vk: Remove unneeded vulkan-138.h

This matches what we do for surface state and makes the dynamic state pool
more opaque to things that need to get dynamic state.

configure.ac

@@ -1153,6 +1153,10 @@ AC_ARG_ENABLE([driglx-direct],
     [driglx_direct="$enableval"],
     [driglx_direct="yes"])
 
+# Check for libcaca
+PKG_CHECK_EXISTS([caca], [have_libcaca=yes], [have_libcaca=no])
+AM_CONDITIONAL([HAVE_LIBCACA], [test x$have_libcaca = xyes])
+
 dnl
 dnl libGL configuration per driver
 dnl
@@ -1529,6 +1533,8 @@ GBM_PC_LIB_PRIV="$DLOPEN_LIBS"
 AC_SUBST([GBM_PC_REQ_PRIV])
 AC_SUBST([GBM_PC_LIB_PRIV])
 
+AM_CONDITIONAL(HAVE_VULKAN, true)
+
 dnl
 dnl EGL configuration
 dnl
@@ -2318,6 +2324,13 @@ AC_SUBST([XA_MINOR], $XA_MINOR)
 AC_SUBST([XA_TINY], $XA_TINY)
 AC_SUBST([XA_VERSION], "$XA_MAJOR.$XA_MINOR.$XA_TINY")
 
+PKG_CHECK_MODULES(VALGRIND, [valgrind],
+                  [have_valgrind=yes], [have_valgrind=no])
+if test "x$have_valgrind" = "xyes"; then
+    AC_DEFINE([HAVE_VALGRIND], 1,
+              [Use valgrind intrinsics to suppress false warnings])
+fi
+
 dnl Restore LDFLAGS and CPPFLAGS
 LDFLAGS="$_SAVE_LDFLAGS"
 CPPFLAGS="$_SAVE_CPPFLAGS"
@@ -2425,6 +2438,7 @@ AC_CONFIG_FILES([Makefile
 		src/mesa/drivers/osmesa/osmesa.pc
 		src/mesa/drivers/x11/Makefile
 		src/mesa/main/tests/Makefile
+		src/vulkan/Makefile
 		src/util/Makefile
 		src/util/tests/hash_table/Makefile])
 

include/vulkan/vk_platform.h

@@ -0,0 +1,90 @@
+//
+// File: vk_platform.h
+//
+/*
+** Copyright (c) 2014-2015 The Khronos Group Inc.
+**
+** Permission is hereby granted, free of charge, to any person obtaining a
+** copy of this software and/or associated documentation files (the
+** "Materials"), to deal in the Materials without restriction, including
+** without limitation the rights to use, copy, modify, merge, publish,
+** distribute, sublicense, and/or sell copies of the Materials, and to
+** permit persons to whom the Materials are furnished to do so, subject to
+** the following conditions:
+**
+** The above copyright notice and this permission notice shall be included
+** in all copies or substantial portions of the Materials.
+**
+** THE MATERIALS ARE 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 NONINFRINGEMENT.
+** IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 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
+** MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
+*/
+
+
+#ifndef __VK_PLATFORM_H__
+#define __VK_PLATFORM_H__
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif // __cplusplus
+
+/*
+***************************************************************************************************
+*   Platform-specific directives and type declarations
+***************************************************************************************************
+*/
+
+#if defined(_WIN32)
+    // On Windows, VKAPI should equate to the __stdcall convention
+    #define VKAPI   __stdcall
+#elif defined(__GNUC__)
+    // On other platforms using GCC, VKAPI stays undefined
+    #define VKAPI
+#else
+    // Unsupported Platform!
+    #error "Unsupported OS Platform detected!"
+#endif
+
+#include <stddef.h>
+
+#if !defined(VK_NO_STDINT_H)
+    #if defined(_MSC_VER) && (_MSC_VER < 1600)
+        typedef signed   __int8  int8_t;
+        typedef unsigned __int8  uint8_t;
+        typedef signed   __int16 int16_t;
+        typedef unsigned __int16 uint16_t;
+        typedef signed   __int32 int32_t;
+        typedef unsigned __int32 uint32_t;
+        typedef signed   __int64 int64_t;
+        typedef unsigned __int64 uint64_t;
+    #else
+        #include <stdint.h>
+    #endif
+#endif // !defined(VK_NO_STDINT_H)
+
+typedef uint64_t   VkDeviceSize;
+typedef uint32_t   VkBool32;
+
+typedef uint32_t   VkSampleMask;
+typedef uint32_t   VkFlags;
+
+#if (UINTPTR_MAX >= UINT64_MAX)
+    #define VK_UINTPTRLEAST64_MAX UINTPTR_MAX
+
+    typedef uintptr_t VkUintPtrLeast64;
+#else
+    #define VK_UINTPTRLEAST64_MAX UINT64_MAX
+
+    typedef uint64_t  VkUintPtrLeast64;
+#endif
+
+#ifdef __cplusplus
+} // extern "C"
+#endif // __cplusplus
+
+#endif // __VK_PLATFORM_H__

include/vulkan/vk_wsi_lunarg.h

@@ -0,0 +1,212 @@
+//
+// File: vk_wsi_display.h
+//
+/*
+** Copyright (c) 2014 The Khronos Group Inc.
+**
+** Permission is hereby granted, free of charge, to any person obtaining a
+** copy of this software and/or associated documentation files (the
+** "Materials"), to deal in the Materials without restriction, including
+** without limitation the rights to use, copy, modify, merge, publish,
+** distribute, sublicense, and/or sell copies of the Materials, and to
+** permit persons to whom the Materials are furnished to do so, subject to
+** the following conditions:
+**
+** The above copyright notice and this permission notice shall be included
+** in all copies or substantial portions of the Materials.
+**
+** THE MATERIALS ARE 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 NONINFRINGEMENT.
+** IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 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
+** MATERIALS OR THE USE OR OTHER DEALINGS IN THE MATERIALS.
+*/
+
+#ifndef __VK_WSI_LUNARG_H__
+#define __VK_WSI_LUNARG_H__
+
+#include "vulkan.h"
+
+#define VK_WSI_LUNARG_REVISION             3
+#define VK_WSI_LUNARG_EXTENSION_NUMBER     1
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif // __cplusplus
+
+// This macro defines INT_MAX in enumerations to force compilers to use 32 bits
+// to represent them. This may or may not be necessary on some compilers. The
+// option to compile it out may allow compilers that warn about missing enumerants
+// in switch statements to be silenced.
+// Using this macro is not needed for flag bit enums because those aren't used
+// as storage type anywhere.
+#define VK_MAX_ENUM(Prefix) VK_##Prefix##_MAX_ENUM = 0x7FFFFFFF
+
+// This macro defines the BEGIN_RANGE, END_RANGE, NUM, and MAX_ENUM constants for
+// the enumerations.
+#define VK_ENUM_RANGE(Prefix, First, Last) \
+    VK_##Prefix##_BEGIN_RANGE                               = VK_##Prefix##_##First, \
+    VK_##Prefix##_END_RANGE                                 = VK_##Prefix##_##Last, \
+    VK_NUM_##Prefix                                         = (VK_##Prefix##_END_RANGE - VK_##Prefix##_BEGIN_RANGE + 1), \
+    VK_MAX_ENUM(Prefix)
+
+// This is a helper macro to define the value of flag bit enum values.
+#define VK_BIT(bit)     (1 << (bit))
+
+// ------------------------------------------------------------------------------------------------
+// Objects
+
+VK_DEFINE_HANDLE(VkDisplayWSI)
+VK_DEFINE_HANDLE(VkSwapChainWSI)
+
+// ------------------------------------------------------------------------------------------------
+// Enumeration constants
+
+#define VK_WSI_LUNARG_ENUM(type,id)    ((type)(VK_WSI_LUNARG_EXTENSION_NUMBER * -1000 + (id)))
+
+// Extend VkPhysicalDeviceInfoType enum with extension specific constants
+#define VK_PHYSICAL_DEVICE_INFO_TYPE_DISPLAY_PROPERTIES_WSI         VK_WSI_LUNARG_ENUM(VkPhysicalDeviceInfoType, 0)
+#define VK_PHYSICAL_DEVICE_INFO_TYPE_QUEUE_PRESENT_PROPERTIES_WSI   VK_WSI_LUNARG_ENUM(VkPhysicalDeviceInfoType, 1)
+
+// Extend VkStructureType enum with extension specific constants
+#define VK_STRUCTURE_TYPE_SWAP_CHAIN_CREATE_INFO_WSI                VK_WSI_LUNARG_ENUM(VkStructureType, 0)
+#define VK_STRUCTURE_TYPE_PRESENT_INFO_WSI                          VK_WSI_LUNARG_ENUM(VkStructureType, 1)
+
+// Extend VkImageLayout enum with extension specific constants
+#define VK_IMAGE_LAYOUT_PRESENT_SOURCE_WSI                          VK_WSI_LUNARG_ENUM(VkImageLayout, 0)
+
+// ------------------------------------------------------------------------------------------------
+// Enumerations
+
+typedef enum VkDisplayInfoTypeWSI_
+{
+    // Info type for vkGetDisplayInfo()
+    VK_DISPLAY_INFO_TYPE_FORMAT_PROPERTIES_WSI              = 0x00000003,   // Return the VkFormat(s) supported for swap chains with the display
+
+    VK_ENUM_RANGE(DISPLAY_INFO_TYPE, FORMAT_PROPERTIES_WSI, FORMAT_PROPERTIES_WSI)
+} VkDisplayInfoTypeWSI;
+
+typedef enum VkSwapChainInfoTypeWSI_
+{
+    // Info type for vkGetSwapChainInfo()
+    VK_SWAP_CHAIN_INFO_TYPE_PERSISTENT_IMAGES_WSI           = 0x00000000,   // Return information about the persistent images of the swapchain
+
+    VK_ENUM_RANGE(SWAP_CHAIN_INFO_TYPE, PERSISTENT_IMAGES_WSI, PERSISTENT_IMAGES_WSI)
+} VkSwapChainInfoTypeWSI;
+
+// ------------------------------------------------------------------------------------------------
+// Flags
+
+typedef VkFlags VkSwapModeFlagsWSI;
+typedef enum VkSwapModeFlagBitsWSI_
+{
+    VK_SWAP_MODE_FLIP_BIT_WSI                               = VK_BIT(0),
+    VK_SWAP_MODE_BLIT_BIT_WSI                               = VK_BIT(1),
+} VkSwapModeFlagBitsWSI;
+
+// ------------------------------------------------------------------------------------------------
+// Structures
+
+typedef struct VkDisplayPropertiesWSI_
+{
+    VkDisplayWSI                            display;            // Handle of the display object
+    VkExtent2D                              physicalResolution; // Max resolution for CRT?
+} VkDisplayPropertiesWSI;
+
+typedef struct VkDisplayFormatPropertiesWSI_
+{
+    VkFormat                                swapChainFormat;    // Format of the images of the swap chain
+} VkDisplayFormatPropertiesWSI;
+
+typedef struct VkSwapChainCreateInfoWSI_
+{
+    VkStructureType                         sType;              // Must be VK_STRUCTURE_TYPE_SWAP_CHAIN_CREATE_INFO_WSI
+    const void*                             pNext;              // Pointer to next structure
+
+    // TBD: It is not yet clear what the use will be for the following two
+    // values.  It seems to be needed for more-global window-system handles
+    // (e.g. X11 display).  If not needed for the SDK, we will drop it from
+    // this extension, and from a future version of this header.
+    const void*                             pNativeWindowSystemHandle; // Pointer to native window system handle
+    const void*                             pNativeWindowHandle; // Pointer to native window handle
+
+    uint32_t                                displayCount;       // Number of displays the swap chain is created for
+    const VkDisplayWSI*                     pDisplays;          // displayCount number of display objects the swap chain is created for
+
+    uint32_t                                imageCount;         // Number of images in the swap chain
+
+    VkFormat                                imageFormat;        // Format of the images of the swap chain
+    VkExtent2D                              imageExtent;        // Width and height of the images of the swap chain
+    uint32_t                                imageArraySize;     // Number of layers of the images of the swap chain (needed for multi-view rendering)
+    VkFlags                                 imageUsageFlags;    // Usage flags for the images of the swap chain (see VkImageUsageFlags)
+
+    VkFlags                                 swapModeFlags;      // Allowed swap modes (see VkSwapModeFlagsWSI)
+} VkSwapChainCreateInfoWSI;
+
+typedef struct VkSwapChainImageInfoWSI_
+{
+    VkImage                                 image;              // Persistent swap chain image handle
+    VkDeviceMemory                          memory;             // Persistent swap chain image's memory handle
+} VkSwapChainImageInfoWSI;
+
+typedef struct VkPhysicalDeviceQueuePresentPropertiesWSI_
+{
+    VkBool32                                supportsPresent;    // Tells whether the queue supports presenting
+} VkPhysicalDeviceQueuePresentPropertiesWSI;
+
+typedef struct VkPresentInfoWSI_
+{
+    VkStructureType                         sType;              // Must be VK_STRUCTURE_TYPE_PRESENT_INFO_WSI
+    const void*                             pNext;              // Pointer to next structure
+    VkImage                                 image;              // Image to present
+    uint32_t                                flipInterval;       // Flip interval
+} VkPresentInfoWSI;
+
+// ------------------------------------------------------------------------------------------------
+// Function types
+
+typedef VkResult (VKAPI *PFN_vkGetDisplayInfoWSI)(VkDisplayWSI display, VkDisplayInfoTypeWSI infoType, size_t* pDataSize, void* pData);
+typedef VkResult (VKAPI *PFN_vkCreateSwapChainWSI)(VkDevice device, const VkSwapChainCreateInfoWSI* pCreateInfo, VkSwapChainWSI* pSwapChain);
+typedef VkResult (VKAPI *PFN_vkDestroySwapChainWSI)(VkSwapChainWSI swapChain);
+typedef VkResult (VKAPI *PFN_vkGetSwapChainInfoWSI)(VkSwapChainWSI swapChain, VkSwapChainInfoTypeWSI infoType, size_t* pDataSize, void* pData);
+typedef VkResult (VKAPI *PFN_vkQueuePresentWSI)(VkQueue queue, const VkPresentInfoWSI* pPresentInfo);
+
+// ------------------------------------------------------------------------------------------------
+// Function prototypes
+
+#ifdef VK_PROTOTYPES
+
+VkResult VKAPI vkGetDisplayInfoWSI(
+    VkDisplayWSI                            display,
+    VkDisplayInfoTypeWSI                    infoType,
+    size_t*                                 pDataSize,
+    void*                                   pData);
+
+VkResult VKAPI vkCreateSwapChainWSI(
+    VkDevice                                device,
+    const VkSwapChainCreateInfoWSI*         pCreateInfo,
+    VkSwapChainWSI*                         pSwapChain);
+
+VkResult VKAPI vkDestroySwapChainWSI(
+    VkSwapChainWSI                          swapChain);
+
+VkResult VKAPI vkGetSwapChainInfoWSI(
+    VkSwapChainWSI                          swapChain,
+    VkSwapChainInfoTypeWSI                  infoType,
+    size_t*                                 pDataSize,
+    void*                                   pData);
+
+VkResult VKAPI vkQueuePresentWSI(
+    VkQueue                                 queue,
+    const VkPresentInfoWSI*                 pPresentInfo);
+
+#endif // VK_PROTOTYPES
+
+#ifdef __cplusplus
+} // extern "C"
+#endif // __cplusplus
+
+#endif // __VK_WSI_LUNARG_H__

include/vulkan/vulkan_intel.h

@@ -0,0 +1,61 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 __VULKAN_INTEL_H__
+#define __VULKAN_INTEL_H__
+
+#include "vulkan.h"
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif // __cplusplus
+
+#define VK_STRUCTURE_TYPE_DMA_BUF_IMAGE_CREATE_INFO_INTEL 1024
+typedef struct VkDmaBufImageCreateInfo_
+{
+    VkStructureType                             sType;                      // Must be VK_STRUCTURE_TYPE_DMA_BUF_IMAGE_CREATE_INFO_INTEL
+    const void*                                 pNext;                      // Pointer to next structure.
+    int                                         fd;
+    VkFormat                                    format;
+    VkExtent3D                                  extent;         // Depth must be 1
+    uint32_t                                    strideInBytes;
+} VkDmaBufImageCreateInfo;
+
+typedef VkResult (VKAPI *PFN_vkCreateDmaBufImageINTEL)(VkDevice device, const VkDmaBufImageCreateInfo* pCreateInfo, VkDeviceMemory* pMem, VkImage* pImage);
+
+#ifdef VK_PROTOTYPES
+
+VkResult VKAPI vkCreateDmaBufImageINTEL(
+    VkDevice                                    _device,
+    const VkDmaBufImageCreateInfo*             pCreateInfo,
+    VkDeviceMemory*                             pMem,
+    VkImage*                                    pImage);
+
+#endif
+
+#ifdef __cplusplus
+} // extern "C"
+#endif // __cplusplus
+
+#endif // __VULKAN_INTEL_H__

src/Makefile.am

@@ -61,6 +61,10 @@ EXTRA_DIST = \
 AM_CFLAGS = $(VISIBILITY_CFLAGS)
 AM_CXXFLAGS = $(VISIBILITY_CXXFLAGS)
 
+if HAVE_VULKAN
+SUBDIRS += vulkan
+endif
+
 AM_CPPFLAGS = \
 	-I$(top_srcdir)/include/ \
 	-I$(top_srcdir)/src/mapi/ \

src/glsl/.gitignore

@@ -4,6 +4,7 @@ glsl_parser.cpp
 glsl_parser.h
 glsl_parser.output
 glsl_test
+spirv2nir
 subtest-cr/
 subtest-lf/
 subtest-cr-lf/

src/glsl/Makefile.am

@@ -77,7 +77,7 @@ check_PROGRAMS =					\
 	tests/sampler-types-test			\
 	tests/uniform-initializer-test
 
-noinst_PROGRAMS = glsl_compiler
+noinst_PROGRAMS = glsl_compiler spirv2nir
 
 tests_blob_test_SOURCES =				\
 	tests/blob_test.c
@@ -158,6 +158,16 @@ glsl_compiler_LDADD =					\
 	$(top_builddir)/src/libglsl_util.la		\
 	$(PTHREAD_LIBS)
 
+spirv2nir_SOURCES = \
+	standalone_scaffolding.cpp \
+	standalone_scaffolding.h \
+	nir/spirv2nir.c
+
+spirv2nir_LDADD =					\
+	libglsl.la					\
+	$(top_builddir)/src/libglsl_util.la		\
+	$(PTHREAD_LIBS)
+
 glsl_test_SOURCES = \
 	standalone_scaffolding.cpp \
 	test.cpp \

src/glsl/Makefile.sources

@@ -59,6 +59,7 @@ NIR_FILES = \
 	nir/nir_remove_dead_variables.c \
 	nir/nir_search.c \
 	nir/nir_search.h \
+	nir/nir_spirv.h \
 	nir/nir_split_var_copies.c \
 	nir/nir_sweep.c \
 	nir/nir_to_ssa.c \
@@ -68,6 +69,8 @@ NIR_FILES = \
 	nir/nir_worklist.c \
 	nir/nir_worklist.h \
 	nir/nir_types.cpp \
+	nir/spirv_to_nir.c \
+	nir/spirv_glsl450_to_nir.c \
 	$(NIR_GENERATED_FILES)
 
 # libglsl

src/glsl/ast.h

@@ -514,6 +514,10 @@ struct ast_type_qualifier {
          unsigned stream:1; /**< Has stream value assigned  */
          unsigned explicit_stream:1; /**< stream value assigned explicitly by shader code */
          /** \} */
+
+         /** \name Vulkan qualifiers */
+         unsigned vk_set:1;
+
       }
       /** \brief Set of flags, accessed by name. */
       q;
@@ -595,6 +599,11 @@ struct ast_type_qualifier {
     */
    glsl_base_type image_base_type;
 
+   /**
+    * Vulkan descriptor set
+    */
+   int set;
+
    /**
     * Return true if and only if an interpolation qualifier is present.
     */

src/glsl/ast_to_hir.cpp

@@ -970,6 +970,7 @@ do_comparison(void *mem_ctx, int operation, ir_rvalue *op0, ir_rvalue *op1)
    case GLSL_TYPE_SAMPLER:
    case GLSL_TYPE_IMAGE:
    case GLSL_TYPE_INTERFACE:
+   case GLSL_TYPE_FUNCTION:
    case GLSL_TYPE_ATOMIC_UINT:
       /* I assume a comparison of a struct containing a sampler just
        * ignores the sampler present in the type.
@@ -2644,7 +2645,16 @@ apply_type_qualifier_to_variable(const struct ast_type_qualifier *qual,
          state->fs_redeclares_gl_fragcoord_with_no_layout_qualifiers;
    }
 
-   if (qual->flags.q.explicit_location) {
+   if (qual->flags.q.vk_set) {
+      if (!qual->flags.q.explicit_binding)
+         _mesa_glsl_error(loc, state,
+                          "Vulkan descriptor set layout requires both set "
+                          "and binding qualifiers");
+
+      var->data.vk_set = true;
+      var->data.set = qual->set;
+      var->data.binding = qual->binding;
+   } else if (qual->flags.q.explicit_location) {
       validate_explicit_location(qual, var, state, loc);
    } else if (qual->flags.q.explicit_index) {
       _mesa_glsl_error(loc, state, "explicit index requires explicit location");
@@ -5886,6 +5896,10 @@ ast_interface_block::hir(exec_list *instructions,
          var->data.explicit_binding = this->layout.flags.q.explicit_binding;
          var->data.binding = this->layout.binding;
 
+         var->data.vk_set = this->layout.flags.q.vk_set;
+         var->data.set = this->layout.set;
+         var->data.binding = this->layout.binding;
+
          state->symbols->add_variable(var);
          instructions->push_tail(var);
       }
@@ -5958,6 +5972,10 @@ ast_interface_block::hir(exec_list *instructions,
          var->data.explicit_binding = this->layout.flags.q.explicit_binding;
          var->data.binding = this->layout.binding;
 
+         var->data.vk_set = this->layout.flags.q.vk_set;
+         var->data.set = this->layout.set;
+         var->data.binding = this->layout.binding;
+
          state->symbols->add_variable(var);
          instructions->push_tail(var);
       }

src/glsl/ast_type.cpp

@@ -252,6 +252,11 @@ ast_type_qualifier::merge_qualifier(YYLTYPE *loc,
       this->image_base_type = q.image_base_type;
    }
 
+   if (q.flags.q.vk_set) {
+      this->set = q.set;
+      this->binding = q.binding;
+   }
+
    return true;
 }
 

src/glsl/glsl_parser.yy

@@ -1424,13 +1424,16 @@ layout_qualifier_id:
          }
       }
 
-      if ((state->ARB_shading_language_420pack_enable ||
-           state->has_atomic_counters()) &&
-          match_layout_qualifier("binding", $1, state) == 0) {
+      if (match_layout_qualifier("binding", $1, state) == 0) {
          $$.flags.q.explicit_binding = 1;
          $$.binding = $3;
       }
 
+      if (match_layout_qualifier("set", $1, state) == 0) {
+         $$.flags.q.vk_set = 1;
+         $$.set = $3;
+      }
+
       if (state->has_atomic_counters() &&
           match_layout_qualifier("offset", $1, state) == 0) {
          $$.flags.q.explicit_offset = 1;

src/glsl/glsl_parser_extras.cpp

@@ -87,6 +87,8 @@ _mesa_glsl_parse_state::_mesa_glsl_parse_state(struct gl_context *_ctx,
 
    this->extensions = &ctx->Extensions;
 
+   this->ARB_compute_shader_enable = true;
+
    this->Const.MaxLights = ctx->Const.MaxLights;
    this->Const.MaxClipPlanes = ctx->Const.MaxClipPlanes;
    this->Const.MaxTextureUnits = ctx->Const.MaxTextureUnits;

src/glsl/glsl_types.cpp

@@ -32,6 +32,7 @@ mtx_t glsl_type::mutex = _MTX_INITIALIZER_NP;
 hash_table *glsl_type::array_types = NULL;
 hash_table *glsl_type::record_types = NULL;
 hash_table *glsl_type::interface_types = NULL;
+hash_table *glsl_type::function_types = NULL;
 void *glsl_type::mem_ctx = NULL;
 
 void
@@ -159,6 +160,39 @@ glsl_type::glsl_type(const glsl_struct_field *fields, unsigned num_fields,
    mtx_unlock(&glsl_type::mutex);
 }
 
+glsl_type::glsl_type(const glsl_type *return_type,
+                     const glsl_function_param *params, unsigned num_params) :
+   gl_type(0),
+   base_type(GLSL_TYPE_FUNCTION),
+   sampler_dimensionality(0), sampler_shadow(0), sampler_array(0),
+   sampler_type(0), interface_packing(0),
+   vector_elements(0), matrix_columns(0),
+   length(num_params)
+{
+   unsigned int i;
+
+   mtx_lock(&glsl_type::mutex);
+
+   init_ralloc_type_ctx();
+
+   this->fields.parameters = rzalloc_array(this->mem_ctx,
+                                           glsl_function_param, num_params + 1);
+
+   /* We store the return type as the first parameter */
+   this->fields.parameters[0].type = return_type;
+   this->fields.parameters[0].in = false;
+   this->fields.parameters[0].out = true;
+
+   /* We store the i'th parameter in slot i+1 */
+   for (i = 0; i < length; i++) {
+      this->fields.parameters[i + 1].type = params[i].type;
+      this->fields.parameters[i + 1].in = params[i].in;
+      this->fields.parameters[i + 1].out = params[i].out;
+   }
+
+   mtx_unlock(&glsl_type::mutex);
+}
+
 
 bool
 glsl_type::contains_sampler() const
@@ -827,6 +861,72 @@ glsl_type::get_interface_instance(const glsl_struct_field *fields,
 }
 
 
+static int
+function_key_compare(const void *a, const void *b)
+{
+   const glsl_type *const key1 = (glsl_type *) a;
+   const glsl_type *const key2 = (glsl_type *) b;
+
+   if (key1->length != key2->length)
+      return 1;
+
+   return memcmp(key1->fields.parameters, key2->fields.parameters,
+                 (key1->length + 1) * sizeof(*key1->fields.parameters));
+}
+
+
+static unsigned
+function_key_hash(const void *a)
+{
+   const glsl_type *const key = (glsl_type *) a;
+   char hash_key[128];
+   unsigned size = 0;
+
+   size = snprintf(hash_key, sizeof(hash_key), "%08x", key->length);
+
+   for (unsigned i = 0; i < key->length; i++) {
+      if (size >= sizeof(hash_key))
+	 break;
+
+      size += snprintf(& hash_key[size], sizeof(hash_key) - size,
+		       "%p", (void *) key->fields.structure[i].type);
+   }
+
+   return hash_table_string_hash(& hash_key);
+}
+
+const glsl_type *
+glsl_type::get_function_instance(const glsl_type *return_type,
+                                 const glsl_function_param *params,
+                                 unsigned num_params)
+{
+   const glsl_type key(return_type, params, num_params);
+
+   mtx_lock(&glsl_type::mutex);
+
+   if (function_types == NULL) {
+      function_types = hash_table_ctor(64, function_key_hash,
+                                       function_key_compare);
+   }
+
+   const glsl_type *t = (glsl_type *) hash_table_find(function_types, &key);
+   if (t == NULL) {
+      mtx_unlock(&glsl_type::mutex);
+      t = new glsl_type(return_type, params, num_params);
+      mtx_lock(&glsl_type::mutex);
+
+      hash_table_insert(function_types, (void *) t, t);
+   }
+
+   assert(t->base_type == GLSL_TYPE_FUNCTION);
+   assert(t->length == num_params);
+
+   mtx_unlock(&glsl_type::mutex);
+
+   return t;
+}
+
+
 const glsl_type *
 glsl_type::get_mul_type(const glsl_type *type_a, const glsl_type *type_b)
 {
@@ -955,6 +1055,7 @@ glsl_type::component_slots() const
    case GLSL_TYPE_IMAGE:
       return 1;
 
+   case GLSL_TYPE_FUNCTION:
    case GLSL_TYPE_SAMPLER:
    case GLSL_TYPE_ATOMIC_UINT:
    case GLSL_TYPE_VOID:
@@ -1326,6 +1427,7 @@ glsl_type::count_attribute_slots() const
    case GLSL_TYPE_ARRAY:
       return this->length * this->fields.array->count_attribute_slots();
 
+   case GLSL_TYPE_FUNCTION:
    case GLSL_TYPE_SAMPLER:
    case GLSL_TYPE_IMAGE:
    case GLSL_TYPE_ATOMIC_UINT:

src/glsl/glsl_types.h

@@ -56,6 +56,7 @@ enum glsl_base_type {
    GLSL_TYPE_IMAGE,
    GLSL_TYPE_ATOMIC_UINT,
    GLSL_TYPE_STRUCT,
+   GLSL_TYPE_FUNCTION,
    GLSL_TYPE_INTERFACE,
    GLSL_TYPE_ARRAY,
    GLSL_TYPE_VOID,
@@ -178,7 +179,7 @@ struct glsl_type {
     */
    union {
       const struct glsl_type *array;            /**< Type of array elements. */
-      const struct glsl_type *parameters;       /**< Parameters to function. */
+      struct glsl_function_param *parameters;   /**< Parameters to function. */
       struct glsl_struct_field *structure;      /**< List of struct fields. */
    } fields;
 
@@ -263,6 +264,13 @@ struct glsl_type {
 						  enum glsl_interface_packing packing,
 						  const char *block_name);
 
+   /**
+    * Get the instance of a function type
+    */
+   static const glsl_type *get_function_instance(const struct glsl_type *return_type,
+                                                 const glsl_function_param *parameters,
+                                                 unsigned num_params);
+
    /**
     * Get the type resulting from a multiplication of \p type_a * \p type_b
     */
@@ -676,6 +684,10 @@ private:
    glsl_type(const glsl_struct_field *fields, unsigned num_fields,
 	     enum glsl_interface_packing packing, const char *name);
 
+   /** Constructor for interface types */
+   glsl_type(const glsl_type *return_type,
+             const glsl_function_param *params, unsigned num_params);
+
    /** Constructor for array types */
    glsl_type(const glsl_type *array, unsigned length);
 
@@ -688,6 +700,9 @@ private:
    /** Hash table containing the known interface types. */
    static struct hash_table *interface_types;
 
+   /** Hash table containing the known function types. */
+   static struct hash_table *function_types;
+
    static int record_key_compare(const void *a, const void *b);
    static unsigned record_key_hash(const void *key);
 
@@ -715,6 +730,10 @@ private:
    /*@}*/
 };
 
+#undef DECL_TYPE
+#undef STRUCT_TYPE
+#endif /* __cplusplus */
+
 struct glsl_struct_field {
    const struct glsl_type *type;
    const char *name;
@@ -758,14 +777,17 @@ struct glsl_struct_field {
    int stream;
 };
 
+struct glsl_function_param {
+   const struct glsl_type *type;
+
+   bool in;
+   bool out;
+};
+
 static inline unsigned int
 glsl_align(unsigned int a, unsigned int align)
 {
    return (a + align - 1) / align * align;
 }
 
-#undef DECL_TYPE
-#undef STRUCT_TYPE
-#endif /* __cplusplus */
-
 #endif /* GLSL_TYPES_H */

src/glsl/ir.h

@@ -683,6 +683,11 @@ public:
       unsigned explicit_location:1;
       unsigned explicit_index:1;
 
+      /**
+       * Do we have a Vulkan (group, index) qualifier for this variable?
+       */
+      unsigned vk_set:1;
+
       /**
        * Was an initial binding explicitly set in the shader?
        *
@@ -752,8 +757,10 @@ public:
        * \note
        * The GLSL spec only allows the values 0 or 1 for the index in \b dual
        * source blending.
+       *
+       * This is now also used for the Vulkan descriptor set index.
        */
-      unsigned index:1;
+      int16_t index;
 
       /**
        * \brief Layout qualifier for gl_FragDepth.
@@ -801,6 +808,11 @@ public:
        */
       int16_t binding;
 
+      /**
+       * Vulkan descriptor set for the resource.
+       */
+      int16_t set;
+
       /**
        * Storage location of the base of this variable
        *

src/glsl/ir_clone.cpp

@@ -357,6 +357,7 @@ ir_constant::clone(void *mem_ctx, struct hash_table *ht) const
       return c;
    }
 
+   case GLSL_TYPE_FUNCTION:
    case GLSL_TYPE_SAMPLER:
    case GLSL_TYPE_IMAGE:
    case GLSL_TYPE_ATOMIC_UINT:

src/glsl/link_uniform_block_active_visitor.cpp

@@ -53,6 +53,11 @@ process_block(void *mem_ctx, struct hash_table *ht, ir_variable *var)
          b->binding = 0;
       }
 
+      if (var->data.vk_set) {
+         b->set = var->data.set;
+         b->index = var->data.index;
+      }
+
       _mesa_hash_table_insert(ht, var->get_interface_type()->name, (void *) b);
       return b;
    } else {

src/glsl/link_uniform_block_active_visitor.h

@@ -35,6 +35,8 @@ struct link_uniform_block_active {
    unsigned num_array_elements;
 
    unsigned binding;
+   unsigned set;
+   unsigned index;
 
    bool has_instance_name;
    bool has_binding;

src/glsl/link_uniform_blocks.cpp

@@ -293,6 +293,9 @@ link_uniform_blocks(void *mem_ctx,
             blocks[i].NumUniforms =
                (unsigned)(ptrdiff_t)(&variables[parcel.index] - blocks[i].Uniforms);
 
+            blocks[i].Set = b->set;
+            blocks[i].Binding = b->binding;
+
             i++;
          }
       } else {
@@ -311,6 +314,9 @@ link_uniform_blocks(void *mem_ctx,
          blocks[i].NumUniforms =
             (unsigned)(ptrdiff_t)(&variables[parcel.index] - blocks[i].Uniforms);
 
+         blocks[i].Set = b->set;
+         blocks[i].Binding = b->binding;
+
          i++;
       }
    }

src/glsl/link_uniform_initializers.cpp

@@ -88,6 +88,7 @@ copy_constant_to_storage(union gl_constant_value *storage,
       case GLSL_TYPE_IMAGE:
       case GLSL_TYPE_ATOMIC_UINT:
       case GLSL_TYPE_INTERFACE:
+      case GLSL_TYPE_FUNCTION:
       case GLSL_TYPE_VOID:
       case GLSL_TYPE_ERROR:
 	 /* All other types should have already been filtered by other

src/glsl/nir/glsl_to_nir.cpp

@@ -322,6 +322,7 @@ nir_visitor::visit(ir_variable *ir)
    }
 
    var->data.index = ir->data.index;
+   var->data.descriptor_set = ir->data.set;
    var->data.binding = ir->data.binding;
    /* XXX Get rid of buffer_index */
    var->data.atomic.buffer_index = ir->data.binding;

src/glsl/nir/nir.h

@@ -300,6 +300,7 @@ typedef struct {
        *
        * For array types, this represents the binding point for the first element.
        */
+      int descriptor_set;
       int binding;
 
       /**
@@ -510,7 +511,11 @@ typedef struct nir_src {
    bool is_ssa;
 } nir_src;
 
-#define NIR_SRC_INIT (nir_src) { { NULL } }
+#ifdef __cplusplus
+#  define NIR_SRC_INIT nir_src()
+#else
+#  define NIR_SRC_INIT (nir_src) { { NULL } }
+#endif
 
 #define nir_foreach_use(reg_or_ssa_def, src) \
    list_for_each_entry(nir_src, src, &(reg_or_ssa_def)->uses, use_link)
@@ -533,7 +538,11 @@ typedef struct {
    bool is_ssa;
 } nir_dest;
 
-#define NIR_DEST_INIT (nir_dest) { { { NULL } } }
+#ifdef __cplusplus
+#  define NIR_DEST_INIT nir_dest()
+#else
+#  define NIR_DEST_INIT (nir_dest) { { { NULL } } }
+#endif
 
 #define nir_foreach_def(reg, dest) \
    list_for_each_entry(nir_dest, dest, &(reg)->defs, reg.def_link)
@@ -782,6 +791,15 @@ NIR_DEFINE_CAST(nir_deref_as_var, nir_deref, nir_deref_var, deref)
 NIR_DEFINE_CAST(nir_deref_as_array, nir_deref, nir_deref_array, deref)
 NIR_DEFINE_CAST(nir_deref_as_struct, nir_deref, nir_deref_struct, deref)
 
+/** Returns the tail of a deref chain */
+static inline nir_deref *
+nir_deref_tail(nir_deref *deref)
+{
+   while (deref->child)
+      deref = deref->child;
+   return deref;
+}
+
 typedef struct {
    nir_instr instr;
 
@@ -976,6 +994,9 @@ typedef struct {
    /* gather component selector */
    unsigned component : 2;
 
+   /* The descriptor set containing this texture */
+   unsigned sampler_set;
+
    /** The sampler index
     *
     * If this texture instruction has a nir_tex_src_sampler_offset source,
@@ -1661,6 +1682,7 @@ void nir_lower_phis_to_scalar(nir_shader *shader);
 void nir_lower_samplers(nir_shader *shader,
                         const struct gl_shader_program *shader_program,
                         gl_shader_stage stage);
+void nir_lower_samplers_for_vk(nir_shader *shader);
 
 void nir_lower_system_values(nir_shader *shader);
 void nir_lower_tex_projector(nir_shader *shader);

src/glsl/nir/nir_builder.h

@@ -28,6 +28,10 @@ struct exec_list;
 
 typedef struct nir_builder {
    struct exec_list *cf_node_list;
+
+   nir_block *before_block;
+   nir_block *after_block;
+
    nir_instr *before_instr;
    nir_instr *after_instr;
 
@@ -48,6 +52,30 @@ nir_builder_insert_after_cf_list(nir_builder *build,
                                  struct exec_list *cf_node_list)
 {
    build->cf_node_list = cf_node_list;
+   build->before_block = NULL;
+   build->after_block = NULL;
+   build->before_instr = NULL;
+   build->after_instr = NULL;
+}
+
+static inline void
+nir_builder_insert_before_block(nir_builder *build,
+                                nir_block *block)
+{
+   build->cf_node_list = NULL;
+   build->before_block = block;
+   build->after_block = NULL;
+   build->before_instr = NULL;
+   build->after_instr = NULL;
+}
+
+static inline void
+nir_builder_insert_after_block(nir_builder *build,
+                                nir_block *block)
+{
+   build->cf_node_list = NULL;
+   build->before_block = NULL;
+   build->after_block = block;
    build->before_instr = NULL;
    build->after_instr = NULL;
 }
@@ -56,6 +84,8 @@ static inline void
 nir_builder_insert_before_instr(nir_builder *build, nir_instr *before_instr)
 {
    build->cf_node_list = NULL;
+   build->before_block = NULL;
+   build->after_block = NULL;
    build->before_instr = before_instr;
    build->after_instr = NULL;
 }
@@ -64,6 +94,8 @@ static inline void
 nir_builder_insert_after_instr(nir_builder *build, nir_instr *after_instr)
 {
    build->cf_node_list = NULL;
+   build->before_block = NULL;
+   build->after_block = NULL;
    build->before_instr = NULL;
    build->after_instr = after_instr;
 }
@@ -73,6 +105,10 @@ nir_builder_instr_insert(nir_builder *build, nir_instr *instr)
 {
    if (build->cf_node_list) {
       nir_instr_insert_after_cf_list(build->cf_node_list, instr);
+   } else if (build->before_block) {
+      nir_instr_insert_before_block(build->before_block, instr);
+   } else if (build->after_block) {
+      nir_instr_insert_after_block(build->after_block, instr);
    } else if (build->before_instr) {
       nir_instr_insert_before(build->before_instr, instr);
    } else {
@@ -240,6 +276,23 @@ nir_swizzle(nir_builder *build, nir_ssa_def *src, unsigned swiz[4],
                      nir_imov_alu(build, alu_src, num_components);
 }
 
+/* Selects the right fdot given the number of components in each source. */
+static inline nir_ssa_def *
+nir_fdot(nir_builder *build, nir_ssa_def *src0, nir_ssa_def *src1)
+{
+   assert(src0->num_components == src1->num_components);
+   switch (src0->num_components) {
+   case 1: return nir_fmul(build, src0, src1);
+   case 2: return nir_fdot2(build, src0, src1);
+   case 3: return nir_fdot3(build, src0, src1);
+   case 4: return nir_fdot4(build, src0, src1);
+   default:
+      unreachable("bad component size");
+   }
+
+   return NULL;
+}
+
 /**
  * Turns a nir_src into a nir_ssa_def * so it can be passed to
  * nir_build_alu()-based builder calls.

src/glsl/nir/nir_lower_io.c

@@ -67,6 +67,7 @@ type_size(const struct glsl_type *type)
       return 0;
    case GLSL_TYPE_IMAGE:
       return 0;
+   case GLSL_TYPE_FUNCTION:
    case GLSL_TYPE_VOID:
    case GLSL_TYPE_ERROR:
    case GLSL_TYPE_DOUBLE:

src/glsl/nir/nir_lower_samplers.cpp

@@ -35,6 +35,30 @@ extern "C" {
 #include "program/program.h"
 }
 
+static void
+add_indirect_to_tex(nir_tex_instr *instr, nir_src indirect)
+{
+   /* First, we have to resize the array of texture sources */
+   nir_tex_src *new_srcs = rzalloc_array(instr, nir_tex_src,
+                                         instr->num_srcs + 1);
+
+   for (unsigned i = 0; i < instr->num_srcs; i++) {
+      new_srcs[i].src_type = instr->src[i].src_type;
+      nir_instr_move_src(&instr->instr, &new_srcs[i].src, &instr->src[i].src);
+   }
+
+   ralloc_free(instr->src);
+   instr->src = new_srcs;
+
+   /* Now we can go ahead and move the source over to being a
+    * first-class texture source.
+    */
+   instr->src[instr->num_srcs].src_type = nir_tex_src_sampler_offset;
+   instr->num_srcs++;
+   nir_instr_rewrite_src(&instr->instr, &instr->src[instr->num_srcs - 1].src,
+                         indirect);
+}
+
 static unsigned
 get_sampler_index(const struct gl_shader_program *shader_program,
                   gl_shader_stage stage, const char *name)
@@ -176,3 +200,49 @@ nir_lower_samplers(nir_shader *shader, const struct gl_shader_program *shader_pr
          lower_impl(overload->impl, shader_program, stage);
    }
 }
+
+static bool
+lower_samplers_for_vk_block(nir_block *block, void *data)
+{
+   nir_foreach_instr(block, instr) {
+      if (instr->type != nir_instr_type_tex)
+         continue;
+
+      nir_tex_instr *tex = nir_instr_as_tex(instr);
+
+      assert(tex->sampler);
+
+      tex->sampler_set = tex->sampler->var->data.descriptor_set;
+      tex->sampler_index = tex->sampler->var->data.binding;
+
+      if (tex->sampler->deref.child) {
+         assert(tex->sampler->deref.child->deref_type == nir_deref_type_array);
+         nir_deref_array *arr = nir_deref_as_array(tex->sampler->deref.child);
+
+         /* Only one-level arrays are allowed in vulkan */
+         assert(arr->deref.child == NULL);
+
+         tex->sampler_index += arr->base_offset;
+         if (arr->deref_array_type == nir_deref_array_type_indirect) {
+            add_indirect_to_tex(tex, arr->indirect);
+            nir_instr_rewrite_src(instr, &arr->indirect, NIR_SRC_INIT);
+
+            tex->sampler_array_size = glsl_get_length(tex->sampler->deref.type);
+         }
+      }
+
+      tex->sampler = NULL;
+   }
+
+   return true;
+}
+
+extern "C" void
+nir_lower_samplers_for_vk(nir_shader *shader)
+{
+   nir_foreach_overload(shader, overload) {
+      if (overload->impl) {
+         nir_foreach_block(overload->impl, lower_samplers_for_vk_block, NULL);
+      }
+   }
+}

src/glsl/nir/nir_lower_var_copies.c

@@ -53,17 +53,6 @@ deref_next_wildcard_parent(nir_deref *deref)
    return NULL;
 }
 
-/* Returns the last deref in the chain.
- */
-static nir_deref *
-get_deref_tail(nir_deref *deref)
-{
-   while (deref->child)
-      deref = deref->child;
-
-   return deref;
-}
-
 /* This function recursively walks the given deref chain and replaces the
  * given copy instruction with an equivalent sequence load/store
  * operations.
@@ -121,8 +110,8 @@ emit_copy_load_store(nir_intrinsic_instr *copy_instr,
    } else {
       /* In this case, we have no wildcards anymore, so all we have to do
        * is just emit the load and store operations. */
-      src_tail = get_deref_tail(src_tail);
-      dest_tail = get_deref_tail(dest_tail);
+      src_tail = nir_deref_tail(src_tail);
+      dest_tail = nir_deref_tail(dest_tail);
 
       assert(src_tail->type == dest_tail->type);
 

src/glsl/nir/nir_spirv.h

@@ -0,0 +1,46 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ *
+ * Authors:
+ *    Jason Ekstrand (jason@jlekstrand.net)
+ *
+ */
+
+#pragma once
+
+#ifndef _NIR_SPIRV_H_
+#define _NIR_SPIRV_H_
+
+#include "nir.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+nir_shader *spirv_to_nir(const uint32_t *words, size_t word_count,
+                         const nir_shader_compiler_options *options);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _NIR_SPIRV_H_ */

src/glsl/nir/nir_split_var_copies.c

@@ -66,14 +66,6 @@ struct split_var_copies_state {
    void *dead_ctx;
 };
 
-static nir_deref *
-get_deref_tail(nir_deref *deref)
-{
-   while (deref->child != NULL)
-      deref = deref->child;
-   return deref;
-}
-
 /* Recursively constructs deref chains to split a copy instruction into
  * multiple (if needed) copy instructions with full-length deref chains.
  * External callers of this function should pass the tail and head of the
@@ -225,8 +217,8 @@ split_var_copies_block(nir_block *block, void *void_state)
 
       nir_deref *dest_head = &intrinsic->variables[0]->deref;
       nir_deref *src_head = &intrinsic->variables[1]->deref;
-      nir_deref *dest_tail = get_deref_tail(dest_head);
-      nir_deref *src_tail = get_deref_tail(src_head);
+      nir_deref *dest_tail = nir_deref_tail(dest_head);
+      nir_deref *src_tail = nir_deref_tail(src_head);
 
       switch (glsl_get_base_type(src_tail->type)) {
       case GLSL_TYPE_ARRAY:

src/glsl/nir/nir_types.cpp

@@ -70,6 +70,18 @@ glsl_get_struct_field(const glsl_type *type, unsigned index)
    return type->fields.structure[index].type;
 }
 
+const glsl_type *
+glsl_get_function_return_type(const glsl_type *type)
+{
+   return type->fields.parameters[0].type;
+}
+
+const glsl_function_param *
+glsl_get_function_param(const glsl_type *type, unsigned index)
+{
+   return &type->fields.parameters[index + 1];
+}
+
 const struct glsl_type *
 glsl_get_column_type(const struct glsl_type *type)
 {
@@ -112,6 +124,20 @@ glsl_get_struct_elem_name(const struct glsl_type *type, unsigned index)
    return type->fields.structure[index].name;
 }
 
+glsl_sampler_dim
+glsl_get_sampler_dim(const struct glsl_type *type)
+{
+   assert(glsl_type_is_sampler(type));
+   return (glsl_sampler_dim)type->sampler_dimensionality;
+}
+
+glsl_base_type
+glsl_get_sampler_result_type(const struct glsl_type *type)
+{
+   assert(glsl_type_is_sampler(type));
+   return (glsl_base_type)type->sampler_type;
+}
+
 bool
 glsl_type_is_void(const glsl_type *type)
 {
@@ -130,12 +156,38 @@ glsl_type_is_scalar(const struct glsl_type *type)
    return type->is_scalar();
 }
 
+bool
+glsl_type_is_vector_or_scalar(const struct glsl_type *type)
+{
+   return type->is_vector() || type->is_scalar();
+}
+
 bool
 glsl_type_is_matrix(const struct glsl_type *type)
 {
    return type->is_matrix();
 }
 
+bool
+glsl_type_is_sampler(const struct glsl_type *type)
+{
+   return type->is_sampler();
+}
+
+bool
+glsl_sampler_type_is_shadow(const struct glsl_type *type)
+{
+   assert(glsl_type_is_sampler(type));
+   return type->sampler_shadow;
+}
+
+bool
+glsl_sampler_type_is_array(const struct glsl_type *type)
+{
+   assert(glsl_type_is_sampler(type));
+   return type->sampler_array;
+}
+
 const glsl_type *
 glsl_void_type(void)
 {
@@ -148,14 +200,80 @@ glsl_float_type(void)
    return glsl_type::float_type;
 }
 
+const glsl_type *
+glsl_int_type(void)
+{
+   return glsl_type::int_type;
+}
+
+const glsl_type *
+glsl_uint_type(void)
+{
+   return glsl_type::uint_type;
+}
+
+const glsl_type *
+glsl_bool_type(void)
+{
+   return glsl_type::bool_type;
+}
+
 const glsl_type *
 glsl_vec4_type(void)
 {
    return glsl_type::vec4_type;
 }
 
+const glsl_type *
+glsl_scalar_type(enum glsl_base_type base_type)
+{
+   return glsl_type::get_instance(base_type, 1, 1);
+}
+
+const glsl_type *
+glsl_vector_type(enum glsl_base_type base_type, unsigned components)
+{
+   assert(components > 1 && components <= 4);
+   return glsl_type::get_instance(base_type, components, 1);
+}
+
+const glsl_type *
+glsl_matrix_type(enum glsl_base_type base_type, unsigned rows, unsigned columns)
+{
+   assert(rows >= 1 && rows <= 4 && columns >= 1 && columns <= 4);
+   return glsl_type::get_instance(base_type, rows, columns);
+}
+
 const glsl_type *
 glsl_array_type(const glsl_type *base, unsigned elements)
 {
    return glsl_type::get_array_instance(base, elements);
 }
+
+const glsl_type *
+glsl_struct_type(const glsl_struct_field *fields,
+                 unsigned num_fields, const char *name)
+{
+   return glsl_type::get_record_instance(fields, num_fields, name);
+}
+
+const struct glsl_type *
+glsl_sampler_type(enum glsl_sampler_dim dim, bool is_shadow, bool is_array,
+                  enum glsl_base_type base_type)
+{
+   return glsl_type::get_sampler_instance(dim, is_shadow, is_array, base_type);
+}
+
+const glsl_type *
+glsl_function_type(const glsl_type *return_type,
+                   const glsl_function_param *params, unsigned num_params)
+{
+   return glsl_type::get_function_instance(return_type, params, num_params);
+}
+
+const glsl_type *
+glsl_transposed_type(const struct glsl_type *type)
+{
+   return glsl_type::get_instance(type->base_type, type->matrix_columns,
+                                  type->vector_elements);
+}

src/glsl/nir/nir_types.h

@@ -49,6 +49,12 @@ const struct glsl_type *glsl_get_array_element(const struct glsl_type *type);
 
 const struct glsl_type *glsl_get_column_type(const struct glsl_type *type);
 
+const struct glsl_type *
+glsl_get_function_return_type(const struct glsl_type *type);
+
+const struct glsl_function_param *
+glsl_get_function_param(const struct glsl_type *type, unsigned index);
+
 enum glsl_base_type glsl_get_base_type(const struct glsl_type *type);
 
 unsigned glsl_get_vector_elements(const struct glsl_type *type);
@@ -62,17 +68,42 @@ unsigned glsl_get_length(const struct glsl_type *type);
 const char *glsl_get_struct_elem_name(const struct glsl_type *type,
                                       unsigned index);
 
+enum glsl_sampler_dim glsl_get_sampler_dim(const struct glsl_type *type);
+enum glsl_base_type glsl_get_sampler_result_type(const struct glsl_type *type);
 
 bool glsl_type_is_void(const struct glsl_type *type);
 bool glsl_type_is_vector(const struct glsl_type *type);
 bool glsl_type_is_scalar(const struct glsl_type *type);
+bool glsl_type_is_vector_or_scalar(const struct glsl_type *type);
 bool glsl_type_is_matrix(const struct glsl_type *type);
+bool glsl_type_is_sampler(const struct glsl_type *type);
+bool glsl_sampler_type_is_shadow(const struct glsl_type *type);
+bool glsl_sampler_type_is_array(const struct glsl_type *type);
 
 const struct glsl_type *glsl_void_type(void);
 const struct glsl_type *glsl_float_type(void);
+const struct glsl_type *glsl_int_type(void);
+const struct glsl_type *glsl_uint_type(void);
+const struct glsl_type *glsl_bool_type(void);
+
 const struct glsl_type *glsl_vec4_type(void);
+const struct glsl_type *glsl_scalar_type(enum glsl_base_type base_type);
+const struct glsl_type *glsl_vector_type(enum glsl_base_type base_type,
+                                         unsigned components);
+const struct glsl_type *glsl_matrix_type(enum glsl_base_type base_type,
+                                         unsigned rows, unsigned columns);
 const struct glsl_type *glsl_array_type(const struct glsl_type *base,
                                         unsigned elements);
+const struct glsl_type *glsl_struct_type(const struct glsl_struct_field *fields,
+                                         unsigned num_fields, const char *name);
+const struct glsl_type *glsl_sampler_type(enum glsl_sampler_dim dim,
+                                          bool is_shadow, bool is_array,
+                                          enum glsl_base_type base_type);
+const struct glsl_type * glsl_function_type(const struct glsl_type *return_type,
+                                            const struct glsl_function_param *params,
+                                            unsigned num_params);
+
+const struct glsl_type *glsl_transposed_type(const struct glsl_type *type);
 
 #ifdef __cplusplus
 }

src/glsl/nir/spirv.h

@@ -0,0 +1,808 @@
+/*
+** Copyright (c) 2015 The Khronos Group Inc.
+** 
+** Permission is hereby granted, free of charge, to any person obtaining a copy
+** of this software and/or associated documentation files (the "Materials"),
+** to deal in the Materials without restriction, including without limitation
+** the rights to use, copy, modify, merge, publish, distribute, sublicense,
+** and/or sell copies of the Materials, and to permit persons to whom the
+** Materials are furnished to do so, subject to the following conditions:
+** 
+** The above copyright notice and this permission notice shall be included in
+** all copies or substantial portions of the Materials.
+** 
+** MODIFICATIONS TO THIS FILE MAY MEAN IT NO LONGER ACCURATELY REFLECTS KHRONOS
+** STANDARDS. THE UNMODIFIED, NORMATIVE VERSIONS OF KHRONOS SPECIFICATIONS AND
+** HEADER INFORMATION ARE LOCATED AT https://www.khronos.org/registry/ 
+** 
+** THE MATERIALS ARE 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 NONINFRINGEMENT. IN NO EVENT SHALL
+** THE AUTHORS OR COPYRIGHT HOLDERS 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 MATERIALS OR THE USE OR OTHER DEALINGS
+** IN THE MATERIALS.
+*/
+
+/*
+** This header is automatically generated by the same tool that creates
+** the Binary Section of the SPIR-V specification.
+*/
+
+/*
+** Specification revision 31.
+** Enumeration tokens for SPIR-V, in various styles:
+**   C, C++, C++11, JSON, Lua, Python
+** 
+** - C will have tokens with a "Spv" prefix, e.g.: SpvSourceLanguageGLSL
+** - C++ will have tokens in the "spv" name space, e.g.: spv::SourceLanguageGLSL
+** - C++11 will use enum classes in the spv namespace, e.g.: spv::SourceLanguage::GLSL
+** - Lua will use tables, e.g.: spv.SourceLanguage.GLSL
+** - Python will use dictionaries, e.g.: spv['SourceLanguage']['GLSL']
+** 
+** Some tokens act like mask values, which can be OR'd together,
+** while others are mutually exclusive.  The mask-like ones have
+** "Mask" in their name, and a parallel enum that has the shift
+** amount (1 << x) for each corresponding enumerant.
+*/
+
+#ifndef spirv_H
+#define spirv_H
+
+typedef unsigned int SpvId;
+
+static const unsigned int SpvMagicNumber = 0x07230203;
+static const unsigned int SpvVersion = 99;
+static const unsigned int SpvOpCodeMask = 0xffff;
+static const unsigned int SpvWordCountShift = 16;
+
+typedef enum SpvSourceLanguage_ {
+    SpvSourceLanguageUnknown = 0,
+    SpvSourceLanguageESSL = 1,
+    SpvSourceLanguageGLSL = 2,
+    SpvSourceLanguageOpenCL = 3,
+} SpvSourceLanguage;
+
+typedef enum SpvExecutionModel_ {
+    SpvExecutionModelVertex = 0,
+    SpvExecutionModelTessellationControl = 1,
+    SpvExecutionModelTessellationEvaluation = 2,
+    SpvExecutionModelGeometry = 3,
+    SpvExecutionModelFragment = 4,
+    SpvExecutionModelGLCompute = 5,
+    SpvExecutionModelKernel = 6,
+} SpvExecutionModel;
+
+typedef enum SpvAddressingModel_ {
+    SpvAddressingModelLogical = 0,
+    SpvAddressingModelPhysical32 = 1,
+    SpvAddressingModelPhysical64 = 2,
+} SpvAddressingModel;
+
+typedef enum SpvMemoryModel_ {
+    SpvMemoryModelSimple = 0,
+    SpvMemoryModelGLSL450 = 1,
+    SpvMemoryModelOpenCL = 2,
+} SpvMemoryModel;
+
+typedef enum SpvExecutionMode_ {
+    SpvExecutionModeInvocations = 0,
+    SpvExecutionModeSpacingEqual = 1,
+    SpvExecutionModeSpacingFractionalEven = 2,
+    SpvExecutionModeSpacingFractionalOdd = 3,
+    SpvExecutionModeVertexOrderCw = 4,
+    SpvExecutionModeVertexOrderCcw = 5,
+    SpvExecutionModePixelCenterInteger = 6,
+    SpvExecutionModeOriginUpperLeft = 7,
+    SpvExecutionModeOriginLowerLeft = 8,
+    SpvExecutionModeEarlyFragmentTests = 9,
+    SpvExecutionModePointMode = 10,
+    SpvExecutionModeXfb = 11,
+    SpvExecutionModeDepthReplacing = 12,
+    SpvExecutionModeDepthAny = 13,
+    SpvExecutionModeDepthGreater = 14,
+    SpvExecutionModeDepthLess = 15,
+    SpvExecutionModeDepthUnchanged = 16,
+    SpvExecutionModeLocalSize = 17,
+    SpvExecutionModeLocalSizeHint = 18,
+    SpvExecutionModeInputPoints = 19,
+    SpvExecutionModeInputLines = 20,
+    SpvExecutionModeInputLinesAdjacency = 21,
+    SpvExecutionModeInputTriangles = 22,
+    SpvExecutionModeInputTrianglesAdjacency = 23,
+    SpvExecutionModeInputQuads = 24,
+    SpvExecutionModeInputIsolines = 25,
+    SpvExecutionModeOutputVertices = 26,
+    SpvExecutionModeOutputPoints = 27,
+    SpvExecutionModeOutputLineStrip = 28,
+    SpvExecutionModeOutputTriangleStrip = 29,
+    SpvExecutionModeVecTypeHint = 30,
+    SpvExecutionModeContractionOff = 31,
+} SpvExecutionMode;
+
+typedef enum SpvStorageClass_ {
+    SpvStorageClassUniformConstant = 0,
+    SpvStorageClassInput = 1,
+    SpvStorageClassUniform = 2,
+    SpvStorageClassOutput = 3,
+    SpvStorageClassWorkgroupLocal = 4,
+    SpvStorageClassWorkgroupGlobal = 5,
+    SpvStorageClassPrivateGlobal = 6,
+    SpvStorageClassFunction = 7,
+    SpvStorageClassGeneric = 8,
+    SpvStorageClassPushConstant = 9,
+    SpvStorageClassAtomicCounter = 10,
+    SpvStorageClassImage = 11,
+} SpvStorageClass;
+
+typedef enum SpvDim_ {
+    SpvDim1D = 0,
+    SpvDim2D = 1,
+    SpvDim3D = 2,
+    SpvDimCube = 3,
+    SpvDimRect = 4,
+    SpvDimBuffer = 5,
+} SpvDim;
+
+typedef enum SpvSamplerAddressingMode_ {
+    SpvSamplerAddressingModeNone = 0,
+    SpvSamplerAddressingModeClampToEdge = 1,
+    SpvSamplerAddressingModeClamp = 2,
+    SpvSamplerAddressingModeRepeat = 3,
+    SpvSamplerAddressingModeRepeatMirrored = 4,
+} SpvSamplerAddressingMode;
+
+typedef enum SpvSamplerFilterMode_ {
+    SpvSamplerFilterModeNearest = 0,
+    SpvSamplerFilterModeLinear = 1,
+} SpvSamplerFilterMode;
+
+typedef enum SpvImageFormat_ {
+    SpvImageFormatUnknown = 0,
+    SpvImageFormatRgba32f = 1,
+    SpvImageFormatRgba16f = 2,
+    SpvImageFormatR32f = 3,
+    SpvImageFormatRgba8 = 4,
+    SpvImageFormatRgba8Snorm = 5,
+    SpvImageFormatRg32f = 6,
+    SpvImageFormatRg16f = 7,
+    SpvImageFormatR11fG11fB10f = 8,
+    SpvImageFormatR16f = 9,
+    SpvImageFormatRgba16 = 10,
+    SpvImageFormatRgb10A2 = 11,
+    SpvImageFormatRg16 = 12,
+    SpvImageFormatRg8 = 13,
+    SpvImageFormatR16 = 14,
+    SpvImageFormatR8 = 15,
+    SpvImageFormatRgba16Snorm = 16,
+    SpvImageFormatRg16Snorm = 17,
+    SpvImageFormatRg8Snorm = 18,
+    SpvImageFormatR16Snorm = 19,
+    SpvImageFormatR8Snorm = 20,
+    SpvImageFormatRgba32i = 21,
+    SpvImageFormatRgba16i = 22,
+    SpvImageFormatRgba8i = 23,
+    SpvImageFormatR32i = 24,
+    SpvImageFormatRg32i = 25,
+    SpvImageFormatRg16i = 26,
+    SpvImageFormatRg8i = 27,
+    SpvImageFormatR16i = 28,
+    SpvImageFormatR8i = 29,
+    SpvImageFormatRgba32ui = 30,
+    SpvImageFormatRgba16ui = 31,
+    SpvImageFormatRgba8ui = 32,
+    SpvImageFormatR32ui = 33,
+    SpvImageFormatRgb10a2ui = 34,
+    SpvImageFormatRg32ui = 35,
+    SpvImageFormatRg16ui = 36,
+    SpvImageFormatRg8ui = 37,
+    SpvImageFormatR16ui = 38,
+    SpvImageFormatR8ui = 39,
+} SpvImageFormat;
+
+typedef enum SpvImageChannelOrder_ {
+    SpvImageChannelOrderR = 0,
+    SpvImageChannelOrderA = 1,
+    SpvImageChannelOrderRG = 2,
+    SpvImageChannelOrderRA = 3,
+    SpvImageChannelOrderRGB = 4,
+    SpvImageChannelOrderRGBA = 5,
+    SpvImageChannelOrderBGRA = 6,
+    SpvImageChannelOrderARGB = 7,
+    SpvImageChannelOrderIntensity = 8,
+    SpvImageChannelOrderLuminance = 9,
+    SpvImageChannelOrderRx = 10,
+    SpvImageChannelOrderRGx = 11,
+    SpvImageChannelOrderRGBx = 12,
+    SpvImageChannelOrderDepth = 13,
+    SpvImageChannelOrderDepthStencil = 14,
+    SpvImageChannelOrdersRGB = 15,
+    SpvImageChannelOrdersRGBx = 16,
+    SpvImageChannelOrdersRGBA = 17,
+    SpvImageChannelOrdersBGRA = 18,
+} SpvImageChannelOrder;
+
+typedef enum SpvImageChannelDataType_ {
+    SpvImageChannelDataTypeSnormInt8 = 0,
+    SpvImageChannelDataTypeSnormInt16 = 1,
+    SpvImageChannelDataTypeUnormInt8 = 2,
+    SpvImageChannelDataTypeUnormInt16 = 3,
+    SpvImageChannelDataTypeUnormShort565 = 4,
+    SpvImageChannelDataTypeUnormShort555 = 5,
+    SpvImageChannelDataTypeUnormInt101010 = 6,
+    SpvImageChannelDataTypeSignedInt8 = 7,
+    SpvImageChannelDataTypeSignedInt16 = 8,
+    SpvImageChannelDataTypeSignedInt32 = 9,
+    SpvImageChannelDataTypeUnsignedInt8 = 10,
+    SpvImageChannelDataTypeUnsignedInt16 = 11,
+    SpvImageChannelDataTypeUnsignedInt32 = 12,
+    SpvImageChannelDataTypeHalfFloat = 13,
+    SpvImageChannelDataTypeFloat = 14,
+    SpvImageChannelDataTypeUnormInt24 = 15,
+} SpvImageChannelDataType;
+
+typedef enum SpvImageOperandsShift_ {
+    SpvImageOperandsBiasShift = 0,
+    SpvImageOperandsLodShift = 1,
+    SpvImageOperandsGradShift = 2,
+    SpvImageOperandsOffsetShift = 3,
+    SpvImageOperandsOffsetsShift = 4,
+    SpvImageOperandsSampleShift = 5,
+} SpvImageOperandsShift;
+
+typedef enum SpvImageOperandsMask_ {
+    SpvImageOperandsMaskNone = 0,
+    SpvImageOperandsBiasMask = 0x00000001,
+    SpvImageOperandsLodMask = 0x00000002,
+    SpvImageOperandsGradMask = 0x00000004,
+    SpvImageOperandsOffsetMask = 0x00000008,
+    SpvImageOperandsOffsetsMask = 0x00000010,
+    SpvImageOperandsSampleMask = 0x00000020,
+} SpvImageOperandsMask;
+
+typedef enum SpvFPFastMathModeShift_ {
+    SpvFPFastMathModeNotNaNShift = 0,
+    SpvFPFastMathModeNotInfShift = 1,
+    SpvFPFastMathModeNSZShift = 2,
+    SpvFPFastMathModeAllowRecipShift = 3,
+    SpvFPFastMathModeFastShift = 4,
+} SpvFPFastMathModeShift;
+
+typedef enum SpvFPFastMathModeMask_ {
+    SpvFPFastMathModeMaskNone = 0,
+    SpvFPFastMathModeNotNaNMask = 0x00000001,
+    SpvFPFastMathModeNotInfMask = 0x00000002,
+    SpvFPFastMathModeNSZMask = 0x00000004,
+    SpvFPFastMathModeAllowRecipMask = 0x00000008,
+    SpvFPFastMathModeFastMask = 0x00000010,
+} SpvFPFastMathModeMask;
+
+typedef enum SpvFPRoundingMode_ {
+    SpvFPRoundingModeRTE = 0,
+    SpvFPRoundingModeRTZ = 1,
+    SpvFPRoundingModeRTP = 2,
+    SpvFPRoundingModeRTN = 3,
+} SpvFPRoundingMode;
+
+typedef enum SpvLinkageType_ {
+    SpvLinkageTypeExport = 0,
+    SpvLinkageTypeImport = 1,
+} SpvLinkageType;
+
+typedef enum SpvAccessQualifier_ {
+    SpvAccessQualifierReadOnly = 0,
+    SpvAccessQualifierWriteOnly = 1,
+    SpvAccessQualifierReadWrite = 2,
+} SpvAccessQualifier;
+
+typedef enum SpvFunctionParameterAttribute_ {
+    SpvFunctionParameterAttributeZext = 0,
+    SpvFunctionParameterAttributeSext = 1,
+    SpvFunctionParameterAttributeByVal = 2,
+    SpvFunctionParameterAttributeSret = 3,
+    SpvFunctionParameterAttributeNoAlias = 4,
+    SpvFunctionParameterAttributeNoCapture = 5,
+    SpvFunctionParameterAttributeSVM = 6,
+    SpvFunctionParameterAttributeNoWrite = 7,
+    SpvFunctionParameterAttributeNoReadWrite = 8,
+} SpvFunctionParameterAttribute;
+
+typedef enum SpvDecoration_ {
+    SpvDecorationRelaxedPrecision = 0,
+    SpvDecorationSpecId = 1,
+    SpvDecorationBlock = 2,
+    SpvDecorationBufferBlock = 3,
+    SpvDecorationRowMajor = 4,
+    SpvDecorationColMajor = 5,
+    SpvDecorationArrayStride = 6,
+    SpvDecorationMatrixStride = 7,
+    SpvDecorationGLSLShared = 8,
+    SpvDecorationGLSLPacked = 9,
+    SpvDecorationCPacked = 10,
+    SpvDecorationBuiltIn = 11,
+    SpvDecorationSmooth = 12,
+    SpvDecorationNoperspective = 13,
+    SpvDecorationFlat = 14,
+    SpvDecorationPatch = 15,
+    SpvDecorationCentroid = 16,
+    SpvDecorationSample = 17,
+    SpvDecorationInvariant = 18,
+    SpvDecorationRestrict = 19,
+    SpvDecorationAliased = 20,
+    SpvDecorationVolatile = 21,
+    SpvDecorationConstant = 22,
+    SpvDecorationCoherent = 23,
+    SpvDecorationNonwritable = 24,
+    SpvDecorationNonreadable = 25,
+    SpvDecorationUniform = 26,
+    SpvDecorationNoStaticUse = 27,
+    SpvDecorationSaturatedConversion = 28,
+    SpvDecorationStream = 29,
+    SpvDecorationLocation = 30,
+    SpvDecorationComponent = 31,
+    SpvDecorationIndex = 32,
+    SpvDecorationBinding = 33,
+    SpvDecorationDescriptorSet = 34,
+    SpvDecorationOffset = 35,
+    SpvDecorationXfbBuffer = 36,
+    SpvDecorationXfbStride = 37,
+    SpvDecorationFuncParamAttr = 38,
+    SpvDecorationFPRoundingMode = 39,
+    SpvDecorationFPFastMathMode = 40,
+    SpvDecorationLinkageAttributes = 41,
+} SpvDecoration;
+
+typedef enum SpvBuiltIn_ {
+    SpvBuiltInPosition = 0,
+    SpvBuiltInPointSize = 1,
+    SpvBuiltInClipVertex = 2,
+    SpvBuiltInClipDistance = 3,
+    SpvBuiltInCullDistance = 4,
+    SpvBuiltInVertexId = 5,
+    SpvBuiltInInstanceId = 6,
+    SpvBuiltInPrimitiveId = 7,
+    SpvBuiltInInvocationId = 8,
+    SpvBuiltInLayer = 9,
+    SpvBuiltInViewportIndex = 10,
+    SpvBuiltInTessLevelOuter = 11,
+    SpvBuiltInTessLevelInner = 12,
+    SpvBuiltInTessCoord = 13,
+    SpvBuiltInPatchVertices = 14,
+    SpvBuiltInFragCoord = 15,
+    SpvBuiltInPointCoord = 16,
+    SpvBuiltInFrontFacing = 17,
+    SpvBuiltInSampleId = 18,
+    SpvBuiltInSamplePosition = 19,
+    SpvBuiltInSampleMask = 20,
+    SpvBuiltInFragColor = 21,
+    SpvBuiltInFragDepth = 22,
+    SpvBuiltInHelperInvocation = 23,
+    SpvBuiltInNumWorkgroups = 24,
+    SpvBuiltInWorkgroupSize = 25,
+    SpvBuiltInWorkgroupId = 26,
+    SpvBuiltInLocalInvocationId = 27,
+    SpvBuiltInGlobalInvocationId = 28,
+    SpvBuiltInLocalInvocationIndex = 29,
+    SpvBuiltInWorkDim = 30,
+    SpvBuiltInGlobalSize = 31,
+    SpvBuiltInEnqueuedWorkgroupSize = 32,
+    SpvBuiltInGlobalOffset = 33,
+    SpvBuiltInGlobalLinearId = 34,
+    SpvBuiltInWorkgroupLinearId = 35,
+    SpvBuiltInSubgroupSize = 36,
+    SpvBuiltInSubgroupMaxSize = 37,
+    SpvBuiltInNumSubgroups = 38,
+    SpvBuiltInNumEnqueuedSubgroups = 39,
+    SpvBuiltInSubgroupId = 40,
+    SpvBuiltInSubgroupLocalInvocationId = 41,
+} SpvBuiltIn;
+
+typedef enum SpvSelectionControlShift_ {
+    SpvSelectionControlFlattenShift = 0,
+    SpvSelectionControlDontFlattenShift = 1,
+} SpvSelectionControlShift;
+
+typedef enum SpvSelectionControlMask_ {
+    SpvSelectionControlMaskNone = 0,
+    SpvSelectionControlFlattenMask = 0x00000001,
+    SpvSelectionControlDontFlattenMask = 0x00000002,
+} SpvSelectionControlMask;
+
+typedef enum SpvLoopControlShift_ {
+    SpvLoopControlUnrollShift = 0,
+    SpvLoopControlDontUnrollShift = 1,
+} SpvLoopControlShift;
+
+typedef enum SpvLoopControlMask_ {
+    SpvLoopControlMaskNone = 0,
+    SpvLoopControlUnrollMask = 0x00000001,
+    SpvLoopControlDontUnrollMask = 0x00000002,
+} SpvLoopControlMask;
+
+typedef enum SpvFunctionControlShift_ {
+    SpvFunctionControlInlineShift = 0,
+    SpvFunctionControlDontInlineShift = 1,
+    SpvFunctionControlPureShift = 2,
+    SpvFunctionControlConstShift = 3,
+} SpvFunctionControlShift;
+
+typedef enum SpvFunctionControlMask_ {
+    SpvFunctionControlMaskNone = 0,
+    SpvFunctionControlInlineMask = 0x00000001,
+    SpvFunctionControlDontInlineMask = 0x00000002,
+    SpvFunctionControlPureMask = 0x00000004,
+    SpvFunctionControlConstMask = 0x00000008,
+} SpvFunctionControlMask;
+
+typedef enum SpvMemorySemanticsShift_ {
+    SpvMemorySemanticsRelaxedShift = 0,
+    SpvMemorySemanticsSequentiallyConsistentShift = 1,
+    SpvMemorySemanticsAcquireShift = 2,
+    SpvMemorySemanticsReleaseShift = 3,
+    SpvMemorySemanticsUniformMemoryShift = 4,
+    SpvMemorySemanticsSubgroupMemoryShift = 5,
+    SpvMemorySemanticsWorkgroupLocalMemoryShift = 6,
+    SpvMemorySemanticsWorkgroupGlobalMemoryShift = 7,
+    SpvMemorySemanticsAtomicCounterMemoryShift = 8,
+    SpvMemorySemanticsImageMemoryShift = 9,
+} SpvMemorySemanticsShift;
+
+typedef enum SpvMemorySemanticsMask_ {
+    SpvMemorySemanticsMaskNone = 0,
+    SpvMemorySemanticsRelaxedMask = 0x00000001,
+    SpvMemorySemanticsSequentiallyConsistentMask = 0x00000002,
+    SpvMemorySemanticsAcquireMask = 0x00000004,
+    SpvMemorySemanticsReleaseMask = 0x00000008,
+    SpvMemorySemanticsUniformMemoryMask = 0x00000010,
+    SpvMemorySemanticsSubgroupMemoryMask = 0x00000020,
+    SpvMemorySemanticsWorkgroupLocalMemoryMask = 0x00000040,
+    SpvMemorySemanticsWorkgroupGlobalMemoryMask = 0x00000080,
+    SpvMemorySemanticsAtomicCounterMemoryMask = 0x00000100,
+    SpvMemorySemanticsImageMemoryMask = 0x00000200,
+} SpvMemorySemanticsMask;
+
+typedef enum SpvMemoryAccessShift_ {
+    SpvMemoryAccessVolatileShift = 0,
+    SpvMemoryAccessAlignedShift = 1,
+} SpvMemoryAccessShift;
+
+typedef enum SpvMemoryAccessMask_ {
+    SpvMemoryAccessMaskNone = 0,
+    SpvMemoryAccessVolatileMask = 0x00000001,
+    SpvMemoryAccessAlignedMask = 0x00000002,
+} SpvMemoryAccessMask;
+
+typedef enum SpvScope_ {
+    SpvScopeCrossDevice = 0,
+    SpvScopeDevice = 1,
+    SpvScopeWorkgroup = 2,
+    SpvScopeSubgroup = 3,
+    SpvScopeInvocation = 4,
+} SpvScope;
+
+typedef enum SpvGroupOperation_ {
+    SpvGroupOperationReduce = 0,
+    SpvGroupOperationInclusiveScan = 1,
+    SpvGroupOperationExclusiveScan = 2,
+} SpvGroupOperation;
+
+typedef enum SpvKernelEnqueueFlags_ {
+    SpvKernelEnqueueFlagsNoWait = 0,
+    SpvKernelEnqueueFlagsWaitKernel = 1,
+    SpvKernelEnqueueFlagsWaitWorkGroup = 2,
+} SpvKernelEnqueueFlags;
+
+typedef enum SpvKernelProfilingInfoShift_ {
+    SpvKernelProfilingInfoCmdExecTimeShift = 0,
+} SpvKernelProfilingInfoShift;
+
+typedef enum SpvKernelProfilingInfoMask_ {
+    SpvKernelProfilingInfoMaskNone = 0,
+    SpvKernelProfilingInfoCmdExecTimeMask = 0x00000001,
+} SpvKernelProfilingInfoMask;
+
+typedef enum SpvCapability_ {
+    SpvCapabilityMatrix = 0,
+    SpvCapabilityShader = 1,
+    SpvCapabilityGeometry = 2,
+    SpvCapabilityTessellation = 3,
+    SpvCapabilityAddresses = 4,
+    SpvCapabilityLinkage = 5,
+    SpvCapabilityKernel = 6,
+    SpvCapabilityVector16 = 7,
+    SpvCapabilityFloat16Buffer = 8,
+    SpvCapabilityFloat16 = 9,
+    SpvCapabilityFloat64 = 10,
+    SpvCapabilityInt64 = 11,
+    SpvCapabilityInt64Atomics = 12,
+    SpvCapabilityImageBasic = 13,
+    SpvCapabilityImageReadWrite = 14,
+    SpvCapabilityImageMipmap = 15,
+    SpvCapabilityImageSRGBWrite = 16,
+    SpvCapabilityPipes = 17,
+    SpvCapabilityGroups = 18,
+    SpvCapabilityDeviceEnqueue = 19,
+    SpvCapabilityLiteralSampler = 20,
+    SpvCapabilityAtomicStorage = 21,
+    SpvCapabilityInt16 = 22,
+} SpvCapability;
+
+typedef enum SpvOp_ {
+    SpvOpNop = 0,
+    SpvOpUndef = 1,
+    SpvOpSource = 3,
+    SpvOpSourceExtension = 4,
+    SpvOpName = 5,
+    SpvOpMemberName = 6,
+    SpvOpString = 7,
+    SpvOpLine = 8,
+    SpvOpExtension = 10,
+    SpvOpExtInstImport = 11,
+    SpvOpExtInst = 12,
+    SpvOpMemoryModel = 14,
+    SpvOpEntryPoint = 15,
+    SpvOpExecutionMode = 16,
+    SpvOpCapability = 17,
+    SpvOpTypeVoid = 19,
+    SpvOpTypeBool = 20,
+    SpvOpTypeInt = 21,
+    SpvOpTypeFloat = 22,
+    SpvOpTypeVector = 23,
+    SpvOpTypeMatrix = 24,
+    SpvOpTypeImage = 25,
+    SpvOpTypeSampler = 26,
+    SpvOpTypeSampledImage = 27,
+    SpvOpTypeArray = 28,
+    SpvOpTypeRuntimeArray = 29,
+    SpvOpTypeStruct = 30,
+    SpvOpTypeOpaque = 31,
+    SpvOpTypePointer = 32,
+    SpvOpTypeFunction = 33,
+    SpvOpTypeEvent = 34,
+    SpvOpTypeDeviceEvent = 35,
+    SpvOpTypeReserveId = 36,
+    SpvOpTypeQueue = 37,
+    SpvOpTypePipe = 38,
+    SpvOpConstantTrue = 41,
+    SpvOpConstantFalse = 42,
+    SpvOpConstant = 43,
+    SpvOpConstantComposite = 44,
+    SpvOpConstantSampler = 45,
+    SpvOpConstantNull = 46,
+    SpvOpSpecConstantTrue = 48,
+    SpvOpSpecConstantFalse = 49,
+    SpvOpSpecConstant = 50,
+    SpvOpSpecConstantComposite = 51,
+    SpvOpSpecConstantOp = 52,
+    SpvOpFunction = 54,
+    SpvOpFunctionParameter = 55,
+    SpvOpFunctionEnd = 56,
+    SpvOpFunctionCall = 57,
+    SpvOpVariable = 59,
+    SpvOpImageTexelPointer = 60,
+    SpvOpLoad = 61,
+    SpvOpStore = 62,
+    SpvOpCopyMemory = 63,
+    SpvOpCopyMemorySized = 64,
+    SpvOpAccessChain = 65,
+    SpvOpInBoundsAccessChain = 66,
+    SpvOpPtrAccessChain = 67,
+    SpvOpArrayLength = 68,
+    SpvOpGenericPtrMemSemantics = 69,
+    SpvOpDecorate = 71,
+    SpvOpMemberDecorate = 72,
+    SpvOpDecorationGroup = 73,
+    SpvOpGroupDecorate = 74,
+    SpvOpGroupMemberDecorate = 75,
+    SpvOpVectorExtractDynamic = 77,
+    SpvOpVectorInsertDynamic = 78,
+    SpvOpVectorShuffle = 79,
+    SpvOpCompositeConstruct = 80,
+    SpvOpCompositeExtract = 81,
+    SpvOpCompositeInsert = 82,
+    SpvOpCopyObject = 83,
+    SpvOpTranspose = 84,
+    SpvOpSampledImage = 86,
+    SpvOpImageSampleImplicitLod = 87,
+    SpvOpImageSampleExplicitLod = 88,
+    SpvOpImageSampleDrefImplicitLod = 89,
+    SpvOpImageSampleDrefExplicitLod = 90,
+    SpvOpImageSampleProjImplicitLod = 91,
+    SpvOpImageSampleProjExplicitLod = 92,
+    SpvOpImageSampleProjDrefImplicitLod = 93,
+    SpvOpImageSampleProjDrefExplicitLod = 94,
+    SpvOpImageFetch = 95,
+    SpvOpImageGather = 96,
+    SpvOpImageDrefGather = 97,
+    SpvOpImageRead = 98,
+    SpvOpImageWrite = 99,
+    SpvOpImageQueryDim = 100,
+    SpvOpImageQueryFormat = 101,
+    SpvOpImageQueryOrder = 102,
+    SpvOpImageQuerySizeLod = 103,
+    SpvOpImageQuerySize = 104,
+    SpvOpImageQueryLod = 105,
+    SpvOpImageQueryLevels = 106,
+    SpvOpImageQuerySamples = 107,
+    SpvOpConvertFToU = 109,
+    SpvOpConvertFToS = 110,
+    SpvOpConvertSToF = 111,
+    SpvOpConvertUToF = 112,
+    SpvOpUConvert = 113,
+    SpvOpSConvert = 114,
+    SpvOpFConvert = 115,
+    SpvOpQuantizeToF16 = 116,
+    SpvOpConvertPtrToU = 117,
+    SpvOpSatConvertSToU = 118,
+    SpvOpSatConvertUToS = 119,
+    SpvOpConvertUToPtr = 120,
+    SpvOpPtrCastToGeneric = 121,
+    SpvOpGenericCastToPtr = 122,
+    SpvOpGenericCastToPtrExplicit = 123,
+    SpvOpBitcast = 124,
+    SpvOpSNegate = 126,
+    SpvOpFNegate = 127,
+    SpvOpIAdd = 128,
+    SpvOpFAdd = 129,
+    SpvOpISub = 130,
+    SpvOpFSub = 131,
+    SpvOpIMul = 132,
+    SpvOpFMul = 133,
+    SpvOpUDiv = 134,
+    SpvOpSDiv = 135,
+    SpvOpFDiv = 136,
+    SpvOpUMod = 137,
+    SpvOpSRem = 138,
+    SpvOpSMod = 139,
+    SpvOpFRem = 140,
+    SpvOpFMod = 141,
+    SpvOpVectorTimesScalar = 142,
+    SpvOpMatrixTimesScalar = 143,
+    SpvOpVectorTimesMatrix = 144,
+    SpvOpMatrixTimesVector = 145,
+    SpvOpMatrixTimesMatrix = 146,
+    SpvOpOuterProduct = 147,
+    SpvOpDot = 148,
+    SpvOpIAddCarry = 149,
+    SpvOpISubBorrow = 150,
+    SpvOpIMulExtended = 151,
+    SpvOpAny = 154,
+    SpvOpAll = 155,
+    SpvOpIsNan = 156,
+    SpvOpIsInf = 157,
+    SpvOpIsFinite = 158,
+    SpvOpIsNormal = 159,
+    SpvOpSignBitSet = 160,
+    SpvOpLessOrGreater = 161,
+    SpvOpOrdered = 162,
+    SpvOpUnordered = 163,
+    SpvOpLogicalEqual = 164,
+    SpvOpLogicalNotEqual = 165,
+    SpvOpLogicalOr = 166,
+    SpvOpLogicalAnd = 167,
+    SpvOpLogicalNot = 168,
+    SpvOpSelect = 169,
+    SpvOpIEqual = 170,
+    SpvOpINotEqual = 171,
+    SpvOpUGreaterThan = 172,
+    SpvOpSGreaterThan = 173,
+    SpvOpUGreaterThanEqual = 174,
+    SpvOpSGreaterThanEqual = 175,
+    SpvOpULessThan = 176,
+    SpvOpSLessThan = 177,
+    SpvOpULessThanEqual = 178,
+    SpvOpSLessThanEqual = 179,
+    SpvOpFOrdEqual = 180,
+    SpvOpFUnordEqual = 181,
+    SpvOpFOrdNotEqual = 182,
+    SpvOpFUnordNotEqual = 183,
+    SpvOpFOrdLessThan = 184,
+    SpvOpFUnordLessThan = 185,
+    SpvOpFOrdGreaterThan = 186,
+    SpvOpFUnordGreaterThan = 187,
+    SpvOpFOrdLessThanEqual = 188,
+    SpvOpFUnordLessThanEqual = 189,
+    SpvOpFOrdGreaterThanEqual = 190,
+    SpvOpFUnordGreaterThanEqual = 191,
+    SpvOpShiftRightLogical = 194,
+    SpvOpShiftRightArithmetic = 195,
+    SpvOpShiftLeftLogical = 196,
+    SpvOpBitwiseOr = 197,
+    SpvOpBitwiseXor = 198,
+    SpvOpBitwiseAnd = 199,
+    SpvOpNot = 200,
+    SpvOpBitFieldInsert = 201,
+    SpvOpBitFieldSExtract = 202,
+    SpvOpBitFieldUExtract = 203,
+    SpvOpBitReverse = 204,
+    SpvOpBitCount = 205,
+    SpvOpDPdx = 207,
+    SpvOpDPdy = 208,
+    SpvOpFwidth = 209,
+    SpvOpDPdxFine = 210,
+    SpvOpDPdyFine = 211,
+    SpvOpFwidthFine = 212,
+    SpvOpDPdxCoarse = 213,
+    SpvOpDPdyCoarse = 214,
+    SpvOpFwidthCoarse = 215,
+    SpvOpEmitVertex = 218,
+    SpvOpEndPrimitive = 219,
+    SpvOpEmitStreamVertex = 220,
+    SpvOpEndStreamPrimitive = 221,
+    SpvOpControlBarrier = 224,
+    SpvOpMemoryBarrier = 225,
+    SpvOpAtomicLoad = 227,
+    SpvOpAtomicStore = 228,
+    SpvOpAtomicExchange = 229,
+    SpvOpAtomicCompareExchange = 230,
+    SpvOpAtomicCompareExchangeWeak = 231,
+    SpvOpAtomicIIncrement = 232,
+    SpvOpAtomicIDecrement = 233,
+    SpvOpAtomicIAdd = 234,
+    SpvOpAtomicISub = 235,
+    SpvOpAtomicIMin = 236,
+    SpvOpAtomicUMin = 237,
+    SpvOpAtomicIMax = 238,
+    SpvOpAtomicUMax = 239,
+    SpvOpAtomicAnd = 240,
+    SpvOpAtomicOr = 241,
+    SpvOpAtomicXor = 242,
+    SpvOpPhi = 245,
+    SpvOpLoopMerge = 246,
+    SpvOpSelectionMerge = 247,
+    SpvOpLabel = 248,
+    SpvOpBranch = 249,
+    SpvOpBranchConditional = 250,
+    SpvOpSwitch = 251,
+    SpvOpKill = 252,
+    SpvOpReturn = 253,
+    SpvOpReturnValue = 254,
+    SpvOpUnreachable = 255,
+    SpvOpLifetimeStart = 256,
+    SpvOpLifetimeStop = 257,
+    SpvOpAsyncGroupCopy = 259,
+    SpvOpWaitGroupEvents = 260,
+    SpvOpGroupAll = 261,
+    SpvOpGroupAny = 262,
+    SpvOpGroupBroadcast = 263,
+    SpvOpGroupIAdd = 264,
+    SpvOpGroupFAdd = 265,
+    SpvOpGroupFMin = 266,
+    SpvOpGroupUMin = 267,
+    SpvOpGroupSMin = 268,
+    SpvOpGroupFMax = 269,
+    SpvOpGroupUMax = 270,
+    SpvOpGroupSMax = 271,
+    SpvOpReadPipe = 274,
+    SpvOpWritePipe = 275,
+    SpvOpReservedReadPipe = 276,
+    SpvOpReservedWritePipe = 277,
+    SpvOpReserveReadPipePackets = 278,
+    SpvOpReserveWritePipePackets = 279,
+    SpvOpCommitReadPipe = 280,
+    SpvOpCommitWritePipe = 281,
+    SpvOpIsValidReserveId = 282,
+    SpvOpGetNumPipePackets = 283,
+    SpvOpGetMaxPipePackets = 284,
+    SpvOpGroupReserveReadPipePackets = 285,
+    SpvOpGroupReserveWritePipePackets = 286,
+    SpvOpGroupCommitReadPipe = 287,
+    SpvOpGroupCommitWritePipe = 288,
+    SpvOpEnqueueMarker = 291,
+    SpvOpEnqueueKernel = 292,
+    SpvOpGetKernelNDrangeSubGroupCount = 293,
+    SpvOpGetKernelNDrangeMaxSubGroupSize = 294,
+    SpvOpGetKernelWorkGroupSize = 295,
+    SpvOpGetKernelPreferredWorkGroupSizeMultiple = 296,
+    SpvOpRetainEvent = 297,
+    SpvOpReleaseEvent = 298,
+    SpvOpCreateUserEvent = 299,
+    SpvOpIsValidEvent = 300,
+    SpvOpSetUserEventStatus = 301,
+    SpvOpCaptureEventProfilingInfo = 302,
+    SpvOpGetDefaultQueue = 303,
+    SpvOpBuildNDRange = 304,
+} SpvOp;
+
+#endif  // #ifndef spirv_H
+

src/glsl/nir/spirv2nir.c

@@ -0,0 +1,54 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ *
+ * Authors:
+ *    Jason Ekstrand (jason@jlekstrand.net)
+ *
+ */
+
+/*
+ * A simple executable that opens a SPIR-V shader, converts it to NIR, and
+ * dumps out the result.  This should be useful for testing the
+ * spirv_to_nir code.
+ */
+
+#include "nir_spirv.h"
+
+#include <sys/mman.h>
+#include <sys/types.h>
+#include <fcntl.h>
+#include <unistd.h>
+
+int main(int argc, char **argv)
+{
+   int fd = open(argv[1], O_RDONLY);
+   off_t len = lseek(fd, 0, SEEK_END);
+
+   assert(len % 4 == 0);
+   size_t word_count = len / 4;
+
+   const void *map = mmap(NULL, len, PROT_READ, MAP_PRIVATE, fd, 0);
+   assert(map != NULL);
+
+   nir_shader *shader = spirv_to_nir(map, word_count, NULL);
+   nir_print_shader(shader, stderr);
+}

src/glsl/nir/spirv_glsl450_to_nir.c

@@ -0,0 +1,285 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ *
+ * Authors:
+ *    Jason Ekstrand (jason@jlekstrand.net)
+ *
+ */
+
+#include "spirv_to_nir_private.h"
+
+enum GLSL450Entrypoint {
+    Round = 0,
+    RoundEven = 1,
+    Trunc = 2,
+    Abs = 3,
+    Sign = 4,
+    Floor = 5,
+    Ceil = 6,
+    Fract = 7,
+
+    Radians = 8,
+    Degrees = 9,
+    Sin = 10,
+    Cos = 11,
+    Tan = 12,
+    Asin = 13,
+    Acos = 14,
+    Atan = 15,
+    Sinh = 16,
+    Cosh = 17,
+    Tanh = 18,
+    Asinh = 19,
+    Acosh = 20,
+    Atanh = 21,
+    Atan2 = 22,
+
+    Pow = 23,
+    Exp = 24,
+    Log = 25,
+    Exp2 = 26,
+    Log2 = 27,
+    Sqrt = 28,
+    InverseSqrt = 29,
+
+    Determinant = 30,
+    MatrixInverse = 31,
+
+    Modf = 32,            // second argument needs the OpVariable = , not an OpLoad
+    Min = 33,
+    Max = 34,
+    Clamp = 35,
+    Mix = 36,
+    Step = 37,
+    SmoothStep = 38,
+
+    FloatBitsToInt = 39,
+    FloatBitsToUint = 40,
+    IntBitsToFloat = 41,
+    UintBitsToFloat = 42,
+
+    Fma = 43,
+    Frexp = 44,
+    Ldexp = 45,
+
+    PackSnorm4x8 = 46,
+    PackUnorm4x8 = 47,
+    PackSnorm2x16 = 48,
+    PackUnorm2x16 = 49,
+    PackHalf2x16 = 50,
+    PackDouble2x32 = 51,
+    UnpackSnorm2x16 = 52,
+    UnpackUnorm2x16 = 53,
+    UnpackHalf2x16 = 54,
+    UnpackSnorm4x8 = 55,
+    UnpackUnorm4x8 = 56,
+    UnpackDouble2x32 = 57,
+
+    Length = 58,
+    Distance = 59,
+    Cross = 60,
+    Normalize = 61,
+    Ftransform = 62,
+    FaceForward = 63,
+    Reflect = 64,
+    Refract = 65,
+
+    UaddCarry = 66,
+    UsubBorrow = 67,
+    UmulExtended = 68,
+    ImulExtended = 69,
+    BitfieldExtract = 70,
+    BitfieldInsert = 71,
+    BitfieldReverse = 72,
+    BitCount = 73,
+    FindLSB = 74,
+    FindMSB = 75,
+
+    InterpolateAtCentroid = 76,
+    InterpolateAtSample = 77,
+    InterpolateAtOffset = 78,
+
+    Count
+};
+
+static nir_ssa_def*
+build_length(nir_builder *b, nir_ssa_def *vec)
+{
+   switch (vec->num_components) {
+   case 1: return nir_fsqrt(b, nir_fmul(b, vec, vec));
+   case 2: return nir_fsqrt(b, nir_fdot2(b, vec, vec));
+   case 3: return nir_fsqrt(b, nir_fdot3(b, vec, vec));
+   case 4: return nir_fsqrt(b, nir_fdot4(b, vec, vec));
+   default:
+      unreachable("Invalid number of components");
+   }
+}
+
+static void
+handle_glsl450_alu(struct vtn_builder *b, enum GLSL450Entrypoint entrypoint,
+                   const uint32_t *w, unsigned count)
+{
+   struct vtn_value *val = vtn_push_value(b, w[2], vtn_value_type_ssa);
+   val->ssa = rzalloc(b, struct vtn_ssa_value);
+   val->ssa->type = vtn_value(b, w[1], vtn_value_type_type)->type->type;
+
+   /* Collect the various SSA sources */
+   unsigned num_inputs = count - 5;
+   nir_ssa_def *src[3];
+   for (unsigned i = 0; i < num_inputs; i++)
+      src[i] = vtn_ssa_value(b, w[i + 5])->def;
+
+   nir_op op;
+   switch (entrypoint) {
+   case Round:       op = nir_op_fround_even;   break; /* TODO */
+   case RoundEven:   op = nir_op_fround_even;   break;
+   case Trunc:       op = nir_op_ftrunc;        break;
+   case Abs:         op = nir_op_fabs;          break;
+   case Sign:        op = nir_op_fsign;         break;
+   case Floor:       op = nir_op_ffloor;        break;
+   case Ceil:        op = nir_op_fceil;         break;
+   case Fract:       op = nir_op_ffract;        break;
+   case Radians:
+      val->ssa->def = nir_fmul(&b->nb, src[0], nir_imm_float(&b->nb, 0.01745329251));
+      return;
+   case Degrees:
+      val->ssa->def = nir_fmul(&b->nb, src[0], nir_imm_float(&b->nb, 57.2957795131));
+      return;
+   case Sin:         op = nir_op_fsin;       break;
+   case Cos:         op = nir_op_fcos;       break;
+   case Tan:
+      val->ssa->def = nir_fdiv(&b->nb, nir_fsin(&b->nb, src[0]),
+                               nir_fcos(&b->nb, src[0]));
+      return;
+   case Pow:         op = nir_op_fpow;       break;
+   case Exp2:        op = nir_op_fexp2;      break;
+   case Log2:        op = nir_op_flog2;      break;
+   case Sqrt:        op = nir_op_fsqrt;      break;
+   case InverseSqrt: op = nir_op_frsq;       break;
+
+   case Modf:        op = nir_op_fmod;       break;
+   case Min:         op = nir_op_fmin;       break;
+   case Max:         op = nir_op_fmax;       break;
+   case Mix:         op = nir_op_flrp;       break;
+   case Step:
+      val->ssa->def = nir_sge(&b->nb, src[1], src[0]);
+      return;
+
+   case FloatBitsToInt:
+   case FloatBitsToUint:
+   case IntBitsToFloat:
+   case UintBitsToFloat:
+      /* Probably going to be removed from the final version of the spec. */
+      val->ssa->def = src[0];
+      return;
+
+   case Fma:         op = nir_op_ffma;       break;
+   case Ldexp:       op = nir_op_ldexp;      break;
+
+   /* Packing/Unpacking functions */
+   case PackSnorm4x8:      op = nir_op_pack_snorm_4x8;      break;
+   case PackUnorm4x8:      op = nir_op_pack_unorm_4x8;      break;
+   case PackSnorm2x16:     op = nir_op_pack_snorm_2x16;     break;
+   case PackUnorm2x16:     op = nir_op_pack_unorm_2x16;     break;
+   case PackHalf2x16:      op = nir_op_pack_half_2x16;      break;
+   case UnpackSnorm4x8:    op = nir_op_unpack_snorm_4x8;    break;
+   case UnpackUnorm4x8:    op = nir_op_unpack_unorm_4x8;    break;
+   case UnpackSnorm2x16:   op = nir_op_unpack_snorm_2x16;   break;
+   case UnpackUnorm2x16:   op = nir_op_unpack_unorm_2x16;   break;
+   case UnpackHalf2x16:    op = nir_op_unpack_half_2x16;    break;
+
+   case Length:
+      val->ssa->def = build_length(&b->nb, src[0]);
+      return;
+   case Distance:
+      val->ssa->def = build_length(&b->nb, nir_fsub(&b->nb, src[0], src[1]));
+      return;
+   case Normalize:
+      val->ssa->def = nir_fdiv(&b->nb, src[0], build_length(&b->nb, src[0]));
+      return;
+
+   case UaddCarry:         op = nir_op_uadd_carry;          break;
+   case UsubBorrow:        op = nir_op_usub_borrow;         break;
+   case BitfieldExtract:   op = nir_op_ubitfield_extract;   break; /* TODO */
+   case BitfieldInsert:    op = nir_op_bitfield_insert;     break;
+   case BitfieldReverse:   op = nir_op_bitfield_reverse;    break;
+   case BitCount:          op = nir_op_bit_count;           break;
+   case FindLSB:           op = nir_op_find_lsb;            break;
+   case FindMSB:           op = nir_op_ufind_msb;           break; /* TODO */
+
+   case Exp:
+   case Log:
+   case Clamp:
+   case Asin:
+   case Acos:
+   case Atan:
+   case Atan2:
+   case Sinh:
+   case Cosh:
+   case Tanh:
+   case Asinh:
+   case Acosh:
+   case Atanh:
+   case SmoothStep:
+   case Frexp:
+   case PackDouble2x32:
+   case UnpackDouble2x32:
+   case Cross:
+   case Ftransform:
+   case FaceForward:
+   case Reflect:
+   case Refract:
+   case UmulExtended:
+   case ImulExtended:
+   default:
+      unreachable("Unhandled opcode");
+   }
+
+   nir_alu_instr *instr = nir_alu_instr_create(b->shader, op);
+   nir_ssa_dest_init(&instr->instr, &instr->dest.dest,
+                     glsl_get_vector_elements(val->ssa->type), val->name);
+   val->ssa->def = &instr->dest.dest.ssa;
+
+   for (unsigned i = 0; i < nir_op_infos[op].num_inputs; i++)
+      instr->src[i].src = nir_src_for_ssa(src[i]);
+
+   nir_builder_instr_insert(&b->nb, &instr->instr);
+}
+
+bool
+vtn_handle_glsl450_instruction(struct vtn_builder *b, uint32_t ext_opcode,
+                               const uint32_t *words, unsigned count)
+{
+   switch ((enum GLSL450Entrypoint)ext_opcode) {
+   case Determinant:
+   case MatrixInverse:
+   case InterpolateAtCentroid:
+   case InterpolateAtSample:
+   case InterpolateAtOffset:
+      unreachable("Unhandled opcode");
+
+   default:
+      handle_glsl450_alu(b, (enum GLSL450Entrypoint)ext_opcode, words, count);
+   }
+
+   return true;
+}

src/glsl/nir/spirv_to_nir_private.h

@@ -0,0 +1,215 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ *
+ * Authors:
+ *    Jason Ekstrand (jason@jlekstrand.net)
+ *
+ */
+
+#include "nir_spirv.h"
+#include "nir_builder.h"
+#include "spirv.h"
+
+struct vtn_builder;
+struct vtn_decoration;
+
+enum vtn_value_type {
+   vtn_value_type_invalid = 0,
+   vtn_value_type_undef,
+   vtn_value_type_string,
+   vtn_value_type_decoration_group,
+   vtn_value_type_type,
+   vtn_value_type_constant,
+   vtn_value_type_deref,
+   vtn_value_type_function,
+   vtn_value_type_block,
+   vtn_value_type_ssa,
+   vtn_value_type_extension,
+};
+
+struct vtn_block {
+   /* Merge opcode if this block contains a merge; SpvOpNop otherwise. */
+   SpvOp merge_op;
+   uint32_t merge_block_id;
+   const uint32_t *label;
+   const uint32_t *branch;
+   nir_block *block;
+};
+
+struct vtn_function {
+   struct exec_node node;
+
+   nir_function_overload *overload;
+   struct vtn_block *start_block;
+
+   const uint32_t *end;
+};
+
+typedef bool (*vtn_instruction_handler)(struct vtn_builder *, uint32_t,
+                                        const uint32_t *, unsigned);
+
+struct vtn_ssa_value {
+   union {
+      nir_ssa_def *def;
+      struct vtn_ssa_value **elems;
+   };
+
+   /* For matrices, a transposed version of the value, or NULL if it hasn't
+    * been computed
+    */
+   struct vtn_ssa_value *transposed;
+
+   const struct glsl_type *type;
+};
+
+struct vtn_type {
+   const struct glsl_type *type;
+
+   /* for matrices, whether the matrix is stored row-major */
+   bool row_major;
+
+   /* for structs, the offset of each member */
+   unsigned *offsets;
+
+   /* for arrays and matrices, the array stride */
+   unsigned stride;
+
+   /* for arrays, the vtn_type for the elements of the array */
+   struct vtn_type *array_element;
+
+   /* for structures, the vtn_type for each member */
+   struct vtn_type **members;
+
+   /* Whether this type, or a parent type, has been decorated as a builtin */
+   bool is_builtin;
+
+   SpvBuiltIn builtin;
+};
+
+struct vtn_value {
+   enum vtn_value_type value_type;
+   const char *name;
+   struct vtn_decoration *decoration;
+   union {
+      void *ptr;
+      char *str;
+      struct vtn_type *type;
+      struct {
+         nir_constant *constant;
+         const struct glsl_type *const_type;
+      };
+      struct {
+         nir_deref_var *deref;
+         struct vtn_type *deref_type;
+      };
+      struct vtn_function *func;
+      struct vtn_block *block;
+      struct vtn_ssa_value *ssa;
+      vtn_instruction_handler ext_handler;
+   };
+};
+
+struct vtn_decoration {
+   struct vtn_decoration *next;
+   int member; /* -1 if not a member decoration */
+   const uint32_t *literals;
+   struct vtn_value *group;
+   SpvDecoration decoration;
+};
+
+struct vtn_builder {
+   nir_builder nb;
+
+   nir_shader *shader;
+   nir_function_impl *impl;
+   struct vtn_block *block;
+
+   /*
+    * In SPIR-V, constants are global, whereas in NIR, the load_const
+    * instruction we use is per-function. So while we parse each function, we
+    * keep a hash table of constants we've resolved to nir_ssa_value's so
+    * far, and we lazily resolve them when we see them used in a function.
+    */
+   struct hash_table *const_table;
+
+   /*
+    * Map from nir_block to the vtn_block which ends with it -- used for
+    * handling phi nodes.
+    */
+   struct hash_table *block_table;
+
+   /*
+    * NIR variable for each SPIR-V builtin.
+    */
+   nir_variable *builtins[42]; /* XXX need symbolic constant from SPIR-V header */
+
+   unsigned value_id_bound;
+   struct vtn_value *values;
+
+   SpvExecutionModel execution_model;
+   struct vtn_value *entry_point;
+
+   struct vtn_function *func;
+   struct exec_list functions;
+};
+
+static inline struct vtn_value *
+vtn_push_value(struct vtn_builder *b, uint32_t value_id,
+               enum vtn_value_type value_type)
+{
+   assert(value_id < b->value_id_bound);
+   assert(b->values[value_id].value_type == vtn_value_type_invalid);
+
+   b->values[value_id].value_type = value_type;
+
+   return &b->values[value_id];
+}
+
+static inline struct vtn_value *
+vtn_untyped_value(struct vtn_builder *b, uint32_t value_id)
+{
+   assert(value_id < b->value_id_bound);
+   return &b->values[value_id];
+}
+
+static inline struct vtn_value *
+vtn_value(struct vtn_builder *b, uint32_t value_id,
+          enum vtn_value_type value_type)
+{
+   struct vtn_value *val = vtn_untyped_value(b, value_id);
+   assert(val->value_type == value_type);
+   return val;
+}
+
+struct vtn_ssa_value *vtn_ssa_value(struct vtn_builder *b, uint32_t value_id);
+
+typedef void (*vtn_decoration_foreach_cb)(struct vtn_builder *,
+                                          struct vtn_value *,
+                                          int member,
+                                          const struct vtn_decoration *,
+                                          void *);
+
+void vtn_foreach_decoration(struct vtn_builder *b, struct vtn_value *value,
+                            vtn_decoration_foreach_cb cb, void *data);
+
+bool vtn_handle_glsl450_instruction(struct vtn_builder *b, uint32_t ext_opcode,
+                                    const uint32_t *words, unsigned count);

src/glsl/shader_enums.h

@@ -23,6 +23,8 @@
  * OTHER DEALINGS IN THE SOFTWARE.
  */
 
+#include "main/config.h"
+
 #ifndef SHADER_ENUMS_H
 #define SHADER_ENUMS_H
 
@@ -184,4 +186,209 @@ enum glsl_interp_qualifier
 };
 
 
+/**
+ * Indexes for vertex program attributes.
+ * GL_NV_vertex_program aliases generic attributes over the conventional
+ * attributes.  In GL_ARB_vertex_program shader the aliasing is optional.
+ * In GL_ARB_vertex_shader / OpenGL 2.0 the aliasing is disallowed (the
+ * generic attributes are distinct/separate).
+ */
+typedef enum
+{
+   VERT_ATTRIB_POS = 0,
+   VERT_ATTRIB_WEIGHT = 1,
+   VERT_ATTRIB_NORMAL = 2,
+   VERT_ATTRIB_COLOR0 = 3,
+   VERT_ATTRIB_COLOR1 = 4,
+   VERT_ATTRIB_FOG = 5,
+   VERT_ATTRIB_COLOR_INDEX = 6,
+   VERT_ATTRIB_EDGEFLAG = 7,
+   VERT_ATTRIB_TEX0 = 8,
+   VERT_ATTRIB_TEX1 = 9,
+   VERT_ATTRIB_TEX2 = 10,
+   VERT_ATTRIB_TEX3 = 11,
+   VERT_ATTRIB_TEX4 = 12,
+   VERT_ATTRIB_TEX5 = 13,
+   VERT_ATTRIB_TEX6 = 14,
+   VERT_ATTRIB_TEX7 = 15,
+   VERT_ATTRIB_POINT_SIZE = 16,
+   VERT_ATTRIB_GENERIC0 = 17,
+   VERT_ATTRIB_GENERIC1 = 18,
+   VERT_ATTRIB_GENERIC2 = 19,
+   VERT_ATTRIB_GENERIC3 = 20,
+   VERT_ATTRIB_GENERIC4 = 21,
+   VERT_ATTRIB_GENERIC5 = 22,
+   VERT_ATTRIB_GENERIC6 = 23,
+   VERT_ATTRIB_GENERIC7 = 24,
+   VERT_ATTRIB_GENERIC8 = 25,
+   VERT_ATTRIB_GENERIC9 = 26,
+   VERT_ATTRIB_GENERIC10 = 27,
+   VERT_ATTRIB_GENERIC11 = 28,
+   VERT_ATTRIB_GENERIC12 = 29,
+   VERT_ATTRIB_GENERIC13 = 30,
+   VERT_ATTRIB_GENERIC14 = 31,
+   VERT_ATTRIB_GENERIC15 = 32,
+   VERT_ATTRIB_MAX = 33
+} gl_vert_attrib;
+
+/**
+ * Symbolic constats to help iterating over
+ * specific blocks of vertex attributes.
+ *
+ * VERT_ATTRIB_FF
+ *   includes all fixed function attributes as well as
+ *   the aliased GL_NV_vertex_program shader attributes.
+ * VERT_ATTRIB_TEX
+ *   include the classic texture coordinate attributes.
+ *   Is a subset of VERT_ATTRIB_FF.
+ * VERT_ATTRIB_GENERIC
+ *   include the OpenGL 2.0+ GLSL generic shader attributes.
+ *   These alias the generic GL_ARB_vertex_shader attributes.
+ */
+#define VERT_ATTRIB_FF(i)           (VERT_ATTRIB_POS + (i))
+#define VERT_ATTRIB_FF_MAX          VERT_ATTRIB_GENERIC0
+
+#define VERT_ATTRIB_TEX(i)          (VERT_ATTRIB_TEX0 + (i))
+#define VERT_ATTRIB_TEX_MAX         MAX_TEXTURE_COORD_UNITS
+
+#define VERT_ATTRIB_GENERIC(i)      (VERT_ATTRIB_GENERIC0 + (i))
+#define VERT_ATTRIB_GENERIC_MAX     MAX_VERTEX_GENERIC_ATTRIBS
+
+/**
+ * Bitflags for vertex attributes.
+ * These are used in bitfields in many places.
+ */
+/*@{*/
+#define VERT_BIT_POS             BITFIELD64_BIT(VERT_ATTRIB_POS)
+#define VERT_BIT_WEIGHT          BITFIELD64_BIT(VERT_ATTRIB_WEIGHT)
+#define VERT_BIT_NORMAL          BITFIELD64_BIT(VERT_ATTRIB_NORMAL)
+#define VERT_BIT_COLOR0          BITFIELD64_BIT(VERT_ATTRIB_COLOR0)
+#define VERT_BIT_COLOR1          BITFIELD64_BIT(VERT_ATTRIB_COLOR1)
+#define VERT_BIT_FOG             BITFIELD64_BIT(VERT_ATTRIB_FOG)
+#define VERT_BIT_COLOR_INDEX     BITFIELD64_BIT(VERT_ATTRIB_COLOR_INDEX)
+#define VERT_BIT_EDGEFLAG        BITFIELD64_BIT(VERT_ATTRIB_EDGEFLAG)
+#define VERT_BIT_TEX0            BITFIELD64_BIT(VERT_ATTRIB_TEX0)
+#define VERT_BIT_TEX1            BITFIELD64_BIT(VERT_ATTRIB_TEX1)
+#define VERT_BIT_TEX2            BITFIELD64_BIT(VERT_ATTRIB_TEX2)
+#define VERT_BIT_TEX3            BITFIELD64_BIT(VERT_ATTRIB_TEX3)
+#define VERT_BIT_TEX4            BITFIELD64_BIT(VERT_ATTRIB_TEX4)
+#define VERT_BIT_TEX5            BITFIELD64_BIT(VERT_ATTRIB_TEX5)
+#define VERT_BIT_TEX6            BITFIELD64_BIT(VERT_ATTRIB_TEX6)
+#define VERT_BIT_TEX7            BITFIELD64_BIT(VERT_ATTRIB_TEX7)
+#define VERT_BIT_POINT_SIZE      BITFIELD64_BIT(VERT_ATTRIB_POINT_SIZE)
+#define VERT_BIT_GENERIC0        BITFIELD64_BIT(VERT_ATTRIB_GENERIC0)
+
+#define VERT_BIT(i)              BITFIELD64_BIT(i)
+#define VERT_BIT_ALL             BITFIELD64_RANGE(0, VERT_ATTRIB_MAX)
+
+#define VERT_BIT_FF(i)           VERT_BIT(i)
+#define VERT_BIT_FF_ALL          BITFIELD64_RANGE(0, VERT_ATTRIB_FF_MAX)
+#define VERT_BIT_TEX(i)          VERT_BIT(VERT_ATTRIB_TEX(i))
+#define VERT_BIT_TEX_ALL         \
+   BITFIELD64_RANGE(VERT_ATTRIB_TEX(0), VERT_ATTRIB_TEX_MAX)
+
+#define VERT_BIT_GENERIC(i)      VERT_BIT(VERT_ATTRIB_GENERIC(i))
+#define VERT_BIT_GENERIC_ALL     \
+   BITFIELD64_RANGE(VERT_ATTRIB_GENERIC(0), VERT_ATTRIB_GENERIC_MAX)
+/*@}*/
+
+
+/**
+ * Indexes for vertex shader outputs, geometry shader inputs/outputs, and
+ * fragment shader inputs.
+ *
+ * Note that some of these values are not available to all pipeline stages.
+ *
+ * When this enum is updated, the following code must be updated too:
+ * - vertResults (in prog_print.c's arb_output_attrib_string())
+ * - fragAttribs (in prog_print.c's arb_input_attrib_string())
+ * - _mesa_varying_slot_in_fs()
+ */
+typedef enum
+{
+   VARYING_SLOT_POS,
+   VARYING_SLOT_COL0, /* COL0 and COL1 must be contiguous */
+   VARYING_SLOT_COL1,
+   VARYING_SLOT_FOGC,
+   VARYING_SLOT_TEX0, /* TEX0-TEX7 must be contiguous */
+   VARYING_SLOT_TEX1,
+   VARYING_SLOT_TEX2,
+   VARYING_SLOT_TEX3,
+   VARYING_SLOT_TEX4,
+   VARYING_SLOT_TEX5,
+   VARYING_SLOT_TEX6,
+   VARYING_SLOT_TEX7,
+   VARYING_SLOT_PSIZ, /* Does not appear in FS */
+   VARYING_SLOT_BFC0, /* Does not appear in FS */
+   VARYING_SLOT_BFC1, /* Does not appear in FS */
+   VARYING_SLOT_EDGE, /* Does not appear in FS */
+   VARYING_SLOT_CLIP_VERTEX, /* Does not appear in FS */ VARYING_SLOT_CLIP_DIST0,
+   VARYING_SLOT_CLIP_DIST1,
+   VARYING_SLOT_PRIMITIVE_ID, /* Does not appear in VS */
+   VARYING_SLOT_LAYER, /* Appears as VS or GS output */
+   VARYING_SLOT_VIEWPORT, /* Appears as VS or GS output */
+   VARYING_SLOT_FACE, /* FS only */
+   VARYING_SLOT_PNTC, /* FS only */
+   VARYING_SLOT_VAR0, /* First generic varying slot */
+   VARYING_SLOT_MAX = VARYING_SLOT_VAR0 + MAX_VARYING
+} gl_varying_slot;
+
+
+/**
+ * Bitflags for varying slots.
+ */
+/*@{*/
+#define VARYING_BIT_POS BITFIELD64_BIT(VARYING_SLOT_POS)
+#define VARYING_BIT_COL0 BITFIELD64_BIT(VARYING_SLOT_COL0)
+#define VARYING_BIT_COL1 BITFIELD64_BIT(VARYING_SLOT_COL1)
+#define VARYING_BIT_FOGC BITFIELD64_BIT(VARYING_SLOT_FOGC)
+#define VARYING_BIT_TEX0 BITFIELD64_BIT(VARYING_SLOT_TEX0)
+#define VARYING_BIT_TEX1 BITFIELD64_BIT(VARYING_SLOT_TEX1)
+#define VARYING_BIT_TEX2 BITFIELD64_BIT(VARYING_SLOT_TEX2)
+#define VARYING_BIT_TEX3 BITFIELD64_BIT(VARYING_SLOT_TEX3)
+#define VARYING_BIT_TEX4 BITFIELD64_BIT(VARYING_SLOT_TEX4)
+#define VARYING_BIT_TEX5 BITFIELD64_BIT(VARYING_SLOT_TEX5)
+#define VARYING_BIT_TEX6 BITFIELD64_BIT(VARYING_SLOT_TEX6)
+#define VARYING_BIT_TEX7 BITFIELD64_BIT(VARYING_SLOT_TEX7)
+#define VARYING_BIT_TEX(U) BITFIELD64_BIT(VARYING_SLOT_TEX0 + (U))
+#define VARYING_BITS_TEX_ANY BITFIELD64_RANGE(VARYING_SLOT_TEX0, \
+                                              MAX_TEXTURE_COORD_UNITS)
+#define VARYING_BIT_PSIZ BITFIELD64_BIT(VARYING_SLOT_PSIZ)
+#define VARYING_BIT_BFC0 BITFIELD64_BIT(VARYING_SLOT_BFC0)
+#define VARYING_BIT_BFC1 BITFIELD64_BIT(VARYING_SLOT_BFC1)
+#define VARYING_BIT_EDGE BITFIELD64_BIT(VARYING_SLOT_EDGE)
+#define VARYING_BIT_CLIP_VERTEX BITFIELD64_BIT(VARYING_SLOT_CLIP_VERTEX)
+#define VARYING_BIT_CLIP_DIST0 BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0)
+#define VARYING_BIT_CLIP_DIST1 BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1)
+#define VARYING_BIT_PRIMITIVE_ID BITFIELD64_BIT(VARYING_SLOT_PRIMITIVE_ID)
+#define VARYING_BIT_LAYER BITFIELD64_BIT(VARYING_SLOT_LAYER)
+#define VARYING_BIT_VIEWPORT BITFIELD64_BIT(VARYING_SLOT_VIEWPORT)
+#define VARYING_BIT_FACE BITFIELD64_BIT(VARYING_SLOT_FACE)
+#define VARYING_BIT_PNTC BITFIELD64_BIT(VARYING_SLOT_PNTC)
+#define VARYING_BIT_VAR(V) BITFIELD64_BIT(VARYING_SLOT_VAR0 + (V))
+/*@}*/
+
+
+/**
+ * Fragment program results
+ */
+typedef enum
+{
+   FRAG_RESULT_DEPTH = 0,
+   FRAG_RESULT_STENCIL = 1,
+   /* If a single color should be written to all render targets, this
+    * register is written.  No FRAG_RESULT_DATAn will be written.
+    */
+   FRAG_RESULT_COLOR = 2,
+   FRAG_RESULT_SAMPLE_MASK = 3,
+
+   /* FRAG_RESULT_DATAn are the per-render-target (GLSL gl_FragData[n]
+    * or ARB_fragment_program fragment.color[n]) color results.  If
+    * any are written, FRAG_RESULT_COLOR will not be written.
+    */
+   FRAG_RESULT_DATA0 = 4,
+   FRAG_RESULT_MAX = (FRAG_RESULT_DATA0 + MAX_DRAW_BUFFERS)
+} gl_frag_result;
+
+
 #endif /* SHADER_ENUMS_H */

src/glsl/standalone_scaffolding.cpp

@@ -34,6 +34,12 @@
 #include <string.h>
 #include "util/ralloc.h"
 
+extern "C" void
+_mesa_error_no_memory(const char *caller)
+{
+   fprintf(stderr, "Mesa error: out of memory in %s", caller);
+}
+
 void
 _mesa_warning(struct gl_context *ctx, const char *fmt, ...)
 {

src/mesa/Makefile.am

@@ -40,8 +40,11 @@ gl_HEADERS = $(top_srcdir)/include/GL/*.h
 
 .PHONY: main/git_sha1.h.tmp
 main/git_sha1.h.tmp:
+	@# Don't assume that $(top_srcdir)/.git is a directory. It may be
+	@# a gitlink file if $(top_srcdir) is a submodule checkout or a linked
+	@# worktree.
 	@touch main/git_sha1.h.tmp
-	@if test -d $(top_srcdir)/.git; then \
+	@if test -e $(top_srcdir)/.git; then \
 		if which git > /dev/null; then \
 		    git --git-dir=$(top_srcdir)/.git log -n 1 --oneline | \
 			sed 's/^\([^ ]*\) .*/#define MESA_GIT_SHA1 "git-\1"/' \

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

@@ -1,3 +1,4 @@
+#include <stdlib.h>
 #include "main/glheader.h"
 #include "main/compiler.h"
 #include "glapi/glapi.h"
@@ -33,12 +34,14 @@ _glapi_check_multithread(void)
 
 PUBLIC void
 _glapi_set_context(void *context)
-{}
+{
+        _glapi_Context = context;
+}
 
 PUBLIC void *
 _glapi_get_context(void)
 {
-	return 0;
+        return _glapi_Context;
 }
 
 PUBLIC void
@@ -84,7 +87,7 @@ _glapi_set_nop_handler(_glapi_nop_handler_proc func)
 PUBLIC struct _glapi_table *
 _glapi_new_nop_table(unsigned num_entries)
 {
-	return NULL;
+        return malloc(16);
 }
 
 #ifndef NO_MAIN

src/mesa/drivers/dri/i965/Makefile.am

@@ -41,7 +41,7 @@ AM_CFLAGS = \
 
 AM_CXXFLAGS = $(AM_CFLAGS)
 
-noinst_LTLIBRARIES = libi965_dri.la
+noinst_LTLIBRARIES = libi965_dri.la libi965_compiler.la
 libi965_dri_la_SOURCES = $(i965_FILES)
 libi965_dri_la_LIBADD = $(INTEL_LIBS)
 
@@ -55,6 +55,16 @@ TEST_LIBS = \
         $(CLOCK_LIB) \
 	../common/libdri_test_stubs.la
 
+libi965_compiler_la_SOURCES = $(i965_FILES)
+libi965_compiler_la_LIBADD = $(INTEL_LIBS) \
+	../common/libdricommon.la \
+	../common/libxmlconfig.la \
+	../common/libmegadriver_stub.la \
+        ../../../libmesa.la \
+	$(DRI_LIB_DEPS) \
+        $(CLOCK_LIB) \
+	../common/libdri_test_stubs.la -lm
+
 TESTS = \
 	test_fs_cmod_propagation \
 	test_fs_saturate_propagation \

src/mesa/drivers/dri/i965/brw_context.c

@@ -311,7 +311,7 @@ brw_init_driver_functions(struct brw_context *brw,
       functions->GetSamplePosition = gen6_get_sample_position;
 }
 
-static void
+void
 brw_initialize_context_constants(struct brw_context *brw)
 {
    struct gl_context *ctx = &brw->ctx;
@@ -390,7 +390,8 @@ brw_initialize_context_constants(struct brw_context *brw)
    int max_samples;
    const int *msaa_modes = intel_supported_msaa_modes(brw->intelScreen);
    const int clamp_max_samples =
-      driQueryOptioni(&brw->optionCache, "clamp_max_samples");
+      brw->optionCache.info != NULL ?
+         driQueryOptioni(&brw->optionCache, "clamp_max_samples") : -1;
 
    if (clamp_max_samples < 0) {
       max_samples = msaa_modes[0];
@@ -801,6 +802,7 @@ brwCreateContext(gl_api api,
 
    intel_batchbuffer_init(brw);
 
+#if 0
    if (brw->gen >= 6) {
       /* Create a new hardware context.  Using a hardware context means that
        * our GPU state will be saved/restored on context switch, allowing us
@@ -819,6 +821,7 @@ brwCreateContext(gl_api api,
    }
 
    brw_init_state(brw);
+#endif
 
    intelInitExtensions(ctx);
 
@@ -882,8 +885,10 @@ brwCreateContext(gl_api api,
 
    _mesa_compute_version(ctx);
 
+#if 0
    _mesa_initialize_dispatch_tables(ctx);
    _mesa_initialize_vbo_vtxfmt(ctx);
+#endif
 
    if (ctx->Extensions.AMD_performance_monitor) {
       brw_init_performance_monitors(brw);

src/mesa/drivers/dri/i965/brw_context.h

@@ -359,6 +359,12 @@ struct brw_stage_prog_data {
       /** @} */
    } binding_table;
 
+   uint32_t *map_entries;
+   struct {
+      uint32_t index_count;
+      uint32_t *index;
+   } bind_map[8]; /* MAX_SETS from vulkan/private.h */
+
    GLuint nr_params;       /**< number of float params/constants */
    GLuint nr_pull_params;
 
@@ -1994,6 +2000,15 @@ gen6_upload_push_constants(struct brw_context *brw,
                            struct brw_stage_state *stage_state,
                            enum aub_state_struct_type type);
 
+struct intel_screen *intel_screen_create(int fd);
+void intel_screen_destroy(struct intel_screen *screen);
+
+struct brw_context *intel_context_create(struct intel_screen *screen);
+void intel_context_destroy(struct brw_context *brw);
+
+void
+brw_initialize_context_constants(struct brw_context *brw);
+
 bool
 gen9_use_linear_1d_layout(const struct brw_context *brw,
                           const struct intel_mipmap_tree *mt);

src/mesa/drivers/dri/i965/brw_cs.cpp

@@ -55,7 +55,7 @@ brw_cs_prog_data_compare(const void *in_a, const void *in_b)
 }
 
 
-static const unsigned *
+const unsigned *
 brw_cs_emit(struct brw_context *brw,
             void *mem_ctx,
             const struct brw_cs_prog_key *key,

src/mesa/drivers/dri/i965/brw_cs.h

@@ -41,6 +41,15 @@ bool brw_cs_prog_data_compare(const void *a, const void *b);
 void
 brw_upload_cs_prog(struct brw_context *brw);
 
+const unsigned *
+brw_cs_emit(struct brw_context *brw,
+            void *mem_ctx,
+            const struct brw_cs_prog_key *key,
+            struct brw_cs_prog_data *prog_data,
+            struct gl_compute_program *cp,
+            struct gl_shader_program *prog,
+            unsigned *final_assembly_size);
+
 #ifdef __cplusplus
 }
 #endif

src/mesa/drivers/dri/i965/brw_defines.h

@@ -57,6 +57,7 @@
 # define GEN7_3DPRIM_VERTEXBUFFER_ACCESS_SEQUENTIAL (0 << 8)
 # define GEN7_3DPRIM_VERTEXBUFFER_ACCESS_RANDOM     (1 << 8)
 
+#ifndef _3DPRIM_POINTLIST /* FIXME: Avoid clashing with defines from bdw_pack.h */
 #define _3DPRIM_POINTLIST         0x01
 #define _3DPRIM_LINELIST          0x02
 #define _3DPRIM_LINESTRIP         0x03
@@ -78,6 +79,7 @@
 #define _3DPRIM_LINESTRIP_BF      0x13
 #define _3DPRIM_LINESTRIP_CONT_BF 0x14
 #define _3DPRIM_TRIFAN_NOSTIPPLE  0x15
+#endif
 
 /* We use this offset to be able to pass native primitive types in struct
  * _mesa_prim::mode.  Native primitive types are BRW_PRIM_OFFSET +

src/mesa/drivers/dri/i965/brw_device_info.c

@@ -353,3 +353,15 @@ brw_get_device_info(int devid, int revision)
 
    return devinfo;
 }
+
+const char *
+brw_get_device_name(int devid)
+{
+   switch (devid) {
+#undef CHIPSET
+#define CHIPSET(id, family, name) case id: return name;
+#include "pci_ids/i965_pci_ids.h"
+   default:
+      return NULL;
+   }
+}

src/mesa/drivers/dri/i965/brw_device_info.h

@@ -85,3 +85,4 @@ struct brw_device_info
 };
 
 const struct brw_device_info *brw_get_device_info(int devid, int revision);
+const char *brw_get_device_name(int devid);

src/mesa/drivers/dri/i965/brw_fs.cpp

@@ -525,6 +525,7 @@ fs_visitor::type_size(const struct glsl_type *type)
    case GLSL_TYPE_ERROR:
    case GLSL_TYPE_INTERFACE:
    case GLSL_TYPE_DOUBLE:
+   case GLSL_TYPE_FUNCTION:
       unreachable("not reached");
    }
 
@@ -1460,6 +1461,10 @@ fs_visitor::assign_vs_urb_setup()
    unsigned vue_entries =
       MAX2(count, vs_prog_data->base.vue_map.num_slots);
 
+   /* URB entry size is counted in units of 64 bytes (for the 3DSTATE_URB_VS
+    * command).  Each attribute is 16 bytes (4 floats/dwords), so each unit
+    * fits four attributes.
+    */
    vs_prog_data->base.urb_entry_size = ALIGN(vue_entries, 4) / 4;
    vs_prog_data->base.urb_read_length = (count + 1) / 2;
 
@@ -2590,9 +2595,22 @@ fs_visitor::emit_repclear_shader()
    brw_wm_prog_key *key = (brw_wm_prog_key*) this->key;
    int base_mrf = 1;
    int color_mrf = base_mrf + 2;
+   fs_inst *mov;
 
-   fs_inst *mov = bld.exec_all().MOV(vec4(brw_message_reg(color_mrf)),
-                                     fs_reg(UNIFORM, 0, BRW_REGISTER_TYPE_F));
+   if (uniforms == 1) {
+      mov = bld.exec_all().MOV(vec4(brw_message_reg(color_mrf)),
+                               fs_reg(UNIFORM, 0, BRW_REGISTER_TYPE_F));
+   } else {
+      struct brw_reg reg =
+         brw_reg(BRW_GENERAL_REGISTER_FILE,
+                 2, 3, 0, 0, BRW_REGISTER_TYPE_F,
+                 BRW_VERTICAL_STRIDE_8,
+                 BRW_WIDTH_2,
+                 BRW_HORIZONTAL_STRIDE_4, BRW_SWIZZLE_XYZW, WRITEMASK_XYZW);
+
+      mov = bld.exec_all().MOV(vec4(brw_message_reg(color_mrf)),
+                               fs_reg(reg));
+   }
 
    fs_inst *write;
    if (key->nr_color_regions == 1) {
@@ -2621,8 +2639,10 @@ fs_visitor::emit_repclear_shader()
    assign_curb_setup();
 
    /* Now that we have the uniform assigned, go ahead and force it to a vec4. */
-   assert(mov->src[0].file == HW_REG);
-   mov->src[0] = brw_vec4_grf(mov->src[0].fixed_hw_reg.nr, 0);
+   if (uniforms == 1) {
+      assert(mov->src[0].file == HW_REG);
+      mov->src[0] = brw_vec4_grf(mov->src[0].fixed_hw_reg.nr, 0);
+   }
 }
 
 /**
@@ -3805,7 +3825,8 @@ fs_visitor::run_vs(gl_clip_plane *clip_planes)
 {
    assert(stage == MESA_SHADER_VERTEX);
 
-   assign_common_binding_table_offsets(0);
+   if (prog_data->map_entries == NULL)
+      assign_common_binding_table_offsets(0);
    setup_vs_payload();
 
    if (shader_time_index >= 0)
@@ -3842,9 +3863,8 @@ fs_visitor::run_fs(bool do_rep_send)
 
    assert(stage == MESA_SHADER_FRAGMENT);
 
-   sanity_param_count = prog->Parameters->NumParameters;
-
-   assign_binding_table_offsets();
+   if (prog_data->map_entries == NULL)
+      assign_binding_table_offsets();
 
    if (devinfo->gen >= 6)
       setup_payload_gen6();
@@ -3914,13 +3934,6 @@ fs_visitor::run_fs(bool do_rep_send)
    else
       wm_prog_data->reg_blocks_16 = brw_register_blocks(grf_used);
 
-   /* If any state parameters were appended, then ParameterValues could have
-    * been realloced, in which case the driver uniform storage set up by
-    * _mesa_associate_uniform_storage() would point to freed memory.  Make
-    * sure that didn't happen.
-    */
-   assert(sanity_param_count == prog->Parameters->NumParameters);
-
    return !failed;
 }
 
@@ -3993,7 +4006,7 @@ brw_wm_fs_emit(struct brw_context *brw,
    if (prog)
       shader = (brw_shader *) prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
 
-   if (unlikely(INTEL_DEBUG & DEBUG_WM))
+   if (unlikely(INTEL_DEBUG & DEBUG_WM) && shader->base.ir)
       brw_dump_ir("fragment", prog, &shader->base, &fp->Base);
 
    int st_index8 = -1, st_index16 = -1;

src/mesa/drivers/dri/i965/brw_fs_nir.cpp

@@ -1368,9 +1368,19 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
       fs_reg surf_index;
 
       if (const_index) {
-         surf_index = fs_reg(stage_prog_data->binding_table.ubo_start +
-                             const_index->u[0]);
+         uint32_t index = const_index->u[0];
+         uint32_t set = shader->base.UniformBlocks[index].Set;
+         uint32_t binding = shader->base.UniformBlocks[index].Binding;
+
+         /* FIXME: We should probably assert here, but dota2 seems to hit
+          * it and we'd like to keep going.
+          */
+         if (binding >= stage_prog_data->bind_map[set].index_count)
+            binding = 0;
+
+         surf_index = fs_reg(stage_prog_data->bind_map[set].index[binding]);
       } else {
+         assert(0 && "need more info from the ir for this.");
          /* The block index is not a constant. Evaluate the index expression
           * per-channel and add the base UBO index; we have to select a value
           * from any live channel.
@@ -1587,7 +1597,13 @@ fs_visitor::nir_emit_intrinsic(const fs_builder &bld, nir_intrinsic_instr *instr
 void
 fs_visitor::nir_emit_texture(const fs_builder &bld, nir_tex_instr *instr)
 {
-   unsigned sampler = instr->sampler_index;
+   uint32_t set = instr->sampler_set;
+   uint32_t binding = instr->sampler_index;
+
+   assert(binding < stage_prog_data->bind_map[set].index_count);
+   assert(stage_prog_data->bind_map[set].index[binding] < 1000);
+
+   unsigned sampler = stage_prog_data->bind_map[set].index[binding];
    fs_reg sampler_reg(sampler);
 
    /* FINISHME: We're failing to recompile our programs when the sampler is
@@ -1758,6 +1774,12 @@ fs_visitor::nir_emit_jump(const fs_builder &bld, nir_jump_instr *instr)
       bld.emit(BRW_OPCODE_CONTINUE);
       break;
    case nir_jump_return:
+      /* This has to be the last block in the shader.  We don't handle
+       * early returns.
+       */
+      assert(nir_cf_node_next(&instr->instr.block->cf_node) == NULL &&
+             instr->instr.block->cf_node.parent->type == nir_cf_node_function);
+      break;
    default:
       unreachable("unknown jump");
    }

src/mesa/drivers/dri/i965/brw_gs.c

@@ -33,19 +33,23 @@
 #include "brw_state.h"
 #include "brw_ff_gs.h"
 
-
 bool
-brw_codegen_gs_prog(struct brw_context *brw,
+brw_compile_gs_prog(struct brw_context *brw,
                     struct gl_shader_program *prog,
                     struct brw_geometry_program *gp,
-                    struct brw_gs_prog_key *key)
+                    struct brw_gs_prog_key *key,
+                    struct brw_gs_compile_output *output)
 {
-   struct brw_stage_state *stage_state = &brw->gs.base;
    struct brw_gs_compile c;
    memset(&c, 0, sizeof(c));
    c.key = *key;
    c.gp = gp;
 
+   /* We get the bind map as input in the output struct...*/
+   c.prog_data.base.base.map_entries = output->prog_data.base.base.map_entries;
+   memcpy(c.prog_data.base.base.bind_map, output->prog_data.base.base.bind_map,
+          sizeof(c.prog_data.base.base.bind_map));
+
    c.prog_data.include_primitive_id =
       (gp->program.Base.InputsRead & VARYING_BIT_PRIMITIVE_ID) != 0;
 
@@ -274,18 +278,41 @@ brw_codegen_gs_prog(struct brw_context *brw,
 
       c.prog_data.base.base.total_scratch
          = brw_get_scratch_size(c.base.last_scratch*REG_SIZE);
+   }
 
+   output->mem_ctx = mem_ctx;
+   output->program = program;
+   output->program_size = program_size;
+   memcpy(&output->prog_data, &c.prog_data,
+          sizeof(output->prog_data));
+
+   return true;
+}
+
+bool
+brw_codegen_gs_prog(struct brw_context *brw,
+                    struct gl_shader_program *prog,
+                    struct brw_geometry_program *gp,
+                    struct brw_gs_prog_key *key)
+{
+   struct brw_gs_compile_output output;
+   struct brw_stage_state *stage_state = &brw->gs.base;
+
+   if (brw_compile_gs_prog(brw, prog, gp, key, &output))
+      return false;
+
+   if (output.prog_data.base.base.total_scratch) {
       brw_get_scratch_bo(brw, &stage_state->scratch_bo,
-			 c.prog_data.base.base.total_scratch *
+			 output.prog_data.base.base.total_scratch *
                          brw->max_gs_threads);
    }
 
    brw_upload_cache(&brw->cache, BRW_CACHE_GS_PROG,
-                    &c.key, sizeof(c.key),
-                    program, program_size,
-                    &c.prog_data, sizeof(c.prog_data),
+                    key, sizeof(*key),
+                    output.program, output.program_size,
+                    &output.prog_data, sizeof(output.prog_data),
                     &stage_state->prog_offset, &brw->gs.prog_data);
-   ralloc_free(mem_ctx);
+   ralloc_free(output.mem_ctx);
 
    return true;
 }

src/mesa/drivers/dri/i965/brw_gs.h

@@ -37,6 +37,22 @@ struct gl_context;
 struct gl_shader_program;
 struct gl_program;
 
+struct brw_gs_compile_output {
+   void *mem_ctx;
+   const void *program;
+   uint32_t program_size;
+   struct brw_gs_prog_data prog_data;
+};
+
+struct brw_gs_prog_key;
+
+bool
+brw_compile_gs_prog(struct brw_context *brw,
+                    struct gl_shader_program *prog,
+                    struct brw_geometry_program *gp,
+                    struct brw_gs_prog_key *key,
+                    struct brw_gs_compile_output *output);
+
 bool brw_gs_prog_data_compare(const void *a, const void *b);
 
 void

src/mesa/drivers/dri/i965/brw_nir.c

@@ -89,7 +89,6 @@ brw_create_nir(struct brw_context *brw,
    const nir_shader_compiler_options *options =
       ctx->Const.ShaderCompilerOptions[stage].NirOptions;
    struct gl_shader *shader = shader_prog ? shader_prog->_LinkedShaders[stage] : NULL;
-   bool debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
    nir_shader *nir;
 
    /* First, lower the GLSL IR or Mesa IR to NIR */
@@ -101,6 +100,28 @@ brw_create_nir(struct brw_context *brw,
    }
    nir_validate_shader(nir);
 
+   brw_process_nir(nir, brw->intelScreen->devinfo, shader_prog, stage);
+
+   static GLuint msg_id = 0;
+   _mesa_gl_debug(&brw->ctx, &msg_id,
+                  MESA_DEBUG_SOURCE_SHADER_COMPILER,
+                  MESA_DEBUG_TYPE_OTHER,
+                  MESA_DEBUG_SEVERITY_NOTIFICATION,
+                  "%s NIR shader: %d inst\n",
+                  _mesa_shader_stage_to_abbrev(stage),
+                  count_nir_instrs(nir));
+
+   return nir;
+}
+
+void
+brw_process_nir(nir_shader *nir,
+                const struct brw_device_info *devinfo,
+                const struct gl_shader_program *shader_prog,
+                gl_shader_stage stage)
+{
+   bool debug_enabled = INTEL_DEBUG & intel_debug_flag_for_shader_stage(stage);
+
    nir_lower_global_vars_to_local(nir);
    nir_validate_shader(nir);
 
@@ -136,8 +157,10 @@ brw_create_nir(struct brw_context *brw,
 
    if (shader_prog) {
       nir_lower_samplers(nir, shader_prog, stage);
-      nir_validate_shader(nir);
+   } else {
+      nir_lower_samplers_for_vk(nir);
    }
+   nir_validate_shader(nir);
 
    nir_lower_system_values(nir);
    nir_validate_shader(nir);
@@ -147,7 +170,7 @@ brw_create_nir(struct brw_context *brw,
 
    nir_optimize(nir);
 
-   if (brw->gen >= 6) {
+   if (devinfo->gen >= 6) {
       /* Try and fuse multiply-adds */
       nir_opt_peephole_ffma(nir);
       nir_validate_shader(nir);
@@ -178,15 +201,6 @@ brw_create_nir(struct brw_context *brw,
       nir_print_shader(nir, stderr);
    }
 
-   static GLuint msg_id = 0;
-   _mesa_gl_debug(&brw->ctx, &msg_id,
-                  MESA_DEBUG_SOURCE_SHADER_COMPILER,
-                  MESA_DEBUG_TYPE_OTHER,
-                  MESA_DEBUG_SEVERITY_NOTIFICATION,
-                  "%s NIR shader: %d inst\n",
-                  _mesa_shader_stage_to_abbrev(stage),
-                  count_nir_instrs(nir));
-
    nir_convert_from_ssa(nir);
    nir_validate_shader(nir);
 
@@ -195,7 +209,7 @@ brw_create_nir(struct brw_context *brw,
     * run it last because it stashes data in instr->pass_flags and we don't
     * want that to be squashed by other NIR passes.
     */
-   if (brw->gen <= 5)
+   if (devinfo->gen <= 5)
       brw_nir_analyze_boolean_resolves(nir);
 
    nir_sweep(nir);
@@ -205,6 +219,4 @@ brw_create_nir(struct brw_context *brw,
               _mesa_shader_stage_to_string(stage));
       nir_print_shader(nir, stderr);
    }
-
-   return nir;
 }

src/mesa/drivers/dri/i965/brw_nir.h

@@ -79,6 +79,12 @@ nir_shader *brw_create_nir(struct brw_context *brw,
                            const struct gl_program *prog,
                            gl_shader_stage stage);
 
+void
+brw_process_nir(nir_shader *nir,
+                const struct brw_device_info *devinfo,
+                const struct gl_shader_program *shader_prog,
+                gl_shader_stage stage);
+
 #ifdef __cplusplus
 }
 #endif

src/mesa/drivers/dri/i965/brw_program.c

@@ -275,7 +275,7 @@ brw_get_scratch_bo(struct brw_context *brw,
 
 void brwInitFragProgFuncs( struct dd_function_table *functions )
 {
-   assert(functions->ProgramStringNotify == _tnl_program_string);
+   /* assert(functions->ProgramStringNotify == _tnl_program_string); */
 
    functions->NewProgram = brwNewProgram;
    functions->DeleteProgram = brwDeleteProgram;

src/mesa/drivers/dri/i965/brw_shader.cpp

@@ -441,6 +441,7 @@ brw_type_for_base_type(const struct glsl_type *type)
    case GLSL_TYPE_ERROR:
    case GLSL_TYPE_INTERFACE:
    case GLSL_TYPE_DOUBLE:
+   case GLSL_TYPE_FUNCTION:
       unreachable("not reached");
    }
 

src/mesa/drivers/dri/i965/brw_state_cache.c

@@ -427,6 +427,9 @@ brw_destroy_cache(struct brw_context *brw, struct brw_cache *cache)
 
    DBG("%s\n", __func__);
 
+   if (cache->bo == NULL)
+      return;
+
    if (brw->has_llc)
       drm_intel_bo_unmap(cache->bo);
    drm_intel_bo_unreference(cache->bo);

src/mesa/drivers/dri/i965/brw_vec4.cpp

@@ -1882,7 +1882,7 @@ brw_vs_emit(struct brw_context *brw,
       st_index = brw_get_shader_time_index(brw, prog, &c->vp->program.Base,
                                            ST_VS);
 
-   if (unlikely(INTEL_DEBUG & DEBUG_VS))
+   if (unlikely(INTEL_DEBUG & DEBUG_VS) && shader->base.ir)
       brw_dump_ir("vertex", prog, &shader->base, &c->vp->program.Base);
 
    if (brw->intelScreen->compiler->scalar_vs) {

src/mesa/drivers/dri/i965/brw_vec4_visitor.cpp

@@ -615,6 +615,7 @@ type_size(const struct glsl_type *type)
    case GLSL_TYPE_DOUBLE:
    case GLSL_TYPE_ERROR:
    case GLSL_TYPE_INTERFACE:
+   case GLSL_TYPE_FUNCTION:
       unreachable("not reached");
    }
 

src/mesa/drivers/dri/i965/brw_wm.c

@@ -47,7 +47,7 @@
  * Return a bitfield where bit n is set if barycentric interpolation mode n
  * (see enum brw_wm_barycentric_interp_mode) is needed by the fragment shader.
  */
-static unsigned
+unsigned
 brw_compute_barycentric_interp_modes(struct brw_context *brw,
                                      bool shade_model_flat,
                                      bool persample_shading,

src/mesa/drivers/dri/i965/brw_wm.h

@@ -90,6 +90,12 @@ bool brw_wm_prog_data_compare(const void *a, const void *b);
 void
 brw_upload_wm_prog(struct brw_context *brw);
 
+unsigned
+brw_compute_barycentric_interp_modes(struct brw_context *brw,
+                                     bool shade_model_flat,
+                                     bool persample_shading,
+                                     const struct gl_fragment_program *fprog);
+
 #ifdef __cplusplus
 } // extern "C"
 #endif

src/mesa/drivers/dri/i965/intel_debug.c

@@ -60,7 +60,7 @@ static const struct dri_debug_control debug_control[] = {
    { "urb",         DEBUG_URB },
    { "vs",          DEBUG_VS },
    { "clip",        DEBUG_CLIP },
-   { "aub",         DEBUG_AUB },
+   { "foob",        DEBUG_AUB }, /* disable aub dumbing in the dri driver */
    { "shader_time", DEBUG_SHADER_TIME },
    { "no16",        DEBUG_NO16 },
    { "blorp",       DEBUG_BLORP },

src/mesa/drivers/dri/i965/intel_extensions.c

@@ -275,9 +275,11 @@ intelInitExtensions(struct gl_context *ctx)
       ctx->Extensions.EXT_shader_integer_mix = ctx->Const.GLSLVersion >= 130;
       ctx->Extensions.EXT_timer_query = true;
 
-      if (brw->gen == 5 || can_write_oacontrol(brw)) {
-         ctx->Extensions.AMD_performance_monitor = true;
-         ctx->Extensions.INTEL_performance_query = true;
+      if (brw->bufmgr) {
+         if (brw->gen == 5 || can_write_oacontrol(brw)) {
+            ctx->Extensions.AMD_performance_monitor = true;
+            ctx->Extensions.INTEL_performance_query = true;
+         }
       }
    }
 
@@ -285,6 +287,7 @@ intelInitExtensions(struct gl_context *ctx)
       uint64_t dummy;
 
       ctx->Extensions.ARB_blend_func_extended =
+         brw->optionCache.info == NULL ||
          !driQueryOptionb(&brw->optionCache, "disable_blend_func_extended");
       ctx->Extensions.ARB_conditional_render_inverted = true;
       ctx->Extensions.ARB_draw_buffers_blend = true;
@@ -308,7 +311,7 @@ intelInitExtensions(struct gl_context *ctx)
       ctx->Extensions.OES_depth_texture_cube_map = true;
 
       /* Test if the kernel has the ioctl. */
-      if (drm_intel_reg_read(brw->bufmgr, TIMESTAMP, &dummy) == 0)
+      if (brw->bufmgr && drm_intel_reg_read(brw->bufmgr, TIMESTAMP, &dummy) == 0)
          ctx->Extensions.ARB_timer_query = true;
 
       /* Only enable this in core profile because other parts of Mesa behave
@@ -331,7 +334,8 @@ intelInitExtensions(struct gl_context *ctx)
       ctx->Extensions.ARB_texture_compression_bptc = true;
       ctx->Extensions.ARB_texture_view = true;
 
-      if (can_do_pipelined_register_writes(brw)) {
+      if (brw->bufmgr &&
+          can_do_pipelined_register_writes(brw)) {
          ctx->Extensions.ARB_draw_indirect = true;
          ctx->Extensions.ARB_transform_feedback2 = true;
          ctx->Extensions.ARB_transform_feedback3 = true;
@@ -359,7 +363,9 @@ intelInitExtensions(struct gl_context *ctx)
    if (ctx->API != API_OPENGL_CORE)
       ctx->Extensions.ARB_color_buffer_float = true;
 
-   if (ctx->Mesa_DXTn || driQueryOptionb(&brw->optionCache, "force_s3tc_enable"))
+   if (ctx->Mesa_DXTn ||
+       (brw->optionCache.info != NULL &&
+        driQueryOptionb(&brw->optionCache, "force_s3tc_enable")))
       ctx->Extensions.EXT_texture_compression_s3tc = true;
 
    ctx->Extensions.ANGLE_texture_compression_dxt = true;

src/mesa/drivers/dri/i965/intel_screen.c

@@ -1426,6 +1426,78 @@ __DRIconfig **intelInitScreen2(__DRIscreen *psp)
    return (const __DRIconfig**) intel_screen_make_configs(psp);
 }
 
+struct intel_screen *
+intel_screen_create(int fd)
+{
+   __DRIscreen *psp;
+   __DRIconfig **configs;
+   int i;
+
+   psp = malloc(sizeof(*psp));
+   if (psp == NULL)
+      return NULL;
+
+   psp->image.loader = (void *) 1; /* Don't complain about this being NULL */
+   psp->fd = fd;
+   psp->dri2.useInvalidate = (void *) 1;
+
+   configs = (__DRIconfig **) intelInitScreen2(psp);
+   for (i = 0; configs[i]; i++)
+      free(configs[i]);
+   free(configs);
+
+   return psp->driverPrivate;
+}
+
+void
+intel_screen_destroy(struct intel_screen *screen)
+{
+   __DRIscreen *psp;
+
+   psp = screen->driScrnPriv;
+   intelDestroyScreen(screen->driScrnPriv);
+   free(psp);
+}
+
+
+struct brw_context *
+intel_context_create(struct intel_screen *screen)
+{
+   __DRIcontext *driContextPriv;
+   struct brw_context *brw;
+   unsigned error;
+
+   driContextPriv = malloc(sizeof(*driContextPriv));
+   if (driContextPriv == NULL)
+      return NULL;
+
+   driContextPriv->driScreenPriv = screen->driScrnPriv;
+
+   brwCreateContext(API_OPENGL_CORE,
+                    NULL, /* visual */
+                    driContextPriv,
+                    3, 0,
+                    0, /* flags */
+                    false, /* notify_reset */
+                    &error,
+                    NULL);
+
+   brw = driContextPriv->driverPrivate;
+   brw->ctx.FirstTimeCurrent = false;
+
+   return driContextPriv->driverPrivate;
+}
+
+void
+intel_context_destroy(struct brw_context *brw)
+{
+   __DRIcontext *driContextPriv;
+
+   driContextPriv = brw->driContext;
+   intelDestroyContext(driContextPriv);
+   free(driContextPriv);
+}
+
 struct intel_buffer {
    __DRIbuffer base;
    drm_intel_bo *bo;

src/mesa/main/mtypes.h

@@ -96,189 +96,6 @@ struct vbo_context;
 
 
 
-/**
- * Indexes for vertex program attributes.
- * GL_NV_vertex_program aliases generic attributes over the conventional
- * attributes.  In GL_ARB_vertex_program shader the aliasing is optional.
- * In GL_ARB_vertex_shader / OpenGL 2.0 the aliasing is disallowed (the
- * generic attributes are distinct/separate).
- */
-typedef enum
-{
-   VERT_ATTRIB_POS = 0,
-   VERT_ATTRIB_WEIGHT = 1,
-   VERT_ATTRIB_NORMAL = 2,
-   VERT_ATTRIB_COLOR0 = 3,
-   VERT_ATTRIB_COLOR1 = 4,
-   VERT_ATTRIB_FOG = 5,
-   VERT_ATTRIB_COLOR_INDEX = 6,
-   VERT_ATTRIB_EDGEFLAG = 7,
-   VERT_ATTRIB_TEX0 = 8,
-   VERT_ATTRIB_TEX1 = 9,
-   VERT_ATTRIB_TEX2 = 10,
-   VERT_ATTRIB_TEX3 = 11,
-   VERT_ATTRIB_TEX4 = 12,
-   VERT_ATTRIB_TEX5 = 13,
-   VERT_ATTRIB_TEX6 = 14,
-   VERT_ATTRIB_TEX7 = 15,
-   VERT_ATTRIB_POINT_SIZE = 16,
-   VERT_ATTRIB_GENERIC0 = 17,
-   VERT_ATTRIB_GENERIC1 = 18,
-   VERT_ATTRIB_GENERIC2 = 19,
-   VERT_ATTRIB_GENERIC3 = 20,
-   VERT_ATTRIB_GENERIC4 = 21,
-   VERT_ATTRIB_GENERIC5 = 22,
-   VERT_ATTRIB_GENERIC6 = 23,
-   VERT_ATTRIB_GENERIC7 = 24,
-   VERT_ATTRIB_GENERIC8 = 25,
-   VERT_ATTRIB_GENERIC9 = 26,
-   VERT_ATTRIB_GENERIC10 = 27,
-   VERT_ATTRIB_GENERIC11 = 28,
-   VERT_ATTRIB_GENERIC12 = 29,
-   VERT_ATTRIB_GENERIC13 = 30,
-   VERT_ATTRIB_GENERIC14 = 31,
-   VERT_ATTRIB_GENERIC15 = 32,
-   VERT_ATTRIB_MAX = 33
-} gl_vert_attrib;
-
-/**
- * Symbolic constats to help iterating over
- * specific blocks of vertex attributes.
- *
- * VERT_ATTRIB_FF
- *   includes all fixed function attributes as well as
- *   the aliased GL_NV_vertex_program shader attributes.
- * VERT_ATTRIB_TEX
- *   include the classic texture coordinate attributes.
- *   Is a subset of VERT_ATTRIB_FF.
- * VERT_ATTRIB_GENERIC
- *   include the OpenGL 2.0+ GLSL generic shader attributes.
- *   These alias the generic GL_ARB_vertex_shader attributes.
- */
-#define VERT_ATTRIB_FF(i)           (VERT_ATTRIB_POS + (i))
-#define VERT_ATTRIB_FF_MAX          VERT_ATTRIB_GENERIC0
-
-#define VERT_ATTRIB_TEX(i)          (VERT_ATTRIB_TEX0 + (i))
-#define VERT_ATTRIB_TEX_MAX         MAX_TEXTURE_COORD_UNITS
-
-#define VERT_ATTRIB_GENERIC(i)      (VERT_ATTRIB_GENERIC0 + (i))
-#define VERT_ATTRIB_GENERIC_MAX     MAX_VERTEX_GENERIC_ATTRIBS
-
-/**
- * Bitflags for vertex attributes.
- * These are used in bitfields in many places.
- */
-/*@{*/
-#define VERT_BIT_POS             BITFIELD64_BIT(VERT_ATTRIB_POS)
-#define VERT_BIT_WEIGHT          BITFIELD64_BIT(VERT_ATTRIB_WEIGHT)
-#define VERT_BIT_NORMAL          BITFIELD64_BIT(VERT_ATTRIB_NORMAL)
-#define VERT_BIT_COLOR0          BITFIELD64_BIT(VERT_ATTRIB_COLOR0)
-#define VERT_BIT_COLOR1          BITFIELD64_BIT(VERT_ATTRIB_COLOR1)
-#define VERT_BIT_FOG             BITFIELD64_BIT(VERT_ATTRIB_FOG)
-#define VERT_BIT_COLOR_INDEX     BITFIELD64_BIT(VERT_ATTRIB_COLOR_INDEX)
-#define VERT_BIT_EDGEFLAG        BITFIELD64_BIT(VERT_ATTRIB_EDGEFLAG)
-#define VERT_BIT_TEX0            BITFIELD64_BIT(VERT_ATTRIB_TEX0)
-#define VERT_BIT_TEX1            BITFIELD64_BIT(VERT_ATTRIB_TEX1)
-#define VERT_BIT_TEX2            BITFIELD64_BIT(VERT_ATTRIB_TEX2)
-#define VERT_BIT_TEX3            BITFIELD64_BIT(VERT_ATTRIB_TEX3)
-#define VERT_BIT_TEX4            BITFIELD64_BIT(VERT_ATTRIB_TEX4)
-#define VERT_BIT_TEX5            BITFIELD64_BIT(VERT_ATTRIB_TEX5)
-#define VERT_BIT_TEX6            BITFIELD64_BIT(VERT_ATTRIB_TEX6)
-#define VERT_BIT_TEX7            BITFIELD64_BIT(VERT_ATTRIB_TEX7)
-#define VERT_BIT_POINT_SIZE      BITFIELD64_BIT(VERT_ATTRIB_POINT_SIZE)
-#define VERT_BIT_GENERIC0        BITFIELD64_BIT(VERT_ATTRIB_GENERIC0)
-
-#define VERT_BIT(i)              BITFIELD64_BIT(i)
-#define VERT_BIT_ALL             BITFIELD64_RANGE(0, VERT_ATTRIB_MAX)
-
-#define VERT_BIT_FF(i)           VERT_BIT(i)
-#define VERT_BIT_FF_ALL          BITFIELD64_RANGE(0, VERT_ATTRIB_FF_MAX)
-#define VERT_BIT_TEX(i)          VERT_BIT(VERT_ATTRIB_TEX(i))
-#define VERT_BIT_TEX_ALL         \
-   BITFIELD64_RANGE(VERT_ATTRIB_TEX(0), VERT_ATTRIB_TEX_MAX)
-
-#define VERT_BIT_GENERIC(i)      VERT_BIT(VERT_ATTRIB_GENERIC(i))
-#define VERT_BIT_GENERIC_ALL     \
-   BITFIELD64_RANGE(VERT_ATTRIB_GENERIC(0), VERT_ATTRIB_GENERIC_MAX)
-/*@}*/
-
-
-/**
- * Indexes for vertex shader outputs, geometry shader inputs/outputs, and
- * fragment shader inputs.
- *
- * Note that some of these values are not available to all pipeline stages.
- *
- * When this enum is updated, the following code must be updated too:
- * - vertResults (in prog_print.c's arb_output_attrib_string())
- * - fragAttribs (in prog_print.c's arb_input_attrib_string())
- * - _mesa_varying_slot_in_fs()
- */
-typedef enum
-{
-   VARYING_SLOT_POS,
-   VARYING_SLOT_COL0, /* COL0 and COL1 must be contiguous */
-   VARYING_SLOT_COL1,
-   VARYING_SLOT_FOGC,
-   VARYING_SLOT_TEX0, /* TEX0-TEX7 must be contiguous */
-   VARYING_SLOT_TEX1,
-   VARYING_SLOT_TEX2,
-   VARYING_SLOT_TEX3,
-   VARYING_SLOT_TEX4,
-   VARYING_SLOT_TEX5,
-   VARYING_SLOT_TEX6,
-   VARYING_SLOT_TEX7,
-   VARYING_SLOT_PSIZ, /* Does not appear in FS */
-   VARYING_SLOT_BFC0, /* Does not appear in FS */
-   VARYING_SLOT_BFC1, /* Does not appear in FS */
-   VARYING_SLOT_EDGE, /* Does not appear in FS */
-   VARYING_SLOT_CLIP_VERTEX, /* Does not appear in FS */
-   VARYING_SLOT_CLIP_DIST0,
-   VARYING_SLOT_CLIP_DIST1,
-   VARYING_SLOT_PRIMITIVE_ID, /* Does not appear in VS */
-   VARYING_SLOT_LAYER, /* Appears as VS or GS output */
-   VARYING_SLOT_VIEWPORT, /* Appears as VS or GS output */
-   VARYING_SLOT_FACE, /* FS only */
-   VARYING_SLOT_PNTC, /* FS only */
-   VARYING_SLOT_VAR0, /* First generic varying slot */
-   VARYING_SLOT_MAX = VARYING_SLOT_VAR0 + MAX_VARYING
-} gl_varying_slot;
-
-
-/**
- * Bitflags for varying slots.
- */
-/*@{*/
-#define VARYING_BIT_POS BITFIELD64_BIT(VARYING_SLOT_POS)
-#define VARYING_BIT_COL0 BITFIELD64_BIT(VARYING_SLOT_COL0)
-#define VARYING_BIT_COL1 BITFIELD64_BIT(VARYING_SLOT_COL1)
-#define VARYING_BIT_FOGC BITFIELD64_BIT(VARYING_SLOT_FOGC)
-#define VARYING_BIT_TEX0 BITFIELD64_BIT(VARYING_SLOT_TEX0)
-#define VARYING_BIT_TEX1 BITFIELD64_BIT(VARYING_SLOT_TEX1)
-#define VARYING_BIT_TEX2 BITFIELD64_BIT(VARYING_SLOT_TEX2)
-#define VARYING_BIT_TEX3 BITFIELD64_BIT(VARYING_SLOT_TEX3)
-#define VARYING_BIT_TEX4 BITFIELD64_BIT(VARYING_SLOT_TEX4)
-#define VARYING_BIT_TEX5 BITFIELD64_BIT(VARYING_SLOT_TEX5)
-#define VARYING_BIT_TEX6 BITFIELD64_BIT(VARYING_SLOT_TEX6)
-#define VARYING_BIT_TEX7 BITFIELD64_BIT(VARYING_SLOT_TEX7)
-#define VARYING_BIT_TEX(U) BITFIELD64_BIT(VARYING_SLOT_TEX0 + (U))
-#define VARYING_BITS_TEX_ANY BITFIELD64_RANGE(VARYING_SLOT_TEX0, \
-                                              MAX_TEXTURE_COORD_UNITS)
-#define VARYING_BIT_PSIZ BITFIELD64_BIT(VARYING_SLOT_PSIZ)
-#define VARYING_BIT_BFC0 BITFIELD64_BIT(VARYING_SLOT_BFC0)
-#define VARYING_BIT_BFC1 BITFIELD64_BIT(VARYING_SLOT_BFC1)
-#define VARYING_BIT_EDGE BITFIELD64_BIT(VARYING_SLOT_EDGE)
-#define VARYING_BIT_CLIP_VERTEX BITFIELD64_BIT(VARYING_SLOT_CLIP_VERTEX)
-#define VARYING_BIT_CLIP_DIST0 BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0)
-#define VARYING_BIT_CLIP_DIST1 BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1)
-#define VARYING_BIT_PRIMITIVE_ID BITFIELD64_BIT(VARYING_SLOT_PRIMITIVE_ID)
-#define VARYING_BIT_LAYER BITFIELD64_BIT(VARYING_SLOT_LAYER)
-#define VARYING_BIT_VIEWPORT BITFIELD64_BIT(VARYING_SLOT_VIEWPORT)
-#define VARYING_BIT_FACE BITFIELD64_BIT(VARYING_SLOT_FACE)
-#define VARYING_BIT_PNTC BITFIELD64_BIT(VARYING_SLOT_PNTC)
-#define VARYING_BIT_VAR(V) BITFIELD64_BIT(VARYING_SLOT_VAR0 + (V))
-/*@}*/
-
 /**
  * Determine if the given gl_varying_slot appears in the fragment shader.
  */
@@ -299,28 +116,6 @@ _mesa_varying_slot_in_fs(gl_varying_slot slot)
 }
 
 
-/**
- * Fragment program results
- */
-typedef enum
-{
-   FRAG_RESULT_DEPTH = 0,
-   FRAG_RESULT_STENCIL = 1,
-   /* If a single color should be written to all render targets, this
-    * register is written.  No FRAG_RESULT_DATAn will be written.
-    */
-   FRAG_RESULT_COLOR = 2,
-   FRAG_RESULT_SAMPLE_MASK = 3,
-
-   /* FRAG_RESULT_DATAn are the per-render-target (GLSL gl_FragData[n]
-    * or ARB_fragment_program fragment.color[n]) color results.  If
-    * any are written, FRAG_RESULT_COLOR will not be written.
-    */
-   FRAG_RESULT_DATA0 = 4,
-   FRAG_RESULT_MAX = (FRAG_RESULT_DATA0 + MAX_DRAW_BUFFERS)
-} gl_frag_result;
-
-
 /**
  * Indexes for all renderbuffers
  */
@@ -2558,6 +2353,11 @@ struct gl_uniform_block
     */
    GLuint Binding;
 
+   /**
+    * Vulkan descriptor set qualifier for this block.
+    */
+   GLuint Set;
+
    /**
     * Minimum size (in bytes) of a buffer object to back this uniform buffer
     * (GL_UNIFORM_BLOCK_DATA_SIZE).

src/mesa/program/ir_to_mesa.cpp

@@ -542,6 +542,7 @@ type_size(const struct glsl_type *type)
    case GLSL_TYPE_VOID:
    case GLSL_TYPE_ERROR:
    case GLSL_TYPE_INTERFACE:
+   case GLSL_TYPE_FUNCTION:
       assert(!"Invalid type in type_size");
       break;
    }
@@ -2460,6 +2461,7 @@ _mesa_associate_uniform_storage(struct gl_context *ctx,
          case GLSL_TYPE_STRUCT:
          case GLSL_TYPE_ERROR:
          case GLSL_TYPE_INTERFACE:
+         case GLSL_TYPE_FUNCTION:
 	    assert(!"Should not get here.");
 	    break;
 	 }

src/vulkan/.gitignore

@@ -0,0 +1,4 @@
+# Generated source files
+/*-spirv.h
+/entrypoints.c
+/entrypoints.h

src/vulkan/Makefile.am

@@ -0,0 +1,95 @@
+# Copyright © 2015 Intel Corporation
+#
+# 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, sublicense,
+# 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+# THE AUTHORS OR COPYRIGHT HOLDERS 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.
+
+
+vulkan_includedir = $(includedir)/vulkan
+
+vulkan_include_HEADERS =				\
+	$(top_srcdir)/include/vulkan/vk_platform.h	\
+	$(top_srcdir)/include/vulkan/vulkan.h		\
+	$(top_srcdir)/include/vulkan/vulkan_intel.h	\
+	$(top_srcdir)/include/vulkan/vk_wsi_lunarg.h
+
+lib_LTLIBRARIES = libvulkan.la
+
+# The gallium includes are for the util/u_math.h include from main/macros.h
+
+AM_CPPFLAGS = \
+	$(INTEL_CFLAGS) \
+	$(VALGRIND_CFLAGS) \
+	$(DEFINES) \
+	-I$(top_srcdir)/include \
+	-I$(top_srcdir)/src \
+	-I$(top_srcdir)/src/mapi \
+	-I$(top_srcdir)/src/mesa \
+	-I$(top_srcdir)/src/mesa/drivers/dri/common \
+	-I$(top_srcdir)/src/mesa/drivers/dri/i965 \
+	-I$(top_srcdir)/src/gallium/auxiliary \
+	-I$(top_srcdir)/src/gallium/include \
+	-I$(top_builddir)/src/vulkan
+
+libvulkan_la_CFLAGS =							\
+	-Wall -Wextra -Wno-unused-parameter -fvisibility=hidden -O0 -g	\
+	-Wstrict-prototypes -Wmissing-prototypes -Wno-override-init
+
+libvulkan_la_CXXFLAGS =							\
+	-Wall -Wextra -Wno-unused-parameter -fvisibility=hidden -O0 -g
+
+libvulkan_la_SOURCES =				\
+	private.h				\
+	gem.c					\
+	device.c				\
+        anv_cmd_buffer.c                        \
+	anv_cmd_emit.c                          \
+	aub.c					\
+	allocator.c				\
+	util.c					\
+	pipeline.c				\
+	image.c					\
+	meta.c					\
+	intel.c					\
+	entrypoints.c				\
+	entrypoints.h				\
+	x11.c					\
+	formats.c				\
+	compiler.cpp				\
+	query.c
+
+BUILT_SOURCES =					\
+	entrypoints.h				\
+	entrypoints.c				\
+	meta-spirv.h
+
+entrypoints.h : vk_gen.py $(vulkan_include_HEADERS)
+	$(AM_V_GEN)cat $(vulkan_include_HEADERS) | $(PYTHON2) $< header > $@
+
+entrypoints.c : vk_gen.py $(vulkan_include_HEADERS)
+	$(AM_V_GEN)cat $(vulkan_include_HEADERS) | $(PYTHON2) $< code > $@
+
+%-spirv.h: %.c glsl_scraper.py
+	$(AM_V_GEN) $(PYTHON2) $(srcdir)/glsl_scraper.py --glsl-only -o $@ $<
+
+CLEANFILES = $(BUILT_SOURCES)
+
+libvulkan_la_LIBADD = -lxcb -lxcb-dri3 \
+	$(top_builddir)/src/mesa/drivers/dri/i965/libi965_compiler.la
+
+include $(top_srcdir)/install-lib-links.mk

src/vulkan/allocator.c

@@ -0,0 +1,665 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ */
+
+#define _DEFAULT_SOURCE
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <values.h>
+#include <assert.h>
+#include <linux/futex.h>
+#include <linux/memfd.h>
+#include <sys/time.h>
+#include <sys/mman.h>
+#include <sys/syscall.h>
+
+#include "private.h"
+
+#ifdef HAVE_VALGRIND
+#define VG_NOACCESS_READ(__ptr) ({                       \
+   VALGRIND_MAKE_MEM_DEFINED((__ptr), sizeof(*(__ptr))); \
+   __typeof(*(__ptr)) __val = *(__ptr);                  \
+   VALGRIND_MAKE_MEM_NOACCESS((__ptr), sizeof(*(__ptr)));\
+   __val;                                                \
+})
+#define VG_NOACCESS_WRITE(__ptr, __val) ({                  \
+   VALGRIND_MAKE_MEM_UNDEFINED((__ptr), sizeof(*(__ptr)));  \
+   *(__ptr) = (__val);                                      \
+   VALGRIND_MAKE_MEM_NOACCESS((__ptr), sizeof(*(__ptr)));   \
+})
+#else
+#define VG_NOACCESS_READ(__ptr) (*(__ptr))
+#define VG_NOACCESS_WRITE(__ptr, __val) (*(__ptr) = (__val))
+#endif
+
+/* Design goals:
+ *
+ *  - Lock free (except when resizing underlying bos)
+ *
+ *  - Constant time allocation with typically only one atomic
+ *
+ *  - Multiple allocation sizes without fragmentation
+ *
+ *  - Can grow while keeping addresses and offset of contents stable
+ *
+ *  - All allocations within one bo so we can point one of the
+ *    STATE_BASE_ADDRESS pointers at it.
+ *
+ * The overall design is a two-level allocator: top level is a fixed size, big
+ * block (8k) allocator, which operates out of a bo.  Allocation is done by
+ * either pulling a block from the free list or growing the used range of the
+ * bo.  Growing the range may run out of space in the bo which we then need to
+ * grow.  Growing the bo is tricky in a multi-threaded, lockless environment:
+ * we need to keep all pointers and contents in the old map valid.  GEM bos in
+ * general can't grow, but we use a trick: we create a memfd and use ftruncate
+ * to grow it as necessary.  We mmap the new size and then create a gem bo for
+ * it using the new gem userptr ioctl.  Without heavy-handed locking around
+ * our allocation fast-path, there isn't really a way to munmap the old mmap,
+ * so we just keep it around until garbage collection time.  While the block
+ * allocator is lockless for normal operations, we block other threads trying
+ * to allocate while we're growing the map.  It sholdn't happen often, and
+ * growing is fast anyway.
+ *
+ * At the next level we can use various sub-allocators.  The state pool is a
+ * pool of smaller, fixed size objects, which operates much like the block
+ * pool.  It uses a free list for freeing objects, but when it runs out of
+ * space it just allocates a new block from the block pool.  This allocator is
+ * intended for longer lived state objects such as SURFACE_STATE and most
+ * other persistent state objects in the API.  We may need to track more info
+ * with these object and a pointer back to the CPU object (eg VkImage).  In
+ * those cases we just allocate a slightly bigger object and put the extra
+ * state after the GPU state object.
+ *
+ * The state stream allocator works similar to how the i965 DRI driver streams
+ * all its state.  Even with Vulkan, we need to emit transient state (whether
+ * surface state base or dynamic state base), and for that we can just get a
+ * block and fill it up.  These cases are local to a command buffer and the
+ * sub-allocator need not be thread safe.  The streaming allocator gets a new
+ * block when it runs out of space and chains them together so they can be
+ * easily freed.
+ */
+
+/* Allocations are always at least 64 byte aligned, so 1 is an invalid value.
+ * We use it to indicate the free list is empty. */
+#define EMPTY 1
+
+struct anv_mmap_cleanup {
+   void *map;
+   size_t size;
+   uint32_t gem_handle;
+};
+
+#define ANV_MMAP_CLEANUP_INIT ((struct anv_mmap_cleanup){0})
+
+static inline long
+sys_futex(void *addr1, int op, int val1,
+          struct timespec *timeout, void *addr2, int val3)
+{
+   return syscall(SYS_futex, addr1, op, val1, timeout, addr2, val3);
+}
+
+static inline int
+futex_wake(uint32_t *addr, int count)
+{
+   return sys_futex(addr, FUTEX_WAKE, count, NULL, NULL, 0);
+}
+
+static inline int
+futex_wait(uint32_t *addr, int32_t value)
+{
+   return sys_futex(addr, FUTEX_WAIT, value, NULL, NULL, 0);
+}
+
+static inline int
+memfd_create(const char *name, unsigned int flags)
+{
+   return syscall(SYS_memfd_create, name, flags);
+}
+
+static inline uint32_t
+ilog2_round_up(uint32_t value)
+{
+   assert(value != 0);
+   return 32 - __builtin_clz(value - 1);
+}
+
+static inline uint32_t
+round_to_power_of_two(uint32_t value)
+{
+   return 1 << ilog2_round_up(value);
+}
+
+static bool
+anv_free_list_pop(union anv_free_list *list, void **map, uint32_t *offset)
+{
+   union anv_free_list current, next, old;
+
+   current = *list;
+   while (current.offset != EMPTY) {
+      /* We have to add a memory barrier here so that the list head (and
+       * offset) gets read before we read the map pointer.  This way we
+       * know that the map pointer is valid for the given offset at the
+       * point where we read it.
+       */
+      __sync_synchronize();
+
+      uint32_t *next_ptr = *map + current.offset;
+      next.offset = VG_NOACCESS_READ(next_ptr);
+      next.count = current.count + 1;
+      old.u64 = __sync_val_compare_and_swap(&list->u64, current.u64, next.u64);
+      if (old.u64 == current.u64) {
+         *offset = current.offset;
+         return true;
+      }
+      current = old;
+   }
+
+   return false;
+}
+
+static void
+anv_free_list_push(union anv_free_list *list, void *map, uint32_t offset)
+{
+   union anv_free_list current, old, new;
+   uint32_t *next_ptr = map + offset;
+
+   old = *list;
+   do {
+      current = old;
+      VG_NOACCESS_WRITE(next_ptr, current.offset);
+      new.offset = offset;
+      new.count = current.count + 1;
+      old.u64 = __sync_val_compare_and_swap(&list->u64, current.u64, new.u64);
+   } while (old.u64 != current.u64);
+}
+
+/* All pointers in the ptr_free_list are assumed to be page-aligned.  This
+ * means that the bottom 12 bits should all be zero.
+ */
+#define PFL_COUNT(x) ((uintptr_t)(x) & 0xfff)
+#define PFL_PTR(x) ((void *)((uintptr_t)(x) & ~0xfff))
+#define PFL_PACK(ptr, count) ({           \
+   assert(((uintptr_t)(ptr) & 0xfff) == 0); \
+   (void *)((uintptr_t)(ptr) | (uintptr_t)((count) & 0xfff)); \
+})
+
+static bool
+anv_ptr_free_list_pop(void **list, void **elem)
+{
+   void *current = *list;
+   while (PFL_PTR(current) != NULL) {
+      void **next_ptr = PFL_PTR(current);
+      void *new_ptr = VG_NOACCESS_READ(next_ptr);
+      unsigned new_count = PFL_COUNT(current) + 1;
+      void *new = PFL_PACK(new_ptr, new_count);
+      void *old = __sync_val_compare_and_swap(list, current, new);
+      if (old == current) {
+         *elem = PFL_PTR(current);
+         return true;
+      }
+      current = old;
+   }
+
+   return false;
+}
+
+static void
+anv_ptr_free_list_push(void **list, void *elem)
+{
+   void *old, *current;
+   void **next_ptr = elem;
+
+   old = *list;
+   do {
+      current = old;
+      VG_NOACCESS_WRITE(next_ptr, PFL_PTR(current));
+      unsigned new_count = PFL_COUNT(current) + 1;
+      void *new = PFL_PACK(elem, new_count);
+      old = __sync_val_compare_and_swap(list, current, new);
+   } while (old != current);
+}
+
+static int
+anv_block_pool_grow(struct anv_block_pool *pool);
+
+void
+anv_block_pool_init(struct anv_block_pool *pool,
+                    struct anv_device *device, uint32_t block_size)
+{
+   assert(is_power_of_two(block_size));
+
+   pool->device = device;
+   pool->bo.gem_handle = 0;
+   pool->bo.offset = 0;
+   pool->size = 0;
+   pool->block_size = block_size;
+   pool->next_block = 0;
+   pool->free_list = ANV_FREE_LIST_EMPTY;
+   anv_vector_init(&pool->mmap_cleanups,
+                   round_to_power_of_two(sizeof(struct anv_mmap_cleanup)), 128);
+
+   /* Immediately grow the pool so we'll have a backing bo. */
+   anv_block_pool_grow(pool);
+}
+
+void
+anv_block_pool_finish(struct anv_block_pool *pool)
+{
+   struct anv_mmap_cleanup *cleanup;
+
+   anv_vector_foreach(cleanup, &pool->mmap_cleanups) {
+      if (cleanup->map)
+         munmap(cleanup->map, cleanup->size);
+      if (cleanup->gem_handle)
+         anv_gem_close(pool->device, cleanup->gem_handle);
+   }
+
+   anv_vector_finish(&pool->mmap_cleanups);
+
+   close(pool->fd);
+}
+
+static int
+anv_block_pool_grow(struct anv_block_pool *pool)
+{
+   size_t size;
+   void *map;
+   int gem_handle;
+   struct anv_mmap_cleanup *cleanup;
+
+   if (pool->size == 0) {
+      size = 32 * pool->block_size;
+   } else {
+      size = pool->size * 2;
+   }
+
+   cleanup = anv_vector_add(&pool->mmap_cleanups);
+   if (!cleanup)
+      return -1;
+   *cleanup = ANV_MMAP_CLEANUP_INIT;
+
+   if (pool->size == 0)
+      pool->fd = memfd_create("block pool", MFD_CLOEXEC);
+
+   if (pool->fd == -1)
+      return -1;
+
+   if (ftruncate(pool->fd, size) == -1)
+      return -1;
+
+   /* First try to see if mremap can grow the map in place. */
+   map = MAP_FAILED;
+   if (pool->size > 0)
+      map = mremap(pool->map, pool->size, size, 0);
+   if (map == MAP_FAILED) {
+      /* Just leak the old map until we destroy the pool.  We can't munmap it
+       * without races or imposing locking on the block allocate fast path. On
+       * the whole the leaked maps adds up to less than the size of the
+       * current map.  MAP_POPULATE seems like the right thing to do, but we
+       * should try to get some numbers.
+       */
+      map = mmap(NULL, size, PROT_READ | PROT_WRITE,
+                 MAP_SHARED | MAP_POPULATE, pool->fd, 0);
+      cleanup->map = map;
+      cleanup->size = size;
+   }
+   if (map == MAP_FAILED)
+      return -1;
+
+   gem_handle = anv_gem_userptr(pool->device, map, size);
+   if (gem_handle == 0)
+      return -1;
+   cleanup->gem_handle = gem_handle;
+
+   /* Now that we successfull allocated everything, we can write the new
+    * values back into pool. */
+   pool->map = map;
+   pool->bo.gem_handle = gem_handle;
+   pool->bo.size = size;
+   pool->bo.map = map;
+   pool->bo.index = 0;
+
+   /* Write size last and after the memory barrier here. We need the memory
+    * barrier to make sure map and gem_handle are written before other threads
+    * see the new size.  A thread could allocate a block and then go try using
+    * the old pool->map and access out of bounds. */
+
+   __sync_synchronize();
+   pool->size = size;
+
+   return 0;
+}
+
+uint32_t
+anv_block_pool_alloc(struct anv_block_pool *pool)
+{
+   uint32_t offset, block, size;
+
+   /* Try free list first. */
+   if (anv_free_list_pop(&pool->free_list, &pool->map, &offset)) {
+      assert(pool->map);
+      return offset;
+   }
+
+ restart:
+   size = pool->size;
+   block = __sync_fetch_and_add(&pool->next_block, pool->block_size);
+   if (block < size) {
+      assert(pool->map);
+      return block;
+   } else if (block == size) {
+      /* We allocated the first block outside the pool, we have to grow it.
+       * pool->next_block acts a mutex: threads who try to allocate now will
+       * get block indexes above the current limit and hit futex_wait
+       * below. */
+      int err = anv_block_pool_grow(pool);
+      assert(err == 0);
+      (void) err;
+      futex_wake(&pool->size, INT_MAX);
+   } else {
+      futex_wait(&pool->size, size);
+      __sync_fetch_and_add(&pool->next_block, -pool->block_size);
+      goto restart;
+   }
+
+   return block;
+}
+
+void
+anv_block_pool_free(struct anv_block_pool *pool, uint32_t offset)
+{
+   anv_free_list_push(&pool->free_list, pool->map, offset);
+}
+
+static void
+anv_fixed_size_state_pool_init(struct anv_fixed_size_state_pool *pool,
+                               size_t state_size)
+{
+   /* At least a cache line and must divide the block size. */
+   assert(state_size >= 64 && is_power_of_two(state_size));
+
+   pool->state_size = state_size;
+   pool->free_list = ANV_FREE_LIST_EMPTY;
+   pool->block.next = 0;
+   pool->block.end = 0;
+}
+
+static uint32_t
+anv_fixed_size_state_pool_alloc(struct anv_fixed_size_state_pool *pool,
+                                struct anv_block_pool *block_pool)
+{
+   uint32_t offset;
+   struct anv_block_state block, old, new;
+
+   /* Try free list first. */
+   if (anv_free_list_pop(&pool->free_list, &block_pool->map, &offset))
+      return offset;
+
+   /* If free list was empty (or somebody raced us and took the items) we
+    * allocate a new item from the end of the block */
+ restart:
+   block.u64 = __sync_fetch_and_add(&pool->block.u64, pool->state_size);
+
+   if (block.next < block.end) {
+      return block.next;
+   } else if (block.next == block.end) {
+      new.next = anv_block_pool_alloc(block_pool);
+      new.end = new.next + block_pool->block_size;
+      old.u64 = __sync_fetch_and_add(&pool->block.u64, new.u64 - block.u64);
+      if (old.next != block.next)
+         futex_wake(&pool->block.end, INT_MAX);
+      return new.next;
+   } else {
+      futex_wait(&pool->block.end, block.end);
+      __sync_fetch_and_add(&pool->block.u64, -pool->state_size);
+      goto restart;
+   }
+}
+
+static void
+anv_fixed_size_state_pool_free(struct anv_fixed_size_state_pool *pool,
+                               struct anv_block_pool *block_pool,
+                               uint32_t offset)
+{
+   anv_free_list_push(&pool->free_list, block_pool->map, offset);
+}
+
+void
+anv_state_pool_init(struct anv_state_pool *pool,
+                    struct anv_block_pool *block_pool)
+{
+   pool->block_pool = block_pool;
+   for (unsigned i = 0; i < ANV_STATE_BUCKETS; i++) {
+      size_t size = 1 << (ANV_MIN_STATE_SIZE_LOG2 + i);
+      anv_fixed_size_state_pool_init(&pool->buckets[i], size);
+   }
+}
+
+struct anv_state
+anv_state_pool_alloc(struct anv_state_pool *pool, size_t size, size_t align)
+{
+   unsigned size_log2 = ilog2_round_up(size < align ? align : size);
+   assert(size_log2 <= ANV_MAX_STATE_SIZE_LOG2);
+   if (size_log2 < ANV_MIN_STATE_SIZE_LOG2)
+      size_log2 = ANV_MIN_STATE_SIZE_LOG2;
+   unsigned bucket = size_log2 - ANV_MIN_STATE_SIZE_LOG2;
+
+   struct anv_state state;
+   state.alloc_size = 1 << size_log2;
+   state.offset = anv_fixed_size_state_pool_alloc(&pool->buckets[bucket],
+                                                  pool->block_pool);
+   state.map = pool->block_pool->map + state.offset;
+   VG(VALGRIND_MALLOCLIKE_BLOCK(state.map, size, 0, false));
+   return state;
+}
+
+void
+anv_state_pool_free(struct anv_state_pool *pool, struct anv_state state)
+{
+   assert(is_power_of_two(state.alloc_size));
+   unsigned size_log2 = ilog2_round_up(state.alloc_size);
+   assert(size_log2 >= ANV_MIN_STATE_SIZE_LOG2 &&
+          size_log2 <= ANV_MAX_STATE_SIZE_LOG2);
+   unsigned bucket = size_log2 - ANV_MIN_STATE_SIZE_LOG2;
+
+   VG(VALGRIND_FREELIKE_BLOCK(state.map, 0));
+   anv_fixed_size_state_pool_free(&pool->buckets[bucket],
+                                  pool->block_pool, state.offset);
+}
+
+#define NULL_BLOCK 1
+struct stream_block {
+   uint32_t next;
+
+   /* The map for the BO at the time the block was givne to us */
+   void *current_map;
+
+#ifdef HAVE_VALGRIND
+   void *_vg_ptr;
+#endif
+};
+
+/* The state stream allocator is a one-shot, single threaded allocator for
+ * variable sized blocks.  We use it for allocating dynamic state.
+ */
+void
+anv_state_stream_init(struct anv_state_stream *stream,
+                      struct anv_block_pool *block_pool)
+{
+   stream->block_pool = block_pool;
+   stream->next = 0;
+   stream->end = 0;
+   stream->current_block = NULL_BLOCK;
+}
+
+void
+anv_state_stream_finish(struct anv_state_stream *stream)
+{
+   struct stream_block *sb;
+   uint32_t block, next_block;
+
+   block = stream->current_block;
+   while (block != NULL_BLOCK) {
+      sb = stream->block_pool->map + block;
+      next_block = VG_NOACCESS_READ(&sb->next);
+      VG(VALGRIND_FREELIKE_BLOCK(VG_NOACCESS_READ(&sb->_vg_ptr), 0));
+      anv_block_pool_free(stream->block_pool, block);
+      block = next_block;
+   }
+}
+
+struct anv_state
+anv_state_stream_alloc(struct anv_state_stream *stream,
+                       uint32_t size, uint32_t alignment)
+{
+   struct stream_block *sb;
+   struct anv_state state;
+   uint32_t block;
+
+   state.offset = align_u32(stream->next, alignment);
+   if (state.offset + size > stream->end) {
+      block = anv_block_pool_alloc(stream->block_pool);
+      void *current_map = stream->block_pool->map;
+      sb = current_map + block;
+      VG_NOACCESS_WRITE(&sb->current_map, current_map);
+      VG_NOACCESS_WRITE(&sb->next, stream->current_block);
+      VG(VG_NOACCESS_WRITE(&sb->_vg_ptr, 0));
+      stream->current_block = block;
+      stream->next = block + sizeof(*sb);
+      stream->end = block + stream->block_pool->block_size;
+      state.offset = align_u32(stream->next, alignment);
+      assert(state.offset + size <= stream->end);
+   }
+
+   sb = stream->block_pool->map + stream->current_block;
+   void *current_map = VG_NOACCESS_READ(&sb->current_map);
+
+   state.map = current_map + state.offset;
+   state.alloc_size = size;
+
+#ifdef HAVE_VALGRIND
+   void *vg_ptr = VG_NOACCESS_READ(&sb->_vg_ptr);
+   if (vg_ptr == NULL) {
+      vg_ptr = state.map;
+      VG_NOACCESS_WRITE(&sb->_vg_ptr, vg_ptr);
+      VALGRIND_MALLOCLIKE_BLOCK(vg_ptr, size, 0, false);
+   } else {
+      ptrdiff_t vg_offset = vg_ptr - current_map;
+      assert(vg_offset >= stream->current_block &&
+             vg_offset < stream->end);
+      VALGRIND_RESIZEINPLACE_BLOCK(vg_ptr,
+                                   stream->next - vg_offset,
+                                   (state.offset + size) - vg_offset,
+                                   0);
+   }
+#endif
+
+   stream->next = state.offset + size;
+
+   return state;
+}
+
+struct bo_pool_bo_link {
+   struct bo_pool_bo_link *next;
+   struct anv_bo bo;
+};
+
+void
+anv_bo_pool_init(struct anv_bo_pool *pool,
+                 struct anv_device *device, uint32_t bo_size)
+{
+   pool->device = device;
+   pool->bo_size = bo_size;
+   pool->free_list = NULL;
+}
+
+void
+anv_bo_pool_finish(struct anv_bo_pool *pool)
+{
+   struct bo_pool_bo_link *link = PFL_PTR(pool->free_list);
+   while (link != NULL) {
+      struct bo_pool_bo_link link_copy = VG_NOACCESS_READ(link);
+
+      /* The anv_gem_m[un]map() functions are also valgrind-safe so they
+       * act as an alloc/free.  In order to avoid a double-free warning, we
+       * need to mark thiss as malloc'd before we unmap it.
+       */
+      VG(VALGRIND_MALLOCLIKE_BLOCK(link_copy.bo.map, pool->bo_size, 0, false));
+
+      anv_gem_munmap(link_copy.bo.map, pool->bo_size);
+      anv_gem_close(pool->device, link_copy.bo.gem_handle);
+      link = link_copy.next;
+   }
+}
+
+VkResult
+anv_bo_pool_alloc(struct anv_bo_pool *pool, struct anv_bo *bo)
+{
+   VkResult result;
+
+   void *next_free_void;
+   if (anv_ptr_free_list_pop(&pool->free_list, &next_free_void)) {
+      struct bo_pool_bo_link *next_free = next_free_void;
+      *bo = VG_NOACCESS_READ(&next_free->bo);
+      assert(bo->map == next_free);
+      assert(bo->size == pool->bo_size);
+
+      VG(VALGRIND_MALLOCLIKE_BLOCK(bo->map, pool->bo_size, 0, false));
+
+      return VK_SUCCESS;
+   }
+
+   struct anv_bo new_bo;
+
+   result = anv_bo_init_new(&new_bo, pool->device, pool->bo_size);
+   if (result != VK_SUCCESS)
+      return result;
+
+   assert(new_bo.size == pool->bo_size);
+
+   new_bo.map = anv_gem_mmap(pool->device, new_bo.gem_handle, 0, pool->bo_size);
+   if (new_bo.map == NULL) {
+      anv_gem_close(pool->device, new_bo.gem_handle);
+      return vk_error(VK_ERROR_MEMORY_MAP_FAILED);
+   }
+
+   /* We don't need to call VALGRIND_MALLOCLIKE_BLOCK here because gem_mmap
+    * calls it for us.  If we really want to be pedantic we could do a
+    * VALGRIND_FREELIKE_BLOCK right after the mmap, but there's no good
+    * reason.
+    */
+
+   *bo = new_bo;
+   return VK_SUCCESS;
+}
+
+void
+anv_bo_pool_free(struct anv_bo_pool *pool, const struct anv_bo *bo)
+{
+   struct bo_pool_bo_link *link = bo->map;
+   link->bo = *bo;
+
+   VG(VALGRIND_FREELIKE_BLOCK(bo->map, 0));
+   anv_ptr_free_list_push(&pool->free_list, link);
+}

src/vulkan/anv_cmd_buffer.c

@@ -0,0 +1,706 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 <assert.h>
+#include <stdbool.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+#include "private.h"
+
+/** \file anv_cmd_buffer.c
+ *
+ * This file contains functions related to anv_cmd_buffer as a data
+ * structure.  This involves everything required to create and destroy
+ * the actual batch buffers as well as link them together and handle
+ * relocations and surface state.  It specifically does *not* contain any
+ * handling of actual vkCmd calls beyond vkCmdExecuteCommands.
+ */
+
+/*-----------------------------------------------------------------------*
+ * Functions related to anv_reloc_list
+ *-----------------------------------------------------------------------*/
+
+VkResult
+anv_reloc_list_init(struct anv_reloc_list *list, struct anv_device *device)
+{
+   list->num_relocs = 0;
+   list->array_length = 256;
+   list->relocs =
+      anv_device_alloc(device, list->array_length * sizeof(*list->relocs), 8,
+                       VK_SYSTEM_ALLOC_TYPE_INTERNAL);
+
+   if (list->relocs == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   list->reloc_bos =
+      anv_device_alloc(device, list->array_length * sizeof(*list->reloc_bos), 8,
+                       VK_SYSTEM_ALLOC_TYPE_INTERNAL);
+
+   if (list->relocs == NULL) {
+      anv_device_free(device, list->relocs);
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+   }
+
+   return VK_SUCCESS;
+}
+
+void
+anv_reloc_list_finish(struct anv_reloc_list *list, struct anv_device *device)
+{
+   anv_device_free(device, list->relocs);
+   anv_device_free(device, list->reloc_bos);
+}
+
+static VkResult
+anv_reloc_list_grow(struct anv_reloc_list *list, struct anv_device *device,
+                    size_t num_additional_relocs)
+{
+   if (list->num_relocs + num_additional_relocs <= list->array_length)
+      return VK_SUCCESS;
+
+   size_t new_length = list->array_length * 2;
+   while (new_length < list->num_relocs + num_additional_relocs)
+      new_length *= 2;
+
+   struct drm_i915_gem_relocation_entry *new_relocs =
+      anv_device_alloc(device, new_length * sizeof(*list->relocs), 8,
+                       VK_SYSTEM_ALLOC_TYPE_INTERNAL);
+   if (new_relocs == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   struct anv_bo **new_reloc_bos =
+      anv_device_alloc(device, new_length * sizeof(*list->reloc_bos), 8,
+                       VK_SYSTEM_ALLOC_TYPE_INTERNAL);
+   if (new_relocs == NULL) {
+      anv_device_free(device, new_relocs);
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+   }
+
+   memcpy(new_relocs, list->relocs, list->num_relocs * sizeof(*list->relocs));
+   memcpy(new_reloc_bos, list->reloc_bos,
+          list->num_relocs * sizeof(*list->reloc_bos));
+
+   anv_device_free(device, list->relocs);
+   anv_device_free(device, list->reloc_bos);
+
+   list->relocs = new_relocs;
+   list->reloc_bos = new_reloc_bos;
+
+   return VK_SUCCESS;
+}
+
+uint64_t
+anv_reloc_list_add(struct anv_reloc_list *list, struct anv_device *device,
+                   uint32_t offset, struct anv_bo *target_bo, uint32_t delta)
+{
+   struct drm_i915_gem_relocation_entry *entry;
+   int index;
+
+   anv_reloc_list_grow(list, device, 1);
+   /* TODO: Handle failure */
+
+   /* XXX: Can we use I915_EXEC_HANDLE_LUT? */
+   index = list->num_relocs++;
+   list->reloc_bos[index] = target_bo;
+   entry = &list->relocs[index];
+   entry->target_handle = target_bo->gem_handle;
+   entry->delta = delta;
+   entry->offset = offset;
+   entry->presumed_offset = target_bo->offset;
+   entry->read_domains = 0;
+   entry->write_domain = 0;
+
+   return target_bo->offset + delta;
+}
+
+static void
+anv_reloc_list_append(struct anv_reloc_list *list, struct anv_device *device,
+                      struct anv_reloc_list *other, uint32_t offset)
+{
+   anv_reloc_list_grow(list, device, other->num_relocs);
+   /* TODO: Handle failure */
+
+   memcpy(&list->relocs[list->num_relocs], &other->relocs[0],
+          other->num_relocs * sizeof(other->relocs[0]));
+   memcpy(&list->reloc_bos[list->num_relocs], &other->reloc_bos[0],
+          other->num_relocs * sizeof(other->reloc_bos[0]));
+
+   for (uint32_t i = 0; i < other->num_relocs; i++)
+      list->relocs[i + list->num_relocs].offset += offset;
+
+   list->num_relocs += other->num_relocs;
+}
+
+/*-----------------------------------------------------------------------*
+ * Functions related to anv_batch
+ *-----------------------------------------------------------------------*/
+
+void *
+anv_batch_emit_dwords(struct anv_batch *batch, int num_dwords)
+{
+   if (batch->next + num_dwords * 4 > batch->end)
+      batch->extend_cb(batch, batch->user_data);
+
+   void *p = batch->next;
+
+   batch->next += num_dwords * 4;
+   assert(batch->next <= batch->end);
+
+   return p;
+}
+
+uint64_t
+anv_batch_emit_reloc(struct anv_batch *batch,
+                     void *location, struct anv_bo *bo, uint32_t delta)
+{
+   return anv_reloc_list_add(&batch->relocs, batch->device,
+                             location - batch->start, bo, delta);
+}
+
+void
+anv_batch_emit_batch(struct anv_batch *batch, struct anv_batch *other)
+{
+   uint32_t size, offset;
+
+   size = other->next - other->start;
+   assert(size % 4 == 0);
+
+   if (batch->next + size > batch->end)
+      batch->extend_cb(batch, batch->user_data);
+
+   assert(batch->next + size <= batch->end);
+
+   memcpy(batch->next, other->start, size);
+
+   offset = batch->next - batch->start;
+   anv_reloc_list_append(&batch->relocs, batch->device,
+                         &other->relocs, offset);
+
+   batch->next += size;
+}
+
+/*-----------------------------------------------------------------------*
+ * Functions related to anv_batch_bo
+ *-----------------------------------------------------------------------*/
+
+static VkResult
+anv_batch_bo_create(struct anv_device *device, struct anv_batch_bo **bbo_out)
+{
+   VkResult result;
+
+   struct anv_batch_bo *bbo =
+      anv_device_alloc(device, sizeof(*bbo), 8, VK_SYSTEM_ALLOC_TYPE_INTERNAL);
+   if (bbo == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   bbo->num_relocs = 0;
+   bbo->prev_batch_bo = NULL;
+
+   result = anv_bo_pool_alloc(&device->batch_bo_pool, &bbo->bo);
+   if (result != VK_SUCCESS) {
+      anv_device_free(device, bbo);
+      return result;
+   }
+
+   *bbo_out = bbo;
+
+   return VK_SUCCESS;
+}
+
+static void
+anv_batch_bo_start(struct anv_batch_bo *bbo, struct anv_batch *batch,
+                   size_t batch_padding)
+{
+   batch->next = batch->start = bbo->bo.map;
+   batch->end = bbo->bo.map + bbo->bo.size - batch_padding;
+   bbo->first_reloc = batch->relocs.num_relocs;
+}
+
+static void
+anv_batch_bo_finish(struct anv_batch_bo *bbo, struct anv_batch *batch)
+{
+   /* Round batch up to an even number of dwords. */
+   if ((batch->next - batch->start) & 4)
+      anv_batch_emit(batch, GEN8_MI_NOOP);
+
+   assert(batch->start == bbo->bo.map);
+   bbo->length = batch->next - batch->start;
+   VG(VALGRIND_CHECK_MEM_IS_DEFINED(batch->start, bbo->length));
+   bbo->num_relocs = batch->relocs.num_relocs - bbo->first_reloc;
+}
+
+static void
+anv_batch_bo_destroy(struct anv_batch_bo *bbo, struct anv_device *device)
+{
+   anv_bo_pool_free(&device->batch_bo_pool, &bbo->bo);
+   anv_device_free(device, bbo);
+}
+
+/*-----------------------------------------------------------------------*
+ * Functions related to anv_batch_bo
+ *-----------------------------------------------------------------------*/
+
+static VkResult
+anv_cmd_buffer_chain_batch(struct anv_batch *batch, void *_data)
+{
+   struct anv_cmd_buffer *cmd_buffer = _data;
+
+   struct anv_batch_bo *new_bbo, *old_bbo = cmd_buffer->last_batch_bo;
+
+   VkResult result = anv_batch_bo_create(cmd_buffer->device, &new_bbo);
+   if (result != VK_SUCCESS)
+      return result;
+
+   /* We set the end of the batch a little short so we would be sure we
+    * have room for the chaining command.  Since we're about to emit the
+    * chaining command, let's set it back where it should go.
+    */
+   batch->end += GEN8_MI_BATCH_BUFFER_START_length * 4;
+   assert(batch->end == old_bbo->bo.map + old_bbo->bo.size);
+
+   anv_batch_emit(batch, GEN8_MI_BATCH_BUFFER_START,
+      GEN8_MI_BATCH_BUFFER_START_header,
+      ._2ndLevelBatchBuffer = _1stlevelbatch,
+      .AddressSpaceIndicator = ASI_PPGTT,
+      .BatchBufferStartAddress = { &new_bbo->bo, 0 },
+   );
+
+   anv_batch_bo_finish(cmd_buffer->last_batch_bo, batch);
+
+   new_bbo->prev_batch_bo = old_bbo;
+   cmd_buffer->last_batch_bo = new_bbo;
+
+   anv_batch_bo_start(new_bbo, batch, GEN8_MI_BATCH_BUFFER_START_length * 4);
+
+   return VK_SUCCESS;
+}
+
+struct anv_state
+anv_cmd_buffer_alloc_surface_state(struct anv_cmd_buffer *cmd_buffer,
+                                   uint32_t size, uint32_t alignment)
+{
+   struct anv_state state;
+
+   state.offset = align_u32(cmd_buffer->surface_next, alignment);
+   if (state.offset + size > cmd_buffer->surface_batch_bo->bo.size)
+      return (struct anv_state) { 0 };
+
+   state.map = cmd_buffer->surface_batch_bo->bo.map + state.offset;
+   state.alloc_size = size;
+   cmd_buffer->surface_next = state.offset + size;
+
+   assert(state.offset + size <= cmd_buffer->surface_batch_bo->bo.size);
+
+   return state;
+}
+
+struct anv_state
+anv_cmd_buffer_alloc_dynamic_state(struct anv_cmd_buffer *cmd_buffer,
+                                   uint32_t size, uint32_t alignment)
+{
+   return anv_state_stream_alloc(&cmd_buffer->dynamic_state_stream,
+                                 size, alignment);
+}
+
+VkResult
+anv_cmd_buffer_new_surface_state_bo(struct anv_cmd_buffer *cmd_buffer)
+{
+   struct anv_batch_bo *new_bbo, *old_bbo = cmd_buffer->surface_batch_bo;
+
+   /* Finish off the old buffer */
+   old_bbo->num_relocs =
+      cmd_buffer->surface_relocs.num_relocs - old_bbo->first_reloc;
+   old_bbo->length = cmd_buffer->surface_next;
+
+   VkResult result = anv_batch_bo_create(cmd_buffer->device, &new_bbo);
+   if (result != VK_SUCCESS)
+      return result;
+
+   new_bbo->first_reloc = cmd_buffer->surface_relocs.num_relocs;
+   cmd_buffer->surface_next = 1;
+
+   new_bbo->prev_batch_bo = old_bbo;
+   cmd_buffer->surface_batch_bo = new_bbo;
+
+   /* Re-emit state base addresses so we get the new surface state base
+    * address before we start emitting binding tables etc.
+    */
+   anv_cmd_buffer_emit_state_base_address(cmd_buffer);
+
+   /* After re-setting the surface state base address, we have to do some
+    * cache flusing so that the sampler engine will pick up the new
+    * SURFACE_STATE objects and binding tables. From the Broadwell PRM,
+    * Shared Function > 3D Sampler > State > State Caching (page 96):
+    *
+    *    Coherency with system memory in the state cache, like the texture
+    *    cache is handled partially by software. It is expected that the
+    *    command stream or shader will issue Cache Flush operation or
+    *    Cache_Flush sampler message to ensure that the L1 cache remains
+    *    coherent with system memory.
+    *
+    *    [...]
+    *
+    *    Whenever the value of the Dynamic_State_Base_Addr,
+    *    Surface_State_Base_Addr are altered, the L1 state cache must be
+    *    invalidated to ensure the new surface or sampler state is fetched
+    *    from system memory.
+    *
+    * The PIPE_CONTROL command has a "State Cache Invalidation Enable" bit
+    * which, according the PIPE_CONTROL instruction documentation in the
+    * Broadwell PRM:
+    *
+    *    Setting this bit is independent of any other bit in this packet.
+    *    This bit controls the invalidation of the L1 and L2 state caches
+    *    at the top of the pipe i.e. at the parsing time.
+    *
+    * Unfortunately, experimentation seems to indicate that state cache
+    * invalidation through a PIPE_CONTROL does nothing whatsoever in
+    * regards to surface state and binding tables.  In stead, it seems that
+    * invalidating the texture cache is what is actually needed.
+    *
+    * XXX:  As far as we have been able to determine through
+    * experimentation, shows that flush the texture cache appears to be
+    * sufficient.  The theory here is that all of the sampling/rendering
+    * units cache the binding table in the texture cache.  However, we have
+    * yet to be able to actually confirm this.
+    */
+   anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
+                  .TextureCacheInvalidationEnable = true);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_CreateCommandBuffer(
+    VkDevice                                    _device,
+    const VkCmdBufferCreateInfo*                pCreateInfo,
+    VkCmdBuffer*                                pCmdBuffer)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   struct anv_cmd_buffer *cmd_buffer;
+   VkResult result;
+
+   assert(pCreateInfo->level == VK_CMD_BUFFER_LEVEL_PRIMARY);
+
+   cmd_buffer = anv_device_alloc(device, sizeof(*cmd_buffer), 8,
+                                 VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (cmd_buffer == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   cmd_buffer->device = device;
+
+   result = anv_batch_bo_create(device, &cmd_buffer->last_batch_bo);
+   if (result != VK_SUCCESS)
+      goto fail;
+
+   result = anv_reloc_list_init(&cmd_buffer->batch.relocs, device);
+   if (result != VK_SUCCESS)
+      goto fail_batch_bo;
+
+   cmd_buffer->batch.device = device;
+   cmd_buffer->batch.extend_cb = anv_cmd_buffer_chain_batch;
+   cmd_buffer->batch.user_data = cmd_buffer;
+
+   anv_batch_bo_start(cmd_buffer->last_batch_bo, &cmd_buffer->batch,
+                      GEN8_MI_BATCH_BUFFER_START_length * 4);
+
+   result = anv_batch_bo_create(device, &cmd_buffer->surface_batch_bo);
+   if (result != VK_SUCCESS)
+      goto fail_batch_relocs;
+   cmd_buffer->surface_batch_bo->first_reloc = 0;
+
+   result = anv_reloc_list_init(&cmd_buffer->surface_relocs, device);
+   if (result != VK_SUCCESS)
+      goto fail_ss_batch_bo;
+
+   /* Start surface_next at 1 so surface offset 0 is invalid. */
+   cmd_buffer->surface_next = 1;
+
+   cmd_buffer->exec2_objects = NULL;
+   cmd_buffer->exec2_bos = NULL;
+   cmd_buffer->exec2_array_length = 0;
+
+   anv_state_stream_init(&cmd_buffer->surface_state_stream,
+                         &device->surface_state_block_pool);
+   anv_state_stream_init(&cmd_buffer->dynamic_state_stream,
+                         &device->dynamic_state_block_pool);
+
+   anv_cmd_state_init(&cmd_buffer->state);
+
+   *pCmdBuffer = anv_cmd_buffer_to_handle(cmd_buffer);
+
+   return VK_SUCCESS;
+
+ fail_ss_batch_bo:
+   anv_batch_bo_destroy(cmd_buffer->surface_batch_bo, device);
+ fail_batch_relocs:
+   anv_reloc_list_finish(&cmd_buffer->batch.relocs, device);
+ fail_batch_bo:
+   anv_batch_bo_destroy(cmd_buffer->last_batch_bo, device);
+ fail:
+   anv_device_free(device, cmd_buffer);
+
+   return result;
+}
+
+VkResult anv_DestroyCommandBuffer(
+    VkDevice                                    _device,
+    VkCmdBuffer                                 _cmd_buffer)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, _cmd_buffer);
+
+   anv_cmd_state_fini(&cmd_buffer->state);
+
+   /* Destroy all of the batch buffers */
+   struct anv_batch_bo *bbo = cmd_buffer->last_batch_bo;
+   while (bbo) {
+      struct anv_batch_bo *prev = bbo->prev_batch_bo;
+      anv_batch_bo_destroy(bbo, device);
+      bbo = prev;
+   }
+   anv_reloc_list_finish(&cmd_buffer->batch.relocs, device);
+
+   /* Destroy all of the surface state buffers */
+   bbo = cmd_buffer->surface_batch_bo;
+   while (bbo) {
+      struct anv_batch_bo *prev = bbo->prev_batch_bo;
+      anv_batch_bo_destroy(bbo, device);
+      bbo = prev;
+   }
+   anv_reloc_list_finish(&cmd_buffer->surface_relocs, device);
+
+   anv_state_stream_finish(&cmd_buffer->surface_state_stream);
+   anv_state_stream_finish(&cmd_buffer->dynamic_state_stream);
+   anv_device_free(device, cmd_buffer->exec2_objects);
+   anv_device_free(device, cmd_buffer->exec2_bos);
+   anv_device_free(device, cmd_buffer);
+
+   return VK_SUCCESS;
+}
+
+static VkResult
+anv_cmd_buffer_add_bo(struct anv_cmd_buffer *cmd_buffer,
+                      struct anv_bo *bo,
+                      struct drm_i915_gem_relocation_entry *relocs,
+                      size_t num_relocs)
+{
+   struct drm_i915_gem_exec_object2 *obj;
+
+   if (bo->index < cmd_buffer->exec2_bo_count &&
+       cmd_buffer->exec2_bos[bo->index] == bo)
+      return VK_SUCCESS;
+
+   if (cmd_buffer->exec2_bo_count >= cmd_buffer->exec2_array_length) {
+      uint32_t new_len = cmd_buffer->exec2_objects ?
+                         cmd_buffer->exec2_array_length * 2 : 64;
+
+      struct drm_i915_gem_exec_object2 *new_objects =
+         anv_device_alloc(cmd_buffer->device, new_len * sizeof(*new_objects),
+                          8, VK_SYSTEM_ALLOC_TYPE_INTERNAL);
+      if (new_objects == NULL)
+         return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+      struct anv_bo **new_bos =
+         anv_device_alloc(cmd_buffer->device, new_len * sizeof(*new_bos),
+                          8, VK_SYSTEM_ALLOC_TYPE_INTERNAL);
+      if (new_objects == NULL) {
+         anv_device_free(cmd_buffer->device, new_objects);
+         return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+      }
+
+      if (cmd_buffer->exec2_objects) {
+         memcpy(new_objects, cmd_buffer->exec2_objects,
+                cmd_buffer->exec2_bo_count * sizeof(*new_objects));
+         memcpy(new_bos, cmd_buffer->exec2_bos,
+                cmd_buffer->exec2_bo_count * sizeof(*new_bos));
+      }
+
+      cmd_buffer->exec2_objects = new_objects;
+      cmd_buffer->exec2_bos = new_bos;
+      cmd_buffer->exec2_array_length = new_len;
+   }
+
+   assert(cmd_buffer->exec2_bo_count < cmd_buffer->exec2_array_length);
+
+   bo->index = cmd_buffer->exec2_bo_count++;
+   obj = &cmd_buffer->exec2_objects[bo->index];
+   cmd_buffer->exec2_bos[bo->index] = bo;
+
+   obj->handle = bo->gem_handle;
+   obj->relocation_count = 0;
+   obj->relocs_ptr = 0;
+   obj->alignment = 0;
+   obj->offset = bo->offset;
+   obj->flags = 0;
+   obj->rsvd1 = 0;
+   obj->rsvd2 = 0;
+
+   if (relocs) {
+      obj->relocation_count = num_relocs;
+      obj->relocs_ptr = (uintptr_t) relocs;
+   }
+
+   return VK_SUCCESS;
+}
+
+static void
+anv_cmd_buffer_add_validate_bos(struct anv_cmd_buffer *cmd_buffer,
+                                struct anv_reloc_list *list)
+{
+   for (size_t i = 0; i < list->num_relocs; i++)
+      anv_cmd_buffer_add_bo(cmd_buffer, list->reloc_bos[i], NULL, 0);
+}
+
+static void
+anv_cmd_buffer_process_relocs(struct anv_cmd_buffer *cmd_buffer,
+                              struct anv_reloc_list *list)
+{
+   struct anv_bo *bo;
+
+   /* If the kernel supports I915_EXEC_NO_RELOC, it will compare offset in
+    * struct drm_i915_gem_exec_object2 against the bos current offset and if
+    * all bos haven't moved it will skip relocation processing alltogether.
+    * If I915_EXEC_NO_RELOC is not supported, the kernel ignores the incoming
+    * value of offset so we can set it either way.  For that to work we need
+    * to make sure all relocs use the same presumed offset.
+    */
+
+   for (size_t i = 0; i < list->num_relocs; i++) {
+      bo = list->reloc_bos[i];
+      if (bo->offset != list->relocs[i].presumed_offset)
+         cmd_buffer->need_reloc = true;
+
+      list->relocs[i].target_handle = bo->index;
+   }
+}
+
+VkResult anv_EndCommandBuffer(
+    VkCmdBuffer                                 cmdBuffer)
+{
+   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
+   struct anv_device *device = cmd_buffer->device;
+   struct anv_batch *batch = &cmd_buffer->batch;
+
+   anv_batch_emit(batch, GEN8_MI_BATCH_BUFFER_END);
+
+   anv_batch_bo_finish(cmd_buffer->last_batch_bo, &cmd_buffer->batch);
+   cmd_buffer->surface_batch_bo->num_relocs =
+      cmd_buffer->surface_relocs.num_relocs - cmd_buffer->surface_batch_bo->first_reloc;
+   cmd_buffer->surface_batch_bo->length = cmd_buffer->surface_next;
+
+   cmd_buffer->exec2_bo_count = 0;
+   cmd_buffer->need_reloc = false;
+
+   /* Lock for access to bo->index. */
+   pthread_mutex_lock(&device->mutex);
+
+   /* Add surface state bos first so we can add them with their relocs. */
+   for (struct anv_batch_bo *bbo = cmd_buffer->surface_batch_bo;
+        bbo != NULL; bbo = bbo->prev_batch_bo) {
+      anv_cmd_buffer_add_bo(cmd_buffer, &bbo->bo,
+                            &cmd_buffer->surface_relocs.relocs[bbo->first_reloc],
+                            bbo->num_relocs);
+   }
+
+   /* Add all of the BOs referenced by surface state */
+   anv_cmd_buffer_add_validate_bos(cmd_buffer, &cmd_buffer->surface_relocs);
+
+   /* Add all but the first batch BO */
+   struct anv_batch_bo *batch_bo = cmd_buffer->last_batch_bo;
+   while (batch_bo->prev_batch_bo) {
+      anv_cmd_buffer_add_bo(cmd_buffer, &batch_bo->bo,
+                            &batch->relocs.relocs[batch_bo->first_reloc],
+                            batch_bo->num_relocs);
+      batch_bo = batch_bo->prev_batch_bo;
+   }
+
+   /* Add everything referenced by the batches */
+   anv_cmd_buffer_add_validate_bos(cmd_buffer, &batch->relocs);
+
+   /* Add the first batch bo last */
+   assert(batch_bo->prev_batch_bo == NULL && batch_bo->first_reloc == 0);
+   anv_cmd_buffer_add_bo(cmd_buffer, &batch_bo->bo,
+                         &batch->relocs.relocs[batch_bo->first_reloc],
+                         batch_bo->num_relocs);
+   assert(batch_bo->bo.index == cmd_buffer->exec2_bo_count - 1);
+
+   anv_cmd_buffer_process_relocs(cmd_buffer, &cmd_buffer->surface_relocs);
+   anv_cmd_buffer_process_relocs(cmd_buffer, &batch->relocs);
+
+   cmd_buffer->execbuf.buffers_ptr = (uintptr_t) cmd_buffer->exec2_objects;
+   cmd_buffer->execbuf.buffer_count = cmd_buffer->exec2_bo_count;
+   cmd_buffer->execbuf.batch_start_offset = 0;
+   cmd_buffer->execbuf.batch_len = batch->next - batch->start;
+   cmd_buffer->execbuf.cliprects_ptr = 0;
+   cmd_buffer->execbuf.num_cliprects = 0;
+   cmd_buffer->execbuf.DR1 = 0;
+   cmd_buffer->execbuf.DR4 = 0;
+
+   cmd_buffer->execbuf.flags = I915_EXEC_HANDLE_LUT;
+   if (!cmd_buffer->need_reloc)
+      cmd_buffer->execbuf.flags |= I915_EXEC_NO_RELOC;
+   cmd_buffer->execbuf.flags |= I915_EXEC_RENDER;
+   cmd_buffer->execbuf.rsvd1 = device->context_id;
+   cmd_buffer->execbuf.rsvd2 = 0;
+
+   pthread_mutex_unlock(&device->mutex);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_ResetCommandBuffer(
+    VkCmdBuffer                                 cmdBuffer,
+    VkCmdBufferResetFlags                       flags)
+{
+   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
+
+   /* Delete all but the first batch bo */
+   while (cmd_buffer->last_batch_bo->prev_batch_bo) {
+      struct anv_batch_bo *prev = cmd_buffer->last_batch_bo->prev_batch_bo;
+      anv_batch_bo_destroy(cmd_buffer->last_batch_bo, cmd_buffer->device);
+      cmd_buffer->last_batch_bo = prev;
+   }
+   assert(cmd_buffer->last_batch_bo->prev_batch_bo == NULL);
+
+   cmd_buffer->batch.relocs.num_relocs = 0;
+   anv_batch_bo_start(cmd_buffer->last_batch_bo, &cmd_buffer->batch,
+                      GEN8_MI_BATCH_BUFFER_START_length * 4);
+
+   /* Delete all but the first batch bo */
+   while (cmd_buffer->surface_batch_bo->prev_batch_bo) {
+      struct anv_batch_bo *prev = cmd_buffer->surface_batch_bo->prev_batch_bo;
+      anv_batch_bo_destroy(cmd_buffer->surface_batch_bo, cmd_buffer->device);
+      cmd_buffer->surface_batch_bo = prev;
+   }
+   assert(cmd_buffer->surface_batch_bo->prev_batch_bo == NULL);
+
+   cmd_buffer->surface_next = 1;
+   cmd_buffer->surface_relocs.num_relocs = 0;
+
+   anv_cmd_state_fini(&cmd_buffer->state);
+   anv_cmd_state_init(&cmd_buffer->state);
+
+   return VK_SUCCESS;
+}

src/vulkan/aub.c

@@ -0,0 +1,310 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 <stdlib.h>
+#include <stdio.h>
+#include <stdint.h>
+#include <string.h>
+#include <unistd.h>
+#include <assert.h>
+#include <sys/types.h>
+#include <sys/mman.h>
+
+#include <drm.h>
+#include <i915_drm.h>
+
+#include "private.h"
+#include "aub.h"
+
+struct anv_aub_writer {
+   FILE *file;
+   uint32_t offset;
+   int gen;
+};
+
+static void
+aub_out(struct anv_aub_writer *writer, uint32_t data)
+{
+   fwrite(&data, 1, 4, writer->file);
+}
+
+static void
+aub_out_data(struct anv_aub_writer *writer, const void *data, size_t size)
+{
+   fwrite(data, 1, size, writer->file);
+}
+
+static struct anv_aub_writer *
+get_anv_aub_writer(struct anv_device *device)
+{
+   struct anv_aub_writer *writer = device->aub_writer;
+   int entry = 0x200003;
+   int i;
+   int gtt_size = 0x10000;
+   const char *filename;
+
+   if (geteuid() != getuid())
+      return NULL;
+
+   if (writer)
+      return writer;
+
+   writer = malloc(sizeof(*writer));
+   if (writer == NULL)
+      return NULL;
+
+   filename = "intel.aub";
+   writer->gen = device->info.gen;
+   writer->file = fopen(filename, "w+");
+   if (!writer->file) {
+      free(writer);
+      return NULL;
+   }
+
+   /* Start allocating objects from just after the GTT. */
+   writer->offset = gtt_size;
+
+   /* Start with a (required) version packet. */
+   aub_out(writer, CMD_AUB_HEADER | (13 - 2));
+   aub_out(writer,
+           (4 << AUB_HEADER_MAJOR_SHIFT) |
+           (0 << AUB_HEADER_MINOR_SHIFT));
+   for (i = 0; i < 8; i++) {
+      aub_out(writer, 0); /* app name */
+   }
+   aub_out(writer, 0); /* timestamp */
+   aub_out(writer, 0); /* timestamp */
+   aub_out(writer, 0); /* comment len */
+
+   /* Set up the GTT. The max we can handle is 256M */
+   aub_out(writer, CMD_AUB_TRACE_HEADER_BLOCK | ((writer->gen >= 8 ? 6 : 5) - 2));
+   aub_out(writer,
+           AUB_TRACE_MEMTYPE_GTT_ENTRY |
+           AUB_TRACE_TYPE_NOTYPE | AUB_TRACE_OP_DATA_WRITE);
+   aub_out(writer, 0); /* subtype */
+   aub_out(writer, 0); /* offset */
+   aub_out(writer, gtt_size); /* size */
+   if (writer->gen >= 8)
+      aub_out(writer, 0);
+   for (i = 0x000; i < gtt_size; i += 4, entry += 0x1000) {
+      aub_out(writer, entry);
+   }
+
+   return device->aub_writer = writer;
+}
+
+void
+anv_aub_writer_destroy(struct anv_aub_writer *writer)
+{
+   fclose(writer->file);
+   free(writer);
+}
+
+
+/**
+ * Break up large objects into multiple writes.  Otherwise a 128kb VBO
+ * would overflow the 16 bits of size field in the packet header and
+ * everything goes badly after that.
+ */
+static void
+aub_write_trace_block(struct anv_aub_writer *writer, uint32_t type,
+                      void *virtual, uint32_t size, uint32_t gtt_offset)
+{
+   uint32_t block_size;
+   uint32_t offset;
+   uint32_t subtype = 0;
+   static const char null_block[8 * 4096];
+
+   for (offset = 0; offset < size; offset += block_size) {
+      block_size = size - offset;
+
+      if (block_size > 8 * 4096)
+         block_size = 8 * 4096;
+
+      aub_out(writer,
+              CMD_AUB_TRACE_HEADER_BLOCK |
+              ((writer->gen >= 8 ? 6 : 5) - 2));
+      aub_out(writer,
+              AUB_TRACE_MEMTYPE_GTT |
+              type | AUB_TRACE_OP_DATA_WRITE);
+      aub_out(writer, subtype);
+      aub_out(writer, gtt_offset + offset);
+      aub_out(writer, align_u32(block_size, 4));
+      if (writer->gen >= 8)
+         aub_out(writer, 0);
+
+      if (virtual)
+         aub_out_data(writer, (char *) virtual + offset, block_size);
+      else
+         aub_out_data(writer, null_block, block_size);
+
+      /* Pad to a multiple of 4 bytes. */
+      aub_out_data(writer, null_block, -block_size & 3);
+   }
+}
+
+/*
+ * Make a ringbuffer on fly and dump it
+ */
+static void
+aub_build_dump_ringbuffer(struct anv_aub_writer *writer,
+                          uint32_t batch_offset, uint32_t offset,
+                          int ring_flag)
+{
+   uint32_t ringbuffer[4096];
+   int ring = AUB_TRACE_TYPE_RING_PRB0; /* The default ring */
+   int ring_count = 0;
+
+   if (ring_flag == I915_EXEC_BSD)
+      ring = AUB_TRACE_TYPE_RING_PRB1;
+   else if (ring_flag == I915_EXEC_BLT)
+      ring = AUB_TRACE_TYPE_RING_PRB2;
+
+   /* Make a ring buffer to execute our batchbuffer. */
+   memset(ringbuffer, 0, sizeof(ringbuffer));
+   if (writer->gen >= 8) {
+      ringbuffer[ring_count++] = AUB_MI_BATCH_BUFFER_START | (3 - 2);
+      ringbuffer[ring_count++] = batch_offset;
+      ringbuffer[ring_count++] = 0;
+   } else {
+      ringbuffer[ring_count++] = AUB_MI_BATCH_BUFFER_START;
+      ringbuffer[ring_count++] = batch_offset;
+   }
+
+   /* Write out the ring.  This appears to trigger execution of
+    * the ring in the simulator.
+    */
+   aub_out(writer,
+           CMD_AUB_TRACE_HEADER_BLOCK |
+           ((writer->gen >= 8 ? 6 : 5) - 2));
+   aub_out(writer,
+           AUB_TRACE_MEMTYPE_GTT | ring | AUB_TRACE_OP_COMMAND_WRITE);
+   aub_out(writer, 0); /* general/surface subtype */
+   aub_out(writer, offset);
+   aub_out(writer, ring_count * 4);
+   if (writer->gen >= 8)
+      aub_out(writer, 0);
+
+   /* FIXME: Need some flush operations here? */
+   aub_out_data(writer, ringbuffer, ring_count * 4);
+}
+
+struct aub_bo {
+   uint32_t offset;
+   void *map;
+   void *relocated;
+};
+
+static void
+relocate_bo(struct anv_bo *bo, struct drm_i915_gem_relocation_entry *relocs,
+            size_t num_relocs, struct aub_bo *bos)
+{
+   struct aub_bo *aub_bo = &bos[bo->index];
+   struct drm_i915_gem_relocation_entry *reloc;
+   uint32_t *dw;
+
+   aub_bo->relocated = malloc(bo->size);
+   memcpy(aub_bo->relocated, aub_bo->map, bo->size);
+   for (size_t i = 0; i < num_relocs; i++) {
+      reloc = &relocs[i];
+      assert(reloc->offset < bo->size);
+      dw = aub_bo->relocated + reloc->offset;
+      *dw = bos[reloc->target_handle].offset + reloc->delta;
+   }
+}
+
+void
+anv_cmd_buffer_dump(struct anv_cmd_buffer *cmd_buffer)
+{
+   struct anv_device *device = cmd_buffer->device;
+   struct anv_batch *batch = &cmd_buffer->batch;
+   struct anv_aub_writer *writer;
+   struct anv_bo *bo;
+   uint32_t ring_flag = 0;
+   uint32_t offset;
+   struct aub_bo *aub_bos;
+
+   writer = get_anv_aub_writer(device);
+   if (writer == NULL)
+      return;
+
+   aub_bos = malloc(cmd_buffer->exec2_bo_count * sizeof(aub_bos[0]));
+   offset = writer->offset;
+   for (uint32_t i = 0; i < cmd_buffer->exec2_bo_count; i++) {
+      bo = cmd_buffer->exec2_bos[i];
+      if (bo->map)
+         aub_bos[i].map = bo->map;
+      else
+         aub_bos[i].map = anv_gem_mmap(device, bo->gem_handle, 0, bo->size);
+      aub_bos[i].relocated = aub_bos[i].map;
+      aub_bos[i].offset = offset;
+      offset = align_u32(offset + bo->size + 4095, 4096);
+   }
+
+   struct anv_batch_bo *first_bbo;
+   for (struct anv_batch_bo *bbo = cmd_buffer->last_batch_bo;
+        bbo != NULL; bbo = bbo->prev_batch_bo) {
+      /* Keep stashing the current BO until we get to the beginning */
+      first_bbo = bbo;
+
+      /* Handle relocations for this batch BO */
+      relocate_bo(&bbo->bo, &batch->relocs.relocs[bbo->first_reloc],
+                  bbo->num_relocs, aub_bos);
+   }
+   assert(first_bbo->prev_batch_bo == NULL);
+
+   for (struct anv_batch_bo *bbo = cmd_buffer->surface_batch_bo;
+        bbo != NULL; bbo = bbo->prev_batch_bo) {
+
+      /* Handle relocations for this surface state BO */
+      relocate_bo(&bbo->bo,
+                  &cmd_buffer->surface_relocs.relocs[bbo->first_reloc],
+                  bbo->num_relocs, aub_bos);
+   }
+
+   for (uint32_t i = 0; i < cmd_buffer->exec2_bo_count; i++) {
+      bo = cmd_buffer->exec2_bos[i];
+      if (i == cmd_buffer->exec2_bo_count - 1) {
+         assert(bo == &first_bbo->bo);
+         aub_write_trace_block(writer, AUB_TRACE_TYPE_BATCH,
+                               aub_bos[i].relocated,
+                               first_bbo->length, aub_bos[i].offset);
+      } else {
+         aub_write_trace_block(writer, AUB_TRACE_TYPE_NOTYPE,
+                               aub_bos[i].relocated,
+                               bo->size, aub_bos[i].offset);
+      }
+      if (aub_bos[i].relocated != aub_bos[i].map)
+         free(aub_bos[i].relocated);
+      if (aub_bos[i].map != bo->map)
+         anv_gem_munmap(aub_bos[i].map, bo->size);
+   }
+
+   /* Dump ring buffer */
+   aub_build_dump_ringbuffer(writer, aub_bos[first_bbo->bo.index].offset,
+                             offset, ring_flag);
+
+   free(aub_bos);
+
+   fflush(writer->file);
+}

src/vulkan/aub.h

@@ -0,0 +1,153 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ *
+ * Authors:
+ *    Eric Anholt <eric@anholt.net>
+ *
+ */
+
+/** @file intel_aub.h
+ *
+ * The AUB file is a file format used by Intel's internal simulation
+ * and other validation tools.  It can be used at various levels by a
+ * driver to input state to the simulated hardware or a replaying
+ * debugger.
+ *
+ * We choose to dump AUB files using the trace block format for ease
+ * of implementation -- dump out the blocks of memory as plain blobs
+ * and insert ring commands to execute the batchbuffer blob.
+ */
+
+#ifndef _INTEL_AUB_H
+#define _INTEL_AUB_H
+
+#define AUB_MI_NOOP			(0)
+#define AUB_MI_BATCH_BUFFER_START 	(0x31 << 23)
+#define AUB_PIPE_CONTROL		(0x7a000002)
+
+/* DW0: instruction type. */
+
+#define CMD_AUB			(7 << 29)
+
+#define CMD_AUB_HEADER		(CMD_AUB | (1 << 23) | (0x05 << 16))
+/* DW1 */
+# define AUB_HEADER_MAJOR_SHIFT		24
+# define AUB_HEADER_MINOR_SHIFT		16
+
+#define CMD_AUB_TRACE_HEADER_BLOCK (CMD_AUB | (1 << 23) | (0x41 << 16))
+#define CMD_AUB_DUMP_BMP           (CMD_AUB | (1 << 23) | (0x9e << 16))
+
+/* DW1 */
+#define AUB_TRACE_OPERATION_MASK	0x000000ff
+#define AUB_TRACE_OP_COMMENT		0x00000000
+#define AUB_TRACE_OP_DATA_WRITE		0x00000001
+#define AUB_TRACE_OP_COMMAND_WRITE	0x00000002
+#define AUB_TRACE_OP_MMIO_WRITE		0x00000003
+// operation = TRACE_DATA_WRITE, Type
+#define AUB_TRACE_TYPE_MASK		0x0000ff00
+#define AUB_TRACE_TYPE_NOTYPE		(0 << 8)
+#define AUB_TRACE_TYPE_BATCH		(1 << 8)
+#define AUB_TRACE_TYPE_VERTEX_BUFFER	(5 << 8)
+#define AUB_TRACE_TYPE_2D_MAP		(6 << 8)
+#define AUB_TRACE_TYPE_CUBE_MAP		(7 << 8)
+#define AUB_TRACE_TYPE_VOLUME_MAP	(9 << 8)
+#define AUB_TRACE_TYPE_1D_MAP		(10 << 8)
+#define AUB_TRACE_TYPE_CONSTANT_BUFFER	(11 << 8)
+#define AUB_TRACE_TYPE_CONSTANT_URB	(12 << 8)
+#define AUB_TRACE_TYPE_INDEX_BUFFER	(13 << 8)
+#define AUB_TRACE_TYPE_GENERAL		(14 << 8)
+#define AUB_TRACE_TYPE_SURFACE		(15 << 8)
+
+
+// operation = TRACE_COMMAND_WRITE, Type =
+#define AUB_TRACE_TYPE_RING_HWB		(1 << 8)
+#define AUB_TRACE_TYPE_RING_PRB0	(2 << 8)
+#define AUB_TRACE_TYPE_RING_PRB1	(3 << 8)
+#define AUB_TRACE_TYPE_RING_PRB2	(4 << 8)
+
+// Address space
+#define AUB_TRACE_ADDRESS_SPACE_MASK	0x00ff0000
+#define AUB_TRACE_MEMTYPE_GTT		(0 << 16)
+#define AUB_TRACE_MEMTYPE_LOCAL		(1 << 16)
+#define AUB_TRACE_MEMTYPE_NONLOCAL	(2 << 16)
+#define AUB_TRACE_MEMTYPE_PCI		(3 << 16)
+#define AUB_TRACE_MEMTYPE_GTT_ENTRY     (4 << 16)
+
+/* DW2 */
+
+/**
+ * aub_state_struct_type enum values are encoded with the top 16 bits
+ * representing the type to be delivered to the .aub file, and the bottom 16
+ * bits representing the subtype.  This macro performs the encoding.
+ */
+#define ENCODE_SS_TYPE(type, subtype) (((type) << 16) | (subtype))
+
+enum aub_state_struct_type {
+   AUB_TRACE_VS_STATE =			ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 1),
+   AUB_TRACE_GS_STATE =			ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 2),
+   AUB_TRACE_CLIP_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 3),
+   AUB_TRACE_SF_STATE =			ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 4),
+   AUB_TRACE_WM_STATE =			ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 5),
+   AUB_TRACE_CC_STATE =			ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 6),
+   AUB_TRACE_CLIP_VP_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 7),
+   AUB_TRACE_SF_VP_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 8),
+   AUB_TRACE_CC_VP_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0x9),
+   AUB_TRACE_SAMPLER_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0xa),
+   AUB_TRACE_KERNEL_INSTRUCTIONS =	ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0xb),
+   AUB_TRACE_SCRATCH_SPACE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0xc),
+   AUB_TRACE_SAMPLER_DEFAULT_COLOR =	ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0xd),
+
+   AUB_TRACE_SCISSOR_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0x15),
+   AUB_TRACE_BLEND_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0x16),
+   AUB_TRACE_DEPTH_STENCIL_STATE =	ENCODE_SS_TYPE(AUB_TRACE_TYPE_GENERAL, 0x17),
+
+   AUB_TRACE_VERTEX_BUFFER =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_VERTEX_BUFFER, 0),
+   AUB_TRACE_BINDING_TABLE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_SURFACE, 0x100),
+   AUB_TRACE_SURFACE_STATE =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_SURFACE, 0x200),
+   AUB_TRACE_VS_CONSTANTS =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_CONSTANT_BUFFER, 0),
+   AUB_TRACE_WM_CONSTANTS =		ENCODE_SS_TYPE(AUB_TRACE_TYPE_CONSTANT_BUFFER, 1),
+};
+
+#undef ENCODE_SS_TYPE
+
+/**
+ * Decode a aub_state_struct_type value to determine the type that should be
+ * stored in the .aub file.
+ */
+static inline uint32_t AUB_TRACE_TYPE(enum aub_state_struct_type ss_type)
+{
+   return (ss_type & 0xFFFF0000) >> 16;
+}
+
+/**
+ * Decode a state_struct_type value to determine the subtype that should be
+ * stored in the .aub file.
+ */
+static inline uint32_t AUB_TRACE_SUBTYPE(enum aub_state_struct_type ss_type)
+{
+   return ss_type & 0xFFFF;
+}
+
+/* DW3: address */
+/* DW4: len */
+
+#endif /* _INTEL_AUB_H */

src/vulkan/formats.c

@@ -0,0 +1,334 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 "private.h"
+
+#define UNSUPPORTED 0xffff
+
+#define fmt(__vk_fmt, ...) \
+   [__vk_fmt] = { .name = #__vk_fmt, __VA_ARGS__ }
+
+static const struct anv_format anv_formats[] = {
+   fmt(VK_FORMAT_UNDEFINED,               RAW,                    .cpp = 1,   .num_channels = 1),
+   fmt(VK_FORMAT_R4G4_UNORM,              UNSUPPORTED),
+   fmt(VK_FORMAT_R4G4_USCALED,            UNSUPPORTED),
+   fmt(VK_FORMAT_R4G4B4A4_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_R4G4B4A4_USCALED,        UNSUPPORTED),
+   fmt(VK_FORMAT_R5G6B5_UNORM,            UNSUPPORTED),
+   fmt(VK_FORMAT_R5G6B5_USCALED,          UNSUPPORTED),
+   fmt(VK_FORMAT_R5G5B5A1_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_R5G5B5A1_USCALED,        UNSUPPORTED),
+   fmt(VK_FORMAT_R8_UNORM,                R8_UNORM,               .cpp = 1,   .num_channels = 1),
+   fmt(VK_FORMAT_R8_SNORM,                R8_SNORM,               .cpp = 1,   .num_channels = 1,),
+   fmt(VK_FORMAT_R8_USCALED,              R8_USCALED,             .cpp = 1,   .num_channels = 1),
+   fmt(VK_FORMAT_R8_SSCALED,              R8_SSCALED,             .cpp = 1,   .num_channels = 1),
+   fmt(VK_FORMAT_R8_UINT,                 R8_UINT,                .cpp = 1,   .num_channels = 1),
+   fmt(VK_FORMAT_R8_SINT,                 R8_SINT,                .cpp = 1,   .num_channels = 1),
+   fmt(VK_FORMAT_R8_SRGB,                 UNSUPPORTED),
+   fmt(VK_FORMAT_R8G8_UNORM,              R8G8_UNORM,             .cpp = 2,   .num_channels = 2),
+   fmt(VK_FORMAT_R8G8_SNORM,              R8G8_SNORM,             .cpp = 2,   .num_channels = 2),
+   fmt(VK_FORMAT_R8G8_USCALED,            R8G8_USCALED,           .cpp = 2,   .num_channels = 2),
+   fmt(VK_FORMAT_R8G8_SSCALED,            R8G8_SSCALED,           .cpp = 2,   .num_channels = 2),
+   fmt(VK_FORMAT_R8G8_UINT,               R8G8_UINT,              .cpp = 2,   .num_channels = 2),
+   fmt(VK_FORMAT_R8G8_SINT,               R8G8_SINT,              .cpp = 2,   .num_channels = 2),
+   fmt(VK_FORMAT_R8G8_SRGB,               UNSUPPORTED), /* L8A8_UNORM_SRGB */
+   fmt(VK_FORMAT_R8G8B8_UNORM,            R8G8B8X8_UNORM,         .cpp = 3,   .num_channels = 3),
+   fmt(VK_FORMAT_R8G8B8_SNORM,            R8G8B8_SNORM,           .cpp = 3,   .num_channels = 3),
+   fmt(VK_FORMAT_R8G8B8_USCALED,          R8G8B8_USCALED,         .cpp = 3,   .num_channels = 3),
+   fmt(VK_FORMAT_R8G8B8_SSCALED,          R8G8B8_SSCALED,         .cpp = 3,   .num_channels = 3),
+   fmt(VK_FORMAT_R8G8B8_UINT,             R8G8B8_UINT,            .cpp = 3,   .num_channels = 3),
+   fmt(VK_FORMAT_R8G8B8_SINT,             R8G8B8_SINT,            .cpp = 3,   .num_channels = 3),
+   fmt(VK_FORMAT_R8G8B8_SRGB,             UNSUPPORTED), /* B8G8R8A8_UNORM_SRGB */
+   fmt(VK_FORMAT_R8G8B8A8_UNORM,          R8G8B8A8_UNORM,         .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R8G8B8A8_SNORM,          R8G8B8A8_SNORM,         .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R8G8B8A8_USCALED,        R8G8B8A8_USCALED,       .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R8G8B8A8_SSCALED,        R8G8B8A8_SSCALED,       .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R8G8B8A8_UINT,           R8G8B8A8_UINT,          .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R8G8B8A8_SINT,           R8G8B8A8_SINT,          .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R8G8B8A8_SRGB,           R8G8B8A8_UNORM_SRGB,    .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R10G10B10A2_UNORM,       R10G10B10A2_UNORM,      .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R10G10B10A2_SNORM,       R10G10B10A2_SNORM,      .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R10G10B10A2_USCALED,     R10G10B10A2_USCALED,    .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R10G10B10A2_SSCALED,     R10G10B10A2_SSCALED,    .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R10G10B10A2_UINT,        R10G10B10A2_UINT,       .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R10G10B10A2_SINT,        R10G10B10A2_SINT,       .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_R16_UNORM,               R16_UNORM,              .cpp = 2,   .num_channels = 1),
+   fmt(VK_FORMAT_R16_SNORM,               R16_SNORM,              .cpp = 2,   .num_channels = 1),
+   fmt(VK_FORMAT_R16_USCALED,             R16_USCALED,            .cpp = 2,   .num_channels = 1),
+   fmt(VK_FORMAT_R16_SSCALED,             R16_SSCALED,            .cpp = 2,   .num_channels = 1),
+   fmt(VK_FORMAT_R16_UINT,                R16_UINT,               .cpp = 2,   .num_channels = 1),
+   fmt(VK_FORMAT_R16_SINT,                R16_SINT,               .cpp = 2,   .num_channels = 1),
+   fmt(VK_FORMAT_R16_SFLOAT,              R16_FLOAT,              .cpp = 2,   .num_channels = 1),
+   fmt(VK_FORMAT_R16G16_UNORM,            R16G16_UNORM,           .cpp = 4,   .num_channels = 2),
+   fmt(VK_FORMAT_R16G16_SNORM,            R16G16_SNORM,           .cpp = 4,   .num_channels = 2),
+   fmt(VK_FORMAT_R16G16_USCALED,          R16G16_USCALED,         .cpp = 4,   .num_channels = 2),
+   fmt(VK_FORMAT_R16G16_SSCALED,          R16G16_SSCALED,         .cpp = 4,   .num_channels = 2),
+   fmt(VK_FORMAT_R16G16_UINT,             R16G16_UINT,            .cpp = 4,   .num_channels = 2),
+   fmt(VK_FORMAT_R16G16_SINT,             R16G16_SINT,            .cpp = 4,   .num_channels = 2),
+   fmt(VK_FORMAT_R16G16_SFLOAT,           R16G16_FLOAT,           .cpp = 4,   .num_channels = 2),
+   fmt(VK_FORMAT_R16G16B16_UNORM,         R16G16B16_UNORM,        .cpp = 6,   .num_channels = 3),
+   fmt(VK_FORMAT_R16G16B16_SNORM,         R16G16B16_SNORM,        .cpp = 6,   .num_channels = 3),
+   fmt(VK_FORMAT_R16G16B16_USCALED,       R16G16B16_USCALED,      .cpp = 6,   .num_channels = 3),
+   fmt(VK_FORMAT_R16G16B16_SSCALED,       R16G16B16_SSCALED,      .cpp = 6,   .num_channels = 3),
+   fmt(VK_FORMAT_R16G16B16_UINT,          R16G16B16_UINT,         .cpp = 6,   .num_channels = 3),
+   fmt(VK_FORMAT_R16G16B16_SINT,          R16G16B16_SINT,         .cpp = 6,   .num_channels = 3),
+   fmt(VK_FORMAT_R16G16B16_SFLOAT,        R16G16B16_FLOAT,        .cpp = 6,   .num_channels = 3),
+   fmt(VK_FORMAT_R16G16B16A16_UNORM,      R16G16B16A16_UNORM,     .cpp = 8,   .num_channels = 4),
+   fmt(VK_FORMAT_R16G16B16A16_SNORM,      R16G16B16A16_SNORM,     .cpp = 8,   .num_channels = 4),
+   fmt(VK_FORMAT_R16G16B16A16_USCALED,    R16G16B16A16_USCALED,   .cpp = 8,   .num_channels = 4),
+   fmt(VK_FORMAT_R16G16B16A16_SSCALED,    R16G16B16A16_SSCALED,   .cpp = 8,   .num_channels = 4),
+   fmt(VK_FORMAT_R16G16B16A16_UINT,       R16G16B16A16_UINT,      .cpp = 8,   .num_channels = 4),
+   fmt(VK_FORMAT_R16G16B16A16_SINT,       R16G16B16A16_SINT,      .cpp = 8,   .num_channels = 4),
+   fmt(VK_FORMAT_R16G16B16A16_SFLOAT,     R16G16B16A16_FLOAT,     .cpp = 8,   .num_channels = 4),
+   fmt(VK_FORMAT_R32_UINT,                R32_UINT,               .cpp = 4,   .num_channels = 1,),
+   fmt(VK_FORMAT_R32_SINT,                R32_SINT,               .cpp = 4,   .num_channels = 1,),
+   fmt(VK_FORMAT_R32_SFLOAT,              R32_FLOAT,              .cpp = 4,   .num_channels = 1,),
+   fmt(VK_FORMAT_R32G32_UINT,             R32G32_UINT,            .cpp = 8,   .num_channels = 2,),
+   fmt(VK_FORMAT_R32G32_SINT,             R32G32_SINT,            .cpp = 8,   .num_channels = 2,),
+   fmt(VK_FORMAT_R32G32_SFLOAT,           R32G32_FLOAT,           .cpp = 8,   .num_channels = 2,),
+   fmt(VK_FORMAT_R32G32B32_UINT,          R32G32B32_UINT,         .cpp = 12,  .num_channels = 3,),
+   fmt(VK_FORMAT_R32G32B32_SINT,          R32G32B32_SINT,         .cpp = 12,  .num_channels = 3,),
+   fmt(VK_FORMAT_R32G32B32_SFLOAT,        R32G32B32_FLOAT,        .cpp = 12,  .num_channels = 3,),
+   fmt(VK_FORMAT_R32G32B32A32_UINT,       R32G32B32A32_UINT,      .cpp = 16,  .num_channels = 4,),
+   fmt(VK_FORMAT_R32G32B32A32_SINT,       R32G32B32A32_SINT,      .cpp = 16,  .num_channels = 4,),
+   fmt(VK_FORMAT_R32G32B32A32_SFLOAT,     R32G32B32A32_FLOAT,     .cpp = 16,  .num_channels = 4,),
+   fmt(VK_FORMAT_R64_SFLOAT,              R64_FLOAT,              .cpp = 8,   .num_channels = 1),
+   fmt(VK_FORMAT_R64G64_SFLOAT,           R64G64_FLOAT,           .cpp = 16,  .num_channels = 2),
+   fmt(VK_FORMAT_R64G64B64_SFLOAT,        R64G64B64_FLOAT,        .cpp = 24,  .num_channels = 3),
+   fmt(VK_FORMAT_R64G64B64A64_SFLOAT,     R64G64B64A64_FLOAT,     .cpp = 32,  .num_channels = 4),
+   fmt(VK_FORMAT_R11G11B10_UFLOAT,        R11G11B10_FLOAT,        .cpp = 4,   .num_channels = 3),
+   fmt(VK_FORMAT_R9G9B9E5_UFLOAT,         R9G9B9E5_SHAREDEXP,     .cpp = 4,   .num_channels = 3),
+
+   fmt(VK_FORMAT_D16_UNORM,               R16_UNORM,              .cpp = 2,   .num_channels = 1, .depth_format = D16_UNORM),
+   fmt(VK_FORMAT_D24_UNORM,               R24_UNORM_X8_TYPELESS,  .cpp = 4,   .num_channels = 1, .depth_format = D24_UNORM_X8_UINT),
+   fmt(VK_FORMAT_D32_SFLOAT,              R32_FLOAT,              .cpp = 4,   .num_channels = 1, .depth_format = D32_FLOAT),
+   fmt(VK_FORMAT_S8_UINT,                 R8_UINT,                .cpp = 1,   .num_channels = 1,                                       .has_stencil = true),
+   fmt(VK_FORMAT_D16_UNORM_S8_UINT,       R16_UNORM,              .cpp = 2,   .num_channels = 2, .depth_format = D16_UNORM,            .has_stencil = true),
+   fmt(VK_FORMAT_D24_UNORM_S8_UINT,       R24_UNORM_X8_TYPELESS,  .cpp = 4,   .num_channels = 2, .depth_format = D24_UNORM_X8_UINT,    .has_stencil = true),
+   fmt(VK_FORMAT_D32_SFLOAT_S8_UINT,      R32_FLOAT,              .cpp = 4,   .num_channels = 2, .depth_format = D32_FLOAT,            .has_stencil = true),
+
+   fmt(VK_FORMAT_BC1_RGB_UNORM,           UNSUPPORTED),
+   fmt(VK_FORMAT_BC1_RGB_SRGB,            UNSUPPORTED),
+   fmt(VK_FORMAT_BC1_RGBA_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_BC1_RGBA_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_BC2_UNORM,               UNSUPPORTED),
+   fmt(VK_FORMAT_BC2_SRGB,                UNSUPPORTED),
+   fmt(VK_FORMAT_BC3_UNORM,               UNSUPPORTED),
+   fmt(VK_FORMAT_BC3_SRGB,                UNSUPPORTED),
+   fmt(VK_FORMAT_BC4_UNORM,               UNSUPPORTED),
+   fmt(VK_FORMAT_BC4_SNORM,               UNSUPPORTED),
+   fmt(VK_FORMAT_BC5_UNORM,               UNSUPPORTED),
+   fmt(VK_FORMAT_BC5_SNORM,               UNSUPPORTED),
+   fmt(VK_FORMAT_BC6H_UFLOAT,             UNSUPPORTED),
+   fmt(VK_FORMAT_BC6H_SFLOAT,             UNSUPPORTED),
+   fmt(VK_FORMAT_BC7_UNORM,               UNSUPPORTED),
+   fmt(VK_FORMAT_BC7_SRGB,                UNSUPPORTED),
+   fmt(VK_FORMAT_ETC2_R8G8B8_UNORM,       UNSUPPORTED),
+   fmt(VK_FORMAT_ETC2_R8G8B8_SRGB,        UNSUPPORTED),
+   fmt(VK_FORMAT_ETC2_R8G8B8A1_UNORM,     UNSUPPORTED),
+   fmt(VK_FORMAT_ETC2_R8G8B8A1_SRGB,      UNSUPPORTED),
+   fmt(VK_FORMAT_ETC2_R8G8B8A8_UNORM,     UNSUPPORTED),
+   fmt(VK_FORMAT_ETC2_R8G8B8A8_SRGB,      UNSUPPORTED),
+   fmt(VK_FORMAT_EAC_R11_UNORM,           UNSUPPORTED),
+   fmt(VK_FORMAT_EAC_R11_SNORM,           UNSUPPORTED),
+   fmt(VK_FORMAT_EAC_R11G11_UNORM,        UNSUPPORTED),
+   fmt(VK_FORMAT_EAC_R11G11_SNORM,        UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_4x4_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_4x4_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_5x4_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_5x4_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_5x5_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_5x5_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_6x5_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_6x5_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_6x6_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_6x6_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_8x5_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_8x5_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_8x6_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_8x6_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_8x8_UNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_8x8_SRGB,           UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x5_UNORM,         UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x5_SRGB,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x6_UNORM,         UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x6_SRGB,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x8_UNORM,         UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x8_SRGB,          UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x10_UNORM,        UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_10x10_SRGB,         UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_12x10_UNORM,        UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_12x10_SRGB,         UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_12x12_UNORM,        UNSUPPORTED),
+   fmt(VK_FORMAT_ASTC_12x12_SRGB,         UNSUPPORTED),
+   fmt(VK_FORMAT_B4G4R4A4_UNORM,          B4G4R4A4_UNORM,         .cpp = 2,   .num_channels = 4),
+   fmt(VK_FORMAT_B5G5R5A1_UNORM,          B5G5R5A1_UNORM,         .cpp = 2,   .num_channels = 4),
+   fmt(VK_FORMAT_B5G6R5_UNORM,            B5G6R5_UNORM,           .cpp = 2,   .num_channels = 3),
+   fmt(VK_FORMAT_B5G6R5_USCALED,          UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8_UNORM,            UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8_SNORM,            UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8_USCALED,          UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8_SSCALED,          UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8_UINT,             UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8_SINT,             UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8_SRGB,             UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8A8_UNORM,          B8G8R8A8_UNORM,         .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_B8G8R8A8_SNORM,          UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8A8_USCALED,        UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8A8_SSCALED,        UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8A8_UINT,           UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8A8_SINT,           UNSUPPORTED),
+   fmt(VK_FORMAT_B8G8R8A8_SRGB,           B8G8R8A8_UNORM_SRGB,    .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_B10G10R10A2_UNORM,       B10G10R10A2_UNORM,      .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_B10G10R10A2_SNORM,       B10G10R10A2_SNORM,      .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_B10G10R10A2_USCALED,     B10G10R10A2_USCALED,    .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_B10G10R10A2_SSCALED,     B10G10R10A2_SSCALED,    .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_B10G10R10A2_UINT,        B10G10R10A2_UINT,       .cpp = 4,   .num_channels = 4),
+   fmt(VK_FORMAT_B10G10R10A2_SINT,        B10G10R10A2_SINT,       .cpp = 4,   .num_channels = 4)
+};
+
+#undef fmt
+
+const struct anv_format *
+anv_format_for_vk_format(VkFormat format)
+{
+   return &anv_formats[format];
+}
+
+bool
+anv_is_vk_format_depth_or_stencil(VkFormat format)
+{
+   const struct anv_format *format_info =
+      anv_format_for_vk_format(format);
+
+   if (format_info->depth_format != UNSUPPORTED &&
+       format_info->depth_format != 0)
+      return true;
+
+   return format_info->has_stencil;
+}
+
+// Format capabilities
+
+struct surface_format_info {
+   bool exists;
+   int sampling;
+   int filtering;
+   int shadow_compare;
+   int chroma_key;
+   int render_target;
+   int alpha_blend;
+   int input_vb;
+   int streamed_output_vb;
+   int color_processing;
+};
+
+extern const struct surface_format_info surface_formats[];
+
+VkResult anv_validate_GetPhysicalDeviceFormatProperties(
+    VkPhysicalDevice                            physicalDevice,
+    VkFormat                                    _format,
+    VkFormatProperties*                         pFormatProperties)
+{
+   const struct anv_format *format = anv_format_for_vk_format(_format);
+   fprintf(stderr, "vkGetFormatProperties(%s)\n", format->name);
+   return anv_GetPhysicalDeviceFormatProperties(physicalDevice, _format, pFormatProperties);
+}
+
+VkResult anv_GetPhysicalDeviceFormatProperties(
+    VkPhysicalDevice                            physicalDevice,
+    VkFormat                                    _format,
+    VkFormatProperties*                         pFormatProperties)
+{
+   ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
+   const struct surface_format_info *info;
+   int gen;
+
+   const struct anv_format *format = anv_format_for_vk_format(_format);
+   if (format == NULL)
+      return vk_error(VK_ERROR_INVALID_VALUE);
+
+   gen = physical_device->info->gen * 10;
+   if (physical_device->info->is_haswell)
+      gen += 5;
+
+   if (format->surface_format == UNSUPPORTED)
+      goto unsupported;
+
+   info = &surface_formats[format->surface_format];
+   if (!info->exists)
+      goto unsupported;
+
+   uint32_t linear = 0, tiled = 0;
+   if (info->sampling <= gen) {
+      linear |= VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT;
+      tiled |= VK_FORMAT_FEATURE_SAMPLED_IMAGE_BIT;
+   }
+   if (info->render_target <= gen) {
+      linear |= VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT;
+      tiled |= VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BIT;
+   }
+   if (info->alpha_blend <= gen) {
+      linear |= VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT;
+      tiled |= VK_FORMAT_FEATURE_COLOR_ATTACHMENT_BLEND_BIT;
+   }
+   if (info->input_vb <= gen) {
+      linear |= VK_FORMAT_FEATURE_VERTEX_BUFFER_BIT;
+   }
+
+   pFormatProperties->linearTilingFeatures = linear;
+   pFormatProperties->optimalTilingFeatures = tiled;
+
+   return VK_SUCCESS;
+
+ unsupported:
+   pFormatProperties->linearTilingFeatures = 0;
+   pFormatProperties->optimalTilingFeatures = 0;
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_GetPhysicalDeviceImageFormatProperties(
+    VkPhysicalDevice                            physicalDevice,
+    VkFormat                                    format,
+    VkImageType                                 type,
+    VkImageTiling                               tiling,
+    VkImageUsageFlags                           usage,
+    VkImageFormatProperties*                    pImageFormatProperties)
+{
+   /* TODO: We should do something here. Chad? */
+   stub_return(VK_UNSUPPORTED);
+}
+
+VkResult anv_GetPhysicalDeviceSparseImageFormatProperties(
+    VkPhysicalDevice                            physicalDevice,
+    VkFormat                                    format,
+    VkImageType                                 type,
+    uint32_t                                    samples,
+    VkImageUsageFlags                           usage,
+    VkImageTiling                               tiling,
+    uint32_t*                                   pNumProperties,
+    VkSparseImageFormatProperties*              pProperties)
+{
+   stub_return(VK_UNSUPPORTED);
+}

src/vulkan/gem.c

@@ -0,0 +1,279 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ */
+
+#define _DEFAULT_SOURCE
+
+#include <sys/ioctl.h>
+#include <sys/mman.h>
+#include <string.h>
+#include <errno.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+#include "private.h"
+
+#define VG_CLEAR(s) VG(memset(&s, 0, sizeof(s)))
+
+static int
+anv_ioctl(int fd, unsigned long request, void *arg)
+{
+   int ret;
+
+   do {
+      ret = ioctl(fd, request, arg);
+   } while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+
+   return ret;
+}
+
+/**
+ * Wrapper around DRM_IOCTL_I915_GEM_CREATE.
+ *
+ * Return gem handle, or 0 on failure. Gem handles are never 0.
+ */
+uint32_t
+anv_gem_create(struct anv_device *device, size_t size)
+{
+   struct drm_i915_gem_create gem_create;
+   int ret;
+
+   VG_CLEAR(gem_create);
+   gem_create.size = size;
+
+   ret = anv_ioctl(device->fd, DRM_IOCTL_I915_GEM_CREATE, &gem_create);
+   if (ret != 0) {
+      /* FIXME: What do we do if this fails? */
+      return 0;
+   }
+
+   return gem_create.handle;
+}
+
+void
+anv_gem_close(struct anv_device *device, int gem_handle)
+{
+   struct drm_gem_close close;
+
+   VG_CLEAR(close);
+   close.handle = gem_handle;
+   anv_ioctl(device->fd, DRM_IOCTL_GEM_CLOSE, &close);
+}
+
+/**
+ * Wrapper around DRM_IOCTL_I915_GEM_MMAP.
+ */
+void*
+anv_gem_mmap(struct anv_device *device, uint32_t gem_handle,
+             uint64_t offset, uint64_t size)
+{
+   struct drm_i915_gem_mmap gem_mmap;
+   int ret;
+
+   gem_mmap.handle = gem_handle;
+   VG_CLEAR(gem_mmap.pad);
+   gem_mmap.offset = offset;
+   gem_mmap.size = size;
+   VG_CLEAR(gem_mmap.addr_ptr);
+
+#ifdef I915_MMAP_WC
+   gem_mmap.flags = 0;
+#endif
+
+   ret = anv_ioctl(device->fd, DRM_IOCTL_I915_GEM_MMAP, &gem_mmap);
+   if (ret != 0) {
+      /* FIXME: Is NULL the right error return? Cf MAP_INVALID */
+      return NULL;
+   }
+
+   VG(VALGRIND_MALLOCLIKE_BLOCK(gem_mmap.addr_ptr, gem_mmap.size, 0, 1));
+   return (void *)(uintptr_t) gem_mmap.addr_ptr;
+}
+
+/* This is just a wrapper around munmap, but it also notifies valgrind that
+ * this map is no longer valid.  Pair this with anv_gem_mmap().
+ */
+void
+anv_gem_munmap(void *p, uint64_t size)
+{
+   VG(VALGRIND_FREELIKE_BLOCK(p, 0));
+   munmap(p, size);
+}
+
+int
+anv_gem_userptr(struct anv_device *device, void *mem, size_t size)
+{
+   struct drm_i915_gem_userptr userptr;
+   int ret;
+
+   VG_CLEAR(userptr);
+   userptr.user_ptr = (__u64)((unsigned long) mem);
+   userptr.user_size = size;
+   userptr.flags = 0;
+
+   ret = anv_ioctl(device->fd, DRM_IOCTL_I915_GEM_USERPTR, &userptr);
+   if (ret == -1)
+      return 0;
+
+   return userptr.handle;
+}
+
+/**
+ * On error, \a timeout_ns holds the remaining time.
+ */
+int
+anv_gem_wait(struct anv_device *device, int gem_handle, int64_t *timeout_ns)
+{
+   struct drm_i915_gem_wait wait;
+   int ret;
+
+   VG_CLEAR(wait);
+   wait.bo_handle = gem_handle;
+   wait.timeout_ns = *timeout_ns;
+   wait.flags = 0;
+
+   ret = anv_ioctl(device->fd, DRM_IOCTL_I915_GEM_WAIT, &wait);
+   *timeout_ns = wait.timeout_ns;
+
+   return ret;
+}
+
+int
+anv_gem_execbuffer(struct anv_device *device,
+                   struct drm_i915_gem_execbuffer2 *execbuf)
+{
+   return anv_ioctl(device->fd, DRM_IOCTL_I915_GEM_EXECBUFFER2, execbuf);
+}
+
+int
+anv_gem_set_tiling(struct anv_device *device,
+                   int gem_handle, uint32_t stride, uint32_t tiling)
+{
+   struct drm_i915_gem_set_tiling set_tiling;
+   int ret;
+
+   /* set_tiling overwrites the input on the error path, so we have to open
+    * code anv_ioctl.
+    */
+
+   do {
+      VG_CLEAR(set_tiling);
+      set_tiling.handle = gem_handle;
+      set_tiling.tiling_mode = I915_TILING_X;
+      set_tiling.stride = stride;
+
+      ret = ioctl(device->fd, DRM_IOCTL_I915_GEM_SET_TILING, &set_tiling);
+   } while (ret == -1 && (errno == EINTR || errno == EAGAIN));
+
+   return ret;
+}
+
+int
+anv_gem_get_param(int fd, uint32_t param)
+{
+   drm_i915_getparam_t gp;
+   int ret, tmp;
+
+   VG_CLEAR(gp);
+   gp.param = param;
+   gp.value = &tmp;
+   ret = anv_ioctl(fd, DRM_IOCTL_I915_GETPARAM, &gp);
+   if (ret == 0)
+      return tmp;
+
+   return 0;
+}
+
+int
+anv_gem_create_context(struct anv_device *device)
+{
+   struct drm_i915_gem_context_create create;
+   int ret;
+
+   VG_CLEAR(create);
+
+   ret = anv_ioctl(device->fd, DRM_IOCTL_I915_GEM_CONTEXT_CREATE, &create);
+   if (ret == -1)
+      return -1;
+
+   return create.ctx_id;
+}
+
+int
+anv_gem_destroy_context(struct anv_device *device, int context)
+{
+   struct drm_i915_gem_context_destroy destroy;
+
+   VG_CLEAR(destroy);
+   destroy.ctx_id = context;
+
+   return anv_ioctl(device->fd, DRM_IOCTL_I915_GEM_CONTEXT_DESTROY, &destroy);
+}
+
+int
+anv_gem_get_aperture(struct anv_physical_device *physical_dev, uint64_t *size)
+{
+   struct drm_i915_gem_get_aperture aperture;
+   int ret;
+
+   VG_CLEAR(aperture);
+   ret = anv_ioctl(physical_dev->fd, DRM_IOCTL_I915_GEM_GET_APERTURE, &aperture);
+   if (ret == -1)
+      return -1;
+
+   *size = aperture.aper_available_size;
+
+   return 0;
+}
+
+int
+anv_gem_handle_to_fd(struct anv_device *device, int gem_handle)
+{
+   struct drm_prime_handle args;
+   int ret;
+
+   VG_CLEAR(args);
+   args.handle = gem_handle;
+   args.flags = DRM_CLOEXEC;
+
+   ret = anv_ioctl(device->fd, DRM_IOCTL_PRIME_HANDLE_TO_FD, &args);
+   if (ret == -1)
+      return -1;
+
+   return args.fd;
+}
+
+int
+anv_gem_fd_to_handle(struct anv_device *device, int fd)
+{
+   struct drm_prime_handle args;
+   int ret;
+
+   VG_CLEAR(args);
+   args.fd = fd;
+
+   ret = anv_ioctl(device->fd, DRM_IOCTL_PRIME_FD_TO_HANDLE, &args);
+   if (ret == -1)
+      return 0;
+
+   return args.handle;
+}

src/vulkan/glsl_scraper.py

@@ -0,0 +1,274 @@
+#! /usr/bin/env python
+
+import argparse
+import cStringIO
+import os
+import re
+import shutil
+import struct
+import subprocess
+import sys
+import tempfile
+from textwrap import dedent
+
+class Shader:
+   def __init__(self, stage):
+      self.stream = cStringIO.StringIO()
+      self.stage = stage
+
+      if self.stage == 'VERTEX':
+         self.ext = 'vert'
+      elif self.stage == 'TESS_CONTROL':
+         self.ext = 'tesc'
+      elif self.stage == 'TESS_EVALUATION':
+         self.ext = 'tese'
+      elif self.stage == 'GEOMETRY':
+         self.ext = 'geom'
+      elif self.stage == 'FRAGMENT':
+         self.ext = 'frag'
+      elif self.stage == 'COMPUTE':
+         self.ext = 'comp'
+      else:
+         assert False
+
+   def add_text(self, s):
+      self.stream.write(s)
+
+   def finish_text(self, line):
+      self.line = line
+
+   def glsl_source(self):
+      return self.stream.getvalue()
+
+   def compile(self):
+      # We can assume if we got here that we have a temp directory and that
+      # we're currently living in it.
+      glsl_fname = 'shader{0}.{1}'.format(self.line, self.ext)
+      spirv_fname = self.ext + '.spv'
+
+      glsl_file = open(glsl_fname, 'w')
+      glsl_file.write('#version 420 core\n')
+      glsl_file.write(self.glsl_source())
+      glsl_file.close()
+
+      out = open('glslang.out', 'wb')
+      err = subprocess.call([glslang, '-V', glsl_fname], stdout=out)
+      if err != 0:
+         out = open('glslang.out', 'r')
+         sys.stderr.write(out.read())
+         out.close()
+         exit(1)
+
+      def dwords(f):
+         while True:
+            dword_str = f.read(4)
+            if not dword_str:
+               return
+            assert len(dword_str) == 4
+            yield struct.unpack('I', dword_str)[0]
+
+      spirv_file = open(spirv_fname, 'rb')
+      self.dwords = list(dwords(spirv_file))
+      spirv_file.close()
+
+      os.remove(glsl_fname)
+      os.remove(spirv_fname)
+
+   def dump_c_code(self, f, glsl_only = False):
+      f.write('\n\n')
+      var_prefix = '_glsl_helpers_shader{0}'.format(self.line)
+
+      # First dump the GLSL source as strings
+      f.write('static const char {0}_glsl_src[] ='.format(var_prefix))
+      f.write('\n_ANV_SPIRV_' + self.stage)
+      f.write('\n"#version 330\\n"')
+      for line in self.glsl_source().splitlines():
+         if not line.strip():
+            continue
+         f.write('\n"{0}\\n"'.format(line))
+      f.write(';\n\n')
+
+      if glsl_only:
+         return
+
+      # Now dump the SPIR-V source
+      f.write('static const uint32_t {0}_spir_v_src[] = {{'.format(var_prefix))
+      line_start = 0
+      while line_start < len(self.dwords):
+         f.write('\n   ')
+         for i in range(line_start, min(line_start + 6, len(self.dwords))):
+            f.write(' 0x{:08x},'.format(self.dwords[i]))
+         line_start += 6
+      f.write('\n};\n')
+
+token_exp = re.compile(r'(GLSL_VK_SHADER_MODULE|\(|\)|,)')
+
+class Parser:
+   def __init__(self, f):
+      self.infile = f
+      self.paren_depth = 0
+      self.shader = None
+      self.line_number = 1
+      self.shaders = []
+
+      def tokenize(f):
+         leftover = ''
+         for line in f:
+            pos = 0
+            while True:
+               m = token_exp.search(line, pos)
+               if m:
+                  if m.start() > pos:
+                     leftover += line[pos:m.start()]
+                  pos = m.end()
+
+                  if leftover:
+                     yield leftover
+                     leftover = ''
+
+                  yield m.group(0)
+
+               else:
+                  leftover += line[pos:]
+                  break
+
+            self.line_number += 1
+
+         if leftover:
+            yield leftover
+
+      self.token_iter = tokenize(self.infile)
+
+   def handle_shader_src(self):
+      paren_depth = 1
+      for t in self.token_iter:
+         if t == '(':
+            paren_depth += 1
+         elif t == ')':
+            paren_depth -= 1
+            if paren_depth == 0:
+               return
+
+         self.current_shader.add_text(t)
+
+   def handle_macro(self):
+      t = self.token_iter.next()
+      assert t == '('
+      t = self.token_iter.next()
+      t = self.token_iter.next()
+      assert t == ','
+
+      stage = self.token_iter.next().strip()
+
+      t = self.token_iter.next()
+      assert t == ','
+
+      self.current_shader = Shader(stage)
+      self.handle_shader_src()
+      self.current_shader.finish_text(self.line_number)
+
+      self.shaders.append(self.current_shader)
+      self.current_shader = None
+
+   def run(self):
+      for t in self.token_iter:
+         if t == 'GLSL_VK_SHADER_MODULE':
+            self.handle_macro()
+
+def open_file(name, mode):
+   if name == '-':
+      if mode == 'w':
+         return sys.stdout
+      elif mode == 'r':
+         return sys.stdin
+      else:
+         assert False
+   else:
+      return open(name, mode)
+
+def parse_args():
+   description = dedent("""\
+      This program scrapes a C file for any instance of the
+      GLSL_VK_SHADER_MODULE macro, grabs the GLSL source code, compiles it
+      to SPIR-V.  The resulting SPIR-V code is written to another C file as
+      an array of 32-bit words.
+
+      If '-' is passed as the input file or output file, stdin or stdout will be
+      used instead of a file on disc.""")
+
+   p = argparse.ArgumentParser(
+         description=description,
+         formatter_class=argparse.RawDescriptionHelpFormatter)
+   p.add_argument('-o', '--outfile', default='-',
+                  help='Output to the given file (default: stdout).')
+   p.add_argument('--with-glslang', metavar='PATH',
+                  default='glslangValidator',
+                  dest='glslang',
+                  help='Full path to the glslangValidator program.')
+   p.add_argument('--glsl-only', action='store_true')
+   p.add_argument('infile', metavar='INFILE')
+
+   return p.parse_args()
+
+
+args = parse_args()
+infname = args.infile
+outfname = args.outfile
+glslang = args.glslang
+glsl_only = args.glsl_only
+
+with open_file(infname, 'r') as infile:
+   parser = Parser(infile)
+   parser.run()
+
+if not glsl_only:
+   # glslangValidator has an absolutely *insane* interface.  We pretty much
+   # have to run in a temporary directory.  Sad day.
+   current_dir = os.getcwd()
+   tmpdir = tempfile.mkdtemp('glsl_scraper')
+
+   try:
+      os.chdir(tmpdir)
+
+      for shader in parser.shaders:
+         shader.compile()
+
+      os.chdir(current_dir)
+   finally:
+      shutil.rmtree(tmpdir)
+
+with open_file(outfname, 'w') as outfile:
+   outfile.write(dedent("""\
+      /* ===========================  DO NOT EDIT!  ===========================
+       *            This file is autogenerated by glsl_scraper.py.
+       */
+
+      #include <stdint.h>
+
+      #define _ANV_SPIRV_MAGIC "\\x03\\x02\\x23\\x07\\0\\0\\0\\0"
+
+      #define _ANV_SPIRV_VERTEX           _ANV_SPIRV_MAGIC "\\0\\0\\0\\0"
+      #define _ANV_SPIRV_TESS_CONTROL     _ANV_SPIRV_MAGIC "\\1\\0\\0\\0"
+      #define _ANV_SPIRV_TESS_EVALUATION  _ANV_SPIRV_MAGIC "\\2\\0\\0\\0"
+      #define _ANV_SPIRV_GEOMETRY         _ANV_SPIRV_MAGIC "\\3\\0\\0\\0"
+      #define _ANV_SPIRV_FRAGMENT         _ANV_SPIRV_MAGIC "\\4\\0\\0\\0"
+      #define _ANV_SPIRV_COMPUTE          _ANV_SPIRV_MAGIC "\\5\\0\\0\\0"
+
+      #define _ANV_GLSL_SRC_VAR2(_line) _glsl_helpers_shader ## _line ## _glsl_src
+      #define _ANV_GLSL_SRC_VAR(_line) _ANV_GLSL_SRC_VAR2(_line)
+
+      #define GLSL_VK_SHADER_MODULE(device, stage, ...) ({                    \\
+         VkShaderModule __module;                                             \\
+         VkShaderModuleCreateInfo __shader_create_info = {                    \\
+            .sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,             \\
+            .codeSize = sizeof(_ANV_GLSL_SRC_VAR(__LINE__)),                  \\
+            .pCode = _ANV_GLSL_SRC_VAR(__LINE__),                             \\
+         };                                                                   \\
+         vkCreateShaderModule(anv_device_to_handle(device),                   \\
+                              &__shader_create_info, &__module);              \\
+         __module;                                                            \\
+      })
+      """))
+
+   for shader in parser.shaders:
+      shader.dump_c_code(outfile, glsl_only)

src/vulkan/image.c

@@ -0,0 +1,743 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 <assert.h>
+#include <stdbool.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+#include "private.h"
+
+struct anv_image_view_info {
+   uint8_t surface_type; /**< RENDER_SURFACE_STATE.SurfaceType */
+   bool is_array:1; /**< RENDER_SURFACE_STATE.SurfaceArray */
+   bool is_cube:1; /**< RENDER_SURFACE_STATE.CubeFaceEnable* */
+};
+
+static const uint8_t anv_halign[] = {
+    [4] = HALIGN4,
+    [8] = HALIGN8,
+    [16] = HALIGN16,
+};
+
+static const uint8_t anv_valign[] = {
+    [4] = VALIGN4,
+    [8] = VALIGN8,
+    [16] = VALIGN16,
+};
+
+static const uint8_t anv_surf_type_from_image_type[] = {
+   [VK_IMAGE_TYPE_1D] = SURFTYPE_1D,
+   [VK_IMAGE_TYPE_2D] = SURFTYPE_2D,
+   [VK_IMAGE_TYPE_3D] = SURFTYPE_3D,
+
+};
+
+static const struct anv_image_view_info
+anv_image_view_info_table[] = {
+   #define INFO(s, ...) { .surface_type = s, __VA_ARGS__ }
+   [VK_IMAGE_VIEW_TYPE_1D]          = INFO(SURFTYPE_1D),
+   [VK_IMAGE_VIEW_TYPE_2D]          = INFO(SURFTYPE_2D),
+   [VK_IMAGE_VIEW_TYPE_3D]          = INFO(SURFTYPE_3D),
+   [VK_IMAGE_VIEW_TYPE_CUBE]        = INFO(SURFTYPE_CUBE,                  .is_cube = 1),
+   [VK_IMAGE_VIEW_TYPE_1D_ARRAY]    = INFO(SURFTYPE_1D,     .is_array = 1),
+   [VK_IMAGE_VIEW_TYPE_2D_ARRAY]    = INFO(SURFTYPE_2D,     .is_array = 1),
+   [VK_IMAGE_VIEW_TYPE_CUBE_ARRAY]  = INFO(SURFTYPE_CUBE,   .is_array = 1, .is_cube = 1),
+   #undef INFO
+};
+
+static const struct anv_surf_type_limits {
+   int32_t width;
+   int32_t height;
+   int32_t depth;
+} anv_surf_type_limits[] = {
+   [SURFTYPE_1D]     = {16384,       0,   2048},
+   [SURFTYPE_2D]     = {16384,   16384,   2048},
+   [SURFTYPE_3D]     = {2048,     2048,   2048},
+   [SURFTYPE_CUBE]   = {16384,   16384,    340},
+   [SURFTYPE_BUFFER] = {128,     16384,     64},
+   [SURFTYPE_STRBUF] = {128,     16384,     64},
+};
+
+static const struct anv_tile_info {
+   uint32_t width;
+   uint32_t height;
+
+   /**
+    * Alignment for RENDER_SURFACE_STATE.SurfaceBaseAddress.
+    *
+    * To simplify calculations, the alignments defined in the table are
+    * sometimes larger than required.  For example, Skylake requires that X and
+    * Y tiled buffers be aligned to 4K, but Broadwell permits smaller
+    * alignment. We choose 4K to accomodate both chipsets.  The alignment of
+    * a linear buffer depends on its element type and usage. Linear depth
+    * buffers have the largest alignment, 64B, so we choose that for all linear
+    * buffers.
+    */
+   uint32_t surface_alignment;
+} anv_tile_info_table[] = {
+   [LINEAR] = {   1,  1,   64 },
+   [XMAJOR] = { 512,  8, 4096 },
+   [YMAJOR] = { 128, 32, 4096 },
+   [WMAJOR] = { 128, 32, 4096 },
+};
+
+static uint32_t
+anv_image_choose_tile_mode(const struct anv_image_create_info *anv_info)
+{
+   if (anv_info->force_tile_mode)
+      return anv_info->tile_mode;
+
+   if (anv_info->vk_info->format == VK_FORMAT_S8_UINT)
+      return WMAJOR;
+
+   switch (anv_info->vk_info->tiling) {
+   case VK_IMAGE_TILING_LINEAR:
+      return LINEAR;
+   case VK_IMAGE_TILING_OPTIMAL:
+      return YMAJOR;
+   default:
+      assert(!"bad VKImageTiling");
+      return LINEAR;
+   }
+}
+
+static VkResult
+anv_image_make_surface(const struct anv_image_create_info *create_info,
+                       uint64_t *inout_image_size,
+                       uint32_t *inout_image_alignment,
+                       struct anv_surface *out_surface)
+{
+   /* See RENDER_SURFACE_STATE.SurfaceQPitch */
+   static const uint16_t min_qpitch UNUSED = 0x4;
+   static const uint16_t max_qpitch UNUSED = 0x1ffc;
+
+   const VkExtent3D *restrict extent = &create_info->vk_info->extent;
+   const uint32_t levels = create_info->vk_info->mipLevels;
+   const uint32_t array_size = create_info->vk_info->arraySize;
+
+   const uint8_t tile_mode = anv_image_choose_tile_mode(create_info);
+
+   const struct anv_tile_info *tile_info =
+       &anv_tile_info_table[tile_mode];
+
+   const struct anv_format *format_info =
+      anv_format_for_vk_format(create_info->vk_info->format);
+
+   const uint32_t i = 4; /* FINISHME: Stop hardcoding subimage alignment */
+   const uint32_t j = 4; /* FINISHME: Stop hardcoding subimage alignment */
+   const uint32_t w0 = align_u32(extent->width, i);
+   const uint32_t h0 = align_u32(extent->height, j);
+
+   uint16_t qpitch;
+   uint32_t mt_width;
+   uint32_t mt_height;
+
+   if (levels == 1 && array_size == 1) {
+      qpitch = min_qpitch;
+      mt_width = w0;
+      mt_height = h0;
+   } else {
+      uint32_t w1 = align_u32(anv_minify(extent->width, 1), i);
+      uint32_t h1 = align_u32(anv_minify(extent->height, 1), j);
+      uint32_t w2 = align_u32(anv_minify(extent->width, 2), i);
+
+      qpitch = h0 + h1 + 11 * j;
+      mt_width = MAX(w0, w1 + w2);
+      mt_height = array_size * qpitch;
+   }
+
+   assert(qpitch >= min_qpitch);
+   if (qpitch > max_qpitch) {
+      anv_loge("image qpitch > 0x%x\n", max_qpitch);
+      return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
+   }
+
+   /* From the Broadwell PRM, RENDER_SURFACE_STATE.SurfaceQpitch:
+    *
+    *   This field must be set an integer multiple of the Surface Vertical
+    *   Alignment.
+    */
+   assert(anv_is_aligned(qpitch, j));
+
+   const uint32_t stride = align_u32(mt_width * format_info->cpp,
+                                     tile_info->width);
+   const uint32_t size = stride * align_u32(mt_height, tile_info->height);
+   const uint32_t offset = align_u32(*inout_image_size,
+                                     tile_info->surface_alignment);
+
+   *inout_image_size = offset + size;
+   *inout_image_alignment = MAX(*inout_image_alignment,
+                                tile_info->surface_alignment);
+
+   *out_surface = (struct anv_surface) {
+      .offset = offset,
+      .stride = stride,
+      .tile_mode = tile_mode,
+      .qpitch = qpitch,
+      .h_align = i,
+      .v_align = j,
+   };
+
+   return VK_SUCCESS;
+}
+
+VkResult
+anv_image_create(VkDevice _device,
+                 const struct anv_image_create_info *create_info,
+                 VkImage *pImage)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   const VkImageCreateInfo *pCreateInfo = create_info->vk_info;
+   const VkExtent3D *restrict extent = &pCreateInfo->extent;
+   struct anv_image *image = NULL;
+   VkResult r;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO);
+
+   /* XXX: We don't handle any of these */
+   anv_assert(pCreateInfo->imageType == VK_IMAGE_TYPE_2D);
+   anv_assert(pCreateInfo->mipLevels > 0);
+   anv_assert(pCreateInfo->arraySize > 0);
+   anv_assert(pCreateInfo->samples == 1);
+   anv_assert(pCreateInfo->extent.width > 0);
+   anv_assert(pCreateInfo->extent.height > 0);
+   anv_assert(pCreateInfo->extent.depth > 0);
+
+   /* TODO(chadv): How should we validate inputs? */
+   const uint8_t surf_type =
+      anv_surf_type_from_image_type[pCreateInfo->imageType];
+
+   const struct anv_surf_type_limits *limits =
+      &anv_surf_type_limits[surf_type];
+
+   if (extent->width > limits->width ||
+       extent->height > limits->height ||
+       extent->depth > limits->depth) {
+      /* TODO(chadv): What is the correct error? */
+      anv_loge("image extent is too large");
+      return vk_error(VK_ERROR_INVALID_MEMORY_SIZE);
+   }
+
+   const struct anv_format *format_info =
+      anv_format_for_vk_format(pCreateInfo->format);
+
+   image = anv_device_alloc(device, sizeof(*image), 8,
+                            VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (!image)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   memset(image, 0, sizeof(*image));
+   image->type = pCreateInfo->imageType;
+   image->extent = pCreateInfo->extent;
+   image->format = pCreateInfo->format;
+   image->levels = pCreateInfo->mipLevels;
+   image->array_size = pCreateInfo->arraySize;
+   image->surf_type = surf_type;
+
+   if (likely(!format_info->has_stencil || format_info->depth_format)) {
+      /* The image's primary surface is a color or depth surface. */
+      r = anv_image_make_surface(create_info, &image->size, &image->alignment,
+                                 &image->primary_surface);
+      if (r != VK_SUCCESS)
+         goto fail;
+   }
+
+   if (format_info->has_stencil) {
+      /* From the GPU's perspective, the depth buffer and stencil buffer are
+       * separate buffers.  From Vulkan's perspective, though, depth and
+       * stencil reside in the same image.  To satisfy Vulkan and the GPU, we
+       * place the depth and stencil buffers in the same bo.
+       */
+      VkImageCreateInfo stencil_info = *pCreateInfo;
+      stencil_info.format = VK_FORMAT_S8_UINT;
+
+      r = anv_image_make_surface(
+            &(struct anv_image_create_info) {
+               .vk_info = &stencil_info,
+            },
+            &image->size, &image->alignment, &image->stencil_surface);
+
+      if (r != VK_SUCCESS)
+         goto fail;
+   }
+
+   *pImage = anv_image_to_handle(image);
+
+   return VK_SUCCESS;
+
+fail:
+   if (image)
+      anv_device_free(device, image);
+
+   return r;
+}
+
+VkResult
+anv_CreateImage(VkDevice device,
+                const VkImageCreateInfo *pCreateInfo,
+                VkImage *pImage)
+{
+   return anv_image_create(device,
+      &(struct anv_image_create_info) {
+         .vk_info = pCreateInfo,
+      },
+      pImage);
+}
+
+VkResult
+anv_DestroyImage(VkDevice _device, VkImage _image)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+
+   anv_device_free(device, anv_image_from_handle(_image));
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_GetImageSubresourceLayout(
+    VkDevice                                    device,
+    VkImage                                     image,
+    const VkImageSubresource*                   pSubresource,
+    VkSubresourceLayout*                        pLayout)
+{
+   stub_return(VK_UNSUPPORTED);
+}
+
+void
+anv_surface_view_fini(struct anv_device *device,
+                      struct anv_surface_view *view)
+{
+   anv_state_pool_free(&device->surface_state_pool, view->surface_state);
+}
+
+void
+anv_image_view_init(struct anv_image_view *iview,
+                    struct anv_device *device,
+                    const VkImageViewCreateInfo* pCreateInfo,
+                    struct anv_cmd_buffer *cmd_buffer)
+{
+   ANV_FROM_HANDLE(anv_image, image, pCreateInfo->image);
+
+   const VkImageSubresourceRange *range = &pCreateInfo->subresourceRange;
+   struct anv_surface_view *view = &iview->view;
+   struct anv_surface *surface;
+
+   const struct anv_format *format_info =
+      anv_format_for_vk_format(pCreateInfo->format);
+
+   const struct anv_image_view_info *view_type_info
+      = &anv_image_view_info_table[pCreateInfo->viewType];
+
+   if (pCreateInfo->viewType != VK_IMAGE_VIEW_TYPE_2D)
+      anv_finishme("non-2D image views");
+
+   switch (pCreateInfo->subresourceRange.aspect) {
+   case VK_IMAGE_ASPECT_STENCIL:
+      anv_finishme("stencil image views");
+      abort();
+      break;
+   case VK_IMAGE_ASPECT_DEPTH:
+   case VK_IMAGE_ASPECT_COLOR:
+      view->offset = image->offset;
+      surface = &image->primary_surface;
+      break;
+   default:
+      unreachable("");
+      break;
+   }
+
+   view->bo = image->bo;
+   view->offset = image->offset + surface->offset;
+   view->format = pCreateInfo->format;
+
+   view->extent = (VkExtent3D) {
+      .width = anv_minify(image->extent.width, range->baseMipLevel),
+      .height = anv_minify(image->extent.height, range->baseMipLevel),
+      .depth = anv_minify(image->extent.depth, range->baseMipLevel),
+   };
+
+   uint32_t depth = 1;
+   if (range->arraySize > 1) {
+      depth = range->arraySize;
+   } else if (image->extent.depth > 1) {
+      depth = image->extent.depth;
+   }
+
+   static const uint32_t vk_to_gen_swizzle[] = {
+      [VK_CHANNEL_SWIZZLE_ZERO]                 = SCS_ZERO,
+      [VK_CHANNEL_SWIZZLE_ONE]                  = SCS_ONE,
+      [VK_CHANNEL_SWIZZLE_R]                    = SCS_RED,
+      [VK_CHANNEL_SWIZZLE_G]                    = SCS_GREEN,
+      [VK_CHANNEL_SWIZZLE_B]                    = SCS_BLUE,
+      [VK_CHANNEL_SWIZZLE_A]                    = SCS_ALPHA
+   };
+
+   struct GEN8_RENDER_SURFACE_STATE surface_state = {
+      .SurfaceType = view_type_info->surface_type,
+      .SurfaceArray = image->array_size > 1,
+      .SurfaceFormat = format_info->surface_format,
+      .SurfaceVerticalAlignment = anv_valign[surface->v_align],
+      .SurfaceHorizontalAlignment = anv_halign[surface->h_align],
+      .TileMode = surface->tile_mode,
+      .VerticalLineStride = 0,
+      .VerticalLineStrideOffset = 0,
+      .SamplerL2BypassModeDisable = true,
+      .RenderCacheReadWriteMode = WriteOnlyCache,
+      .MemoryObjectControlState = GEN8_MOCS,
+
+      /* The driver sets BaseMipLevel in SAMPLER_STATE, not here in
+       * RENDER_SURFACE_STATE. The Broadwell PRM says "it is illegal to have
+       * both Base Mip Level fields nonzero".
+       */
+      .BaseMipLevel = 0.0,
+
+      .SurfaceQPitch = surface->qpitch >> 2,
+      .Height = image->extent.height - 1,
+      .Width = image->extent.width - 1,
+      .Depth = depth - 1,
+      .SurfacePitch = surface->stride - 1,
+      .MinimumArrayElement = range->baseArraySlice,
+      .NumberofMultisamples = MULTISAMPLECOUNT_1,
+      .XOffset = 0,
+      .YOffset = 0,
+
+      /* For sampler surfaces, the hardware interprets field MIPCount/LOD as
+       * MIPCount.  The range of levels accessible by the sampler engine is
+       * [SurfaceMinLOD, SurfaceMinLOD + MIPCountLOD].
+       */
+      .MIPCountLOD = range->mipLevels - 1,
+      .SurfaceMinLOD = range->baseMipLevel,
+
+      .AuxiliarySurfaceMode = AUX_NONE,
+      .RedClearColor = 0,
+      .GreenClearColor = 0,
+      .BlueClearColor = 0,
+      .AlphaClearColor = 0,
+      .ShaderChannelSelectRed = vk_to_gen_swizzle[pCreateInfo->channels.r],
+      .ShaderChannelSelectGreen = vk_to_gen_swizzle[pCreateInfo->channels.g],
+      .ShaderChannelSelectBlue = vk_to_gen_swizzle[pCreateInfo->channels.b],
+      .ShaderChannelSelectAlpha = vk_to_gen_swizzle[pCreateInfo->channels.a],
+      .ResourceMinLOD = 0.0,
+      .SurfaceBaseAddress = { NULL, view->offset },
+   };
+
+   if (cmd_buffer)
+      view->surface_state =
+         anv_state_stream_alloc(&cmd_buffer->surface_state_stream, 64, 64);
+   else
+      view->surface_state =
+         anv_state_pool_alloc(&device->surface_state_pool, 64, 64);
+
+   GEN8_RENDER_SURFACE_STATE_pack(NULL, view->surface_state.map, &surface_state);
+}
+
+VkResult
+anv_validate_CreateImageView(VkDevice _device,
+                             const VkImageViewCreateInfo *pCreateInfo,
+                             VkImageView *pView)
+{
+   ANV_FROM_HANDLE(anv_image, image, pCreateInfo->image);
+   const VkImageSubresourceRange *subresource;
+   const struct anv_image_view_info *view_info;
+   const struct anv_format *view_format_info;
+   const struct anv_format *image_format_info;
+
+   /* Validate structure type before dereferencing it. */
+   assert(pCreateInfo);
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO);
+   subresource = &pCreateInfo->subresourceRange;
+
+   /* Validate viewType is in range before using it. */
+   assert(pCreateInfo->viewType >= VK_IMAGE_VIEW_TYPE_BEGIN_RANGE);
+   assert(pCreateInfo->viewType <= VK_IMAGE_VIEW_TYPE_END_RANGE);
+   view_info = &anv_image_view_info_table[pCreateInfo->viewType];
+
+   /* Validate format is in range before using it. */
+   assert(pCreateInfo->format >= VK_FORMAT_BEGIN_RANGE);
+   assert(pCreateInfo->format <= VK_FORMAT_END_RANGE);
+   image_format_info = anv_format_for_vk_format(image->format);
+   view_format_info = anv_format_for_vk_format(pCreateInfo->format);
+
+   /* Validate channel swizzles. */
+   assert(pCreateInfo->channels.r >= VK_CHANNEL_SWIZZLE_BEGIN_RANGE);
+   assert(pCreateInfo->channels.r <= VK_CHANNEL_SWIZZLE_END_RANGE);
+   assert(pCreateInfo->channels.g >= VK_CHANNEL_SWIZZLE_BEGIN_RANGE);
+   assert(pCreateInfo->channels.g <= VK_CHANNEL_SWIZZLE_END_RANGE);
+   assert(pCreateInfo->channels.b >= VK_CHANNEL_SWIZZLE_BEGIN_RANGE);
+   assert(pCreateInfo->channels.b <= VK_CHANNEL_SWIZZLE_END_RANGE);
+   assert(pCreateInfo->channels.a >= VK_CHANNEL_SWIZZLE_BEGIN_RANGE);
+   assert(pCreateInfo->channels.a <= VK_CHANNEL_SWIZZLE_END_RANGE);
+
+   /* Validate subresource. */
+   assert(subresource->aspect >= VK_IMAGE_ASPECT_BEGIN_RANGE);
+   assert(subresource->aspect <= VK_IMAGE_ASPECT_END_RANGE);
+   assert(subresource->mipLevels > 0);
+   assert(subresource->arraySize > 0);
+   assert(subresource->baseMipLevel < image->levels);
+   assert(subresource->baseMipLevel + subresource->mipLevels <= image->levels);
+   assert(subresource->baseArraySlice < image->array_size);
+   assert(subresource->baseArraySlice + subresource->arraySize <= image->array_size);
+   assert(pView);
+
+   if (view_info->is_cube) {
+      assert(subresource->baseArraySlice % 6 == 0);
+      assert(subresource->arraySize % 6 == 0);
+   }
+
+   /* Validate format. */
+   switch (subresource->aspect) {
+   case VK_IMAGE_ASPECT_COLOR:
+      assert(!image_format_info->depth_format);
+      assert(!image_format_info->has_stencil);
+      assert(!view_format_info->depth_format);
+      assert(!view_format_info->has_stencil);
+      assert(view_format_info->cpp == image_format_info->cpp);
+      break;
+   case VK_IMAGE_ASPECT_DEPTH:
+      assert(image_format_info->depth_format);
+      assert(view_format_info->depth_format);
+      assert(view_format_info->cpp == image_format_info->cpp);
+      break;
+   case VK_IMAGE_ASPECT_STENCIL:
+      /* FINISHME: Is it legal to have an R8 view of S8? */
+      assert(image_format_info->has_stencil);
+      assert(view_format_info->has_stencil);
+      break;
+   default:
+      assert(!"bad VkImageAspect");
+      break;
+   }
+
+   return anv_CreateImageView(_device, pCreateInfo, pView);
+}
+
+VkResult
+anv_CreateImageView(VkDevice _device,
+                    const VkImageViewCreateInfo *pCreateInfo,
+                    VkImageView *pView)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   struct anv_image_view *view;
+
+   view = anv_device_alloc(device, sizeof(*view), 8,
+                           VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (view == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   anv_image_view_init(view, device, pCreateInfo, NULL);
+
+   *pView = anv_image_view_to_handle(view);
+
+   return VK_SUCCESS;
+}
+
+VkResult
+anv_DestroyImageView(VkDevice _device, VkImageView _iview)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_image_view, iview, _iview);
+
+   anv_surface_view_fini(device, &iview->view);
+   anv_device_free(device, iview);
+
+   return VK_SUCCESS;
+}
+
+void
+anv_color_attachment_view_init(struct anv_color_attachment_view *aview,
+                               struct anv_device *device,
+                               const VkAttachmentViewCreateInfo* pCreateInfo,
+                               struct anv_cmd_buffer *cmd_buffer)
+{
+   ANV_FROM_HANDLE(anv_image, image, pCreateInfo->image);
+   struct anv_surface_view *view = &aview->view;
+   struct anv_surface *surface = &image->primary_surface;
+   const struct anv_format *format_info =
+      anv_format_for_vk_format(pCreateInfo->format);
+
+   aview->base.attachment_type = ANV_ATTACHMENT_VIEW_TYPE_COLOR;
+
+   anv_assert(pCreateInfo->arraySize > 0);
+   anv_assert(pCreateInfo->mipLevel < image->levels);
+   anv_assert(pCreateInfo->baseArraySlice + pCreateInfo->arraySize <= image->array_size);
+
+   view->bo = image->bo;
+   view->offset = image->offset + surface->offset;
+   view->format = pCreateInfo->format;
+
+   view->extent = (VkExtent3D) {
+      .width = anv_minify(image->extent.width, pCreateInfo->mipLevel),
+      .height = anv_minify(image->extent.height, pCreateInfo->mipLevel),
+      .depth = anv_minify(image->extent.depth, pCreateInfo->mipLevel),
+   };
+
+   uint32_t depth = 1;
+   if (pCreateInfo->arraySize > 1) {
+      depth = pCreateInfo->arraySize;
+   } else if (image->extent.depth > 1) {
+      depth = image->extent.depth;
+   }
+
+   if (cmd_buffer)
+      view->surface_state =
+         anv_state_stream_alloc(&cmd_buffer->surface_state_stream, 64, 64);
+   else
+      view->surface_state =
+         anv_state_pool_alloc(&device->surface_state_pool, 64, 64);
+
+   struct GEN8_RENDER_SURFACE_STATE surface_state = {
+      .SurfaceType = SURFTYPE_2D,
+      .SurfaceArray = image->array_size > 1,
+      .SurfaceFormat = format_info->surface_format,
+      .SurfaceVerticalAlignment = anv_valign[surface->v_align],
+      .SurfaceHorizontalAlignment = anv_halign[surface->h_align],
+      .TileMode = surface->tile_mode,
+      .VerticalLineStride = 0,
+      .VerticalLineStrideOffset = 0,
+      .SamplerL2BypassModeDisable = true,
+      .RenderCacheReadWriteMode = WriteOnlyCache,
+      .MemoryObjectControlState = GEN8_MOCS,
+
+      /* The driver sets BaseMipLevel in SAMPLER_STATE, not here in
+       * RENDER_SURFACE_STATE. The Broadwell PRM says "it is illegal to have
+       * both Base Mip Level fields nonzero".
+       */
+      .BaseMipLevel = 0.0,
+
+      .SurfaceQPitch = surface->qpitch >> 2,
+      .Height = image->extent.height - 1,
+      .Width = image->extent.width - 1,
+      .Depth = depth - 1,
+      .SurfacePitch = surface->stride - 1,
+      .MinimumArrayElement = pCreateInfo->baseArraySlice,
+      .NumberofMultisamples = MULTISAMPLECOUNT_1,
+      .XOffset = 0,
+      .YOffset = 0,
+
+      /* For render target surfaces, the hardware interprets field MIPCount/LOD as
+       * LOD. The Broadwell PRM says:
+       *
+       *    MIPCountLOD defines the LOD that will be rendered into.
+       *    SurfaceMinLOD is ignored.
+       */
+      .SurfaceMinLOD = 0,
+      .MIPCountLOD = pCreateInfo->mipLevel,
+
+      .AuxiliarySurfaceMode = AUX_NONE,
+      .RedClearColor = 0,
+      .GreenClearColor = 0,
+      .BlueClearColor = 0,
+      .AlphaClearColor = 0,
+      .ShaderChannelSelectRed = SCS_RED,
+      .ShaderChannelSelectGreen = SCS_GREEN,
+      .ShaderChannelSelectBlue = SCS_BLUE,
+      .ShaderChannelSelectAlpha = SCS_ALPHA,
+      .ResourceMinLOD = 0.0,
+      .SurfaceBaseAddress = { NULL, view->offset },
+   };
+
+   GEN8_RENDER_SURFACE_STATE_pack(NULL, view->surface_state.map, &surface_state);
+}
+
+static void
+anv_depth_stencil_view_init(struct anv_depth_stencil_view *view,
+                            const VkAttachmentViewCreateInfo *pCreateInfo)
+{
+   ANV_FROM_HANDLE(anv_image, image, pCreateInfo->image);
+   struct anv_surface *depth_surface = &image->primary_surface;
+   struct anv_surface *stencil_surface = &image->stencil_surface;
+   const struct anv_format *format =
+      anv_format_for_vk_format(image->format);
+
+   view->base.attachment_type = ANV_ATTACHMENT_VIEW_TYPE_DEPTH_STENCIL;
+
+   /* XXX: We don't handle any of these */
+   anv_assert(pCreateInfo->mipLevel == 0);
+   anv_assert(pCreateInfo->baseArraySlice == 0);
+   anv_assert(pCreateInfo->arraySize == 1);
+
+   view->bo = image->bo;
+
+   view->depth_stride = depth_surface->stride;
+   view->depth_offset = image->offset + depth_surface->offset;
+   view->depth_format = format->depth_format;
+   view->depth_qpitch = 0; /* FINISHME: QPitch */
+
+   view->stencil_stride = stencil_surface->stride;
+   view->stencil_offset = image->offset + stencil_surface->offset;
+   view->stencil_qpitch = 0; /* FINISHME: QPitch */
+}
+
+VkResult
+anv_CreateAttachmentView(VkDevice _device,
+                         const VkAttachmentViewCreateInfo *pCreateInfo,
+                         VkAttachmentView *pView)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_ATTACHMENT_VIEW_CREATE_INFO);
+
+   if (anv_is_vk_format_depth_or_stencil(pCreateInfo->format)) {
+      struct anv_depth_stencil_view *view =
+         anv_device_alloc(device, sizeof(*view), 8,
+                          VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+      if (view == NULL)
+         return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+      anv_depth_stencil_view_init(view, pCreateInfo);
+
+      *pView = anv_attachment_view_to_handle(&view->base);
+   } else {
+      struct anv_color_attachment_view *view =
+         anv_device_alloc(device, sizeof(*view), 8,
+                          VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+      if (view == NULL)
+         return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+      anv_color_attachment_view_init(view, device, pCreateInfo, NULL);
+
+      *pView = anv_attachment_view_to_handle(&view->base);
+   }
+
+   return VK_SUCCESS;
+}
+
+VkResult
+anv_DestroyAttachmentView(VkDevice _device, VkAttachmentView _view)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_attachment_view, view, _view);
+
+   if (view->attachment_type == ANV_ATTACHMENT_VIEW_TYPE_COLOR) {
+      struct anv_color_attachment_view *aview =
+         (struct anv_color_attachment_view *)view;
+
+      anv_surface_view_fini(device, &aview->view);
+   }
+
+   anv_device_free(device, view);
+
+   return VK_SUCCESS;
+}

src/vulkan/intel.c

@@ -0,0 +1,97 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 <assert.h>
+#include <stdbool.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+#include "private.h"
+
+VkResult anv_CreateDmaBufImageINTEL(
+    VkDevice                                    _device,
+    const VkDmaBufImageCreateInfo*              pCreateInfo,
+    VkDeviceMemory*                             pMem,
+    VkImage*                                    pImage)
+{
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_device_memory *mem;
+   struct anv_image *image;
+   VkResult result;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DMA_BUF_IMAGE_CREATE_INFO_INTEL);
+
+   mem = anv_device_alloc(device, sizeof(*mem), 8,
+                          VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (mem == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   mem->bo.gem_handle = anv_gem_fd_to_handle(device, pCreateInfo->fd);
+   if (!mem->bo.gem_handle) {
+      result = vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
+      goto fail;
+   }
+
+   mem->bo.map = NULL;
+   mem->bo.index = 0;
+   mem->bo.offset = 0;
+   mem->bo.size = pCreateInfo->strideInBytes * pCreateInfo->extent.height;
+
+   image = anv_device_alloc(device, sizeof(*image), 8,
+                            VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (image == NULL) {
+      result = vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+      goto fail_mem;
+   }
+
+   *image = (struct anv_image) {
+      .bo = &mem->bo,
+      .offset = 0,
+      .type = VK_IMAGE_TYPE_2D,
+      .extent = pCreateInfo->extent,
+      .size = mem->bo.size,
+
+      .primary_surface = {
+         .offset = 0,
+         .stride = pCreateInfo->strideInBytes,
+         .tile_mode = XMAJOR,
+      },
+   };
+
+   assert(image->extent.width > 0);
+   assert(image->extent.height > 0);
+   assert(image->extent.depth == 1);
+
+   *pMem = anv_device_memory_to_handle(mem);
+   *pImage = anv_image_to_handle(image);
+
+   return VK_SUCCESS;
+
+ fail_mem:
+   anv_gem_close(device, mem->bo.gem_handle);
+ fail:
+   anv_device_free(device, mem);
+
+   return result;
+}

src/vulkan/pipeline.c

@@ -0,0 +1,950 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 <assert.h>
+#include <stdbool.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+#include "private.h"
+
+// Shader functions
+
+VkResult anv_CreateShaderModule(
+    VkDevice                                    _device,
+    const VkShaderModuleCreateInfo*             pCreateInfo,
+    VkShaderModule*                             pShaderModule)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   struct anv_shader_module *module;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO);
+   assert(pCreateInfo->flags == 0);
+
+   module = anv_device_alloc(device, sizeof(*module) + pCreateInfo->codeSize, 8,
+                             VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (module == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   module->size = pCreateInfo->codeSize;
+   memcpy(module->data, pCreateInfo->pCode, module->size);
+
+   *pShaderModule = anv_shader_module_to_handle(module);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_DestroyShaderModule(
+    VkDevice                                    _device,
+    VkShaderModule                              _module)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_shader_module, module, _module);
+
+   anv_device_free(device, module);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_CreateShader(
+    VkDevice                                    _device,
+    const VkShaderCreateInfo*                   pCreateInfo,
+    VkShader*                                   pShader)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_shader_module, module, pCreateInfo->module);
+   struct anv_shader *shader;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SHADER_CREATE_INFO);
+   assert(pCreateInfo->flags == 0);
+
+   size_t name_len = strlen(pCreateInfo->pName);
+
+   if (strcmp(pCreateInfo->pName, "main") != 0) {
+      anv_finishme("Multiple shaders per module not really supported");
+   }
+
+   shader = anv_device_alloc(device, sizeof(*shader) + name_len + 1, 8,
+                             VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (shader == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   shader->module = module;
+   memcpy(shader->entrypoint, pCreateInfo->pName, name_len + 1);
+
+   *pShader = anv_shader_to_handle(shader);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_DestroyShader(
+    VkDevice                                    _device,
+    VkShader                                    _shader)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_shader, shader, _shader);
+
+   anv_device_free(device, shader);
+
+   return VK_SUCCESS;
+}
+
+
+VkResult anv_CreatePipelineCache(
+    VkDevice                                    device,
+    const VkPipelineCacheCreateInfo*            pCreateInfo,
+    VkPipelineCache*                            pPipelineCache)
+{
+   pPipelineCache->handle = 1;
+
+   stub_return(VK_SUCCESS);
+}
+
+VkResult anv_DestroyPipelineCache(
+    VkDevice                                    _device,
+    VkPipelineCache                             _cache)
+{
+   /* VkPipelineCache is a dummy object. */
+   return VK_SUCCESS;
+}
+
+size_t anv_GetPipelineCacheSize(
+    VkDevice                                    device,
+    VkPipelineCache                             pipelineCache)
+{
+   stub_return(0);
+}
+
+VkResult anv_GetPipelineCacheData(
+    VkDevice                                    device,
+    VkPipelineCache                             pipelineCache,
+    void*                                       pData)
+{
+   stub_return(VK_UNSUPPORTED);
+}
+
+VkResult anv_MergePipelineCaches(
+    VkDevice                                    device,
+    VkPipelineCache                             destCache,
+    uint32_t                                    srcCacheCount,
+    const VkPipelineCache*                      pSrcCaches)
+{
+   stub_return(VK_UNSUPPORTED);
+}
+
+// Pipeline functions
+
+static void
+emit_vertex_input(struct anv_pipeline *pipeline,
+                  const VkPipelineVertexInputStateCreateInfo *info)
+{
+   const uint32_t num_dwords = 1 + info->attributeCount * 2;
+   uint32_t *p;
+   bool instancing_enable[32];
+
+   pipeline->vb_used = 0;
+   for (uint32_t i = 0; i < info->bindingCount; i++) {
+      const VkVertexInputBindingDescription *desc =
+         &info->pVertexBindingDescriptions[i];
+
+      pipeline->vb_used |= 1 << desc->binding;
+      pipeline->binding_stride[desc->binding] = desc->strideInBytes;
+
+      /* Step rate is programmed per vertex element (attribute), not
+       * binding. Set up a map of which bindings step per instance, for
+       * reference by vertex element setup. */
+      switch (desc->stepRate) {
+      default:
+      case VK_VERTEX_INPUT_STEP_RATE_VERTEX:
+         instancing_enable[desc->binding] = false;
+         break;
+      case VK_VERTEX_INPUT_STEP_RATE_INSTANCE:
+         instancing_enable[desc->binding] = true;
+         break;
+      }
+   }
+
+   p = anv_batch_emitn(&pipeline->batch, num_dwords,
+                       GEN8_3DSTATE_VERTEX_ELEMENTS);
+
+   for (uint32_t i = 0; i < info->attributeCount; i++) {
+      const VkVertexInputAttributeDescription *desc =
+         &info->pVertexAttributeDescriptions[i];
+      const struct anv_format *format = anv_format_for_vk_format(desc->format);
+
+      struct GEN8_VERTEX_ELEMENT_STATE element = {
+         .VertexBufferIndex = desc->binding,
+         .Valid = true,
+         .SourceElementFormat = format->surface_format,
+         .EdgeFlagEnable = false,
+         .SourceElementOffset = desc->offsetInBytes,
+         .Component0Control = VFCOMP_STORE_SRC,
+         .Component1Control = format->num_channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
+         .Component2Control = format->num_channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
+         .Component3Control = format->num_channels >= 4 ? VFCOMP_STORE_SRC : VFCOMP_STORE_1_FP
+      };
+      GEN8_VERTEX_ELEMENT_STATE_pack(NULL, &p[1 + i * 2], &element);
+
+      anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_INSTANCING,
+                     .InstancingEnable = instancing_enable[desc->binding],
+                     .VertexElementIndex = i,
+                     /* Vulkan so far doesn't have an instance divisor, so
+                      * this is always 1 (ignored if not instancing). */
+                     .InstanceDataStepRate = 1);
+   }
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_SGVS,
+                  .VertexIDEnable = pipeline->vs_prog_data.uses_vertexid,
+                  .VertexIDComponentNumber = 2,
+                  .VertexIDElementOffset = info->bindingCount,
+                  .InstanceIDEnable = pipeline->vs_prog_data.uses_instanceid,
+                  .InstanceIDComponentNumber = 3,
+                  .InstanceIDElementOffset = info->bindingCount);
+}
+
+static void
+emit_ia_state(struct anv_pipeline *pipeline,
+              const VkPipelineInputAssemblyStateCreateInfo *info,
+              const struct anv_pipeline_create_info *extra)
+{
+   static const uint32_t vk_to_gen_primitive_type[] = {
+      [VK_PRIMITIVE_TOPOLOGY_POINT_LIST]        = _3DPRIM_POINTLIST,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_LIST]         = _3DPRIM_LINELIST,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP]        = _3DPRIM_LINESTRIP,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST]     = _3DPRIM_TRILIST,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP]    = _3DPRIM_TRISTRIP,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_FAN]      = _3DPRIM_TRIFAN,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_LIST_ADJ]     = _3DPRIM_LINELIST_ADJ,
+      [VK_PRIMITIVE_TOPOLOGY_LINE_STRIP_ADJ]    = _3DPRIM_LINESTRIP_ADJ,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_LIST_ADJ] = _3DPRIM_TRILIST_ADJ,
+      [VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP_ADJ] = _3DPRIM_TRISTRIP_ADJ,
+      [VK_PRIMITIVE_TOPOLOGY_PATCH]             = _3DPRIM_PATCHLIST_1
+   };
+   uint32_t topology = vk_to_gen_primitive_type[info->topology];
+
+   if (extra && extra->use_rectlist)
+      topology = _3DPRIM_RECTLIST;
+
+   struct GEN8_3DSTATE_VF vf = {
+      GEN8_3DSTATE_VF_header,
+      .IndexedDrawCutIndexEnable = info->primitiveRestartEnable,
+   };
+   GEN8_3DSTATE_VF_pack(NULL, pipeline->state_vf, &vf);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_TOPOLOGY,
+                  .PrimitiveTopologyType = topology);
+}
+
+static void
+emit_rs_state(struct anv_pipeline *pipeline,
+              const VkPipelineRasterStateCreateInfo *info,
+              const struct anv_pipeline_create_info *extra)
+{
+   static const uint32_t vk_to_gen_cullmode[] = {
+      [VK_CULL_MODE_NONE]                       = CULLMODE_NONE,
+      [VK_CULL_MODE_FRONT]                      = CULLMODE_FRONT,
+      [VK_CULL_MODE_BACK]                       = CULLMODE_BACK,
+      [VK_CULL_MODE_FRONT_AND_BACK]             = CULLMODE_BOTH
+   };
+
+   static const uint32_t vk_to_gen_fillmode[] = {
+      [VK_FILL_MODE_POINTS]                     = RASTER_POINT,
+      [VK_FILL_MODE_WIREFRAME]                  = RASTER_WIREFRAME,
+      [VK_FILL_MODE_SOLID]                      = RASTER_SOLID
+   };
+
+   static const uint32_t vk_to_gen_front_face[] = {
+      [VK_FRONT_FACE_CCW]                       = CounterClockwise,
+      [VK_FRONT_FACE_CW]                        = Clockwise
+   };
+
+   struct GEN8_3DSTATE_SF sf = {
+      GEN8_3DSTATE_SF_header,
+      .ViewportTransformEnable = !(extra && extra->disable_viewport),
+      .TriangleStripListProvokingVertexSelect = 0,
+      .LineStripListProvokingVertexSelect = 0,
+      .TriangleFanProvokingVertexSelect = 0,
+      .PointWidthSource = pipeline->writes_point_size ? Vertex : State,
+      .PointWidth = 1.0,
+   };
+
+   /* FINISHME: VkBool32 rasterizerDiscardEnable; */
+
+   GEN8_3DSTATE_SF_pack(NULL, pipeline->state_sf, &sf);
+
+   struct GEN8_3DSTATE_RASTER raster = {
+      GEN8_3DSTATE_RASTER_header,
+      .FrontWinding = vk_to_gen_front_face[info->frontFace],
+      .CullMode = vk_to_gen_cullmode[info->cullMode],
+      .FrontFaceFillMode = vk_to_gen_fillmode[info->fillMode],
+      .BackFaceFillMode = vk_to_gen_fillmode[info->fillMode],
+      .ScissorRectangleEnable = !(extra && extra->disable_scissor),
+      .ViewportZClipTestEnable = info->depthClipEnable
+   };
+
+   GEN8_3DSTATE_RASTER_pack(NULL, pipeline->state_raster, &raster);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE,
+                  .ForceVertexURBEntryReadLength = false,
+                  .ForceVertexURBEntryReadOffset = false,
+                  .PointSpriteTextureCoordinateOrigin = UPPERLEFT,
+                  .NumberofSFOutputAttributes =
+                     pipeline->wm_prog_data.num_varying_inputs);
+
+}
+
+static void
+emit_cb_state(struct anv_pipeline *pipeline,
+              const VkPipelineColorBlendStateCreateInfo *info)
+{
+   struct anv_device *device = pipeline->device;
+
+   static const uint32_t vk_to_gen_logic_op[] = {
+      [VK_LOGIC_OP_COPY]                        = LOGICOP_COPY,
+      [VK_LOGIC_OP_CLEAR]                       = LOGICOP_CLEAR,
+      [VK_LOGIC_OP_AND]                         = LOGICOP_AND,
+      [VK_LOGIC_OP_AND_REVERSE]                 = LOGICOP_AND_REVERSE,
+      [VK_LOGIC_OP_AND_INVERTED]                = LOGICOP_AND_INVERTED,
+      [VK_LOGIC_OP_NOOP]                        = LOGICOP_NOOP,
+      [VK_LOGIC_OP_XOR]                         = LOGICOP_XOR,
+      [VK_LOGIC_OP_OR]                          = LOGICOP_OR,
+      [VK_LOGIC_OP_NOR]                         = LOGICOP_NOR,
+      [VK_LOGIC_OP_EQUIV]                       = LOGICOP_EQUIV,
+      [VK_LOGIC_OP_INVERT]                      = LOGICOP_INVERT,
+      [VK_LOGIC_OP_OR_REVERSE]                  = LOGICOP_OR_REVERSE,
+      [VK_LOGIC_OP_COPY_INVERTED]               = LOGICOP_COPY_INVERTED,
+      [VK_LOGIC_OP_OR_INVERTED]                 = LOGICOP_OR_INVERTED,
+      [VK_LOGIC_OP_NAND]                        = LOGICOP_NAND,
+      [VK_LOGIC_OP_SET]                         = LOGICOP_SET,
+   };
+
+   static const uint32_t vk_to_gen_blend[] = {
+      [VK_BLEND_ZERO]                           = BLENDFACTOR_ZERO,
+      [VK_BLEND_ONE]                            = BLENDFACTOR_ONE,
+      [VK_BLEND_SRC_COLOR]                      = BLENDFACTOR_SRC_COLOR,
+      [VK_BLEND_ONE_MINUS_SRC_COLOR]            = BLENDFACTOR_INV_SRC_COLOR,
+      [VK_BLEND_DEST_COLOR]                     = BLENDFACTOR_DST_COLOR,
+      [VK_BLEND_ONE_MINUS_DEST_COLOR]           = BLENDFACTOR_INV_DST_COLOR,
+      [VK_BLEND_SRC_ALPHA]                      = BLENDFACTOR_SRC_ALPHA,
+      [VK_BLEND_ONE_MINUS_SRC_ALPHA]            = BLENDFACTOR_INV_SRC_ALPHA,
+      [VK_BLEND_DEST_ALPHA]                     = BLENDFACTOR_DST_ALPHA,
+      [VK_BLEND_ONE_MINUS_DEST_ALPHA]           = BLENDFACTOR_INV_DST_ALPHA,
+      [VK_BLEND_CONSTANT_COLOR]                 = BLENDFACTOR_CONST_COLOR,
+      [VK_BLEND_ONE_MINUS_CONSTANT_COLOR]       = BLENDFACTOR_INV_CONST_COLOR,
+      [VK_BLEND_CONSTANT_ALPHA]                 = BLENDFACTOR_CONST_ALPHA,
+      [VK_BLEND_ONE_MINUS_CONSTANT_ALPHA]       = BLENDFACTOR_INV_CONST_ALPHA,
+      [VK_BLEND_SRC_ALPHA_SATURATE]             = BLENDFACTOR_SRC_ALPHA_SATURATE,
+      [VK_BLEND_SRC1_COLOR]                     = BLENDFACTOR_SRC1_COLOR,
+      [VK_BLEND_ONE_MINUS_SRC1_COLOR]           = BLENDFACTOR_INV_SRC1_COLOR,
+      [VK_BLEND_SRC1_ALPHA]                     = BLENDFACTOR_SRC1_ALPHA,
+      [VK_BLEND_ONE_MINUS_SRC1_ALPHA]           = BLENDFACTOR_INV_SRC1_ALPHA,
+   };
+
+   static const uint32_t vk_to_gen_blend_op[] = {
+      [VK_BLEND_OP_ADD]                         = BLENDFUNCTION_ADD,
+      [VK_BLEND_OP_SUBTRACT]                    = BLENDFUNCTION_SUBTRACT,
+      [VK_BLEND_OP_REVERSE_SUBTRACT]            = BLENDFUNCTION_REVERSE_SUBTRACT,
+      [VK_BLEND_OP_MIN]                         = BLENDFUNCTION_MIN,
+      [VK_BLEND_OP_MAX]                         = BLENDFUNCTION_MAX,
+   };
+
+   uint32_t num_dwords = 1 + info->attachmentCount * 2;
+   pipeline->blend_state =
+      anv_state_pool_alloc(&device->dynamic_state_pool, num_dwords * 4, 64);
+
+   struct GEN8_BLEND_STATE blend_state = {
+      .AlphaToCoverageEnable = info->alphaToCoverageEnable,
+   };
+
+   uint32_t *state = pipeline->blend_state.map;
+   GEN8_BLEND_STATE_pack(NULL, state, &blend_state);
+
+   for (uint32_t i = 0; i < info->attachmentCount; i++) {
+      const VkPipelineColorBlendAttachmentState *a = &info->pAttachments[i];
+
+      struct GEN8_BLEND_STATE_ENTRY entry = {
+         .LogicOpEnable = info->logicOpEnable,
+         .LogicOpFunction = vk_to_gen_logic_op[info->logicOp],
+         .ColorBufferBlendEnable = a->blendEnable,
+         .PreBlendSourceOnlyClampEnable = false,
+         .PreBlendColorClampEnable = false,
+         .PostBlendColorClampEnable = false,
+         .SourceBlendFactor = vk_to_gen_blend[a->srcBlendColor],
+         .DestinationBlendFactor = vk_to_gen_blend[a->destBlendColor],
+         .ColorBlendFunction = vk_to_gen_blend_op[a->blendOpColor],
+         .SourceAlphaBlendFactor = vk_to_gen_blend[a->srcBlendAlpha],
+         .DestinationAlphaBlendFactor = vk_to_gen_blend[a->destBlendAlpha],
+         .AlphaBlendFunction = vk_to_gen_blend_op[a->blendOpAlpha],
+         .WriteDisableAlpha = !(a->channelWriteMask & VK_CHANNEL_A_BIT),
+         .WriteDisableRed = !(a->channelWriteMask & VK_CHANNEL_R_BIT),
+         .WriteDisableGreen = !(a->channelWriteMask & VK_CHANNEL_G_BIT),
+         .WriteDisableBlue = !(a->channelWriteMask & VK_CHANNEL_B_BIT),
+      };
+
+      GEN8_BLEND_STATE_ENTRY_pack(NULL, state + i * 2 + 1, &entry);
+   }
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_BLEND_STATE_POINTERS,
+                  .BlendStatePointer = pipeline->blend_state.offset,
+                  .BlendStatePointerValid = true);
+}
+
+static const uint32_t vk_to_gen_compare_op[] = {
+   [VK_COMPARE_OP_NEVER]                        = COMPAREFUNCTION_NEVER,
+   [VK_COMPARE_OP_LESS]                         = COMPAREFUNCTION_LESS,
+   [VK_COMPARE_OP_EQUAL]                        = COMPAREFUNCTION_EQUAL,
+   [VK_COMPARE_OP_LESS_EQUAL]                   = COMPAREFUNCTION_LEQUAL,
+   [VK_COMPARE_OP_GREATER]                      = COMPAREFUNCTION_GREATER,
+   [VK_COMPARE_OP_NOT_EQUAL]                    = COMPAREFUNCTION_NOTEQUAL,
+   [VK_COMPARE_OP_GREATER_EQUAL]                = COMPAREFUNCTION_GEQUAL,
+   [VK_COMPARE_OP_ALWAYS]                       = COMPAREFUNCTION_ALWAYS,
+};
+
+static const uint32_t vk_to_gen_stencil_op[] = {
+   [VK_STENCIL_OP_KEEP]                         = 0,
+   [VK_STENCIL_OP_ZERO]                         = 0,
+   [VK_STENCIL_OP_REPLACE]                      = 0,
+   [VK_STENCIL_OP_INC_CLAMP]                    = 0,
+   [VK_STENCIL_OP_DEC_CLAMP]                    = 0,
+   [VK_STENCIL_OP_INVERT]                       = 0,
+   [VK_STENCIL_OP_INC_WRAP]                     = 0,
+   [VK_STENCIL_OP_DEC_WRAP]                     = 0
+};
+
+static void
+emit_ds_state(struct anv_pipeline *pipeline,
+              const VkPipelineDepthStencilStateCreateInfo *info)
+{
+   if (info == NULL) {
+      /* We're going to OR this together with the dynamic state.  We need
+       * to make sure it's initialized to something useful.
+       */
+      memset(pipeline->state_wm_depth_stencil, 0,
+             sizeof(pipeline->state_wm_depth_stencil));
+      return;
+   }
+
+   /* VkBool32 depthBoundsEnable;          // optional (depth_bounds_test) */
+
+   struct GEN8_3DSTATE_WM_DEPTH_STENCIL wm_depth_stencil = {
+      .DepthTestEnable = info->depthTestEnable,
+      .DepthBufferWriteEnable = info->depthWriteEnable,
+      .DepthTestFunction = vk_to_gen_compare_op[info->depthCompareOp],
+      .DoubleSidedStencilEnable = true,
+
+      .StencilTestEnable = info->stencilTestEnable,
+      .StencilFailOp = vk_to_gen_stencil_op[info->front.stencilFailOp],
+      .StencilPassDepthPassOp = vk_to_gen_stencil_op[info->front.stencilPassOp],
+      .StencilPassDepthFailOp = vk_to_gen_stencil_op[info->front.stencilDepthFailOp],
+      .StencilTestFunction = vk_to_gen_compare_op[info->front.stencilCompareOp],
+      .BackfaceStencilFailOp = vk_to_gen_stencil_op[info->back.stencilFailOp],
+      .BackfaceStencilPassDepthPassOp = vk_to_gen_stencil_op[info->back.stencilPassOp],
+      .BackfaceStencilPassDepthFailOp =vk_to_gen_stencil_op[info->back.stencilDepthFailOp],
+      .BackfaceStencilTestFunction = vk_to_gen_compare_op[info->back.stencilCompareOp],
+   };
+
+   GEN8_3DSTATE_WM_DEPTH_STENCIL_pack(NULL, pipeline->state_wm_depth_stencil, &wm_depth_stencil);
+}
+
+VkResult
+anv_pipeline_create(
+    VkDevice                                    _device,
+    const VkGraphicsPipelineCreateInfo*         pCreateInfo,
+    const struct anv_pipeline_create_info *     extra,
+    VkPipeline*                                 pPipeline)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   struct anv_pipeline *pipeline;
+   VkResult result;
+   uint32_t offset, length;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO);
+   
+   pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
+                               VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (pipeline == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   pipeline->device = device;
+   pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
+   memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
+
+   result = anv_reloc_list_init(&pipeline->batch.relocs, device);
+   if (result != VK_SUCCESS) {
+      anv_device_free(device, pipeline);
+      return result;
+   }
+   pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
+   pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
+
+   anv_state_stream_init(&pipeline->program_stream,
+                         &device->instruction_block_pool);
+
+   for (uint32_t i = 0; i < pCreateInfo->stageCount; i++) {
+      pipeline->shaders[pCreateInfo->pStages[i].stage] =
+         anv_shader_from_handle(pCreateInfo->pStages[i].shader);
+   }
+
+   if (pCreateInfo->pTessellationState)
+      anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_TESSELLATION_STATE_CREATE_INFO");
+   if (pCreateInfo->pViewportState)
+      anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_VIEWPORT_STATE_CREATE_INFO");
+   if (pCreateInfo->pMultisampleState)
+      anv_finishme("VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO");
+
+   pipeline->use_repclear = extra && extra->use_repclear;
+
+   anv_compiler_run(device->compiler, pipeline);
+
+   /* FIXME: The compiler dead-codes FS inputs when we don't have a VS, so we
+    * hard code this to num_attributes - 2. This is because the attributes
+    * include VUE header and position, which aren't counted as varying
+    * inputs. */
+   if (pipeline->vs_simd8 == NO_KERNEL) {
+      pipeline->wm_prog_data.num_varying_inputs =
+         pCreateInfo->pVertexInputState->attributeCount - 2;
+   }
+
+   assert(pCreateInfo->pVertexInputState);
+   emit_vertex_input(pipeline, pCreateInfo->pVertexInputState);
+   assert(pCreateInfo->pInputAssemblyState);
+   emit_ia_state(pipeline, pCreateInfo->pInputAssemblyState, extra);
+   assert(pCreateInfo->pRasterState);
+   emit_rs_state(pipeline, pCreateInfo->pRasterState, extra);
+   emit_ds_state(pipeline, pCreateInfo->pDepthStencilState);
+   emit_cb_state(pipeline, pCreateInfo->pColorBlendState);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_STATISTICS,
+                   .StatisticsEnable = true);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_HS, .Enable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_TE, .TEEnable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_DS, .FunctionEnable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_STREAMOUT, .SOFunctionEnable = false);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_VS,
+                  .ConstantBufferOffset = 0,
+                  .ConstantBufferSize = 4);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_GS,
+                  .ConstantBufferOffset = 4,
+                  .ConstantBufferSize = 4);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PUSH_CONSTANT_ALLOC_PS,
+                  .ConstantBufferOffset = 8,
+                  .ConstantBufferSize = 4);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM_CHROMAKEY,
+                  .ChromaKeyKillEnable = false);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE_SWIZ);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_AA_LINE_PARAMETERS);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_CLIP,
+                  .ClipEnable = true,
+                  .ViewportXYClipTestEnable = !(extra && extra->disable_viewport),
+                  .MinimumPointWidth = 0.125,
+                  .MaximumPointWidth = 255.875);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_WM,
+                  .StatisticsEnable = true,
+                  .LineEndCapAntialiasingRegionWidth = _05pixels,
+                  .LineAntialiasingRegionWidth = _10pixels,
+                  .EarlyDepthStencilControl = NORMAL,
+                  .ForceThreadDispatchEnable = NORMAL,
+                  .PointRasterizationRule = RASTRULE_UPPER_RIGHT,
+                  .BarycentricInterpolationMode =
+                     pipeline->wm_prog_data.barycentric_interp_modes);
+
+   uint32_t samples = 1;
+   uint32_t log2_samples = __builtin_ffs(samples) - 1;
+   bool enable_sampling = samples > 1 ? true : false;
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_MULTISAMPLE,
+                  .PixelPositionOffsetEnable = enable_sampling,
+                  .PixelLocation = CENTER,
+                  .NumberofMultisamples = log2_samples);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SAMPLE_MASK,
+                  .SampleMask = 0xffff);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_VS,
+                  .VSURBStartingAddress = pipeline->urb.vs_start,
+                  .VSURBEntryAllocationSize = pipeline->urb.vs_size - 1,
+                  .VSNumberofURBEntries = pipeline->urb.nr_vs_entries);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_GS,
+                  .GSURBStartingAddress = pipeline->urb.gs_start,
+                  .GSURBEntryAllocationSize = pipeline->urb.gs_size - 1,
+                  .GSNumberofURBEntries = pipeline->urb.nr_gs_entries);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_HS,
+                  .HSURBStartingAddress = pipeline->urb.vs_start,
+                  .HSURBEntryAllocationSize = 0,
+                  .HSNumberofURBEntries = 0);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_URB_DS,
+                  .DSURBStartingAddress = pipeline->urb.vs_start,
+                  .DSURBEntryAllocationSize = 0,
+                  .DSNumberofURBEntries = 0);
+
+   const struct brw_gs_prog_data *gs_prog_data = &pipeline->gs_prog_data;
+   offset = 1;
+   length = (gs_prog_data->base.vue_map.num_slots + 1) / 2 - offset;
+
+   if (pipeline->gs_vec4 == NO_KERNEL)
+      anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS, .Enable = false);
+   else
+      anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_GS,
+                     .SingleProgramFlow = false,
+                     .KernelStartPointer = pipeline->gs_vec4,
+                     .VectorMaskEnable = Vmask,
+                     .SamplerCount = 0,
+                     .BindingTableEntryCount = 0,
+                     .ExpectedVertexCount = pipeline->gs_vertex_count,
+                        
+                     .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_GEOMETRY],
+                     .PerThreadScratchSpace = ffs(gs_prog_data->base.base.total_scratch / 2048),
+
+                     .OutputVertexSize = gs_prog_data->output_vertex_size_hwords * 2 - 1,
+                     .OutputTopology = gs_prog_data->output_topology,
+                     .VertexURBEntryReadLength = gs_prog_data->base.urb_read_length,
+                     .DispatchGRFStartRegisterForURBData =
+                        gs_prog_data->base.base.dispatch_grf_start_reg,
+
+                     .MaximumNumberofThreads = device->info.max_gs_threads,
+                     .ControlDataHeaderSize = gs_prog_data->control_data_header_size_hwords,
+                     //pipeline->gs_prog_data.dispatch_mode |
+                     .StatisticsEnable = true,
+                     .IncludePrimitiveID = gs_prog_data->include_primitive_id,
+                     .ReorderMode = TRAILING,
+                     .Enable = true,
+
+                     .ControlDataFormat = gs_prog_data->control_data_format,
+
+                     /* FIXME: mesa sets this based on ctx->Transform.ClipPlanesEnabled:
+                      * UserClipDistanceClipTestEnableBitmask_3DSTATE_GS(v)
+                      * UserClipDistanceCullTestEnableBitmask(v)
+                      */
+
+                     .VertexURBEntryOutputReadOffset = offset,
+                     .VertexURBEntryOutputLength = length);
+
+   const struct brw_vue_prog_data *vue_prog_data = &pipeline->vs_prog_data.base;
+   /* Skip the VUE header and position slots */
+   offset = 1;
+   length = (vue_prog_data->vue_map.num_slots + 1) / 2 - offset;
+
+   if (pipeline->vs_simd8 == NO_KERNEL || (extra && extra->disable_vs))
+      anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
+                     .FunctionEnable = false,
+                     .VertexURBEntryOutputReadOffset = 1,
+                     /* Even if VS is disabled, SBE still gets the amount of
+                      * vertex data to read from this field. We use attribute
+                      * count - 1, as we don't count the VUE header here. */
+                     .VertexURBEntryOutputLength =
+                        DIV_ROUND_UP(pCreateInfo->pVertexInputState->attributeCount - 1, 2));
+   else
+      anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VS,
+                     .KernelStartPointer = pipeline->vs_simd8,
+                     .SingleVertexDispatch = Multiple,
+                     .VectorMaskEnable = Dmask,
+                     .SamplerCount = 0,
+                     .BindingTableEntryCount =
+                     vue_prog_data->base.binding_table.size_bytes / 4,
+                     .ThreadDispatchPriority = Normal,
+                     .FloatingPointMode = IEEE754,
+                     .IllegalOpcodeExceptionEnable = false,
+                     .AccessesUAV = false,
+                     .SoftwareExceptionEnable = false,
+
+                     .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_VERTEX],
+                     .PerThreadScratchSpace = ffs(vue_prog_data->base.total_scratch / 2048),
+
+                     .DispatchGRFStartRegisterForURBData =
+                     vue_prog_data->base.dispatch_grf_start_reg,
+                     .VertexURBEntryReadLength = vue_prog_data->urb_read_length,
+                     .VertexURBEntryReadOffset = 0,
+
+                     .MaximumNumberofThreads = device->info.max_vs_threads - 1,
+                     .StatisticsEnable = false,
+                     .SIMD8DispatchEnable = true,
+                     .VertexCacheDisable = false,
+                     .FunctionEnable = true,
+
+                     .VertexURBEntryOutputReadOffset = offset,
+                     .VertexURBEntryOutputLength = length,
+                     .UserClipDistanceClipTestEnableBitmask = 0,
+                     .UserClipDistanceCullTestEnableBitmask = 0);
+
+   const struct brw_wm_prog_data *wm_prog_data = &pipeline->wm_prog_data;
+   uint32_t ksp0, ksp2, grf_start0, grf_start2;
+
+   ksp2 = 0;
+   grf_start2 = 0;
+   if (pipeline->ps_simd8 != NO_KERNEL) {
+      ksp0 = pipeline->ps_simd8;
+      grf_start0 = wm_prog_data->base.dispatch_grf_start_reg;
+      if (pipeline->ps_simd16 != NO_KERNEL) {
+         ksp2 = pipeline->ps_simd16;
+         grf_start2 = wm_prog_data->dispatch_grf_start_reg_16;
+      }
+   } else if (pipeline->ps_simd16 != NO_KERNEL) {
+      ksp0 = pipeline->ps_simd16;
+      grf_start0 = wm_prog_data->dispatch_grf_start_reg_16;
+   } else {
+      unreachable("no ps shader");
+   }
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS,
+                  .KernelStartPointer0 = ksp0,
+   
+                  .SingleProgramFlow = false,
+                  .VectorMaskEnable = true,
+                  .SamplerCount = 1,
+
+                  .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
+                  .PerThreadScratchSpace = ffs(wm_prog_data->base.total_scratch / 2048),
+                  
+                  .MaximumNumberofThreadsPerPSD = 64 - 2,
+                  .PositionXYOffsetSelect = wm_prog_data->uses_pos_offset ?
+                     POSOFFSET_SAMPLE: POSOFFSET_NONE,
+                  .PushConstantEnable = wm_prog_data->base.nr_params > 0,
+                  ._8PixelDispatchEnable = pipeline->ps_simd8 != NO_KERNEL,
+                  ._16PixelDispatchEnable = pipeline->ps_simd16 != NO_KERNEL,
+                  ._32PixelDispatchEnable = false,
+
+                  .DispatchGRFStartRegisterForConstantSetupData0 = grf_start0,
+                  .DispatchGRFStartRegisterForConstantSetupData1 = 0,
+                  .DispatchGRFStartRegisterForConstantSetupData2 = grf_start2,
+
+                  .KernelStartPointer1 = 0,
+                  .KernelStartPointer2 = ksp2);
+
+   bool per_sample_ps = false;
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_PS_EXTRA,
+                  .PixelShaderValid = true,
+                  .PixelShaderKillsPixel = wm_prog_data->uses_kill,
+                  .PixelShaderComputedDepthMode = wm_prog_data->computed_depth_mode,
+                  .AttributeEnable = wm_prog_data->num_varying_inputs > 0,
+                  .oMaskPresenttoRenderTarget = wm_prog_data->uses_omask,
+                  .PixelShaderIsPerSample = per_sample_ps);
+
+   *pPipeline = anv_pipeline_to_handle(pipeline);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_DestroyPipeline(
+    VkDevice                                    _device,
+    VkPipeline                                  _pipeline)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_pipeline, pipeline, _pipeline);
+
+   anv_compiler_free(pipeline);
+   anv_reloc_list_finish(&pipeline->batch.relocs, pipeline->device);
+   anv_state_stream_finish(&pipeline->program_stream);
+   anv_state_pool_free(&device->dynamic_state_pool, pipeline->blend_state);
+   anv_device_free(pipeline->device, pipeline);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_CreateGraphicsPipelines(
+    VkDevice                                    _device,
+    VkPipelineCache                             pipelineCache,
+    uint32_t                                    count,
+    const VkGraphicsPipelineCreateInfo*         pCreateInfos,
+    VkPipeline*                                 pPipelines)
+{
+   VkResult result = VK_SUCCESS;
+
+   unsigned i = 0;
+   for (; i < count; i++) {
+      result = anv_pipeline_create(_device, &pCreateInfos[i],
+                                   NULL, &pPipelines[i]);
+      if (result != VK_SUCCESS) {
+         for (unsigned j = 0; j < i; j++) {
+            anv_DestroyPipeline(_device, pPipelines[j]);
+         }
+
+         return result;
+      }
+   }
+
+   return VK_SUCCESS;
+}
+
+static VkResult anv_compute_pipeline_create(
+    VkDevice                                    _device,
+    const VkComputePipelineCreateInfo*          pCreateInfo,
+    VkPipeline*                                 pPipeline)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   struct anv_pipeline *pipeline;
+   VkResult result;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO);
+
+   pipeline = anv_device_alloc(device, sizeof(*pipeline), 8,
+                               VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (pipeline == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   pipeline->device = device;
+   pipeline->layout = anv_pipeline_layout_from_handle(pCreateInfo->layout);
+
+   result = anv_reloc_list_init(&pipeline->batch.relocs, device);
+   if (result != VK_SUCCESS) {
+      anv_device_free(device, pipeline);
+      return result;
+   }
+   pipeline->batch.next = pipeline->batch.start = pipeline->batch_data;
+   pipeline->batch.end = pipeline->batch.start + sizeof(pipeline->batch_data);
+
+   anv_state_stream_init(&pipeline->program_stream,
+                         &device->instruction_block_pool);
+
+   memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
+
+   pipeline->shaders[VK_SHADER_STAGE_COMPUTE] =
+      anv_shader_from_handle(pCreateInfo->cs.shader);
+
+   pipeline->use_repclear = false;
+
+   anv_compiler_run(device->compiler, pipeline);
+
+   const struct brw_cs_prog_data *cs_prog_data = &pipeline->cs_prog_data;
+
+   anv_batch_emit(&pipeline->batch, GEN8_MEDIA_VFE_STATE,
+                  .ScratchSpaceBasePointer = pipeline->scratch_start[VK_SHADER_STAGE_FRAGMENT],
+                  .PerThreadScratchSpace = ffs(cs_prog_data->base.total_scratch / 2048),
+                  .ScratchSpaceBasePointerHigh = 0,
+                  .StackSize = 0,
+
+                  .MaximumNumberofThreads = device->info.max_cs_threads - 1,
+                  .NumberofURBEntries = 2,
+                  .ResetGatewayTimer = true,
+                  .BypassGatewayControl = true,
+                  .URBEntryAllocationSize = 2,
+                  .CURBEAllocationSize = 0);
+
+   struct brw_cs_prog_data *prog_data = &pipeline->cs_prog_data;
+   uint32_t group_size = prog_data->local_size[0] *
+      prog_data->local_size[1] * prog_data->local_size[2];
+   pipeline->cs_thread_width_max = DIV_ROUND_UP(group_size, prog_data->simd_size);
+   uint32_t remainder = group_size & (prog_data->simd_size - 1);
+
+   if (remainder > 0)
+      pipeline->cs_right_mask = ~0u >> (32 - remainder);
+   else
+      pipeline->cs_right_mask = ~0u >> (32 - prog_data->simd_size);
+
+
+   *pPipeline = anv_pipeline_to_handle(pipeline);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_CreateComputePipelines(
+    VkDevice                                    _device,
+    VkPipelineCache                             pipelineCache,
+    uint32_t                                    count,
+    const VkComputePipelineCreateInfo*          pCreateInfos,
+    VkPipeline*                                 pPipelines)
+{
+   VkResult result = VK_SUCCESS;
+
+   unsigned i = 0;
+   for (; i < count; i++) {
+      result = anv_compute_pipeline_create(_device, &pCreateInfos[i],
+                                           &pPipelines[i]);
+      if (result != VK_SUCCESS) {
+         for (unsigned j = 0; j < i; j++) {
+            anv_DestroyPipeline(_device, pPipelines[j]);
+         }
+
+         return result;
+      }
+   }
+
+   return VK_SUCCESS;
+}
+
+// Pipeline layout functions
+
+VkResult anv_CreatePipelineLayout(
+    VkDevice                                    _device,
+    const VkPipelineLayoutCreateInfo*           pCreateInfo,
+    VkPipelineLayout*                           pPipelineLayout)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   struct anv_pipeline_layout *layout;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO);
+
+   layout = anv_device_alloc(device, sizeof(*layout), 8,
+                             VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (layout == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   layout->num_sets = pCreateInfo->descriptorSetCount;
+
+   uint32_t surface_start[VK_SHADER_STAGE_NUM] = { 0, };
+   uint32_t sampler_start[VK_SHADER_STAGE_NUM] = { 0, };
+
+   for (uint32_t s = 0; s < VK_SHADER_STAGE_NUM; s++) {
+      layout->stage[s].surface_count = 0;
+      layout->stage[s].sampler_count = 0;
+   }
+
+   for (uint32_t i = 0; i < pCreateInfo->descriptorSetCount; i++) {
+      ANV_FROM_HANDLE(anv_descriptor_set_layout, set_layout,
+                      pCreateInfo->pSetLayouts[i]);
+
+      layout->set[i].layout = set_layout;
+      for (uint32_t s = 0; s < VK_SHADER_STAGE_NUM; s++) {
+         layout->set[i].surface_start[s] = surface_start[s];
+         surface_start[s] += set_layout->stage[s].surface_count;
+         layout->set[i].sampler_start[s] = sampler_start[s];
+         sampler_start[s] += set_layout->stage[s].sampler_count;
+
+         layout->stage[s].surface_count += set_layout->stage[s].surface_count;
+         layout->stage[s].sampler_count += set_layout->stage[s].sampler_count;
+      }
+   }
+
+   *pPipelineLayout = anv_pipeline_layout_to_handle(layout);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_DestroyPipelineLayout(
+    VkDevice                                    _device,
+    VkPipelineLayout                            _pipelineLayout)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_pipeline_layout, pipeline_layout, _pipelineLayout);
+
+   anv_device_free(device, pipeline_layout);
+
+   return VK_SUCCESS;
+}

src/vulkan/query.c

@@ -0,0 +1,352 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 <assert.h>
+#include <stdbool.h>
+#include <string.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+#include "private.h"
+
+struct anv_query_pool_slot {
+   uint64_t begin;
+   uint64_t end;
+   uint64_t available;
+};
+
+struct anv_query_pool {
+   VkQueryType                                  type;
+   uint32_t                                     slots;
+   struct anv_bo                                bo;
+};
+
+VkResult anv_CreateQueryPool(
+    VkDevice                                    _device,
+    const VkQueryPoolCreateInfo*                pCreateInfo,
+    VkQueryPool*                                pQueryPool)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   struct anv_query_pool *pool;
+   VkResult result;
+   size_t size;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_QUERY_POOL_CREATE_INFO);
+
+   switch (pCreateInfo->queryType) {
+   case VK_QUERY_TYPE_OCCLUSION:
+      break;
+   case VK_QUERY_TYPE_PIPELINE_STATISTICS:
+      return VK_UNSUPPORTED;
+   default:
+      unreachable("");
+   }
+
+   pool = anv_device_alloc(device, sizeof(*pool), 8,
+                            VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (pool == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   size = pCreateInfo->slots * sizeof(struct anv_query_pool_slot);
+   result = anv_bo_init_new(&pool->bo, device, size);
+   if (result != VK_SUCCESS)
+      goto fail;
+
+   pool->bo.map = anv_gem_mmap(device, pool->bo.gem_handle, 0, size);
+
+   *pQueryPool = anv_query_pool_to_handle(pool);
+
+   return VK_SUCCESS;
+
+ fail:
+   anv_device_free(device, pool);
+
+   return result;
+}
+
+VkResult anv_DestroyQueryPool(
+    VkDevice                                    _device,
+    VkQueryPool                                 _pool)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_query_pool, pool, _pool);
+
+   anv_gem_munmap(pool->bo.map, pool->bo.size);
+   anv_gem_close(device, pool->bo.gem_handle);
+   anv_device_free(device, pool);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_GetQueryPoolResults(
+    VkDevice                                    _device,
+    VkQueryPool                                 queryPool,
+    uint32_t                                    startQuery,
+    uint32_t                                    queryCount,
+    size_t*                                     pDataSize,
+    void*                                       pData,
+    VkQueryResultFlags                          flags)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+   ANV_FROM_HANDLE(anv_query_pool, pool, queryPool);
+   struct anv_query_pool_slot *slot = pool->bo.map;
+   int64_t timeout = INT64_MAX;
+   uint32_t *dst32 = pData;
+   uint64_t *dst64 = pData;
+   uint64_t result;
+   int ret;
+
+   if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) {
+      /* Where is the availabilty info supposed to go? */
+      anv_finishme("VK_QUERY_RESULT_WITH_AVAILABILITY_BIT");
+      return VK_UNSUPPORTED;
+   }
+
+   assert(pool->type == VK_QUERY_TYPE_OCCLUSION);
+
+   if (flags & VK_QUERY_RESULT_64_BIT)
+      *pDataSize = queryCount * sizeof(uint64_t);
+   else
+      *pDataSize = queryCount * sizeof(uint32_t);
+
+   if (pData == NULL)
+      return VK_SUCCESS;
+
+   if (flags & VK_QUERY_RESULT_WAIT_BIT) {
+      ret = anv_gem_wait(device, pool->bo.gem_handle, &timeout);
+      if (ret == -1)
+         return vk_error(VK_ERROR_UNKNOWN);
+   }
+
+   for (uint32_t i = 0; i < queryCount; i++) {
+      result = slot[startQuery + i].end - slot[startQuery + i].begin;
+      if (flags & VK_QUERY_RESULT_64_BIT) {
+         *dst64++ = result;
+      } else {
+         if (result > UINT32_MAX)
+            result = UINT32_MAX;
+         *dst32++ = result;
+      }
+   }
+
+   return VK_SUCCESS;
+}
+
+static void
+anv_batch_emit_ps_depth_count(struct anv_batch *batch,
+                              struct anv_bo *bo, uint32_t offset)
+{
+   anv_batch_emit(batch, GEN8_PIPE_CONTROL,
+                  .DestinationAddressType = DAT_PPGTT,
+                  .PostSyncOperation = WritePSDepthCount,
+                  .Address = { bo, offset });  /* FIXME: This is only lower 32 bits */
+}
+
+void anv_CmdBeginQuery(
+    VkCmdBuffer                                 cmdBuffer,
+    VkQueryPool                                 queryPool,
+    uint32_t                                    slot,
+    VkQueryControlFlags                         flags)
+{
+   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
+   ANV_FROM_HANDLE(anv_query_pool, pool, queryPool);
+
+   switch (pool->type) {
+   case VK_QUERY_TYPE_OCCLUSION:
+      anv_batch_emit_ps_depth_count(&cmd_buffer->batch, &pool->bo,
+                                    slot * sizeof(struct anv_query_pool_slot));
+      break;
+
+   case VK_QUERY_TYPE_PIPELINE_STATISTICS:
+   default:
+      unreachable("");
+   }
+}
+
+void anv_CmdEndQuery(
+    VkCmdBuffer                                 cmdBuffer,
+    VkQueryPool                                 queryPool,
+    uint32_t                                    slot)
+{
+   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
+   ANV_FROM_HANDLE(anv_query_pool, pool, queryPool);
+
+   switch (pool->type) {
+   case VK_QUERY_TYPE_OCCLUSION:
+      anv_batch_emit_ps_depth_count(&cmd_buffer->batch, &pool->bo,
+                                    slot * sizeof(struct anv_query_pool_slot) + 8);
+      break;
+
+   case VK_QUERY_TYPE_PIPELINE_STATISTICS:
+   default:
+      unreachable("");
+   }
+}
+
+void anv_CmdResetQueryPool(
+    VkCmdBuffer                                 cmdBuffer,
+    VkQueryPool                                 queryPool,
+    uint32_t                                    startQuery,
+    uint32_t                                    queryCount)
+{
+   stub();
+}
+
+#define TIMESTAMP 0x2358
+
+void anv_CmdWriteTimestamp(
+    VkCmdBuffer                                 cmdBuffer,
+    VkTimestampType                             timestampType,
+    VkBuffer                                    destBuffer,
+    VkDeviceSize                                destOffset)
+{
+   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
+   ANV_FROM_HANDLE(anv_buffer, buffer, destBuffer);
+   struct anv_bo *bo = buffer->bo;
+
+   switch (timestampType) {
+   case VK_TIMESTAMP_TYPE_TOP:
+      anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
+                     .RegisterAddress = TIMESTAMP,
+                     .MemoryAddress = { bo, buffer->offset + destOffset });
+      anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
+                     .RegisterAddress = TIMESTAMP + 4,
+                     .MemoryAddress = { bo, buffer->offset + destOffset + 4 });
+      break;
+
+   case VK_TIMESTAMP_TYPE_BOTTOM:
+      anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
+                     .DestinationAddressType = DAT_PPGTT,
+                     .PostSyncOperation = WriteTimestamp,
+                     .Address = /* FIXME: This is only lower 32 bits */
+                        { bo, buffer->offset + destOffset });
+      break;
+
+   default:
+      break;
+   }
+}
+
+#define alu_opcode(v)   __gen_field((v),  20, 31)
+#define alu_operand1(v) __gen_field((v),  10, 19)
+#define alu_operand2(v) __gen_field((v),   0,  9)
+#define alu(opcode, operand1, operand2) \
+   alu_opcode(opcode) | alu_operand1(operand1) | alu_operand2(operand2)
+
+#define OPCODE_NOOP      0x000
+#define OPCODE_LOAD      0x080
+#define OPCODE_LOADINV   0x480
+#define OPCODE_LOAD0     0x081
+#define OPCODE_LOAD1     0x481
+#define OPCODE_ADD       0x100
+#define OPCODE_SUB       0x101
+#define OPCODE_AND       0x102
+#define OPCODE_OR        0x103
+#define OPCODE_XOR       0x104
+#define OPCODE_STORE     0x180
+#define OPCODE_STOREINV  0x580
+
+#define OPERAND_R0   0x00
+#define OPERAND_R1   0x01
+#define OPERAND_R2   0x02
+#define OPERAND_R3   0x03
+#define OPERAND_R4   0x04
+#define OPERAND_SRCA 0x20
+#define OPERAND_SRCB 0x21
+#define OPERAND_ACCU 0x31
+#define OPERAND_ZF   0x32
+#define OPERAND_CF   0x33
+
+#define CS_GPR(n) (0x2600 + (n) * 8)
+
+static void
+emit_load_alu_reg_u64(struct anv_batch *batch, uint32_t reg,
+                      struct anv_bo *bo, uint32_t offset)
+{
+   anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_MEM,
+                  .RegisterAddress = reg,
+                  .MemoryAddress = { bo, offset });
+   anv_batch_emit(batch, GEN8_MI_LOAD_REGISTER_MEM,
+                  .RegisterAddress = reg + 4,
+                  .MemoryAddress = { bo, offset + 4 });
+}
+
+void anv_CmdCopyQueryPoolResults(
+    VkCmdBuffer                                 cmdBuffer,
+    VkQueryPool                                 queryPool,
+    uint32_t                                    startQuery,
+    uint32_t                                    queryCount,
+    VkBuffer                                    destBuffer,
+    VkDeviceSize                                destOffset,
+    VkDeviceSize                                destStride,
+    VkQueryResultFlags                          flags)
+{
+   ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, cmdBuffer);
+   ANV_FROM_HANDLE(anv_query_pool, pool, queryPool);
+   ANV_FROM_HANDLE(anv_buffer, buffer, destBuffer);
+   uint32_t slot_offset, dst_offset;
+
+   if (flags & VK_QUERY_RESULT_WITH_AVAILABILITY_BIT) {
+      /* Where is the availabilty info supposed to go? */
+      anv_finishme("VK_QUERY_RESULT_WITH_AVAILABILITY_BIT");
+      return;
+   }
+
+   assert(pool->type == VK_QUERY_TYPE_OCCLUSION);
+
+   /* FIXME: If we're not waiting, should we just do this on the CPU? */
+   if (flags & VK_QUERY_RESULT_WAIT_BIT)
+      anv_batch_emit(&cmd_buffer->batch, GEN8_PIPE_CONTROL,
+                     .CommandStreamerStallEnable = true,
+                     .StallAtPixelScoreboard = true);
+
+   dst_offset = buffer->offset + destOffset;
+   for (uint32_t i = 0; i < queryCount; i++) {
+
+      slot_offset = (startQuery + i) * sizeof(struct anv_query_pool_slot);
+
+      emit_load_alu_reg_u64(&cmd_buffer->batch, CS_GPR(0), &pool->bo, slot_offset);
+      emit_load_alu_reg_u64(&cmd_buffer->batch, CS_GPR(1), &pool->bo, slot_offset + 8);
+
+      /* FIXME: We need to clamp the result for 32 bit. */
+
+      uint32_t *dw = anv_batch_emitn(&cmd_buffer->batch, 5, GEN8_MI_MATH);
+      dw[1] = alu(OPCODE_LOAD, OPERAND_SRCA, OPERAND_R1);
+      dw[2] = alu(OPCODE_LOAD, OPERAND_SRCB, OPERAND_R0);
+      dw[3] = alu(OPCODE_SUB, 0, 0);
+      dw[4] = alu(OPCODE_STORE, OPERAND_R2, OPERAND_ACCU);
+
+      anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
+                     .RegisterAddress = CS_GPR(2),
+                     /* FIXME: This is only lower 32 bits */
+                     .MemoryAddress = { buffer->bo, dst_offset });
+
+      if (flags & VK_QUERY_RESULT_64_BIT)
+         anv_batch_emit(&cmd_buffer->batch, GEN8_MI_STORE_REGISTER_MEM,
+                        .RegisterAddress = CS_GPR(2) + 4,
+                        /* FIXME: This is only lower 32 bits */
+                        .MemoryAddress = { buffer->bo, dst_offset + 4 });
+
+      dst_offset += destStride;
+   }
+}

src/vulkan/util.c

@@ -0,0 +1,151 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+#include <assert.h>
+
+#include "private.h"
+
+/** Log an error message.  */
+void anv_printflike(1, 2)
+anv_loge(const char *format, ...)
+{
+   va_list va;
+
+   va_start(va, format);
+   anv_loge_v(format, va);
+   va_end(va);
+}
+
+/** \see anv_loge() */
+void
+anv_loge_v(const char *format, va_list va)
+{
+   fprintf(stderr, "vk: error: ");
+   vfprintf(stderr, format, va);
+   fprintf(stderr, "\n");
+}
+
+void anv_printflike(3, 4)
+__anv_finishme(const char *file, int line, const char *format, ...)
+{
+   va_list ap;
+   char buffer[256];
+
+   va_start(ap, format);
+   vsnprintf(buffer, sizeof(buffer), format, ap);
+   va_end(ap);
+
+   fprintf(stderr, "%s:%d: FINISHME: %s\n", file, line, buffer);
+}
+
+void anv_noreturn anv_printflike(1, 2)
+anv_abortf(const char *format, ...)
+{
+   va_list va;
+
+   va_start(va, format);
+   anv_abortfv(format, va);
+   va_end(va);
+}
+
+void anv_noreturn
+anv_abortfv(const char *format, va_list va)
+{
+   fprintf(stderr, "vk: error: ");
+   vfprintf(stderr, format, va);
+   fprintf(stderr, "\n");
+   abort();
+}
+
+int
+anv_vector_init(struct anv_vector *vector, uint32_t element_size, uint32_t size)
+{
+   assert(is_power_of_two(size));
+   assert(element_size < size && is_power_of_two(element_size));
+
+   vector->head = 0;
+   vector->tail = 0;
+   vector->element_size = element_size;
+   vector->size = size;
+   vector->data = malloc(size);
+
+   return vector->data != NULL;
+}
+
+void *
+anv_vector_add(struct anv_vector *vector)
+{
+   uint32_t offset, size, split, tail;
+   void *data;
+
+   if (vector->head - vector->tail == vector->size) {
+      size = vector->size * 2;
+      data = malloc(size);
+      if (data == NULL)
+         return NULL;
+      split = align_u32(vector->tail, vector->size);
+      tail = vector->tail & (vector->size - 1);
+      if (vector->head - split < vector->size) {
+         memcpy(data + tail,
+                vector->data + tail,
+                split - vector->tail);
+         memcpy(data + vector->size,
+                vector->data, vector->head - split);
+      } else {
+         memcpy(data + tail,
+                vector->data + tail,
+                vector->head - vector->tail);
+      }
+      free(vector->data);
+      vector->data = data;
+      vector->size = size;
+   }
+
+   assert(vector->head - vector->tail < vector->size);
+
+   offset = vector->head & (vector->size - 1);
+   vector->head += vector->element_size;
+
+   return vector->data + offset;
+}
+
+void *
+anv_vector_remove(struct anv_vector *vector)
+{
+   uint32_t offset;
+
+   if (vector->head == vector->tail)
+      return NULL;
+
+   assert(vector->head - vector->tail <= vector->size);
+
+   offset = vector->tail & (vector->size - 1);
+   vector->tail += vector->element_size;
+
+   return vector->data + offset;
+}

src/vulkan/vk_gen.py

@@ -0,0 +1,269 @@
+# coding=utf-8
+#
+# Copyright © 2015 Intel Corporation
+#
+# 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, sublicense,
+# 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+# THE AUTHORS OR COPYRIGHT HOLDERS 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.
+#
+
+import fileinput, re, sys
+
+# Each function typedef in the vulkan.h header is all on one line and matches
+# this regepx. We hope that won't change.
+
+p = re.compile('typedef ([^ ]*) *\(VKAPI \*PFN_vk([^(]*)\)(.*);')
+
+entrypoints = []
+
+# We generate a static hash table for entry point lookup
+# (vkGetProcAddress). We use a linear congruential generator for our hash
+# function and a power-of-two size table. The prime numbers are determined
+# experimentally.
+
+none = 0xffff
+hash_size = 256
+u32_mask = 2**32 - 1
+hash_mask = hash_size - 1
+
+prime_factor = 5024183
+prime_step = 19
+
+def hash(name):
+    h = 0;
+    for c in name:
+        h = (h * prime_factor + ord(c)) & u32_mask
+
+    return h
+
+opt_header = False
+opt_code = False
+
+if (sys.argv[1] == "header"):
+    opt_header = True
+    sys.argv.pop()
+elif (sys.argv[1] == "code"):
+    opt_code = True
+    sys.argv.pop()
+
+# Parse the entry points in the header
+
+i = 0
+for line in fileinput.input():
+    m  = p.match(line)
+    if (m):
+        if m.group(2) == 'VoidFunction':
+            continue
+        fullname = "vk" + m.group(2)
+        h = hash(fullname)
+        entrypoints.append((m.group(1), m.group(2), m.group(3), i, h))
+        i = i + 1
+
+# For outputting entrypoints.h we generate a anv_EntryPoint() prototype
+# per entry point.
+
+if opt_header:
+    for type, name, args, num, h in entrypoints:
+        print "%s anv_%s%s;" % (type, name, args)
+        print "%s anv_validate_%s%s;" % (type, name, args)
+    exit()
+
+
+
+print """/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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.
+ */
+
+/* This file generated from vk_gen.py, don't edit directly. */
+
+#include "private.h"
+
+struct anv_entrypoint {
+   uint32_t name;
+   uint32_t hash;
+   void *function;
+   void *validate;
+};
+
+/* We use a big string constant to avoid lots of reloctions from the entry
+ * point table to lots of little strings. The entries in the entry point table
+ * store the index into this big string.
+ */
+
+static const char strings[] ="""
+
+offsets = []
+i = 0;
+for type, name, args, num, h in entrypoints:
+    print "   \"vk%s\\0\"" % name
+    offsets.append(i)
+    i += 2 + len(name) + 1
+print """   ;
+
+/* Weak aliases for all potential validate functions. These will resolve to
+ * NULL if they're not defined, which lets the resolve_entrypoint() function
+ * either pick a validate wrapper if available or just plug in the actual
+ * entry point.
+ */
+"""
+
+for type, name, args, num, h in entrypoints:
+    print "%s anv_validate_%s%s __attribute__ ((weak));" % (type, name, args)
+
+# Now generate the table of all entry points and their validation functions
+
+print "\nstatic const struct anv_entrypoint entrypoints[] = {"
+for type, name, args, num, h in entrypoints:
+    print "   { %5d, 0x%08x, anv_%s, anv_validate_%s }," % (offsets[num], h, name, name)
+print "};\n"
+
+print """
+#ifdef DEBUG
+static bool enable_validate = true;
+#else
+static bool enable_validate = false;
+#endif
+
+/* We can't use symbols that need resolving (like, oh, getenv) in the resolve
+ * function. This means that we have to determine whether or not to use the
+ * validation layer sometime before that. The constructor function attribute asks
+ * the dynamic linker to invoke determine_validate() at dlopen() time which
+ * works.
+ */
+static void __attribute__ ((constructor))
+determine_validate(void)
+{
+   const char *s = getenv("ANV_VALIDATE");
+
+   if (s)
+      enable_validate = atoi(s);
+}
+
+static void * __attribute__ ((noinline))
+resolve_entrypoint(uint32_t index)
+{
+   if (enable_validate && entrypoints[index].validate)
+      return entrypoints[index].validate;
+
+   return entrypoints[index].function;
+}
+"""
+
+# Now output ifuncs and their resolve helpers for all entry points. The
+# resolve helper calls resolve_entrypoint() with the entry point index, which
+# lets the resolver look it up in the table.
+
+for type, name, args, num, h in entrypoints:
+    print "static void *resolve_%s(void) { return resolve_entrypoint(%d); }" % (name, num)
+    print "%s vk%s%s\n   __attribute__ ((ifunc (\"resolve_%s\"), visibility (\"default\")));\n" % (type, name, args, name)
+
+
+# Now generate the hash table used for entry point look up.  This is a
+# uint16_t table of entry point indices. We use 0xffff to indicate an entry
+# in the hash table is empty.
+
+map = [none for f in xrange(hash_size)]
+collisions = [0 for f in xrange(10)]
+for type, name, args, num, h in entrypoints:
+    level = 0
+    while map[h & hash_mask] != none:
+        h = h + prime_step
+        level = level + 1
+    if level > 9:
+        collisions[9] += 1
+    else:
+        collisions[level] += 1
+    map[h & hash_mask] = num
+
+print "/* Hash table stats:"
+print " * size %d entries" % hash_size
+print " * collisions  entries"
+for i in xrange(10):
+    if (i == 9):
+        plus = "+"
+    else:
+        plus = " "
+
+    print " *     %2d%s     %4d" % (i, plus, collisions[i])
+print " */\n"
+
+print "#define none 0x%04x\n" % none
+
+print "static const uint16_t map[] = {"
+for i in xrange(0, hash_size, 8):
+    print "   ",
+    for j in xrange(i, i + 8):
+        if map[j] & 0xffff == 0xffff:
+            print "  none,",
+        else:
+            print "0x%04x," % (map[j] & 0xffff),
+    print
+
+print "};"    
+
+# Finally we generate the hash table lookup function.  The hash function and
+# linear probing algorithm matches the hash table generated above.
+
+print """
+void *
+anv_lookup_entrypoint(const char *name)
+{
+   static const uint32_t prime_factor = %d;
+   static const uint32_t prime_step = %d;
+   const struct anv_entrypoint *e;
+   uint32_t hash, h, i;
+   const char *p;
+
+   hash = 0;
+   for (p = name; *p; p++)
+      hash = hash * prime_factor + *p;
+
+   h = hash;
+   do {
+      i = map[h & %d];
+      if (i == none)
+         return NULL;
+      e = &entrypoints[i];
+      h += prime_step;
+   } while (e->hash != hash);
+
+   if (strcmp(name, strings + e->name) != 0)
+      return NULL;
+
+   return resolve_entrypoint(i);
+}
+""" % (prime_factor, prime_step, hash_mask)

src/vulkan/x11.c

@@ -0,0 +1,299 @@
+/*
+ * Copyright © 2015 Intel Corporation
+ *
+ * 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, sublicense,
+ * 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 NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE AUTHORS OR COPYRIGHT HOLDERS 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 "private.h"
+
+#include <xcb/xcb.h>
+#include <xcb/dri3.h>
+#include <xcb/present.h>
+
+static const VkFormat formats[] = {
+   VK_FORMAT_B5G6R5_UNORM,
+   VK_FORMAT_B8G8R8A8_UNORM,
+   VK_FORMAT_B8G8R8A8_SRGB,
+};
+
+VkResult anv_GetDisplayInfoWSI(
+    VkDisplayWSI                            display,
+    VkDisplayInfoTypeWSI                    infoType,
+    size_t*                                 pDataSize,
+    void*                                   pData)
+{
+   VkDisplayFormatPropertiesWSI *properties = pData;
+   size_t size;
+
+   if (pDataSize == NULL)
+      return VK_ERROR_INVALID_POINTER;
+
+   switch (infoType) {
+   case VK_DISPLAY_INFO_TYPE_FORMAT_PROPERTIES_WSI:
+      size = sizeof(properties[0]) * ARRAY_SIZE(formats);
+
+      if (pData == NULL) {
+         *pDataSize = size;
+         return VK_SUCCESS;
+      }
+
+      if (*pDataSize < size)
+         return vk_error(VK_ERROR_INVALID_VALUE);
+
+      *pDataSize = size;
+
+      for (uint32_t i = 0; i < ARRAY_SIZE(formats); i++)
+         properties[i].swapChainFormat = formats[i];
+
+      return VK_SUCCESS;
+
+   default:
+      return VK_UNSUPPORTED;
+   }
+}
+
+struct anv_swap_chain {
+   struct anv_device *                          device;
+   xcb_connection_t *                           conn;
+   xcb_window_t                                 window;
+   xcb_gc_t                                     gc;
+   VkExtent2D                                   extent;
+   uint32_t                                     count;
+   struct {
+      struct anv_image *                        image;
+      struct anv_device_memory *                memory;
+      xcb_pixmap_t                              pixmap;
+   }                                            images[0];
+};
+
+VkResult anv_CreateSwapChainWSI(
+    VkDevice                                _device,
+    const VkSwapChainCreateInfoWSI*         pCreateInfo,
+    VkSwapChainWSI*                         pSwapChain)
+{
+   ANV_FROM_HANDLE(anv_device, device, _device);
+
+   struct anv_swap_chain *chain;
+   xcb_void_cookie_t cookie;
+   VkResult result;
+   size_t size;
+   int ret;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SWAP_CHAIN_CREATE_INFO_WSI);
+
+   size = sizeof(*chain) + pCreateInfo->imageCount * sizeof(chain->images[0]);
+   chain = anv_device_alloc(device, size, 8,
+                            VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (chain == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   chain->device = device;
+   chain->conn = (xcb_connection_t *) pCreateInfo->pNativeWindowSystemHandle;
+   chain->window = (xcb_window_t) (uintptr_t) pCreateInfo->pNativeWindowHandle;
+   chain->count = pCreateInfo->imageCount;
+   chain->extent = pCreateInfo->imageExtent;
+
+   for (uint32_t i = 0; i < chain->count; i++) {
+      VkDeviceMemory memory_h;
+      VkImage image_h;
+      struct anv_image *image;
+      struct anv_surface *surface;
+      struct anv_device_memory *memory;
+
+      anv_image_create(_device,
+         &(struct anv_image_create_info) {
+            .force_tile_mode = true,
+            .tile_mode = XMAJOR,
+            .vk_info =
+         &(VkImageCreateInfo) {
+            .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+            .imageType = VK_IMAGE_TYPE_2D,
+            .format = pCreateInfo->imageFormat,
+            .extent = {
+               .width = pCreateInfo->imageExtent.width,
+               .height = pCreateInfo->imageExtent.height,
+               .depth = 1
+            },
+            .mipLevels = 1,
+            .arraySize = 1,
+            .samples = 1,
+            /* FIXME: Need a way to use X tiling to allow scanout */
+            .tiling = VK_IMAGE_TILING_OPTIMAL,
+            .usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
+            .flags = 0,
+         }},
+         &image_h);
+
+      image = anv_image_from_handle(image_h);
+      surface = &image->primary_surface;
+
+      anv_AllocMemory(_device,
+                      &(VkMemoryAllocInfo) {
+                         .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOC_INFO,
+                         .allocationSize = image->size,
+                         .memoryTypeIndex = 0,
+                      },
+                      &memory_h);
+
+      memory = anv_device_memory_from_handle(memory_h);
+
+      anv_BindImageMemory(VK_NULL_HANDLE, anv_image_to_handle(image),
+                          memory_h, 0);
+
+      ret = anv_gem_set_tiling(device, memory->bo.gem_handle,
+                               surface->stride, I915_TILING_X);
+      if (ret) {
+         result = vk_error(VK_ERROR_UNKNOWN);
+         goto fail;
+      }
+
+      int fd = anv_gem_handle_to_fd(device, memory->bo.gem_handle);
+      if (fd == -1) {
+         result = vk_error(VK_ERROR_UNKNOWN);
+         goto fail;
+      }
+
+      uint32_t bpp = 32;
+      uint32_t depth = 24;
+      xcb_pixmap_t pixmap = xcb_generate_id(chain->conn);
+
+      cookie =
+         xcb_dri3_pixmap_from_buffer_checked(chain->conn,
+                                             pixmap,
+                                             chain->window,
+                                             image->size,
+                                             pCreateInfo->imageExtent.width,
+                                             pCreateInfo->imageExtent.height,
+                                             surface->stride,
+                                             depth, bpp, fd);
+
+      chain->images[i].image = image;
+      chain->images[i].memory = memory;
+      chain->images[i].pixmap = pixmap;
+      image->swap_chain = chain;
+
+      xcb_discard_reply(chain->conn, cookie.sequence);
+   }
+
+   chain->gc = xcb_generate_id(chain->conn);
+   if (!chain->gc) {
+      result = vk_error(VK_ERROR_UNKNOWN);
+      goto fail;
+   }
+
+   cookie = xcb_create_gc(chain->conn,
+                          chain->gc,
+                          chain->window,
+                          XCB_GC_GRAPHICS_EXPOSURES,
+                          (uint32_t []) { 0 });
+   xcb_discard_reply(chain->conn, cookie.sequence);
+
+   *pSwapChain = (VkSwapChainWSI) chain;
+
+   return VK_SUCCESS;
+
+ fail:
+   return result;
+}
+
+VkResult anv_DestroySwapChainWSI(
+    VkSwapChainWSI                          swapChain)
+{
+   struct anv_swap_chain *chain = (struct anv_swap_chain *) swapChain;
+   struct anv_device *device = chain->device;
+
+   anv_device_free(device, chain);
+
+   return VK_SUCCESS;
+}
+
+VkResult anv_GetSwapChainInfoWSI(
+    VkSwapChainWSI                          swapChain,
+    VkSwapChainInfoTypeWSI                  infoType,
+    size_t*                                 pDataSize,
+    void*                                   pData)
+{
+   struct anv_swap_chain *chain = (struct anv_swap_chain *) swapChain;
+   VkSwapChainImageInfoWSI *images;
+   size_t size;
+
+   switch (infoType) {
+   case VK_SWAP_CHAIN_INFO_TYPE_PERSISTENT_IMAGES_WSI:
+      size = sizeof(*images) * chain->count;
+      if (pData && *pDataSize < size)
+         return VK_ERROR_INVALID_VALUE;
+
+      *pDataSize = size;
+      if (!pData)
+         return VK_SUCCESS;
+
+      images = pData;
+      for (uint32_t i = 0; i < chain->count; i++) {
+         images[i].image = anv_image_to_handle(chain->images[i].image);
+         images[i].memory = anv_device_memory_to_handle(chain->images[i].memory);
+      }
+
+      return VK_SUCCESS;
+
+   default:
+      return VK_UNSUPPORTED;
+   }
+}
+
+VkResult anv_QueuePresentWSI(
+    VkQueue                                 queue_,
+    const VkPresentInfoWSI*                 pPresentInfo)
+{
+   ANV_FROM_HANDLE(anv_image, image, pPresentInfo->image);
+
+   struct anv_swap_chain *chain = image->swap_chain;
+   xcb_void_cookie_t cookie;
+   xcb_pixmap_t pixmap;
+
+   assert(pPresentInfo->sType == VK_STRUCTURE_TYPE_PRESENT_INFO_WSI);
+
+   if (chain == NULL)
+      return vk_error(VK_ERROR_INVALID_VALUE);
+
+   pixmap = XCB_NONE;
+   for (uint32_t i = 0; i < chain->count; i++) {
+      if (image == chain->images[i].image) {
+         pixmap = chain->images[i].pixmap;
+         break;
+      }
+   }
+
+   if (pixmap == XCB_NONE)
+      return vk_error(VK_ERROR_INVALID_VALUE);
+
+   cookie = xcb_copy_area(chain->conn,
+                          pixmap,
+                          chain->window,
+                          chain->gc,
+                          0, 0,
+                          0, 0,
+                          chain->extent.width,
+                          chain->extent.height);
+   xcb_discard_reply(chain->conn, cookie.sequence);
+
+   xcb_flush(chain->conn);
+
+   return VK_SUCCESS;
+}