 use = val;
-         break;
-      default:
-         err = EGL_BAD_ATTRIBUTE;
-         break;
-      }
-
-      if (err != EGL_SUCCESS) {
-         _eglLog(_EGL_WARNING, "bad image attribute 0x%04x", attr);
-	 goto cleanup_img;
-      }
-   }
-
-   if (width <= 0 || height <= 0) {
-      _eglLog(_EGL_WARNING, "bad width or height (%dx%d)", width, height);
+   if (attrs.Width <= 0 || attrs.Height <= 0) {
+      _eglLog(_EGL_WARNING, "bad width or height (%dx%d)",
+            attrs.Width, attrs.Height);
       goto cleanup_img;
    }
 
-   switch (format) {
+   switch (attrs.DRMBufferFormatMESA) {
    case EGL_DRM_BUFFER_FORMAT_ARGB32_MESA:
       format = __DRI_IMAGE_FORMAT_ARGB8888;
       break;
    default:
-      _eglLog(_EGL_WARNING, "bad image format value 0x%04x", format);
+      _eglLog(_EGL_WARNING, "bad image format value 0x%04x",
+            attrs.DRMBufferFormatMESA);
       goto cleanup_img;
    }
 
    valid_mask =
       EGL_DRM_BUFFER_USE_SCANOUT_MESA |
       EGL_DRM_BUFFER_USE_SHARE_MESA; 
-   if (use & ~valid_mask) {
-      _eglLog(_EGL_WARNING, "bad image use bit 0x%04x", use & ~valid_mask);
+   if (attrs.DRMBufferUseMESA & ~valid_mask) {
+      _eglLog(_EGL_WARNING, "bad image use bit 0x%04x",
+            attrs.DRMBufferUseMESA & ~valid_mask);
       goto cleanup_img;
    }
 
    dri_use = 0;
-   if (use & EGL_DRM_BUFFER_USE_SHARE_MESA)
+   if (attrs.DRMBufferUseMESA & EGL_DRM_BUFFER_USE_SHARE_MESA)
       dri_use |= __DRI_IMAGE_USE_SHARE;
-   if (use & EGL_DRM_BUFFER_USE_SCANOUT_MESA)
+   if (attrs.DRMBufferUseMESA & EGL_DRM_BUFFER_USE_SCANOUT_MESA)
       dri_use |= __DRI_IMAGE_USE_SCANOUT;
 
    dri2_img->dri_image = 
       dri2_dpy->image->createImage(dri2_dpy->dri_screen,
-				   width, height, format, dri_use, dri2_img);
+				   attrs.Width, attrs.Height,
+                                   format, dri_use, dri2_img);
    if (dri2_img->dri_image == NULL) {
       err = EGL_BAD_ALLOC;
       goto cleanup_img;
@@ -1853,6 +1854,8 @@ dri2_export_drm_image_mesa(_EGLDriver *drv, _EGLDisplay *disp, _EGLImage *img,
    struct dri2_egl_display *dri2_dpy = dri2_egl_display(disp);
    struct dri2_egl_image *dri2_img = dri2_egl_image(img);
 
+   (void) drv;
+
    if (name && !dri2_dpy->image->queryImage(dri2_img->dri_image,
 					    __DRI_IMAGE_ATTRIB_NAME, name)) {
       _eglError(EGL_BAD_ALLOC, "dri2_export_drm_image_mesa");
@@ -1879,6 +1882,8 @@ _eglMain(const char *args)
 {
    struct dri2_egl_driver *dri2_drv;
 
+   (void) args;
+
    dri2_drv = malloc(sizeof *dri2_drv);
    if (!dri2_drv)
       return NULL;

src/egl/drivers/glx/egl_glx.c

@@ -132,29 +132,38 @@ static const struct {
    int egl_attr;
 } fbconfig_attributes[] = {
    /* table 3.1 of GLX 1.4 */
-   { GLX_BUFFER_SIZE,			EGL_BUFFER_SIZE },
-   { GLX_LEVEL,				EGL_LEVEL },
-   { GLX_RED_SIZE,			EGL_RED_SIZE },
-   { GLX_GREEN_SIZE,			EGL_GREEN_SIZE },
-   { GLX_BLUE_SIZE,			EGL_BLUE_SIZE },
-   { GLX_ALPHA_SIZE,			EGL_ALPHA_SIZE },
-   { GLX_DEPTH_SIZE,			EGL_DEPTH_SIZE },
-   { GLX_STENCIL_SIZE,			EGL_STENCIL_SIZE },
-   { GLX_SAMPLE_BUFFERS,		EGL_SAMPLE_BUFFERS },
-   { GLX_SAMPLES,			EGL_SAMPLES },
-   { GLX_RENDER_TYPE,			EGL_RENDERABLE_TYPE },
-   { GLX_X_RENDERABLE,			EGL_NATIVE_RENDERABLE },
-   { GLX_X_VISUAL_TYPE,			EGL_NATIVE_VISUAL_TYPE },
-   { GLX_CONFIG_CAVEAT,			EGL_CONFIG_CAVEAT },
-   { GLX_TRANSPARENT_TYPE,		EGL_TRANSPARENT_TYPE },
-   { GLX_TRANSPARENT_RED_VALUE,		EGL_TRANSPARENT_RED_VALUE },
-   { GLX_TRANSPARENT_GREEN_VALUE,	EGL_TRANSPARENT_GREEN_VALUE },
-   { GLX_TRANSPARENT_BLUE_VALUE,	EGL_TRANSPARENT_BLUE_VALUE },
-   { GLX_MAX_PBUFFER_WIDTH,		EGL_MAX_PBUFFER_WIDTH },
-   { GLX_MAX_PBUFFER_HEIGHT,		EGL_MAX_PBUFFER_HEIGHT },
-   { GLX_MAX_PBUFFER_PIXELS,		EGL_MAX_PBUFFER_PIXELS },
-   { GLX_VISUAL_ID,			EGL_NATIVE_VISUAL_ID },
-   { GLX_X_VISUAL_TYPE,			EGL_NATIVE_VISUAL_TYPE },
+   { GLX_FBCONFIG_ID,                  0 },
+   { GLX_BUFFER_SIZE,                  EGL_BUFFER_SIZE },
+   { GLX_LEVEL,                        EGL_LEVEL },
+   { GLX_DOUBLEBUFFER,                 0 },
+   { GLX_STEREO,                       0 },
+   { GLX_AUX_BUFFERS,                  0 },
+   { GLX_RED_SIZE,                     EGL_RED_SIZE },
+   { GLX_GREEN_SIZE,                   EGL_GREEN_SIZE },
+   { GLX_BLUE_SIZE,                    EGL_BLUE_SIZE },
+   { GLX_ALPHA_SIZE,                   EGL_ALPHA_SIZE },
+   { GLX_DEPTH_SIZE,                   EGL_DEPTH_SIZE },
+   { GLX_STENCIL_SIZE,                 EGL_STENCIL_SIZE },
+   { GLX_ACCUM_RED_SIZE,               0 },
+   { GLX_ACCUM_GREEN_SIZE,             0 },
+   { GLX_ACCUM_BLUE_SIZE,              0 },
+   { GLX_ACCUM_ALPHA_SIZE,             0 },
+   { GLX_SAMPLE_BUFFERS,               EGL_SAMPLE_BUFFERS },
+   { GLX_SAMPLES,                      EGL_SAMPLES },
+   { GLX_RENDER_TYPE,                  0 },
+   { GLX_DRAWABLE_TYPE,                EGL_SURFACE_TYPE },
+   { GLX_X_RENDERABLE,                 EGL_NATIVE_RENDERABLE },
+   { GLX_X_VISUAL_TYPE,                EGL_NATIVE_VISUAL_TYPE },
+   { GLX_CONFIG_CAVEAT,                EGL_CONFIG_CAVEAT },
+   { GLX_TRANSPARENT_TYPE,             EGL_TRANSPARENT_TYPE },
+   { GLX_TRANSPARENT_INDEX_VALUE,      0 },
+   { GLX_TRANSPARENT_RED_VALUE,        EGL_TRANSPARENT_RED_VALUE },
+   { GLX_TRANSPARENT_GREEN_VALUE,      EGL_TRANSPARENT_GREEN_VALUE },
+   { GLX_TRANSPARENT_BLUE_VALUE,       EGL_TRANSPARENT_BLUE_VALUE },
+   { GLX_MAX_PBUFFER_WIDTH,            EGL_MAX_PBUFFER_WIDTH },
+   { GLX_MAX_PBUFFER_HEIGHT,           EGL_MAX_PBUFFER_HEIGHT },
+   { GLX_MAX_PBUFFER_PIXELS,           EGL_MAX_PBUFFER_PIXELS },
+   { GLX_VISUAL_ID,                    EGL_NATIVE_VISUAL_ID }
 };
 
 
@@ -162,12 +171,31 @@ static EGLBoolean
 convert_fbconfig(Display *dpy, GLXFBConfig fbconfig,
                  struct GLX_egl_config *GLX_conf)
 {
-   int err = 0, attr, egl_attr, val, i;
-   EGLint conformant, config_caveat, surface_type;
+   int err, attr, val;
+   unsigned i;
+
+   /* must have rgba bit */
+   err = glXGetFBConfigAttrib(dpy, fbconfig, GLX_RENDER_TYPE, &val);
+   if (err || !(val & GLX_RGBA_BIT))
+      return EGL_FALSE;
+
+   /* must know whether it is double-buffered */
+   err = glXGetFBConfigAttrib(dpy, fbconfig, GLX_DOUBLEBUFFER, &val);
+   if (err)
+      return EGL_FALSE;
+   GLX_conf->double_buffered = val;
+
+   GLX_conf->Base.RenderableType = EGL_OPENGL_BIT;
+   GLX_conf->Base.Conformant = EGL_OPENGL_BIT;
 
    for (i = 0; i < ARRAY_SIZE(fbconfig_attributes); i++) {
+      EGLint egl_attr, egl_val;
+
       attr = fbconfig_attributes[i].attr;
       egl_attr = fbconfig_attributes[i].egl_attr;
+      if (!egl_attr)
+         continue;
+
       err = glXGetFBConfigAttrib(dpy, fbconfig, attr, &val);
       if (err) {
          if (err == GLX_BAD_ATTRIBUTE) {
@@ -177,47 +205,71 @@ convert_fbconfig(Display *dpy, GLXFBConfig fbconfig,
          break;
       }
 
-      _eglSetConfigKey(&GLX_conf->Base, egl_attr, val);
+      switch (egl_attr) {
+      case EGL_SURFACE_TYPE:
+         egl_val = 0;
+         if (val & GLX_WINDOW_BIT)
+            egl_val |= EGL_WINDOW_BIT;
+         /* pixmap and pbuffer surfaces must be single-buffered in EGL */
+         if (!GLX_conf->double_buffered) {
+            if (val & GLX_PIXMAP_BIT)
+               egl_val |= EGL_PIXMAP_BIT;
+            if (val & GLX_PBUFFER_BIT)
+               egl_val |= EGL_PBUFFER_BIT;
+         }
+         break;
+      case EGL_NATIVE_VISUAL_TYPE:
+         switch (val) {
+         case GLX_TRUE_COLOR:
+            egl_val = TrueColor;
+            break;
+         case GLX_DIRECT_COLOR:
+            egl_val = DirectColor;
+            break;
+         case GLX_PSEUDO_COLOR:
+            egl_val = PseudoColor;
+            break;
+         case GLX_STATIC_COLOR:
+            egl_val = StaticColor;
+            break;
+         case GLX_GRAY_SCALE:
+            egl_val = GrayScale;
+            break;
+         case GLX_STATIC_GRAY:
+            egl_val = StaticGray;
+            break;
+         default:
+            egl_val = EGL_NONE;
+            break;
+         }
+         break;
+      case EGL_CONFIG_CAVEAT:
+         egl_val = EGL_NONE;
+         if (val == GLX_SLOW_CONFIG) {
+            egl_val = EGL_SLOW_CONFIG;
+         }
+         else if (val == GLX_NON_CONFORMANT_CONFIG) {
+            GLX_conf->Base.Conformant &= ~EGL_OPENGL_BIT;
+            egl_val = EGL_NONE;
+         }
+         break;
+      case EGL_TRANSPARENT_TYPE:
+         egl_val = (val == GLX_TRANSPARENT_RGB) ?
+            EGL_TRANSPARENT_RGB : EGL_NONE;
+         break;
+      default:
+         egl_val = val;
+         break;
+      }
+
+      _eglSetConfigKey(&GLX_conf->Base, egl_attr, egl_val);
    }
    if (err)
       return EGL_FALSE;
 
-   /* must have rgba bit */
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_RENDER_TYPE, &val);
-   if (!(val & GLX_RGBA_BIT))
+   if (!GLX_conf->Base.SurfaceType)
       return EGL_FALSE;
 
-   conformant = EGL_OPENGL_BIT;
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_CONFIG_CAVEAT, &val);
-   if (val == GLX_SLOW_CONFIG)
-      config_caveat = EGL_SLOW_CONFIG;
-   if (val == GLX_NON_CONFORMANT_CONFIG)
-      conformant &= ~EGL_OPENGL_BIT;
-   if (!(conformant & EGL_OPENGL_ES_BIT))
-      config_caveat = EGL_NON_CONFORMANT_CONFIG;
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_CONFIG_CAVEAT, config_caveat);
-
-   surface_type = 0;
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_DRAWABLE_TYPE, &val);
-   if (val & GLX_WINDOW_BIT)
-      surface_type |= EGL_WINDOW_BIT;
-   if (val & GLX_PIXMAP_BIT)
-      surface_type |= EGL_PIXMAP_BIT;
-   if (val & GLX_PBUFFER_BIT)
-      surface_type |= EGL_PBUFFER_BIT;
-
-   /* pixmap and pbuffer surfaces must be single-buffered in EGL */
-   glXGetFBConfigAttrib(dpy, fbconfig, GLX_DOUBLEBUFFER, &val);
-   GLX_conf->double_buffered = val;
-   if (GLX_conf->double_buffered) {
-      surface_type &= ~(EGL_PIXMAP_BIT | EGL_PBUFFER_BIT);
-      if (!surface_type)
-         return EGL_FALSE;
-   }
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_SURFACE_TYPE, surface_type);
-
    return EGL_TRUE;
 }
 
@@ -226,34 +278,69 @@ static const struct {
    int egl_attr;
 } visual_attributes[] = {
    /* table 3.7 of GLX 1.4 */
-   /* no GLX_USE_GL */
-   { GLX_BUFFER_SIZE,		EGL_BUFFER_SIZE },
-   { GLX_LEVEL,			EGL_LEVEL },
-   { GLX_RED_SIZE,		EGL_RED_SIZE },
-   { GLX_GREEN_SIZE,		EGL_GREEN_SIZE },
-   { GLX_BLUE_SIZE,		EGL_BLUE_SIZE },
-   { GLX_ALPHA_SIZE,		EGL_ALPHA_SIZE },
-   { GLX_DEPTH_SIZE,		EGL_DEPTH_SIZE },
-   { GLX_STENCIL_SIZE,		EGL_STENCIL_SIZE },
-   { GLX_SAMPLE_BUFFERS,	EGL_SAMPLE_BUFFERS },
-   { GLX_SAMPLES,		EGL_SAMPLES },
+   { GLX_USE_GL,              0 },
+   { GLX_BUFFER_SIZE,         EGL_BUFFER_SIZE },
+   { GLX_LEVEL,               EGL_LEVEL },
+   { GLX_RGBA,                0 },
+   { GLX_DOUBLEBUFFER,        0 },
+   { GLX_STEREO,              0 },
+   { GLX_AUX_BUFFERS,         0 },
+   { GLX_RED_SIZE,            EGL_RED_SIZE },
+   { GLX_GREEN_SIZE,          EGL_GREEN_SIZE },
+   { GLX_BLUE_SIZE,           EGL_BLUE_SIZE },
+   { GLX_ALPHA_SIZE,          EGL_ALPHA_SIZE },
+   { GLX_DEPTH_SIZE,          EGL_DEPTH_SIZE },
+   { GLX_STENCIL_SIZE,        EGL_STENCIL_SIZE },
+   { GLX_ACCUM_RED_SIZE,      0 },
+   { GLX_ACCUM_GREEN_SIZE,    0 },
+   { GLX_ACCUM_BLUE_SIZE,     0 },
+   { GLX_ACCUM_ALPHA_SIZE,    0 },
+   { GLX_SAMPLE_BUFFERS,      EGL_SAMPLE_BUFFERS },
+   { GLX_SAMPLES,             EGL_SAMPLES },
+   { GLX_FBCONFIG_ID,         0 },
+   /* GLX_EXT_visual_rating */
+   { GLX_VISUAL_CAVEAT_EXT,   EGL_CONFIG_CAVEAT }
 };
 
 static EGLBoolean
 convert_visual(Display *dpy, XVisualInfo *vinfo,
                struct GLX_egl_config *GLX_conf)
 {
-   int err, attr, egl_attr, val, i;
-   EGLint conformant, config_caveat, surface_type;
+   int err, attr, val;
+   unsigned i;
 
-   /* the visual must support OpenGL */
+   /* the visual must support OpenGL and RGBA buffer */
    err = glXGetConfig(dpy, vinfo, GLX_USE_GL, &val);
+   if (!err && val)
+      err = glXGetConfig(dpy, vinfo, GLX_RGBA, &val);
    if (err || !val)
       return EGL_FALSE;
 
+   /* must know whether it is double-buffered */
+   err = glXGetConfig(dpy, vinfo, GLX_DOUBLEBUFFER, &val);
+   if (err)
+      return EGL_FALSE;
+   GLX_conf->double_buffered = val;
+
+   GLX_conf->Base.RenderableType = EGL_OPENGL_BIT;
+   GLX_conf->Base.Conformant = EGL_OPENGL_BIT;
+   GLX_conf->Base.SurfaceType = EGL_WINDOW_BIT;
+   /* pixmap surfaces must be single-buffered in EGL */
+   if (!GLX_conf->double_buffered)
+      GLX_conf->Base.SurfaceType |= EGL_PIXMAP_BIT;
+
+   GLX_conf->Base.NativeVisualID = vinfo->visualid;
+   GLX_conf->Base.NativeVisualType = vinfo->class;
+   GLX_conf->Base.NativeRenderable = EGL_TRUE;
+
    for (i = 0; i < ARRAY_SIZE(visual_attributes); i++) {
+      EGLint egl_attr, egl_val;
+
       attr = visual_attributes[i].attr;
-      egl_attr = fbconfig_attributes[i].egl_attr;
+      egl_attr = visual_attributes[i].egl_attr;
+      if (!egl_attr)
+         continue;
+
       err = glXGetConfig(dpy, vinfo, attr, &val);
       if (err) {
          if (err == GLX_BAD_ATTRIBUTE) {
@@ -263,41 +350,26 @@ convert_visual(Display *dpy, XVisualInfo *vinfo,
          break;
       }
 
-      _eglSetConfigKey(&GLX_conf->Base, egl_attr, val);
+      switch (egl_attr) {
+      case EGL_CONFIG_CAVEAT:
+         egl_val = EGL_NONE;
+         if (val == GLX_SLOW_VISUAL_EXT) {
+            egl_val = EGL_SLOW_CONFIG;
+         }
+         else if (val == GLX_NON_CONFORMANT_VISUAL_EXT) {
+            GLX_conf->Base.Conformant &= ~EGL_OPENGL_BIT;
+            egl_val = EGL_NONE;
+         }
+         break;
+         break;
+      default:
+         egl_val = val;
+         break;
+      }
+      _eglSetConfigKey(&GLX_conf->Base, egl_attr, egl_val);
    }
-   if (err)
-      return EGL_FALSE;
 
-   glXGetConfig(dpy, vinfo, GLX_RGBA, &val);
-   if (!val)
-      return EGL_FALSE;
-
-   conformant = EGL_OPENGL_BIT;
-   glXGetConfig(dpy, vinfo, GLX_VISUAL_CAVEAT_EXT, &val);
-   if (val == GLX_SLOW_CONFIG)
-      config_caveat = EGL_SLOW_CONFIG;
-   if (val == GLX_NON_CONFORMANT_CONFIG)
-      conformant &= ~EGL_OPENGL_BIT;
-   if (!(conformant & EGL_OPENGL_ES_BIT))
-      config_caveat = EGL_NON_CONFORMANT_CONFIG;
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_CONFIG_CAVEAT, config_caveat);
-   _eglSetConfigKey(&GLX_conf->Base, EGL_NATIVE_VISUAL_ID, vinfo->visualid);
-   _eglSetConfigKey(&GLX_conf->Base, EGL_NATIVE_VISUAL_TYPE, vinfo->class);
-
-   /* pixmap and pbuffer surfaces must be single-buffered in EGL */
-   glXGetConfig(dpy, vinfo, GLX_DOUBLEBUFFER, &val);
-   GLX_conf->double_buffered = val;
-   surface_type = EGL_WINDOW_BIT;
-   /* pixmap surfaces must be single-buffered in EGL */
-   if (!GLX_conf->double_buffered)
-      surface_type |= EGL_PIXMAP_BIT;
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_SURFACE_TYPE, surface_type);
-
-   _eglSetConfigKey(&GLX_conf->Base, EGL_NATIVE_RENDERABLE, EGL_TRUE);
-
-   return EGL_TRUE;
+   return (err) ? EGL_FALSE : EGL_TRUE;
 }
 
 
@@ -305,30 +377,31 @@ static void
 fix_config(struct GLX_egl_display *GLX_dpy, struct GLX_egl_config *GLX_conf)
 {
    _EGLConfig *conf = &GLX_conf->Base;
-   EGLint surface_type, r, g, b, a;
 
-   surface_type = GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE);
    if (!GLX_conf->double_buffered && GLX_dpy->single_buffered_quirk) {
       /* some GLX impls do not like single-buffered window surface */
-      surface_type &= ~EGL_WINDOW_BIT;
+      conf->SurfaceType &= ~EGL_WINDOW_BIT;
       /* pbuffer bit is usually not set */
       if (GLX_dpy->have_pbuffer)
-         surface_type |= EGL_PBUFFER_BIT;
-      SET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE, surface_type);
+         conf->SurfaceType |= EGL_PBUFFER_BIT;
    }
 
    /* no visual attribs unless window bit is set */
-   if (!(surface_type & EGL_WINDOW_BIT)) {
-      SET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_ID, 0);
-      SET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_TYPE, EGL_NONE);
+   if (!(conf->SurfaceType & EGL_WINDOW_BIT)) {
+      conf->NativeVisualID = 0;
+      conf->NativeVisualType = EGL_NONE;
+   }
+
+   if (conf->TransparentType != EGL_TRANSPARENT_RGB) {
+      /* some impls set them to -1 (GLX_DONT_CARE) */
+      conf->TransparentRedValue = 0;
+      conf->TransparentGreenValue = 0;
+      conf->TransparentBlueValue = 0;
    }
 
    /* make sure buffer size is set correctly */
-   r = GET_CONFIG_ATTRIB(conf, EGL_RED_SIZE);
-   g = GET_CONFIG_ATTRIB(conf, EGL_GREEN_SIZE);
-   b = GET_CONFIG_ATTRIB(conf, EGL_BLUE_SIZE);
-   a = GET_CONFIG_ATTRIB(conf, EGL_ALPHA_SIZE);
-   SET_CONFIG_ATTRIB(conf, EGL_BUFFER_SIZE, r + g + b + a);
+   conf->BufferSize =
+      conf->RedSize + conf->GreenSize + conf->BlueSize + conf->AlphaSize;
 }
 
 
@@ -379,7 +452,7 @@ create_configs(_EGLDisplay *dpy, struct GLX_egl_display *GLX_dpy,
          memcpy(GLX_conf, &template, sizeof(template));
          GLX_conf->index = i;
 
-         _eglAddConfig(dpy, &GLX_conf->Base);
+         _eglLinkConfig(&GLX_conf->Base);
          id++;
       }
    }
@@ -457,6 +530,8 @@ GLX_eglInitialize(_EGLDriver *drv, _EGLDisplay *disp,
 {
    struct GLX_egl_display *GLX_dpy;
 
+   (void) drv;
+
    if (disp->Platform != _EGL_PLATFORM_X11)
       return EGL_FALSE;
 
@@ -541,6 +616,8 @@ GLX_eglCreateContext(_EGLDriver *drv, _EGLDisplay *disp, _EGLConfig *conf,
    struct GLX_egl_display *GLX_dpy = GLX_egl_display(disp);
    struct GLX_egl_context *GLX_ctx_shared = GLX_egl_context(share_list);
 
+   (void) drv;
+
    if (!GLX_ctx) {
       _eglError(EGL_BAD_ALLOC, "eglCreateContext");
       return NULL;
@@ -600,12 +677,16 @@ GLX_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
    struct GLX_egl_surface *GLX_dsurf = GLX_egl_surface(dsurf);
    struct GLX_egl_surface *GLX_rsurf = GLX_egl_surface(rsurf);
    struct GLX_egl_context *GLX_ctx = GLX_egl_context(ctx);
+   _EGLContext *old_ctx;
+   _EGLSurface *old_dsurf, *old_rsurf;
    GLXDrawable ddraw, rdraw;
    GLXContext cctx;
    EGLBoolean ret = EGL_FALSE;
 
-   /* bind the new context and return the "orphaned" one */
-   if (!_eglBindContext(&ctx, &dsurf, &rsurf))
+   (void) drv;
+
+   /* make new bindings */
+   if (!_eglBindContext(ctx, dsurf, rsurf, &old_ctx, &old_dsurf, &old_rsurf))
       return EGL_FALSE;
 
    ddraw = (GLX_dsurf) ? GLX_dsurf->glx_drawable : None;
@@ -618,13 +699,27 @@ GLX_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *dsurf,
       ret = glXMakeCurrent(GLX_dpy->dpy, ddraw, cctx);
 
    if (ret) {
-      if (dsurf && !_eglIsSurfaceLinked(dsurf))
-         destroy_surface(disp, dsurf);
-      if (rsurf && rsurf != dsurf && !_eglIsSurfaceLinked(rsurf))
-         destroy_surface(disp, rsurf);
+      if (_eglPutSurface(old_dsurf))
+         destroy_surface(disp, old_dsurf);
+      if (_eglPutSurface(old_rsurf))
+         destroy_surface(disp, old_rsurf);
+      /* no destroy? */
+      _eglPutContext(old_ctx);
    }
    else {
-      _eglBindContext(&ctx, &dsurf, &rsurf);
+      /* undo the previous _eglBindContext */
+      _eglBindContext(old_ctx, old_dsurf, old_rsurf, &ctx, &dsurf, &rsurf);
+      assert(&GLX_ctx->Base == ctx &&
+             &GLX_dsurf->Base == dsurf &&
+             &GLX_rsurf->Base == rsurf);
+
+      _eglPutSurface(dsurf);
+      _eglPutSurface(rsurf);
+      _eglPutContext(ctx);
+
+      _eglPutSurface(old_dsurf);
+      _eglPutSurface(old_rsurf);
+      _eglPutContext(old_ctx);
    }
 
    return ret;
@@ -656,6 +751,8 @@ GLX_eglCreateWindowSurface(_EGLDriver *drv, _EGLDisplay *disp,
    struct GLX_egl_surface *GLX_surf;
    uint width, height;
 
+   (void) drv;
+
    GLX_surf = CALLOC_STRUCT(GLX_egl_surface);
    if (!GLX_surf) {
       _eglError(EGL_BAD_ALLOC, "eglCreateWindowSurface");
@@ -702,6 +799,8 @@ GLX_eglCreatePixmapSurface(_EGLDriver *drv, _EGLDisplay *disp,
    struct GLX_egl_surface *GLX_surf;
    uint width, height;
 
+   (void) drv;
+
    GLX_surf = CALLOC_STRUCT(GLX_egl_surface);
    if (!GLX_surf) {
       _eglError(EGL_BAD_ALLOC, "eglCreatePixmapSurface");
@@ -762,6 +861,8 @@ GLX_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *disp,
    int attribs[5];
    int i;
 
+   (void) drv;
+
    GLX_surf = CALLOC_STRUCT(GLX_egl_surface);
    if (!GLX_surf) {
       _eglError(EGL_BAD_ALLOC, "eglCreatePbufferSurface");
@@ -820,7 +921,9 @@ GLX_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *disp,
 static EGLBoolean
 GLX_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *surf)
 {
-   if (!_eglIsSurfaceBound(surf))
+   (void) drv;
+
+   if (_eglPutSurface(surf))
       destroy_surface(disp, surf);
 
    return EGL_TRUE;
@@ -833,6 +936,8 @@ GLX_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *draw)
    struct GLX_egl_display *GLX_dpy = GLX_egl_display(disp);
    struct GLX_egl_surface *GLX_surf = GLX_egl_surface(draw);
 
+   (void) drv;
+
    glXSwapBuffers(GLX_dpy->dpy, GLX_surf->glx_drawable);
 
    return EGL_TRUE;
@@ -844,12 +949,18 @@ GLX_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *disp, _EGLSurface *draw)
 static _EGLProc
 GLX_eglGetProcAddress(_EGLDriver *drv, const char *procname)
 {
+   (void) drv;
+
    return (_EGLProc) glXGetProcAddress((const GLubyte *) procname);
 }
 
 static EGLBoolean
 GLX_eglWaitClient(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
 {
+   (void) drv;
+   (void) dpy;
+   (void) ctx;
+
    glXWaitGL();
    return EGL_TRUE;
 }
@@ -857,6 +968,9 @@ GLX_eglWaitClient(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
 static EGLBoolean
 GLX_eglWaitNative(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine)
 {
+   (void) drv;
+   (void) dpy;
+
    if (engine != EGL_CORE_NATIVE_ENGINE)
       return _eglError(EGL_BAD_PARAMETER, "eglWaitNative");
    glXWaitX();
@@ -880,6 +994,8 @@ _eglMain(const char *args)
 {
    struct GLX_egl_driver *GLX_drv = CALLOC_STRUCT(GLX_egl_driver);
 
+   (void) args;
+
    if (!GLX_drv)
       return NULL;
 

src/egl/main/Makefile

@@ -36,6 +36,7 @@ SOURCES = \
 	eglcurrent.c \
 	egldisplay.c \
 	egldriver.c \
+	eglfallbacks.c \
 	eglglobals.c \
 	eglimage.c \
 	egllog.c \
@@ -57,7 +58,7 @@ EGL_NATIVE_PLATFORM=_EGL_INVALID_PLATFORM
 ifeq ($(firstword $(EGL_PLATFORMS)),x11)
 EGL_NATIVE_PLATFORM=_EGL_PLATFORM_X11
 endif
-ifeq ($(firstword $(EGL_PLATFORMS)),kms)
+ifeq ($(firstword $(EGL_PLATFORMS)),drm)
 EGL_NATIVE_PLATFORM=_EGL_PLATFORM_DRM
 endif
 ifeq ($(firstword $(EGL_PLATFORMS)),fbdev)

src/egl/main/SConscript

@@ -4,48 +4,49 @@
 
 Import('*')
 
-if env['platform'] != 'winddk':
+env = env.Clone()
 
-	env = env.Clone()
+env.Append(CPPDEFINES = [
+    '_EGL_NATIVE_PLATFORM=_EGL_PLATFORM_WINDOWS',
+    '_EGL_DRIVER_SEARCH_DIR=\\"\\"',
+    '_EGL_OS_WINDOWS',
+    '_EGL_GET_CORE_ADDRESSES',
+    'KHRONOS_DLL_EXPORTS',
+])
 
-	env.Append(CPPDEFINES = [
-		'_EGL_NATIVE_PLATFORM=_EGL_PLATFORM_WINDOWS',
-		'_EGL_DRIVER_SEARCH_DIR=\\"\\"',
-		'_EGL_OS_WINDOWS',
-		'_EGL_GET_CORE_ADDRESSES',
-		'KHRONOS_DLL_EXPORTS',
-	])
+env.Append(CPPPATH = [
+    '#/include',
+])
 
-	env.Append(CPPPATH = [
-		'#/include',
-	])
+egl_sources = [
+    'eglapi.c',
+    'eglarray.c',
+    'eglconfig.c',
+    'eglcontext.c',
+    'eglcurrent.c',
+    'egldisplay.c',
+    'egldriver.c',
+    'eglfallbacks.c',
+    'eglglobals.c',
+    'eglimage.c',
+    'egllog.c',
+    'eglmisc.c',
+    'eglmode.c',
+    'eglscreen.c',
+    'eglstring.c',
+    'eglsurface.c',
+    'eglsync.c',
+]
 
-	egl_sources = [
-		'eglapi.c',
-		'eglarray.c',
-		'eglconfig.c',
-		'eglcontext.c',
-		'eglcurrent.c',
-		'egldisplay.c',
-		'egldriver.c',
-		'eglglobals.c',
-		'eglimage.c',
-		'egllog.c',
-		'eglmisc.c',
-		'eglmode.c',
-		'eglscreen.c',
-		'eglstring.c',
-		'eglsurface.c',
-		'eglsync.c',
-	]
+egl = env.SharedLibrary(
+    target = 'libEGL',
+    source = egl_sources + ['egl.def'],
+)
 
-	egl = env.SharedLibrary(
-		target = 'libEGL',
-		source = egl_sources + ['egl.def'],
-	)
+installed_egl = env.InstallSharedLibrary(egl, version=(1, 4, 0))
 
-	env.InstallSharedLibrary(egl, version=(1, 4, 0))
+env.Alias('egl', installed_egl)
 
-	egl = [env.FindIxes(egl, 'LIBPREFIX', 'LIBSUFFIX')]
+egl = [env.FindIxes(egl, 'LIBPREFIX', 'LIBSUFFIX')]
 
-	Export('egl')
+Export('egl')

src/egl/main/eglapi.c

@@ -402,16 +402,21 @@ eglCreateContext(EGLDisplay dpy, EGLConfig config, EGLContext share_list,
    _EGLContext *context;
    EGLContext ret;
 
-   if (config)
-      _EGL_CHECK_CONFIG(disp, conf, EGL_NO_CONTEXT, drv);
-   else
-      _EGL_CHECK_DISPLAY(disp, EGL_NO_CONTEXT, drv);
+   _EGL_CHECK_DISPLAY(disp, EGL_NO_CONTEXT, drv);
+
+   if (!config) {
+      /* config may be NULL if surfaceless */
+      if (!disp->Extensions.KHR_surfaceless_gles1 &&
+          !disp->Extensions.KHR_surfaceless_gles2 &&
+          !disp->Extensions.KHR_surfaceless_opengl)
+         RETURN_EGL_ERROR(disp, EGL_BAD_CONFIG, EGL_NO_CONTEXT);
+   }
 
    if (!share && share_list != EGL_NO_CONTEXT)
       RETURN_EGL_ERROR(disp, EGL_BAD_CONTEXT, EGL_NO_CONTEXT);
 
    context = drv->API.CreateContext(drv, disp, conf, share, attrib_list);
-   ret = (context) ? _eglLinkContext(context, disp) : EGL_NO_CONTEXT;
+   ret = (context) ? _eglLinkContext(context) : EGL_NO_CONTEXT;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -459,9 +464,19 @@ eglMakeCurrent(EGLDisplay dpy, EGLSurface draw, EGLSurface read,
 
    if (!context && ctx != EGL_NO_CONTEXT)
       RETURN_EGL_ERROR(disp, EGL_BAD_CONTEXT, EGL_FALSE);
-   if ((!draw_surf && draw != EGL_NO_SURFACE) ||
-       (!read_surf && read != EGL_NO_SURFACE))
-      RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
+   if (!draw_surf || !read_surf) {
+      /* surfaces may be NULL if surfaceless */
+      if (!disp->Extensions.KHR_surfaceless_gles1 &&
+          !disp->Extensions.KHR_surfaceless_gles2 &&
+          !disp->Extensions.KHR_surfaceless_opengl)
+         RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
+
+      if ((!draw_surf && draw != EGL_NO_SURFACE) ||
+          (!read_surf && read != EGL_NO_SURFACE))
+         RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
+      if (draw_surf || read_surf)
+         RETURN_EGL_ERROR(disp, EGL_BAD_MATCH, EGL_FALSE);
+   }
 
    ret = drv->API.MakeCurrent(drv, disp, draw_surf, read_surf, context);
 
@@ -500,7 +515,7 @@ eglCreateWindowSurface(EGLDisplay dpy, EGLConfig config,
       RETURN_EGL_ERROR(disp, EGL_BAD_NATIVE_WINDOW, EGL_NO_SURFACE);
 
    surf = drv->API.CreateWindowSurface(drv, disp, conf, window, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -521,7 +536,7 @@ eglCreatePixmapSurface(EGLDisplay dpy, EGLConfig config,
       RETURN_EGL_ERROR(disp, EGL_BAD_NATIVE_PIXMAP, EGL_NO_SURFACE);
 
    surf = drv->API.CreatePixmapSurface(drv, disp, conf, pixmap, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -540,7 +555,7 @@ eglCreatePbufferSurface(EGLDisplay dpy, EGLConfig config,
    _EGL_CHECK_CONFIG(disp, conf, EGL_NO_SURFACE, drv);
 
    surf = drv->API.CreatePbufferSurface(drv, disp, conf, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -633,11 +648,12 @@ eglSwapInterval(EGLDisplay dpy, EGLint interval)
 
    _EGL_CHECK_DISPLAY(disp, EGL_FALSE, drv);
 
-   if (!ctx || !_eglIsContextLinked(ctx) || ctx->Resource.Display != disp)
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       ctx->Resource.Display != disp)
       RETURN_EGL_ERROR(disp, EGL_BAD_CONTEXT, EGL_FALSE);
 
    surf = ctx->DrawSurface;
-   if (!_eglIsSurfaceLinked(surf))
+   if (_eglGetSurfaceHandle(surf) == EGL_NO_SURFACE)
       RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
 
    ret = drv->API.SwapInterval(drv, disp, surf, interval);
@@ -658,7 +674,8 @@ eglSwapBuffers(EGLDisplay dpy, EGLSurface surface)
    _EGL_CHECK_SURFACE(disp, surf, EGL_FALSE, drv);
 
    /* surface must be bound to current context in EGL 1.4 */
-   if (!ctx || !_eglIsContextLinked(ctx) || surf != ctx->DrawSurface)
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       surf != ctx->DrawSurface)
       RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
 
    ret = drv->API.SwapBuffers(drv, disp, surf);
@@ -699,7 +716,8 @@ eglWaitClient(void)
    _eglLockMutex(&disp->Mutex);
 
    /* let bad current context imply bad current surface */
-   if (!_eglIsContextLinked(ctx) || !_eglIsSurfaceLinked(ctx->DrawSurface))
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       _eglGetSurfaceHandle(ctx->DrawSurface) == EGL_NO_SURFACE)
       RETURN_EGL_ERROR(disp, EGL_BAD_CURRENT_SURFACE, EGL_FALSE);
 
    /* a valid current context implies an initialized current display */
@@ -748,7 +766,8 @@ eglWaitNative(EGLint engine)
    _eglLockMutex(&disp->Mutex);
 
    /* let bad current context imply bad current surface */
-   if (!_eglIsContextLinked(ctx) || !_eglIsSurfaceLinked(ctx->DrawSurface))
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       _eglGetSurfaceHandle(ctx->DrawSurface) == EGL_NO_SURFACE)
       RETURN_EGL_ERROR(disp, EGL_BAD_CURRENT_SURFACE, EGL_FALSE);
 
    /* a valid current context implies an initialized current display */
@@ -1028,7 +1047,7 @@ eglCreateScreenSurfaceMESA(EGLDisplay dpy, EGLConfig config,
    _EGL_CHECK_CONFIG(disp, conf, EGL_NO_SURFACE, drv);
 
    surf = drv->API.CreateScreenSurfaceMESA(drv, disp, conf, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1220,7 +1239,7 @@ eglCreatePbufferFromClientBuffer(EGLDisplay dpy, EGLenum buftype,
 
    surf = drv->API.CreatePbufferFromClientBuffer(drv, disp, buftype, buffer,
                                                  conf, attrib_list);
-   ret = (surf) ? _eglLinkSurface(surf, disp) : EGL_NO_SURFACE;
+   ret = (surf) ? _eglLinkSurface(surf) : EGL_NO_SURFACE;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1276,12 +1295,14 @@ eglCreateImageKHR(EGLDisplay dpy, EGLContext ctx, EGLenum target,
    EGLImageKHR ret;
 
    _EGL_CHECK_DISPLAY(disp, EGL_NO_IMAGE_KHR, drv);
+   if (!disp->Extensions.KHR_image_base)
+      RETURN_EGL_EVAL(disp, EGL_NO_IMAGE_KHR);
    if (!context && ctx != EGL_NO_CONTEXT)
       RETURN_EGL_ERROR(disp, EGL_BAD_CONTEXT, EGL_NO_IMAGE_KHR);
 
    img = drv->API.CreateImageKHR(drv,
          disp, context, target, buffer, attr_list);
-   ret = (img) ? _eglLinkImage(img, disp) : EGL_NO_IMAGE_KHR;
+   ret = (img) ? _eglLinkImage(img) : EGL_NO_IMAGE_KHR;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1296,6 +1317,8 @@ eglDestroyImageKHR(EGLDisplay dpy, EGLImageKHR image)
    EGLBoolean ret;
 
    _EGL_CHECK_DISPLAY(disp, EGL_FALSE, drv);
+   if (!disp->Extensions.KHR_image_base)
+      RETURN_EGL_EVAL(disp, EGL_FALSE);
    if (!img)
       RETURN_EGL_ERROR(disp, EGL_BAD_PARAMETER, EGL_FALSE);
 
@@ -1321,9 +1344,11 @@ eglCreateSyncKHR(EGLDisplay dpy, EGLenum type, const EGLint *attrib_list)
    EGLSyncKHR ret;
 
    _EGL_CHECK_DISPLAY(disp, EGL_NO_SYNC_KHR, drv);
+   if (!disp->Extensions.KHR_reusable_sync)
+      RETURN_EGL_EVAL(disp, EGL_NO_SYNC_KHR);
 
    sync = drv->API.CreateSyncKHR(drv, disp, type, attrib_list);
-   ret = (sync) ? _eglLinkSync(sync, disp) : EGL_NO_SYNC_KHR;
+   ret = (sync) ? _eglLinkSync(sync) : EGL_NO_SYNC_KHR;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1338,6 +1363,8 @@ eglDestroySyncKHR(EGLDisplay dpy, EGLSyncKHR sync)
    EGLBoolean ret;
 
    _EGL_CHECK_SYNC(disp, s, EGL_FALSE, drv);
+   assert(disp->Extensions.KHR_reusable_sync);
+
    _eglUnlinkSync(s);
    ret = drv->API.DestroySyncKHR(drv, disp, s);
 
@@ -1354,6 +1381,7 @@ eglClientWaitSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint flags, EGLTimeKHR t
    EGLint ret;
 
    _EGL_CHECK_SYNC(disp, s, EGL_FALSE, drv);
+   assert(disp->Extensions.KHR_reusable_sync);
    ret = drv->API.ClientWaitSyncKHR(drv, disp, s, flags, timeout);
 
    RETURN_EGL_EVAL(disp, ret);
@@ -1369,6 +1397,7 @@ eglSignalSyncKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLenum mode)
    EGLBoolean ret;
 
    _EGL_CHECK_SYNC(disp, s, EGL_FALSE, drv);
+   assert(disp->Extensions.KHR_reusable_sync);
    ret = drv->API.SignalSyncKHR(drv, disp, s, mode);
 
    RETURN_EGL_EVAL(disp, ret);
@@ -1384,6 +1413,7 @@ eglGetSyncAttribKHR(EGLDisplay dpy, EGLSyncKHR sync, EGLint attribute, EGLint *v
    EGLBoolean ret;
 
    _EGL_CHECK_SYNC(disp, s, EGL_FALSE, drv);
+   assert(disp->Extensions.KHR_reusable_sync);
    ret = drv->API.GetSyncAttribKHR(drv, disp, s, attribute, value);
 
    RETURN_EGL_EVAL(disp, ret);
@@ -1407,14 +1437,15 @@ eglSwapBuffersRegionNOK(EGLDisplay dpy, EGLSurface surface,
 
    _EGL_CHECK_SURFACE(disp, surf, EGL_FALSE, drv);
 
+   if (!disp->Extensions.NOK_swap_region)
+      RETURN_EGL_EVAL(disp, EGL_FALSE);
+
    /* surface must be bound to current context in EGL 1.4 */
-   if (!ctx || !_eglIsContextLinked(ctx) || surf != ctx->DrawSurface)
+   if (_eglGetContextHandle(ctx) == EGL_NO_CONTEXT ||
+       surf != ctx->DrawSurface)
       RETURN_EGL_ERROR(disp, EGL_BAD_SURFACE, EGL_FALSE);
 
-   if (drv->API.SwapBuffersRegionNOK)
-      ret = drv->API.SwapBuffersRegionNOK(drv, disp, surf, numRects, rects);
-   else
-      ret = drv->API.SwapBuffers(drv, disp, surf);
+   ret = drv->API.SwapBuffersRegionNOK(drv, disp, surf, numRects, rects);
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1433,9 +1464,11 @@ eglCreateDRMImageMESA(EGLDisplay dpy, const EGLint *attr_list)
    EGLImageKHR ret;
 
    _EGL_CHECK_DISPLAY(disp, EGL_NO_IMAGE_KHR, drv);
+   if (!disp->Extensions.MESA_drm_image)
+      RETURN_EGL_EVAL(disp, EGL_NO_IMAGE_KHR);
 
    img = drv->API.CreateDRMImageMESA(drv, disp, attr_list);
-   ret = (img) ? _eglLinkImage(img, disp) : EGL_NO_IMAGE_KHR;
+   ret = (img) ? _eglLinkImage(img) : EGL_NO_IMAGE_KHR;
 
    RETURN_EGL_EVAL(disp, ret);
 }
@@ -1450,6 +1483,8 @@ eglExportDRMImageMESA(EGLDisplay dpy, EGLImageKHR image,
    EGLBoolean ret;
 
    _EGL_CHECK_DISPLAY(disp, EGL_FALSE, drv);
+   assert(disp->Extensions.MESA_drm_image);
+
    if (!img)
       RETURN_EGL_ERROR(disp, EGL_BAD_PARAMETER, EGL_FALSE);
 

src/egl/main/eglarray.h

@@ -37,7 +37,7 @@ void *
 _eglFindArray(_EGLArray *array, void *elem);
 
 
-void **
+PUBLIC void **
 _eglFilterArray(_EGLArray *array, EGLint *size,
                 _EGLArrayForEach filter, void *filter_data);
 

src/egl/main/eglconfig.c

@@ -24,34 +24,34 @@
  * IDs are from 1 to N respectively.
  */
 void
-_eglInitConfig(_EGLConfig *config, _EGLDisplay *dpy, EGLint id)
+_eglInitConfig(_EGLConfig *conf, _EGLDisplay *dpy, EGLint id)
 {
-   memset(config, 0, sizeof(*config));
+   memset(conf, 0, sizeof(*conf));
 
-   config->Display = dpy;
+   conf->Display = dpy;
 
    /* some attributes take non-zero default values */
-   SET_CONFIG_ATTRIB(config, EGL_CONFIG_ID,               id);
-   SET_CONFIG_ATTRIB(config, EGL_CONFIG_CAVEAT,           EGL_NONE);
-   SET_CONFIG_ATTRIB(config, EGL_TRANSPARENT_TYPE,        EGL_NONE);
-   SET_CONFIG_ATTRIB(config, EGL_NATIVE_VISUAL_TYPE,      EGL_NONE);
-#ifdef EGL_VERSION_1_2
-   SET_CONFIG_ATTRIB(config, EGL_COLOR_BUFFER_TYPE,       EGL_RGB_BUFFER);
-#endif /* EGL_VERSION_1_2 */
+   conf->ConfigID = id;
+   conf->ConfigCaveat = EGL_NONE;
+   conf->TransparentType = EGL_NONE;
+   conf->NativeVisualType = EGL_NONE;
+   conf->ColorBufferType = EGL_RGB_BUFFER;
 }
 
 
 /**
- * Link a config to a display and return the handle of the link.
+ * Link a config to its display and return the handle of the link.
  * The handle can be passed to client directly.
  *
  * Note that we just save the ptr to the config (we don't copy the config).
  */
-EGLConfig
-_eglAddConfig(_EGLDisplay *dpy, _EGLConfig *conf)
+PUBLIC EGLConfig
+_eglLinkConfig(_EGLConfig *conf)
 {
+   _EGLDisplay *dpy = conf->Display;
+
    /* sanity check */
-   assert(GET_CONFIG_ATTRIB(conf, EGL_CONFIG_ID) > 0);
+   assert(dpy && conf->ConfigID > 0);
 
    if (!dpy->Configs) {
       dpy->Configs = _eglCreateArray("Config", 16);
@@ -59,23 +59,29 @@ _eglAddConfig(_EGLDisplay *dpy, _EGLConfig *conf)
          return (EGLConfig) NULL;
    }
 
-   conf->Display = dpy;
    _eglAppendArray(dpy->Configs, (void *) conf);
 
    return (EGLConfig) conf;
 }
 
 
-EGLBoolean
-_eglCheckConfigHandle(EGLConfig config, _EGLDisplay *dpy)
+/**
+ * Lookup a handle to find the linked config.
+ * Return NULL if the handle has no corresponding linked config.
+ */
+_EGLConfig *
+_eglLookupConfig(EGLConfig config, _EGLDisplay *dpy)
 {
    _EGLConfig *conf;
 
+   if (!dpy)
+      return NULL;
+
    conf = (_EGLConfig *) _eglFindArray(dpy->Configs, (void *) config);
    if (conf)
       assert(conf->Display == dpy);
 
-   return (conf != NULL);
+   return conf;
 }
 
 
@@ -104,6 +110,7 @@ static const struct {
    EGLint default_value;
 } _eglValidationTable[] =
 {
+   /* core */
    { EGL_BUFFER_SIZE,               ATTRIB_TYPE_INTEGER,
                                     ATTRIB_CRITERION_ATLEAST,
                                     0 },
@@ -200,22 +207,13 @@ static const struct {
    { EGL_TRANSPARENT_BLUE_VALUE,    ATTRIB_TYPE_INTEGER,
                                     ATTRIB_CRITERION_EXACT,
                                     EGL_DONT_CARE },
-   /* these are not real attributes */
    { EGL_MATCH_NATIVE_PIXMAP,       ATTRIB_TYPE_PSEUDO,
                                     ATTRIB_CRITERION_SPECIAL,
                                     EGL_NONE },
-   /* there is a gap before EGL_SAMPLES */
-   { 0x3030,                        ATTRIB_TYPE_PSEUDO,
-                                    ATTRIB_CRITERION_IGNORE,
-                                    0 },
-   { EGL_NONE,                      ATTRIB_TYPE_PSEUDO,
-                                    ATTRIB_CRITERION_IGNORE,
-                                    0 },
-
+   /* extensions */
    { EGL_Y_INVERTED_NOK,            ATTRIB_TYPE_BOOLEAN,
                                     ATTRIB_CRITERION_EXACT,
-                                    EGL_DONT_CARE },
-
+                                    EGL_DONT_CARE }
 };
 
 
@@ -232,18 +230,13 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
 {
    EGLint i, attr, val;
    EGLBoolean valid = EGL_TRUE;
-   EGLint red_size = 0, green_size = 0, blue_size = 0, luminance_size = 0;
-   EGLint alpha_size = 0, buffer_size = 0;
-
-   /* all attributes should have been listed */
-   assert(ARRAY_SIZE(_eglValidationTable) == _EGL_CONFIG_NUM_ATTRIBS);
 
    /* check attributes by their types */
    for (i = 0; i < ARRAY_SIZE(_eglValidationTable); i++) {
       EGLint mask;
 
       attr = _eglValidationTable[i].attr;
-      val = GET_CONFIG_ATTRIB(conf, attr);
+      val = _eglGetConfigKey(conf, attr);
 
       switch (_eglValidationTable[i].type) {
       case ATTRIB_TYPE_INTEGER:
@@ -255,30 +248,14 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
             break;
          case EGL_SAMPLE_BUFFERS:
             /* there can be at most 1 sample buffer */
-            if (val > 1)
+            if (val > 1 || val < 0)
                valid = EGL_FALSE;
             break;
-         case EGL_RED_SIZE:
-            red_size = val;
-            break;
-         case EGL_GREEN_SIZE:
-            green_size = val;
-            break;
-         case EGL_BLUE_SIZE:
-            blue_size = val;
-            break;
-         case EGL_LUMINANCE_SIZE:
-            luminance_size = val;
-            break;
-         case EGL_ALPHA_SIZE:
-            alpha_size = val;
-            break;
-         case EGL_BUFFER_SIZE:
-            buffer_size = val;
+         default:
+            if (val < 0)
+               valid = EGL_FALSE;
             break;
          }
-         if (val < 0)
-            valid = EGL_FALSE;
          break;
       case ATTRIB_TYPE_BOOLEAN:
          if (val != EGL_TRUE && val != EGL_FALSE)
@@ -366,17 +343,18 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
 
    /* now check for conflicting attribute values */
 
-   switch (GET_CONFIG_ATTRIB(conf, EGL_COLOR_BUFFER_TYPE)) {
+   switch (conf->ColorBufferType) {
    case EGL_RGB_BUFFER:
-      if (luminance_size)
+      if (conf->LuminanceSize)
          valid = EGL_FALSE;
-      if (red_size + green_size + blue_size + alpha_size != buffer_size)
+      if (conf->RedSize + conf->GreenSize +
+            conf->BlueSize + conf->AlphaSize != conf->BufferSize)
          valid = EGL_FALSE;
       break;
    case EGL_LUMINANCE_BUFFER:
-      if (red_size || green_size || blue_size)
+      if (conf->RedSize || conf->GreenSize || conf->BlueSize)
          valid = EGL_FALSE;
-      if (luminance_size + alpha_size != buffer_size)
+      if (conf->LuminanceSize + conf->AlphaSize != conf->BufferSize)
          valid = EGL_FALSE;
       break;
    }
@@ -385,23 +363,19 @@ _eglValidateConfig(const _EGLConfig *conf, EGLBoolean for_matching)
       return EGL_FALSE;
    }
 
-   val = GET_CONFIG_ATTRIB(conf, EGL_SAMPLE_BUFFERS);
-   if (!val && GET_CONFIG_ATTRIB(conf, EGL_SAMPLES))
+   if (!conf->SampleBuffers && conf->Samples)
       valid = EGL_FALSE;
    if (!valid) {
       _eglLog(_EGL_DEBUG, "conflicting samples and sample buffers");
       return EGL_FALSE;
    }
 
-   val = GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE);
-   if (!(val & EGL_WINDOW_BIT)) {
-      if (GET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_ID) != 0 ||
-          GET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_TYPE) != EGL_NONE)
+   if (!(conf->SurfaceType & EGL_WINDOW_BIT)) {
+      if (conf->NativeVisualID != 0 || conf->NativeVisualType != EGL_NONE)
          valid = EGL_FALSE;
    }
-   if (!(val & EGL_PBUFFER_BIT)) {
-      if (GET_CONFIG_ATTRIB(conf, EGL_BIND_TO_TEXTURE_RGB) ||
-          GET_CONFIG_ATTRIB(conf, EGL_BIND_TO_TEXTURE_RGBA))
+   if (!(conf->SurfaceType & EGL_PBUFFER_BIT)) {
+      if (conf->BindToTextureRGB || conf->BindToTextureRGBA)
          valid = EGL_FALSE;
    }
    if (!valid) {
@@ -433,11 +407,11 @@ _eglMatchConfig(const _EGLConfig *conf, const _EGLConfig *criteria)
          continue;
 
       attr = _eglValidationTable[i].attr;
-      cmp = GET_CONFIG_ATTRIB(criteria, attr);
+      cmp = _eglGetConfigKey(criteria, attr);
       if (cmp == EGL_DONT_CARE)
          continue;
 
-      val = GET_CONFIG_ATTRIB(conf, attr);
+      val = _eglGetConfigKey(conf, attr);
       switch (_eglValidationTable[i].criterion) {
       case ATTRIB_CRITERION_EXACT:
          if (val != cmp)
@@ -478,16 +452,11 @@ _eglMatchConfig(const _EGLConfig *conf, const _EGLConfig *criteria)
 static INLINE EGLBoolean
 _eglIsConfigAttribValid(_EGLConfig *conf, EGLint attr)
 {
-   if (_eglIndexConfig(conf, attr) < 0)
+   if (_eglOffsetOfConfig(attr) < 0)
       return EGL_FALSE;
 
-   /* there are some holes in the range */
    switch (attr) {
-   case 0x3030 /* a gap before EGL_SAMPLES */:
-   case EGL_NONE:
-#ifdef EGL_VERSION_1_4
    case EGL_MATCH_NATIVE_PIXMAP:
-#endif
       return EGL_FALSE;
    case EGL_Y_INVERTED_NOK:
       return conf->Display->Extensions.NOK_texture_from_pixmap;
@@ -503,18 +472,18 @@ _eglIsConfigAttribValid(_EGLConfig *conf, EGLint attr)
  * Return EGL_FALSE if any of the attribute is invalid.
  */
 EGLBoolean
-_eglParseConfigAttribList(_EGLConfig *conf, const EGLint *attrib_list)
+_eglParseConfigAttribList(_EGLConfig *conf, _EGLDisplay *dpy,
+                          const EGLint *attrib_list)
 {
    EGLint attr, val, i;
-   EGLint config_id = 0, level = 0;
-   EGLBoolean has_native_visual_type = EGL_FALSE;
-   EGLBoolean has_transparent_color = EGL_FALSE;
+
+   _eglInitConfig(conf, dpy, EGL_DONT_CARE);
 
    /* reset to default values */
    for (i = 0; i < ARRAY_SIZE(_eglValidationTable); i++) {
       attr = _eglValidationTable[i].attr;
       val = _eglValidationTable[i].default_value;
-      SET_CONFIG_ATTRIB(conf, attr, val);
+      _eglSetConfigKey(conf, attr, val);
    }
 
    /* parse the list */
@@ -524,59 +493,33 @@ _eglParseConfigAttribList(_EGLConfig *conf, const EGLint *attrib_list)
 
       if (!_eglIsConfigAttribValid(conf, attr))
 	 return EGL_FALSE;
-	      
-      SET_CONFIG_ATTRIB(conf, attr, val);
 
-      /* rememeber some attributes for post-processing */
-      switch (attr) {
-      case EGL_CONFIG_ID:
-         config_id = val;
-         break;
-      case EGL_LEVEL:
-         level = val;
-         break;
-      case EGL_NATIVE_VISUAL_TYPE:
-         has_native_visual_type = EGL_TRUE;
-         break;
-      case EGL_TRANSPARENT_RED_VALUE:
-      case EGL_TRANSPARENT_GREEN_VALUE:
-      case EGL_TRANSPARENT_BLUE_VALUE:
-         has_transparent_color = EGL_TRUE;
-         break;
-      default:
-         break;
-      }
+      _eglSetConfigKey(conf, attr, val);
    }
 
    if (!_eglValidateConfig(conf, EGL_TRUE))
       return EGL_FALSE;
 
    /* the spec says that EGL_LEVEL cannot be EGL_DONT_CARE */
-   if (level == EGL_DONT_CARE)
+   if (conf->Level == EGL_DONT_CARE)
       return EGL_FALSE;
 
    /* ignore other attributes when EGL_CONFIG_ID is given */
-   if (config_id > 0) {
-      _eglResetConfigKeys(conf, EGL_DONT_CARE);
-      SET_CONFIG_ATTRIB(conf, EGL_CONFIG_ID, config_id);
+   if (conf->ConfigID != EGL_DONT_CARE) {
+      for (i = 0; i < ARRAY_SIZE(_eglValidationTable); i++) {
+         attr = _eglValidationTable[i].attr;
+         if (attr != EGL_CONFIG_ID)
+            _eglSetConfigKey(conf, attr, EGL_DONT_CARE);
+      }
    }
    else {
-      if (has_native_visual_type) {
-         val = GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE);
-         if (!(val & EGL_WINDOW_BIT))
-            SET_CONFIG_ATTRIB(conf, EGL_NATIVE_VISUAL_TYPE, EGL_DONT_CARE);
-      }
+      if (!(conf->SurfaceType & EGL_WINDOW_BIT))
+         conf->NativeVisualType = EGL_DONT_CARE;
 
-      if (has_transparent_color) {
-         val = GET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_TYPE);
-         if (val == EGL_NONE) {
-            SET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_RED_VALUE,
-                              EGL_DONT_CARE);
-            SET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_GREEN_VALUE,
-                              EGL_DONT_CARE);
-            SET_CONFIG_ATTRIB(conf, EGL_TRANSPARENT_BLUE_VALUE,
-                              EGL_DONT_CARE);
-         }
+      if (conf->TransparentType == EGL_NONE) {
+         conf->TransparentRedValue = EGL_DONT_CARE;
+         conf->TransparentGreenValue = EGL_DONT_CARE;
+         conf->TransparentBlueValue = EGL_DONT_CARE;
       }
    }
 
@@ -610,7 +553,6 @@ _eglCompareConfigs(const _EGLConfig *conf1, const _EGLConfig *conf2,
       EGL_ALPHA_MASK_SIZE,
    };
    EGLint val1, val2;
-   EGLBoolean rgb_buffer;
    EGLint i;
 
    if (conf1 == conf2)
@@ -619,44 +561,41 @@ _eglCompareConfigs(const _EGLConfig *conf1, const _EGLConfig *conf2,
    /* the enum values have the desired ordering */
    assert(EGL_NONE < EGL_SLOW_CONFIG);
    assert(EGL_SLOW_CONFIG < EGL_NON_CONFORMANT_CONFIG);
-   val1 = GET_CONFIG_ATTRIB(conf1, EGL_CONFIG_CAVEAT);
-   val2 = GET_CONFIG_ATTRIB(conf2, EGL_CONFIG_CAVEAT);
-   if (val1 != val2)
-      return (val1 - val2);
+   val1 = conf1->ConfigCaveat - conf2->ConfigCaveat;
+   if (val1)
+      return val1;
 
    /* the enum values have the desired ordering */
    assert(EGL_RGB_BUFFER < EGL_LUMINANCE_BUFFER);
-   val1 = GET_CONFIG_ATTRIB(conf1, EGL_COLOR_BUFFER_TYPE);
-   val2 = GET_CONFIG_ATTRIB(conf2, EGL_COLOR_BUFFER_TYPE);
-   if (val1 != val2)
-      return (val1 - val2);
-   rgb_buffer = (val1 == EGL_RGB_BUFFER);
+   val1 = conf1->ColorBufferType - conf2->ColorBufferType;
+   if (val1)
+      return val1;
 
    if (criteria) {
       val1 = val2 = 0;
-      if (rgb_buffer) {
-         if (GET_CONFIG_ATTRIB(criteria, EGL_RED_SIZE) > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_RED_SIZE);
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_RED_SIZE);
+      if (conf1->ColorBufferType == EGL_RGB_BUFFER) {
+         if (criteria->RedSize > 0) {
+            val1 += conf1->RedSize;
+            val2 += conf2->RedSize;
          }
-         if (GET_CONFIG_ATTRIB(criteria, EGL_GREEN_SIZE) > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_GREEN_SIZE);
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_GREEN_SIZE);
+         if (criteria->GreenSize > 0) {
+            val1 += conf1->GreenSize;
+            val2 += conf2->GreenSize;
          }
-         if (GET_CONFIG_ATTRIB(criteria, EGL_BLUE_SIZE) > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_BLUE_SIZE);
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_BLUE_SIZE);
+         if (criteria->BlueSize > 0) {
+            val1 += conf1->BlueSize;
+            val2 += conf2->BlueSize;
          }
       }
       else {
-         if (GET_CONFIG_ATTRIB(criteria, EGL_LUMINANCE_SIZE) > 0) {
-            val1 += GET_CONFIG_ATTRIB(conf1, EGL_LUMINANCE_SIZE);
-            val2 += GET_CONFIG_ATTRIB(conf2, EGL_LUMINANCE_SIZE);
+         if (criteria->LuminanceSize > 0) {
+            val1 += conf1->LuminanceSize;
+            val2 += conf2->LuminanceSize;
          }
       }
-      if (GET_CONFIG_ATTRIB(criteria, EGL_ALPHA_SIZE) > 0) {
-         val1 += GET_CONFIG_ATTRIB(conf1, EGL_ALPHA_SIZE);
-         val2 += GET_CONFIG_ATTRIB(conf2, EGL_ALPHA_SIZE);
+      if (criteria->AlphaSize > 0) {
+         val1 += conf1->AlphaSize;
+         val2 += conf2->AlphaSize;
       }
    }
    else {
@@ -669,24 +608,15 @@ _eglCompareConfigs(const _EGLConfig *conf1, const _EGLConfig *conf2,
       return (val2 - val1);
 
    for (i = 0; i < ARRAY_SIZE(compare_attribs); i++) {
-      val1 = GET_CONFIG_ATTRIB(conf1, compare_attribs[i]);
-      val2 = GET_CONFIG_ATTRIB(conf2, compare_attribs[i]);
+      val1 = _eglGetConfigKey(conf1, compare_attribs[i]);
+      val2 = _eglGetConfigKey(conf2, compare_attribs[i]);
       if (val1 != val2)
          return (val1 - val2);
    }
 
    /* EGL_NATIVE_VISUAL_TYPE cannot be compared here */
 
-   if (compare_id) {
-      val1 = GET_CONFIG_ATTRIB(conf1, EGL_CONFIG_ID);
-      val2 = GET_CONFIG_ATTRIB(conf2, EGL_CONFIG_ID);
-      assert(val1 != val2);
-   }
-   else {
-      val1 = val2 = 0;
-   }
-
-   return (val1 - val2);
+   return (compare_id) ? (conf1->ConfigID - conf2->ConfigID) : 0;
 }
 
 
@@ -764,8 +694,7 @@ _eglChooseConfig(_EGLDriver *drv, _EGLDisplay *disp, const EGLint *attrib_list,
    if (!num_configs)
       return _eglError(EGL_BAD_PARAMETER, "eglChooseConfigs");
 
-   _eglInitConfig(&criteria, disp, 0);
-   if (!_eglParseConfigAttribList(&criteria, attrib_list))
+   if (!_eglParseConfigAttribList(&criteria, disp, attrib_list))
       return _eglError(EGL_BAD_ATTRIBUTE, "eglChooseConfig");
 
    configList = (_EGLConfig **) _eglFilterArray(disp->Configs, &count,
@@ -802,7 +731,7 @@ _eglGetConfigAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
    if (!value)
       return _eglError(EGL_BAD_PARAMETER, "eglGetConfigAttrib");
 
-   *value = GET_CONFIG_ATTRIB(conf, attribute);
+   *value = _eglGetConfigKey(conf, attribute);
    return EGL_TRUE;
 }
 

src/egl/main/eglconfig.h

@@ -3,72 +3,107 @@
 
 
 #include <assert.h>
+#include <stddef.h>
 #include "egltypedefs.h"
 
 
-#define _EGL_CONFIG_FIRST_ATTRIB EGL_BUFFER_SIZE
-#define _EGL_CONFIG_LAST_ATTRIB EGL_CONFORMANT
-#define _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS \
-   (_EGL_CONFIG_LAST_ATTRIB - _EGL_CONFIG_FIRST_ATTRIB + 1)
-
-/* Attributes outside the contiguous block:
- *
- *   EGL_Y_INVERTED_NOK
- */
-#define _EGL_CONFIG_FIRST_EXTRA_ATTRIB _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS
-#define _EGL_CONFIG_NUM_EXTRA_ATTRIBS 1
-
-#define _EGL_CONFIG_NUM_ATTRIBS \
-   _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS + _EGL_CONFIG_NUM_EXTRA_ATTRIBS
-
-
+/* update _eglValidationTable and _eglOffsetOfConfig before updating this
+ * struct */
 struct _egl_config
 {
    _EGLDisplay *Display;
-   EGLint Storage[_EGL_CONFIG_NUM_ATTRIBS];
+
+   /* core */
+   EGLint BufferSize;
+   EGLint AlphaSize;
+   EGLint BlueSize;
+   EGLint GreenSize;
+   EGLint RedSize;
+   EGLint DepthSize;
+   EGLint StencilSize;
+   EGLint ConfigCaveat;
+   EGLint ConfigID;
+   EGLint Level;
+   EGLint MaxPbufferHeight;
+   EGLint MaxPbufferPixels;
+   EGLint MaxPbufferWidth;
+   EGLint NativeRenderable;
+   EGLint NativeVisualID;
+   EGLint NativeVisualType;
+   EGLint Samples;
+   EGLint SampleBuffers;
+   EGLint SurfaceType;
+   EGLint TransparentType;
+   EGLint TransparentBlueValue;
+   EGLint TransparentGreenValue;
+   EGLint TransparentRedValue;
+   EGLint BindToTextureRGB;
+   EGLint BindToTextureRGBA;
+   EGLint MinSwapInterval;
+   EGLint MaxSwapInterval;
+   EGLint LuminanceSize;
+   EGLint AlphaMaskSize;
+   EGLint ColorBufferType;
+   EGLint RenderableType;
+   EGLint MatchNativePixmap;
+   EGLint Conformant;
+
+   /* extensions */
+   EGLint YInvertedNOK;
 };
 
 
 /**
- * Macros for source level compatibility.
- */
-#define SET_CONFIG_ATTRIB(CONF, ATTR, VAL) _eglSetConfigKey(CONF, ATTR, VAL)
-#define GET_CONFIG_ATTRIB(CONF, ATTR) _eglGetConfigKey(CONF, ATTR)
-
-
-/**
- * Given a key, return an index into the storage of the config.
- * Return -1 if the key is invalid.
+ * Map an EGL attribute enum to the offset of the member in _EGLConfig.
  */
 static INLINE EGLint
-_eglIndexConfig(const _EGLConfig *conf, EGLint key)
+_eglOffsetOfConfig(EGLint attr)
 {
-   (void) conf;
-   if (key >= _EGL_CONFIG_FIRST_ATTRIB &&
-       key < _EGL_CONFIG_FIRST_ATTRIB + _EGL_CONFIG_NUM_CONTIGUOUS_ATTRIBS)
-      return key - _EGL_CONFIG_FIRST_ATTRIB;
-   
-   switch (key) {
-   case EGL_Y_INVERTED_NOK:
-      return _EGL_CONFIG_FIRST_EXTRA_ATTRIB;
+   switch (attr) {
+#define ATTRIB_MAP(attr, memb) case attr: return offsetof(_EGLConfig, memb)
+   /* core */
+   ATTRIB_MAP(EGL_BUFFER_SIZE,               BufferSize);
+   ATTRIB_MAP(EGL_ALPHA_SIZE,                AlphaSize);
+   ATTRIB_MAP(EGL_BLUE_SIZE,                 BlueSize);
+   ATTRIB_MAP(EGL_GREEN_SIZE,                GreenSize);
+   ATTRIB_MAP(EGL_RED_SIZE,                  RedSize);
+   ATTRIB_MAP(EGL_DEPTH_SIZE,                DepthSize);
+   ATTRIB_MAP(EGL_STENCIL_SIZE,              StencilSize);
+   ATTRIB_MAP(EGL_CONFIG_CAVEAT,             ConfigCaveat);
+   ATTRIB_MAP(EGL_CONFIG_ID,                 ConfigID);
+   ATTRIB_MAP(EGL_LEVEL,                     Level);
+   ATTRIB_MAP(EGL_MAX_PBUFFER_HEIGHT,        MaxPbufferHeight);
+   ATTRIB_MAP(EGL_MAX_PBUFFER_PIXELS,        MaxPbufferPixels);
+   ATTRIB_MAP(EGL_MAX_PBUFFER_WIDTH,         MaxPbufferWidth);
+   ATTRIB_MAP(EGL_NATIVE_RENDERABLE,         NativeRenderable);
+   ATTRIB_MAP(EGL_NATIVE_VISUAL_ID,          NativeVisualID);
+   ATTRIB_MAP(EGL_NATIVE_VISUAL_TYPE,        NativeVisualType);
+   ATTRIB_MAP(EGL_SAMPLES,                   Samples);
+   ATTRIB_MAP(EGL_SAMPLE_BUFFERS,            SampleBuffers);
+   ATTRIB_MAP(EGL_SURFACE_TYPE,              SurfaceType);
+   ATTRIB_MAP(EGL_TRANSPARENT_TYPE,          TransparentType);
+   ATTRIB_MAP(EGL_TRANSPARENT_BLUE_VALUE,    TransparentBlueValue);
+   ATTRIB_MAP(EGL_TRANSPARENT_GREEN_VALUE,   TransparentGreenValue);
+   ATTRIB_MAP(EGL_TRANSPARENT_RED_VALUE,     TransparentRedValue);
+   ATTRIB_MAP(EGL_BIND_TO_TEXTURE_RGB,       BindToTextureRGB);
+   ATTRIB_MAP(EGL_BIND_TO_TEXTURE_RGBA,      BindToTextureRGBA);
+   ATTRIB_MAP(EGL_MIN_SWAP_INTERVAL,         MinSwapInterval);
+   ATTRIB_MAP(EGL_MAX_SWAP_INTERVAL,         MaxSwapInterval);
+   ATTRIB_MAP(EGL_LUMINANCE_SIZE,            LuminanceSize);
+   ATTRIB_MAP(EGL_ALPHA_MASK_SIZE,           AlphaMaskSize);
+   ATTRIB_MAP(EGL_COLOR_BUFFER_TYPE,         ColorBufferType);
+   ATTRIB_MAP(EGL_RENDERABLE_TYPE,           RenderableType);
+   ATTRIB_MAP(EGL_MATCH_NATIVE_PIXMAP,       MatchNativePixmap);
+   ATTRIB_MAP(EGL_CONFORMANT,                Conformant);
+   /* extensions */
+   ATTRIB_MAP(EGL_Y_INVERTED_NOK,            YInvertedNOK);
+#undef ATTRIB_MAP
    default:
       return -1;
    }
 }
 
 
-/**
- * Reset all keys in the config to a given value.
- */
-static INLINE void
-_eglResetConfigKeys(_EGLConfig *conf, EGLint val)
-{
-   EGLint i;
-   for (i = 0; i < _EGL_CONFIG_NUM_ATTRIBS; i++)
-      conf->Storage[i] = val;
-}
-
-
 /**
  * Update a config for a given key.
  *
@@ -79,9 +114,9 @@ _eglResetConfigKeys(_EGLConfig *conf, EGLint val)
 static INLINE void
 _eglSetConfigKey(_EGLConfig *conf, EGLint key, EGLint val)
 {
-   EGLint idx = _eglIndexConfig(conf, key);
-   assert(idx >= 0);
-   conf->Storage[idx] = val;
+   EGLint offset = _eglOffsetOfConfig(key);
+   assert(offset >= 0);
+   *((EGLint *) ((char *) conf + offset)) = val;
 }
 
 
@@ -91,9 +126,9 @@ _eglSetConfigKey(_EGLConfig *conf, EGLint key, EGLint val)
 static INLINE EGLint
 _eglGetConfigKey(const _EGLConfig *conf, EGLint key)
 {
-   EGLint idx = _eglIndexConfig(conf, key);
-   assert(idx >= 0);
-   return conf->Storage[idx];
+   EGLint offset = _eglOffsetOfConfig(key);
+   assert(offset >= 0);
+   return *((EGLint *) ((char *) conf + offset));
 }
 
 
@@ -102,34 +137,20 @@ _eglInitConfig(_EGLConfig *config, _EGLDisplay *dpy, EGLint id);
 
 
 PUBLIC EGLConfig
-_eglAddConfig(_EGLDisplay *dpy, _EGLConfig *conf);
+_eglLinkConfig(_EGLConfig *conf);
 
 
-extern EGLBoolean
-_eglCheckConfigHandle(EGLConfig config, _EGLDisplay *dpy);
+extern _EGLConfig *
+_eglLookupConfig(EGLConfig config, _EGLDisplay *dpy);
 
 
 /**
- * Lookup a handle to find the linked config.
- * Return NULL if the handle has no corresponding linked config.
- */
-static INLINE _EGLConfig *
-_eglLookupConfig(EGLConfig config, _EGLDisplay *dpy)
-{
-   _EGLConfig *conf = (_EGLConfig *) config;
-   if (!dpy || !_eglCheckConfigHandle(config, dpy))
-      conf = NULL;
-   return conf;
-}
-
-
-/**
- * Return the handle of a linked config, or NULL.
+ * Return the handle of a linked config.
  */
 static INLINE EGLConfig
 _eglGetConfigHandle(_EGLConfig *conf)
 {
-   return (EGLConfig) ((conf && conf->Display) ? conf : NULL);
+   return (EGLConfig) conf;
 }
 
 
@@ -142,7 +163,8 @@ _eglMatchConfig(const _EGLConfig *conf, const _EGLConfig *criteria);
 
 
 PUBLIC EGLBoolean
-_eglParseConfigAttribList(_EGLConfig *conf, const EGLint *attrib_list);
+_eglParseConfigAttribList(_EGLConfig *conf, _EGLDisplay *dpy,
+                          const EGLint *attrib_list);
 
 
 PUBLIC EGLint

src/egl/main/eglcontext.c

@@ -103,8 +103,7 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy, _EGLConfig *conf,
       return EGL_FALSE;
    }
 
-   memset(ctx, 0, sizeof(_EGLContext));
-   ctx->Resource.Display = dpy;
+   _eglInitResource(&ctx->Resource, sizeof(*ctx), dpy);
    ctx->ClientAPI = api;
    ctx->Config = conf;
    ctx->WindowRenderBuffer = EGL_NONE;
@@ -113,13 +112,12 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy, _EGLConfig *conf,
 
    err = _eglParseContextAttribList(ctx, attrib_list);
    if (err == EGL_SUCCESS && ctx->Config) {
-      EGLint renderable_type, api_bit;
+      EGLint api_bit;
 
-      renderable_type = GET_CONFIG_ATTRIB(ctx->Config, EGL_RENDERABLE_TYPE);
       api_bit = _eglGetContextAPIBit(ctx);
-      if (!(renderable_type & api_bit)) {
+      if (!(ctx->Config->RenderableType & api_bit)) {
          _eglLog(_EGL_DEBUG, "context api is 0x%x while config supports 0x%x",
-               api_bit, renderable_type);
+               api_bit, ctx->Config->RenderableType);
          err = EGL_BAD_CONFIG;
       }
    }
@@ -130,29 +128,6 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy, _EGLConfig *conf,
 }
 
 
-/**
- * Just a placeholder/demo function.  Real driver will never use this!
- */
-_EGLContext *
-_eglCreateContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                  _EGLContext *share_list, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Default fallback routine - drivers should usually override this.
- */
-EGLBoolean
-_eglDestroyContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
-{
-   if (!_eglIsContextBound(ctx))
-      free(ctx);
-   return EGL_TRUE;
-}
-
-
 #ifdef EGL_VERSION_1_2
 static EGLint
 _eglQueryContextRenderBuffer(_EGLContext *ctx)
@@ -183,7 +158,9 @@ _eglQueryContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *c,
 
    switch (attribute) {
    case EGL_CONFIG_ID:
-      *value = GET_CONFIG_ATTRIB(c->Config, EGL_CONFIG_ID);
+      if (!c->Config)
+         return _eglError(EGL_BAD_ATTRIBUTE, "eglQueryContext");
+      *value = c->Config->ConfigID;
       break;
    case EGL_CONTEXT_CLIENT_VERSION:
       *value = c->ClientVersion;
@@ -272,10 +249,6 @@ _eglCheckMakeCurrent(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read)
    if (!surfaceless && (draw == NULL || read == NULL))
       return _eglError(EGL_BAD_MATCH, "eglMakeCurrent");
 
-   /* context stealing from another thread is not allowed */
-   if (ctx->Binding && ctx->Binding != t)
-      return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
-
    /*
     * The spec says
     *
@@ -283,16 +256,23 @@ _eglCheckMakeCurrent(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read)
     * bound to contexts in another thread, an EGL_BAD_ACCESS error is
     * generated."
     *
-    * But it also says
+    * and
     *
     * "at most one context may be bound to a particular surface at a given
     * time"
-    *
-    * The latter is more restrictive so we can check only the latter case.
     */
-   if ((draw && draw->CurrentContext && draw->CurrentContext != ctx) ||
-       (read && read->CurrentContext && read->CurrentContext != ctx))
+   if (ctx->Binding && ctx->Binding != t)
       return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
+   if (draw && draw->CurrentContext && draw->CurrentContext != ctx) {
+      if (draw->CurrentContext->Binding != t ||
+          draw->CurrentContext->ClientAPI != ctx->ClientAPI)
+         return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
+   }
+   if (read && read->CurrentContext && read->CurrentContext != ctx) {
+      if (read->CurrentContext->Binding != t ||
+          read->CurrentContext->ClientAPI != ctx->ClientAPI)
+         return _eglError(EGL_BAD_ACCESS, "eglMakeCurrent");
+   }
 
    /* simply require the configs to be equal */
    if ((draw && draw->Config != ctx->Config) ||
@@ -323,79 +303,65 @@ _eglCheckMakeCurrent(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read)
 
 /**
  * Bind the context to the current thread and given surfaces.  Return the
- * "orphaned" context and surfaces.  Each argument is both input and output.
+ * previous bound context and surfaces.  The caller should unreference the
+ * returned context and surfaces.
+ *
+ * Making a second call with the resources returned by the first call
+ * unsurprisingly undoes the first call, except for the resouce reference
+ * counts.
  */
 EGLBoolean
-_eglBindContext(_EGLContext **ctx, _EGLSurface **draw, _EGLSurface **read)
+_eglBindContext(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read,
+                _EGLContext **old_ctx,
+                _EGLSurface **old_draw, _EGLSurface **old_read)
 {
    _EGLThreadInfo *t = _eglGetCurrentThread();
-   _EGLContext *newCtx = *ctx, *oldCtx;
-   _EGLSurface *newDraw = *draw, *newRead = *read;
+   _EGLContext *prev_ctx;
+   _EGLSurface *prev_draw, *prev_read;
 
-   if (!_eglCheckMakeCurrent(newCtx, newDraw, newRead))
+   if (!_eglCheckMakeCurrent(ctx, draw, read))
       return EGL_FALSE;
 
+   /* increment refcounts before binding */
+   _eglGetContext(ctx);
+   _eglGetSurface(draw);
+   _eglGetSurface(read);
+
    /* bind the new context */
-   oldCtx = _eglBindContextToThread(newCtx, t);
+   prev_ctx = _eglBindContextToThread(ctx, t);
 
-   /* break old bindings */
-   if (oldCtx) {
-      *ctx = oldCtx;
-      *draw = oldCtx->DrawSurface;
-      *read = oldCtx->ReadSurface;
+   /* break previous bindings */
+   if (prev_ctx) {
+      prev_draw = prev_ctx->DrawSurface;
+      prev_read = prev_ctx->ReadSurface;
 
-      if (*draw)
-         (*draw)->CurrentContext = NULL;
-      if (*read)
-         (*read)->CurrentContext = NULL;
+      if (prev_draw)
+         prev_draw->CurrentContext = NULL;
+      if (prev_read)
+         prev_read->CurrentContext = NULL;
 
-      oldCtx->DrawSurface = NULL;
-      oldCtx->ReadSurface = NULL;
+      prev_ctx->DrawSurface = NULL;
+      prev_ctx->ReadSurface = NULL;
+   }
+   else {
+      prev_draw = prev_read = NULL;
    }
 
    /* establish new bindings */
-   if (newCtx) {
-      if (newDraw)
-         newDraw->CurrentContext = newCtx;
-      if (newRead)
-         newRead->CurrentContext = newCtx;
+   if (ctx) {
+      if (draw)
+         draw->CurrentContext = ctx;
+      if (read)
+         read->CurrentContext = ctx;
 
-      newCtx->DrawSurface = newDraw;
-      newCtx->ReadSurface = newRead;
+      ctx->DrawSurface = draw;
+      ctx->ReadSurface = read;
    }
 
-   /* an old context or surface is not orphaned if it is still bound */
-   if (*ctx == newCtx)
-      *ctx = NULL;
-   if (*draw == newDraw || *draw == newRead)
-      *draw = NULL;
-   if (*read == newDraw || *read == newRead)
-      *read = NULL;
+   assert(old_ctx && old_draw && old_read);
+   *old_ctx = prev_ctx;
+   *old_draw = prev_draw;
+   *old_read = prev_read;
 
    return EGL_TRUE;
 }
-
-
-/**
- * Just a placeholder/demo function.  Drivers should override this.
- */
-EGLBoolean
-_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw,
-                _EGLSurface *read, _EGLContext *ctx)
-{
-   return EGL_FALSE;
-}
-
-
-/**
- * This is defined by the EGL_MESA_copy_context extension.
- */
-EGLBoolean
-_eglCopyContextMESA(_EGLDriver *drv, EGLDisplay dpy, EGLContext source,
-                    EGLContext dest, EGLint mask)
-{
-   /* This function will always have to be overridden/implemented in the
-    * device driver.  If the driver is based on Mesa, use _mesa_copy_context().
-    */
-   return EGL_FALSE;
-}

src/egl/main/eglcontext.h

@@ -34,51 +34,46 @@ _eglInitContext(_EGLContext *ctx, _EGLDisplay *dpy,
                 _EGLConfig *config, const EGLint *attrib_list);
 
 
-extern _EGLContext *
-_eglCreateContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, _EGLContext *share_list, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglDestroyContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx);
-
-
 extern EGLBoolean
 _eglQueryContext(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx, EGLint attribute, EGLint *value);
 
 
 PUBLIC EGLBoolean
-_eglBindContext(_EGLContext **ctx, _EGLSurface **draw, _EGLSurface **read);
-
-
-extern EGLBoolean
-_eglMakeCurrent(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *draw, _EGLSurface *read, _EGLContext *ctx);
-
-
-extern EGLBoolean
-_eglCopyContextMESA(_EGLDriver *drv, EGLDisplay dpy, EGLContext source, EGLContext dest, EGLint mask);
+_eglBindContext(_EGLContext *ctx, _EGLSurface *draw, _EGLSurface *read,
+                _EGLContext **old_ctx,
+                _EGLSurface **old_draw, _EGLSurface **old_read);
 
 
 /**
- * Return true if the context is bound to a thread.
- *
- * The binding is considered a reference to the context.  Drivers should not
- * destroy a context when it is bound.
+ * Increment reference count for the context.
  */
-static INLINE EGLBoolean
-_eglIsContextBound(_EGLContext *ctx)
+static INLINE _EGLContext *
+_eglGetContext(_EGLContext *ctx)
 {
-   return (ctx->Binding != NULL);
+   if (ctx)
+      _eglGetResource(&ctx->Resource);
+   return ctx;
 }
 
 
 /**
- * Link a context to a display and return the handle of the link.
+ * Decrement reference count for the context.
+ */
+static INLINE EGLBoolean
+_eglPutContext(_EGLContext *ctx)
+{
+   return (ctx) ? _eglPutResource(&ctx->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link a context to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLContext
-_eglLinkContext(_EGLContext *ctx, _EGLDisplay *dpy)
+_eglLinkContext(_EGLContext *ctx)
 {
-   _eglLinkResource(&ctx->Resource, _EGL_RESOURCE_CONTEXT, dpy);
+   _eglLinkResource(&ctx->Resource, _EGL_RESOURCE_CONTEXT);
    return (EGLContext) ctx;
 }
 
@@ -120,18 +115,4 @@ _eglGetContextHandle(_EGLContext *ctx)
 }
 
 
-/**
- * Return true if the context is linked to a display.
- *
- * The link is considered a reference to the context (the display is owning the
- * context).  Drivers should not destroy a context when it is linked.
- */
-static INLINE EGLBoolean
-_eglIsContextLinked(_EGLContext *ctx)
-{
-   _EGLResource *res = (_EGLResource *) ctx;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGLCONTEXT_INCLUDED */

src/egl/main/eglcurrent.c

@@ -14,33 +14,7 @@
 static _EGLThreadInfo dummy_thread = _EGL_THREAD_INFO_INITIALIZER;
 
 
-#ifdef GLX_USE_TLS
-static __thread const _EGLThreadInfo *_egl_TSD
-   __attribute__ ((tls_model("initial-exec")));
-
-static INLINE void _eglSetTSD(const _EGLThreadInfo *t)
-{
-   _egl_TSD = t;
-}
-
-static INLINE _EGLThreadInfo *_eglGetTSD(void)
-{
-   return (_EGLThreadInfo *) _egl_TSD;
-}
-
-static INLINE void _eglFiniTSD(void)
-{
-}
-
-static INLINE EGLBoolean _eglInitTSD(void (*dtor)(_EGLThreadInfo *))
-{
-   /* TODO destroy TSD */
-   (void) dtor;
-   (void) _eglFiniTSD;
-   return EGL_TRUE;
-}
-
-#elif PTHREADS
+#if PTHREADS
 #include <pthread.h>
 
 static _EGL_DECLARE_MUTEX(_egl_TSDMutex);
@@ -48,14 +22,26 @@ static EGLBoolean _egl_TSDInitialized;
 static pthread_key_t _egl_TSD;
 static void (*_egl_FreeTSD)(_EGLThreadInfo *);
 
+#ifdef GLX_USE_TLS
+static __thread const _EGLThreadInfo *_egl_TLS
+   __attribute__ ((tls_model("initial-exec")));
+#endif
+
 static INLINE void _eglSetTSD(const _EGLThreadInfo *t)
 {
    pthread_setspecific(_egl_TSD, (const void *) t);
+#ifdef GLX_USE_TLS
+   _egl_TLS = t;
+#endif
 }
 
 static INLINE _EGLThreadInfo *_eglGetTSD(void)
 {
+#ifdef GLX_USE_TLS
+   return (_EGLThreadInfo *) _egl_TLS;
+#else
    return (_EGLThreadInfo *) pthread_getspecific(_egl_TSD);
+#endif
 }
 
 static INLINE void _eglFiniTSD(void)

src/egl/main/egldisplay.c

@@ -27,7 +27,7 @@ _eglGetNativePlatformFromEnv(void)
    } egl_platforms[_EGL_NUM_PLATFORMS] = {
       { _EGL_PLATFORM_WINDOWS, "gdi" },
       { _EGL_PLATFORM_X11, "x11" },
-      { _EGL_PLATFORM_DRM, "kms" },
+      { _EGL_PLATFORM_DRM, "drm" },
       { _EGL_PLATFORM_FBDEV, "fbdev" }
    };
    _EGLPlatformType plat = _EGL_INVALID_PLATFORM;
@@ -233,17 +233,53 @@ _eglCheckResource(void *res, _EGLResourceType type, _EGLDisplay *dpy)
 
 
 /**
- * Link a resource to a display.
+ * Initialize a display resource.
  */
 void
-_eglLinkResource(_EGLResource *res, _EGLResourceType type, _EGLDisplay *dpy)
+_eglInitResource(_EGLResource *res, EGLint size, _EGLDisplay *dpy)
 {
-   assert(!res->Display || res->Display == dpy);
-
+   memset(res, 0, size);
    res->Display = dpy;
+   res->RefCount = 1;
+}
+
+
+/**
+ * Increment reference count for the resource.
+ */
+void
+_eglGetResource(_EGLResource *res)
+{
+   assert(res && res->RefCount > 0);
+   /* hopefully a resource is always manipulated with its display locked */
+   res->RefCount++;
+}
+
+
+/**
+ * Decrement reference count for the resource.
+ */
+EGLBoolean
+_eglPutResource(_EGLResource *res)
+{
+   assert(res && res->RefCount > 0);
+   res->RefCount--;
+   return (!res->RefCount);
+}
+
+
+/**
+ * Link a resource to its display.
+ */
+void
+_eglLinkResource(_EGLResource *res, _EGLResourceType type)
+{
+   assert(res->Display);
+
    res->IsLinked = EGL_TRUE;
-   res->Next = dpy->ResourceLists[type];
-   dpy->ResourceLists[type] = res;
+   res->Next = res->Display->ResourceLists[type];
+   res->Display->ResourceLists[type] = res;
+   _eglGetResource(res);
 }
 
 
@@ -270,6 +306,9 @@ _eglUnlinkResource(_EGLResource *res, _EGLResourceType type)
    }
 
    res->Next = NULL;
-   /* do not reset res->Display */
    res->IsLinked = EGL_FALSE;
+   _eglPutResource(res);
+
+   /* We always unlink before destroy.  The driver still owns a reference */
+   assert(res->RefCount);
 }

src/egl/main/egldisplay.h

@@ -40,6 +40,7 @@ struct _egl_resource
    /* which display the resource belongs to */
    _EGLDisplay *Display;
    EGLBoolean IsLinked;
+   EGLint RefCount;
 
    /* used to link resources of the same type */
    _EGLResource *Next;
@@ -162,7 +163,19 @@ _eglGetDisplayHandle(_EGLDisplay *dpy)
 
 
 extern void
-_eglLinkResource(_EGLResource *res, _EGLResourceType type, _EGLDisplay *dpy);
+_eglInitResource(_EGLResource *res, EGLint size, _EGLDisplay *dpy);
+
+
+PUBLIC void
+_eglGetResource(_EGLResource *res);
+
+
+PUBLIC EGLBoolean
+_eglPutResource(_EGLResource *res);
+
+
+extern void
+_eglLinkResource(_EGLResource *res, _EGLResourceType type);
 
 
 extern void

src/egl/main/egldriver.c

@@ -9,19 +9,10 @@
 #include <stdlib.h>
 
 #include "eglstring.h"
-#include "eglconfig.h"
-#include "eglcontext.h"
 #include "egldefines.h"
 #include "egldisplay.h"
 #include "egldriver.h"
 #include "egllog.h"
-#include "eglmisc.h"
-#include "eglmode.h"
-#include "eglscreen.h"
-#include "eglstring.h"
-#include "eglsurface.h"
-#include "eglimage.h"
-#include "eglsync.h"
 #include "eglmutex.h"
 
 #if defined(_EGL_OS_UNIX)
@@ -404,35 +395,62 @@ _eglPreloadForEach(const char *search_path,
 static const char *
 _eglGetSearchPath(void)
 {
-   static const char *search_path;
+   static char search_path[1024];
 
 #if defined(_EGL_OS_UNIX) || defined(_EGL_OS_WINDOWS)
-   if (!search_path) {
-      static char buffer[1024];
-      const char *p;
+   if (search_path[0] == '\0') {
+      char *buf = search_path;
+      size_t len = sizeof(search_path);
+      EGLBoolean use_env;
+      char dir_sep;
       int ret;
 
-      p = getenv("EGL_DRIVERS_PATH");
 #if defined(_EGL_OS_UNIX)
-      if (p && (geteuid() != getuid() || getegid() != getgid())) {
+      use_env = (geteuid() == getuid() && getegid() == getgid());
+      dir_sep = '/';
+#else
+      use_env = EGL_TRUE;
+      dir_sep = '\\';
+#endif
+
+      if (use_env) {
+         char *p;
+
+         /* extract the dirname from EGL_DRIVER */
+         p = getenv("EGL_DRIVER");
+         if (p && strchr(p, dir_sep)) {
+            ret = _eglsnprintf(buf, len, "%s", p);
+            if (ret > 0 && ret < len) {
+               p = strrchr(buf, dir_sep);
+               *p++ = ':';
+
+               len -= p - buf;
+               buf = p;
+            }
+         }
+
+         /* append EGL_DRIVERS_PATH */
+         p = getenv("EGL_DRIVERS_PATH");
+         if (p) {
+            ret = _eglsnprintf(buf, len, "%s:", p);
+            if (ret > 0 && ret < len) {
+               buf += ret;
+               len -= ret;
+            }
+         }
+      }
+      else {
          _eglLog(_EGL_DEBUG,
                "ignore EGL_DRIVERS_PATH for setuid/setgid binaries");
-         p = NULL;
       }
-#endif /* _EGL_OS_UNIX */
 
-      if (p) {
-         ret = _eglsnprintf(buffer, sizeof(buffer),
-               "%s:%s", p, _EGL_DRIVER_SEARCH_DIR);
-         if (ret > 0 && ret < sizeof(buffer))
-            search_path = buffer;
-      }
+      ret = _eglsnprintf(buf, len, "%s", _EGL_DRIVER_SEARCH_DIR);
+      if (ret < 0 || ret >= len)
+         search_path[0] = '\0';
+
+      _eglLog(_EGL_DEBUG, "EGL search path is %s", search_path);
    }
-   if (!search_path)
-      search_path = _EGL_DRIVER_SEARCH_DIR;
-#else
-   search_path = "";
-#endif
+#endif /* defined(_EGL_OS_UNIX) || defined(_EGL_OS_WINDOWS) */
 
    return search_path;
 }
@@ -663,77 +681,6 @@ _eglUnloadDrivers(void)
 }
 
 
-/**
- * Plug all the available fallback routines into the given driver's
- * dispatch table.
- */
-void
-_eglInitDriverFallbacks(_EGLDriver *drv)
-{
-   /* If a pointer is set to NULL, then the device driver _really_ has
-    * to implement it.
-    */
-   drv->API.Initialize = NULL;
-   drv->API.Terminate = NULL;
-
-   drv->API.GetConfigs = _eglGetConfigs;
-   drv->API.ChooseConfig = _eglChooseConfig;
-   drv->API.GetConfigAttrib = _eglGetConfigAttrib;
-
-   drv->API.CreateContext = _eglCreateContext;
-   drv->API.DestroyContext = _eglDestroyContext;
-   drv->API.MakeCurrent = _eglMakeCurrent;
-   drv->API.QueryContext = _eglQueryContext;
-
-   drv->API.CreateWindowSurface = _eglCreateWindowSurface;
-   drv->API.CreatePixmapSurface = _eglCreatePixmapSurface;
-   drv->API.CreatePbufferSurface = _eglCreatePbufferSurface;
-   drv->API.DestroySurface = _eglDestroySurface;
-   drv->API.QuerySurface = _eglQuerySurface;
-   drv->API.SurfaceAttrib = _eglSurfaceAttrib;
-   drv->API.BindTexImage = _eglBindTexImage;
-   drv->API.ReleaseTexImage = _eglReleaseTexImage;
-   drv->API.SwapInterval = _eglSwapInterval;
-   drv->API.SwapBuffers = _eglSwapBuffers;
-   drv->API.CopyBuffers = _eglCopyBuffers;
-
-   drv->API.QueryString = _eglQueryString;
-   drv->API.WaitClient = _eglWaitClient;
-   drv->API.WaitNative = _eglWaitNative;
-
-#ifdef EGL_MESA_screen_surface
-   drv->API.ChooseModeMESA = _eglChooseModeMESA; 
-   drv->API.GetModesMESA = _eglGetModesMESA;
-   drv->API.GetModeAttribMESA = _eglGetModeAttribMESA;
-   drv->API.GetScreensMESA = _eglGetScreensMESA;
-   drv->API.CreateScreenSurfaceMESA = _eglCreateScreenSurfaceMESA;
-   drv->API.ShowScreenSurfaceMESA = _eglShowScreenSurfaceMESA;
-   drv->API.ScreenPositionMESA = _eglScreenPositionMESA;
-   drv->API.QueryScreenMESA = _eglQueryScreenMESA;
-   drv->API.QueryScreenSurfaceMESA = _eglQueryScreenSurfaceMESA;
-   drv->API.QueryScreenModeMESA = _eglQueryScreenModeMESA;
-   drv->API.QueryModeStringMESA = _eglQueryModeStringMESA;
-#endif /* EGL_MESA_screen_surface */
-
-#ifdef EGL_VERSION_1_2
-   drv->API.CreatePbufferFromClientBuffer = _eglCreatePbufferFromClientBuffer;
-#endif /* EGL_VERSION_1_2 */
-
-#ifdef EGL_KHR_image_base
-   drv->API.CreateImageKHR = _eglCreateImageKHR;
-   drv->API.DestroyImageKHR = _eglDestroyImageKHR;
-#endif /* EGL_KHR_image_base */
-
-#ifdef EGL_KHR_reusable_sync
-   drv->API.CreateSyncKHR = _eglCreateSyncKHR;
-   drv->API.DestroySyncKHR = _eglDestroySyncKHR;
-   drv->API.ClientWaitSyncKHR = _eglClientWaitSyncKHR;
-   drv->API.SignalSyncKHR = _eglSignalSyncKHR;
-   drv->API.GetSyncAttribKHR = _eglGetSyncAttribKHR;
-#endif /* EGL_KHR_reusable_sync */
-}
-
-
 /**
  * Invoke a callback function on each EGL search path.
  *

src/egl/main/egldriver.h

@@ -80,6 +80,7 @@ extern void
 _eglUnloadDrivers(void);
 
 
+/* defined in eglfallbacks.c */
 PUBLIC void
 _eglInitDriverFallbacks(_EGLDriver *drv);
 

src/egl/main/eglfallbacks.c

@@ -0,0 +1,99 @@
+#include <string.h>
+#include "egltypedefs.h"
+#include "egldriver.h"
+#include "eglconfig.h"
+#include "eglcontext.h"
+#include "eglsurface.h"
+#include "eglmisc.h"
+#include "eglscreen.h"
+#include "eglmode.h"
+#include "eglsync.h"
+
+
+static EGLBoolean
+_eglReturnFalse(void)
+{
+   return EGL_FALSE;
+}
+
+
+/**
+ * Plug all the available fallback routines into the given driver's
+ * dispatch table.
+ */
+void
+_eglInitDriverFallbacks(_EGLDriver *drv)
+{
+   memset(&drv->API, 0, sizeof(drv->API));
+
+   /* the driver has to implement these */
+   drv->API.Initialize = NULL;
+   drv->API.Terminate = NULL;
+
+   drv->API.GetConfigs = _eglGetConfigs;
+   drv->API.ChooseConfig = _eglChooseConfig;
+   drv->API.GetConfigAttrib = _eglGetConfigAttrib;
+
+   drv->API.CreateContext = (CreateContext_t) _eglReturnFalse;
+   drv->API.DestroyContext = (DestroyContext_t) _eglReturnFalse;
+   drv->API.MakeCurrent = (MakeCurrent_t) _eglReturnFalse;
+   drv->API.QueryContext = _eglQueryContext;
+
+   drv->API.CreateWindowSurface = (CreateWindowSurface_t) _eglReturnFalse;
+   drv->API.CreatePixmapSurface = (CreatePixmapSurface_t) _eglReturnFalse;
+   drv->API.CreatePbufferSurface = (CreatePbufferSurface_t) _eglReturnFalse;
+   drv->API.CreatePbufferFromClientBuffer =
+      (CreatePbufferFromClientBuffer_t) _eglReturnFalse;
+   drv->API.DestroySurface = (DestroySurface_t) _eglReturnFalse;
+   drv->API.QuerySurface = _eglQuerySurface;
+   drv->API.SurfaceAttrib = _eglSurfaceAttrib;
+
+   drv->API.BindTexImage = (BindTexImage_t) _eglReturnFalse;
+   drv->API.ReleaseTexImage = (ReleaseTexImage_t) _eglReturnFalse;
+   drv->API.CopyBuffers = (CopyBuffers_t) _eglReturnFalse;
+   drv->API.SwapBuffers = (SwapBuffers_t) _eglReturnFalse;
+   drv->API.SwapInterval = _eglSwapInterval;
+
+   drv->API.WaitClient = (WaitClient_t) _eglReturnFalse;
+   drv->API.WaitNative = (WaitNative_t) _eglReturnFalse;
+   drv->API.GetProcAddress = (GetProcAddress_t) _eglReturnFalse;
+   drv->API.QueryString = _eglQueryString;
+
+#ifdef EGL_MESA_screen_surface
+   drv->API.CopyContextMESA = (CopyContextMESA_t) _eglReturnFalse;
+   drv->API.CreateScreenSurfaceMESA =
+      (CreateScreenSurfaceMESA_t) _eglReturnFalse;
+   drv->API.ShowScreenSurfaceMESA = (ShowScreenSurfaceMESA_t) _eglReturnFalse;
+   drv->API.ChooseModeMESA = _eglChooseModeMESA;
+   drv->API.GetModesMESA = _eglGetModesMESA;
+   drv->API.GetModeAttribMESA = _eglGetModeAttribMESA;
+   drv->API.GetScreensMESA = _eglGetScreensMESA;
+   drv->API.ScreenPositionMESA = _eglScreenPositionMESA;
+   drv->API.QueryScreenMESA = _eglQueryScreenMESA;
+   drv->API.QueryScreenSurfaceMESA = _eglQueryScreenSurfaceMESA;
+   drv->API.QueryScreenModeMESA = _eglQueryScreenModeMESA;
+   drv->API.QueryModeStringMESA = _eglQueryModeStringMESA;
+#endif /* EGL_MESA_screen_surface */
+
+#ifdef EGL_KHR_image_base
+   drv->API.CreateImageKHR = NULL;
+   drv->API.DestroyImageKHR = NULL;
+#endif /* EGL_KHR_image_base */
+
+#ifdef EGL_KHR_reusable_sync
+   drv->API.CreateSyncKHR = NULL;
+   drv->API.DestroySyncKHR = NULL;
+   drv->API.ClientWaitSyncKHR = NULL;
+   drv->API.SignalSyncKHR = NULL;
+   drv->API.GetSyncAttribKHR = _eglGetSyncAttribKHR;
+#endif /* EGL_KHR_reusable_sync */
+
+#ifdef EGL_MESA_drm_image
+   drv->API.CreateDRMImageMESA = NULL;
+   drv->API.ExportDRMImageMESA = NULL;
+#endif
+
+#ifdef EGL_NOK_swap_region
+   drv->API.SwapBuffersRegionNOK = NULL;
+#endif
+}

src/egl/main/eglimage.c

@@ -2,7 +2,6 @@
 #include <string.h>
 
 #include "eglimage.h"
-#include "eglcurrent.h"
 #include "egllog.h"
 
 
@@ -12,28 +11,57 @@
 /**
  * Parse the list of image attributes and return the proper error code.
  */
-static EGLint
-_eglParseImageAttribList(_EGLImage *img, const EGLint *attrib_list)
+EGLint
+_eglParseImageAttribList(_EGLImageAttribs *attrs, _EGLDisplay *dpy,
+                         const EGLint *attrib_list)
 {
    EGLint i, err = EGL_SUCCESS;
 
+   (void) dpy;
+
+   memset(attrs, 0, sizeof(attrs));
+   attrs->ImagePreserved = EGL_FALSE;
+   attrs->GLTextureLevel = 0;
+   attrs->GLTextureZOffset = 0;
+
    if (!attrib_list)
-      return EGL_SUCCESS;
+      return err;
 
    for (i = 0; attrib_list[i] != EGL_NONE; i++) {
       EGLint attr = attrib_list[i++];
       EGLint val = attrib_list[i];
 
       switch (attr) {
+      /* EGL_KHR_image_base */
       case EGL_IMAGE_PRESERVED_KHR:
-         img->Preserved = val;
+         attrs->ImagePreserved = val;
          break;
+
+      /* EGL_KHR_gl_image */
       case EGL_GL_TEXTURE_LEVEL_KHR:
-         img->GLTextureLevel = val;
+         attrs->GLTextureLevel = val;
          break;
       case EGL_GL_TEXTURE_ZOFFSET_KHR:
-         img->GLTextureZOffset = val;
+         attrs->GLTextureZOffset = val;
          break;
+
+      /* EGL_MESA_drm_image */
+      case EGL_WIDTH:
+         attrs->Width = val;
+         break;
+      case EGL_HEIGHT:
+         attrs->Height = val;
+         break;
+      case EGL_DRM_BUFFER_FORMAT_MESA:
+         attrs->DRMBufferFormatMESA = val;
+         break;
+      case EGL_DRM_BUFFER_USE_MESA:
+         attrs->DRMBufferUseMESA = val;
+         break;
+      case EGL_DRM_BUFFER_STRIDE_MESA:
+         attrs->DRMBufferStrideMESA = val;
+         break;
+
       default:
          /* unknown attrs are ignored */
          break;
@@ -50,41 +78,12 @@ _eglParseImageAttribList(_EGLImage *img, const EGLint *attrib_list)
 
 
 EGLBoolean
-_eglInitImage(_EGLImage *img, _EGLDisplay *dpy, const EGLint *attrib_list)
+_eglInitImage(_EGLImage *img, _EGLDisplay *dpy)
 {
-   EGLint err;
-
-   memset(img, 0, sizeof(_EGLImage));
-   img->Resource.Display = dpy;
-
-   img->Preserved = EGL_FALSE;
-   img->GLTextureLevel = 0;
-   img->GLTextureZOffset = 0;
-
-   err = _eglParseImageAttribList(img, attrib_list);
-   if (err != EGL_SUCCESS)
-      return _eglError(err, "eglCreateImageKHR");
+   _eglInitResource(&img->Resource, sizeof(*img), dpy);
 
    return EGL_TRUE;
 }
 
 
-_EGLImage *
-_eglCreateImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx,
-                   EGLenum target, EGLClientBuffer buffer,
-                   const EGLint *attr_list)
-{
-   /* driver should override this function */
-   return NULL;
-}
-
-
-EGLBoolean
-_eglDestroyImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLImage *image)
-{
-   /* driver should override this function */
-   return EGL_FALSE;
-}
-
-
 #endif /* EGL_KHR_image_base */

src/egl/main/eglimage.h

@@ -6,6 +6,23 @@
 #include "egldisplay.h"
 
 
+struct _egl_image_attribs
+{
+   /* EGL_KHR_image_base */
+   EGLBoolean ImagePreserved;
+
+   /* EGL_KHR_gl_image */
+   EGLint GLTextureLevel;
+   EGLint GLTextureZOffset;
+
+   /* EGL_MESA_drm_image */
+   EGLint Width;
+   EGLint Height;
+   EGLint DRMBufferFormatMESA;
+   EGLint DRMBufferUseMESA;
+   EGLint DRMBufferStrideMESA;
+};
+
 /**
  * "Base" class for device driver images.
  */
@@ -13,34 +30,48 @@ struct _egl_image
 {
    /* An image is a display resource */
    _EGLResource Resource;
-
-   EGLBoolean Preserved;
-   EGLint GLTextureLevel;
-   EGLint GLTextureZOffset;
 };
 
 
+PUBLIC EGLint
+_eglParseImageAttribList(_EGLImageAttribs *attrs, _EGLDisplay *dpy,
+                         const EGLint *attrib_list);
+
+
 PUBLIC EGLBoolean
-_eglInitImage(_EGLImage *img, _EGLDisplay *dpy, const EGLint *attrib_list);
-
-
-extern _EGLImage *
-_eglCreateImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx,
-                   EGLenum target, EGLClientBuffer buffer, const EGLint *attr_list);
-
-
-extern EGLBoolean
-_eglDestroyImageKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLImage *image);
+_eglInitImage(_EGLImage *img, _EGLDisplay *dpy);
 
 
 /**
- * Link an image to a display and return the handle of the link.
+ * Increment reference count for the image.
+ */
+static INLINE _EGLImage *
+_eglGetImage(_EGLImage *img)
+{
+   if (img)
+      _eglGetResource(&img->Resource);
+   return img;
+}
+
+
+/**
+ * Decrement reference count for the image.
+ */
+static INLINE EGLBoolean
+_eglPutImage(_EGLImage *img)
+{
+   return (img) ? _eglPutResource(&img->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link an image to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLImageKHR
-_eglLinkImage(_EGLImage *img, _EGLDisplay *dpy)
+_eglLinkImage(_EGLImage *img)
 {
-   _eglLinkResource(&img->Resource, _EGL_RESOURCE_IMAGE, dpy);
+   _eglLinkResource(&img->Resource, _EGL_RESOURCE_IMAGE);
    return (EGLImageKHR) img;
 }
 
@@ -82,15 +113,4 @@ _eglGetImageHandle(_EGLImage *img)
 }
 
 
-/**
- * Return true if the image is linked to a display.
- */
-static INLINE EGLBoolean
-_eglIsImageLinked(_EGLImage *img)
-{
-   _EGLResource *res = (_EGLResource *) img;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGLIMAGE_INCLUDED */

src/egl/main/egllog.c

@@ -151,6 +151,7 @@ _eglLog(EGLint level, const char *fmtStr, ...)
 {
    va_list args;
    char msg[MAXSTRING];
+   int ret;
 
    /* one-time initialization; a little race here is fine */
    if (!logging.initialized)
@@ -162,7 +163,9 @@ _eglLog(EGLint level, const char *fmtStr, ...)
 
    if (logging.logger) {
       va_start(args, fmtStr);
-      vsnprintf(msg, MAXSTRING, fmtStr, args);
+      ret = vsnprintf(msg, MAXSTRING, fmtStr, args);
+      if (ret < 0 || ret >= MAXSTRING)
+         strcpy(msg, "<message truncated>");
       va_end(args);
 
       logging.logger(level, msg);

src/egl/main/eglmisc.c

@@ -158,32 +158,3 @@ _eglQueryString(_EGLDriver *drv, _EGLDisplay *dpy, EGLint name)
       return NULL;
    }
 }
-
-
-EGLBoolean
-_eglWaitClient(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx)
-{
-   /* just a placeholder */
-   (void) drv;
-   (void) dpy;
-   (void) ctx;
-   return EGL_TRUE;
-}
-
-
-EGLBoolean
-_eglWaitNative(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine)
-{
-   /* just a placeholder */
-   (void) drv;
-   (void) dpy;
-   switch (engine) {
-   case EGL_CORE_NATIVE_ENGINE:
-      break;
-   default:
-      _eglError(EGL_BAD_PARAMETER, "eglWaitNative(engine)");
-      return EGL_FALSE;
-   }
-
-   return EGL_TRUE;
-}

src/egl/main/eglmisc.h

@@ -37,12 +37,4 @@ extern const char *
 _eglQueryString(_EGLDriver *drv, _EGLDisplay *dpy, EGLint name);
 
 
-extern EGLBoolean
-_eglWaitClient(_EGLDriver *drv, _EGLDisplay *dpy, _EGLContext *ctx);
-
-
-extern EGLBoolean
-_eglWaitNative(_EGLDriver *drv, _EGLDisplay *dpy, EGLint engine);
-
-
 #endif /* EGLMISC_INCLUDED */

src/egl/main/eglmode.c

@@ -3,11 +3,9 @@
 #include <string.h>
 
 #include "egldisplay.h"
-#include "egldriver.h"
 #include "eglmode.h"
 #include "eglcurrent.h"
 #include "eglscreen.h"
-#include "eglstring.h"
 
 
 #ifdef EGL_MESA_screen_surface
@@ -31,12 +29,19 @@ _eglLookupMode(EGLModeMESA mode, _EGLDisplay *disp)
    /* loop over all screens on the display */
    for (scrnum = 0; scrnum < disp->Screens->Size; scrnum++) {
       const _EGLScreen *scrn = disp->Screens->Elements[scrnum];
-      EGLint i;
-      /* search list of modes for handle */
-      for (i = 0; i < scrn->NumModes; i++) {
-         if (scrn->Modes[i].Handle == mode) {
-            return scrn->Modes + i;
-         }
+      EGLint idx;
+
+      /*
+       * the mode ids of a screen ranges from scrn->Handle to scrn->Handle +
+       * scrn->NumModes
+       */
+      if (mode >= scrn->Handle &&
+          mode < scrn->Handle + _EGL_SCREEN_MAX_MODES) {
+         idx = mode - scrn->Handle;
+
+         assert(idx < scrn->NumModes && scrn->Modes[idx].Handle == mode);
+
+         return &scrn->Modes[idx];
       }
    }
 
@@ -44,45 +49,6 @@ _eglLookupMode(EGLModeMESA mode, _EGLDisplay *disp)
 }
 
 
-/**
- * Add a new mode with the given attributes (width, height, depth, refreshRate)
- * to the given screen.
- * Assign a new mode ID/handle to the mode as well.
- * \return pointer to the new _EGLMode
- */
-_EGLMode *
-_eglAddNewMode(_EGLScreen *screen, EGLint width, EGLint height,
-               EGLint refreshRate, const char *name)
-{
-   EGLint n;
-   _EGLMode *newModes;
-
-   assert(screen);
-   assert(width > 0);
-   assert(height > 0);
-   assert(refreshRate > 0);
-
-   n = screen->NumModes;
-   newModes = (_EGLMode *) realloc(screen->Modes, (n+1) * sizeof(_EGLMode));
-   if (newModes) {
-      screen->Modes = newModes;
-      screen->Modes[n].Handle = n + 1;
-      screen->Modes[n].Width = width;
-      screen->Modes[n].Height = height;
-      screen->Modes[n].RefreshRate = refreshRate;
-      screen->Modes[n].Optimal = EGL_FALSE;
-      screen->Modes[n].Interlaced = EGL_FALSE;
-      screen->Modes[n].Name = _eglstrdup(name);
-      screen->NumModes++;
-      return screen->Modes + n;
-   }
-   else {
-      return NULL;
-   }
-}
-
-
-
 /**
  * Parse the attrib_list to fill in the fields of the given _eglMode
  * Return EGL_FALSE if any errors, EGL_TRUE otherwise.

src/egl/main/eglmode.h

@@ -32,11 +32,6 @@ extern _EGLMode *
 _eglLookupMode(EGLModeMESA mode, _EGLDisplay *dpy);
 
 
-PUBLIC _EGLMode *
-_eglAddNewMode(_EGLScreen *screen, EGLint width, EGLint height,
-               EGLint refreshRate, const char *name);
-
-
 extern EGLBoolean
 _eglChooseModeMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn,
                    const EGLint *attrib_list, EGLModeMESA *modes,

src/egl/main/eglscreen.c

@@ -18,7 +18,6 @@
 #include "egldisplay.h"
 #include "eglcurrent.h"
 #include "eglmode.h"
-#include "eglconfig.h"
 #include "eglsurface.h"
 #include "eglscreen.h"
 #include "eglmutex.h"
@@ -42,7 +41,8 @@ _eglAllocScreenHandle(void)
    EGLScreenMESA s;
 
    _eglLockMutex(&_eglNextScreenHandleMutex);
-   s = _eglNextScreenHandle++;
+   s = _eglNextScreenHandle;
+   _eglNextScreenHandle += _EGL_SCREEN_MAX_MODES;
    _eglUnlockMutex(&_eglNextScreenHandleMutex);
 
    return s;
@@ -53,16 +53,54 @@ _eglAllocScreenHandle(void)
  * Initialize an _EGLScreen object to default values.
  */
 void
-_eglInitScreen(_EGLScreen *screen)
+_eglInitScreen(_EGLScreen *screen, _EGLDisplay *dpy, EGLint num_modes)
 {
    memset(screen, 0, sizeof(_EGLScreen));
+
+   screen->Display = dpy;
+   screen->NumModes = num_modes;
    screen->StepX = 1;
    screen->StepY = 1;
+
+   if (num_modes > _EGL_SCREEN_MAX_MODES)
+      num_modes = _EGL_SCREEN_MAX_MODES;
+   screen->Modes = (_EGLMode *) calloc(num_modes, sizeof(*screen->Modes));
+   screen->NumModes = (screen->Modes) ? num_modes : 0;
 }
 
 
 /**
- * Given a public screen handle, return the internal _EGLScreen object.
+ * Link a screen to its display and return the handle of the link.
+ * The handle can be passed to client directly.
+ */
+EGLScreenMESA
+_eglLinkScreen(_EGLScreen *screen)
+{
+   _EGLDisplay *display;
+   EGLint i;
+
+   assert(screen && screen->Display);
+   display = screen->Display;
+
+   if (!display->Screens) {
+      display->Screens = _eglCreateArray("Screen", 4);
+      if (!display->Screens)
+         return (EGLScreenMESA) 0;
+   }
+
+   screen->Handle = _eglAllocScreenHandle();
+   for (i = 0; i < screen->NumModes; i++)
+      screen->Modes[i].Handle = screen->Handle + i;
+
+   _eglAppendArray(display->Screens, (void *) screen);
+
+   return screen->Handle;
+}
+
+
+/**
+ * Lookup a handle to find the linked config.
+ * Return NULL if the handle has no corresponding linked config.
  */
 _EGLScreen *
 _eglLookupScreen(EGLScreenMESA screen, _EGLDisplay *display)
@@ -74,39 +112,21 @@ _eglLookupScreen(EGLScreenMESA screen, _EGLDisplay *display)
 
    for (i = 0; i < display->Screens->Size; i++) {
       _EGLScreen *scr = (_EGLScreen *) display->Screens->Elements[i];
-      if (scr->Handle == screen)
+      if (scr->Handle == screen) {
+         assert(scr->Display == display);
          return scr;
+      }
    }
    return NULL;
 }
 
 
-/**
- * Add the given _EGLScreen to the display's list of screens.
- */
-void
-_eglAddScreen(_EGLDisplay *display, _EGLScreen *screen)
-{
-   assert(display);
-   assert(screen);
-
-   if (!display->Screens) {
-      display->Screens = _eglCreateArray("Screen", 4);
-      if (!display->Screens)
-         return;
-   }
-   screen->Handle = _eglAllocScreenHandle();
-   _eglAppendArray(display->Screens, (void *) screen);
-}
-
-
-
 static EGLBoolean
 _eglFlattenScreen(void *elem, void *buffer)
 {
    _EGLScreen *scr = (_EGLScreen *) elem;
    EGLScreenMESA *handle = (EGLScreenMESA *) buffer;
-   *handle = scr->Handle;
+   *handle = _eglGetScreenHandle(scr);
    return EGL_TRUE;
 }
 
@@ -122,66 +142,6 @@ _eglGetScreensMESA(_EGLDriver *drv, _EGLDisplay *display, EGLScreenMESA *screens
 }
 
 
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreateScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                            const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Show the given surface on the named screen.
- * If surface is EGL_NO_SURFACE, disable the screen's output.
- * 
- * This is just a placeholder function; drivers will always override
- * this with code that _really_ shows the surface.
- */
-EGLBoolean
-_eglShowScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy,
-                          _EGLScreen *scrn, _EGLSurface *surf,
-                          _EGLMode *mode)
-{
-   if (!surf) {
-      scrn->CurrentSurface = NULL;
-   }
-   else {
-      if (surf->Type != EGL_SCREEN_BIT_MESA) {
-         _eglError(EGL_BAD_SURFACE, "eglShowSurfaceMESA");
-         return EGL_FALSE;
-      }
-      if (surf->Width < mode->Width || surf->Height < mode->Height) {
-         _eglError(EGL_BAD_SURFACE,
-                   "eglShowSurfaceMESA(surface smaller than screen size)");
-         return EGL_FALSE;
-      }
-
-      scrn->CurrentSurface = surf;
-      scrn->CurrentMode = mode;
-   }
-   return EGL_TRUE;
-}
-
-
-/**
- * Set a screen's current display mode.
- * Note: mode = EGL_NO_MODE is valid (turns off the screen)
- *
- * This is just a placeholder function; drivers will always override
- * this with code that _really_ sets the mode.
- */
-EGLBoolean
-_eglScreenModeMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn,
-                   _EGLMode *m)
-{
-   scrn->CurrentMode = m;
-   return EGL_TRUE;
-}
-
-
 /**
  * Set a screen's surface origin.
  */
@@ -242,33 +202,4 @@ _eglQueryScreenMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn,
 }
 
 
-/**
- * Delete the modes associated with given screen.
- */
-void
-_eglDestroyScreenModes(_EGLScreen *scrn)
-{
-   EGLint i;
-   for (i = 0; i < scrn->NumModes; i++) {
-      if (scrn->Modes[i].Name)
-         free((char *) scrn->Modes[i].Name); /* cast away const */
-   }
-   if (scrn->Modes)
-      free(scrn->Modes);
-   scrn->Modes = NULL;
-   scrn->NumModes = 0;
-}
-
-      
-/**
- * Default fallback routine - drivers should usually override this.
- */
-void
-_eglDestroyScreen(_EGLScreen *scrn)
-{
-   _eglDestroyScreenModes(scrn);
-   free(scrn);
-}
-
-
 #endif /* EGL_MESA_screen_surface */

src/egl/main/eglscreen.h

@@ -8,6 +8,9 @@
 #ifdef EGL_MESA_screen_surface
 
 
+#define _EGL_SCREEN_MAX_MODES 16
+
+
 /**
  * Per-screen information.
  * Note that an EGL screen doesn't have a size.  A screen may be set to
@@ -19,6 +22,8 @@
  */
 struct _egl_screen
 {
+   _EGLDisplay *Display;
+
    EGLScreenMESA Handle; /* The public/opaque handle which names this object */
 
    _EGLMode *CurrentMode;
@@ -33,41 +38,35 @@ struct _egl_screen
 
 
 PUBLIC void
-_eglInitScreen(_EGLScreen *screen);
+_eglInitScreen(_EGLScreen *screen, _EGLDisplay *dpy, EGLint num_modes);
+
+
+PUBLIC EGLScreenMESA
+_eglLinkScreen(_EGLScreen *screen);
 
 
 extern _EGLScreen *
 _eglLookupScreen(EGLScreenMESA screen, _EGLDisplay *dpy);
 
 
-PUBLIC void
-_eglAddScreen(_EGLDisplay *display, _EGLScreen *screen);
+/**
+ * Return the handle of a linked screen.
+ */
+static INLINE EGLScreenMESA
+_eglGetScreenHandle(_EGLScreen *screen)
+{
+   return (screen) ? screen->Handle : (EGLScreenMESA) 0;
+}
 
 
 extern EGLBoolean
 _eglGetScreensMESA(_EGLDriver *drv, _EGLDisplay *dpy, EGLScreenMESA *screens, EGLint max_screens, EGLint *num_screens);
 
 
-extern _EGLSurface *
-_eglCreateScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglShowScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, _EGLSurface *surf, _EGLMode *m);
-
-
-extern EGLBoolean
-_eglScreenModeMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, _EGLMode *m);
-
-
 extern EGLBoolean
 _eglScreenPositionMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, EGLint x, EGLint y);
 
 
-extern EGLBoolean
-_eglQueryDisplayMESA(_EGLDriver *drv, _EGLDisplay *dpy, EGLint attribute, EGLint *value);
-
-
 extern EGLBoolean
 _eglQueryScreenSurfaceMESA(_EGLDriver *drv, _EGLDisplay *dpy,
                            _EGLScreen *scrn, _EGLSurface **surface);
@@ -81,14 +80,6 @@ extern EGLBoolean
 _eglQueryScreenMESA(_EGLDriver *drv, _EGLDisplay *dpy, _EGLScreen *scrn, EGLint attribute, EGLint *value);
 
 
-extern void
-_eglDestroyScreenModes(_EGLScreen *scrn);
-
-
-PUBLIC void
-_eglDestroyScreen(_EGLScreen *scrn);
-
-
 #endif /* EGL_MESA_screen_surface */
 
 

src/egl/main/eglsurface.c

@@ -17,12 +17,12 @@
 static void
 _eglClampSwapInterval(_EGLSurface *surf, EGLint interval)
 {
-   EGLint bound = GET_CONFIG_ATTRIB(surf->Config, EGL_MAX_SWAP_INTERVAL);
+   EGLint bound = surf->Config->MaxSwapInterval;
    if (interval >= bound) {
       interval = bound;
    }
    else {
-      bound = GET_CONFIG_ATTRIB(surf->Config, EGL_MIN_SWAP_INTERVAL);
+      bound = surf->Config->MinSwapInterval;
       if (interval < bound)
          interval = bound;
    }
@@ -263,14 +263,13 @@ _eglInitSurface(_EGLSurface *surf, _EGLDisplay *dpy, EGLint type,
       return EGL_FALSE;
    }
 
-   if ((GET_CONFIG_ATTRIB(conf, EGL_SURFACE_TYPE) & type) == 0) {
+   if ((conf->SurfaceType & type) == 0) {
       /* The config can't be used to create a surface of this type */
       _eglError(EGL_BAD_CONFIG, func);
       return EGL_FALSE;
    }
 
-   memset(surf, 0, sizeof(_EGLSurface));
-   surf->Resource.Display = dpy;
+   _eglInitResource(&surf->Resource, sizeof(*surf), dpy);
    surf->Type = type;
    surf->Config = conf;
 
@@ -303,24 +302,6 @@ _eglInitSurface(_EGLSurface *surf, _EGLDisplay *dpy, EGLint type,
 }
 
 
-EGLBoolean
-_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
-{
-   /* Drivers have to do the actual buffer swap.  */
-   return EGL_TRUE;
-}
-
-
-EGLBoolean
-_eglCopyBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
-                EGLNativePixmapType target)
-{
-   /* copy surface to native pixmap */
-   /* All implementation burdon for this is in the device driver */
-   return EGL_FALSE;
-}
-
-
 EGLBoolean
 _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
                  EGLint attribute, EGLint *value)
@@ -333,7 +314,7 @@ _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
       *value = surface->Height;
       break;
    case EGL_CONFIG_ID:
-      *value = GET_CONFIG_ATTRIB(surface->Config, EGL_CONFIG_ID);
+      *value = surface->Config->ConfigID;
       break;
    case EGL_LARGEST_PBUFFER:
       *value = surface->LargestPbuffer;
@@ -388,51 +369,6 @@ _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
 }
 
 
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreateWindowSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                        EGLNativeWindowType window, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreatePixmapSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                        EGLNativePixmapType pixmap, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf,
-                         const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-/**
- * Default fallback routine - drivers should usually override this.
- */
-EGLBoolean
-_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf)
-{
-   if (!_eglIsSurfaceBound(surf))
-      free(surf);
-   return EGL_TRUE;
-}
-
-
 /**
  * Default fallback routine - drivers might override this.
  */
@@ -445,7 +381,7 @@ _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
 
    switch (attribute) {
    case EGL_MIPMAP_LEVEL:
-      confval = GET_CONFIG_ATTRIB(surface->Config, EGL_RENDERABLE_TYPE);
+      confval = surface->Config->RenderableType;
       if (!(confval & (EGL_OPENGL_ES_BIT | EGL_OPENGL_ES2_BIT))) {
          err = EGL_BAD_PARAMETER;
          break;
@@ -457,7 +393,7 @@ _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
       case EGL_MULTISAMPLE_RESOLVE_DEFAULT:
          break;
       case EGL_MULTISAMPLE_RESOLVE_BOX:
-         confval = GET_CONFIG_ATTRIB(surface->Config, EGL_SURFACE_TYPE);
+         confval = surface->Config->SurfaceType;
          if (!(confval & EGL_MULTISAMPLE_RESOLVE_BOX_BIT))
             err = EGL_BAD_MATCH;
          break;
@@ -474,7 +410,7 @@ _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
       case EGL_BUFFER_DESTROYED:
          break;
       case EGL_BUFFER_PRESERVED:
-         confval = GET_CONFIG_ATTRIB(surface->Config, EGL_SURFACE_TYPE);
+         confval = surface->Config->SurfaceType;
          if (!(confval & EGL_SWAP_BEHAVIOR_PRESERVED_BIT))
             err = EGL_BAD_MATCH;
          break;
@@ -536,40 +472,6 @@ _eglBindTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
 }
 
 
-EGLBoolean
-_eglReleaseTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surface,
-                    EGLint buffer)
-{
-   /* Just do basic error checking and return success/fail.
-    * Drivers must implement the real stuff.
-    */
-
-   if (surface->Type != EGL_PBUFFER_BIT) {
-      _eglError(EGL_BAD_SURFACE, "eglBindTexImage");
-      return EGL_FALSE;
-   }
-
-   if (surface->TextureFormat == EGL_NO_TEXTURE) {
-      _eglError(EGL_BAD_MATCH, "eglBindTexImage");
-      return EGL_FALSE;
-   }
-
-   if (buffer != EGL_BACK_BUFFER) {
-      _eglError(EGL_BAD_PARAMETER, "eglReleaseTexImage");
-      return EGL_FALSE;
-   }
-
-   if (!surface->BoundToTexture) {
-      _eglError(EGL_BAD_SURFACE, "eglReleaseTexImage");
-      return EGL_FALSE;
-   }
-
-   surface->BoundToTexture = EGL_FALSE;
-
-   return EGL_TRUE;
-}
-
-
 EGLBoolean
 _eglSwapInterval(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
                  EGLint interval)
@@ -577,24 +479,3 @@ _eglSwapInterval(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf,
    _eglClampSwapInterval(surf, interval);
    return EGL_TRUE;
 }
-
-
-#ifdef EGL_VERSION_1_2
-
-/**
- * Example function - drivers should do a proper implementation.
- */
-_EGLSurface *
-_eglCreatePbufferFromClientBuffer(_EGLDriver *drv, _EGLDisplay *dpy,
-                                  EGLenum buftype, EGLClientBuffer buffer,
-                                  _EGLConfig *conf, const EGLint *attrib_list)
-{
-   if (buftype != EGL_OPENVG_IMAGE) {
-      _eglError(EGL_BAD_PARAMETER, "eglCreatePbufferFromClientBuffer");
-      return NULL;
-   }
-
-   return NULL;
-}
-
-#endif /* EGL_VERSION_1_2 */

src/egl/main/eglsurface.h

@@ -51,34 +51,10 @@ _eglInitSurface(_EGLSurface *surf, _EGLDisplay *dpy, EGLint type,
                 _EGLConfig *config, const EGLint *attrib_list);
 
 
-extern EGLBoolean
-_eglSwapBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf);
-
-
-extern EGLBoolean
-_eglCopyBuffers(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLNativePixmapType target);
-
-
 extern EGLBoolean
 _eglQuerySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint attribute, EGLint *value);
 
 
-extern _EGLSurface *
-_eglCreateWindowSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, EGLNativeWindowType window, const EGLint *attrib_list);
-
-
-extern _EGLSurface *
-_eglCreatePixmapSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, EGLNativePixmapType pixmap, const EGLint *attrib_list);
-
-
-extern _EGLSurface *
-_eglCreatePbufferSurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLConfig *conf, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglDestroySurface(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf);
-
-
 extern EGLBoolean
 _eglSurfaceAttrib(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint attribute, EGLint value);
 
@@ -87,45 +63,40 @@ PUBLIC extern EGLBoolean
 _eglBindTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint buffer);
 
 
-extern EGLBoolean
-_eglReleaseTexImage(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint buffer);
-
-
 extern EGLBoolean
 _eglSwapInterval(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSurface *surf, EGLint interval);
 
 
-#ifdef EGL_VERSION_1_2
-
-extern _EGLSurface *
-_eglCreatePbufferFromClientBuffer(_EGLDriver *drv, _EGLDisplay *dpy,
-                                  EGLenum buftype, EGLClientBuffer buffer,
-                                  _EGLConfig *conf, const EGLint *attrib_list);
-
-#endif /* EGL_VERSION_1_2 */
-
-
 /**
- * Return true if there is a context bound to the surface.
- *
- * The binding is considered a reference to the surface.  Drivers should not
- * destroy a surface when it is bound.
+ * Increment reference count for the surface.
  */
-static INLINE EGLBoolean
-_eglIsSurfaceBound(_EGLSurface *surf)
+static INLINE _EGLSurface *
+_eglGetSurface(_EGLSurface *surf)
 {
-   return (surf->CurrentContext != NULL);
+   if (surf)
+      _eglGetResource(&surf->Resource);
+   return surf;
 }
 
 
 /**
- * Link a surface to a display and return the handle of the link.
+ * Decrement reference count for the surface.
+ */
+static INLINE EGLBoolean
+_eglPutSurface(_EGLSurface *surf)
+{
+   return (surf) ? _eglPutResource(&surf->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link a surface to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLSurface
-_eglLinkSurface(_EGLSurface *surf, _EGLDisplay *dpy)
+_eglLinkSurface(_EGLSurface *surf)
 {
-   _eglLinkResource(&surf->Resource, _EGL_RESOURCE_SURFACE, dpy);
+   _eglLinkResource(&surf->Resource, _EGL_RESOURCE_SURFACE);
    return (EGLSurface) surf;
 }
 
@@ -167,18 +138,4 @@ _eglGetSurfaceHandle(_EGLSurface *surf)
 }
 
 
-/**
- * Return true if the surface is linked to a display.
- *
- * The link is considered a reference to the surface (the display is owning the
- * surface).  Drivers should not destroy a surface when it is linked.
- */
-static INLINE EGLBoolean
-_eglIsSurfaceLinked(_EGLSurface *surf)
-{
-   _EGLResource *res = (_EGLResource *) surf;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGLSURFACE_INCLUDED */

src/egl/main/eglsync.c

@@ -50,10 +50,7 @@ _eglInitSync(_EGLSync *sync, _EGLDisplay *dpy, EGLenum type,
        !(type == EGL_SYNC_FENCE_KHR && dpy->Extensions.KHR_fence_sync))
       return _eglError(EGL_BAD_ATTRIBUTE, "eglCreateSyncKHR");
 
-   memset(sync, 0, sizeof(*sync));
-
-   sync->Resource.Display = dpy;
-
+   _eglInitResource(&sync->Resource, sizeof(*sync), dpy);
    sync->Type = type;
    sync->SyncStatus = EGL_UNSIGNALED_KHR;
    sync->SyncCondition = EGL_SYNC_PRIOR_COMMANDS_COMPLETE_KHR;
@@ -66,37 +63,6 @@ _eglInitSync(_EGLSync *sync, _EGLDisplay *dpy, EGLenum type,
 }
 
 
-_EGLSync *
-_eglCreateSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy,
-                  EGLenum type, const EGLint *attrib_list)
-{
-   return NULL;
-}
-
-
-EGLBoolean
-_eglDestroySyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync)
-{
-   return EGL_TRUE;
-}
-
-
-EGLint
-_eglClientWaitSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                      EGLint flags, EGLTimeKHR timeout)
-{
-   return EGL_FALSE;
-}
-
-
-EGLBoolean
-_eglSignalSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                  EGLenum mode)
-{
-   return EGL_FALSE;
-}
-
-
 EGLBoolean
 _eglGetSyncAttribKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
                      EGLint attribute, EGLint *value)

src/egl/main/eglsync.h

@@ -28,38 +28,41 @@ _eglInitSync(_EGLSync *sync, _EGLDisplay *dpy, EGLenum type,
              const EGLint *attrib_list);
 
 
-extern _EGLSync *
-_eglCreateSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy,
-                  EGLenum type, const EGLint *attrib_list);
-
-
-extern EGLBoolean
-_eglDestroySyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync);
-
-
-extern EGLint
-_eglClientWaitSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                      EGLint flags, EGLTimeKHR timeout);
-
-
-extern EGLBoolean
-_eglSignalSyncKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
-                  EGLenum mode);
-
-
 extern EGLBoolean
 _eglGetSyncAttribKHR(_EGLDriver *drv, _EGLDisplay *dpy, _EGLSync *sync,
                      EGLint attribute, EGLint *value);
 
 
 /**
- * Link a sync to a display and return the handle of the link.
+ * Increment reference count for the sync.
+ */
+static INLINE _EGLSync *
+_eglGetSync(_EGLSync *sync)
+{
+   if (sync)
+      _eglGetResource(&sync->Resource);
+   return sync;
+}
+
+
+/**
+ * Decrement reference count for the sync.
+ */
+static INLINE EGLBoolean
+_eglPutSync(_EGLSync *sync)
+{
+   return (sync) ? _eglPutResource(&sync->Resource) : EGL_FALSE;
+}
+
+
+/**
+ * Link a sync to its display and return the handle of the link.
  * The handle can be passed to client directly.
  */
 static INLINE EGLSyncKHR
-_eglLinkSync(_EGLSync *sync, _EGLDisplay *dpy)
+_eglLinkSync(_EGLSync *sync)
 {
-   _eglLinkResource(&sync->Resource, _EGL_RESOURCE_SYNC, dpy);
+   _eglLinkResource(&sync->Resource, _EGL_RESOURCE_SYNC);
    return (EGLSyncKHR) sync;
 }
 
@@ -100,20 +103,6 @@ _eglGetSyncHandle(_EGLSync *sync)
 }
 
 
-/**
- * Return true if the sync is linked to a display.
- *
- * The link is considered a reference to the sync (the display is owning the
- * sync).  Drivers should not destroy a sync when it is linked.
- */
-static INLINE EGLBoolean
-_eglIsSyncLinked(_EGLSync *sync)
-{
-   _EGLResource *res = (_EGLResource *) sync;
-   return (res && _eglIsResourceLinked(res));
-}
-
-
 #endif /* EGL_KHR_reusable_sync */
 
 

src/egl/main/egltypedefs.h

@@ -24,6 +24,8 @@ typedef struct _egl_extensions _EGLExtensions;
 
 typedef struct _egl_image _EGLImage;
 
+typedef struct _egl_image_attribs _EGLImageAttribs;
+
 typedef struct _egl_mode _EGLMode;
 
 typedef struct _egl_resource _EGLResource;

src/gallium/Makefile.template

@@ -40,7 +40,7 @@ depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(SYMLINKS) $(GENERATED_SOURC
 	touch depend
 	$(MKDEP) $(MKDEP_OPTIONS) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(GENERATED_SOURCES) 2> /dev/null
 
-$(PROGS): % : %.o
+$(PROGS): % : %.o $(PROGS_DEPS)
 	$(LD) $(LDFLAGS) $(filter %.o,$^) -o $@ -Wl,--start-group  $(LIBS) -Wl,--end-group
 
 # Emacs tags

src/gallium/SConscript

@@ -1,30 +1,135 @@
-import os
+Import('env')
 
-Import('*')
+# 
+# Auxiliary modules
+#
 
 SConscript('auxiliary/SConscript')
 
-for driver in env['drivers']:
-	SConscript(os.path.join('drivers', driver, 'SConscript'))
+#
+# Drivers
+#
+
+SConscript([
+    'drivers/failover/SConscript', 
+    'drivers/galahad/SConscript',
+    'drivers/identity/SConscript', 
+    'drivers/llvmpipe/SConscript', 
+    'drivers/rbug/SConscript',
+    'drivers/softpipe/SConscript',
+    'drivers/svga/SConscript', 
+    'drivers/trace/SConscript', 
+])
+
+if not env['msvc']:
+    # These drivers do not build on MSVC compilers
+    SConscript([
+        'drivers/i915/SConscript', 
+        'drivers/i965/SConscript',
+        'drivers/r300/SConscript',
+    ])
+
+if env['drm']:
+    # These drivers depend on drm headers
+    if env['drm_radeon']:
+        SConscript([
+            'drivers/r600/SConscript',
+        ])
+    # XXX: nouveau drivers have a tight dependency on libdrm, so to enable
+    # we need some version logic before we enable them. Also, ATM there is
+    # no nouveau target in scons
+    # if env['drm_nouveau']:
+    #     SConscript([
+    #         'drivers/nouveau/SConscript', 
+    #         'drivers/nv50/SConscript', 
+    #         'drivers/nvfx/SConscript',
+    #     ])
+
+#
+# State trackers
+#
 
 # Needed by some state trackers
 SConscript('winsys/sw/null/SConscript')
 
 SConscript('state_trackers/python/SConscript')
-if platform != 'embedded':
-	SConscript('state_trackers/glx/xlib/SConscript')
-	SConscript('state_trackers/dri/SConscript')
-	SConscript('state_trackers/xorg/SConscript')
-	SConscript('state_trackers/egl/SConscript')
-	SConscript('state_trackers/vega/SConscript')
+if env['platform'] != 'embedded':
+    SConscript('state_trackers/vega/SConscript')
 
-if platform == 'windows':
-	SConscript('state_trackers/wgl/SConscript')
+    if env['x11']:
+        SConscript('state_trackers/glx/xlib/SConscript')
+
+    if env['dri']:
+        SConscript('state_trackers/dri/SConscript')
+
+    if env['dri'] and env['xorg']:
+        SConscript('state_trackers/xorg/SConscript')
+
+if env['platform'] == 'windows':
+    SConscript([
+        'state_trackers/egl/SConscript',
+        'state_trackers/wgl/SConscript',
+    ])
+
+#
+# Winsys
+# 
 
 SConscript('winsys/SConscript')
 
-SConscript('targets/SConscript')
+#
+# Targets
+#
 
-if platform != 'embedded':
-	SConscript('tests/unit/SConscript')
-	SConscript('tests/graw/SConscript')
+SConscript([
+    'targets/graw-null/SConscript',
+])
+
+if env['x11']:
+    SConscript([
+        'targets/graw-xlib/SConscript',
+        'targets/libgl-xlib/SConscript',
+    ])
+
+if env['platform'] == 'windows':
+    SConscript([
+        'targets/graw-gdi/SConscript',
+        'targets/libgl-gdi/SConscript',
+        #'egl-gdi/SConscript',
+    ])
+
+if env['dri']:
+    SConscript([
+        'targets/SConscript.dri',
+        'targets/dri-swrast/SConscript',
+        'targets/dri-vmwgfx/SConscript',
+        #'targets/dri-nouveau/SConscript',
+    ])
+    if env['drm_intel']:
+        SConscript([
+            'targets/dri-i915/SConscript',
+            'targets/dri-i965/SConscript',
+        ])
+    if env['drm_radeon']:
+        SConscript([
+            'targets/dri-r300/SConscript',
+            'targets/dri-r600/SConscript',
+        ])
+
+if env['xorg']:
+    SConscript([
+        #'targets/xorg-i915/SConscript',
+        #'targets/xorg-i965/SConscript',
+        #'targets/xorg-nouveau/SConscript',
+        #'targets/xorg-radeon/SConscript',
+        'targets/xorg-vmwgfx/SConscript',
+    ])
+
+
+#
+# Unit tests & tools
+#
+
+if env['platform'] != 'embedded':
+    SConscript('tests/unit/SConscript')
+    SConscript('tests/graw/SConscript')

src/gallium/auxiliary/Makefile

@@ -8,6 +8,7 @@ C_SOURCES = \
 	cso_cache/cso_context.c \
 	cso_cache/cso_hash.c \
 	draw/draw_context.c \
+	draw/draw_fs.c \
 	draw/draw_gs.c \
 	draw/draw_pipe.c \
 	draw/draw_pipe_aaline.c \
@@ -121,6 +122,7 @@ C_SOURCES = \
 	util/u_handle_table.c \
 	util/u_hash.c \
 	util/u_hash_table.c \
+	util/u_index_modify.c \
 	util/u_keymap.c \
 	util/u_linear.c \
 	util/u_linkage.c \
@@ -153,6 +155,7 @@ C_SOURCES = \
 GALLIVM_SOURCES = \
         gallivm/lp_bld_arit.c \
         gallivm/lp_bld_assert.c \
+        gallivm/lp_bld_bitarit.c \
         gallivm/lp_bld_const.c \
         gallivm/lp_bld_conv.c \
         gallivm/lp_bld_debug.c \
@@ -168,10 +171,12 @@ GALLIVM_SOURCES = \
         gallivm/lp_bld_printf.c \
         gallivm/lp_bld_quad.c \
         gallivm/lp_bld_sample.c \
+        gallivm/lp_bld_sample_aos.c \
         gallivm/lp_bld_sample_soa.c \
         gallivm/lp_bld_struct.c \
         gallivm/lp_bld_swizzle.c \
         gallivm/lp_bld_tgsi_aos.c \
+        gallivm/lp_bld_tgsi_info.c \
         gallivm/lp_bld_tgsi_soa.c \
         gallivm/lp_bld_type.c \
         draw/draw_llvm.c \
@@ -206,16 +211,16 @@ include ../Makefile.template
 
 
 indices/u_indices_gen.c: indices/u_indices_gen.py
-	python $< > $@
+	$(PYTHON2) $< > $@
 
 indices/u_unfilled_gen.c: indices/u_unfilled_gen.py
-	python $< > $@
+	$(PYTHON2) $< > $@
 
 util/u_format_srgb.c: util/u_format_srgb.py
-	python $< > $@
+	$(PYTHON2) $< > $@
 
 util/u_format_table.c: util/u_format_table.py util/u_format_pack.py util/u_format_parse.py util/u_format.csv
-	python util/u_format_table.py util/u_format.csv > $@
+	$(PYTHON2) util/u_format_table.py util/u_format.csv > $@
 
 util/u_half.c: util/u_half.py
-	python util/u_half.py > $@
+	$(PYTHON2) util/u_half.py > $@

src/gallium/auxiliary/SConscript

@@ -7,8 +7,6 @@ env.Append(CPPPATH = [
     'util',
 ])
 
-env.Tool('udis86')
-
 env.CodeGenerate(
     target = 'indices/u_indices_gen.c', 
     script = 'indices/u_indices_gen.py', 
@@ -54,6 +52,7 @@ source = [
     'cso_cache/cso_context.c',
     'cso_cache/cso_hash.c',
     'draw/draw_context.c',
+    'draw/draw_fs.c',
     'draw/draw_gs.c',
     'draw/draw_pipe.c',
     'draw/draw_pipe_aaline.c',
@@ -170,6 +169,7 @@ source = [
     'util/u_handle_table.c',
     'util/u_hash.c',
     'util/u_hash_table.c',
+    'util/u_index_modify.c',
     'util/u_keymap.c',
     'util/u_linear.c',
     'util/u_linkage.c',
@@ -200,36 +200,42 @@ source = [
 ]
 
 if env['llvm']:
+    if env['UDIS86']:
+        env.Append(CPPDEFINES = [('HAVE_UDIS86', '1')])
+
     source += [
-    'gallivm/lp_bld_arit.c',
-    'gallivm/lp_bld_assert.c',
-    'gallivm/lp_bld_const.c',
-    'gallivm/lp_bld_conv.c',
-    'gallivm/lp_bld_debug.c',
-    'gallivm/lp_bld_flow.c',
-    'gallivm/lp_bld_format_aos.c',
-    'gallivm/lp_bld_format_soa.c',
-    'gallivm/lp_bld_format_yuv.c',
-    'gallivm/lp_bld_gather.c',
-    'gallivm/lp_bld_init.c',
-    'gallivm/lp_bld_intr.c',
-    'gallivm/lp_bld_logic.c',
-    'gallivm/lp_bld_misc.cpp',
-    'gallivm/lp_bld_pack.c',
-    'gallivm/lp_bld_printf.c',
-    'gallivm/lp_bld_quad.c',
-    'gallivm/lp_bld_sample.c',
-    'gallivm/lp_bld_sample_soa.c',
-    'gallivm/lp_bld_struct.c',
-    'gallivm/lp_bld_swizzle.c',
-    'gallivm/lp_bld_tgsi_aos.c',
-    'gallivm/lp_bld_tgsi_soa.c',
-    'gallivm/lp_bld_type.c',
-    'draw/draw_llvm.c',
-    'draw/draw_llvm_sample.c',
-    'draw/draw_llvm_translate.c',
-    'draw/draw_pt_fetch_shade_pipeline_llvm.c',
-    'draw/draw_vs_llvm.c'
+        'gallivm/lp_bld_arit.c',
+        'gallivm/lp_bld_assert.c',
+        'gallivm/lp_bld_bitarit.c',
+        'gallivm/lp_bld_const.c',
+        'gallivm/lp_bld_conv.c',
+        'gallivm/lp_bld_debug.c',
+        'gallivm/lp_bld_flow.c',
+        'gallivm/lp_bld_format_aos.c',
+        'gallivm/lp_bld_format_soa.c',
+        'gallivm/lp_bld_format_yuv.c',
+        'gallivm/lp_bld_gather.c',
+        'gallivm/lp_bld_init.c',
+        'gallivm/lp_bld_intr.c',
+        'gallivm/lp_bld_logic.c',
+        'gallivm/lp_bld_misc.cpp',
+        'gallivm/lp_bld_pack.c',
+        'gallivm/lp_bld_printf.c',
+        'gallivm/lp_bld_quad.c',
+        'gallivm/lp_bld_sample.c',
+        'gallivm/lp_bld_sample_aos.c',
+        'gallivm/lp_bld_sample_soa.c',
+        'gallivm/lp_bld_struct.c',
+        'gallivm/lp_bld_swizzle.c',
+        'gallivm/lp_bld_tgsi_aos.c',
+        'gallivm/lp_bld_tgsi_info.c',
+        'gallivm/lp_bld_tgsi_soa.c',
+        'gallivm/lp_bld_type.c',
+        'draw/draw_llvm.c',
+        'draw/draw_llvm_sample.c',
+        'draw/draw_llvm_translate.c',
+        'draw/draw_pt_fetch_shade_pipeline_llvm.c',
+        'draw/draw_vs_llvm.c'
     ]
 
 gallium = env.ConvenienceLibrary(
@@ -237,4 +243,6 @@ gallium = env.ConvenienceLibrary(
     source = source,
 )
 
+env.Alias('gallium', gallium)
+
 Export('gallium')

src/gallium/auxiliary/draw/draw_context.c

@@ -335,6 +335,7 @@ draw_set_mapped_constant_buffer(struct draw_context *draw,
    case PIPE_SHADER_VERTEX:
       draw->pt.user.vs_constants[slot] = buffer;
       draw->pt.user.vs_constants_size[slot] = size;
+      draw->pt.user.planes = (float (*) [12][4]) &(draw->plane[0]);
       draw_vs_set_constants(draw, slot, buffer, size);
       break;
    case PIPE_SHADER_GEOMETRY:
@@ -413,6 +414,42 @@ draw_set_force_passthrough( struct draw_context *draw, boolean enable )
 }
 
 
+
+/**
+ * Allocate an extra vertex/geometry shader vertex attribute.
+ * This is used by some of the optional draw module stages such
+ * as wide_point which may need to allocate additional generic/texcoord
+ * attributes.
+ */
+int
+draw_alloc_extra_vertex_attrib(struct draw_context *draw,
+                               uint semantic_name, uint semantic_index)
+{
+   const int num_outputs = draw_current_shader_outputs(draw);
+   const int n = draw->extra_shader_outputs.num;
+
+   assert(n < Elements(draw->extra_shader_outputs.semantic_name));
+
+   draw->extra_shader_outputs.semantic_name[n] = semantic_name;
+   draw->extra_shader_outputs.semantic_index[n] = semantic_index;
+   draw->extra_shader_outputs.slot[n] = num_outputs + n;
+   draw->extra_shader_outputs.num++;
+
+   return draw->extra_shader_outputs.slot[n];
+}
+
+
+/**
+ * Remove all extra vertex attributes that were allocated with
+ * draw_alloc_extra_vertex_attrib().
+ */
+void
+draw_remove_extra_vertex_attribs(struct draw_context *draw)
+{
+   draw->extra_shader_outputs.num = 0;
+}
+
+
 /**
  * Ask the draw module for the location/slot of the given vertex attribute in
  * a post-transformed vertex.
@@ -446,12 +483,12 @@ draw_find_shader_output(const struct draw_context *draw,
          return i;
    }
 
-   /* XXX there may be more than one extra vertex attrib.
-    * For example, simulated gl_FragCoord and gl_PointCoord.
-    */
-   if (draw->extra_shader_outputs.semantic_name == semantic_name &&
-       draw->extra_shader_outputs.semantic_index == semantic_index) {
-      return draw->extra_shader_outputs.slot;
+   /* Search the extra vertex attributes */
+   for (i = 0; i < draw->extra_shader_outputs.num; i++) {
+      if (draw->extra_shader_outputs.semantic_name[i] == semantic_name &&
+          draw->extra_shader_outputs.semantic_index[i] == semantic_index) {
+         return draw->extra_shader_outputs.slot[i];
+      }
    }
 
    return 0;
@@ -470,16 +507,18 @@ draw_find_shader_output(const struct draw_context *draw,
 uint
 draw_num_shader_outputs(const struct draw_context *draw)
 {
-   uint count = draw->vs.vertex_shader->info.num_outputs;
+   uint count;
 
    /* If a geometry shader is present, its outputs go to the
     * driver, else the vertex shader's outputs.
     */
    if (draw->gs.geometry_shader)
       count = draw->gs.geometry_shader->info.num_outputs;
+   else
+      count = draw->vs.vertex_shader->info.num_outputs;
+
+   count += draw->extra_shader_outputs.num;
 
-   if (draw->extra_shader_outputs.slot > 0)
-      count++;
    return count;
 }
 
@@ -671,6 +710,11 @@ draw_set_samplers(struct draw_context *draw,
       draw->samplers[i] = NULL;
 
    draw->num_samplers = num;
+
+#ifdef HAVE_LLVM
+   if (draw->llvm)
+      draw_llvm_set_sampler_state(draw);
+#endif
 }
 
 void
@@ -678,9 +722,9 @@ draw_set_mapped_texture(struct draw_context *draw,
                         unsigned sampler_idx,
                         uint32_t width, uint32_t height, uint32_t depth,
                         uint32_t last_level,
-                        uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
-                        uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
-                        const void *data[DRAW_MAX_TEXTURE_LEVELS])
+                        uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
+                        uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
+                        const void *data[PIPE_MAX_TEXTURE_LEVELS])
 {
 #ifdef HAVE_LLVM
    if(draw->llvm)

src/gallium/auxiliary/draw/draw_context.h

@@ -46,9 +46,9 @@ struct draw_context;
 struct draw_stage;
 struct draw_vertex_shader;
 struct draw_geometry_shader;
+struct draw_fragment_shader;
 struct tgsi_sampler;
 
-#define DRAW_MAX_TEXTURE_LEVELS 13  /* 4K x 4K for now */
 
 struct draw_context *draw_create( struct pipe_context *pipe );
 
@@ -119,9 +119,9 @@ draw_set_mapped_texture(struct draw_context *draw,
                         unsigned sampler_idx,
                         uint32_t width, uint32_t height, uint32_t depth,
                         uint32_t last_level,
-                        uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
-                        uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
-                        const void *data[DRAW_MAX_TEXTURE_LEVELS]);
+                        uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
+                        uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
+                        const void *data[PIPE_MAX_TEXTURE_LEVELS]);
 
 
 /*
@@ -137,6 +137,17 @@ void draw_delete_vertex_shader(struct draw_context *draw,
                                struct draw_vertex_shader *dvs);
 
 
+/*
+ * Fragment shader functions
+ */
+struct draw_fragment_shader *
+draw_create_fragment_shader(struct draw_context *draw,
+                            const struct pipe_shader_state *shader);
+void draw_bind_fragment_shader(struct draw_context *draw,
+                               struct draw_fragment_shader *dvs);
+void draw_delete_fragment_shader(struct draw_context *draw,
+                                 struct draw_fragment_shader *dvs);
+
 /*
  * Geometry shader functions
  */

src/gallium/auxiliary/draw/draw_fs.c

@@ -0,0 +1,73 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * 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 VMWARE 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 "pipe/p_shader_tokens.h"
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "util/u_prim.h"
+
+#include "tgsi/tgsi_parse.h"
+
+#include "draw_fs.h"
+#include "draw_private.h"
+#include "draw_context.h"
+
+
+struct draw_fragment_shader *
+draw_create_fragment_shader(struct draw_context *draw,
+                            const struct pipe_shader_state *shader)
+{
+   struct draw_fragment_shader *dfs;
+
+   dfs = CALLOC_STRUCT(draw_fragment_shader);
+   if (dfs) {
+      dfs->base = *shader;
+      tgsi_scan_shader(shader->tokens, &dfs->info);
+   }
+
+   return dfs;
+}
+
+
+void
+draw_bind_fragment_shader(struct draw_context *draw,
+                          struct draw_fragment_shader *dfs)
+{
+   draw_do_flush(draw, DRAW_FLUSH_STATE_CHANGE);
+
+   draw->fs.fragment_shader = dfs;
+}
+
+
+void
+draw_delete_fragment_shader(struct draw_context *draw,
+                            struct draw_fragment_shader *dfs)
+{
+   FREE(dfs);
+}
+

src/gallium/auxiliary/draw/draw_fs.h

@@ -0,0 +1,42 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * 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 VMWARE 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 DRAW_FS_H
+#define DRAW_FS_H
+
+
+#include "tgsi/tgsi_scan.h"
+
+
+struct draw_fragment_shader
+{
+   struct pipe_shader_state base;
+   struct tgsi_shader_info info;
+};
+
+
+#endif /* DRAW_FS_H */

src/gallium/auxiliary/draw/draw_llvm.c

@@ -31,6 +31,9 @@
 #include "draw_vs.h"
 
 #include "gallivm/lp_bld_arit.h"
+#include "gallivm/lp_bld_logic.h"
+#include "gallivm/lp_bld_const.h"
+#include "gallivm/lp_bld_swizzle.h"
 #include "gallivm/lp_bld_struct.h"
 #include "gallivm/lp_bld_type.h"
 #include "gallivm/lp_bld_flow.h"
@@ -43,7 +46,7 @@
 #include "tgsi/tgsi_exec.h"
 #include "tgsi/tgsi_dump.h"
 
-#include "util/u_cpu_detect.h"
+#include "util/u_math.h"
 #include "util/u_pointer.h"
 #include "util/u_string.h"
 
@@ -71,12 +74,17 @@ init_globals(struct draw_llvm *llvm)
       elem_types[DRAW_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
       elem_types[DRAW_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
       elem_types[DRAW_JIT_TEXTURE_ROW_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS);
+         LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
       elem_types[DRAW_JIT_TEXTURE_IMG_STRIDE] =
-         LLVMArrayType(LLVMInt32Type(), DRAW_MAX_TEXTURE_LEVELS);
+         LLVMArrayType(LLVMInt32Type(), PIPE_MAX_TEXTURE_LEVELS);
       elem_types[DRAW_JIT_TEXTURE_DATA] =
          LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
-                       DRAW_MAX_TEXTURE_LEVELS);
+                       PIPE_MAX_TEXTURE_LEVELS);
+      elem_types[DRAW_JIT_TEXTURE_MIN_LOD] = LLVMFloatType();
+      elem_types[DRAW_JIT_TEXTURE_MAX_LOD] = LLVMFloatType();
+      elem_types[DRAW_JIT_TEXTURE_LOD_BIAS] = LLVMFloatType();
+      elem_types[DRAW_JIT_TEXTURE_BORDER_COLOR] = 
+         LLVMArrayType(LLVMFloatType(), 4);
 
       texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
 
@@ -101,6 +109,18 @@ init_globals(struct draw_llvm *llvm)
       LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, data,
                              llvm->target, texture_type,
                              DRAW_JIT_TEXTURE_DATA);
+      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, min_lod,
+                             llvm->target, texture_type,
+                             DRAW_JIT_TEXTURE_MIN_LOD);
+      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, max_lod,
+                             llvm->target, texture_type,
+                             DRAW_JIT_TEXTURE_MAX_LOD);
+      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, lod_bias,
+                             llvm->target, texture_type,
+                             DRAW_JIT_TEXTURE_LOD_BIAS);
+      LP_CHECK_MEMBER_OFFSET(struct draw_jit_texture, border_color,
+                             llvm->target, texture_type,
+                             DRAW_JIT_TEXTURE_BORDER_COLOR);
       LP_CHECK_STRUCT_SIZE(struct draw_jit_texture,
                            llvm->target, texture_type);
 
@@ -110,12 +130,14 @@ init_globals(struct draw_llvm *llvm)
 
    /* struct draw_jit_context */
    {
-      LLVMTypeRef elem_types[3];
+      LLVMTypeRef elem_types[5];
       LLVMTypeRef context_type;
 
       elem_types[0] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */
-      elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* vs_constants */
-      elem_types[2] = LLVMArrayType(texture_type,
+      elem_types[1] = LLVMPointerType(LLVMFloatType(), 0); /* gs_constants */
+      elem_types[2] = LLVMPointerType(LLVMArrayType(LLVMArrayType(LLVMFloatType(), 4), 12), 0); /* planes */
+      elem_types[3] = LLVMPointerType(LLVMFloatType(), 0); /* viewport */
+      elem_types[4] = LLVMArrayType(texture_type,
                                     PIPE_MAX_VERTEX_SAMPLERS); /* textures */
 
       context_type = LLVMStructType(elem_types, Elements(elem_types), 0);
@@ -124,6 +146,8 @@ init_globals(struct draw_llvm *llvm)
                              llvm->target, context_type, 0);
       LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, gs_constants,
                              llvm->target, context_type, 1);
+      LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, planes,
+                             llvm->target, context_type, 2);
       LP_CHECK_MEMBER_OFFSET(struct draw_jit_context, textures,
                              llvm->target, context_type,
                              DRAW_JIT_CTX_TEXTURES);
@@ -249,13 +273,7 @@ draw_llvm_create(struct draw_context *draw)
          LLVMAddConstantPropagationPass(llvm->pass);
       }
 
-      if(util_cpu_caps.has_sse4_1) {
-         /* FIXME: There is a bug in this pass, whereby the combination of fptosi
-          * and sitofp (necessary for trunc/floor/ceil/round implementation)
-          * somehow becomes invalid code.
-          */
-         LLVMAddInstructionCombiningPass(llvm->pass);
-      }
+      LLVMAddInstructionCombiningPass(llvm->pass);
       LLVMAddGVNPass(llvm->pass);
    } else {
       /* We need at least this pass to prevent the backends to fail in
@@ -403,7 +421,7 @@ generate_fetch(LLVMBuilderRef builder,
                             "instance_divisor");
    }
 
-   /* limit index to min(inex, vb_max_index) */
+   /* limit index to min(index, vb_max_index) */
    cond = LLVMBuildICmp(builder, LLVMIntULE, index, vb_max_index, "");
    index = LLVMBuildSelect(builder, cond, index, vb_max_index, "");
 
@@ -532,19 +550,28 @@ static void
 store_aos(LLVMBuilderRef builder,
           LLVMValueRef io_ptr,
           LLVMValueRef index,
-          LLVMValueRef value)
+          LLVMValueRef value,
+          LLVMValueRef clipmask)
 {
    LLVMValueRef id_ptr = draw_jit_header_id(builder, io_ptr);
    LLVMValueRef data_ptr = draw_jit_header_data(builder, io_ptr);
    LLVMValueRef indices[3];
+   LLVMValueRef val, shift;
 
    indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
    indices[1] = index;
    indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0);
 
-   /* undefined vertex */
-   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(),
-                                        0xffff, 0), id_ptr);
+   /* initialize vertex id:16 = 0xffff, pad:3 = 0, edgeflag:1 = 1 */
+   val = LLVMConstInt(LLVMInt32Type(), 0xffff1, 0); 
+   shift  = LLVMConstInt(LLVMInt32Type(), 12, 0);          
+   val = LLVMBuildShl(builder, val, shift, "");
+   /* add clipmask:12 */   
+   val = LLVMBuildOr(builder, val, clipmask, "");               
+
+   /* store vertex header */
+   LLVMBuildStore(builder, val, id_ptr);
+
 
 #if DEBUG_STORE
    lp_build_printf(builder, "    ---- %p storing attribute %d (io = %p)\n", data_ptr, index, io_ptr);
@@ -599,7 +626,8 @@ store_aos_array(LLVMBuilderRef builder,
                 LLVMValueRef io_ptr,
                 LLVMValueRef aos[NUM_CHANNELS],
                 int attrib,
-                int num_outputs)
+                int num_outputs,
+                LLVMValueRef clipmask)
 {
    LLVMValueRef attr_index = LLVMConstInt(LLVMInt32Type(), attrib, 0);
    LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
@@ -607,7 +635,8 @@ store_aos_array(LLVMBuilderRef builder,
    LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
    LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
    LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr;
-
+   LLVMValueRef clipmask0, clipmask1, clipmask2, clipmask3;
+   
    debug_assert(NUM_CHANNELS == 4);
 
    io0_ptr = LLVMBuildGEP(builder, io_ptr,
@@ -619,21 +648,31 @@ store_aos_array(LLVMBuilderRef builder,
    io3_ptr = LLVMBuildGEP(builder, io_ptr,
                           &ind3, 1, "");
 
-#if DEBUG_STORE
-   lp_build_printf(builder, "   io = %p, indexes[%d, %d, %d, %d]\n",
-                   io_ptr, ind0, ind1, ind2, ind3);
-#endif
+   clipmask0 = LLVMBuildExtractElement(builder, clipmask,
+                                       ind0, "");
+   clipmask1 = LLVMBuildExtractElement(builder, clipmask,
+                                       ind1, "");
+   clipmask2 = LLVMBuildExtractElement(builder, clipmask,
+                                       ind2, "");
+   clipmask3 = LLVMBuildExtractElement(builder, clipmask,
+                                       ind3, "");
 
-   store_aos(builder, io0_ptr, attr_index, aos[0]);
-   store_aos(builder, io1_ptr, attr_index, aos[1]);
-   store_aos(builder, io2_ptr, attr_index, aos[2]);
-   store_aos(builder, io3_ptr, attr_index, aos[3]);
+#if DEBUG_STORE
+   lp_build_printf(builder, "io = %p, indexes[%d, %d, %d, %d]\n, clipmask0 = %x, clipmask1 = %x, clipmask2 = %x, clipmask3 = %x\n",
+                   io_ptr, ind0, ind1, ind2, ind3, clipmask0, clipmask1, clipmask2, clipmask3);
+#endif
+   /* store for each of the 4 vertices */
+   store_aos(builder, io0_ptr, attr_index, aos[0], clipmask0);
+   store_aos(builder, io1_ptr, attr_index, aos[1], clipmask1);
+   store_aos(builder, io2_ptr, attr_index, aos[2], clipmask2);
+   store_aos(builder, io3_ptr, attr_index, aos[3], clipmask3);
 }
 
 static void
 convert_to_aos(LLVMBuilderRef builder,
                LLVMValueRef io,
                LLVMValueRef (*outputs)[NUM_CHANNELS],
+               LLVMValueRef clipmask,
                int num_outputs,
                int max_vertices)
 {
@@ -662,13 +701,305 @@ convert_to_aos(LLVMBuilderRef builder,
                       io,
                       aos,
                       attrib,
-                      num_outputs);
+                      num_outputs,
+                      clipmask);
    }
 #if DEBUG_STORE
    lp_build_printf(builder, "   # storing end\n");
 #endif
 }
 
+/*
+ * Stores original vertex positions in clip coordinates
+ * There is probably a more efficient way to do this, 4 floats at once
+ * rather than extracting each element one by one.
+ */
+static void
+store_clip(LLVMBuilderRef builder,
+           LLVMValueRef io_ptr,           
+           LLVMValueRef (*outputs)[NUM_CHANNELS])
+{
+   LLVMValueRef out[4];
+   LLVMValueRef indices[2]; 
+   LLVMValueRef io0_ptr, io1_ptr, io2_ptr, io3_ptr;
+   LLVMValueRef clip_ptr0, clip_ptr1, clip_ptr2, clip_ptr3;
+   LLVMValueRef clip0_ptr, clip1_ptr, clip2_ptr, clip3_ptr;    
+   LLVMValueRef out0elem, out1elem, out2elem, out3elem;
+   int i;
+
+   LLVMValueRef ind0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   LLVMValueRef ind1 = LLVMConstInt(LLVMInt32Type(), 1, 0);
+   LLVMValueRef ind2 = LLVMConstInt(LLVMInt32Type(), 2, 0);
+   LLVMValueRef ind3 = LLVMConstInt(LLVMInt32Type(), 3, 0);
+   
+   indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indices[1] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   
+   out[0] = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/
+   out[1] = LLVMBuildLoad(builder, outputs[0][1], ""); /*y0 y1 y2 y3*/
+   out[2] = LLVMBuildLoad(builder, outputs[0][2], ""); /*z0 z1 z2 z3*/
+   out[3] = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/  
+
+   io0_ptr = LLVMBuildGEP(builder, io_ptr, &ind0, 1, "");
+   io1_ptr = LLVMBuildGEP(builder, io_ptr, &ind1, 1, "");
+   io2_ptr = LLVMBuildGEP(builder, io_ptr, &ind2, 1, "");
+   io3_ptr = LLVMBuildGEP(builder, io_ptr, &ind3, 1, "");
+
+   clip_ptr0 = draw_jit_header_clip(builder, io0_ptr);
+   clip_ptr1 = draw_jit_header_clip(builder, io1_ptr);
+   clip_ptr2 = draw_jit_header_clip(builder, io2_ptr);
+   clip_ptr3 = draw_jit_header_clip(builder, io3_ptr);
+
+   for (i = 0; i<4; i++){
+      clip0_ptr = LLVMBuildGEP(builder, clip_ptr0,
+                               indices, 2, ""); //x0
+      clip1_ptr = LLVMBuildGEP(builder, clip_ptr1,
+                               indices, 2, ""); //x1
+      clip2_ptr = LLVMBuildGEP(builder, clip_ptr2,
+                               indices, 2, ""); //x2
+      clip3_ptr = LLVMBuildGEP(builder, clip_ptr3,
+                               indices, 2, ""); //x3
+
+      out0elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind0, ""); //x0
+      out1elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind1, ""); //x1
+      out2elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind2, ""); //x2
+      out3elem = LLVMBuildExtractElement(builder, out[i],
+                                         ind3, ""); //x3
+  
+      LLVMBuildStore(builder, out0elem, clip0_ptr);
+      LLVMBuildStore(builder, out1elem, clip1_ptr);
+      LLVMBuildStore(builder, out2elem, clip2_ptr);
+      LLVMBuildStore(builder, out3elem, clip3_ptr);
+
+      indices[1]= LLVMBuildAdd(builder, indices[1], ind1, "");
+   }
+
+}
+
+/* Equivalent of _mm_set1_ps(a)
+ */
+static LLVMValueRef vec4f_from_scalar(LLVMBuilderRef bld,
+				      LLVMValueRef a,
+				      const char *name)
+{
+   LLVMValueRef res = LLVMGetUndef(LLVMVectorType(LLVMFloatType(), 4));
+   int i;
+
+   for(i = 0; i < 4; ++i) {
+      LLVMValueRef index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      res = LLVMBuildInsertElement(bld, res, a, index, i == 3 ? name : "");
+   }
+
+   return res;
+}
+
+/*
+ * Transforms the outputs for viewport mapping
+ */
+static void
+generate_viewport(struct draw_llvm *llvm,
+                  LLVMBuilderRef builder,
+                  LLVMValueRef (*outputs)[NUM_CHANNELS],
+                  LLVMValueRef context_ptr)
+{
+   int i;
+   struct lp_type f32_type = lp_type_float_vec(32);
+   LLVMValueRef out3 = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/   
+   LLVMValueRef const1 = lp_build_const_vec(f32_type, 1.0);       /*1.0 1.0 1.0 1.0*/ 
+   LLVMValueRef vp_ptr = draw_jit_context_viewport(builder, context_ptr);
+
+   /* for 1/w convention*/
+   out3 = LLVMBuildFDiv(builder, const1, out3, "");
+   LLVMBuildStore(builder, out3, outputs[0][3]);
+  
+   /* Viewport Mapping */
+   for (i=0; i<3; i++){
+      LLVMValueRef out = LLVMBuildLoad(builder, outputs[0][i], ""); /*x0 x1 x2 x3*/
+      LLVMValueRef scale;
+      LLVMValueRef trans;
+      LLVMValueRef scale_i;
+      LLVMValueRef trans_i;
+      LLVMValueRef index;
+      
+      index = LLVMConstInt(LLVMInt32Type(), i, 0);
+      scale_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
+
+      index = LLVMConstInt(LLVMInt32Type(), i+4, 0);
+      trans_i = LLVMBuildGEP(builder, vp_ptr, &index, 1, "");
+
+      scale = vec4f_from_scalar(builder, LLVMBuildLoad(builder, scale_i, ""), "scale");
+      trans = vec4f_from_scalar(builder, LLVMBuildLoad(builder, trans_i, ""), "trans");
+
+      /* divide by w */
+      out = LLVMBuildFMul(builder, out, out3, "");
+      /* mult by scale */
+      out = LLVMBuildFMul(builder, out, scale, "");
+      /* add translation */
+      out = LLVMBuildFAdd(builder, out, trans, "");
+
+      /* store transformed outputs */
+      LLVMBuildStore(builder, out, outputs[0][i]);
+   }
+   
+}
+
+
+/*
+ * Returns clipmask as 4xi32 bitmask for the 4 vertices
+ */
+static LLVMValueRef 
+generate_clipmask(LLVMBuilderRef builder,
+                  LLVMValueRef (*outputs)[NUM_CHANNELS],
+                  boolean clip_xy,
+                  boolean clip_z,
+                  boolean clip_user,
+                  boolean clip_halfz,
+                  unsigned nr,
+                  LLVMValueRef context_ptr)
+{
+   LLVMValueRef mask; /* stores the <4xi32> clipmasks */     
+   LLVMValueRef test, temp; 
+   LLVMValueRef zero, shift;
+   LLVMValueRef pos_x, pos_y, pos_z, pos_w;
+   LLVMValueRef plane1, planes, plane_ptr, sum;
+
+   unsigned i;
+
+   struct lp_type f32_type = lp_type_float_vec(32); 
+
+   mask = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+   temp = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+   zero = lp_build_const_vec(f32_type, 0);                    /* 0.0f 0.0f 0.0f 0.0f */
+   shift = lp_build_const_int_vec(lp_type_int_vec(32), 1);    /* 1 1 1 1 */
+
+   /* Assuming position stored at output[0] */
+   pos_x = LLVMBuildLoad(builder, outputs[0][0], ""); /*x0 x1 x2 x3*/
+   pos_y = LLVMBuildLoad(builder, outputs[0][1], ""); /*y0 y1 y2 y3*/
+   pos_z = LLVMBuildLoad(builder, outputs[0][2], ""); /*z0 z1 z2 z3*/
+   pos_w = LLVMBuildLoad(builder, outputs[0][3], ""); /*w0 w1 w2 w3*/   
+
+   /* Cliptest, for hardwired planes */
+   if (clip_xy){
+      /* plane 1 */
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_x , pos_w);
+      temp = shift;
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = test;
+   
+      /* plane 2 */
+      test = LLVMBuildFAdd(builder, pos_x, pos_w, "");
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+   
+      /* plane 3 */
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_y, pos_w);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+
+      /* plane 4 */
+      test = LLVMBuildFAdd(builder, pos_y, pos_w, "");
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+   }
+
+   if (clip_z){
+      temp = lp_build_const_int_vec(lp_type_int_vec(32), 16);
+      if (clip_halfz){
+         /* plane 5 */
+         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, pos_z);
+         test = LLVMBuildAnd(builder, test, temp, ""); 
+         mask = LLVMBuildOr(builder, mask, test, "");
+      }  
+      else{
+         /* plane 5 */
+         test = LLVMBuildFAdd(builder, pos_z, pos_w, "");
+         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, test);
+         test = LLVMBuildAnd(builder, test, temp, ""); 
+         mask = LLVMBuildOr(builder, mask, test, "");
+      }
+      /* plane 6 */
+      test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, pos_z, pos_w);
+      temp = LLVMBuildShl(builder, temp, shift, "");
+      test = LLVMBuildAnd(builder, test, temp, ""); 
+      mask = LLVMBuildOr(builder, mask, test, "");
+   }   
+
+   if (clip_user){
+      LLVMValueRef planes_ptr = draw_jit_context_planes(builder, context_ptr);
+      LLVMValueRef indices[3];
+      temp = lp_build_const_int_vec(lp_type_int_vec(32), 32);
+
+      /* userclip planes */
+      for (i = 6; i < nr; i++) {
+         indices[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         indices[1] = LLVMConstInt(LLVMInt32Type(), i, 0);
+
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_x");
+         planes = vec4f_from_scalar(builder, plane1, "plane4_x");
+         sum = LLVMBuildFMul(builder, planes, pos_x, "");
+
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 1, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_y"); 
+         planes = vec4f_from_scalar(builder, plane1, "plane4_y");
+         test = LLVMBuildFMul(builder, planes, pos_y, "");
+         sum = LLVMBuildFAdd(builder, sum, test, "");
+         
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 2, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_z"); 
+         planes = vec4f_from_scalar(builder, plane1, "plane4_z");
+         test = LLVMBuildFMul(builder, planes, pos_z, "");
+         sum = LLVMBuildFAdd(builder, sum, test, "");
+
+         indices[2] = LLVMConstInt(LLVMInt32Type(), 3, 0);
+         plane_ptr = LLVMBuildGEP(builder, planes_ptr, indices, 3, "");
+         plane1 = LLVMBuildLoad(builder, plane_ptr, "plane_w"); 
+         planes = vec4f_from_scalar(builder, plane1, "plane4_w");
+         test = LLVMBuildFMul(builder, planes, pos_w, "");
+         sum = LLVMBuildFAdd(builder, sum, test, "");
+
+         test = lp_build_compare(builder, f32_type, PIPE_FUNC_GREATER, zero, sum);
+         temp = LLVMBuildShl(builder, temp, shift, "");
+         test = LLVMBuildAnd(builder, test, temp, ""); 
+         mask = LLVMBuildOr(builder, mask, test, "");
+      }
+   }
+   return mask;
+}
+
+/*
+ * Returns boolean if any clipping has occurred
+ * Used zero/non-zero i32 value to represent boolean 
+ */
+static void
+clipmask_bool(LLVMBuilderRef builder, 
+              LLVMValueRef clipmask,
+              LLVMValueRef ret_ptr)
+{
+   LLVMValueRef ret = LLVMBuildLoad(builder, ret_ptr, "");   
+   LLVMValueRef temp;
+   int i;
+
+   for (i=0; i<4; i++){   
+      temp = LLVMBuildExtractElement(builder, clipmask,
+                                     LLVMConstInt(LLVMInt32Type(), i, 0) , "");
+      ret = LLVMBuildOr(builder, ret, temp, "");
+   }
+   
+   LLVMBuildStore(builder, ret, ret_ptr);
+}
+
 static void
 draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 {
@@ -688,7 +1019,12 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    LLVMValueRef outputs[PIPE_MAX_SHADER_OUTPUTS][NUM_CHANNELS];
    void *code;
    struct lp_build_sampler_soa *sampler = 0;
-
+   LLVMValueRef ret, ret_ptr;
+   boolean bypass_viewport = variant->key.bypass_viewport;
+   boolean enable_cliptest = variant->key.clip_xy || 
+                             variant->key.clip_z  ||
+                             variant->key.clip_user;
+   
    arg_types[0] = llvm->context_ptr_type;           /* context */
    arg_types[1] = llvm->vertex_header_ptr_type;     /* vertex_header */
    arg_types[2] = llvm->buffer_ptr_type;            /* vbuffers */
@@ -698,7 +1034,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
    arg_types[6] = llvm->vb_ptr_type;                /* pipe_vertex_buffer's */
    arg_types[7] = LLVMInt32Type();                  /* instance_id */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0);
 
    variant->function = LLVMAddFunction(llvm->module, "draw_llvm_shader", func_type);
    LLVMSetFunctionCallConv(variant->function, LLVMCCallConv);
@@ -738,6 +1074,10 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
 
    step = LLVMConstInt(LLVMInt32Type(), max_vertices, 0);
 
+   /* function will return non-zero i32 value if any clipped vertices */     
+   ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), "");   
+   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr);
+
    /* code generated texture sampling */
    sampler = draw_llvm_sampler_soa_create(
       draw_llvm_variant_key_samplers(&variant->key),
@@ -752,6 +1092,7 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
       LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
       LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } };
       LLVMValueRef io;
+      LLVMValueRef clipmask;   /* holds the clipmask value */
       const LLVMValueRef (*ptr_aos)[NUM_CHANNELS];
 
       io_itr = LLVMBuildSub(builder, lp_loop.counter, start, "");
@@ -788,21 +1129,44 @@ draw_llvm_generate(struct draw_llvm *llvm, struct draw_llvm_variant *variant)
                   context_ptr,
                   sampler);
 
-      convert_to_aos(builder, io, outputs,
+      /* store original positions in clip before further manipulation */
+      store_clip(builder, io, outputs);
+
+      /* do cliptest */
+      if (enable_cliptest){
+         /* allocate clipmask, assign it integer type */
+         clipmask = generate_clipmask(builder, outputs,
+                                      variant->key.clip_xy,
+                                      variant->key.clip_z, 
+                                      variant->key.clip_user,
+                                      variant->key.clip_halfz,
+                                      variant->key.nr_planes,
+                                      context_ptr);
+         /* return clipping boolean value for function */
+         clipmask_bool(builder, clipmask, ret_ptr);
+      }
+      else{
+         clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0);    
+      }
+      
+      /* do viewport mapping */
+      if (!bypass_viewport){
+         generate_viewport(llvm, builder, outputs, context_ptr);
+      }
+
+      /* store clipmask in vertex header and positions in data */
+      convert_to_aos(builder, io, outputs, clipmask,
                      draw->vs.vertex_shader->info.num_outputs,
                      max_vertices);
    }
+
    lp_build_loop_end_cond(builder, end, step, LLVMIntUGE, &lp_loop);
 
    sampler->destroy(sampler);
 
-#ifdef PIPE_ARCH_X86
-   /* Avoid corrupting the FPU stack on 32bit OSes. */
-   lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
-#endif
-
-   LLVMBuildRetVoid(builder);
-
+   ret = LLVMBuildLoad(builder, ret_ptr,"");
+   LLVMBuildRet(builder, ret);
+      
    LLVMDisposeBuilder(builder);
 
    /*
@@ -852,7 +1216,12 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    LLVMValueRef fetch_max;
    void *code;
    struct lp_build_sampler_soa *sampler = 0;
-
+   LLVMValueRef ret, ret_ptr;
+   boolean bypass_viewport = variant->key.bypass_viewport;
+   boolean enable_cliptest = variant->key.clip_xy || 
+                             variant->key.clip_z  ||
+                             variant->key.clip_user;
+   
    arg_types[0] = llvm->context_ptr_type;               /* context */
    arg_types[1] = llvm->vertex_header_ptr_type;         /* vertex_header */
    arg_types[2] = llvm->buffer_ptr_type;                /* vbuffers */
@@ -862,10 +1231,9 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
    arg_types[6] = llvm->vb_ptr_type;                    /* pipe_vertex_buffer's */
    arg_types[7] = LLVMInt32Type();                      /* instance_id */
 
-   func_type = LLVMFunctionType(LLVMVoidType(), arg_types, Elements(arg_types), 0);
+   func_type = LLVMFunctionType(LLVMInt32Type(), arg_types, Elements(arg_types), 0);
 
-   variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts",
-                                            func_type);
+   variant->function_elts = LLVMAddFunction(llvm->module, "draw_llvm_shader_elts", func_type);
    LLVMSetFunctionCallConv(variant->function_elts, LLVMCCallConv);
    for(i = 0; i < Elements(arg_types); ++i)
       if(LLVMGetTypeKind(arg_types[i]) == LLVMPointerTypeKind)
@@ -911,11 +1279,16 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
                             LLVMConstInt(LLVMInt32Type(), 1, 0),
                             "fetch_max");
 
+   /* function returns non-zero i32 value if any clipped vertices */
+   ret_ptr = lp_build_alloca(builder, LLVMInt32Type(), ""); 
+   LLVMBuildStore(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), ret_ptr);
+
    lp_build_loop_begin(builder, LLVMConstInt(LLVMInt32Type(), 0, 0), &lp_loop);
    {
       LLVMValueRef inputs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS];
       LLVMValueRef aos_attribs[PIPE_MAX_SHADER_INPUTS][NUM_CHANNELS] = { { 0 } };
       LLVMValueRef io;
+      LLVMValueRef clipmask;   /* holds the clipmask value */
       const LLVMValueRef (*ptr_aos)[NUM_CHANNELS];
 
       io_itr = lp_loop.counter;
@@ -962,21 +1335,47 @@ draw_llvm_generate_elts(struct draw_llvm *llvm, struct draw_llvm_variant *varian
                   context_ptr,
                   sampler);
 
-      convert_to_aos(builder, io, outputs,
+      /* store original positions in clip before further manipulation */
+      store_clip(builder, io, outputs);
+
+      /* do cliptest */
+      if (enable_cliptest){
+         /* allocate clipmask, assign it integer type */
+         clipmask = generate_clipmask(builder, outputs,
+                                      variant->key.clip_xy,
+                                      variant->key.clip_z, 
+                                      variant->key.clip_user,
+                                      variant->key.clip_halfz,
+                                      variant->key.nr_planes,
+                                      context_ptr);
+         /* return clipping boolean value for function */
+         clipmask_bool(builder, clipmask, ret_ptr);
+      }
+      else{
+         clipmask = lp_build_const_int_vec(lp_type_int_vec(32), 0);
+      }
+      
+      /* do viewport mapping */
+      if (!bypass_viewport){
+         generate_viewport(llvm, builder, outputs, context_ptr);
+      }
+
+      /* store clipmask in vertex header, 
+       * original positions in clip 
+       * and transformed positions in data 
+       */   
+      convert_to_aos(builder, io, outputs, clipmask,
                      draw->vs.vertex_shader->info.num_outputs,
                      max_vertices);
    }
+
    lp_build_loop_end_cond(builder, fetch_count, step, LLVMIntUGE, &lp_loop);
 
    sampler->destroy(sampler);
 
-#ifdef PIPE_ARCH_X86
-   /* Avoid corrupting the FPU stack on 32bit OSes. */
-   lp_build_intrinsic(builder, "llvm.x86.mmx.emms", LLVMVoidType(), NULL, 0);
-#endif
-
-   LLVMBuildRetVoid(builder);
-
+   ret = LLVMBuildLoad(builder, ret_ptr,"");   
+   LLVMBuildRet(builder, ret);
+   
    LLVMDisposeBuilder(builder);
 
    /*
@@ -1020,6 +1419,16 @@ draw_llvm_make_variant_key(struct draw_llvm *llvm, char *store)
     */
    key->nr_vertex_elements = llvm->draw->pt.nr_vertex_elements;
 
+   /* will have to rig this up properly later */
+   key->clip_xy = llvm->draw->clip_xy;
+   key->clip_z = llvm->draw->clip_z;
+   key->clip_user = llvm->draw->clip_user;
+   key->bypass_viewport = llvm->draw->identity_viewport;
+   key->clip_halfz = !llvm->draw->rasterizer->gl_rasterization_rules;
+   key->need_edgeflags = (llvm->draw->vs.edgeflag_output ? TRUE : FALSE);
+   key->nr_planes = llvm->draw->nr_planes;
+   key->pad = 0;
+
    /* All variants of this shader will have the same value for
     * nr_samplers.  Not yet trying to compact away holes in the
     * sampler array.
@@ -1048,9 +1457,9 @@ draw_llvm_set_mapped_texture(struct draw_context *draw,
                              unsigned sampler_idx,
                              uint32_t width, uint32_t height, uint32_t depth,
                              uint32_t last_level,
-                             uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
-                             uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
-                             const void *data[DRAW_MAX_TEXTURE_LEVELS])
+                             uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
+                             uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
+                             const void *data[PIPE_MAX_TEXTURE_LEVELS])
 {
    unsigned j;
    struct draw_jit_texture *jit_tex;
@@ -1072,6 +1481,25 @@ draw_llvm_set_mapped_texture(struct draw_context *draw,
    }
 }
 
+
+void
+draw_llvm_set_sampler_state(struct draw_context *draw)
+{
+   unsigned i;
+
+   for (i = 0; i < draw->num_samplers; i++) {
+      struct draw_jit_texture *jit_tex = &draw->llvm->jit_context.textures[i];
+
+      if (draw->samplers[i]) {
+         jit_tex->min_lod = draw->samplers[i]->min_lod;
+         jit_tex->max_lod = draw->samplers[i]->max_lod;
+         jit_tex->lod_bias = draw->samplers[i]->lod_bias;
+         COPY_4V(jit_tex->border_color, draw->samplers[i]->border_color);
+      }
+   }
+}
+
+
 void
 draw_llvm_destroy_variant(struct draw_llvm_variant *variant)
 {

src/gallium/auxiliary/draw/draw_llvm.h

@@ -41,7 +41,6 @@
 #include <llvm-c/Target.h>
 #include <llvm-c/ExecutionEngine.h>
 
-#define DRAW_MAX_TEXTURE_LEVELS 13  /* 4K x 4K for now */
 
 struct draw_llvm;
 struct llvm_vertex_shader;
@@ -52,9 +51,13 @@ struct draw_jit_texture
    uint32_t height;
    uint32_t depth;
    uint32_t last_level;
-   uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS];
-   uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS];
-   const void *data[DRAW_MAX_TEXTURE_LEVELS];
+   uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS];
+   uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS];
+   const void *data[PIPE_MAX_TEXTURE_LEVELS];
+   float min_lod;
+   float max_lod;
+   float lod_bias;
+   float border_color[4];
 };
 
 enum {
@@ -65,6 +68,10 @@ enum {
    DRAW_JIT_TEXTURE_ROW_STRIDE,
    DRAW_JIT_TEXTURE_IMG_STRIDE,
    DRAW_JIT_TEXTURE_DATA,
+   DRAW_JIT_TEXTURE_MIN_LOD,
+   DRAW_JIT_TEXTURE_MAX_LOD,
+   DRAW_JIT_TEXTURE_LOD_BIAS,
+   DRAW_JIT_TEXTURE_BORDER_COLOR,
    DRAW_JIT_TEXTURE_NUM_FIELDS  /* number of fields above */
 };
 
@@ -89,7 +96,8 @@ struct draw_jit_context
 {
    const float *vs_constants;
    const float *gs_constants;
-
+   float (*planes) [12][4];
+   float *viewport;
 
    struct draw_jit_texture textures[PIPE_MAX_VERTEX_SAMPLERS];
 };
@@ -101,18 +109,22 @@ struct draw_jit_context
 #define draw_jit_context_gs_constants(_builder, _ptr) \
    lp_build_struct_get(_builder, _ptr, 1, "gs_constants")
 
-#define DRAW_JIT_CTX_TEXTURES 2
+#define draw_jit_context_planes(_builder, _ptr) \
+   lp_build_struct_get(_builder, _ptr, 2, "planes")
+
+#define draw_jit_context_viewport(_builder, _ptr) \
+   lp_build_struct_get(_builder, _ptr, 3, "viewport")
+
+#define DRAW_JIT_CTX_TEXTURES 4
 
 #define draw_jit_context_textures(_builder, _ptr) \
    lp_build_struct_get_ptr(_builder, _ptr, DRAW_JIT_CTX_TEXTURES, "textures")
 
-
-
 #define draw_jit_header_id(_builder, _ptr)              \
    lp_build_struct_get_ptr(_builder, _ptr, 0, "id")
 
 #define draw_jit_header_clip(_builder, _ptr) \
-   lp_build_struct_get(_builder, _ptr, 1, "clip")
+   lp_build_struct_get_ptr(_builder, _ptr, 1, "clip")
 
 #define draw_jit_header_data(_builder, _ptr)            \
    lp_build_struct_get_ptr(_builder, _ptr, 2, "data")
@@ -128,7 +140,7 @@ struct draw_jit_context
    lp_build_struct_get(_builder, _ptr, 2, "buffer_offset")
 
 
-typedef void
+typedef int
 (*draw_jit_vert_func)(struct draw_jit_context *context,
                       struct vertex_header *io,
                       const char *vbuffers[PIPE_MAX_ATTRIBS],
@@ -139,7 +151,7 @@ typedef void
                       unsigned instance_id);
 
 
-typedef void
+typedef int
 (*draw_jit_vert_func_elts)(struct draw_jit_context *context,
                            struct vertex_header *io,
                            const char *vbuffers[PIPE_MAX_ATTRIBS],
@@ -151,8 +163,16 @@ typedef void
 
 struct draw_llvm_variant_key
 {
-   unsigned nr_vertex_elements:16;
-   unsigned nr_samplers:16;
+   unsigned nr_vertex_elements:8;
+   unsigned nr_samplers:8;
+   unsigned clip_xy:1;
+   unsigned clip_z:1;
+   unsigned clip_user:1;
+   unsigned clip_halfz:1;
+   unsigned bypass_viewport:1;
+   unsigned need_edgeflags:1;
+   unsigned nr_planes:4;
+   unsigned pad:6;
 
    /* Variable number of vertex elements:
     */
@@ -274,13 +294,16 @@ struct lp_build_sampler_soa *
 draw_llvm_sampler_soa_create(const struct lp_sampler_static_state *static_state,
                              LLVMValueRef context_ptr);
 
+void
+draw_llvm_set_sampler_state(struct draw_context *draw);
+
 void
 draw_llvm_set_mapped_texture(struct draw_context *draw,
                              unsigned sampler_idx,
                              uint32_t width, uint32_t height, uint32_t depth,
                              uint32_t last_level,
-                             uint32_t row_stride[DRAW_MAX_TEXTURE_LEVELS],
-                             uint32_t img_stride[DRAW_MAX_TEXTURE_LEVELS],
-                             const void *data[DRAW_MAX_TEXTURE_LEVELS]);
+                             uint32_t row_stride[PIPE_MAX_TEXTURE_LEVELS],
+                             uint32_t img_stride[PIPE_MAX_TEXTURE_LEVELS],
+                             const void *data[PIPE_MAX_TEXTURE_LEVELS]);
 
 #endif

src/gallium/auxiliary/draw/draw_llvm_sample.c

@@ -146,6 +146,10 @@ DRAW_LLVM_TEXTURE_MEMBER(last_level, DRAW_JIT_TEXTURE_LAST_LEVEL, TRUE)
 DRAW_LLVM_TEXTURE_MEMBER(row_stride, DRAW_JIT_TEXTURE_ROW_STRIDE, FALSE)
 DRAW_LLVM_TEXTURE_MEMBER(img_stride, DRAW_JIT_TEXTURE_IMG_STRIDE, FALSE)
 DRAW_LLVM_TEXTURE_MEMBER(data_ptr,   DRAW_JIT_TEXTURE_DATA, FALSE)
+DRAW_LLVM_TEXTURE_MEMBER(min_lod,    DRAW_JIT_TEXTURE_MIN_LOD, TRUE)
+DRAW_LLVM_TEXTURE_MEMBER(max_lod,    DRAW_JIT_TEXTURE_MAX_LOD, TRUE)
+DRAW_LLVM_TEXTURE_MEMBER(lod_bias,   DRAW_JIT_TEXTURE_LOD_BIAS, TRUE)
+DRAW_LLVM_TEXTURE_MEMBER(border_color, DRAW_JIT_TEXTURE_BORDER_COLOR, FALSE)
 
 
 static void
@@ -207,6 +211,10 @@ draw_llvm_sampler_soa_create(const struct lp_sampler_static_state *static_state,
    sampler->dynamic_state.base.row_stride = draw_llvm_texture_row_stride;
    sampler->dynamic_state.base.img_stride = draw_llvm_texture_img_stride;
    sampler->dynamic_state.base.data_ptr = draw_llvm_texture_data_ptr;
+   sampler->dynamic_state.base.min_lod = draw_llvm_texture_min_lod;
+   sampler->dynamic_state.base.max_lod = draw_llvm_texture_max_lod;
+   sampler->dynamic_state.base.lod_bias = draw_llvm_texture_lod_bias;
+   sampler->dynamic_state.base.border_color = draw_llvm_texture_border_color;
    sampler->dynamic_state.static_state = static_state;
    sampler->dynamic_state.context_ptr = context_ptr;
 

src/gallium/auxiliary/draw/draw_llvm_translate.c

@@ -3,7 +3,6 @@
 
 #include "draw_llvm.h"
 
-#include "gallivm/lp_bld_arit.h"
 #include "gallivm/lp_bld_struct.h"
 #include "gallivm/lp_bld_format.h"
 #include "gallivm/lp_bld_debug.h"

src/gallium/auxiliary/draw/draw_pipe_aaline.c

@@ -688,10 +688,9 @@ aaline_first_line(struct draw_stage *stage, struct prim_header *header)
    aaline->tex_slot = draw_current_shader_outputs(draw);
    aaline->pos_slot = draw_current_shader_position_output(draw);;
 
-   /* advertise the extra post-transformed vertex attribute */
-   draw->extra_shader_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
-   draw->extra_shader_outputs.semantic_index = aaline->fs->generic_attrib;
-   draw->extra_shader_outputs.slot = aaline->tex_slot;
+   /* allocate the extra post-transformed vertex attribute */
+   (void) draw_alloc_extra_vertex_attrib(draw, TGSI_SEMANTIC_GENERIC,
+                                         aaline->fs->generic_attrib);
 
    /* how many samplers? */
    /* we'll use sampler/texture[pstip->sampler_unit] for the stipple */
@@ -744,7 +743,7 @@ aaline_flush(struct draw_stage *stage, unsigned flags)
 
    draw->suspend_flushing = FALSE;
 
-   draw->extra_shader_outputs.slot = 0;
+   draw_remove_extra_vertex_attribs(draw);
 }
 
 

src/gallium/auxiliary/draw/draw_pipe_aapoint.c

@@ -701,9 +701,9 @@ aapoint_first_point(struct draw_stage *stage, struct prim_header *header)
 
    aapoint->pos_slot = draw_current_shader_position_output(draw);
 
-   draw->extra_shader_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
-   draw->extra_shader_outputs.semantic_index = aapoint->fs->generic_attrib;
-   draw->extra_shader_outputs.slot = aapoint->tex_slot;
+   /* allocate the extra post-transformed vertex attribute */
+   (void) draw_alloc_extra_vertex_attrib(draw, TGSI_SEMANTIC_GENERIC,
+                                         aapoint->fs->generic_attrib);
 
    /* find psize slot in post-transform vertex */
    aapoint->psize_slot = -1;
@@ -754,7 +754,7 @@ aapoint_flush(struct draw_stage *stage, unsigned flags)
 
    draw->suspend_flushing = FALSE;
 
-   draw->extra_shader_outputs.slot = 0;
+   draw_remove_extra_vertex_attribs(draw);
 }
 
 

src/gallium/auxiliary/draw/draw_pipe_clip.c

@@ -263,6 +263,8 @@ do_clip_tri( struct draw_stage *stage,
       clipmask &= ~(1<<plane_idx);
 
       assert(n < MAX_CLIPPED_VERTICES);
+      if (n >= MAX_CLIPPED_VERTICES)
+         return;
       inlist[n] = inlist[0]; /* prevent rotation of vertices */
 
       for (i = 1; i <= n; i++) {
@@ -272,16 +274,22 @@ do_clip_tri( struct draw_stage *stage,
 
 	 if (!IS_NEGATIVE(dp_prev)) {
             assert(outcount < MAX_CLIPPED_VERTICES);
+            if (outcount >= MAX_CLIPPED_VERTICES)
+               return;
 	    outlist[outcount++] = vert_prev;
 	 }
 
 	 if (DIFFERENT_SIGNS(dp, dp_prev)) {
 	    struct vertex_header *new_vert;
 
-            assert(tmpnr < MAX_CLIPPED_VERTICES+1);
+            assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
+            if (tmpnr >= MAX_CLIPPED_VERTICES + 1)
+               return;
             new_vert = clipper->stage.tmp[tmpnr++];
 
             assert(outcount < MAX_CLIPPED_VERTICES);
+            if (outcount >= MAX_CLIPPED_VERTICES)
+               return;
 	    outlist[outcount++] = new_vert;
 
 	    if (IS_NEGATIVE(dp)) {
@@ -321,27 +329,32 @@ do_clip_tri( struct draw_stage *stage,
 
    /* If flat-shading, copy provoking vertex color to polygon vertex[0]
     */
-   if (clipper->flat) {
-      if (stage->draw->rasterizer->flatshade_first) {
-         if (inlist[0] != header->v[0]) {
-            assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
-            inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
-            copy_colors(stage, inlist[0], header->v[0]);
+   if (n >= 3) {
+      if (clipper->flat) {
+         if (stage->draw->rasterizer->flatshade_first) {
+            if (inlist[0] != header->v[0]) {
+               assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
+               if (tmpnr >= MAX_CLIPPED_VERTICES + 1)
+                  return;
+               inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
+               copy_colors(stage, inlist[0], header->v[0]);
+            }
+         }
+         else {
+            if (inlist[0] != header->v[2]) {
+               assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
+               if (tmpnr >= MAX_CLIPPED_VERTICES + 1)
+                  return;
+               inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
+               copy_colors(stage, inlist[0], header->v[2]);
+            }
          }
       }
-      else {
-         if (inlist[0] != header->v[2]) {
-            assert(tmpnr < MAX_CLIPPED_VERTICES + 1);
-            inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
-            copy_colors(stage, inlist[0], header->v[2]);
-         }
-      }
-   }
-
-   /* Emit the polygon as triangles to the setup stage:
-    */
-   if (n >= 3)
+      
+      /* Emit the polygon as triangles to the setup stage:
+       */
       emit_poly( stage, inlist, n, header );
+   }
 }
 
 

src/gallium/auxiliary/draw/draw_pipe_validate.c

@@ -172,7 +172,7 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
    wide_lines = (rast->line_width > draw->pipeline.wide_line_threshold
                  && !rast->line_smooth);
 
-   /* drawing large points? */
+   /* drawing large/sprite points (but not AA points)? */
    if (rast->sprite_coord_enable && draw->pipeline.point_sprite)
       wide_points = TRUE;
    else if (rast->point_smooth && draw->pipeline.aapoint)
@@ -207,7 +207,7 @@ static struct draw_stage *validate_pipeline( struct draw_stage *stage )
       precalc_flat = TRUE;
    }
 
-   if (wide_points || rast->sprite_coord_enable) {
+   if (wide_points) {
       draw->pipeline.wide_point->next = next;
       next = draw->pipeline.wide_point;
    }

src/gallium/auxiliary/draw/draw_pipe_wide_point.c

@@ -57,26 +57,24 @@
 #include "util/u_memory.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_shader_tokens.h"
+#include "draw_fs.h"
 #include "draw_vs.h"
 #include "draw_pipe.h"
 
 
 struct widepoint_stage {
-   struct draw_stage stage;
+   struct draw_stage stage;  /**< base class */
 
    float half_point_size;
 
    float xbias;
    float ybias;
 
-   uint texcoord_slot[PIPE_MAX_SHADER_OUTPUTS];
-   uint texcoord_enable[PIPE_MAX_SHADER_OUTPUTS];
-   uint num_texcoords;
-   uint texcoord_mode;
+   /** for automatic texcoord generation/replacement */
+   uint num_texcoord_gen;
+   uint texcoord_gen_slot[PIPE_MAX_SHADER_OUTPUTS];
 
    int psize_slot;
-
-   int point_coord_fs_input;  /**< input for pointcoord */
 };
 
 
@@ -96,30 +94,20 @@ widepoint_stage( struct draw_stage *stage )
 static void set_texcoords(const struct widepoint_stage *wide,
                           struct vertex_header *v, const float tc[4])
 {
+   const struct draw_context *draw = wide->stage.draw;
+   const struct pipe_rasterizer_state *rast = draw->rasterizer;
+   const uint texcoord_mode = rast->sprite_coord_mode;
    uint i;
-   for (i = 0; i < wide->num_texcoords; i++) {
-      if (wide->texcoord_enable[i]) {
-         uint j = wide->texcoord_slot[i];
-         v->data[j][0] = tc[0];
-         if (wide->texcoord_mode == PIPE_SPRITE_COORD_LOWER_LEFT)
-            v->data[j][1] = 1.0f - tc[1];
-         else
-            v->data[j][1] = tc[1];
-         v->data[j][2] = tc[2];
-         v->data[j][3] = tc[3];
-      }
-   }
 
-   if (wide->point_coord_fs_input >= 0) {
-      /* put gl_PointCoord into the extra vertex slot */
-      uint slot = wide->stage.draw->extra_shader_outputs.slot;
+   for (i = 0; i < wide->num_texcoord_gen; i++) {
+      const uint slot = wide->texcoord_gen_slot[i];
       v->data[slot][0] = tc[0];
-      if (wide->texcoord_mode == PIPE_SPRITE_COORD_LOWER_LEFT)
+      if (texcoord_mode == PIPE_SPRITE_COORD_LOWER_LEFT)
          v->data[slot][1] = 1.0f - tc[1];
       else
          v->data[slot][1] = tc[1];
-      v->data[slot][2] = 0.0F;
-      v->data[slot][3] = 1.0F;
+      v->data[slot][2] = tc[2];
+      v->data[slot][3] = tc[3];
    }
 }
 
@@ -201,18 +189,9 @@ static void widepoint_point( struct draw_stage *stage,
 }
 
 
-static int
-find_pntc_input_attrib(struct draw_context *draw)
-{
-   /* Scan the fragment program's input decls to find the pointcoord
-    * attribute.  The xy components will store the point coord.
-    */
-   return 0; /* XXX fix this */
-}
-
-
-static void widepoint_first_point( struct draw_stage *stage, 
-			      struct prim_header *header )
+static void
+widepoint_first_point(struct draw_stage *stage, 
+                      struct prim_header *header)
 {
    struct widepoint_stage *wide = widepoint_stage(stage);
    struct draw_context *draw = stage->draw;
@@ -244,31 +223,49 @@ static void widepoint_first_point( struct draw_stage *stage,
       stage->point = draw_pipe_passthrough_point;
    }
 
+   draw_remove_extra_vertex_attribs(draw);
+
    if (rast->point_quad_rasterization) {
-      /* find vertex shader texcoord outputs */
-      const struct draw_vertex_shader *vs = draw->vs.vertex_shader;
-      uint i, j = 0;
-      wide->texcoord_mode = rast->sprite_coord_mode;
-      for (i = 0; i < vs->info.num_outputs; i++) {
-         if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_GENERIC) {
-            wide->texcoord_slot[j] = i;
-            wide->texcoord_enable[j] = (rast->sprite_coord_enable >> j) & 1;
-            j++;
+      const struct draw_fragment_shader *fs = draw->fs.fragment_shader;
+      uint i;
+
+      wide->num_texcoord_gen = 0;
+
+      /* Loop over fragment shader inputs looking for generic inputs
+       * for which bit 'k' in sprite_coord_enable is set.
+       */
+      for (i = 0; i < fs->info.num_inputs; i++) {
+         if (fs->info.input_semantic_name[i] == TGSI_SEMANTIC_GENERIC) {
+            const int generic_index = fs->info.input_semantic_index[i];
+            /* Note that sprite_coord enable is a bitfield of
+             * PIPE_MAX_SHADER_OUTPUTS bits.
+             */
+            if (generic_index < PIPE_MAX_SHADER_OUTPUTS &&
+                (rast->sprite_coord_enable & (1 << generic_index))) {
+               /* OK, this generic attribute needs to be replaced with a
+                * texcoord (see above).
+                */
+               int slot = draw_find_shader_output(draw,
+                                                  TGSI_SEMANTIC_GENERIC,
+                                                  generic_index);
+
+               if (slot > 0) {
+                  /* there's already a post-vertex shader attribute
+                   * for this fragment shader input attribute.
+                   */
+               }
+               else {
+                  /* need to allocate a new post-vertex shader attribute */
+                  slot = draw_alloc_extra_vertex_attrib(draw,
+                                                        TGSI_SEMANTIC_GENERIC,
+                                                        generic_index);
+               }
+
+               /* add this slot to the texcoord-gen list */
+               wide->texcoord_gen_slot[wide->num_texcoord_gen++] = slot;
+            }
          }
       }
-      wide->num_texcoords = j;
-
-      /* find fragment shader PointCoord input */
-      wide->point_coord_fs_input = find_pntc_input_attrib(draw);
-
-      /* setup extra vp output (point coord implemented as a texcoord) */
-      draw->extra_shader_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
-      draw->extra_shader_outputs.semantic_index = 0;
-      draw->extra_shader_outputs.slot = draw_current_shader_outputs(draw);
-   }
-   else {
-      wide->point_coord_fs_input = -1;
-      draw->extra_shader_outputs.slot = 0;
    }
 
    wide->psize_slot = -1;
@@ -295,7 +292,8 @@ static void widepoint_flush( struct draw_stage *stage, unsigned flags )
 
    stage->point = widepoint_first_point;
    stage->next->flush( stage->next, flags );
-   stage->draw->extra_shader_outputs.slot = 0;
+
+   draw_remove_extra_vertex_attribs(draw);
 
    /* restore original rasterizer state */
    if (draw->rast_handle) {

src/gallium/auxiliary/draw/draw_private.h

@@ -169,6 +169,9 @@ struct draw_context
          unsigned vs_constants_size[PIPE_MAX_CONSTANT_BUFFERS];
          const void *gs_constants[PIPE_MAX_CONSTANT_BUFFERS];
          unsigned gs_constants_size[PIPE_MAX_CONSTANT_BUFFERS];
+         
+         /* pointer to planes */
+         float (*planes)[12][4]; 
       } user;
 
       boolean test_fse;         /* enable FSE even though its not correct (eg for softpipe) */
@@ -250,6 +253,11 @@ struct draw_context
       struct tgsi_sampler **samplers;
    } gs;
 
+   /** Fragment shader state */
+   struct {
+      struct draw_fragment_shader *fragment_shader;
+   } fs;
+
    /** Stream output (vertex feedback) state */
    struct {
       struct pipe_stream_output_state state;
@@ -266,9 +274,10 @@ struct draw_context
    /* If a prim stage introduces new vertex attributes, they'll be stored here
     */
    struct {
-      uint semantic_name;
-      uint semantic_index;
-      int slot;
+      uint num;
+      uint semantic_name[10];
+      uint semantic_index[10];
+      uint slot[10];
    } extra_shader_outputs;
 
    unsigned reduced_prim;
@@ -362,6 +371,11 @@ void draw_gs_destroy( struct draw_context *draw );
 uint draw_current_shader_outputs(const struct draw_context *draw);
 uint draw_current_shader_position_output(const struct draw_context *draw);
 
+int draw_alloc_extra_vertex_attrib(struct draw_context *draw,
+                                   uint semantic_name, uint semantic_index);
+void draw_remove_extra_vertex_attribs(struct draw_context *draw);
+
+
 /*******************************************************************************
  * Vertex processing (was passthrough) code:
  */

src/gallium/auxiliary/draw/draw_pt.c

@@ -287,6 +287,84 @@ draw_print_arrays(struct draw_context *draw, uint prim, int start, uint count)
 }
 
 
+/** Helper code for below */
+#define PRIM_RESTART_LOOP(elements) \
+   do { \
+      for (i = start; i < end; i++) { \
+         if (elements[i] == info->restart_index) { \
+            if (cur_count > 0) { \
+               /* draw elts up to prev pos */ \
+               draw_pt_arrays(draw, prim, cur_start, cur_count); \
+            } \
+            /* begin new prim at next elt */ \
+            cur_start = i + 1; \
+            cur_count = 0; \
+         } \
+         else { \
+            cur_count++; \
+         } \
+      } \
+      if (cur_count > 0) { \
+         draw_pt_arrays(draw, prim, cur_start, cur_count); \
+      } \
+   } while (0)
+
+
+/**
+ * For drawing prims with primitive restart enabled.
+ * Scan for restart indexes and draw the runs of elements/vertices between
+ * the restarts.
+ */
+static void
+draw_pt_arrays_restart(struct draw_context *draw,
+                       const struct pipe_draw_info *info)
+{
+   const unsigned prim = info->mode;
+   const unsigned start = info->start;
+   const unsigned count = info->count;
+   const unsigned end = start + count;
+   unsigned i, cur_start, cur_count;
+
+   assert(info->primitive_restart);
+
+   if (draw->pt.user.elts) {
+      /* indexed prims (draw_elements) */
+      cur_start = start;
+      cur_count = 0;
+
+      switch (draw->pt.user.eltSize) {
+      case 1:
+         {
+            const ubyte *elt_ub = (const ubyte *) draw->pt.user.elts;
+            PRIM_RESTART_LOOP(elt_ub);
+         }
+         break;
+      case 2:
+         {
+            const ushort *elt_us = (const ushort *) draw->pt.user.elts;
+            PRIM_RESTART_LOOP(elt_us);
+         }
+         break;
+      case 4:
+         {
+            const uint *elt_ui = (const uint *) draw->pt.user.elts;
+            PRIM_RESTART_LOOP(elt_ui);
+         }
+         break;
+      default:
+         assert(0 && "bad eltSize in draw_arrays()");
+      }
+   }
+   else {
+      /* Non-indexed prims (draw_arrays).
+       * Primitive restart should have been handled in the state tracker.
+       */
+      draw_pt_arrays(draw, prim, start, count);
+   }
+}
+
+
+
 /**
  * Non-instanced drawing.
  * \sa draw_arrays_instanced
@@ -395,6 +473,12 @@ draw_vbo(struct draw_context *draw,
 
    for (instance = 0; instance < info->instance_count; instance++) {
       draw->instance_id = instance + info->start_instance;
-      draw_pt_arrays(draw, info->mode, info->start, info->count);
+
+      if (info->primitive_restart) {
+         draw_pt_arrays_restart(draw, info);
+      }
+      else {
+         draw_pt_arrays(draw, info->mode, info->start, info->count);
+      }
    }
 }

src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline_llvm.c

@@ -175,6 +175,11 @@ llvm_middle_end_prepare( struct draw_pt_middle_end *middle,
       draw->pt.user.vs_constants[0];
    fpme->llvm->jit_context.gs_constants =
       draw->pt.user.gs_constants[0];
+   fpme->llvm->jit_context.planes =
+      (float (*) [12][4]) draw->pt.user.planes[0];
+   fpme->llvm->jit_context.viewport =
+      (float *)draw->viewport.scale;
+    
 }
 
 
@@ -217,6 +222,7 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
    struct draw_vertex_info gs_vert_info;
    struct draw_vertex_info *vert_info;
    unsigned opt = fpme->opt;
+   unsigned clipped = 0;
 
    llvm_vert_info.count = fetch_info->count;
    llvm_vert_info.vertex_size = fpme->vertex_size;
@@ -230,7 +236,7 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
    }
 
    if (fetch_info->linear)
-      fpme->current_variant->jit_func( &fpme->llvm->jit_context,
+      clipped = fpme->current_variant->jit_func( &fpme->llvm->jit_context,
                                        llvm_vert_info.verts,
                                        (const char **)draw->pt.user.vbuffer,
                                        fetch_info->start,
@@ -239,7 +245,7 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
                                        draw->pt.vertex_buffer,
                                        draw->instance_id);
    else
-      fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
+      clipped = fpme->current_variant->jit_func_elts( &fpme->llvm->jit_context,
                                             llvm_vert_info.verts,
                                             (const char **)draw->pt.user.vbuffer,
                                             fetch_info->elts,
@@ -266,6 +272,9 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
       FREE(vert_info->verts);
       vert_info = &gs_vert_info;
       prim_info = &gs_prim_info;
+
+      clipped = draw_pt_post_vs_run( fpme->post_vs, vert_info );
+
    }
 
    /* stream output needs to be done before clipping */
@@ -273,11 +282,11 @@ llvm_pipeline_generic( struct draw_pt_middle_end *middle,
 		    vert_info,
                     prim_info );
 
-   if (draw_pt_post_vs_run( fpme->post_vs, vert_info )) {
+   if (clipped) {
       opt |= PT_PIPELINE;
    }
 
-   /* Do we need to run the pipeline?
+   /* Do we need to run the pipeline? Now will come here if clipped
     */
    if (opt & PT_PIPELINE) {
       pipeline( fpme,

src/gallium/auxiliary/gallivm/lp_bld_arit.c

@@ -614,17 +614,15 @@ lp_build_div(struct lp_build_context *bld,
 
 
 /**
- * Linear interpolation.
- *
- * This also works for integer values with a few caveats.
+ * Linear interpolation -- without any checks.
  *
  * @sa http://www.stereopsis.com/doubleblend.html
  */
-LLVMValueRef
-lp_build_lerp(struct lp_build_context *bld,
-              LLVMValueRef x,
-              LLVMValueRef v0,
-              LLVMValueRef v1)
+static INLINE LLVMValueRef
+lp_build_lerp_simple(struct lp_build_context *bld,
+                     LLVMValueRef x,
+                     LLVMValueRef v0,
+                     LLVMValueRef v1)
 {
    LLVMValueRef delta;
    LLVMValueRef res;
@@ -639,12 +637,80 @@ lp_build_lerp(struct lp_build_context *bld,
 
    res = lp_build_add(bld, v0, res);
 
-   if(bld->type.fixed)
+   if (bld->type.fixed) {
       /* XXX: This step is necessary for lerping 8bit colors stored on 16bits,
        * but it will be wrong for other uses. Basically we need a more
        * powerful lp_type, capable of further distinguishing the values
        * interpretation from the value storage. */
       res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(bld->type, (1 << bld->type.width/2) - 1), "");
+   }
+
+   return res;
+}
+
+
+/**
+ * Linear interpolation.
+ */
+LLVMValueRef
+lp_build_lerp(struct lp_build_context *bld,
+              LLVMValueRef x,
+              LLVMValueRef v0,
+              LLVMValueRef v1)
+{
+   const struct lp_type type = bld->type;
+   LLVMValueRef res;
+
+   assert(lp_check_value(type, x));
+   assert(lp_check_value(type, v0));
+   assert(lp_check_value(type, v1));
+
+   if (type.norm) {
+      struct lp_type wide_type;
+      struct lp_build_context wide_bld;
+      LLVMValueRef xl, xh, v0l, v0h, v1l, v1h, resl, resh;
+      LLVMValueRef shift;
+
+      assert(type.length >= 2);
+      assert(!type.sign);
+
+      /*
+       * Create a wider type, enough to hold the intermediate result of the
+       * multiplication.
+       */
+      memset(&wide_type, 0, sizeof wide_type);
+      wide_type.fixed  = TRUE;
+      wide_type.width  = type.width*2;
+      wide_type.length = type.length/2;
+
+      lp_build_context_init(&wide_bld, bld->builder, wide_type);
+
+      lp_build_unpack2(bld->builder, type, wide_type, x,  &xl,  &xh);
+      lp_build_unpack2(bld->builder, type, wide_type, v0, &v0l, &v0h);
+      lp_build_unpack2(bld->builder, type, wide_type, v1, &v1l, &v1h);
+
+      /*
+       * Scale x from [0, 255] to [0, 256]
+       */
+
+      shift = lp_build_const_int_vec(wide_type, type.width - 1);
+
+      xl = lp_build_add(&wide_bld, xl,
+                        LLVMBuildAShr(bld->builder, xl, shift, ""));
+      xh = lp_build_add(&wide_bld, xh,
+                        LLVMBuildAShr(bld->builder, xh, shift, ""));
+
+      /*
+       * Lerp both halves.
+       */
+
+      resl = lp_build_lerp_simple(&wide_bld, xl, v0l, v1l);
+      resh = lp_build_lerp_simple(&wide_bld, xh, v0h, v1h);
+
+      res = lp_build_pack2(bld->builder, wide_type, type, resl, resh);
+   } else {
+      res = lp_build_lerp_simple(bld, x, v0, v1);
+   }
 
    return res;
 }
@@ -917,41 +983,134 @@ enum lp_build_round_sse41_mode
 };
 
 
+/**
+ * Helper for SSE4.1's ROUNDxx instructions.
+ *
+ * NOTE: In the SSE4.1's nearest mode, if two values are equally close, the
+ * result is the even value.  That is, rounding 2.5 will be 2.0, and not 3.0.
+ */
 static INLINE LLVMValueRef
 lp_build_round_sse41(struct lp_build_context *bld,
                      LLVMValueRef a,
                      enum lp_build_round_sse41_mode mode)
 {
    const struct lp_type type = bld->type;
-   LLVMTypeRef vec_type = lp_build_vec_type(type);
+   LLVMTypeRef i32t = LLVMInt32Type();
    const char *intrinsic;
+   LLVMValueRef res;
 
    assert(type.floating);
-   assert(type.width*type.length == 128);
+
    assert(lp_check_value(type, a));
    assert(util_cpu_caps.has_sse4_1);
 
-   switch(type.width) {
-   case 32:
-      intrinsic = "llvm.x86.sse41.round.ps";
-      break;
-   case 64:
-      intrinsic = "llvm.x86.sse41.round.pd";
-      break;
-   default:
-      assert(0);
-      return bld->undef;
+   if (type.length == 1) {
+      LLVMTypeRef vec_type;
+      LLVMValueRef undef;
+      LLVMValueRef args[3];
+      LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
+
+      switch(type.width) {
+      case 32:
+         intrinsic = "llvm.x86.sse41.round.ss";
+         break;
+      case 64:
+         intrinsic = "llvm.x86.sse41.round.sd";
+         break;
+      default:
+         assert(0);
+         return bld->undef;
+      }
+
+      vec_type = LLVMVectorType(bld->elem_type, 4);
+
+      undef = LLVMGetUndef(vec_type);
+
+      args[0] = undef;
+      args[1] = LLVMBuildInsertElement(bld->builder, undef, a, index0, "");
+      args[2] = LLVMConstInt(i32t, mode, 0);
+
+      res = lp_build_intrinsic(bld->builder, intrinsic,
+                               vec_type, args, Elements(args));
+
+      res = LLVMBuildExtractElement(bld->builder, res, index0, "");
+   }
+   else {
+      assert(type.width*type.length == 128);
+
+      switch(type.width) {
+      case 32:
+         intrinsic = "llvm.x86.sse41.round.ps";
+         break;
+      case 64:
+         intrinsic = "llvm.x86.sse41.round.pd";
+         break;
+      default:
+         assert(0);
+         return bld->undef;
+      }
+
+      res = lp_build_intrinsic_binary(bld->builder, intrinsic,
+                                      bld->vec_type, a,
+                                      LLVMConstInt(i32t, mode, 0));
    }
 
-   return lp_build_intrinsic_binary(bld->builder, intrinsic, vec_type, a,
-                                    LLVMConstInt(LLVMInt32Type(), mode, 0));
+   return res;
+}
+
+
+static INLINE LLVMValueRef
+lp_build_iround_nearest_sse2(struct lp_build_context *bld,
+                             LLVMValueRef a)
+{
+   const struct lp_type type = bld->type;
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMTypeRef ret_type = lp_build_int_vec_type(type);
+   const char *intrinsic;
+   LLVMValueRef res;
+
+   assert(type.floating);
+   /* using the double precision conversions is a bit more complicated */
+   assert(type.width == 32);
+
+   assert(lp_check_value(type, a));
+   assert(util_cpu_caps.has_sse2);
+
+   /* This is relying on MXCSR rounding mode, which should always be nearest. */
+   if (type.length == 1) {
+      LLVMTypeRef vec_type;
+      LLVMValueRef undef;
+      LLVMValueRef arg;
+      LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
+
+      vec_type = LLVMVectorType(bld->elem_type, 4);
+
+      intrinsic = "llvm.x86.sse.cvtss2si";
+
+      undef = LLVMGetUndef(vec_type);
+
+      arg = LLVMBuildInsertElement(bld->builder, undef, a, index0, "");
+
+      res = lp_build_intrinsic_unary(bld->builder, intrinsic,
+                                     ret_type, arg);
+   }
+   else {
+      assert(type.width*type.length == 128);
+
+      intrinsic = "llvm.x86.sse2.cvtps2dq";
+
+      res = lp_build_intrinsic_unary(bld->builder, intrinsic,
+                                     ret_type, a);
+   }
+
+   return res;
 }
 
 
 /**
- * Return the integer part of a float (vector) value.  The returned value is
- * a float (vector).
- * Ex: trunc(-1.5) = 1.0
+ * Return the integer part of a float (vector) value (== round toward zero).
+ * The returned value is a float (vector).
+ * Ex: trunc(-1.5) = -1.0
  */
 LLVMValueRef
 lp_build_trunc(struct lp_build_context *bld,
@@ -962,8 +1121,10 @@ lp_build_trunc(struct lp_build_context *bld,
    assert(type.floating);
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 && type.width*type.length == 128)
+   if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_TRUNCATE);
+   }
    else {
       LLVMTypeRef vec_type = lp_build_vec_type(type);
       LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
@@ -990,8 +1151,10 @@ lp_build_round(struct lp_build_context *bld,
    assert(type.floating);
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 && type.width*type.length == 128)
+   if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
+   }
    else {
       LLVMTypeRef vec_type = lp_build_vec_type(type);
       LLVMValueRef res;
@@ -1016,8 +1179,10 @@ lp_build_floor(struct lp_build_context *bld,
    assert(type.floating);
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 && type.width*type.length == 128)
+   if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR);
+   }
    else {
       LLVMTypeRef vec_type = lp_build_vec_type(type);
       LLVMValueRef res;
@@ -1042,8 +1207,10 @@ lp_build_ceil(struct lp_build_context *bld,
    assert(type.floating);
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 && type.width*type.length == 128)
+   if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
       return lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
+   }
    else {
       LLVMTypeRef vec_type = lp_build_vec_type(type);
       LLVMValueRef res;
@@ -1068,9 +1235,9 @@ lp_build_fract(struct lp_build_context *bld,
 
 
 /**
- * Return the integer part of a float (vector) value.  The returned value is
- * an integer (vector).
- * Ex: itrunc(-1.5) = 1
+ * Return the integer part of a float (vector) value (== round toward zero).
+ * The returned value is an integer (vector).
+ * Ex: itrunc(-1.5) = -1
  */
 LLVMValueRef
 lp_build_itrunc(struct lp_build_context *bld,
@@ -1097,31 +1264,40 @@ lp_build_iround(struct lp_build_context *bld,
                 LLVMValueRef a)
 {
    const struct lp_type type = bld->type;
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMTypeRef int_vec_type = bld->int_vec_type;
    LLVMValueRef res;
 
    assert(type.floating);
 
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 && type.width*type.length == 128) {
+   if (util_cpu_caps.has_sse2 &&
+       ((type.width == 32) && (type.length == 1 || type.length == 4))) {
+      return lp_build_iround_nearest_sse2(bld, a);
+   }
+   else if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
       res = lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_NEAREST);
    }
    else {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
-      LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
-      LLVMValueRef sign;
       LLVMValueRef half;
 
-      /* get sign bit */
-      sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
-
-      /* sign * 0.5 */
       half = lp_build_const_vec(type, 0.5);
-      half = LLVMBuildBitCast(bld->builder, half, int_vec_type, "");
-      half = LLVMBuildOr(bld->builder, sign, half, "");
-      half = LLVMBuildBitCast(bld->builder, half, vec_type, "");
+
+      if (type.sign) {
+         LLVMTypeRef vec_type = bld->vec_type;
+         LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
+         LLVMValueRef sign;
+
+         /* get sign bit */
+         sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+         sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+
+         /* sign * 0.5 */
+         half = LLVMBuildBitCast(bld->builder, half, int_vec_type, "");
+         half = LLVMBuildOr(bld->builder, sign, half, "");
+         half = LLVMBuildBitCast(bld->builder, half, vec_type, "");
+      }
 
       res = LLVMBuildFAdd(bld->builder, a, half, "");
    }
@@ -1142,37 +1318,42 @@ lp_build_ifloor(struct lp_build_context *bld,
                 LLVMValueRef a)
 {
    const struct lp_type type = bld->type;
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMTypeRef int_vec_type = bld->int_vec_type;
    LLVMValueRef res;
 
    assert(type.floating);
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 && type.width*type.length == 128) {
+   if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
       res = lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_FLOOR);
    }
    else {
-      /* Take the sign bit and add it to 1 constant */
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
-      unsigned mantissa = lp_mantissa(type);
-      LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
-      LLVMValueRef sign;
-      LLVMValueRef offset;
+      res = a;
 
-      /* sign = a < 0 ? ~0 : 0 */
-      sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
-      sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "ifloor.sign");
+      if (type.sign) {
+         /* Take the sign bit and add it to 1 constant */
+         LLVMTypeRef vec_type = bld->vec_type;
+         unsigned mantissa = lp_mantissa(type);
+         LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
+         LLVMValueRef sign;
+         LLVMValueRef offset;
 
-      /* offset = -0.99999(9)f */
-      offset = lp_build_const_vec(type, -(double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa));
-      offset = LLVMConstBitCast(offset, int_vec_type);
+         /* sign = a < 0 ? ~0 : 0 */
+         sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+         sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+         sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "ifloor.sign");
 
-      /* offset = a < 0 ? offset : 0.0f */
-      offset = LLVMBuildAnd(bld->builder, offset, sign, "");
-      offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "ifloor.offset");
+         /* offset = -0.99999(9)f */
+         offset = lp_build_const_vec(type, -(double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa));
+         offset = LLVMConstBitCast(offset, int_vec_type);
 
-      res = LLVMBuildFAdd(bld->builder, a, offset, "ifloor.res");
+         /* offset = a < 0 ? offset : 0.0f */
+         offset = LLVMBuildAnd(bld->builder, offset, sign, "");
+         offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "ifloor.offset");
+
+         res = LLVMBuildFAdd(bld->builder, res, offset, "ifloor.res");
+      }
    }
 
    /* round to nearest (toward zero) */
@@ -1192,35 +1373,39 @@ lp_build_iceil(struct lp_build_context *bld,
                LLVMValueRef a)
 {
    const struct lp_type type = bld->type;
-   LLVMTypeRef int_vec_type = lp_build_int_vec_type(type);
+   LLVMTypeRef int_vec_type = bld->int_vec_type;
    LLVMValueRef res;
 
    assert(type.floating);
    assert(lp_check_value(type, a));
 
-   if (util_cpu_caps.has_sse4_1 && type.width*type.length == 128) {
+   if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
       res = lp_build_round_sse41(bld, a, LP_BUILD_ROUND_SSE41_CEIL);
    }
    else {
-      LLVMTypeRef vec_type = lp_build_vec_type(type);
+      LLVMTypeRef vec_type = bld->vec_type;
       unsigned mantissa = lp_mantissa(type);
-      LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
-      LLVMValueRef sign;
       LLVMValueRef offset;
 
-      /* sign = a < 0 ? 0 : ~0 */
-      sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
-      sign = LLVMBuildAnd(bld->builder, sign, mask, "");
-      sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "iceil.sign");
-      sign = LLVMBuildNot(bld->builder, sign, "iceil.not");
-
       /* offset = 0.99999(9)f */
       offset = lp_build_const_vec(type, (double)(((unsigned long long)1 << mantissa) - 10)/((unsigned long long)1 << mantissa));
-      offset = LLVMConstBitCast(offset, int_vec_type);
 
-      /* offset = a < 0 ? 0.0 : offset */
-      offset = LLVMBuildAnd(bld->builder, offset, sign, "");
-      offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "iceil.offset");
+      if (type.sign) {
+         LLVMValueRef mask = lp_build_const_int_vec(type, (unsigned long long)1 << (type.width - 1));
+         LLVMValueRef sign;
+
+         /* sign = a < 0 ? 0 : ~0 */
+         sign = LLVMBuildBitCast(bld->builder, a, int_vec_type, "");
+         sign = LLVMBuildAnd(bld->builder, sign, mask, "");
+         sign = LLVMBuildAShr(bld->builder, sign, lp_build_const_int_vec(type, type.width - 1), "iceil.sign");
+         sign = LLVMBuildNot(bld->builder, sign, "iceil.not");
+
+         /* offset = a < 0 ? 0.0 : offset */
+         offset = LLVMConstBitCast(offset, int_vec_type);
+         offset = LLVMBuildAnd(bld->builder, offset, sign, "");
+         offset = LLVMBuildBitCast(bld->builder, offset, vec_type, "iceil.offset");
+      }
 
       res = LLVMBuildFAdd(bld->builder, a, offset, "iceil.res");
    }
@@ -1232,6 +1417,46 @@ lp_build_iceil(struct lp_build_context *bld,
 }
 
 
+/**
+ * Combined ifloor() & fract().
+ *
+ * Preferred to calling the functions separately, as it will ensure that the
+ * stratergy (floor() vs ifloor()) that results in less redundant work is used.
+ */
+void
+lp_build_ifloor_fract(struct lp_build_context *bld,
+                      LLVMValueRef a,
+                      LLVMValueRef *out_ipart,
+                      LLVMValueRef *out_fpart)
+{
+   const struct lp_type type = bld->type;
+   LLVMValueRef ipart;
+
+   assert(type.floating);
+   assert(lp_check_value(type, a));
+
+   if (util_cpu_caps.has_sse4_1 &&
+       (type.length == 1 || type.width*type.length == 128)) {
+      /*
+       * floor() is easier.
+       */
+
+      ipart = lp_build_floor(bld, a);
+      *out_fpart = LLVMBuildFSub(bld->builder, a, ipart, "fpart");
+      *out_ipart = LLVMBuildFPToSI(bld->builder, ipart, bld->int_vec_type, "ipart");
+   }
+   else {
+      /*
+       * ifloor() is easier.
+       */
+
+      *out_ipart = lp_build_ifloor(bld, a);
+      ipart = LLVMBuildSIToFP(bld->builder, *out_ipart, bld->vec_type, "ipart");
+      *out_fpart = LLVMBuildFSub(bld->builder, a, ipart, "fpart");
+   }
+}
+
+
 LLVMValueRef
 lp_build_sqrt(struct lp_build_context *bld,
               LLVMValueRef a)
@@ -2040,6 +2265,71 @@ lp_build_exp2(struct lp_build_context *bld,
 }
 
 
+/**
+ * Extract the exponent of a IEEE-754 floating point value.
+ *
+ * Optionally apply an integer bias.
+ *
+ * Result is an integer value with
+ *
+ *   ifloor(log2(x)) + bias
+ */
+LLVMValueRef
+lp_build_extract_exponent(struct lp_build_context *bld,
+                          LLVMValueRef x,
+                          int bias)
+{
+   const struct lp_type type = bld->type;
+   unsigned mantissa = lp_mantissa(type);
+   LLVMValueRef res;
+
+   assert(type.floating);
+
+   assert(lp_check_value(bld->type, x));
+
+   x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, "");
+
+   res = LLVMBuildLShr(bld->builder, x, lp_build_const_int_vec(type, mantissa), "");
+   res = LLVMBuildAnd(bld->builder, res, lp_build_const_int_vec(type, 255), "");
+   res = LLVMBuildSub(bld->builder, res, lp_build_const_int_vec(type, 127 - bias), "");
+
+   return res;
+}
+
+
+/**
+ * Extract the mantissa of the a floating.
+ *
+ * Result is a floating point value with
+ *
+ *   x / floor(log2(x))
+ */
+LLVMValueRef
+lp_build_extract_mantissa(struct lp_build_context *bld,
+                          LLVMValueRef x)
+{
+   const struct lp_type type = bld->type;
+   unsigned mantissa = lp_mantissa(type);
+   LLVMValueRef mantmask = lp_build_const_int_vec(type, (1ULL << mantissa) - 1);
+   LLVMValueRef one = LLVMConstBitCast(bld->one, bld->int_vec_type);
+   LLVMValueRef res;
+
+   assert(lp_check_value(bld->type, x));
+
+   assert(type.floating);
+
+   x = LLVMBuildBitCast(bld->builder, x, bld->int_vec_type, "");
+
+   /* res = x / 2**ipart */
+   res = LLVMBuildAnd(bld->builder, x, mantmask, "");
+   res = LLVMBuildOr(bld->builder, res, one, "");
+   res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+
+   return res;
+}
+
+
+
 /**
  * Minimax polynomial fit of log2(x)/(x - 1), for x in range [1, 2[
  * These coefficients can be generate with
@@ -2159,3 +2449,62 @@ lp_build_log2(struct lp_build_context *bld,
    lp_build_log2_approx(bld, x, NULL, NULL, &res);
    return res;
 }
+
+
+/**
+ * Faster (and less accurate) log2.
+ *
+ *    log2(x) = floor(log2(x)) - 1 + x / 2**floor(log2(x))
+ *
+ * Piece-wise linear approximation, with exact results when x is a
+ * power of two.
+ *
+ * See http://www.flipcode.com/archives/Fast_log_Function.shtml
+ */
+LLVMValueRef
+lp_build_fast_log2(struct lp_build_context *bld,
+                   LLVMValueRef x)
+{
+   LLVMValueRef ipart;
+   LLVMValueRef fpart;
+
+   assert(lp_check_value(bld->type, x));
+
+   assert(bld->type.floating);
+
+   /* ipart = floor(log2(x)) - 1 */
+   ipart = lp_build_extract_exponent(bld, x, -1);
+   ipart = LLVMBuildSIToFP(bld->builder, ipart, bld->vec_type, "");
+
+   /* fpart = x / 2**ipart */
+   fpart = lp_build_extract_mantissa(bld, x);
+
+   /* ipart + fpart */
+   return LLVMBuildFAdd(bld->builder, ipart, fpart, "");
+}
+
+
+/**
+ * Fast implementation of iround(log2(x)).
+ *
+ * Not an approximation -- it should give accurate results all the time.
+ */
+LLVMValueRef
+lp_build_ilog2(struct lp_build_context *bld,
+               LLVMValueRef x)
+{
+   LLVMValueRef sqrt2 = lp_build_const_vec(bld->type, M_SQRT2);
+   LLVMValueRef ipart;
+
+   assert(bld->type.floating);
+
+   assert(lp_check_value(bld->type, x));
+
+   /* x * 2^(0.5)   i.e., add 0.5 to the log2(x) */
+   x = LLVMBuildFMul(bld->builder, x, sqrt2, "");
+
+   /* ipart = floor(log2(x) + 0.5)  */
+   ipart = lp_build_extract_exponent(bld, x, 0);
+
+   return ipart;
+}

src/gallium/auxiliary/gallivm/lp_bld_arit.h

@@ -171,6 +171,12 @@ LLVMValueRef
 lp_build_itrunc(struct lp_build_context *bld,
                 LLVMValueRef a);
 
+void
+lp_build_ifloor_fract(struct lp_build_context *bld,
+                      LLVMValueRef a,
+                      LLVMValueRef *out_ipart,
+                      LLVMValueRef *out_fpart);
+
 LLVMValueRef
 lp_build_sqrt(struct lp_build_context *bld,
               LLVMValueRef a);
@@ -208,10 +214,27 @@ LLVMValueRef
 lp_build_exp2(struct lp_build_context *bld,
               LLVMValueRef a);
 
+LLVMValueRef
+lp_build_extract_exponent(struct lp_build_context *bld,
+                          LLVMValueRef x,
+                          int bias);
+
+LLVMValueRef
+lp_build_extract_mantissa(struct lp_build_context *bld,
+                          LLVMValueRef x);
+
 LLVMValueRef
 lp_build_log2(struct lp_build_context *bld,
               LLVMValueRef a);
 
+LLVMValueRef
+lp_build_fast_log2(struct lp_build_context *bld,
+                   LLVMValueRef a);
+
+LLVMValueRef
+lp_build_ilog2(struct lp_build_context *bld,
+               LLVMValueRef x);
+
 void
 lp_build_exp2_approx(struct lp_build_context *bld,
                      LLVMValueRef x,

src/gallium/auxiliary/gallivm/lp_bld_bitarit.c

@@ -0,0 +1,187 @@
+/**************************************************************************
+ *
+ * Copyright 2010 VMware, Inc.
+ * 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.
+ *
+ **************************************************************************/
+
+
+#include "util/u_debug.h"
+
+#include "lp_bld_type.h"
+#include "lp_bld_debug.h"
+#include "lp_bld_const.h"
+#include "lp_bld_bitarit.h"
+
+
+/**
+ * Return (a | b)
+ */
+LLVMValueRef
+lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
+{
+   const struct lp_type type = bld->type;
+   LLVMValueRef res;
+
+   assert(lp_check_value(type, a));
+   assert(lp_check_value(type, b));
+
+   /* can't do bitwise ops on floating-point values */
+   if (type.floating) {
+      a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, "");
+      b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, "");
+   }
+
+   res = LLVMBuildOr(bld->builder, a, b, "");
+
+   if (type.floating) {
+      res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+   }
+
+   return res;
+}
+
+
+/**
+ * Return (a & b)
+ */
+LLVMValueRef
+lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
+{
+   const struct lp_type type = bld->type;
+   LLVMValueRef res;
+
+   assert(lp_check_value(type, a));
+   assert(lp_check_value(type, b));
+
+   /* can't do bitwise ops on floating-point values */
+   if (type.floating) {
+      a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, "");
+      b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, "");
+   }
+
+   res = LLVMBuildAnd(bld->builder, a, b, "");
+
+   if (type.floating) {
+      res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+   }
+
+   return res;
+}
+
+
+/**
+ * Return (a & ~b)
+ */
+LLVMValueRef
+lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
+{
+   const struct lp_type type = bld->type;
+   LLVMValueRef res;
+
+   assert(lp_check_value(type, a));
+   assert(lp_check_value(type, b));
+
+   /* can't do bitwise ops on floating-point values */
+   if (type.floating) {
+      a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, "");
+      b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, "");
+   }
+
+   res = LLVMBuildNot(bld->builder, b, "");
+   res = LLVMBuildAnd(bld->builder, a, res, "");
+
+   if (type.floating) {
+      res = LLVMBuildBitCast(bld->builder, res, bld->vec_type, "");
+   }
+
+   return res;
+}
+
+
+/**
+ * Shift left.
+ */
+LLVMValueRef
+lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
+{
+   const struct lp_type type = bld->type;
+   LLVMValueRef res;
+
+   assert(!type.floating);
+
+   assert(lp_check_value(type, a));
+   assert(lp_check_value(type, b));
+
+   res = LLVMBuildShl(bld->builder, a, b, "");
+
+   return res;
+}
+
+
+/**
+ * Shift right.
+ */
+LLVMValueRef
+lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
+{
+   const struct lp_type type = bld->type;
+   LLVMValueRef res;
+
+   assert(!type.floating);
+
+   assert(lp_check_value(type, a));
+   assert(lp_check_value(type, b));
+
+   if (type.sign) {
+      res = LLVMBuildAShr(bld->builder, a, b, "");
+   } else {
+      res = LLVMBuildLShr(bld->builder, a, b, "");
+   }
+
+   return res;
+}
+
+
+/**
+ * Shift left with immediate.
+ */
+LLVMValueRef
+lp_build_shl_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm)
+{
+   LLVMValueRef b = lp_build_const_int_vec(bld->type, imm);
+   assert(imm <= bld->type.width);
+   return lp_build_shl(bld, a, b);
+}
+
+
+/**
+ * Shift right with immediate.
+ */
+LLVMValueRef
+lp_build_shr_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm)
+{
+   LLVMValueRef b = lp_build_const_int_vec(bld->type, imm);
+   assert(imm <= bld->type.width);
+   return lp_build_shr(bld, a, b);
+}

src/gallium/auxiliary/gallivm/lp_bld_bitarit.h

@@ -0,0 +1,69 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * 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 VMWARE 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.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Helper bitwise arithmetic functions.
+ *
+ * @author Jose Fonseca <jfonseca@vmware.com>
+ */
+
+
+#ifndef LP_BLD_BITARIT_H
+#define LP_BLD_BITARIT_H
+
+
+#include "gallivm/lp_bld.h"
+
+
+struct lp_type;
+struct lp_build_context;
+
+
+LLVMValueRef
+lp_build_or(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
+
+LLVMValueRef
+lp_build_and(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
+
+LLVMValueRef
+lp_build_andnot(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
+
+LLVMValueRef
+lp_build_shl(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
+
+LLVMValueRef
+lp_build_shr(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
+
+LLVMValueRef
+lp_build_shl_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm);
+
+LLVMValueRef
+lp_build_shr_imm(struct lp_build_context *bld, LLVMValueRef a, unsigned imm);
+
+
+#endif /* !LP_BLD_ARIT_H */

src/gallium/auxiliary/gallivm/lp_bld_conv.c

@@ -63,6 +63,7 @@
 
 #include "util/u_debug.h"
 #include "util/u_math.h"
+#include "util/u_cpu_detect.h"
 
 #include "lp_bld_type.h"
 #include "lp_bld_const.h"
@@ -96,58 +97,104 @@ lp_build_clamped_float_to_unsigned_norm(LLVMBuilderRef builder,
    LLVMTypeRef int_vec_type = lp_build_int_vec_type(src_type);
    LLVMValueRef res;
    unsigned mantissa;
-   unsigned n;
-   unsigned long long ubound;
-   unsigned long long mask;
-   double scale;
-   double bias;
 
    assert(src_type.floating);
+   assert(dst_width <= src_type.width);
+   src_type.sign = FALSE;
 
    mantissa = lp_mantissa(src_type);
 
-   /* We cannot carry more bits than the mantissa */
-   n = MIN2(mantissa, dst_width);
-
-   /* This magic coefficients will make the desired result to appear in the
-    * lowest significant bits of the mantissa.
-    */
-   ubound = ((unsigned long long)1 << n);
-   mask = ubound - 1;
-   scale = (double)mask/ubound;
-   bias = (double)((unsigned long long)1 << (mantissa - n));
-
-   res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
-   res = LLVMBuildFAdd(builder, res, lp_build_const_vec(src_type, bias), "");
-   res = LLVMBuildBitCast(builder, res, int_vec_type, "");
-
-   if(dst_width > n) {
-      int shift = dst_width - n;
-      res = LLVMBuildShl(builder, res, lp_build_const_int_vec(src_type, shift), "");
-
-      /* TODO: Fill in the empty lower bits for additional precision? */
-      /* YES: this fixes progs/trivial/tri-z-eq.c.
-       * Otherwise vertex Z=1.0 values get converted to something like
-       * 0xfffffb00 and the test for equality with 0xffffffff fails.
+   if (dst_width <= mantissa) {
+      /*
+       * Apply magic coefficients that will make the desired result to appear
+       * in the lowest significant bits of the mantissa, with correct rounding.
+       *
+       * This only works if the destination width fits in the mantissa.
        */
-#if 0
-      {
-         LLVMValueRef msb;
-         msb = LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, dst_width - 1), "");
-         msb = LLVMBuildShl(builder, msb, lp_build_const_int_vec(src_type, shift), "");
-         msb = LLVMBuildSub(builder, msb, lp_build_const_int_vec(src_type, 1), "");
-         res = LLVMBuildOr(builder, res, msb, "");
-      }
-#elif 0
-      while(shift > 0) {
-         res = LLVMBuildOr(builder, res, LLVMBuildLShr(builder, res, lp_build_const_int_vec(src_type, n), ""), "");
-         shift -= n;
-         n *= 2;
-      }
-#endif
-   }
-   else
+
+      unsigned long long ubound;
+      unsigned long long mask;
+      double scale;
+      double bias;
+
+      ubound = (1ULL << dst_width);
+      mask = ubound - 1;
+      scale = (double)mask/ubound;
+      bias = (double)(1ULL << (mantissa - dst_width));
+
+      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
+      res = LLVMBuildFAdd(builder, res, lp_build_const_vec(src_type, bias), "");
+      res = LLVMBuildBitCast(builder, res, int_vec_type, "");
       res = LLVMBuildAnd(builder, res, lp_build_const_int_vec(src_type, mask), "");
+   }
+   else if (dst_width == (mantissa + 1)) {
+      /*
+       * The destination width matches exactly what can be represented in
+       * floating point (i.e., mantissa + 1 bits). So do a straight
+       * multiplication followed by casting. No further rounding is necessary.
+       */
+
+      double scale;
+
+      scale = (double)((1ULL << dst_width) - 1);
+
+      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
+      res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
+   }
+   else {
+      /*
+       * The destination exceeds what can be represented in the floating point.
+       * So multiply by the largest power two we get away with, and when
+       * subtract the most significant bit to rescale to normalized values.
+       *
+       * The largest power of two factor we can get away is
+       * (1 << (src_type.width - 1)), because we need to use signed . In theory it
+       * should be (1 << (src_type.width - 2)), but IEEE 754 rules states
+       * INT_MIN should be returned in FPToSI, which is the correct result for
+       * values near 1.0!
+       *
+       * This means we get (src_type.width - 1) correct bits for values near 0.0,
+       * and (mantissa + 1) correct bits for values near 1.0. Equally or more
+       * important, we also get exact results for 0.0 and 1.0.
+       */
+
+      unsigned n = MIN2(src_type.width - 1, dst_width);
+
+      double scale = (double)(1ULL << n);
+      unsigned lshift = dst_width - n;
+      unsigned rshift = n;
+      LLVMValueRef lshifted;
+      LLVMValueRef rshifted;
+
+      res = LLVMBuildFMul(builder, src, lp_build_const_vec(src_type, scale), "");
+      res = LLVMBuildFPToSI(builder, res, int_vec_type, "");
+
+      /*
+       * Align the most significant bit to its final place.
+       *
+       * This will cause 1.0 to overflow to 0, but the later adjustment will
+       * get it right.
+       */
+      if (lshift) {
+         lshifted = LLVMBuildShl(builder, res,
+                                 lp_build_const_int_vec(src_type, lshift), "");
+      } else {
+         lshifted = res;
+      }
+
+      /*
+       * Align the most significant bit to the right.
+       */
+      rshifted =  LLVMBuildAShr(builder, res,
+                                lp_build_const_int_vec(src_type, rshift), "");
+
+      /*
+       * Subtract the MSB to the LSB, therefore re-scaling from
+       * (1 << dst_width) to ((1 << dst_width) - 1).
+       */
+
+      res = LLVMBuildSub(builder, lshifted, rshifted, "");
+   }
 
    return res;
 }
@@ -177,6 +224,16 @@ lp_build_unsigned_norm_to_float(LLVMBuilderRef builder,
 
    assert(dst_type.floating);
 
+   /* Special-case int8->float, though most cases could be handled
+    * this way:
+    */
+   if (src_width == 8) {
+      scale = 1.0/255.0;
+      res = LLVMBuildSIToFP(builder, src, vec_type, "");
+      res = LLVMBuildFMul(builder, res, lp_build_const_vec(dst_type, scale), "");
+      return res;
+   }
+
    mantissa = lp_mantissa(dst_type);
 
    n = MIN2(mantissa, src_width);
@@ -241,6 +298,87 @@ lp_build_conv(LLVMBuilderRef builder,
    }
    num_tmps = num_srcs;
 
+
+   /* Special case 4x4f --> 1x16ub 
+    */
+   if (src_type.floating == 1 &&
+       src_type.fixed    == 0 &&
+       src_type.sign     == 1 &&
+       src_type.norm     == 0 &&
+       src_type.width    == 32 &&
+       src_type.length   == 4 &&
+
+       dst_type.floating == 0 &&
+       dst_type.fixed    == 0 &&
+       dst_type.sign     == 0 &&
+       dst_type.norm     == 1 &&
+       dst_type.width    == 8 &&
+       dst_type.length   == 16 &&
+
+       util_cpu_caps.has_sse2)
+   {
+      int i;
+
+      for (i = 0; i < num_dsts; i++, src += 4) {
+         struct lp_type int16_type = dst_type;
+         struct lp_type int32_type = dst_type;
+         LLVMValueRef lo, hi;
+         LLVMValueRef src_int0;
+         LLVMValueRef src_int1;
+         LLVMValueRef src_int2;
+         LLVMValueRef src_int3;
+         LLVMTypeRef int16_vec_type;
+         LLVMTypeRef int32_vec_type;
+         LLVMTypeRef src_vec_type;
+         LLVMTypeRef dst_vec_type;
+         LLVMValueRef const_255f;
+         LLVMValueRef a, b, c, d;
+
+         int16_type.width *= 2;
+         int16_type.length /= 2;
+         int16_type.sign = 1;
+
+         int32_type.width *= 4;
+         int32_type.length /= 4;
+         int32_type.sign = 1;
+
+         src_vec_type   = lp_build_vec_type(src_type);
+         dst_vec_type   = lp_build_vec_type(dst_type);
+         int16_vec_type = lp_build_vec_type(int16_type);
+         int32_vec_type = lp_build_vec_type(int32_type);
+
+         const_255f = lp_build_const_vec(src_type, 255.0f);
+
+         a = LLVMBuildFMul(builder, src[0], const_255f, "");
+         b = LLVMBuildFMul(builder, src[1], const_255f, "");
+         c = LLVMBuildFMul(builder, src[2], const_255f, "");
+         d = LLVMBuildFMul(builder, src[3], const_255f, "");
+
+         {
+            struct lp_build_context bld;
+
+            bld.builder = builder;
+            bld.type = src_type;
+            bld.vec_type = src_vec_type;
+            bld.int_elem_type = lp_build_elem_type(int32_type);
+            bld.int_vec_type = int32_vec_type;
+            bld.undef = lp_build_undef(src_type);
+            bld.zero = lp_build_zero(src_type);
+            bld.one = lp_build_one(src_type);
+
+            src_int0 = lp_build_iround(&bld, a);
+            src_int1 = lp_build_iround(&bld, b);
+            src_int2 = lp_build_iround(&bld, c);
+            src_int3 = lp_build_iround(&bld, d);
+         }
+         /* relying on clamping behavior of sse2 intrinsics here */
+         lo = lp_build_pack2(builder, int32_type, int16_type, src_int0, src_int1);
+         hi = lp_build_pack2(builder, int32_type, int16_type, src_int2, src_int3);
+         dst[i] = lp_build_pack2(builder, int16_type, dst_type, lo, hi);
+      }
+      return; 
+   }
+
    /*
     * Clamp if necessary
     */

src/gallium/auxiliary/gallivm/lp_bld_debug.c

@@ -57,6 +57,8 @@ lp_disassemble(const void* func)
 #ifdef HAVE_UDIS86
    ud_t ud_obj;
    uint64_t max_jmp_pc;
+   uint inst_no;
+   boolean emit_addrs = TRUE, emit_line_nos = FALSE;
 
    ud_init(&ud_obj);
 
@@ -76,13 +78,18 @@ lp_disassemble(const void* func)
 
    while (ud_disassemble(&ud_obj)) {
 
+      if (emit_addrs) {
 #ifdef PIPE_ARCH_X86
-      debug_printf("0x%08lx:\t", (unsigned long)ud_insn_off(&ud_obj));
+         debug_printf("0x%08lx:\t", (unsigned long)ud_insn_off(&ud_obj));
 #endif
 #ifdef PIPE_ARCH_X86_64
-      debug_printf("0x%016llx:\t", (unsigned long long)ud_insn_off(&ud_obj));
+         debug_printf("0x%016llx:\t", (unsigned long long)ud_insn_off(&ud_obj));
 #endif
-
+      }
+      else if (emit_line_nos) {
+         debug_printf("%6d:\t", inst_no);
+         inst_no++;
+      }
 #if 0
       debug_printf("%-16s ", ud_insn_hex(&ud_obj));
 #endif
@@ -115,8 +122,10 @@ lp_disassemble(const void* func)
          }
       }
 
-      if ((ud_insn_off(&ud_obj) >= max_jmp_pc && ud_obj.mnemonic == UD_Iret) ||
-           ud_obj.mnemonic == UD_Iinvalid)
+      if (ud_obj.mnemonic == UD_Iinvalid ||
+          (ud_insn_off(&ud_obj) >= max_jmp_pc &&
+           (ud_obj.mnemonic == UD_Iret ||
+            ud_obj.mnemonic == UD_Ijmp)))
          break;
    }
 

src/gallium/auxiliary/gallivm/lp_bld_debug.h

@@ -36,11 +36,12 @@
 #include "util/u_string.h"
 
 
-#define GALLIVM_DEBUG_TGSI      0x1
-#define GALLIVM_DEBUG_IR        0x2
-#define GALLIVM_DEBUG_ASM       0x4
-#define GALLIVM_DEBUG_NO_OPT    0x8
-#define GALLIVM_DEBUG_PERF      0x10
+#define GALLIVM_DEBUG_TGSI          (1 << 0)
+#define GALLIVM_DEBUG_IR            (1 << 1)
+#define GALLIVM_DEBUG_ASM           (1 << 2)
+#define GALLIVM_DEBUG_NO_OPT        (1 << 3)
+#define GALLIVM_DEBUG_PERF          (1 << 4)
+#define GALLIVM_DEBUG_NO_BRILINEAR  (1 << 5)
 
 
 #ifdef DEBUG

src/gallium/auxiliary/gallivm/lp_bld_flow.c

@@ -38,273 +38,15 @@
 #include "lp_bld_flow.h"
 
 
-#define LP_BUILD_FLOW_MAX_VARIABLES 64
-#define LP_BUILD_FLOW_MAX_DEPTH 32
-
 /**
- * Enumeration of all possible flow constructs.
- */
-enum lp_build_flow_construct_kind {
-   LP_BUILD_FLOW_SCOPE,
-   LP_BUILD_FLOW_SKIP,
-   LP_BUILD_FLOW_IF
-};
-
-
-/**
- * Variable declaration scope.
- */
-struct lp_build_flow_scope
-{
-   /** Number of variables declared in this scope */
-   unsigned num_variables;
-};
-
-
-/**
- * Early exit. Useful to skip to the end of a function or block when
- * the execution mask becomes zero or when there is an error condition.
- */
-struct lp_build_flow_skip
-{
-   /** Block to skip to */
-   LLVMBasicBlockRef block;
-
-   /** Number of variables declared at the beginning */
-   unsigned num_variables;
-
-   LLVMValueRef *phi;  /**< array [num_variables] */
-};
-
-
-/**
- * if/else/endif.
- */
-struct lp_build_flow_if
-{
-   unsigned num_variables;
-
-   LLVMValueRef *phi;  /**< array [num_variables] */
-
-   LLVMValueRef condition;
-   LLVMBasicBlockRef entry_block, true_block, false_block, merge_block;
-};
-
-
-/**
- * Union of all possible flow constructs' data
- */
-union lp_build_flow_construct_data
-{
-   struct lp_build_flow_scope scope;
-   struct lp_build_flow_skip skip;
-   struct lp_build_flow_if ifthen;
-};
-
-
-/**
- * Element of the flow construct stack.
- */
-struct lp_build_flow_construct
-{
-   enum lp_build_flow_construct_kind kind;
-   union lp_build_flow_construct_data data;
-};
-
-
-/**
- * All necessary data to generate LLVM control flow constructs.
+ * Insert a new block, right where builder is pointing to.
  *
- * Besides keeping track of the control flow construct themselves we also
- * need to keep track of variables in order to generate SSA Phi values.
- */
-struct lp_build_flow_context
-{
-   LLVMBuilderRef builder;
-
-   /**
-    * Control flow stack.
-    */
-   struct lp_build_flow_construct constructs[LP_BUILD_FLOW_MAX_DEPTH];
-   unsigned num_constructs;
-
-   /**
-    * Variable stack
-    */
-   LLVMValueRef *variables[LP_BUILD_FLOW_MAX_VARIABLES];
-   unsigned num_variables;
-};
-
-
-struct lp_build_flow_context *
-lp_build_flow_create(LLVMBuilderRef builder)
-{
-   struct lp_build_flow_context *flow;
-
-   flow = CALLOC_STRUCT(lp_build_flow_context);
-   if(!flow)
-      return NULL;
-
-   flow->builder = builder;
-
-   return flow;
-}
-
-
-void
-lp_build_flow_destroy(struct lp_build_flow_context *flow)
-{
-   assert(flow->num_constructs == 0);
-   assert(flow->num_variables == 0);
-   FREE(flow);
-}
-
-
-/**
- * Begin/push a new flow control construct, such as a loop, skip block
- * or variable scope.
- */
-static union lp_build_flow_construct_data *
-lp_build_flow_push(struct lp_build_flow_context *flow,
-                   enum lp_build_flow_construct_kind kind)
-{
-   assert(flow->num_constructs < LP_BUILD_FLOW_MAX_DEPTH);
-   if(flow->num_constructs >= LP_BUILD_FLOW_MAX_DEPTH)
-      return NULL;
-
-   flow->constructs[flow->num_constructs].kind = kind;
-   return &flow->constructs[flow->num_constructs++].data;
-}
-
-
-/**
- * Return the current/top flow control construct on the stack.
- * \param kind  the expected type of the top-most construct
- */
-static union lp_build_flow_construct_data *
-lp_build_flow_peek(struct lp_build_flow_context *flow,
-                   enum lp_build_flow_construct_kind kind)
-{
-   assert(flow->num_constructs);
-   if(!flow->num_constructs)
-      return NULL;
-
-   assert(flow->constructs[flow->num_constructs - 1].kind == kind);
-   if(flow->constructs[flow->num_constructs - 1].kind != kind)
-      return NULL;
-
-   return &flow->constructs[flow->num_constructs - 1].data;
-}
-
-
-/**
- * End/pop the current/top flow control construct on the stack.
- * \param kind  the expected type of the top-most construct
- */
-static union lp_build_flow_construct_data *
-lp_build_flow_pop(struct lp_build_flow_context *flow,
-                  enum lp_build_flow_construct_kind kind)
-{
-   assert(flow->num_constructs);
-   if(!flow->num_constructs)
-      return NULL;
-
-   assert(flow->constructs[flow->num_constructs - 1].kind == kind);
-   if(flow->constructs[flow->num_constructs - 1].kind != kind)
-      return NULL;
-
-   return &flow->constructs[--flow->num_constructs].data;
-}
-
-
-/**
- * Begin a variable scope.
+ * This is useful important not only for aesthetic reasons, but also for
+ * performance reasons, as frequently run blocks should be laid out next to
+ * each other and fall-throughs maximized.
  *
+ * See also llvm/lib/Transforms/Scalar/BasicBlockPlacement.cpp.
  *
- */
-void
-lp_build_flow_scope_begin(struct lp_build_flow_context *flow)
-{
-   struct lp_build_flow_scope *scope;
-
-   scope = &lp_build_flow_push(flow, LP_BUILD_FLOW_SCOPE)->scope;
-   if(!scope)
-      return;
-
-   scope->num_variables = 0;
-}
-
-
-/**
- * Declare a variable.
- *
- * A variable is a named entity which can have different LLVMValueRef's at
- * different points of the program. This is relevant for control flow because
- * when there are multiple branches to a same location we need to replace
- * the variable's value with a Phi function as explained in
- * http://en.wikipedia.org/wiki/Static_single_assignment_form .
- *
- * We keep track of variables by keeping around a pointer to where they're
- * current.
- *
- * There are a few cautions to observe:
- *
- * - Variable's value must not be NULL. If there is no initial value then
- *   LLVMGetUndef() should be used.
- *
- * - Variable's value must be kept up-to-date. If the variable is going to be
- *   modified by a function then a pointer should be passed so that its value
- *   is accurate. Failure to do this will cause some of the variables'
- *   transient values to be lost, leading to wrong results.
- *
- * - A program should be written from top to bottom, by always appending
- *   instructions to the bottom with a single LLVMBuilderRef. Inserting and/or
- *   modifying existing statements will most likely lead to wrong results.
- *
- */
-void
-lp_build_flow_scope_declare(struct lp_build_flow_context *flow,
-                            LLVMValueRef *variable)
-{
-   struct lp_build_flow_scope *scope;
-
-   scope = &lp_build_flow_peek(flow, LP_BUILD_FLOW_SCOPE)->scope;
-   if(!scope)
-      return;
-
-   assert(*variable);
-   if(!*variable)
-      return;
-
-   assert(flow->num_variables < LP_BUILD_FLOW_MAX_VARIABLES);
-   if(flow->num_variables >= LP_BUILD_FLOW_MAX_VARIABLES)
-      return;
-
-   flow->variables[flow->num_variables++] = variable;
-   ++scope->num_variables;
-}
-
-
-void
-lp_build_flow_scope_end(struct lp_build_flow_context *flow)
-{
-   struct lp_build_flow_scope *scope;
-
-   scope = &lp_build_flow_pop(flow, LP_BUILD_FLOW_SCOPE)->scope;
-   if(!scope)
-      return;
-
-   assert(flow->num_variables >= scope->num_variables);
-   if(flow->num_variables < scope->num_variables) {
-      flow->num_variables = 0;
-      return;
-   }
-
-   flow->num_variables -= scope->num_variables;
-}
-
-
-/**
  * Note: this function has no dependencies on the flow code and could
  * be used elsewhere.
  */
@@ -334,52 +76,18 @@ lp_build_insert_new_block(LLVMBuilderRef builder, const char *name)
 }
 
 
-static LLVMBasicBlockRef
-lp_build_flow_insert_block(struct lp_build_flow_context *flow)
-{
-   return lp_build_insert_new_block(flow->builder, "");
-}
-
-
 /**
  * Begin a "skip" block.  Inside this block we can test a condition and
  * skip to the end of the block if the condition is false.
  */
 void
-lp_build_flow_skip_begin(struct lp_build_flow_context *flow)
+lp_build_flow_skip_begin(struct lp_build_skip_context *skip,
+                         LLVMBuilderRef builder)
 {
-   struct lp_build_flow_skip *skip;
-   LLVMBuilderRef builder;
-   unsigned i;
-
-   skip = &lp_build_flow_push(flow, LP_BUILD_FLOW_SKIP)->skip;
-   if(!skip)
-      return;
+   skip->builder = builder;
 
    /* create new basic block */
-   skip->block = lp_build_flow_insert_block(flow);
-
-   skip->num_variables = flow->num_variables;
-   if(!skip->num_variables) {
-      skip->phi = NULL;
-      return;
-   }
-
-   /* Allocate a Phi node for each variable in this skip scope */
-   skip->phi = MALLOC(skip->num_variables * sizeof *skip->phi);
-   if(!skip->phi) {
-      skip->num_variables = 0;
-      return;
-   }
-
-   builder = LLVMCreateBuilder();
-   LLVMPositionBuilderAtEnd(builder, skip->block);
-
-   /* create a Phi node for each variable */
-   for(i = 0; i < skip->num_variables; ++i)
-      skip->phi[i] = LLVMBuildPhi(builder, LLVMTypeOf(*flow->variables[i]), "");
-
-   LLVMDisposeBuilder(builder);
+   skip->block = lp_build_insert_new_block(skip->builder, "skip");
 }
 
 
@@ -388,83 +96,50 @@ lp_build_flow_skip_begin(struct lp_build_flow_context *flow)
  * skip block if the condition is true.
  */
 void
-lp_build_flow_skip_cond_break(struct lp_build_flow_context *flow,
+lp_build_flow_skip_cond_break(struct lp_build_skip_context *skip,
                               LLVMValueRef cond)
 {
-   struct lp_build_flow_skip *skip;
-   LLVMBasicBlockRef current_block;
    LLVMBasicBlockRef new_block;
-   unsigned i;
 
-   skip = &lp_build_flow_peek(flow, LP_BUILD_FLOW_SKIP)->skip;
-   if(!skip)
-      return;
-
-   current_block = LLVMGetInsertBlock(flow->builder);
-
-   new_block = lp_build_flow_insert_block(flow);
-
-   /* for each variable, update the Phi node with a (variable, block) pair */
-   for(i = 0; i < skip->num_variables; ++i) {
-      assert(*flow->variables[i]);
-      assert(LLVMTypeOf(skip->phi[i]) == LLVMTypeOf(*flow->variables[i]));
-      LLVMAddIncoming(skip->phi[i], flow->variables[i], &current_block, 1);
-   }
+   new_block = lp_build_insert_new_block(skip->builder, "");
 
    /* if cond is true, goto skip->block, else goto new_block */
-   LLVMBuildCondBr(flow->builder, cond, skip->block, new_block);
+   LLVMBuildCondBr(skip->builder, cond, skip->block, new_block);
 
-   LLVMPositionBuilderAtEnd(flow->builder, new_block);
+   LLVMPositionBuilderAtEnd(skip->builder, new_block);
 }
 
 
 void
-lp_build_flow_skip_end(struct lp_build_flow_context *flow)
+lp_build_flow_skip_end(struct lp_build_skip_context *skip)
 {
-   struct lp_build_flow_skip *skip;
-   LLVMBasicBlockRef current_block;
-   unsigned i;
-
-   skip = &lp_build_flow_pop(flow, LP_BUILD_FLOW_SKIP)->skip;
-   if(!skip)
-      return;
-
-   current_block = LLVMGetInsertBlock(flow->builder);
-
-   /* add (variable, block) tuples to the phi nodes */
-   for(i = 0; i < skip->num_variables; ++i) {
-      assert(*flow->variables[i]);
-      assert(LLVMTypeOf(skip->phi[i]) == LLVMTypeOf(*flow->variables[i]));
-      LLVMAddIncoming(skip->phi[i], flow->variables[i], &current_block, 1);
-      *flow->variables[i] = skip->phi[i];
-   }
-
    /* goto block */
-   LLVMBuildBr(flow->builder, skip->block);
-   LLVMPositionBuilderAtEnd(flow->builder, skip->block);
-
-   FREE(skip->phi);
+   LLVMBuildBr(skip->builder, skip->block);
+   LLVMPositionBuilderAtEnd(skip->builder, skip->block);
 }
 
 
 /**
  * Check if the mask predicate is zero.  If so, jump to the end of the block.
  */
-static void
+void
 lp_build_mask_check(struct lp_build_mask_context *mask)
 {
-   LLVMBuilderRef builder = mask->flow->builder;
+   LLVMBuilderRef builder = mask->skip.builder;
+   LLVMValueRef value;
    LLVMValueRef cond;
 
+   value = lp_build_mask_value(mask);
+
    /* cond = (mask == 0) */
    cond = LLVMBuildICmp(builder,
                         LLVMIntEQ,
-                        LLVMBuildBitCast(builder, mask->value, mask->reg_type, ""),
+                        LLVMBuildBitCast(builder, value, mask->reg_type, ""),
                         LLVMConstNull(mask->reg_type),
                         "");
 
    /* if cond, goto end of block */
-   lp_build_flow_skip_cond_break(mask->flow, cond);
+   lp_build_flow_skip_cond_break(&mask->skip, cond);
 }
 
 
@@ -477,21 +152,27 @@ lp_build_mask_check(struct lp_build_mask_context *mask)
  */
 void
 lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    struct lp_build_flow_context *flow,
+                    LLVMBuilderRef builder,
                     struct lp_type type,
                     LLVMValueRef value)
 {
    memset(mask, 0, sizeof *mask);
 
-   mask->flow = flow;
    mask->reg_type = LLVMIntType(type.width * type.length);
-   mask->value = value;
+   mask->var = lp_build_alloca(builder,
+                               lp_build_int_vec_type(type),
+                               "execution_mask");
 
-   lp_build_flow_scope_begin(flow);
-   lp_build_flow_scope_declare(flow, &mask->value);
-   lp_build_flow_skip_begin(flow);
+   LLVMBuildStore(builder, value, mask->var);
 
-   lp_build_mask_check(mask);
+   lp_build_flow_skip_begin(&mask->skip, builder);
+}
+
+
+LLVMValueRef
+lp_build_mask_value(struct lp_build_mask_context *mask)
+{
+   return LLVMBuildLoad(mask->skip.builder, mask->var, "");
 }
 
 
@@ -504,9 +185,10 @@ void
 lp_build_mask_update(struct lp_build_mask_context *mask,
                      LLVMValueRef value)
 {
-   mask->value = LLVMBuildAnd( mask->flow->builder, mask->value, value, "");
-
-   lp_build_mask_check(mask);
+   value = LLVMBuildAnd(mask->skip.builder,
+                        lp_build_mask_value(mask),
+                        value, "");
+   LLVMBuildStore(mask->skip.builder, value, mask->var);
 }
 
 
@@ -516,9 +198,8 @@ lp_build_mask_update(struct lp_build_mask_context *mask,
 LLVMValueRef
 lp_build_mask_end(struct lp_build_mask_context *mask)
 {
-   lp_build_flow_skip_end(mask->flow);
-   lp_build_flow_scope_end(mask->flow);
-   return mask->value;
+   lp_build_flow_skip_end(&mask->skip);
+   return lp_build_mask_value(mask);
 }
 
 
@@ -528,19 +209,47 @@ lp_build_loop_begin(LLVMBuilderRef builder,
                     LLVMValueRef start,
                     struct lp_build_loop_state *state)
 {
-   LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
-   LLVMValueRef function = LLVMGetBasicBlockParent(block);
+   state->block = lp_build_insert_new_block(builder, "loop_begin");
 
-   state->block = LLVMAppendBasicBlock(function, "loop");
+   state->counter_var = lp_build_alloca(builder, LLVMTypeOf(start), "loop_counter");
+
+   LLVMBuildStore(builder, start, state->counter_var);
 
    LLVMBuildBr(builder, state->block);
 
    LLVMPositionBuilderAtEnd(builder, state->block);
 
-   state->counter = LLVMBuildPhi(builder, LLVMTypeOf(start), "");
+   state->counter = LLVMBuildLoad(builder, state->counter_var, "");
+}
 
-   LLVMAddIncoming(state->counter, &start, &block, 1);
 
+void
+lp_build_loop_end_cond(LLVMBuilderRef builder,
+                       LLVMValueRef end,
+                       LLVMValueRef step,
+                       LLVMIntPredicate llvm_cond,
+                       struct lp_build_loop_state *state)
+{
+   LLVMValueRef next;
+   LLVMValueRef cond;
+   LLVMBasicBlockRef after_block;
+
+   if (!step)
+      step = LLVMConstInt(LLVMTypeOf(end), 1, 0);
+
+   next = LLVMBuildAdd(builder, state->counter, step, "");
+
+   LLVMBuildStore(builder, next, state->counter_var);
+
+   cond = LLVMBuildICmp(builder, llvm_cond, next, end, "");
+
+   after_block = lp_build_insert_new_block(builder, "loop_end");
+
+   LLVMBuildCondBr(builder, cond, after_block, state->block);
+
+   LLVMPositionBuilderAtEnd(builder, after_block);
+
+   state->counter = LLVMBuildLoad(builder, state->counter_var, "");
 }
 
 
@@ -550,55 +259,7 @@ lp_build_loop_end(LLVMBuilderRef builder,
                   LLVMValueRef step,
                   struct lp_build_loop_state *state)
 {
-   LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
-   LLVMValueRef function = LLVMGetBasicBlockParent(block);
-   LLVMValueRef next;
-   LLVMValueRef cond;
-   LLVMBasicBlockRef after_block;
-
-   if (!step)
-      step = LLVMConstInt(LLVMTypeOf(end), 1, 0);
-
-   next = LLVMBuildAdd(builder, state->counter, step, "");
-
-   cond = LLVMBuildICmp(builder, LLVMIntNE, next, end, "");
-
-   after_block = LLVMAppendBasicBlock(function, "");
-
-   LLVMBuildCondBr(builder, cond, after_block, state->block);
-
-   LLVMAddIncoming(state->counter, &next, &block, 1);
-
-   LLVMPositionBuilderAtEnd(builder, after_block);
-}
-
-void
-lp_build_loop_end_cond(LLVMBuilderRef builder,
-                       LLVMValueRef end,
-                       LLVMValueRef step,
-                       int llvm_cond,
-                       struct lp_build_loop_state *state)
-{
-   LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
-   LLVMValueRef function = LLVMGetBasicBlockParent(block);
-   LLVMValueRef next;
-   LLVMValueRef cond;
-   LLVMBasicBlockRef after_block;
-
-   if (!step)
-      step = LLVMConstInt(LLVMTypeOf(end), 1, 0);
-
-   next = LLVMBuildAdd(builder, state->counter, step, "");
-
-   cond = LLVMBuildICmp(builder, llvm_cond, next, end, "");
-
-   after_block = LLVMAppendBasicBlock(function, "");
-
-   LLVMBuildCondBr(builder, cond, after_block, state->block);
-
-   LLVMAddIncoming(state->counter, &next, &block, 1);
-
-   LLVMPositionBuilderAtEnd(builder, after_block);
+   lp_build_loop_end_cond(builder, end, step, LLVMIntNE, state);
 }
 
 
@@ -616,24 +277,16 @@ lp_build_loop_end_cond(LLVMBuilderRef builder,
 
   Is built with:
 
-     LLVMValueRef x = LLVMGetUndef();  // or something else
+     // x needs an alloca variable
+     x = lp_build_alloca(builder, type, "x");
 
-     flow = lp_build_flow_create(builder);
 
-        lp_build_flow_scope_begin(flow);
+     lp_build_if(ctx, builder, cond);
+        LLVMBuildStore(LLVMBuildAdd(1, 2), x);
+     lp_build_else(ctx);
+        LLVMBuildStore(LLVMBuildAdd(2, 3). x);
+     lp_build_endif(ctx);
 
-           // x needs a phi node
-           lp_build_flow_scope_declare(flow, &x);
-
-           lp_build_if(ctx, flow, builder, cond);
-              x = LLVMAdd(1, 2);
-           lp_build_else(ctx);
-              x = LLVMAdd(2, 3);
-           lp_build_endif(ctx);
-
-        lp_build_flow_scope_end(flow);
-
-     lp_build_flow_destroy(flow);
  */
 
 
@@ -642,47 +295,19 @@ lp_build_loop_end_cond(LLVMBuilderRef builder,
  * Begin an if/else/endif construct.
  */
 void
-lp_build_if(struct lp_build_if_state *ctx,
-            struct lp_build_flow_context *flow,
+lp_build_if(struct lp_build_if_state *ifthen,
             LLVMBuilderRef builder,
             LLVMValueRef condition)
 {
    LLVMBasicBlockRef block = LLVMGetInsertBlock(builder);
-   struct lp_build_flow_if *ifthen;
-   unsigned i;
 
-   memset(ctx, 0, sizeof(*ctx));
-   ctx->builder = builder;
-   ctx->flow = flow;
-
-   /* push/create new scope */
-   ifthen = &lp_build_flow_push(flow, LP_BUILD_FLOW_IF)->ifthen;
-   assert(ifthen);
-
-   ifthen->num_variables = flow->num_variables;
+   memset(ifthen, 0, sizeof *ifthen);
+   ifthen->builder = builder;
    ifthen->condition = condition;
    ifthen->entry_block = block;
 
-   /* create a Phi node for each variable in this flow scope */
-   ifthen->phi = MALLOC(ifthen->num_variables * sizeof(*ifthen->phi));
-   if (!ifthen->phi) {
-      ifthen->num_variables = 0;
-      return;
-   }
-
    /* create endif/merge basic block for the phi functions */
    ifthen->merge_block = lp_build_insert_new_block(builder, "endif-block");
-   LLVMPositionBuilderAtEnd(builder, ifthen->merge_block);
-
-   /* create a phi node for each variable */
-   for (i = 0; i < flow->num_variables; i++) {
-      ifthen->phi[i] = LLVMBuildPhi(builder, LLVMTypeOf(*flow->variables[i]), "");
-
-      /* add add the initial value of the var from the entry block */
-      if (!LLVMIsUndef(*flow->variables[i]))
-         LLVMAddIncoming(ifthen->phi[i], flow->variables[i],
-                         &ifthen->entry_block, 1);
-   }
 
    /* create/insert true_block before merge_block */
    ifthen->true_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-true-block");
@@ -696,27 +321,16 @@ lp_build_if(struct lp_build_if_state *ctx,
  * Begin else-part of a conditional
  */
 void
-lp_build_else(struct lp_build_if_state *ctx)
+lp_build_else(struct lp_build_if_state *ifthen)
 {
-   struct lp_build_flow_context *flow = ctx->flow;
-   struct lp_build_flow_if *ifthen;
-   unsigned i;
-
-   ifthen = &lp_build_flow_peek(flow, LP_BUILD_FLOW_IF)->ifthen;
-   assert(ifthen);
-
-   /* for each variable, update the Phi node with a (variable, block) pair */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
-   for (i = 0; i < flow->num_variables; i++) {
-      assert(*flow->variables[i]);
-      LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &ifthen->true_block, 1);
-   }
+   /* Append an unconditional Br(anch) instruction on the true_block */
+   LLVMBuildBr(ifthen->builder, ifthen->merge_block);
 
    /* create/insert false_block before the merge block */
    ifthen->false_block = LLVMInsertBasicBlock(ifthen->merge_block, "if-false-block");
 
    /* successive code goes into the else block */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->false_block);
+   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->false_block);
 }
 
 
@@ -724,75 +338,30 @@ lp_build_else(struct lp_build_if_state *ctx)
  * End a conditional.
  */
 void
-lp_build_endif(struct lp_build_if_state *ctx)
+lp_build_endif(struct lp_build_if_state *ifthen)
 {
-   struct lp_build_flow_context *flow = ctx->flow;
-   struct lp_build_flow_if *ifthen;
-   LLVMBasicBlockRef curBlock = LLVMGetInsertBlock(ctx->builder);
-   unsigned i;
-
-   ifthen = &lp_build_flow_pop(flow, LP_BUILD_FLOW_IF)->ifthen;
-   assert(ifthen);
-
    /* Insert branch to the merge block from current block */
-   LLVMBuildBr(ctx->builder, ifthen->merge_block);
+   LLVMBuildBr(ifthen->builder, ifthen->merge_block);
 
-   if (ifthen->false_block) {
-      LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
-      /* for each variable, update the Phi node with a (variable, block) pair */
-      for (i = 0; i < flow->num_variables; i++) {
-         assert(*flow->variables[i]);
-         LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &curBlock, 1);
-         /* replace the variable ref with the phi function */
-         *flow->variables[i] = ifthen->phi[i];
-      }
-   }
-   else {
-      /* no else clause */
-      LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
-      for (i = 0; i < flow->num_variables; i++) {
-         assert(*flow->variables[i]);
-         LLVMAddIncoming(ifthen->phi[i], flow->variables[i], &ifthen->true_block, 1);
-
-         /* replace the variable ref with the phi function */
-         *flow->variables[i] = ifthen->phi[i];
-      }
-   }
-
-   FREE(ifthen->phi);
-
-   /***
-    *** Now patch in the various branch instructions.
-    ***/
+   /*
+    * Now patch in the various branch instructions.
+    */
 
    /* Insert the conditional branch instruction at the end of entry_block */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->entry_block);
+   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->entry_block);
    if (ifthen->false_block) {
       /* we have an else clause */
-      LLVMBuildCondBr(ctx->builder, ifthen->condition,
+      LLVMBuildCondBr(ifthen->builder, ifthen->condition,
                       ifthen->true_block, ifthen->false_block);
    }
    else {
       /* no else clause */
-      LLVMBuildCondBr(ctx->builder, ifthen->condition,
+      LLVMBuildCondBr(ifthen->builder, ifthen->condition,
                       ifthen->true_block, ifthen->merge_block);
    }
 
-   /* Insert branch from end of true_block to merge_block */
-   if (ifthen->false_block) {
-      /* Append an unconditional Br(anch) instruction on the true_block */
-      LLVMPositionBuilderAtEnd(ctx->builder, ifthen->true_block);
-      LLVMBuildBr(ctx->builder, ifthen->merge_block);
-   }
-   else {
-      /* No else clause.
-       * Note that we've already inserted the branch at the end of
-       * true_block.  See the very first LLVMBuildBr() call in this function.
-       */
-   }
-
    /* Resume building code at end of the ifthen->merge_block */
-   LLVMPositionBuilderAtEnd(ctx->builder, ifthen->merge_block);
+   LLVMPositionBuilderAtEnd(ifthen->builder, ifthen->merge_block);
 }
 
 
@@ -830,6 +399,7 @@ lp_build_alloca(LLVMBuilderRef builder,
    }
 
    res = LLVMBuildAlloca(first_builder, type, name);
+   LLVMBuildStore(builder, LLVMConstNull(type), res);
 
    LLVMDisposeBuilder(first_builder);
 

src/gallium/auxiliary/gallivm/lp_bld_flow.h

@@ -41,52 +41,49 @@
 struct lp_type;
 
 
-struct lp_build_flow_context;
+/**
+ * Early exit. Useful to skip to the end of a function or block when
+ * the execution mask becomes zero or when there is an error condition.
+ */
+struct lp_build_skip_context
+{
+   LLVMBuilderRef builder;
 
-
-struct lp_build_flow_context *
-lp_build_flow_create(LLVMBuilderRef builder);
+   /** Block to skip to */
+   LLVMBasicBlockRef block;
+};
 
 void
-lp_build_flow_destroy(struct lp_build_flow_context *flow);
+lp_build_flow_skip_begin(struct lp_build_skip_context *ctx,
+                         LLVMBuilderRef builder);
 
 void
-lp_build_flow_scope_begin(struct lp_build_flow_context *flow);
-
-void
-lp_build_flow_scope_declare(struct lp_build_flow_context *flow,
-                            LLVMValueRef *variable);
-
-void
-lp_build_flow_scope_end(struct lp_build_flow_context *flow);
-
-void
-lp_build_flow_skip_begin(struct lp_build_flow_context *flow);
-
-void
-lp_build_flow_skip_cond_break(struct lp_build_flow_context *flow,
+lp_build_flow_skip_cond_break(struct lp_build_skip_context *ctx,
                               LLVMValueRef cond);
 
 void
-lp_build_flow_skip_end(struct lp_build_flow_context *flow);
+lp_build_flow_skip_end(struct lp_build_skip_context *ctx);
 
 
 struct lp_build_mask_context
 {
-   struct lp_build_flow_context *flow;
+   struct lp_build_skip_context skip;
 
    LLVMTypeRef reg_type;
 
-   LLVMValueRef value;
+   LLVMValueRef var;
 };
 
 
 void
 lp_build_mask_begin(struct lp_build_mask_context *mask,
-                    struct lp_build_flow_context *flow,
+                    LLVMBuilderRef builder,
                     struct lp_type type,
                     LLVMValueRef value);
 
+LLVMValueRef
+lp_build_mask_value(struct lp_build_mask_context *mask);
+
 /**
  * Bitwise AND the mask with the given value, if a previous mask was set.
  */
@@ -94,6 +91,9 @@ void
 lp_build_mask_update(struct lp_build_mask_context *mask,
                      LLVMValueRef value);
 
+void
+lp_build_mask_check(struct lp_build_mask_context *mask);
+
 LLVMValueRef
 lp_build_mask_end(struct lp_build_mask_context *mask);
 
@@ -108,6 +108,7 @@ lp_build_mask_end(struct lp_build_mask_context *mask);
 struct lp_build_loop_state
 {
   LLVMBasicBlockRef block;
+  LLVMValueRef counter_var;
   LLVMValueRef counter;
 };
 
@@ -128,22 +129,28 @@ void
 lp_build_loop_end_cond(LLVMBuilderRef builder,
                        LLVMValueRef end,
                        LLVMValueRef step,
-                       int cond, /* LLVM condition */
+                       LLVMIntPredicate cond,
                        struct lp_build_loop_state *state);
 
 
 
 
+/**
+ * if/else/endif.
+ */
 struct lp_build_if_state
 {
    LLVMBuilderRef builder;
-   struct lp_build_flow_context *flow;
+   LLVMValueRef condition;
+   LLVMBasicBlockRef entry_block;
+   LLVMBasicBlockRef true_block;
+   LLVMBasicBlockRef false_block;
+   LLVMBasicBlockRef merge_block;
 };
 
 
 void
 lp_build_if(struct lp_build_if_state *ctx,
-            struct lp_build_flow_context *flow,
             LLVMBuilderRef builder,
             LLVMValueRef condition);
 

src/gallium/auxiliary/gallivm/lp_bld_format_yuv.c

@@ -35,6 +35,7 @@
 
 
 #include "util/u_format.h"
+#include "util/u_cpu_detect.h"
 
 #include "lp_bld_arit.h"
 #include "lp_bld_type.h"
@@ -42,7 +43,7 @@
 #include "lp_bld_conv.h"
 #include "lp_bld_gather.h"
 #include "lp_bld_format.h"
-
+#include "lp_bld_logic.h"
 
 /**
  * Extract Y, U, V channels from packed UYVY.
@@ -59,7 +60,7 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder,
                 LLVMValueRef *v)
 {
    struct lp_type type;
-   LLVMValueRef shift, mask;
+   LLVMValueRef mask;
 
    memset(&type, 0, sizeof type);
    type.width = 32;
@@ -69,14 +70,37 @@ uyvy_to_yuv_soa(LLVMBuilderRef builder,
    assert(lp_check_value(type, i));
 
    /*
-    * y = (uyvy >> 16*i) & 0xff
+    * y = (uyvy >> (16*i + 8)) & 0xff
     * u = (uyvy        ) & 0xff
     * v = (uyvy >> 16  ) & 0xff
     */
 
-   shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), "");
-   shift = LLVMBuildAdd(builder, shift, lp_build_const_int_vec(type, 8), "");
-   *y = LLVMBuildLShr(builder, packed, shift, "");
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   /*
+    * Avoid shift with per-element count.
+    * No support on x86, gets translated to roughly 5 instructions
+    * per element. Didn't measure performance but cuts shader size
+    * by quite a bit (less difference if cpu has no sse4.1 support).
+    */
+   if (util_cpu_caps.has_sse2 && n == 4) {
+      LLVMValueRef sel, tmp, tmp2;
+      struct lp_build_context bld32;
+
+      lp_build_context_init(&bld32, builder, type);
+
+      tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 8), "");
+      tmp2 = LLVMBuildLShr(builder, tmp, lp_build_const_int_vec(type, 16), "");
+      sel = lp_build_compare(builder, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(type, 0));
+      *y = lp_build_select(&bld32, sel, tmp, tmp2);
+   } else
+#endif
+   {
+      LLVMValueRef shift;
+      shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), "");
+      shift = LLVMBuildAdd(builder, shift, lp_build_const_int_vec(type, 8), "");
+      *y = LLVMBuildLShr(builder, packed, shift, "");
+   }
+
    *u = packed;
    *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 16), "");
 
@@ -103,7 +127,7 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder,
                 LLVMValueRef *v)
 {
    struct lp_type type;
-   LLVMValueRef shift, mask;
+   LLVMValueRef mask;
 
    memset(&type, 0, sizeof type);
    type.width = 32;
@@ -118,8 +142,30 @@ yuyv_to_yuv_soa(LLVMBuilderRef builder,
     * v = (yuyv >> 24  ) & 0xff
     */
 
-   shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), "");
-   *y = LLVMBuildLShr(builder, packed, shift, "");
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   /*
+    * Avoid shift with per-element count.
+    * No support on x86, gets translated to roughly 5 instructions
+    * per element. Didn't measure performance but cuts shader size
+    * by quite a bit (less difference if cpu has no sse4.1 support).
+    */
+   if (util_cpu_caps.has_sse2 && n == 4) {
+      LLVMValueRef sel, tmp;
+      struct lp_build_context bld32;
+
+      lp_build_context_init(&bld32, builder, type);
+
+      tmp = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 16), "");
+      sel = lp_build_compare(builder, type, PIPE_FUNC_EQUAL, i, lp_build_const_int_vec(type, 0));
+       *y = lp_build_select(&bld32, sel, packed, tmp);
+   } else
+#endif
+   {
+      LLVMValueRef shift;
+      shift = LLVMBuildMul(builder, i, lp_build_const_int_vec(type, 16), "");
+      *y = LLVMBuildLShr(builder, packed, shift, "");
+   }
+
    *u = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 8), "");
    *v = LLVMBuildLShr(builder, packed, lp_build_const_int_vec(type, 24), "");
 

src/gallium/auxiliary/gallivm/lp_bld_init.c

@@ -44,6 +44,7 @@ static const struct debug_named_value lp_bld_debug_flags[] = {
    { "asm",    GALLIVM_DEBUG_ASM, NULL },
    { "nopt",   GALLIVM_DEBUG_NO_OPT, NULL },
    { "perf",   GALLIVM_DEBUG_PERF, NULL },
+   { "no_brilinear", GALLIVM_DEBUG_NO_BRILINEAR, NULL },
    DEBUG_NAMED_VALUE_END
 };
 
@@ -144,13 +145,7 @@ lp_build_init(void)
          LLVMAddCFGSimplificationPass(lp_build_pass);
          LLVMAddPromoteMemoryToRegisterPass(lp_build_pass);
          LLVMAddConstantPropagationPass(lp_build_pass);
-         if(util_cpu_caps.has_sse4_1) {
-            /* FIXME: There is a bug in this pass, whereby the combination of fptosi
-             * and sitofp (necessary for trunc/floor/ceil/round implementation)
-             * somehow becomes invalid code.
-             */
-            LLVMAddInstructionCombiningPass(lp_build_pass);
-         }
+         LLVMAddInstructionCombiningPass(lp_build_pass);
          LLVMAddGVNPass(lp_build_pass);
       } else {
          /* We need at least this pass to prevent the backends to fail in

src/gallium/auxiliary/gallivm/lp_bld_init.h

@@ -47,4 +47,10 @@ lp_build_init(void);
 extern void
 lp_func_delete_body(LLVMValueRef func);
 
+
+extern LLVMValueRef
+lp_build_load_volatile(LLVMBuilderRef B, LLVMValueRef PointerVal,
+                       const char *Name);
+
+
 #endif /* !LP_BLD_INIT_H */

src/gallium/auxiliary/gallivm/lp_bld_logic.c

@@ -92,9 +92,23 @@ lp_build_compare(LLVMBuilderRef builder,
    if(func == PIPE_FUNC_ALWAYS)
       return ones;
 
-   /* TODO: optimize the constant case */
+#if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
+   /*
+    * There are no unsigned integer comparison instructions in SSE.
+    */
 
-   /* XXX: It is not clear if we should use the ordered or unordered operators */
+   if (!type.floating && !type.sign &&
+       type.width * type.length == 128 &&
+       util_cpu_caps.has_sse2 &&
+       (func == PIPE_FUNC_LESS ||
+        func == PIPE_FUNC_LEQUAL ||
+        func == PIPE_FUNC_GREATER ||
+        func == PIPE_FUNC_GEQUAL) &&
+       (gallivm_debug & GALLIVM_DEBUG_PERF)) {
+         debug_printf("%s: inefficient <%u x i%u> unsigned comparison\n",
+                      __FUNCTION__, type.length, type.width);
+   }
+#endif
 
 #if HAVE_LLVM < 0x0207
 #if defined(PIPE_ARCH_X86) || defined(PIPE_ARCH_X86_64)
@@ -225,6 +239,8 @@ lp_build_compare(LLVMBuilderRef builder,
 #endif
 #endif /* HAVE_LLVM < 0x0207 */
 
+   /* XXX: It is not clear if we should use the ordered or unordered operators */
+
    if(type.floating) {
       LLVMRealPredicate op;
       switch(func) {
@@ -446,10 +462,12 @@ lp_build_select(struct lp_build_context *bld,
       LLVMTypeRef arg_type;
       LLVMValueRef args[3];
 
-      if (type.width == 64) {
+      if (type.floating &&
+          type.width == 64) {
          intrinsic = "llvm.x86.sse41.blendvpd";
          arg_type = LLVMVectorType(LLVMDoubleType(), 2);
-      } else if (type.width == 32) {
+      } else if (type.floating &&
+                 type.width == 32) {
          intrinsic = "llvm.x86.sse41.blendvps";
          arg_type = LLVMVectorType(LLVMFloatType(), 4);
       } else {
@@ -556,28 +574,3 @@ lp_build_select_aos(struct lp_build_context *bld,
 #endif
    }
 }
-
-
-/** Return (a & ~b) */
-LLVMValueRef
-lp_build_andc(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b)
-{
-   const struct lp_type type = bld->type;
-
-   assert(lp_check_value(type, a));
-   assert(lp_check_value(type, b));
-
-   /* can't do bitwise ops on floating-point values */
-   if(type.floating) {
-      a = LLVMBuildBitCast(bld->builder, a, bld->int_vec_type, "");
-      b = LLVMBuildBitCast(bld->builder, b, bld->int_vec_type, "");
-   }
-
-   b = LLVMBuildNot(bld->builder, b, "");
-   b = LLVMBuildAnd(bld->builder, a, b, "");
-
-   if(type.floating) {
-      b = LLVMBuildBitCast(bld->builder, b, bld->vec_type, "");
-   }
-   return b;
-}

src/gallium/auxiliary/gallivm/lp_bld_logic.h

@@ -82,8 +82,4 @@ lp_build_select_aos(struct lp_build_context *bld,
                     LLVMValueRef b);
 
 
-LLVMValueRef
-lp_build_andc(struct lp_build_context *bld, LLVMValueRef a, LLVMValueRef b);
-
-
 #endif /* !LP_BLD_LOGIC_H */

src/gallium/auxiliary/gallivm/lp_bld_misc.cpp

@@ -178,3 +178,13 @@ lp_func_delete_body(LLVMValueRef FF)
    llvm::Function *func = llvm::unwrap<llvm::Function>(FF);
    func->deleteBody();
 }
+
+
+extern "C"
+LLVMValueRef
+lp_build_load_volatile(LLVMBuilderRef B, LLVMValueRef PointerVal,
+                       const char *Name)
+{
+   return llvm::wrap(llvm::unwrap(B)->CreateLoad(llvm::unwrap(PointerVal), true, Name));
+}
+

src/gallium/auxiliary/gallivm/lp_bld_printf.c

@@ -29,6 +29,8 @@
 
 #include "util/u_debug.h"
 #include "util/u_memory.h"
+#include "util/u_string.h"
+#include "lp_bld_const.h"
 #include "lp_bld_printf.h"
 
 
@@ -119,3 +121,22 @@ lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...)
    return LLVMBuildCall(builder, func_printf, params, argcount + 1, "");
 }
 
+
+
+/**
+ * Print a float[4] vector.
+ */
+LLVMValueRef
+lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec)
+{
+   char format[1000];
+   LLVMValueRef x, y, z, w;
+
+   x = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(0), "");
+   y = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(1), "");
+   z = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(2), "");
+   w = LLVMBuildExtractElement(builder, vec, lp_build_const_int32(3), "");
+
+   util_snprintf(format, sizeof(format), "%s %%f %%f %%f %%f\n", msg);
+   return lp_build_printf(builder, format, x, y, z, w);
+}

src/gallium/auxiliary/gallivm/lp_bld_printf.h

@@ -35,5 +35,9 @@
 LLVMValueRef lp_build_const_string_variable(LLVMModuleRef module, const char *str, int len);
 LLVMValueRef lp_build_printf(LLVMBuilderRef builder, const char *fmt, ...);
 
+LLVMValueRef
+lp_build_print_vec4(LLVMBuilderRef builder, const char *msg, LLVMValueRef vec);
+
+
 #endif
 

src/gallium/auxiliary/gallivm/lp_bld_quad.c

@@ -81,11 +81,15 @@ LLVMValueRef
 lp_build_scalar_ddx(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMValueRef idx_left  = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef idx_right = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_RIGHT, 0);
-   LLVMValueRef a_left  = LLVMBuildExtractElement(bld->builder, a, idx_left, "");
-   LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "");
-   return lp_build_sub(bld, a_right, a_left);
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMValueRef idx_left  = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
+   LLVMValueRef idx_right = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_RIGHT, 0);
+   LLVMValueRef a_left  = LLVMBuildExtractElement(bld->builder, a, idx_left, "left");
+   LLVMValueRef a_right = LLVMBuildExtractElement(bld->builder, a, idx_right, "right");
+   if (bld->type.floating)
+      return LLVMBuildFSub(bld->builder, a_right, a_left, "ddx");
+   else
+      return LLVMBuildSub(bld->builder, a_right, a_left, "ddx");
 }
 
 
@@ -93,9 +97,13 @@ LLVMValueRef
 lp_build_scalar_ddy(struct lp_build_context *bld,
                     LLVMValueRef a)
 {
-   LLVMValueRef idx_top    = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_TOP_LEFT, 0);
-   LLVMValueRef idx_bottom = LLVMConstInt(LLVMInt32Type(), LP_BLD_QUAD_BOTTOM_LEFT, 0);
-   LLVMValueRef a_top    = LLVMBuildExtractElement(bld->builder, a, idx_top, "");
-   LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "");
-   return lp_build_sub(bld, a_bottom, a_top);
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMValueRef idx_top    = LLVMConstInt(i32t, LP_BLD_QUAD_TOP_LEFT, 0);
+   LLVMValueRef idx_bottom = LLVMConstInt(i32t, LP_BLD_QUAD_BOTTOM_LEFT, 0);
+   LLVMValueRef a_top    = LLVMBuildExtractElement(bld->builder, a, idx_top, "top");
+   LLVMValueRef a_bottom = LLVMBuildExtractElement(bld->builder, a, idx_bottom, "bottom");
+   if (bld->type.floating)
+      return LLVMBuildFSub(bld->builder, a_bottom, a_top, "ddy");
+   else
+      return LLVMBuildSub(bld->builder, a_bottom, a_top, "ddy");
 }

src/gallium/auxiliary/gallivm/lp_bld_sample.c

@@ -36,11 +36,53 @@
 #include "pipe/p_state.h"
 #include "util/u_format.h"
 #include "util/u_math.h"
-#include "lp_bld_debug.h"
-#include "lp_bld_const.h"
 #include "lp_bld_arit.h"
-#include "lp_bld_type.h"
+#include "lp_bld_const.h"
+#include "lp_bld_debug.h"
+#include "lp_bld_printf.h"
+#include "lp_bld_flow.h"
 #include "lp_bld_sample.h"
+#include "lp_bld_swizzle.h"
+#include "lp_bld_type.h"
+
+
+/*
+ * Bri-linear factor. Should be greater than one.
+ */
+#define BRILINEAR_FACTOR 2
+
+
+/**
+ * Does the given texture wrap mode allow sampling the texture border color?
+ * XXX maybe move this into gallium util code.
+ */
+boolean
+lp_sampler_wrap_mode_uses_border_color(unsigned mode,
+                                       unsigned min_img_filter,
+                                       unsigned mag_img_filter)
+{
+   switch (mode) {
+   case PIPE_TEX_WRAP_REPEAT:
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+   case PIPE_TEX_WRAP_MIRROR_REPEAT:
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_EDGE:
+      return FALSE;
+   case PIPE_TEX_WRAP_CLAMP:
+   case PIPE_TEX_WRAP_MIRROR_CLAMP:
+      if (min_img_filter == PIPE_TEX_FILTER_NEAREST &&
+          mag_img_filter == PIPE_TEX_FILTER_NEAREST) {
+         return FALSE;
+      } else {
+         return TRUE;
+      }
+   case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
+   case PIPE_TEX_WRAP_MIRROR_CLAMP_TO_BORDER:
+      return TRUE;
+   default:
+      assert(0 && "unexpected wrap mode");
+      return FALSE;
+   }
+}
 
 
 /**
@@ -91,31 +133,40 @@ lp_sampler_static_state(struct lp_sampler_static_state *state,
    state->wrap_r            = sampler->wrap_r;
    state->min_img_filter    = sampler->min_img_filter;
    state->mag_img_filter    = sampler->mag_img_filter;
-   if (view->last_level) {
+
+   if (view->last_level && sampler->max_lod > 0.0f) {
       state->min_mip_filter = sampler->min_mip_filter;
    } else {
       state->min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
    }
 
+   if (state->min_mip_filter != PIPE_TEX_MIPFILTER_NONE) {
+      if (sampler->lod_bias != 0.0f) {
+         state->lod_bias_non_zero = 1;
+      }
+
+      /* If min_lod == max_lod we can greatly simplify mipmap selection.
+       * This is a case that occurs during automatic mipmap generation.
+       */
+      if (sampler->min_lod == sampler->max_lod) {
+         state->min_max_lod_equal = 1;
+      } else {
+         if (sampler->min_lod > 0.0f) {
+            state->apply_min_lod = 1;
+         }
+
+         if (sampler->max_lod < (float)view->last_level) {
+            state->apply_max_lod = 1;
+         }
+      }
+   }
+
    state->compare_mode      = sampler->compare_mode;
    if (sampler->compare_mode != PIPE_TEX_COMPARE_NONE) {
       state->compare_func   = sampler->compare_func;
    }
 
    state->normalized_coords = sampler->normalized_coords;
-   state->lod_bias          = sampler->lod_bias;
-   if (!view->last_level &&
-       sampler->min_img_filter == sampler->mag_img_filter) {
-      state->min_lod        = 0.0f;
-      state->max_lod        = 0.0f;
-   } else {
-      state->min_lod        = MAX2(sampler->min_lod, 0.0f);
-      state->max_lod        = sampler->max_lod;
-   }
-   state->border_color[0]   = sampler->border_color[0];
-   state->border_color[1]   = sampler->border_color[1];
-   state->border_color[2]   = sampler->border_color[2];
-   state->border_color[3]   = sampler->border_color[3];
 
    /*
     * FIXME: Handle the remainder of pipe_sampler_view.
@@ -123,6 +174,825 @@ lp_sampler_static_state(struct lp_sampler_static_state *state,
 }
 
 
+/**
+ * Generate code to compute coordinate gradient (rho).
+ * \param ddx  partial derivatives of (s, t, r, q) with respect to X
+ * \param ddy  partial derivatives of (s, t, r, q) with respect to Y
+ *
+ * XXX: The resulting rho is scalar, so we ignore all but the first element of
+ * derivatives that are passed by the shader.
+ */
+static LLVMValueRef
+lp_build_rho(struct lp_build_sample_context *bld,
+             const LLVMValueRef ddx[4],
+             const LLVMValueRef ddy[4])
+{
+   struct lp_build_context *float_size_bld = &bld->float_size_bld;
+   struct lp_build_context *float_bld = &bld->float_bld;
+   const unsigned dims = bld->dims;
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMValueRef index0 = LLVMConstInt(i32t, 0, 0);
+   LLVMValueRef index1 = LLVMConstInt(i32t, 1, 0);
+   LLVMValueRef index2 = LLVMConstInt(i32t, 2, 0);
+   LLVMValueRef dsdx, dsdy, dtdx, dtdy, drdx, drdy;
+   LLVMValueRef rho_x, rho_y;
+   LLVMValueRef rho_vec;
+   LLVMValueRef float_size;
+   LLVMValueRef rho;
+
+   dsdx = ddx[0];
+   dsdy = ddy[0];
+
+   if (dims <= 1) {
+      rho_x = dsdx;
+      rho_y = dsdy;
+   }
+   else {
+      rho_x = float_size_bld->undef;
+      rho_y = float_size_bld->undef;
+
+      rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dsdx, index0, "");
+      rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dsdy, index0, "");
+
+      dtdx = ddx[1];
+      dtdy = ddy[1];
+
+      rho_x = LLVMBuildInsertElement(bld->builder, rho_x, dtdx, index1, "");
+      rho_y = LLVMBuildInsertElement(bld->builder, rho_y, dtdy, index1, "");
+
+      if (dims >= 3) {
+         drdx = ddx[2];
+         drdy = ddy[2];
+
+         rho_x = LLVMBuildInsertElement(bld->builder, rho_x, drdx, index2, "");
+         rho_y = LLVMBuildInsertElement(bld->builder, rho_y, drdy, index2, "");
+      }
+   }
+
+   rho_x = lp_build_abs(float_size_bld, rho_x);
+   rho_y = lp_build_abs(float_size_bld, rho_y);
+
+   rho_vec = lp_build_max(float_size_bld, rho_x, rho_y);
+
+   float_size = lp_build_int_to_float(float_size_bld, bld->int_size);
+
+   rho_vec = lp_build_mul(float_size_bld, rho_vec, float_size);
+
+   if (dims <= 1) {
+      rho = rho_vec;
+   }
+   else {
+      if (dims >= 2) {
+         LLVMValueRef rho_s, rho_t, rho_r;
+
+         rho_s = LLVMBuildExtractElement(bld->builder, rho_vec, index0, "");
+         rho_t = LLVMBuildExtractElement(bld->builder, rho_vec, index1, "");
+
+         rho = lp_build_max(float_bld, rho_s, rho_t);
+
+         if (dims >= 3) {
+            rho_r = LLVMBuildExtractElement(bld->builder, rho_vec, index0, "");
+            rho = lp_build_max(float_bld, rho, rho_r);
+         }
+      }
+   }
+
+   return rho;
+}
+
+
+/*
+ * Bri-linear lod computation
+ *
+ * Use a piece-wise linear approximation of log2 such that:
+ * - round to nearest, for values in the neighborhood of -1, 0, 1, 2, etc.
+ * - linear approximation for values in the neighborhood of 0.5, 1.5., etc,
+ *   with the steepness specified in 'factor'
+ * - exact result for 0.5, 1.5, etc.
+ *
+ *
+ *   1.0 -              /----*
+ *                     /
+ *                    /
+ *                   /
+ *   0.5 -          *
+ *                 /
+ *                /
+ *               /
+ *   0.0 - *----/
+ *
+ *         |                 |
+ *        2^0               2^1
+ *
+ * This is a technique also commonly used in hardware:
+ * - http://ixbtlabs.com/articles2/gffx/nv40-rx800-3.html
+ *
+ * TODO: For correctness, this should only be applied when texture is known to
+ * have regular mipmaps, i.e., mipmaps derived from the base level.
+ *
+ * TODO: This could be done in fixed point, where applicable.
+ */
+static void
+lp_build_brilinear_lod(struct lp_build_context *bld,
+                       LLVMValueRef lod,
+                       double factor,
+                       LLVMValueRef *out_lod_ipart,
+                       LLVMValueRef *out_lod_fpart)
+{
+   LLVMValueRef lod_fpart;
+   double pre_offset = (factor - 0.5)/factor - 0.5;
+   double post_offset = 1 - factor;
+
+   if (0) {
+      lp_build_printf(bld->builder, "lod = %f\n", lod);
+   }
+
+   lod = lp_build_add(bld, lod,
+                      lp_build_const_vec(bld->type, pre_offset));
+
+   lp_build_ifloor_fract(bld, lod, out_lod_ipart, &lod_fpart);
+
+   lod_fpart = lp_build_mul(bld, lod_fpart,
+                            lp_build_const_vec(bld->type, factor));
+
+   lod_fpart = lp_build_add(bld, lod_fpart,
+                            lp_build_const_vec(bld->type, post_offset));
+
+   /*
+    * It's not necessary to clamp lod_fpart since:
+    * - the above expression will never produce numbers greater than one.
+    * - the mip filtering branch is only taken if lod_fpart is positive
+    */
+
+   *out_lod_fpart = lod_fpart;
+
+   if (0) {
+      lp_build_printf(bld->builder, "lod_ipart = %i\n", *out_lod_ipart);
+      lp_build_printf(bld->builder, "lod_fpart = %f\n\n", *out_lod_fpart);
+   }
+}
+
+
+/*
+ * Combined log2 and brilinear lod computation.
+ *
+ * It's in all identical to calling lp_build_fast_log2() and
+ * lp_build_brilinear_lod() above, but by combining we can compute the interger
+ * and fractional part independently.
+ */
+static void
+lp_build_brilinear_rho(struct lp_build_context *bld,
+                       LLVMValueRef rho,
+                       double factor,
+                       LLVMValueRef *out_lod_ipart,
+                       LLVMValueRef *out_lod_fpart)
+{
+   LLVMValueRef lod_ipart;
+   LLVMValueRef lod_fpart;
+
+   const double pre_factor = (2*factor - 0.5)/(M_SQRT2*factor);
+   const double post_offset = 1 - 2*factor;
+
+   assert(bld->type.floating);
+
+   assert(lp_check_value(bld->type, rho));
+
+   /*
+    * The pre factor will make the intersections with the exact powers of two
+    * happen precisely where we want then to be, which means that the integer
+    * part will not need any post adjustments.
+    */
+   rho = lp_build_mul(bld, rho,
+                      lp_build_const_vec(bld->type, pre_factor));
+
+   /* ipart = ifloor(log2(rho)) */
+   lod_ipart = lp_build_extract_exponent(bld, rho, 0);
+
+   /* fpart = rho / 2**ipart */
+   lod_fpart = lp_build_extract_mantissa(bld, rho);
+
+   lod_fpart = lp_build_mul(bld, lod_fpart,
+                            lp_build_const_vec(bld->type, factor));
+
+   lod_fpart = lp_build_add(bld, lod_fpart,
+                            lp_build_const_vec(bld->type, post_offset));
+
+   /*
+    * Like lp_build_brilinear_lod, it's not necessary to clamp lod_fpart since:
+    * - the above expression will never produce numbers greater than one.
+    * - the mip filtering branch is only taken if lod_fpart is positive
+    */
+
+   *out_lod_ipart = lod_ipart;
+   *out_lod_fpart = lod_fpart;
+}
+
+
+/**
+ * Generate code to compute texture level of detail (lambda).
+ * \param ddx  partial derivatives of (s, t, r, q) with respect to X
+ * \param ddy  partial derivatives of (s, t, r, q) with respect to Y
+ * \param lod_bias  optional float vector with the shader lod bias
+ * \param explicit_lod  optional float vector with the explicit lod
+ * \param width  scalar int texture width
+ * \param height  scalar int texture height
+ * \param depth  scalar int texture depth
+ *
+ * XXX: The resulting lod is scalar, so ignore all but the first element of
+ * derivatives, lod_bias, etc that are passed by the shader.
+ */
+void
+lp_build_lod_selector(struct lp_build_sample_context *bld,
+                      unsigned unit,
+                      const LLVMValueRef ddx[4],
+                      const LLVMValueRef ddy[4],
+                      LLVMValueRef lod_bias, /* optional */
+                      LLVMValueRef explicit_lod, /* optional */
+                      unsigned mip_filter,
+                      LLVMValueRef *out_lod_ipart,
+                      LLVMValueRef *out_lod_fpart)
+
+{
+   struct lp_build_context *float_bld = &bld->float_bld;
+   LLVMValueRef lod;
+
+   *out_lod_ipart = bld->int_bld.zero;
+   *out_lod_fpart = bld->float_bld.zero;
+
+   if (bld->static_state->min_max_lod_equal) {
+      /* User is forcing sampling from a particular mipmap level.
+       * This is hit during mipmap generation.
+       */
+      LLVMValueRef min_lod =
+         bld->dynamic_state->min_lod(bld->dynamic_state, bld->builder, unit);
+
+      lod = min_lod;
+   }
+   else {
+      LLVMValueRef sampler_lod_bias =
+         bld->dynamic_state->lod_bias(bld->dynamic_state, bld->builder, unit);
+      LLVMValueRef index0 = LLVMConstInt(LLVMInt32Type(), 0, 0);
+
+      if (explicit_lod) {
+         lod = LLVMBuildExtractElement(bld->builder, explicit_lod,
+                                       index0, "");
+      }
+      else {
+         LLVMValueRef rho;
+
+         rho = lp_build_rho(bld, ddx, ddy);
+
+         /*
+          * Compute lod = log2(rho)
+          */
+
+         if (!lod_bias &&
+             !bld->static_state->lod_bias_non_zero &&
+             !bld->static_state->apply_max_lod &&
+             !bld->static_state->apply_min_lod) {
+            /*
+             * Special case when there are no post-log2 adjustments, which
+             * saves instructions but keeping the integer and fractional lod
+             * computations separate from the start.
+             */
+
+            if (mip_filter == PIPE_TEX_MIPFILTER_NONE ||
+                mip_filter == PIPE_TEX_MIPFILTER_NEAREST) {
+               *out_lod_ipart = lp_build_ilog2(float_bld, rho);
+               *out_lod_fpart = bld->float_bld.zero;
+               return;
+            }
+            if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR &&
+                !(gallivm_debug & GALLIVM_DEBUG_NO_BRILINEAR)) {
+               lp_build_brilinear_rho(float_bld, rho, BRILINEAR_FACTOR,
+                                      out_lod_ipart, out_lod_fpart);
+               return;
+            }
+         }
+
+         if (0) {
+            lod = lp_build_log2(float_bld, rho);
+         }
+         else {
+            lod = lp_build_fast_log2(float_bld, rho);
+         }
+
+         /* add shader lod bias */
+         if (lod_bias) {
+            lod_bias = LLVMBuildExtractElement(bld->builder, lod_bias,
+                                               index0, "");
+            lod = LLVMBuildFAdd(bld->builder, lod, lod_bias, "shader_lod_bias");
+         }
+      }
+
+      /* add sampler lod bias */
+      if (bld->static_state->lod_bias_non_zero)
+         lod = LLVMBuildFAdd(bld->builder, lod, sampler_lod_bias, "sampler_lod_bias");
+
+
+      /* clamp lod */
+      if (bld->static_state->apply_max_lod) {
+         LLVMValueRef max_lod =
+            bld->dynamic_state->max_lod(bld->dynamic_state, bld->builder, unit);
+
+         lod = lp_build_min(float_bld, lod, max_lod);
+      }
+      if (bld->static_state->apply_min_lod) {
+         LLVMValueRef min_lod =
+            bld->dynamic_state->min_lod(bld->dynamic_state, bld->builder, unit);
+
+         lod = lp_build_max(float_bld, lod, min_lod);
+      }
+   }
+
+   if (mip_filter == PIPE_TEX_MIPFILTER_LINEAR) {
+      if (!(gallivm_debug & GALLIVM_DEBUG_NO_BRILINEAR)) {
+         lp_build_brilinear_lod(float_bld, lod, BRILINEAR_FACTOR,
+                                out_lod_ipart, out_lod_fpart);
+      }
+      else {
+         lp_build_ifloor_fract(float_bld, lod, out_lod_ipart, out_lod_fpart);
+      }
+
+      lp_build_name(*out_lod_fpart, "lod_fpart");
+   }
+   else {
+      *out_lod_ipart = lp_build_iround(float_bld, lod);
+   }
+
+   lp_build_name(*out_lod_ipart, "lod_ipart");
+
+   return;
+}
+
+
+/**
+ * For PIPE_TEX_MIPFILTER_NEAREST, convert float LOD to integer
+ * mipmap level index.
+ * Note: this is all scalar code.
+ * \param lod  scalar float texture level of detail
+ * \param level_out  returns integer 
+ */
+void
+lp_build_nearest_mip_level(struct lp_build_sample_context *bld,
+                           unsigned unit,
+                           LLVMValueRef lod_ipart,
+                           LLVMValueRef *level_out)
+{
+   struct lp_build_context *int_bld = &bld->int_bld;
+   LLVMValueRef last_level, level;
+
+   LLVMValueRef zero = LLVMConstInt(LLVMInt32Type(), 0, 0);
+
+   last_level = bld->dynamic_state->last_level(bld->dynamic_state,
+                                               bld->builder, unit);
+
+   /* convert float lod to integer */
+   level = lod_ipart;
+
+   /* clamp level to legal range of levels */
+   *level_out = lp_build_clamp(int_bld, level, zero, last_level);
+}
+
+
+/**
+ * For PIPE_TEX_MIPFILTER_LINEAR, convert float LOD to integer to
+ * two (adjacent) mipmap level indexes.  Later, we'll sample from those
+ * two mipmap levels and interpolate between them.
+ */
+void
+lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
+                           unsigned unit,
+                           LLVMValueRef lod_ipart,
+                           LLVMValueRef *lod_fpart_inout,
+                           LLVMValueRef *level0_out,
+                           LLVMValueRef *level1_out)
+{
+   LLVMBuilderRef builder = bld->builder;
+   struct lp_build_context *int_bld = &bld->int_bld;
+   struct lp_build_context *float_bld = &bld->float_bld;
+   LLVMValueRef last_level;
+   LLVMValueRef clamp_min;
+   LLVMValueRef clamp_max;
+
+   *level0_out = lod_ipart;
+   *level1_out = lp_build_add(int_bld, lod_ipart, int_bld->one);
+
+   last_level = bld->dynamic_state->last_level(bld->dynamic_state,
+                                               bld->builder, unit);
+
+   /*
+    * Clamp both lod_ipart and lod_ipart + 1 to [0, last_level], with the
+    * minimum number of comparisons, and zeroing lod_fpart in the extreme
+    * ends in the process.
+    */
+
+   /* lod_ipart < 0 */
+   clamp_min = LLVMBuildICmp(builder, LLVMIntSLT,
+                             lod_ipart, int_bld->zero,
+                             "clamp_lod_to_zero");
+
+   *level0_out = LLVMBuildSelect(builder, clamp_min,
+                                 int_bld->zero, *level0_out, "");
+
+   *level1_out = LLVMBuildSelect(builder, clamp_min,
+                                 int_bld->zero, *level1_out, "");
+
+   *lod_fpart_inout = LLVMBuildSelect(builder, clamp_min,
+                                      float_bld->zero, *lod_fpart_inout, "");
+
+   /* lod_ipart >= last_level */
+   clamp_max = LLVMBuildICmp(builder, LLVMIntSGE,
+                             lod_ipart, last_level,
+                             "clamp_lod_to_last");
+
+   *level0_out = LLVMBuildSelect(builder, clamp_max,
+                                 last_level, *level0_out, "");
+
+   *level1_out = LLVMBuildSelect(builder, clamp_max,
+                                 last_level, *level1_out, "");
+
+   *lod_fpart_inout = LLVMBuildSelect(builder, clamp_max,
+                                      float_bld->zero, *lod_fpart_inout, "");
+
+   lp_build_name(*level0_out, "sampler%u_miplevel0", unit);
+   lp_build_name(*level1_out, "sampler%u_miplevel1", unit);
+   lp_build_name(*lod_fpart_inout, "sampler%u_mipweight", unit);
+}
+
+
+/**
+ * Return pointer to a single mipmap level.
+ * \param data_array  array of pointers to mipmap levels
+ * \param level  integer mipmap level
+ */
+LLVMValueRef
+lp_build_get_mipmap_level(struct lp_build_sample_context *bld,
+                          LLVMValueRef level)
+{
+   LLVMValueRef indexes[2], data_ptr;
+   indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indexes[1] = level;
+   data_ptr = LLVMBuildGEP(bld->builder, bld->data_array, indexes, 2, "");
+   data_ptr = LLVMBuildLoad(bld->builder, data_ptr, "");
+   return data_ptr;
+}
+
+
+LLVMValueRef
+lp_build_get_const_mipmap_level(struct lp_build_sample_context *bld,
+                                int level)
+{
+   LLVMValueRef lvl = LLVMConstInt(LLVMInt32Type(), level, 0);
+   return lp_build_get_mipmap_level(bld, lvl);
+}
+
+
+/**
+ * Codegen equivalent for u_minify().
+ * Return max(1, base_size >> level);
+ */
+static LLVMValueRef
+lp_build_minify(struct lp_build_context *bld,
+                LLVMValueRef base_size,
+                LLVMValueRef level)
+{
+   assert(lp_check_value(bld->type, base_size));
+   assert(lp_check_value(bld->type, level));
+
+   if (level == bld->zero) {
+      /* if we're using mipmap level zero, no minification is needed */
+      return base_size;
+   }
+   else {
+      LLVMValueRef size =
+         LLVMBuildLShr(bld->builder, base_size, level, "minify");
+      assert(bld->type.sign);
+      size = lp_build_max(bld, size, bld->one);
+      return size;
+   }
+}
+
+
+/**
+ * Dereference stride_array[mipmap_level] array to get a stride.
+ * Return stride as a vector.
+ */
+static LLVMValueRef
+lp_build_get_level_stride_vec(struct lp_build_sample_context *bld,
+                              LLVMValueRef stride_array, LLVMValueRef level)
+{
+   LLVMValueRef indexes[2], stride;
+   indexes[0] = LLVMConstInt(LLVMInt32Type(), 0, 0);
+   indexes[1] = level;
+   stride = LLVMBuildGEP(bld->builder, stride_array, indexes, 2, "");
+   stride = LLVMBuildLoad(bld->builder, stride, "");
+   stride = lp_build_broadcast_scalar(&bld->int_coord_bld, stride);
+   return stride;
+}
+
+
+/**
+ * When sampling a mipmap, we need to compute the width, height, depth
+ * of the source levels from the level indexes.  This helper function
+ * does that.
+ */
+void
+lp_build_mipmap_level_sizes(struct lp_build_sample_context *bld,
+                            LLVMValueRef ilevel,
+                            LLVMValueRef *out_size,
+                            LLVMValueRef *row_stride_vec,
+                            LLVMValueRef *img_stride_vec)
+{
+   const unsigned dims = bld->dims;
+   LLVMValueRef ilevel_vec;
+
+   ilevel_vec = lp_build_broadcast_scalar(&bld->int_size_bld, ilevel);
+
+   /*
+    * Compute width, height, depth at mipmap level 'ilevel'
+    */
+   *out_size = lp_build_minify(&bld->int_size_bld, bld->int_size, ilevel_vec);
+
+   if (dims >= 2) {
+      *row_stride_vec = lp_build_get_level_stride_vec(bld,
+                                                      bld->row_stride_array,
+                                                      ilevel);
+      if (dims == 3 || bld->static_state->target == PIPE_TEXTURE_CUBE) {
+         *img_stride_vec = lp_build_get_level_stride_vec(bld,
+                                                         bld->img_stride_array,
+                                                         ilevel);
+      }
+   }
+}
+
+
+/**
+ * Extract and broadcast texture size.
+ *
+ * @param size_type   type of the texture size vector (either
+ *                    bld->int_size_type or bld->float_size_type)
+ * @param coord_type  type of the texture size vector (either
+ *                    bld->int_coord_type or bld->coord_type)
+ * @param int_size    vector with the integer texture size (width, height,
+ *                    depth)
+ */
+void
+lp_build_extract_image_sizes(struct lp_build_sample_context *bld,
+                             struct lp_type size_type,
+                             struct lp_type coord_type,
+                             LLVMValueRef size,
+                             LLVMValueRef *out_width,
+                             LLVMValueRef *out_height,
+                             LLVMValueRef *out_depth)
+{
+   const unsigned dims = bld->dims;
+   LLVMTypeRef i32t = LLVMInt32Type();
+
+   *out_width = lp_build_extract_broadcast(bld->builder,
+                                           size_type,
+                                           coord_type,
+                                           size,
+                                           LLVMConstInt(i32t, 0, 0));
+   if (dims >= 2) {
+      *out_height = lp_build_extract_broadcast(bld->builder,
+                                               size_type,
+                                               coord_type,
+                                               size,
+                                               LLVMConstInt(i32t, 1, 0));
+      if (dims == 3) {
+         *out_depth = lp_build_extract_broadcast(bld->builder,
+                                                 size_type,
+                                                 coord_type,
+                                                 size,
+                                                 LLVMConstInt(i32t, 2, 0));
+      }
+   }
+}
+
+
+/**
+ * Unnormalize coords.
+ *
+ * @param int_size  vector with the integer texture size (width, height, depth)
+ */
+void
+lp_build_unnormalized_coords(struct lp_build_sample_context *bld,
+                             LLVMValueRef flt_size,
+                             LLVMValueRef *s,
+                             LLVMValueRef *t,
+                             LLVMValueRef *r)
+{
+   const unsigned dims = bld->dims;
+   LLVMValueRef width;
+   LLVMValueRef height;
+   LLVMValueRef depth;
+
+   lp_build_extract_image_sizes(bld,
+                                bld->float_size_type,
+                                bld->coord_type,
+                                flt_size,
+                                &width,
+                                &height,
+                                &depth);
+
+   /* s = s * width, t = t * height */
+   *s = lp_build_mul(&bld->coord_bld, *s, width);
+   if (dims >= 2) {
+      *t = lp_build_mul(&bld->coord_bld, *t, height);
+      if (dims >= 3) {
+         *r = lp_build_mul(&bld->coord_bld, *r, depth);
+      }
+   }
+}
+
+
+/** Helper used by lp_build_cube_lookup() */
+static LLVMValueRef
+lp_build_cube_ima(struct lp_build_context *coord_bld, LLVMValueRef coord)
+{
+   /* ima = -0.5 / abs(coord); */
+   LLVMValueRef negHalf = lp_build_const_vec(coord_bld->type, -0.5);
+   LLVMValueRef absCoord = lp_build_abs(coord_bld, coord);
+   LLVMValueRef ima = lp_build_div(coord_bld, negHalf, absCoord);
+   return ima;
+}
+
+
+/**
+ * Helper used by lp_build_cube_lookup()
+ * \param sign  scalar +1 or -1
+ * \param coord  float vector
+ * \param ima  float vector
+ */
+static LLVMValueRef
+lp_build_cube_coord(struct lp_build_context *coord_bld,
+                    LLVMValueRef sign, int negate_coord,
+                    LLVMValueRef coord, LLVMValueRef ima)
+{
+   /* return negate(coord) * ima * sign + 0.5; */
+   LLVMValueRef half = lp_build_const_vec(coord_bld->type, 0.5);
+   LLVMValueRef res;
+
+   assert(negate_coord == +1 || negate_coord == -1);
+
+   if (negate_coord == -1) {
+      coord = lp_build_negate(coord_bld, coord);
+   }
+
+   res = lp_build_mul(coord_bld, coord, ima);
+   if (sign) {
+      sign = lp_build_broadcast_scalar(coord_bld, sign);
+      res = lp_build_mul(coord_bld, res, sign);
+   }
+   res = lp_build_add(coord_bld, res, half);
+
+   return res;
+}
+
+
+/** Helper used by lp_build_cube_lookup()
+ * Return (major_coord >= 0) ? pos_face : neg_face;
+ */
+static LLVMValueRef
+lp_build_cube_face(struct lp_build_sample_context *bld,
+                   LLVMValueRef major_coord,
+                   unsigned pos_face, unsigned neg_face)
+{
+   LLVMValueRef cmp = LLVMBuildFCmp(bld->builder, LLVMRealUGE,
+                                    major_coord,
+                                    bld->float_bld.zero, "");
+   LLVMValueRef pos = LLVMConstInt(LLVMInt32Type(), pos_face, 0);
+   LLVMValueRef neg = LLVMConstInt(LLVMInt32Type(), neg_face, 0);
+   LLVMValueRef res = LLVMBuildSelect(bld->builder, cmp, pos, neg, "");
+   return res;
+}
+
+
+
+/**
+ * Generate code to do cube face selection and compute per-face texcoords.
+ */
+void
+lp_build_cube_lookup(struct lp_build_sample_context *bld,
+                     LLVMValueRef s,
+                     LLVMValueRef t,
+                     LLVMValueRef r,
+                     LLVMValueRef *face,
+                     LLVMValueRef *face_s,
+                     LLVMValueRef *face_t)
+{
+   struct lp_build_context *float_bld = &bld->float_bld;
+   struct lp_build_context *coord_bld = &bld->coord_bld;
+   LLVMValueRef rx, ry, rz;
+   LLVMValueRef arx, ary, arz;
+   LLVMValueRef c25 = LLVMConstReal(LLVMFloatType(), 0.25);
+   LLVMValueRef arx_ge_ary, arx_ge_arz;
+   LLVMValueRef ary_ge_arx, ary_ge_arz;
+   LLVMValueRef arx_ge_ary_arz, ary_ge_arx_arz;
+   LLVMValueRef rx_pos, ry_pos, rz_pos;
+
+   assert(bld->coord_bld.type.length == 4);
+
+   /*
+    * Use the average of the four pixel's texcoords to choose the face.
+    */
+   rx = lp_build_mul(float_bld, c25,
+                     lp_build_sum_vector(&bld->coord_bld, s));
+   ry = lp_build_mul(float_bld, c25,
+                     lp_build_sum_vector(&bld->coord_bld, t));
+   rz = lp_build_mul(float_bld, c25,
+                     lp_build_sum_vector(&bld->coord_bld, r));
+
+   arx = lp_build_abs(float_bld, rx);
+   ary = lp_build_abs(float_bld, ry);
+   arz = lp_build_abs(float_bld, rz);
+
+   /*
+    * Compare sign/magnitude of rx,ry,rz to determine face
+    */
+   arx_ge_ary = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, ary, "");
+   arx_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, arx, arz, "");
+   ary_ge_arx = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arx, "");
+   ary_ge_arz = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ary, arz, "");
+
+   arx_ge_ary_arz = LLVMBuildAnd(bld->builder, arx_ge_ary, arx_ge_arz, "");
+   ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, "");
+
+   rx_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rx, float_bld->zero, "");
+   ry_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, ry, float_bld->zero, "");
+   rz_pos = LLVMBuildFCmp(bld->builder, LLVMRealUGE, rz, float_bld->zero, "");
+
+   {
+      struct lp_build_if_state if_ctx;
+      LLVMValueRef face_s_var;
+      LLVMValueRef face_t_var;
+      LLVMValueRef face_var;
+
+      face_s_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_s_var");
+      face_t_var = lp_build_alloca(bld->builder, bld->coord_bld.vec_type, "face_t_var");
+      face_var = lp_build_alloca(bld->builder, bld->int_bld.vec_type, "face_var");
+
+      lp_build_if(&if_ctx, bld->builder, arx_ge_ary_arz);
+      {
+         /* +/- X face */
+         LLVMValueRef sign = lp_build_sgn(float_bld, rx);
+         LLVMValueRef ima = lp_build_cube_ima(coord_bld, s);
+         *face_s = lp_build_cube_coord(coord_bld, sign, +1, r, ima);
+         *face_t = lp_build_cube_coord(coord_bld, NULL, +1, t, ima);
+         *face = lp_build_cube_face(bld, rx,
+                                    PIPE_TEX_FACE_POS_X,
+                                    PIPE_TEX_FACE_NEG_X);
+         LLVMBuildStore(bld->builder, *face_s, face_s_var);
+         LLVMBuildStore(bld->builder, *face_t, face_t_var);
+         LLVMBuildStore(bld->builder, *face, face_var);
+      }
+      lp_build_else(&if_ctx);
+      {
+         struct lp_build_if_state if_ctx2;
+
+         ary_ge_arx_arz = LLVMBuildAnd(bld->builder, ary_ge_arx, ary_ge_arz, "");
+
+         lp_build_if(&if_ctx2, bld->builder, ary_ge_arx_arz);
+         {
+            /* +/- Y face */
+            LLVMValueRef sign = lp_build_sgn(float_bld, ry);
+            LLVMValueRef ima = lp_build_cube_ima(coord_bld, t);
+            *face_s = lp_build_cube_coord(coord_bld, NULL, -1, s, ima);
+            *face_t = lp_build_cube_coord(coord_bld, sign, -1, r, ima);
+            *face = lp_build_cube_face(bld, ry,
+                                       PIPE_TEX_FACE_POS_Y,
+                                       PIPE_TEX_FACE_NEG_Y);
+            LLVMBuildStore(bld->builder, *face_s, face_s_var);
+            LLVMBuildStore(bld->builder, *face_t, face_t_var);
+            LLVMBuildStore(bld->builder, *face, face_var);
+         }
+         lp_build_else(&if_ctx2);
+         {
+            /* +/- Z face */
+            LLVMValueRef sign = lp_build_sgn(float_bld, rz);
+            LLVMValueRef ima = lp_build_cube_ima(coord_bld, r);
+            *face_s = lp_build_cube_coord(coord_bld, sign, -1, s, ima);
+            *face_t = lp_build_cube_coord(coord_bld, NULL, +1, t, ima);
+            *face = lp_build_cube_face(bld, rz,
+                                       PIPE_TEX_FACE_POS_Z,
+                                       PIPE_TEX_FACE_NEG_Z);
+            LLVMBuildStore(bld->builder, *face_s, face_s_var);
+            LLVMBuildStore(bld->builder, *face_t, face_t_var);
+            LLVMBuildStore(bld->builder, *face, face_var);
+         }
+         lp_build_endif(&if_ctx2);
+      }
+
+      lp_build_endif(&if_ctx);
+
+      *face_s = LLVMBuildLoad(bld->builder, face_s_var, "face_s");
+      *face_t = LLVMBuildLoad(bld->builder, face_t_var, "face_t");
+      *face   = LLVMBuildLoad(bld->builder, face_var, "face");
+   }
+}
+
+
 /**
  * Compute the partial offset of a pixel block along an arbitrary axis.
  *
@@ -152,11 +1022,21 @@ lp_build_sample_partial_offset(struct lp_build_context *bld,
        * Pixel blocks have power of two dimensions. LLVM should convert the
        * rem/div to bit arithmetic.
        * TODO: Verify this.
+       * It does indeed BUT it does transform it to scalar (and back) when doing so
+       * (using roughly extract, shift/and, mov, unpack) (llvm 2.7).
+       * The generated code looks seriously unfunny and is quite expensive.
        */
-
+#if 0
       LLVMValueRef block_width = lp_build_const_int_vec(bld->type, block_length);
       subcoord = LLVMBuildURem(bld->builder, coord, block_width, "");
       coord    = LLVMBuildUDiv(bld->builder, coord, block_width, "");
+#else
+      unsigned logbase2 = util_unsigned_logbase2(block_length);
+      LLVMValueRef block_shift = lp_build_const_int_vec(bld->type, logbase2);
+      LLVMValueRef block_mask = lp_build_const_int_vec(bld->type, block_length - 1);
+      subcoord = LLVMBuildAnd(bld->builder, coord, block_mask, "");
+      coord = LLVMBuildLShr(bld->builder, coord, block_shift, "");
+#endif
    }
 
    offset = lp_build_mul(bld, coord, stride);

src/gallium/auxiliary/gallivm/lp_bld_sample.h

@@ -37,8 +37,11 @@
 
 
 #include "pipe/p_format.h"
-
+#include "util/u_debug.h"
 #include "gallivm/lp_bld.h"
+#include "gallivm/lp_bld_type.h"
+#include "gallivm/lp_bld_swizzle.h"
+
 
 struct pipe_resource;
 struct pipe_sampler_view;
@@ -79,8 +82,10 @@ struct lp_sampler_static_state
    unsigned compare_mode:1;
    unsigned compare_func:3;
    unsigned normalized_coords:1;
-   float lod_bias, min_lod, max_lod;
-   float border_color[4];
+   unsigned min_max_lod_equal:1;  /**< min_lod == max_lod ? */
+   unsigned lod_bias_non_zero:1;
+   unsigned apply_min_lod:1;  /**< min_lod > 0 ? */
+   unsigned apply_max_lod:1;  /**< max_lod < last_level ? */
 };
 
 
@@ -97,48 +102,187 @@ struct lp_sampler_static_state
 struct lp_sampler_dynamic_state
 {
 
-   /** Obtain the base texture width. */
+   /** Obtain the base texture width (returns int32) */
    LLVMValueRef
    (*width)( const struct lp_sampler_dynamic_state *state,
              LLVMBuilderRef builder,
              unsigned unit);
 
-   /** Obtain the base texture height. */
+   /** Obtain the base texture height (returns int32) */
    LLVMValueRef
    (*height)( const struct lp_sampler_dynamic_state *state,
               LLVMBuilderRef builder,
               unsigned unit);
 
-   /** Obtain the base texture depth. */
+   /** Obtain the base texture depth (returns int32) */
    LLVMValueRef
    (*depth)( const struct lp_sampler_dynamic_state *state,
              LLVMBuilderRef builder,
              unsigned unit);
 
-   /** Obtain the number of mipmap levels (minus one). */
+   /** Obtain the number of mipmap levels minus one (returns int32) */
    LLVMValueRef
    (*last_level)( const struct lp_sampler_dynamic_state *state,
                   LLVMBuilderRef builder,
                   unsigned unit);
 
+   /** Obtain stride in bytes between image rows/blocks (returns int32) */
    LLVMValueRef
    (*row_stride)( const struct lp_sampler_dynamic_state *state,
                   LLVMBuilderRef builder,
                   unsigned unit);
 
+   /** Obtain stride in bytes between image slices (returns int32) */
    LLVMValueRef
    (*img_stride)( const struct lp_sampler_dynamic_state *state,
                   LLVMBuilderRef builder,
                   unsigned unit);
 
+   /** Obtain pointer to array of pointers to mimpap levels */
    LLVMValueRef
    (*data_ptr)( const struct lp_sampler_dynamic_state *state,
                 LLVMBuilderRef builder,
                 unsigned unit);
 
+   /** Obtain texture min lod (returns float) */
+   LLVMValueRef
+   (*min_lod)(const struct lp_sampler_dynamic_state *state,
+              LLVMBuilderRef builder, unsigned unit);
+
+   /** Obtain texture max lod (returns float) */
+   LLVMValueRef
+   (*max_lod)(const struct lp_sampler_dynamic_state *state,
+              LLVMBuilderRef builder, unsigned unit);
+
+   /** Obtain texture lod bias (returns float) */
+   LLVMValueRef
+   (*lod_bias)(const struct lp_sampler_dynamic_state *state,
+               LLVMBuilderRef builder, unsigned unit);
+
+   /** Obtain texture border color (returns ptr to float[4]) */
+   LLVMValueRef
+   (*border_color)(const struct lp_sampler_dynamic_state *state,
+                   LLVMBuilderRef builder, unsigned unit);
 };
 
 
+/**
+ * Keep all information for sampling code generation in a single place.
+ */
+struct lp_build_sample_context
+{
+   LLVMBuilderRef builder;
+
+   const struct lp_sampler_static_state *static_state;
+
+   struct lp_sampler_dynamic_state *dynamic_state;
+
+   const struct util_format_description *format_desc;
+
+   /* See texture_dims() */
+   unsigned dims;
+
+   /** regular scalar float type */
+   struct lp_type float_type;
+   struct lp_build_context float_bld;
+
+   /** float vector type */
+   struct lp_build_context float_vec_bld;
+
+   /** regular scalar float type */
+   struct lp_type int_type;
+   struct lp_build_context int_bld;
+
+   /** Incoming coordinates type and build context */
+   struct lp_type coord_type;
+   struct lp_build_context coord_bld;
+
+   /** Signed integer coordinates */
+   struct lp_type int_coord_type;
+   struct lp_build_context int_coord_bld;
+
+   /** Unsigned integer texture size */
+   struct lp_type int_size_type;
+   struct lp_build_context int_size_bld;
+
+   /** Unsigned integer texture size */
+   struct lp_type float_size_type;
+   struct lp_build_context float_size_bld;
+
+   /** Output texels type and build context */
+   struct lp_type texel_type;
+   struct lp_build_context texel_bld;
+
+   /* Common dynamic state values */
+   LLVMValueRef width;
+   LLVMValueRef height;
+   LLVMValueRef depth;
+   LLVMValueRef row_stride_array;
+   LLVMValueRef img_stride_array;
+   LLVMValueRef data_array;
+
+   /** Integer vector with texture width, height, depth */
+   LLVMValueRef int_size;
+};
+
+
+
+/**
+ * We only support a few wrap modes in lp_build_sample_wrap_linear_int() at
+ * this time.  Return whether the given mode is supported by that function.
+ */
+static INLINE boolean
+lp_is_simple_wrap_mode(unsigned mode)
+{
+   switch (mode) {
+   case PIPE_TEX_WRAP_REPEAT:
+   case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return TRUE;
+   default:
+      return FALSE;
+   }
+}
+
+
+static INLINE void
+apply_sampler_swizzle(struct lp_build_sample_context *bld,
+                      LLVMValueRef *texel)
+{
+   unsigned char swizzles[4];
+
+   swizzles[0] = bld->static_state->swizzle_r;
+   swizzles[1] = bld->static_state->swizzle_g;
+   swizzles[2] = bld->static_state->swizzle_b;
+   swizzles[3] = bld->static_state->swizzle_a;
+
+   lp_build_swizzle_soa_inplace(&bld->texel_bld, texel, swizzles);
+}
+
+
+static INLINE unsigned
+texture_dims(enum pipe_texture_target tex)
+{
+   switch (tex) {
+   case PIPE_TEXTURE_1D:
+      return 1;
+   case PIPE_TEXTURE_2D:
+   case PIPE_TEXTURE_RECT:
+   case PIPE_TEXTURE_CUBE:
+      return 2;
+   case PIPE_TEXTURE_3D:
+      return 3;
+   default:
+      assert(0 && "bad texture target in texture_dims()");
+      return 2;
+   }
+}
+
+
+boolean
+lp_sampler_wrap_mode_uses_border_color(unsigned mode,
+                                       unsigned min_img_filter,
+                                       unsigned mag_img_filter);
+
 /**
  * Derive the sampler static state.
  */
@@ -148,6 +292,76 @@ lp_sampler_static_state(struct lp_sampler_static_state *state,
                         const struct pipe_sampler_state *sampler);
 
 
+void
+lp_build_lod_selector(struct lp_build_sample_context *bld,
+                      unsigned unit,
+                      const LLVMValueRef ddx[4],
+                      const LLVMValueRef ddy[4],
+                      LLVMValueRef lod_bias, /* optional */
+                      LLVMValueRef explicit_lod, /* optional */
+                      unsigned mip_filter,
+                      LLVMValueRef *out_lod_ipart,
+                      LLVMValueRef *out_lod_fpart);
+
+void
+lp_build_nearest_mip_level(struct lp_build_sample_context *bld,
+                           unsigned unit,
+                           LLVMValueRef lod,
+                           LLVMValueRef *level_out);
+
+void
+lp_build_linear_mip_levels(struct lp_build_sample_context *bld,
+                           unsigned unit,
+                           LLVMValueRef lod_ipart,
+                           LLVMValueRef *lod_fpart_inout,
+                           LLVMValueRef *level0_out,
+                           LLVMValueRef *level1_out);
+
+LLVMValueRef
+lp_build_get_mipmap_level(struct lp_build_sample_context *bld,
+                          LLVMValueRef level);
+
+LLVMValueRef
+lp_build_get_const_mipmap_level(struct lp_build_sample_context *bld,
+                                int level);
+
+
+void
+lp_build_mipmap_level_sizes(struct lp_build_sample_context *bld,
+                            LLVMValueRef ilevel,
+                            LLVMValueRef *out_size_vec,
+                            LLVMValueRef *row_stride_vec,
+                            LLVMValueRef *img_stride_vec);
+
+
+void
+lp_build_extract_image_sizes(struct lp_build_sample_context *bld,
+                             struct lp_type size_type,
+                             struct lp_type coord_type,
+                             LLVMValueRef size,
+                             LLVMValueRef *out_width,
+                             LLVMValueRef *out_height,
+                             LLVMValueRef *out_depth);
+
+
+void
+lp_build_unnormalized_coords(struct lp_build_sample_context *bld,
+                             LLVMValueRef flt_size,
+                             LLVMValueRef *s,
+                             LLVMValueRef *t,
+                             LLVMValueRef *r);
+
+
+void
+lp_build_cube_lookup(struct lp_build_sample_context *bld,
+                     LLVMValueRef s,
+                     LLVMValueRef t,
+                     LLVMValueRef r,
+                     LLVMValueRef *face,
+                     LLVMValueRef *face_s,
+                     LLVMValueRef *face_t);
+
+
 void
 lp_build_sample_partial_offset(struct lp_build_context *bld,
                                unsigned block_length,
@@ -184,6 +398,9 @@ lp_build_sample_soa(LLVMBuilderRef builder,
                     LLVMValueRef explicit_lod,
                     LLVMValueRef texel_out[4]);
 
+void
+lp_build_sample_nop(struct lp_type type,
+                    LLVMValueRef texel_out[4]);
 
 
 #endif /* LP_BLD_SAMPLE_H */

src/gallium/auxiliary/gallivm/lp_bld_sample_aos.h

@@ -25,40 +25,32 @@
  *
  **************************************************************************/
 
-
 /**
- * The setup code is concerned with point/line/triangle setup and
- * putting commands/data into the bins.
+ * @file
+ * Texture sampling -- SoA.
+ *
+ * @author Jose Fonseca <jfonseca@vmware.com>
+ * @author Brian Paul <brianp@vmware.com>
  */
 
-
-#ifndef LP_SETUP_COEF_H
-#define LP_SETUP_COEF_H
+#ifndef LP_BLD_SAMPLE_AOS_H
+#define LP_BLD_SAMPLE_AOS_H
 
 
-struct lp_tri_info {
+#include "lp_bld_sample.h"
 
-   float x0_center;
-   float y0_center;
 
-   /* turn these into an aligned float[4] */
-   float dy01_ooa;
-   float dy20_ooa;
-   float dx01_ooa;
-   float dx20_ooa;
+void
+lp_build_sample_aos(struct lp_build_sample_context *bld,
+                    unsigned unit,
+                    LLVMValueRef s,
+                    LLVMValueRef t,
+                    LLVMValueRef r,
+                    const LLVMValueRef *ddx,
+                    const LLVMValueRef *ddy,
+                    LLVMValueRef lod_bias, /* optional */
+                    LLVMValueRef explicit_lod, /* optional */
+                    LLVMValueRef texel_out[4]);
 
-   const float (*v0)[4];
-   const float (*v1)[4];
-   const float (*v2)[4];
 
-   boolean frontfacing;		/* remove eventually */
-};
-
-void lp_setup_tri_coef( struct lp_setup_context *setup,
-			struct lp_rast_shader_inputs *inputs,
-                        const float (*v0)[4],
-                        const float (*v1)[4],
-                        const float (*v2)[4],
-                        boolean frontfacing);
-
-#endif
+#endif /* LP_BLD_SAMPLE_AOS_H */

src/gallium/auxiliary/gallivm/lp_bld_swizzle.c

@@ -76,11 +76,16 @@ lp_build_broadcast_scalar(struct lp_build_context *bld,
    }
    else {
       LLVMValueRef res;
+
 #if HAVE_LLVM >= 0x207
+      /* The shuffle vector is always made of int32 elements */
+      struct lp_type i32_vec_type = lp_type_int_vec(32);
+      i32_vec_type.length = type.length;
+
       res = LLVMBuildInsertElement(bld->builder, bld->undef, scalar,
                                    LLVMConstInt(LLVMInt32Type(), 0, 0), "");
       res = LLVMBuildShuffleVector(bld->builder, res, bld->undef,
-                                   lp_build_const_int_vec(type, 0), "");
+                                   lp_build_const_int_vec(i32_vec_type, 0), "");
 #else
       /* XXX: The above path provokes a bug in LLVM 2.6 */
       unsigned i;
@@ -95,6 +100,83 @@ lp_build_broadcast_scalar(struct lp_build_context *bld,
 }
 
 
+/**
+ * Combined extract and broadcast (or a mere shuffle when the two types match)
+ */
+LLVMValueRef
+lp_build_extract_broadcast(LLVMBuilderRef builder,
+                           struct lp_type src_type,
+                           struct lp_type dst_type,
+                           LLVMValueRef vector,
+                           LLVMValueRef index)
+{
+   LLVMTypeRef i32t = LLVMInt32Type();
+   LLVMValueRef res;
+
+   assert(src_type.floating == dst_type.floating);
+   assert(src_type.width    == dst_type.width);
+
+   assert(lp_check_value(src_type, vector));
+   assert(LLVMTypeOf(index) == i32t);
+
+   if (src_type.length == 1) {
+      if (dst_type.length == 1) {
+         /*
+          * Trivial scalar -> scalar.
+          */
+
+         res = vector;
+      }
+      else {
+         /*
+          * Broadcast scalar -> vector.
+          */
+
+         res = lp_build_broadcast(builder,
+                                  lp_build_vec_type(dst_type),
+                                  vector);
+      }
+   }
+   else {
+      if (dst_type.length == src_type.length) {
+         /*
+          * Special shuffle of the same size.
+          */
+
+         LLVMValueRef shuffle;
+         shuffle = lp_build_broadcast(builder,
+                                      LLVMVectorType(i32t, dst_type.length),
+                                      index);
+         res = LLVMBuildShuffleVector(builder, vector,
+                                      LLVMGetUndef(lp_build_vec_type(dst_type)),
+                                      shuffle, "");
+      }
+      else {
+         LLVMValueRef scalar;
+         scalar = LLVMBuildExtractElement(builder, vector, index, "");
+         if (dst_type.length == 1) {
+            /*
+             * Trivial extract scalar from vector.
+             */
+
+            res = scalar;
+         }
+         else {
+            /*
+             * General case of different sized vectors.
+             */
+
+            res = lp_build_broadcast(builder,
+                                     lp_build_vec_type(dst_type),
+                                     vector);
+         }
+      }
+   }
+
+   return res;
+}
+
+
 /**
  * Swizzle one channel into all other three channels.
  */