Add vulkan driver for BDW

configure.ac

@@ -1158,6 +1158,10 @@ 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$have_libcaca = xyes])
+
 dnl
 dnl libGL configuration per driver
 dnl
@@ -1535,6 +1539,8 @@ GBM_PC_LIB_PRIV="$DLOPEN_LIBS"
 AC_SUBST([GBM_PC_REQ_PRIV])
 AC_SUBST([GBM_PC_LIB_PRIV])
 
+AM_CONDITIONAL(HAVE_VULKAN, true)
+
 dnl
 dnl EGL configuration
 dnl
@@ -2311,6 +2317,13 @@ 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"
@@ -2419,6 +2432,7 @@ 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   bool32_t;
+
+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/vulkan_intel.h

@@ -0,0 +1,55 @@
+/*
+ * 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;
+
+VkResult VKAPI vkCreateDmaBufImageINTEL(
+    VkDevice                                    _device,
+    const VkDmaBufImageCreateInfo*             pCreateInfo,
+    VkDeviceMemory*                             pMem,
+    VkImage*                                    pImage);
+
+#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/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

@@ -2645,7 +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_index)
+         _mesa_glsl_error(loc, state,
+                          "Vulkan descriptor set layout requires both group and index "
+                          "qualifiers");
+
+      var->data.vk_set = true;
+      var->data.set = qual->set;
+      var->data.index = qual->index;
+   } 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");
@@ -5782,6 +5791,10 @@ 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.index = this->layout.index;
+
          state->symbols->add_variable(var);
          instructions->push_tail(var);
       }
@@ -5854,6 +5867,10 @@ 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.index = this->layout.index;
+
          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->index = q.index;
+   }
+
    return true;
 }
 

src/glsl/glsl_parser.yy

@@ -1431,6 +1431,11 @@ layout_qualifier_id:
          $$.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/ir.h

@@ -682,6 +682,11 @@ 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?
        *
@@ -751,8 +756,10 @@ 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.
@@ -800,6 +807,11 @@ public:
        */
       int16_t binding;
 
+      /**
+       * Vulkan descriptor set for the resource.
+       */
+      int16_t set;
+
       /**
        * Storage location of the base of this variable
        *

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].Index = b->index;
+
             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].Index = b->index;
+
          i++;
       }
    }

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)
 
@@ -54,6 +54,15 @@ 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/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

@@ -830,6 +830,7 @@ 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
@@ -848,6 +849,7 @@ brwCreateContext(gl_api api,
    }
 
    brw_init_state(brw);
+#endif
 
    intelInitExtensions(ctx);
 
@@ -909,8 +911,10 @@ 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 +359,9 @@ struct brw_stage_prog_data {
       /** @} */
    } binding_table;
 
+   uint32_t *map_entries;
+   uint32_t *bind_map[4];
+
    GLuint nr_params;       /**< number of float params/constants */
    GLuint nr_pull_params;
 
@@ -1976,6 +1979,12 @@ 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);
+
 #ifdef __cplusplus
 }
 #endif

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

@@ -55,6 +55,7 @@
 # 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
@@ -76,6 +77,7 @@
 #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

@@ -1910,6 +1910,10 @@ 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;
 
@@ -3033,9 +3037,22 @@ 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;
+
+   if (uniforms == 1) {
+      mov = emit(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 = emit(MOV(vec4(brw_message_reg(color_mrf)), fs_reg(reg)));
+   }
 
-   fs_inst *mov = emit(MOV(vec4(brw_message_reg(color_mrf)),
-                           fs_reg(UNIFORM, 0, BRW_REGISTER_TYPE_F)));
    mov->force_writemask_all = true;
 
    fs_inst *write;
@@ -3065,8 +3082,10 @@ 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);
+   }
 }
 
 /**
@@ -4081,7 +4100,8 @@ fs_visitor::run_vs()
 {
    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 (INTEL_DEBUG & DEBUG_SHADER_TIME)
@@ -4129,7 +4149,8 @@ fs_visitor::run_fs()
 
    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();

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

@@ -1198,14 +1198,20 @@ fs_visitor::visit(ir_expression *ir)
       ir_constant *const_uniform_block = ir->operands[0]->as_constant();
       ir_constant *const_offset = ir->operands[1]->as_constant();
       fs_reg surf_index;
+      uint32_t binding, set, index, set_index;
 
       if (const_uniform_block) {
          /* The block index is a constant, so just emit the binding table entry
           * as an immediate.
           */
-         surf_index = fs_reg(stage_prog_data->binding_table.ubo_start +
-                                 const_uniform_block->value.u[0]);
+         index = const_uniform_block->value.u[0];
+         set = shader->base.UniformBlocks[index].Set;
+         set_index = shader->base.UniformBlocks[index].Index;
+         binding = stage_prog_data->bind_map[set][set_index];
+         surf_index = fs_reg(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.
@@ -2289,8 +2295,13 @@ fs_visitor::emit_texture(ir_texture_opcode op,
 void
 fs_visitor::visit(ir_texture *ir)
 {
-   uint32_t sampler =
-      _mesa_get_sampler_uniform_value(ir->sampler, shader_prog, prog);
+   uint32_t sampler;
+
+   ir_dereference_variable *deref_var = ir->sampler->as_dereference_variable();
+   assert(deref_var);
+   ir_variable *var = deref_var->var;
+
+   sampler = stage_prog_data->bind_map[var->data.set][var->data.index];
 
    ir_rvalue *nonconst_sampler_index =
       _mesa_get_sampler_array_nonconst_index(ir->sampler);

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_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_wm.c

@@ -46,7 +46,7 @@
  * 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_screen.c

@@ -1416,6 +1416,78 @@ __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

@@ -2570,6 +2570,12 @@ struct gl_uniform_block
     */
    GLuint Binding;
 
+   /**
+    * Vulkan descriptor set and index qualifiers for this block.
+    */
+   GLuint Set;
+   GLuint Index;
+
    /**
     * Minimum size (in bytes) of a buffer object to back this uniform buffer
     * (GL_UNIFORM_BLOCK_DATA_SIZE).

src/vulkan/Makefile.am

@@ -0,0 +1,67 @@
+# 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.
+
+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
+
+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				\
+	aub.c					\
+	allocator.c				\
+	util.c					\
+	pipeline.c				\
+	image.c					\
+	meta.c					\
+	intel.c					\
+	compiler.cpp
+
+bin_PROGRAMS = vk
+
+vk_SOURCES = vk.c
+vk_LDADD = libvulkan.la -lpng16
+
+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 +1,499 @@
+/*
+ * 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"
+
+/* 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();
+
+      next.offset = *(uint32_t *)(*map + current.offset);
+      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;
+      *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);
+}
+
+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);
+}
+
+/* The memfd path lets us create a map for an fd and lets us grow and remap
+ * without copying.  It breaks valgrind however, so we have a MAP_ANONYMOUS
+ * path we can take for valgrind debugging. */
+
+#define USE_MEMFD 0
+
+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);
+
+#if USE_MEMFD
+   close(pool->fd);
+#endif
+}
+
+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 USE_MEMFD
+   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;
+#else
+   /* The MAP_ANONYMOUS fallback can't grow without races, so just bail here
+    * if we're trying to grow the pool. */
+   assert(pool->size == 0);
+   map = mmap(NULL, size, PROT_READ | PROT_WRITE,
+              MAP_PRIVATE | MAP_ANONYMOUS | MAP_POPULATE, -1, 0);
+   if (map == MAP_FAILED)
+      return -1;
+   cleanup->map = map;
+   cleanup->size = size;
+#endif
+
+   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))
+      return offset;
+
+ restart:
+   size = pool->size;
+   block = __sync_fetch_and_add(&pool->next_block, pool->block_size);
+   if (block < size) {
+      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. */
+      anv_block_pool_grow(pool);
+      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;
+   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;
+
+   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 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 != 1) {
+      sb = stream->block_pool->map + block;
+      next_block = sb->next;
+      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);
+      sb = stream->block_pool->map + block;
+      sb->next = stream->current_block;
+      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);
+   }
+
+   stream->next = state.offset + size;
+
+   state.alloc_size = size;
+   state.map = stream->block_pool->map + state.offset;
+
+   return state;
+}

src/vulkan/aub.c

@@ -0,0 +1,292 @@
+/*
+ * 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 anv_reloc_list *list, 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 < list->num_relocs; i++) {
+      reloc = &list->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, length;
+   struct aub_bo *aub_bos;
+
+   writer = get_anv_aub_writer(device);
+   if (writer == NULL)
+      return;
+
+   aub_bos = malloc(cmd_buffer->bo_count * sizeof(aub_bos[0]));
+   offset = writer->offset;
+   for (uint32_t i = 0; i < cmd_buffer->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);
+   }
+
+   relocate_bo(&batch->bo, &batch->cmd_relocs, aub_bos);
+   relocate_bo(&device->surface_state_block_pool.bo,
+               &batch->surf_relocs, aub_bos);
+
+   for (uint32_t i = 0; i < cmd_buffer->bo_count; i++) {
+      bo = cmd_buffer->exec2_bos[i];
+      if (i == cmd_buffer->bo_count - 1) {
+         length = batch->next - batch->bo.map;
+         aub_write_trace_block(writer, AUB_TRACE_TYPE_BATCH,
+                               aub_bos[i].relocated,
+                               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[batch->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/compiler.cpp

@@ -0,0 +1,931 @@
+/*
+ * 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 <sys/stat.h>
+#include <unistd.h>
+#include <fcntl.h>
+
+#include <brw_context.h>
+#include <brw_wm.h> /* brw_new_shader_program is here */
+
+#include <brw_vs.h>
+#include <brw_gs.h>
+
+#include <mesa/main/shaderobj.h>
+#include <mesa/main/fbobject.h>
+#include <mesa/program/program.h>
+#include <glsl/program.h>
+
+#include "private.h"
+
+static void
+fail_if(int cond, const char *format, ...)
+{
+   va_list args;
+
+   if (!cond)
+      return;
+
+   va_start(args, format);
+   vfprintf(stderr, format, args);
+   va_end(args);
+
+   exit(1);
+}
+
+static VkResult
+set_binding_table_layout(struct brw_stage_prog_data *prog_data,
+                         struct anv_pipeline *pipeline, uint32_t stage)
+{
+   uint32_t count, bias, set, *map;
+
+   struct anv_pipeline_layout_entry *entries;
+
+   if (stage == VK_SHADER_STAGE_FRAGMENT)
+      bias = MAX_RTS;
+   else
+      bias = 0;
+
+   count = pipeline->layout->stage[stage].count;
+   entries = pipeline->layout->stage[stage].entries;
+
+   prog_data->map_entries =
+      (uint32_t *) malloc(count * sizeof(prog_data->map_entries[0]));
+   if (prog_data->map_entries == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   set = 0;
+   map = prog_data->map_entries;
+   for (uint32_t i = 0; i < count; i++) {
+      if (entries[i].set == set) {
+         prog_data->bind_map[set] = map;
+         set++;
+      }
+      *map++ = bias + i;
+   }
+
+   return VK_SUCCESS;
+}
+
+static void
+brw_vs_populate_key(struct brw_context *brw,
+                    struct brw_vertex_program *vp,
+                    struct brw_vs_prog_key *key)
+{
+   struct gl_context *ctx = &brw->ctx;
+   /* BRW_NEW_VERTEX_PROGRAM */
+   struct gl_program *prog = (struct gl_program *) vp;
+
+   memset(key, 0, sizeof(*key));
+
+   /* Just upload the program verbatim for now.  Always send it all
+    * the inputs it asks for, whether they are varying or not.
+    */
+   key->base.program_string_id = vp->id;
+   brw_setup_vue_key_clip_info(brw, &key->base,
+                               vp->program.Base.UsesClipDistanceOut);
+
+   /* _NEW_POLYGON */
+   if (brw->gen < 6) {
+      key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
+                           ctx->Polygon.BackMode != GL_FILL);
+   }
+
+   if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
+                               VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
+      /* _NEW_LIGHT | _NEW_BUFFERS */
+      key->clamp_vertex_color = ctx->Light._ClampVertexColor;
+   }
+
+   /* _NEW_POINT */
+   if (brw->gen < 6 && ctx->Point.PointSprite) {
+      for (int i = 0; i < 8; i++) {
+         if (ctx->Point.CoordReplace[i])
+            key->point_coord_replace |= (1 << i);
+      }
+   }
+
+   /* _NEW_TEXTURE */
+   brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
+                                      &key->base.tex);
+}
+
+static bool
+really_do_vs_prog(struct brw_context *brw,
+                  struct gl_shader_program *prog,
+                  struct brw_vertex_program *vp,
+                  struct brw_vs_prog_key *key, struct anv_pipeline *pipeline)
+{
+   GLuint program_size;
+   const GLuint *program;
+   struct brw_vs_compile c;
+   struct brw_vs_prog_data *prog_data = &pipeline->vs_prog_data;
+   struct brw_stage_prog_data *stage_prog_data = &prog_data->base.base;
+   void *mem_ctx;
+   struct gl_shader *vs = NULL;
+
+   if (prog)
+      vs = prog->_LinkedShaders[MESA_SHADER_VERTEX];
+
+   memset(&c, 0, sizeof(c));
+   memcpy(&c.key, key, sizeof(*key));
+   memset(prog_data, 0, sizeof(*prog_data));
+
+   mem_ctx = ralloc_context(NULL);
+
+   c.vp = vp;
+
+   /* Allocate the references to the uniforms that will end up in the
+    * prog_data associated with the compiled program, and which will be freed
+    * by the state cache.
+    */
+   int param_count;
+   if (vs) {
+      /* We add padding around uniform values below vec4 size, with the worst
+       * case being a float value that gets blown up to a vec4, so be
+       * conservative here.
+       */
+      param_count = vs->num_uniform_components * 4;
+
+   } else {
+      param_count = vp->program.Base.Parameters->NumParameters * 4;
+   }
+   /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
+    * planes as uniforms.
+    */
+   param_count += c.key.base.nr_userclip_plane_consts * 4;
+
+   /* Setting nr_params here NOT to the size of the param and pull_param
+    * arrays, but to the number of uniform components vec4_visitor
+    * needs. vec4_visitor::setup_uniforms() will set it back to a proper value.
+    */
+   stage_prog_data->nr_params = ALIGN(param_count, 4) / 4;
+   if (vs) {
+      stage_prog_data->nr_params += vs->num_samplers;
+   }
+
+   GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
+   prog_data->inputs_read = vp->program.Base.InputsRead;
+
+   if (c.key.copy_edgeflag) {
+      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
+      prog_data->inputs_read |= VERT_BIT_EDGEFLAG;
+   }
+
+   if (brw->gen < 6) {
+      /* Put dummy slots into the VUE for the SF to put the replaced
+       * point sprite coords in.  We shouldn't need these dummy slots,
+       * which take up precious URB space, but it would mean that the SF
+       * doesn't get nice aligned pairs of input coords into output
+       * coords, which would be a pain to handle.
+       */
+      for (int i = 0; i < 8; i++) {
+         if (c.key.point_coord_replace & (1 << i))
+            outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
+      }
+
+      /* if back colors are written, allocate slots for front colors too */
+      if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
+         outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
+      if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
+         outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
+   }
+
+   /* In order for legacy clipping to work, we need to populate the clip
+    * distance varying slots whenever clipping is enabled, even if the vertex
+    * shader doesn't write to gl_ClipDistance.
+    */
+   if (c.key.base.userclip_active) {
+      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
+      outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
+   }
+
+   brw_compute_vue_map(brw->intelScreen->devinfo,
+                       &prog_data->base.vue_map, outputs_written);
+\
+   set_binding_table_layout(&prog_data->base.base, pipeline,
+                            VK_SHADER_STAGE_VERTEX);
+
+   /* Emit GEN4 code.
+    */
+   program = brw_vs_emit(brw, prog, &c, prog_data, mem_ctx, &program_size);
+   if (program == NULL) {
+      ralloc_free(mem_ctx);
+      return false;
+   }
+
+   pipeline->vs_simd8 = pipeline->program_next;
+   memcpy((char *) pipeline->device->instruction_block_pool.map +
+          pipeline->vs_simd8, program, program_size);
+
+   pipeline->program_next = align(pipeline->program_next + program_size, 64);
+
+   ralloc_free(mem_ctx);
+
+   if (stage_prog_data->total_scratch > 0)
+      if (!anv_bo_init_new(&pipeline->vs_scratch_bo,
+                           pipeline->device,
+                           stage_prog_data->total_scratch))
+         return false;
+
+
+   return true;
+}
+
+void brw_wm_populate_key(struct brw_context *brw,
+                         struct brw_fragment_program *fp,
+                         struct brw_wm_prog_key *key)
+{
+   struct gl_context *ctx = &brw->ctx;
+   struct gl_program *prog = (struct gl_program *) brw->fragment_program;
+   GLuint lookup = 0;
+   GLuint line_aa;
+   bool program_uses_dfdy = fp->program.UsesDFdy;
+   struct gl_framebuffer draw_buffer;
+   bool multisample_fbo;
+
+   memset(key, 0, sizeof(*key));
+
+   for (int i = 0; i < MAX_SAMPLERS; i++) {
+      /* Assume color sampler, no swizzling. */
+      key->tex.swizzles[i] = SWIZZLE_XYZW;
+   }
+
+   /* A non-zero framebuffer name indicates that the framebuffer was created by
+    * the user rather than the window system. */
+   draw_buffer.Name = 1;
+   draw_buffer.Visual.samples = 1;
+   draw_buffer._NumColorDrawBuffers = 1;
+   draw_buffer._NumColorDrawBuffers = 1;
+   draw_buffer.Width = 400;
+   draw_buffer.Height = 400;
+   ctx->DrawBuffer = &draw_buffer;
+
+   multisample_fbo = ctx->DrawBuffer->Visual.samples > 1;
+
+   /* Build the index for table lookup
+    */
+   if (brw->gen < 6) {
+      /* _NEW_COLOR */
+      if (fp->program.UsesKill || ctx->Color.AlphaEnabled)
+         lookup |= IZ_PS_KILL_ALPHATEST_BIT;
+
+      if (fp->program.Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH))
+         lookup |= IZ_PS_COMPUTES_DEPTH_BIT;
+
+      /* _NEW_DEPTH */
+      if (ctx->Depth.Test)
+         lookup |= IZ_DEPTH_TEST_ENABLE_BIT;
+
+      if (ctx->Depth.Test && ctx->Depth.Mask) /* ?? */
+         lookup |= IZ_DEPTH_WRITE_ENABLE_BIT;
+
+      /* _NEW_STENCIL | _NEW_BUFFERS */
+      if (ctx->Stencil._Enabled) {
+         lookup |= IZ_STENCIL_TEST_ENABLE_BIT;
+
+         if (ctx->Stencil.WriteMask[0] ||
+             ctx->Stencil.WriteMask[ctx->Stencil._BackFace])
+            lookup |= IZ_STENCIL_WRITE_ENABLE_BIT;
+      }
+      key->iz_lookup = lookup;
+   }
+
+   line_aa = AA_NEVER;
+
+   /* _NEW_LINE, _NEW_POLYGON, BRW_NEW_REDUCED_PRIMITIVE */
+   if (ctx->Line.SmoothFlag) {
+      if (brw->reduced_primitive == GL_LINES) {
+         line_aa = AA_ALWAYS;
+      }
+      else if (brw->reduced_primitive == GL_TRIANGLES) {
+         if (ctx->Polygon.FrontMode == GL_LINE) {
+            line_aa = AA_SOMETIMES;
+
+            if (ctx->Polygon.BackMode == GL_LINE ||
+                (ctx->Polygon.CullFlag &&
+                 ctx->Polygon.CullFaceMode == GL_BACK))
+               line_aa = AA_ALWAYS;
+         }
+         else if (ctx->Polygon.BackMode == GL_LINE) {
+            line_aa = AA_SOMETIMES;
+
+            if ((ctx->Polygon.CullFlag &&
+                 ctx->Polygon.CullFaceMode == GL_FRONT))
+               line_aa = AA_ALWAYS;
+         }
+      }
+   }
+
+   key->line_aa = line_aa;
+
+   /* _NEW_HINT */
+   key->high_quality_derivatives =
+      ctx->Hint.FragmentShaderDerivative == GL_NICEST;
+
+   if (brw->gen < 6)
+      key->stats_wm = brw->stats_wm;
+
+   /* _NEW_LIGHT */
+   key->flat_shade = (ctx->Light.ShadeModel == GL_FLAT);
+
+   /* _NEW_FRAG_CLAMP | _NEW_BUFFERS */
+   key->clamp_fragment_color = ctx->Color._ClampFragmentColor;
+
+   /* _NEW_TEXTURE */
+   brw_populate_sampler_prog_key_data(ctx, prog, brw->wm.base.sampler_count,
+                                      &key->tex);
+
+   /* _NEW_BUFFERS */
+   /*
+    * Include the draw buffer origin and height so that we can calculate
+    * fragment position values relative to the bottom left of the drawable,
+    * from the incoming screen origin relative position we get as part of our
+    * payload.
+    *
+    * This is only needed for the WM_WPOSXY opcode when the fragment program
+    * uses the gl_FragCoord input.
+    *
+    * We could avoid recompiling by including this as a constant referenced by
+    * our program, but if we were to do that it would also be nice to handle
+    * getting that constant updated at batchbuffer submit time (when we
+    * hold the lock and know where the buffer really is) rather than at emit
+    * time when we don't hold the lock and are just guessing.  We could also
+    * just avoid using this as key data if the program doesn't use
+    * fragment.position.
+    *
+    * For DRI2 the origin_x/y will always be (0,0) but we still need the
+    * drawable height in order to invert the Y axis.
+    */
+   if (fp->program.Base.InputsRead & VARYING_BIT_POS) {
+      key->drawable_height = ctx->DrawBuffer->Height;
+   }
+
+   if ((fp->program.Base.InputsRead & VARYING_BIT_POS) || program_uses_dfdy) {
+      key->render_to_fbo = _mesa_is_user_fbo(ctx->DrawBuffer);
+   }
+
+   /* _NEW_BUFFERS */
+   key->nr_color_regions = ctx->DrawBuffer->_NumColorDrawBuffers;
+
+   /* _NEW_MULTISAMPLE, _NEW_COLOR, _NEW_BUFFERS */
+   key->replicate_alpha = ctx->DrawBuffer->_NumColorDrawBuffers > 1 &&
+      (ctx->Multisample.SampleAlphaToCoverage || ctx->Color.AlphaEnabled);
+
+   /* _NEW_BUFFERS _NEW_MULTISAMPLE */
+   /* Ignore sample qualifier while computing this flag. */
+   key->persample_shading =
+      _mesa_get_min_invocations_per_fragment(ctx, &fp->program, true) > 1;
+   if (key->persample_shading)
+      key->persample_2x = ctx->DrawBuffer->Visual.samples == 2;
+
+   key->compute_pos_offset =
+      _mesa_get_min_invocations_per_fragment(ctx, &fp->program, false) > 1 &&
+      fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_POS;
+
+   key->compute_sample_id =
+      multisample_fbo &&
+      ctx->Multisample.Enabled &&
+      (fp->program.Base.SystemValuesRead & SYSTEM_BIT_SAMPLE_ID);
+
+   /* BRW_NEW_VUE_MAP_GEOM_OUT */
+   if (brw->gen < 6 || _mesa_bitcount_64(fp->program.Base.InputsRead &
+                                         BRW_FS_VARYING_INPUT_MASK) > 16)
+      key->input_slots_valid = brw->vue_map_geom_out.slots_valid;
+
+
+   /* _NEW_COLOR | _NEW_BUFFERS */
+   /* Pre-gen6, the hardware alpha test always used each render
+    * target's alpha to do alpha test, as opposed to render target 0's alpha
+    * like GL requires.  Fix that by building the alpha test into the
+    * shader, and we'll skip enabling the fixed function alpha test.
+    */
+   if (brw->gen < 6 && ctx->DrawBuffer->_NumColorDrawBuffers > 1 && ctx->Color.AlphaEnabled) {
+      key->alpha_test_func = ctx->Color.AlphaFunc;
+      key->alpha_test_ref = ctx->Color.AlphaRef;
+   }
+
+   /* The unique fragment program ID */
+   key->program_string_id = fp->id;
+
+   ctx->DrawBuffer = NULL;
+}
+
+static uint8_t
+computed_depth_mode(struct gl_fragment_program *fp)
+{
+   if (fp->Base.OutputsWritten & BITFIELD64_BIT(FRAG_RESULT_DEPTH)) {
+      switch (fp->FragDepthLayout) {
+      case FRAG_DEPTH_LAYOUT_NONE:
+      case FRAG_DEPTH_LAYOUT_ANY:
+         return BRW_PSCDEPTH_ON;
+      case FRAG_DEPTH_LAYOUT_GREATER:
+         return BRW_PSCDEPTH_ON_GE;
+      case FRAG_DEPTH_LAYOUT_LESS:
+         return BRW_PSCDEPTH_ON_LE;
+      case FRAG_DEPTH_LAYOUT_UNCHANGED:
+         return BRW_PSCDEPTH_OFF;
+      }
+   }
+   return BRW_PSCDEPTH_OFF;
+}
+
+static bool
+really_do_wm_prog(struct brw_context *brw,
+                  struct gl_shader_program *prog,
+                  struct brw_fragment_program *fp,
+                  struct brw_wm_prog_key *key, struct anv_pipeline *pipeline)
+{
+   struct gl_context *ctx = &brw->ctx;
+   void *mem_ctx = ralloc_context(NULL);
+   struct brw_wm_prog_data *prog_data = &pipeline->wm_prog_data;
+   struct gl_shader *fs = NULL;
+   unsigned int program_size;
+   const uint32_t *program;
+   uint32_t offset;
+
+   if (prog)
+      fs = prog->_LinkedShaders[MESA_SHADER_FRAGMENT];
+
+   memset(prog_data, 0, sizeof(*prog_data));
+
+   /* key->alpha_test_func means simulating alpha testing via discards,
+    * so the shader definitely kills pixels.
+    */
+   prog_data->uses_kill = fp->program.UsesKill || key->alpha_test_func;
+
+   prog_data->computed_depth_mode = computed_depth_mode(&fp->program);
+
+   /* Allocate the references to the uniforms that will end up in the
+    * prog_data associated with the compiled program, and which will be freed
+    * by the state cache.
+    */
+   int param_count;
+   if (fs) {
+      param_count = fs->num_uniform_components;
+   } else {
+      param_count = fp->program.Base.Parameters->NumParameters * 4;
+   }
+   /* The backend also sometimes adds params for texture size. */
+   param_count += 2 * ctx->Const.Program[MESA_SHADER_FRAGMENT].MaxTextureImageUnits;
+   prog_data->base.param =
+      rzalloc_array(NULL, const gl_constant_value *, param_count);
+   prog_data->base.pull_param =
+      rzalloc_array(NULL, const gl_constant_value *, param_count);
+   prog_data->base.nr_params = param_count;
+
+   prog_data->barycentric_interp_modes =
+      brw_compute_barycentric_interp_modes(brw, key->flat_shade,
+                                           key->persample_shading,
+                                           &fp->program);
+
+   set_binding_table_layout(&prog_data->base, pipeline,
+                            VK_SHADER_STAGE_FRAGMENT);
+   /* This needs to come after shader time and pull constant entries, but we
+    * don't have those set up now, so just put it after the layout entries.
+    */
+   prog_data->binding_table.render_target_start = 0;
+
+   program = brw_wm_fs_emit(brw, mem_ctx, key, prog_data,
+                            &fp->program, prog, &program_size);
+   if (program == NULL) {
+      ralloc_free(mem_ctx);
+      return false;
+   }
+
+   offset = pipeline->program_next;
+   pipeline->program_next = align(pipeline->program_next + program_size, 64);
+
+   if (prog_data->no_8)
+      pipeline->ps_simd8 = NO_KERNEL;
+   else
+      pipeline->ps_simd8 = offset;
+
+   if (prog_data->no_8 || prog_data->prog_offset_16)
+      pipeline->ps_simd16 = offset + prog_data->prog_offset_16;
+   else
+      pipeline->ps_simd16 = NO_KERNEL;
+
+   memcpy((char *) pipeline->device->instruction_block_pool.map +
+          offset, program, program_size);
+
+   ralloc_free(mem_ctx);
+
+   if (prog_data->base.total_scratch > 0)
+      if (!anv_bo_init_new(&pipeline->ps_scratch_bo,
+                           pipeline->device,
+                           prog_data->base.total_scratch))
+         return false;
+
+   return true;
+}
+
+static void
+brw_gs_populate_key(struct brw_context *brw,
+                    struct anv_pipeline *pipeline,
+                    struct brw_geometry_program *gp,
+                    struct brw_gs_prog_key *key)
+{
+   struct gl_context *ctx = &brw->ctx;
+   struct brw_stage_state *stage_state = &brw->gs.base;
+   struct gl_program *prog = &gp->program.Base;
+
+   memset(key, 0, sizeof(*key));
+
+   key->base.program_string_id = gp->id;
+   brw_setup_vue_key_clip_info(brw, &key->base,
+                               gp->program.Base.UsesClipDistanceOut);
+
+   /* _NEW_TEXTURE */
+   brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count,
+                                      &key->base.tex);
+
+   struct brw_vs_prog_data *prog_data = &pipeline->vs_prog_data;
+
+   /* BRW_NEW_VUE_MAP_VS */
+   key->input_varyings = prog_data->base.vue_map.slots_valid;
+}
+
+static bool
+really_do_gs_prog(struct brw_context *brw,
+                  struct gl_shader_program *prog,
+                  struct brw_geometry_program *gp,
+                  struct brw_gs_prog_key *key, struct anv_pipeline *pipeline)
+{
+   struct brw_gs_compile_output output;
+   uint32_t offset;
+
+   /* FIXME: We pass the bind map to the compile in the output struct. Need
+    * something better. */
+   set_binding_table_layout(&output.prog_data.base.base,
+                            pipeline, VK_SHADER_STAGE_GEOMETRY);
+
+   brw_compile_gs_prog(brw, prog, gp, key, &output);
+
+   offset = pipeline->program_next;
+   pipeline->program_next = align(pipeline->program_next + output.program_size, 64);
+
+   pipeline->gs_vec4 = offset;
+   pipeline->gs_vertex_count = gp->program.VerticesIn;
+
+   memcpy((char *) pipeline->device->instruction_block_pool.map +
+          offset, output.program, output.program_size);
+
+   ralloc_free(output.mem_ctx);
+
+   if (output.prog_data.base.base.total_scratch) {
+      if (!anv_bo_init_new(&pipeline->gs_scratch_bo,
+                           pipeline->device,
+                           output.prog_data.base.base.total_scratch))
+         return false;
+   }
+
+   memcpy(&pipeline->gs_prog_data, &output.prog_data, sizeof pipeline->gs_prog_data);
+
+   return true;
+}
+
+static void
+fail_on_compile_error(int status, const char *msg)
+{
+   int source, line, column;
+   char error[256];
+
+   if (status)
+      return;
+
+   if (sscanf(msg, "%d:%d(%d): error: %255[^\n]", &source, &line, &column, error) == 4)
+      fail_if(!status, "%d:%s\n", line, error);
+   else
+      fail_if(!status, "%s\n", msg);
+}
+
+struct anv_compiler {
+   struct intel_screen *screen;
+   struct brw_context *brw;
+};
+
+
+extern "C" {
+
+struct anv_compiler *
+anv_compiler_create(int fd)
+{
+   struct anv_compiler *compiler;
+
+   compiler = (struct anv_compiler *) malloc(sizeof *compiler);
+   if (compiler == NULL)
+      return NULL;
+
+   compiler->screen = intel_screen_create(fd);
+   if (compiler->screen == NULL) {
+      free(compiler);
+      return NULL;
+   }
+
+   compiler->brw = intel_context_create(compiler->screen);
+   if (compiler->brw == NULL) {
+      free(compiler);
+      return NULL;
+   }
+
+   compiler->brw->precompile = false;
+
+   return compiler;
+}
+
+void
+anv_compiler_destroy(struct anv_compiler *compiler)
+{
+   intel_context_destroy(compiler->brw);
+   intel_screen_destroy(compiler->screen);
+   free(compiler);
+}
+
+/* From gen7_urb.c */
+
+/* FIXME: Add to struct intel_device_info */
+
+static const int gen8_push_size = 32 * 1024;
+
+static void
+gen7_compute_urb_partition(struct anv_pipeline *pipeline)
+{
+   const struct brw_device_info *devinfo = &pipeline->device->info;
+   unsigned vs_size = pipeline->vs_prog_data.base.urb_entry_size;
+   unsigned vs_entry_size_bytes = vs_size * 64;
+   bool gs_present = pipeline->gs_vec4 != NO_KERNEL;
+   unsigned gs_size = gs_present ? pipeline->gs_prog_data.base.urb_entry_size : 1;
+   unsigned gs_entry_size_bytes = gs_size * 64;
+
+   /* From p35 of the Ivy Bridge PRM (section 1.7.1: 3DSTATE_URB_GS):
+    *
+    *     VS Number of URB Entries must be divisible by 8 if the VS URB Entry
+    *     Allocation Size is less than 9 512-bit URB entries.
+    *
+    * Similar text exists for GS.
+    */
+   unsigned vs_granularity = (vs_size < 9) ? 8 : 1;
+   unsigned gs_granularity = (gs_size < 9) ? 8 : 1;
+
+   /* URB allocations must be done in 8k chunks. */
+   unsigned chunk_size_bytes = 8192;
+
+   /* Determine the size of the URB in chunks. */
+   unsigned urb_chunks = devinfo->urb.size * 1024 / chunk_size_bytes;
+
+   /* Reserve space for push constants */
+   unsigned push_constant_bytes = gen8_push_size;
+   unsigned push_constant_chunks =
+      push_constant_bytes / chunk_size_bytes;
+
+   /* Initially, assign each stage the minimum amount of URB space it needs,
+    * and make a note of how much additional space it "wants" (the amount of
+    * additional space it could actually make use of).
+    */
+
+   /* VS has a lower limit on the number of URB entries */
+   unsigned vs_chunks =
+      ALIGN(devinfo->urb.min_vs_entries * vs_entry_size_bytes,
+            chunk_size_bytes) / chunk_size_bytes;
+   unsigned vs_wants =
+      ALIGN(devinfo->urb.max_vs_entries * vs_entry_size_bytes,
+            chunk_size_bytes) / chunk_size_bytes - vs_chunks;
+
+   unsigned gs_chunks = 0;
+   unsigned gs_wants = 0;
+   if (gs_present) {
+      /* There are two constraints on the minimum amount of URB space we can
+       * allocate:
+       *
+       * (1) We need room for at least 2 URB entries, since we always operate
+       * the GS in DUAL_OBJECT mode.
+       *
+       * (2) We can't allocate less than nr_gs_entries_granularity.
+       */
+      gs_chunks = ALIGN(MAX2(gs_granularity, 2) * gs_entry_size_bytes,
+                        chunk_size_bytes) / chunk_size_bytes;
+      gs_wants =
+         ALIGN(devinfo->urb.max_gs_entries * gs_entry_size_bytes,
+               chunk_size_bytes) / chunk_size_bytes - gs_chunks;
+   }
+
+   /* There should always be enough URB space to satisfy the minimum
+    * requirements of each stage.
+    */
+   unsigned total_needs = push_constant_chunks + vs_chunks + gs_chunks;
+   assert(total_needs <= urb_chunks);
+
+   /* Mete out remaining space (if any) in proportion to "wants". */
+   unsigned total_wants = vs_wants + gs_wants;
+   unsigned remaining_space = urb_chunks - total_needs;
+   if (remaining_space > total_wants)
+      remaining_space = total_wants;
+   if (remaining_space > 0) {
+      unsigned vs_additional = (unsigned)
+         round(vs_wants * (((double) remaining_space) / total_wants));
+      vs_chunks += vs_additional;
+      remaining_space -= vs_additional;
+      gs_chunks += remaining_space;
+   }
+
+   /* Sanity check that we haven't over-allocated. */
+   assert(push_constant_chunks + vs_chunks + gs_chunks <= urb_chunks);
+
+   /* Finally, compute the number of entries that can fit in the space
+    * allocated to each stage.
+    */
+   unsigned nr_vs_entries = vs_chunks * chunk_size_bytes / vs_entry_size_bytes;
+   unsigned nr_gs_entries = gs_chunks * chunk_size_bytes / gs_entry_size_bytes;
+
+   /* Since we rounded up when computing *_wants, this may be slightly more
+    * than the maximum allowed amount, so correct for that.
+    */
+   nr_vs_entries = MIN2(nr_vs_entries, devinfo->urb.max_vs_entries);
+   nr_gs_entries = MIN2(nr_gs_entries, devinfo->urb.max_gs_entries);
+
+   /* Ensure that we program a multiple of the granularity. */
+   nr_vs_entries = ROUND_DOWN_TO(nr_vs_entries, vs_granularity);
+   nr_gs_entries = ROUND_DOWN_TO(nr_gs_entries, gs_granularity);
+
+   /* Finally, sanity check to make sure we have at least the minimum number
+    * of entries needed for each stage.
+    */
+   assert(nr_vs_entries >= devinfo->urb.min_vs_entries);
+   if (gs_present)
+      assert(nr_gs_entries >= 2);
+
+   /* Lay out the URB in the following order:
+    * - push constants
+    * - VS
+    * - GS
+    */
+   pipeline->urb.vs_start = push_constant_chunks;
+   pipeline->urb.vs_size = vs_size;
+   pipeline->urb.nr_vs_entries = nr_vs_entries;
+
+   pipeline->urb.gs_start = push_constant_chunks + vs_chunks;
+   pipeline->urb.gs_size = gs_size;
+   pipeline->urb.nr_gs_entries = nr_gs_entries;
+}
+
+static const struct {
+   uint32_t token;
+   const char *name;
+} stage_info[] = {
+   { GL_VERTEX_SHADER, "vertex" },
+   { GL_TESS_CONTROL_SHADER, "tess control" },
+   { GL_TESS_EVALUATION_SHADER, "tess evaluation" },
+   { GL_GEOMETRY_SHADER, "geometry" },
+   { GL_FRAGMENT_SHADER, "fragment" },
+   { GL_COMPUTE_SHADER, "compute" },
+};
+
+static void
+anv_compile_shader(struct anv_compiler *compiler,
+                   struct gl_shader_program *program,
+                   struct anv_pipeline *pipeline, uint32_t stage)
+{
+   struct brw_context *brw = compiler->brw;
+   struct gl_shader *shader;
+   int name = 0;
+
+   shader = brw_new_shader(&brw->ctx, name, stage_info[stage].token);
+   fail_if(shader == NULL, "failed to create %s shader\n", stage_info[stage].name);
+   shader->Source = strdup(pipeline->shaders[stage]->data);
+   _mesa_glsl_compile_shader(&brw->ctx, shader, false, false);
+   fail_on_compile_error(shader->CompileStatus, shader->InfoLog);
+
+   program->Shaders[program->NumShaders] = shader;
+   program->NumShaders++;
+}
+
+int
+anv_compiler_run(struct anv_compiler *compiler, struct anv_pipeline *pipeline)
+{
+   struct gl_shader_program *program;
+   int name = 0;
+   struct brw_context *brw = compiler->brw;
+   struct anv_device *device = pipeline->device;
+
+   brw->use_rep_send = pipeline->use_repclear;
+   brw->no_simd8 = pipeline->use_repclear;
+
+   program = brw->ctx.Driver.NewShaderProgram(name);
+   program->Shaders = (struct gl_shader **)
+      calloc(VK_NUM_SHADER_STAGE, sizeof(struct gl_shader *));
+   fail_if(program == NULL || program->Shaders == NULL,
+           "failed to create program\n");
+
+   /* FIXME: Only supports vs and fs combo at the moment */
+   assert(pipeline->shaders[VK_SHADER_STAGE_VERTEX]);
+   assert(pipeline->shaders[VK_SHADER_STAGE_FRAGMENT]);
+
+   anv_compile_shader(compiler, program, pipeline, VK_SHADER_STAGE_VERTEX);
+   anv_compile_shader(compiler, program, pipeline, VK_SHADER_STAGE_FRAGMENT);
+   if (pipeline->shaders[VK_SHADER_STAGE_GEOMETRY])
+      anv_compile_shader(compiler, program, pipeline, VK_SHADER_STAGE_GEOMETRY);
+
+   _mesa_glsl_link_shader(&brw->ctx, program);
+   fail_on_compile_error(program->LinkStatus,
+                         program->InfoLog);
+
+   pipeline->program_block =
+      anv_block_pool_alloc(&device->instruction_block_pool);
+   pipeline->program_next = pipeline->program_block;
+
+
+   bool success;
+   struct brw_wm_prog_key wm_key;
+   struct gl_fragment_program *fp = (struct gl_fragment_program *)
+      program->_LinkedShaders[MESA_SHADER_FRAGMENT]->Program;
+   struct brw_fragment_program *bfp = brw_fragment_program(fp);
+
+   brw_wm_populate_key(brw, bfp, &wm_key);
+
+   success = really_do_wm_prog(brw, program, bfp, &wm_key, pipeline);
+   fail_if(!success, "do_wm_prog failed\n");
+   pipeline->prog_data[VK_SHADER_STAGE_FRAGMENT] = &pipeline->wm_prog_data.base;
+
+
+   struct brw_vs_prog_key vs_key;
+   struct gl_vertex_program *vp = (struct gl_vertex_program *)
+      program->_LinkedShaders[MESA_SHADER_VERTEX]->Program;
+   struct brw_vertex_program *bvp = brw_vertex_program(vp);
+
+   brw_vs_populate_key(brw, bvp, &vs_key);
+
+   success = really_do_vs_prog(brw, program, bvp, &vs_key, pipeline);
+   fail_if(!success, "do_wm_prog failed\n");
+   pipeline->prog_data[VK_SHADER_STAGE_VERTEX] = &pipeline->vs_prog_data.base.base;
+
+   if (pipeline->shaders[VK_SHADER_STAGE_GEOMETRY]) {
+      struct brw_gs_prog_key gs_key;
+      struct gl_geometry_program *gp = (struct gl_geometry_program *)
+         program->_LinkedShaders[MESA_SHADER_GEOMETRY]->Program;
+      struct brw_geometry_program *bgp = brw_geometry_program(gp);
+
+      brw_gs_populate_key(brw, pipeline, bgp, &gs_key);
+
+      success = really_do_gs_prog(brw, program, bgp, &gs_key, pipeline);
+      fail_if(!success, "do_gs_prog failed\n");
+      pipeline->active_stages = VK_SHADER_STAGE_VERTEX_BIT |
+         VK_SHADER_STAGE_GEOMETRY_BIT | VK_SHADER_STAGE_FRAGMENT_BIT;
+      pipeline->prog_data[VK_SHADER_STAGE_GEOMETRY] = &pipeline->gs_prog_data.base.base;
+
+   } else {
+      pipeline->gs_vec4 = NO_KERNEL;
+      pipeline->active_stages =
+         VK_SHADER_STAGE_VERTEX_BIT | VK_SHADER_STAGE_FRAGMENT_BIT;
+   }
+
+
+   /* FIXME: Allocate more blocks if we fill up this one and worst case,
+    * allocate multiple continuous blocks from end of pool to hold really big
+    * programs. */
+   assert(pipeline->program_next - pipeline->program_block < 8192);
+
+   brw->ctx.Driver.DeleteShaderProgram(&brw->ctx, program);
+
+   gen7_compute_urb_partition(pipeline);
+
+   return 0;
+}
+
+/* This badly named function frees the struct anv_pipeline data that the compiler
+ * allocates.  Currently just the prog_data structs.
+ */
+void
+anv_compiler_free(struct anv_pipeline *pipeline)
+{
+   struct anv_device *device = pipeline->device;
+
+   for (uint32_t stage = 0; stage < VK_NUM_SHADER_STAGE; stage++)
+      if (pipeline->prog_data[stage])
+         free(pipeline->prog_data[stage]->map_entries);
+
+   anv_block_pool_free(&device->instruction_block_pool,
+                       pipeline->program_block);
+}
+
+}

src/vulkan/gem.c

@@ -0,0 +1,283 @@
+/*
+ * 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"
+
+#ifdef HAVE_VALGRIND
+#include <valgrind.h>
+#include <memcheck.h>
+#define VG(x) x
+#else
+#define VG(x)
+#endif
+
+#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;
+
+   VG_CLEAR(gem_mmap);
+   gem_mmap.handle = gem_handle;
+   gem_mmap.offset = offset;
+   gem_mmap.size = size;
+   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)
+{
+   munmap(p, size);
+   VG(VALGRIND_FREELIKE_BLOCK(p, 0));
+}
+
+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;
+   if (ret == -1)
+      return -errno;
+
+   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_device *device, uint64_t *size)
+{
+   struct drm_i915_gem_get_aperture aperture;
+   int ret;
+
+   VG_CLEAR(aperture);
+   ret = anv_ioctl(device->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/image.c

@@ -0,0 +1,404 @@
+/*
+ * 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"
+
+// Image functions
+
+static const struct anv_format anv_formats[] = {
+   [VK_FORMAT_UNDEFINED] = { .format = RAW },
+   // [VK_FORMAT_R4G4_UNORM] = { .format = R4G4_UNORM },
+   // [VK_FORMAT_R4G4_USCALED] = { .format = R4G4_USCALED },
+   // [VK_FORMAT_R4G4B4A4_UNORM] = { .format = R4G4B4A4_UNORM },
+   // [VK_FORMAT_R4G4B4A4_USCALED] = { .format = R4G4B4A4_USCALED },
+   // [VK_FORMAT_R5G6B5_UNORM] = { .format = R5G6B5_UNORM },
+   // [VK_FORMAT_R5G6B5_USCALED] = { .format = R5G6B5_USCALED },
+   // [VK_FORMAT_R5G5B5A1_UNORM] = { .format = R5G5B5A1_UNORM },
+   // [VK_FORMAT_R5G5B5A1_USCALED] = { .format = R5G5B5A1_USCALED },
+   [VK_FORMAT_R8_UNORM] = { .format = R8_UNORM, .cpp = 1, .channels = 1 },
+   [VK_FORMAT_R8_SNORM] = { .format = R8_SNORM, .cpp = 1, .channels = 1 },
+   [VK_FORMAT_R8_USCALED] = { .format = R8_USCALED, .cpp = 1, .channels = 1 },
+   [VK_FORMAT_R8_SSCALED] = { .format = R8_SSCALED, .cpp = 1, .channels = 1 },
+   [VK_FORMAT_R8_UINT] = { .format = R8_UINT, .cpp = 1, .channels = 1 },
+   [VK_FORMAT_R8_SINT] = { .format = R8_SINT, .cpp = 1, .channels = 1 },
+   // [VK_FORMAT_R8_SRGB] = { .format = R8_SRGB, .cpp = 1 },
+   [VK_FORMAT_R8G8_UNORM] = { .format = R8G8_UNORM, .cpp = 2, .channels = 2 },
+   [VK_FORMAT_R8G8_SNORM] = { .format = R8G8_SNORM, .cpp = 2, .channels = 2 },
+   [VK_FORMAT_R8G8_USCALED] = { .format = R8G8_USCALED, .cpp = 2, .channels = 2 },
+   [VK_FORMAT_R8G8_SSCALED] = { .format = R8G8_SSCALED, .cpp = 2, .channels = 2 },
+   [VK_FORMAT_R8G8_UINT] = { .format = R8G8_UINT, .cpp = 2, .channels = 2 },
+   [VK_FORMAT_R8G8_SINT] = { .format = R8G8_SINT, .cpp = 2, .channels = 2 },
+   // [VK_FORMAT_R8G8_SRGB] = { .format = R8G8_SRGB },
+   [VK_FORMAT_R8G8B8_UNORM] = { .format = R8G8B8X8_UNORM, .cpp = 3, .channels = 3  },
+   // [VK_FORMAT_R8G8B8_SNORM] = { .format = R8G8B8X8_SNORM, .cpp = 4 },
+   [VK_FORMAT_R8G8B8_USCALED] = { .format = R8G8B8_USCALED, .cpp = 3, .channels = 3 },
+   [VK_FORMAT_R8G8B8_SSCALED] = { .format = R8G8B8_SSCALED, .cpp = 3, .channels = 3 },
+   [VK_FORMAT_R8G8B8_UINT] = { .format = R8G8B8_UINT, .cpp = 3, .channels = 3 },
+   [VK_FORMAT_R8G8B8_SINT] = { .format = R8G8B8_SINT, .cpp = 3, .channels = 3 },
+   // [VK_FORMAT_R8G8B8_SRGB] = { .format = R8G8B8_SRGB },
+   [VK_FORMAT_R8G8B8A8_UNORM] = { .format = R8G8B8A8_UNORM, .cpp = 4, .channels = 4 },
+   [VK_FORMAT_R8G8B8A8_SNORM] = { .format = R8G8B8A8_SNORM, .cpp = 4, .channels = 4 },
+   [VK_FORMAT_R8G8B8A8_USCALED] = { .format = R8G8B8A8_USCALED, .cpp = 4, .channels = 4 },
+   [VK_FORMAT_R8G8B8A8_SSCALED] = { .format = R8G8B8A8_SSCALED, .cpp = 4, .channels = 4 },
+   [VK_FORMAT_R8G8B8A8_UINT] = { .format = R8G8B8A8_UINT, .cpp = 4, .channels = 4 },
+   [VK_FORMAT_R8G8B8A8_SINT] = { .format = R8G8B8A8_SINT, .cpp = 4, .channels = 4 },
+   // [VK_FORMAT_R8G8B8A8_SRGB] = { .format = R8G8B8A8_SRGB },
+   // [VK_FORMAT_R10G10B10A2_UNORM] = { .format = R10G10B10A2_UNORM },
+   // [VK_FORMAT_R10G10B10A2_SNORM] = { .format = R10G10B10A2_SNORM },
+   // [VK_FORMAT_R10G10B10A2_USCALED] = { .format = R10G10B10A2_USCALED },
+   // [VK_FORMAT_R10G10B10A2_SSCALED] = { .format = R10G10B10A2_SSCALED },
+   // [VK_FORMAT_R10G10B10A2_UINT] = { .format = R10G10B10A2_UINT },
+   // [VK_FORMAT_R10G10B10A2_SINT] = { .format = R10G10B10A2_SINT },
+   // [VK_FORMAT_R16_UNORM] = { .format = R16_UNORM },
+   // [VK_FORMAT_R16_SNORM] = { .format = R16_SNORM },
+   // [VK_FORMAT_R16_USCALED] = { .format = R16_USCALED },
+   // [VK_FORMAT_R16_SSCALED] = { .format = R16_SSCALED },
+   // [VK_FORMAT_R16_UINT] = { .format = R16_UINT },
+   // [VK_FORMAT_R16_SINT] = { .format = R16_SINT },
+   [VK_FORMAT_R16_SFLOAT] = { .format = R16_FLOAT, .cpp = 2, .channels = 1 },
+   // [VK_FORMAT_R16G16_UNORM] = { .format = R16G16_UNORM },
+   // [VK_FORMAT_R16G16_SNORM] = { .format = R16G16_SNORM },
+   // [VK_FORMAT_R16G16_USCALED] = { .format = R16G16_USCALED },
+   // [VK_FORMAT_R16G16_SSCALED] = { .format = R16G16_SSCALED },
+   // [VK_FORMAT_R16G16_UINT] = { .format = R16G16_UINT },
+   // [VK_FORMAT_R16G16_SINT] = { .format = R16G16_SINT },
+   [VK_FORMAT_R16G16_SFLOAT] = { .format = R16G16_FLOAT, .cpp = 4, .channels = 2 },
+   // [VK_FORMAT_R16G16B16_UNORM] = { .format = R16G16B16_UNORM },
+   // [VK_FORMAT_R16G16B16_SNORM] = { .format = R16G16B16_SNORM },
+   // [VK_FORMAT_R16G16B16_USCALED] = { .format = R16G16B16_USCALED },
+   // [VK_FORMAT_R16G16B16_SSCALED] = { .format = R16G16B16_SSCALED },
+   // [VK_FORMAT_R16G16B16_UINT] = { .format = R16G16B16_UINT },
+   // [VK_FORMAT_R16G16B16_SINT] = { .format = R16G16B16_SINT },
+   [VK_FORMAT_R16G16B16_SFLOAT] = { .format = R16G16B16_FLOAT, .cpp = 6, .channels = 3 },
+   // [VK_FORMAT_R16G16B16A16_UNORM] = { .format = R16G16B16A16_UNORM },
+   // [VK_FORMAT_R16G16B16A16_SNORM] = { .format = R16G16B16A16_SNORM },
+   // [VK_FORMAT_R16G16B16A16_USCALED] = { .format = R16G16B16A16_USCALED },
+   // [VK_FORMAT_R16G16B16A16_SSCALED] = { .format = R16G16B16A16_SSCALED },
+   // [VK_FORMAT_R16G16B16A16_UINT] = { .format = R16G16B16A16_UINT },
+   // [VK_FORMAT_R16G16B16A16_SINT] = { .format = R16G16B16A16_SINT },
+   [VK_FORMAT_R16G16B16A16_SFLOAT] = { .format = R16G16B16A16_FLOAT, .cpp = 8, .channels = 4 },
+   // [VK_FORMAT_R32_UINT] = { .format = R32_UINT },
+   // [VK_FORMAT_R32_SINT] = { .format = R32_SINT },
+   [VK_FORMAT_R32_SFLOAT] = { .format = R32_FLOAT, .cpp = 4, .channels = 1 },
+   // [VK_FORMAT_R32G32_UINT] = { .format = R32G32_UINT },
+   // [VK_FORMAT_R32G32_SINT] = { .format = R32G32_SINT },
+   [VK_FORMAT_R32G32_SFLOAT] = { .format = R32G32_FLOAT, .cpp = 8, .channels = 2 },
+   // [VK_FORMAT_R32G32B32_UINT] = { .format = R32G32B32_UINT },
+   // [VK_FORMAT_R32G32B32_SINT] = { .format = R32G32B32_SINT },
+   [VK_FORMAT_R32G32B32_SFLOAT] = { .format = R32G32B32_FLOAT, .cpp = 12, .channels = 3 },
+   // [VK_FORMAT_R32G32B32A32_UINT] = { .format = R32G32B32A32_UINT },
+   // [VK_FORMAT_R32G32B32A32_SINT] = { .format = R32G32B32A32_SINT },
+   [VK_FORMAT_R32G32B32A32_SFLOAT] = { .format = R32G32B32A32_FLOAT, .cpp = 16, .channels = 4 },
+   [VK_FORMAT_R64_SFLOAT] = { .format = R64_FLOAT, .cpp = 8, .channels = 1 },
+   [VK_FORMAT_R64G64_SFLOAT] = { .format = R64G64_FLOAT, .cpp = 16, .channels = 2 },
+   [VK_FORMAT_R64G64B64_SFLOAT] = { .format = R64G64B64_FLOAT, .cpp = 24, .channels = 3 },
+   [VK_FORMAT_R64G64B64A64_SFLOAT] = { .format = R64G64B64A64_FLOAT, .cpp = 32, .channels = 4 },
+   // [VK_FORMAT_R11G11B10_UFLOAT] = { .format = R11G11B10_UFLOAT },
+   // [VK_FORMAT_R9G9B9E5_UFLOAT] = { .format = R9G9B9E5_UFLOAT },
+   // [VK_FORMAT_D16_UNORM] = { .format = D16_UNORM },
+   // [VK_FORMAT_D24_UNORM] = { .format = D24_UNORM },
+   // [VK_FORMAT_D32_SFLOAT] = { .format = D32_SFLOAT },
+   // [VK_FORMAT_S8_UINT] = { .format = S8_UINT },
+   // [VK_FORMAT_D16_UNORM_S8_UINT] = { .format = D16_UNORM },
+   // [VK_FORMAT_D24_UNORM_S8_UINT] = { .format = D24_UNORM },
+   // [VK_FORMAT_D32_SFLOAT_S8_UINT] = { .format = D32_SFLOAT },
+   // [VK_FORMAT_BC1_RGB_UNORM] = { .format = BC1_RGB },
+   // [VK_FORMAT_BC1_RGB_SRGB] = { .format = BC1_RGB },
+   // [VK_FORMAT_BC1_RGBA_UNORM] = { .format = BC1_RGBA },
+   // [VK_FORMAT_BC1_RGBA_SRGB] = { .format = BC1_RGBA },
+   // [VK_FORMAT_BC2_UNORM] = { .format = BC2_UNORM },
+   // [VK_FORMAT_BC2_SRGB] = { .format = BC2_SRGB },
+   // [VK_FORMAT_BC3_UNORM] = { .format = BC3_UNORM },
+   // [VK_FORMAT_BC3_SRGB] = { .format = BC3_SRGB },
+   // [VK_FORMAT_BC4_UNORM] = { .format = BC4_UNORM },
+   // [VK_FORMAT_BC4_SNORM] = { .format = BC4_SNORM },
+   // [VK_FORMAT_BC5_UNORM] = { .format = BC5_UNORM },
+   // [VK_FORMAT_BC5_SNORM] = { .format = BC5_SNORM },
+   // [VK_FORMAT_BC6H_UFLOAT] = { .format = BC6H_UFLOAT },
+   // [VK_FORMAT_BC6H_SFLOAT] = { .format = BC6H_SFLOAT },
+   // [VK_FORMAT_BC7_UNORM] = { .format = BC7_UNORM },
+   // [VK_FORMAT_BC7_SRGB] = { .format = BC7_SRGB },
+   // [VK_FORMAT_ETC2_R8G8B8_UNORM] = { .format = ETC2_R8G8B8 },
+   // [VK_FORMAT_ETC2_R8G8B8_SRGB] = { .format = ETC2_R8G8B8 },
+   // [VK_FORMAT_ETC2_R8G8B8A1_UNORM] = { .format = ETC2_R8G8B8A1 },
+   // [VK_FORMAT_ETC2_R8G8B8A1_SRGB] = { .format = ETC2_R8G8B8A1 },
+   // [VK_FORMAT_ETC2_R8G8B8A8_UNORM] = { .format = ETC2_R8G8B8A8 },
+   // [VK_FORMAT_ETC2_R8G8B8A8_SRGB] = { .format = ETC2_R8G8B8A8 },
+   // [VK_FORMAT_EAC_R11_UNORM] = { .format = EAC_R11 },
+   // [VK_FORMAT_EAC_R11_SNORM] = { .format = EAC_R11 },
+   // [VK_FORMAT_EAC_R11G11_UNORM] = { .format = EAC_R11G11 },
+   // [VK_FORMAT_EAC_R11G11_SNORM] = { .format = EAC_R11G11 },
+   // [VK_FORMAT_ASTC_4x4_UNORM] = { .format = ASTC_4x4 },
+   // [VK_FORMAT_ASTC_4x4_SRGB] = { .format = ASTC_4x4 },
+   // [VK_FORMAT_ASTC_5x4_UNORM] = { .format = ASTC_5x4 },
+   // [VK_FORMAT_ASTC_5x4_SRGB] = { .format = ASTC_5x4 },
+   // [VK_FORMAT_ASTC_5x5_UNORM] = { .format = ASTC_5x5 },
+   // [VK_FORMAT_ASTC_5x5_SRGB] = { .format = ASTC_5x5 },
+   // [VK_FORMAT_ASTC_6x5_UNORM] = { .format = ASTC_6x5 },
+   // [VK_FORMAT_ASTC_6x5_SRGB] = { .format = ASTC_6x5 },
+   // [VK_FORMAT_ASTC_6x6_UNORM] = { .format = ASTC_6x6 },
+   // [VK_FORMAT_ASTC_6x6_SRGB] = { .format = ASTC_6x6 },
+   // [VK_FORMAT_ASTC_8x5_UNORM] = { .format = ASTC_8x5 },
+   // [VK_FORMAT_ASTC_8x5_SRGB] = { .format = ASTC_8x5 },
+   // [VK_FORMAT_ASTC_8x6_UNORM] = { .format = ASTC_8x6 },
+   // [VK_FORMAT_ASTC_8x6_SRGB] = { .format = ASTC_8x6 },
+   // [VK_FORMAT_ASTC_8x8_UNORM] = { .format = ASTC_8x8 },
+   // [VK_FORMAT_ASTC_8x8_SRGB] = { .format = ASTC_8x8 },
+   // [VK_FORMAT_ASTC_10x5_UNORM] = { .format = ASTC_10x5 },
+   // [VK_FORMAT_ASTC_10x5_SRGB] = { .format = ASTC_10x5 },
+   // [VK_FORMAT_ASTC_10x6_UNORM] = { .format = ASTC_10x6 },
+   // [VK_FORMAT_ASTC_10x6_SRGB] = { .format = ASTC_10x6 },
+   // [VK_FORMAT_ASTC_10x8_UNORM] = { .format = ASTC_10x8 },
+   // [VK_FORMAT_ASTC_10x8_SRGB] = { .format = ASTC_10x8 },
+   // [VK_FORMAT_ASTC_10x10_UNORM] = { .format = ASTC_10x10 },
+   // [VK_FORMAT_ASTC_10x10_SRGB] = { .format = ASTC_10x10 },
+   // [VK_FORMAT_ASTC_12x10_UNORM] = { .format = ASTC_12x10 },
+   // [VK_FORMAT_ASTC_12x10_SRGB] = { .format = ASTC_12x10 },
+   // [VK_FORMAT_ASTC_12x12_UNORM] = { .format = ASTC_12x12 },
+   // [VK_FORMAT_ASTC_12x12_SRGB] = { .format = ASTC_12x12 },
+   // [VK_FORMAT_B4G4R4A4_UNORM] = { .format = B4G4R4A4_UNORM },
+   // [VK_FORMAT_B5G5R5A1_UNORM] = { .format = B5G5R5A1_UNORM },
+   // [VK_FORMAT_B5G6R5_UNORM] = { .format = B5G6R5_UNORM },
+   // [VK_FORMAT_B5G6R5_USCALED] = { .format = B5G6R5_USCALED },
+   // [VK_FORMAT_B8G8R8_UNORM] = { .format = B8G8R8_UNORM },
+   // [VK_FORMAT_B8G8R8_SNORM] = { .format = B8G8R8_SNORM },
+   // [VK_FORMAT_B8G8R8_USCALED] = { .format = B8G8R8_USCALED },
+   // [VK_FORMAT_B8G8R8_SSCALED] = { .format = B8G8R8_SSCALED },
+   // [VK_FORMAT_B8G8R8_UINT] = { .format = B8G8R8_UINT },
+   // [VK_FORMAT_B8G8R8_SINT] = { .format = B8G8R8_SINT },
+   // [VK_FORMAT_B8G8R8_SRGB] = { .format = B8G8R8_SRGB },
+   [VK_FORMAT_B8G8R8A8_UNORM] = { .format = B8G8R8A8_UNORM, .cpp = 4, .channels = 4 },
+   // [VK_FORMAT_B8G8R8A8_SNORM] = { .format = B8G8R8A8_SNORM },
+   // [VK_FORMAT_B8G8R8A8_USCALED] = { .format = B8G8R8A8_USCALED },
+   // [VK_FORMAT_B8G8R8A8_SSCALED] = { .format = B8G8R8A8_SSCALED },
+   // [VK_FORMAT_B8G8R8A8_UINT] = { .format = B8G8R8A8_UINT },
+   // [VK_FORMAT_B8G8R8A8_SINT] = { .format = B8G8R8A8_SINT },
+   // [VK_FORMAT_B8G8R8A8_SRGB] = { .format = B8G8R8A8_SRGB },
+   // [VK_FORMAT_B10G10R10A2_UNORM] = { .format = B10G10R10A2_UNORM },
+   // [VK_FORMAT_B10G10R10A2_SNORM] = { .format = B10G10R10A2_SNORM },
+   // [VK_FORMAT_B10G10R10A2_USCALED] = { .format = B10G10R10A2_USCALED },
+   // [VK_FORMAT_B10G10R10A2_SSCALED] = { .format = B10G10R10A2_SSCALED },
+   // [VK_FORMAT_B10G10R10A2_UINT] = { .format = B10G10R10A2_UINT },
+   // [VK_FORMAT_B10G10R10A2_SINT] = { .format = B10G10R10A2_SINT }
+};
+
+const struct anv_format *
+anv_format_for_vk_format(VkFormat format)
+{
+   return &anv_formats[format];
+}
+
+static const struct anv_tile_mode_info {
+   int32_t tile_width;
+   int32_t tile_height;
+} tile_mode_info[] = {
+   [LINEAR] = {   1,  1 },
+   [XMAJOR] = { 512,  8 },
+   [YMAJOR] = { 128, 32 },
+   [WMAJOR] = { 128, 32 }
+};
+
+VkResult VKAPI vkCreateImage(
+    VkDevice                                    _device,
+    const VkImageCreateInfo*                    pCreateInfo,
+    VkImage*                                    pImage)
+{
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_image *image;
+   const struct anv_format *format;
+   int32_t aligned_height;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO);
+
+   image = anv_device_alloc(device, sizeof(*image), 8,
+                            VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (image == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   image->mem = NULL;
+   image->offset = 0;
+   image->type = pCreateInfo->imageType;
+   image->extent = pCreateInfo->extent;
+
+   assert(image->extent.width > 0);
+   assert(image->extent.height > 0);
+   assert(image->extent.depth > 0);
+
+   switch (pCreateInfo->tiling) {
+   case VK_IMAGE_TILING_LINEAR:
+      image->tile_mode = LINEAR;
+      /* Linear depth buffers must be 64 byte aligned, which is the strictest
+       * requirement for all kinds of linear surfaces.
+       */
+      image->alignment = 64;
+      break;
+   case VK_IMAGE_TILING_OPTIMAL:
+      image->tile_mode = YMAJOR;
+      image->alignment = 4096;
+      break;
+   default:
+      break;
+   }
+   
+   format = anv_format_for_vk_format(pCreateInfo->format);
+   image->stride = ALIGN_I32(image->extent.width * format->cpp,
+                             tile_mode_info[image->tile_mode].tile_width);
+   aligned_height = ALIGN_I32(image->extent.height,
+                              tile_mode_info[image->tile_mode].tile_height);
+   image->size = image->stride * aligned_height;
+
+   *pImage = (VkImage) image;
+
+   return VK_SUCCESS;
+}
+
+VkResult VKAPI vkGetImageSubresourceInfo(
+    VkDevice                                    device,
+    VkImage                                     image,
+    const VkImageSubresource*                   pSubresource,
+    VkSubresourceInfoType                       infoType,
+    size_t*                                     pDataSize,
+    void*                                       pData)
+{
+   return VK_UNSUPPORTED;
+}
+
+// Image view functions
+
+static struct anv_state
+create_surface_state(struct anv_device *device,
+                     struct anv_image *image, const struct anv_format *format)
+{
+   struct anv_state state =
+      anv_state_pool_alloc(&device->surface_state_pool, 64, 64);
+
+   struct GEN8_RENDER_SURFACE_STATE surface_state = {
+      .SurfaceType = SURFTYPE_2D,
+      .SurfaceArray = false,
+      .SurfaceFormat = format->format,
+      .SurfaceVerticalAlignment = VALIGN4,
+      .SurfaceHorizontalAlignment = HALIGN4,
+      .TileMode = image->tile_mode,
+      .VerticalLineStride = 0,
+      .VerticalLineStrideOffset = 0,
+      .SamplerL2BypassModeDisable = true,
+      .RenderCacheReadWriteMode = WriteOnlyCache,
+      .MemoryObjectControlState = 0, /* FIXME: MOCS */
+      .BaseMipLevel = 0,
+      .SurfaceQPitch = 0,
+      .Height = image->extent.height - 1,
+      .Width = image->extent.width - 1,
+      .Depth = image->extent.depth - 1,
+      .SurfacePitch = image->stride - 1,
+      .MinimumArrayElement = 0,
+      .NumberofMultisamples = MULTISAMPLECOUNT_1,
+      .XOffset = 0,
+      .YOffset = 0,
+      .SurfaceMinLOD = 0,
+      .MIPCountLOD = 0,
+      .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,
+      /* FIXME: We assume that the image must be bound at this time. */
+      .SurfaceBaseAddress = { NULL, image->offset },
+   };
+
+   GEN8_RENDER_SURFACE_STATE_pack(NULL, state.map, &surface_state);
+
+   return state;
+}
+
+VkResult VKAPI vkCreateImageView(
+    VkDevice                                    _device,
+    const VkImageViewCreateInfo*                pCreateInfo,
+    VkImageView*                                pView)
+{
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_image_view *view;
+   const struct anv_format *format =
+      anv_format_for_vk_format(pCreateInfo->format);
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO);
+
+   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);
+
+   view->image = (struct anv_image *) pCreateInfo->image;
+
+   view->surface_state = create_surface_state(device, view->image, format);
+
+   *pView = (VkImageView) view;
+
+   return VK_SUCCESS;
+}
+
+VkResult VKAPI vkCreateColorAttachmentView(
+    VkDevice                                    _device,
+    const VkColorAttachmentViewCreateInfo*      pCreateInfo,
+    VkColorAttachmentView*                      pView)
+{
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_color_attachment_view *view;
+   struct anv_image *image;
+   const struct anv_format *format =
+      anv_format_for_vk_format(pCreateInfo->format);
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_COLOR_ATTACHMENT_VIEW_CREATE_INFO);
+
+   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);
+
+   view->image = (struct anv_image *) pCreateInfo->image;
+   image = view->image;
+
+   view->surface_state = create_surface_state(device, image, format);
+
+   *pView = (VkColorAttachmentView) view;
+
+   return VK_SUCCESS;
+}
+
+VkResult VKAPI vkCreateDepthStencilView(
+    VkDevice                                    device,
+    const VkDepthStencilViewCreateInfo*         pCreateInfo,
+    VkDepthStencilView*                         pView)
+{
+   return VK_UNSUPPORTED;
+}

src/vulkan/intel.c

@@ -0,0 +1,93 @@
+/*
+ * 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"
+
+#include <vulkan/vulkan_intel.h>
+
+VkResult VKAPI vkCreateDmaBufImageINTEL(
+    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->mem = mem;
+   image->offset = 0;
+   image->type = VK_IMAGE_TYPE_2D;
+   image->extent = pCreateInfo->extent;
+   image->tile_mode = XMAJOR;
+   image->stride = pCreateInfo->strideInBytes;
+   image->size = mem->bo.size;
+
+   assert(image->extent.width > 0);
+   assert(image->extent.height > 0);
+   assert(image->extent.depth == 1);
+
+   *pMem = (VkDeviceMemory) mem;
+   *pImage = (VkImage) image;
+
+   return VK_SUCCESS;
+
+ fail_mem:
+   anv_gem_close(device, mem->bo.gem_handle);
+ fail:
+   anv_device_free(device, mem);
+
+   return result;
+}

src/vulkan/meta.c

@@ -0,0 +1,140 @@
+/*
+ * 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"
+
+void VKAPI vkCmdCopyBuffer(
+    VkCmdBuffer                                 cmdBuffer,
+    VkBuffer                                    srcBuffer,
+    VkBuffer                                    destBuffer,
+    uint32_t                                    regionCount,
+    const VkBufferCopy*                         pRegions)
+{
+}
+
+void VKAPI vkCmdCopyImage(
+    VkCmdBuffer                                 cmdBuffer,
+    VkImage                                     srcImage,
+    VkImageLayout                               srcImageLayout,
+    VkImage                                     destImage,
+    VkImageLayout                               destImageLayout,
+    uint32_t                                    regionCount,
+    const VkImageCopy*                          pRegions)
+{
+}
+
+void VKAPI vkCmdBlitImage(
+    VkCmdBuffer                                 cmdBuffer,
+    VkImage                                     srcImage,
+    VkImageLayout                               srcImageLayout,
+    VkImage                                     destImage,
+    VkImageLayout                               destImageLayout,
+    uint32_t                                    regionCount,
+    const VkImageBlit*                          pRegions)
+{
+}
+
+void VKAPI vkCmdCopyBufferToImage(
+    VkCmdBuffer                                 cmdBuffer,
+    VkBuffer                                    srcBuffer,
+    VkImage                                     destImage,
+    VkImageLayout                               destImageLayout,
+    uint32_t                                    regionCount,
+    const VkBufferImageCopy*                    pRegions)
+{
+}
+
+void VKAPI vkCmdCopyImageToBuffer(
+    VkCmdBuffer                                 cmdBuffer,
+    VkImage                                     srcImage,
+    VkImageLayout                               srcImageLayout,
+    VkBuffer                                    destBuffer,
+    uint32_t                                    regionCount,
+    const VkBufferImageCopy*                    pRegions)
+{
+}
+
+void VKAPI vkCmdCloneImageData(
+    VkCmdBuffer                                 cmdBuffer,
+    VkImage                                     srcImage,
+    VkImageLayout                               srcImageLayout,
+    VkImage                                     destImage,
+    VkImageLayout                               destImageLayout)
+{
+}
+
+void VKAPI vkCmdUpdateBuffer(
+    VkCmdBuffer                                 cmdBuffer,
+    VkBuffer                                    destBuffer,
+    VkDeviceSize                                destOffset,
+    VkDeviceSize                                dataSize,
+    const uint32_t*                             pData)
+{
+}
+
+void VKAPI vkCmdFillBuffer(
+    VkCmdBuffer                                 cmdBuffer,
+    VkBuffer                                    destBuffer,
+    VkDeviceSize                                destOffset,
+    VkDeviceSize                                fillSize,
+    uint32_t                                    data)
+{
+}
+
+void VKAPI vkCmdClearColorImage(
+    VkCmdBuffer                                 cmdBuffer,
+    VkImage                                     image,
+    VkImageLayout                               imageLayout,
+    const VkClearColor*                         color,
+    uint32_t                                    rangeCount,
+    const VkImageSubresourceRange*              pRanges)
+{
+}
+
+void VKAPI vkCmdClearDepthStencil(
+    VkCmdBuffer                                 cmdBuffer,
+    VkImage                                     image,
+    VkImageLayout                               imageLayout,
+    float                                       depth,
+    uint32_t                                    stencil,
+    uint32_t                                    rangeCount,
+    const VkImageSubresourceRange*              pRanges)
+{
+}
+
+void VKAPI vkCmdResolveImage(
+    VkCmdBuffer                                 cmdBuffer,
+    VkImage                                     srcImage,
+    VkImageLayout                               srcImageLayout,
+    VkImage                                     destImage,
+    VkImageLayout                               destImageLayout,
+    uint32_t                                    regionCount,
+    const VkImageResolve*                       pRegions)
+{
+}

src/vulkan/pipeline.c

@@ -0,0 +1,565 @@
+/*
+ * 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 VKAPI vkCreateShader(
+    VkDevice                                    _device,
+    const VkShaderCreateInfo*                   pCreateInfo,
+    VkShader*                                   pShader)
+{
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_shader *shader;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_SHADER_CREATE_INFO);
+
+   shader = anv_device_alloc(device, sizeof(*shader) + pCreateInfo->codeSize, 8,
+                               VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (shader == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   shader->size = pCreateInfo->codeSize;
+   memcpy(shader->data, pCreateInfo->pCode, shader->size);
+
+   *pShader = (VkShader) shader;
+
+   return VK_SUCCESS;
+}
+
+// Pipeline functions
+
+static void
+emit_vertex_input(struct anv_pipeline *pipeline, VkPipelineVertexInputCreateInfo *info)
+{
+   const uint32_t num_dwords = 1 + info->attributeCount * 2;
+   uint32_t *p;
+   bool instancing_enable[32];
+   
+   for (uint32_t i = 0; i < info->bindingCount; i++) {
+      const VkVertexInputBindingDescription *desc =
+         &info->pVertexBindingDescriptions[i];
+      
+      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->location,
+         .Valid = true,
+         .SourceElementFormat = format->format,
+         .EdgeFlagEnable = false,
+         .SourceElementOffset = desc->offsetInBytes,
+         .Component0Control = VFCOMP_STORE_SRC,
+         .Component1Control = format->channels >= 2 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
+         .Component2Control = format->channels >= 3 ? VFCOMP_STORE_SRC : VFCOMP_STORE_0,
+         .Component3Control = format->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, VkPipelineIaStateCreateInfo *info)
+{
+   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_LISTSTRIP_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
+   };
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF,
+                  .IndexedDrawCutIndexEnable = info->primitiveRestartEnable,
+                  .CutIndex = info->primitiveRestartIndex);
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_VF_TOPOLOGY,
+                  .PrimitiveTopologyType = vk_to_gen_primitive_type[info->topology]);
+}
+
+static void
+emit_rs_state(struct anv_pipeline *pipeline, VkPipelineRsStateCreateInfo *info)
+{
+   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
+   };
+   
+   static const uint32_t vk_to_gen_coordinate_origin[] = {
+      [VK_COORDINATE_ORIGIN_UPPER_LEFT] = UPPERLEFT,
+      [VK_COORDINATE_ORIGIN_LOWER_LEFT] = LOWERLEFT
+   };
+
+   struct GEN8_3DSTATE_SF sf = {
+      GEN8_3DSTATE_SF_header,
+      .ViewportTransformEnable = true,
+      .TriangleStripListProvokingVertexSelect =
+         info->provokingVertex == VK_PROVOKING_VERTEX_FIRST ? 0 : 2,
+      .LineStripListProvokingVertexSelect =
+         info->provokingVertex == VK_PROVOKING_VERTEX_FIRST ? 0 : 1,
+      .TriangleFanProvokingVertexSelect =
+         info->provokingVertex == VK_PROVOKING_VERTEX_FIRST ? 0 : 2,
+      .PointWidthSource = info->programPointSize ? Vertex : State,
+   };
+
+   /* bool32_t                                    rasterizerDiscardEnable; */
+
+
+   GEN8_3DSTATE_SF_pack(NULL, pipeline->state_sf, &sf);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_RASTER,
+                  .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 = true,
+                  .ViewportZClipTestEnable = info->depthClipEnable);
+
+   anv_batch_emit(&pipeline->batch, GEN8_3DSTATE_SBE,
+                  .ForceVertexURBEntryReadLength = false,
+                  .ForceVertexURBEntryReadOffset = false,
+                  .PointSpriteTextureCoordinateOrigin =
+                     vk_to_gen_coordinate_origin[info->pointOrigin],
+                  .NumberofSFOutputAttributes =
+                     pipeline->wm_prog_data.num_varying_inputs);
+
+}
+
+VkResult VKAPI vkCreateGraphicsPipeline(
+    VkDevice                                    _device,
+    const VkGraphicsPipelineCreateInfo*         pCreateInfo,
+    VkPipeline*                                 pPipeline)
+{
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_pipeline *pipeline;
+   const struct anv_common *common;
+   VkPipelineShaderStageCreateInfo *shader_create_info;
+   VkPipelineIaStateCreateInfo *ia_info;
+   VkPipelineRsStateCreateInfo *rs_info;
+   VkPipelineVertexInputCreateInfo *vi_info;
+   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 = (struct anv_pipeline_layout *) pCreateInfo->layout;
+   memset(pipeline->shaders, 0, sizeof(pipeline->shaders));
+   result = anv_batch_init(&pipeline->batch, device);
+   if (result != VK_SUCCESS)
+      goto fail;
+
+   for (common = pCreateInfo->pNext; common; common = common->pNext) {
+      switch (common->sType) {
+      case VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_CREATE_INFO:
+         vi_info = (VkPipelineVertexInputCreateInfo *) common;
+         break;
+      case VK_STRUCTURE_TYPE_PIPELINE_IA_STATE_CREATE_INFO:
+         ia_info = (VkPipelineIaStateCreateInfo *) common;
+         break;
+      case VK_STRUCTURE_TYPE_PIPELINE_TESS_STATE_CREATE_INFO:
+      case VK_STRUCTURE_TYPE_PIPELINE_VP_STATE_CREATE_INFO:
+         break;
+      case VK_STRUCTURE_TYPE_PIPELINE_RS_STATE_CREATE_INFO:
+         rs_info = (VkPipelineRsStateCreateInfo *) common;
+         break;
+      case VK_STRUCTURE_TYPE_PIPELINE_MS_STATE_CREATE_INFO:
+      case VK_STRUCTURE_TYPE_PIPELINE_CB_STATE_CREATE_INFO:
+      case VK_STRUCTURE_TYPE_PIPELINE_DS_STATE_CREATE_INFO:
+      case VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO:
+         shader_create_info = (VkPipelineShaderStageCreateInfo *) common;
+         pipeline->shaders[shader_create_info->shader.stage] =
+            (struct anv_shader *) shader_create_info->shader.shader;
+         break;
+      default:
+         break;
+      }
+   }
+
+   pipeline->use_repclear = false;
+
+   anv_compiler_run(device->compiler, pipeline);
+
+   emit_vertex_input(pipeline, vi_info);
+   emit_ia_state(pipeline, ia_info);
+   emit_rs_state(pipeline, rs_info);
+
+   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_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,
+                        
+                     .PerThreadScratchSpace = 0,
+                     .ScratchSpaceBasePointer = 0,
+
+                     .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);
+
+   //trp_generate_blend_hw_cmds(batch, pipeline);
+
+   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;
+
+   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,
+
+                  /* FIXME: pointer needs to be assigned outside as it aliases
+                   * PerThreadScratchSpace.
+                   */
+                  .ScratchSpaceBasePointer = 0,
+                  .PerThreadScratchSpace = 0,
+
+                  .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 = ia_info->disableVertexReuse,
+                  .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 = 0,
+
+                  .ScratchSpaceBasePointer = 0,
+                  .PerThreadScratchSpace = 0,
+                  
+                  .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 = (VkPipeline) pipeline;
+
+   return VK_SUCCESS;
+
+ fail:
+   anv_device_free(device, pipeline);
+   
+   return result;
+}
+
+VkResult
+anv_pipeline_destroy(struct anv_pipeline *pipeline)
+{
+   anv_compiler_free(pipeline);
+   anv_batch_finish(&pipeline->batch, pipeline->device);
+   anv_device_free(pipeline->device, pipeline);
+
+   return VK_SUCCESS;
+}
+
+VkResult VKAPI vkCreateGraphicsPipelineDerivative(
+    VkDevice                                    device,
+    const VkGraphicsPipelineCreateInfo*         pCreateInfo,
+    VkPipeline                                  basePipeline,
+    VkPipeline*                                 pPipeline)
+{
+   return VK_UNSUPPORTED;
+}
+
+VkResult VKAPI vkCreateComputePipeline(
+    VkDevice                                    device,
+    const VkComputePipelineCreateInfo*          pCreateInfo,
+    VkPipeline*                                 pPipeline)
+{
+   return VK_UNSUPPORTED;
+}
+
+VkResult VKAPI vkStorePipeline(
+    VkDevice                                    device,
+    VkPipeline                                  pipeline,
+    size_t*                                     pDataSize,
+    void*                                       pData)
+{
+   return VK_UNSUPPORTED;
+}
+
+VkResult VKAPI vkLoadPipeline(
+    VkDevice                                    device,
+    size_t                                      dataSize,
+    const void*                                 pData,
+    VkPipeline*                                 pPipeline)
+{
+   return VK_UNSUPPORTED;
+}
+
+VkResult VKAPI vkLoadPipelineDerivative(
+    VkDevice                                    device,
+    size_t                                      dataSize,
+    const void*                                 pData,
+    VkPipeline                                  basePipeline,
+    VkPipeline*                                 pPipeline)
+{
+   return VK_UNSUPPORTED;
+}
+
+// Pipeline layout functions
+
+VkResult VKAPI vkCreatePipelineLayout(
+    VkDevice                                    _device,
+    const VkPipelineLayoutCreateInfo*           pCreateInfo,
+    VkPipelineLayout*                           pPipelineLayout)
+{
+   struct anv_device *device = (struct anv_device *) _device;
+   struct anv_pipeline_layout *layout;
+   struct anv_pipeline_layout_entry *entry;
+   uint32_t total;
+   size_t size;
+
+   assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO);
+   
+   total = 0;
+   for (uint32_t i = 0; i < pCreateInfo->descriptorSetCount; i++) {
+      struct anv_descriptor_set_layout *set_layout =
+         (struct anv_descriptor_set_layout *) pCreateInfo->pSetLayouts[i];
+      for (uint32_t j = 0; j < set_layout->count; j++)
+         total += set_layout->total;
+   }
+
+   size = sizeof(*layout) + total * sizeof(layout->entries[0]);
+   layout = anv_device_alloc(device, size, 8,
+                             VK_SYSTEM_ALLOC_TYPE_API_OBJECT);
+   if (layout == NULL)
+      return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+
+   entry = layout->entries;
+   for (uint32_t s = 0; s < VK_NUM_SHADER_STAGE; s++) {
+      layout->stage[s].entries = entry;
+
+      for (uint32_t i = 0; i < pCreateInfo->descriptorSetCount; i++) {
+         struct anv_descriptor_set_layout *set_layout =
+            (struct anv_descriptor_set_layout *) pCreateInfo->pSetLayouts[i];
+         for (uint32_t j = 0; j < set_layout->count; j++)
+            if (set_layout->bindings[j].mask & (1 << s)) {
+               entry->type = set_layout->bindings[j].type;
+               entry->set = i;
+               entry->index = j;
+               entry++;
+            }
+      }
+
+      layout->stage[s].count = entry - layout->stage[s].entries;
+   }
+
+   *pPipelineLayout = (VkPipelineLayout) layout;
+
+   return VK_SUCCESS;
+}

src/vulkan/private.h

@@ -0,0 +1,594 @@
+/*
+ * 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.
+ */
+
+#pragma once
+
+#include <stdlib.h>
+#include <stdio.h>
+#include <stdbool.h>
+#include <pthread.h>
+#include <assert.h>
+#include <i915_drm.h>
+
+#include "brw_device_info.h"
+#include "util/macros.h"
+
+#define VK_PROTOTYPES
+#include <vulkan/vulkan.h>
+
+#undef VKAPI
+#define VKAPI __attribute__ ((visibility ("default")))
+
+#include "brw_context.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static inline uint32_t
+ALIGN_U32(uint32_t v, uint32_t a)
+{
+   return (v + a - 1) & ~(a - 1);
+}
+
+static inline int32_t
+ALIGN_I32(int32_t v, int32_t a)
+{
+   return (v + a - 1) & ~(a - 1);
+}
+
+#define for_each_bit(b, dword)                          \
+   for (uint32_t __dword = (dword);                     \
+        (b) = __builtin_ffs(__dword) - 1, __dword;      \
+        __dword &= ~(1 << (b)))
+
+/* Define no kernel as 1, since that's an illegal offset for a kernel */
+#define NO_KERNEL 1
+
+struct anv_common {
+    VkStructureType                             sType;
+    const void*                                 pNext;
+};
+
+/* Whenever we generate an error, pass it through this function. Useful for
+ * debugging, where we can break on it. Only call at error site, not when
+ * propagating errors. Might be useful to plug in a stack trace here.
+ */
+
+static inline VkResult
+vk_error(VkResult error)
+{
+#ifdef DEBUG
+   fprintf(stderr, "vk_error: %x\n", error);
+#endif
+
+   return error;
+}
+
+/**
+ * A dynamically growable, circular buffer.  Elements are added at head and
+ * removed from tail. head and tail are free-running uint32_t indices and we
+ * only compute the modulo with size when accessing the array.  This way,
+ * number of bytes in the queue is always head - tail, even in case of
+ * wraparound.
+ */
+
+struct anv_vector {
+   uint32_t head;
+   uint32_t tail;
+   uint32_t element_size;
+   uint32_t size;
+   void *data;
+};
+
+int anv_vector_init(struct anv_vector *queue, uint32_t element_size, uint32_t size);
+void *anv_vector_add(struct anv_vector *queue);
+void *anv_vector_remove(struct anv_vector *queue);
+
+static inline int
+anv_vector_length(struct anv_vector *queue)
+{
+   return (queue->head - queue->tail) / queue->element_size;
+}
+
+static inline void
+anv_vector_finish(struct anv_vector *queue)
+{
+   free(queue->data);
+}
+
+#define anv_vector_foreach(elem, queue)                                  \
+   static_assert(__builtin_types_compatible_p(__typeof__(queue), struct anv_vector *), ""); \
+   for (uint32_t __anv_vector_offset = (queue)->tail;                                \
+        elem = (queue)->data + (__anv_vector_offset & ((queue)->size - 1)), __anv_vector_offset < (queue)->head; \
+        __anv_vector_offset += (queue)->element_size)
+
+struct anv_bo {
+   int gem_handle;
+   uint32_t index;
+   uint64_t offset;
+   uint64_t size;
+
+   /* This field is here for the benefit of the aub dumper.  It can (and for
+    * userptr bos it must) be set to the cpu map of the buffer.  Destroying
+    * the bo won't clean up the mmap, it's still the responsibility of the bo
+    * user to do that. */
+   void *map;
+};
+
+/* Represents a lock-free linked list of "free" things.  This is used by
+ * both the block pool and the state pools.  Unfortunately, in order to
+ * solve the ABA problem, we can't use a single uint32_t head.
+ */
+union anv_free_list {
+   struct {
+      uint32_t offset;
+
+      /* A simple count that is incremented every time the head changes. */
+      uint32_t count;
+   };
+   uint64_t u64;
+};
+
+#define ANV_FREE_LIST_EMPTY ((union anv_free_list) { { 1, 0 } })
+
+struct anv_block_pool {
+   struct anv_device *device;
+
+   struct anv_bo bo;
+   void *map;
+   int fd;
+   uint32_t size;
+
+   /**
+    * Array of mmaps and gem handles owned by the block pool, reclaimed when
+    * the block pool is destroyed.
+    */
+   struct anv_vector mmap_cleanups;
+
+   uint32_t block_size;
+
+   uint32_t next_block;
+   union anv_free_list free_list;
+};
+
+struct anv_block_state {
+   union {
+      struct {
+         uint32_t next;
+         uint32_t end;
+      };
+      uint64_t u64;
+   };
+};
+
+struct anv_state {
+   uint32_t offset;
+   uint32_t alloc_size;
+   void *map;
+};
+
+struct anv_fixed_size_state_pool {
+   size_t state_size;
+   union anv_free_list free_list;
+   struct anv_block_state block;
+};
+
+#define ANV_MIN_STATE_SIZE_LOG2 6
+#define ANV_MAX_STATE_SIZE_LOG2 10
+
+#define ANV_STATE_BUCKETS (ANV_MAX_STATE_SIZE_LOG2 - ANV_MIN_STATE_SIZE_LOG2)
+
+struct anv_state_pool {
+   struct anv_block_pool *block_pool;
+   struct anv_fixed_size_state_pool buckets[ANV_STATE_BUCKETS];
+};
+
+struct anv_state_stream {
+   struct anv_block_pool *block_pool;
+   uint32_t next;
+   uint32_t current_block;
+   uint32_t end;
+};
+
+void anv_block_pool_init(struct anv_block_pool *pool,
+                         struct anv_device *device, uint32_t block_size);
+void anv_block_pool_finish(struct anv_block_pool *pool);
+uint32_t anv_block_pool_alloc(struct anv_block_pool *pool);
+void anv_block_pool_free(struct anv_block_pool *pool, uint32_t offset);
+void anv_state_pool_init(struct anv_state_pool *pool,
+                         struct anv_block_pool *block_pool);
+struct anv_state anv_state_pool_alloc(struct anv_state_pool *pool,
+                                      size_t state_size, size_t alignment);
+void anv_state_pool_free(struct anv_state_pool *pool, struct anv_state state);
+void anv_state_stream_init(struct anv_state_stream *stream,
+                           struct anv_block_pool *block_pool);
+void anv_state_stream_finish(struct anv_state_stream *stream);
+struct anv_state anv_state_stream_alloc(struct anv_state_stream *stream,
+                                        uint32_t size, uint32_t alignment);
+
+struct anv_physical_device {
+    struct anv_instance *                       instance;
+    uint32_t                                    chipset_id;
+    bool                                        no_hw;
+    const char *                                path;
+    const char *                                name;
+    const struct brw_device_info *              info;
+};
+
+struct anv_instance {
+    void *                                      pAllocUserData;
+    PFN_vkAllocFunction                         pfnAlloc;
+    PFN_vkFreeFunction                          pfnFree;
+    uint32_t                                    apiVersion;
+    uint32_t                                    physicalDeviceCount;
+    struct anv_physical_device                  physicalDevice;
+};
+
+struct anv_device {
+    struct anv_instance *                       instance;
+    uint32_t                                    chipset_id;
+    struct brw_device_info                      info;
+    int                                         context_id;
+    int                                         fd;
+    bool                                        no_hw;
+    bool                                        dump_aub;
+
+    struct anv_block_pool                       dyn_state_block_pool;
+    struct anv_state_pool                       dyn_state_pool;
+
+    struct anv_block_pool                       instruction_block_pool;
+    struct anv_block_pool                       surface_state_block_pool;
+    struct anv_state_pool                       surface_state_pool;
+
+    struct anv_compiler *                       compiler;
+    struct anv_aub_writer *                     aub_writer;
+    pthread_mutex_t                             mutex;
+};
+
+struct anv_queue {
+    struct anv_device *                         device;
+
+    struct anv_state_pool *                     pool;
+
+    /**
+     * Serial number of the most recently completed batch executed on the
+     * engine.
+     */
+    struct anv_state                            completed_serial;
+
+    /**
+     * The next batch submitted to the engine will be assigned this serial
+     * number.
+     */
+    uint32_t                                    next_serial;
+
+    uint32_t                                    last_collected_serial;
+};
+
+void *
+anv_device_alloc(struct anv_device *            device,
+                 size_t                         size,
+                 size_t                         alignment,
+                 VkSystemAllocType              allocType);
+
+void
+anv_device_free(struct anv_device *             device,
+                void *                          mem);
+
+void* anv_gem_mmap(struct anv_device *device,
+                   uint32_t gem_handle, uint64_t offset, uint64_t size);
+void anv_gem_munmap(void *p, uint64_t size);
+uint32_t anv_gem_create(struct anv_device *device, size_t size);
+void anv_gem_close(struct anv_device *device, int gem_handle);
+int anv_gem_userptr(struct anv_device *device, void *mem, size_t size);
+int anv_gem_wait(struct anv_device *device, int gem_handle, int64_t *timeout_ns);
+int anv_gem_execbuffer(struct anv_device *device,
+                       struct drm_i915_gem_execbuffer2 *execbuf);
+int anv_gem_set_tiling(struct anv_device *device, int gem_handle,
+                       uint32_t stride, uint32_t tiling);
+int anv_gem_create_context(struct anv_device *device);
+int anv_gem_destroy_context(struct anv_device *device, int context);
+int anv_gem_get_param(int fd, uint32_t param);
+int anv_gem_get_aperture(struct anv_device *device, uint64_t *size);
+int anv_gem_handle_to_fd(struct anv_device *device, int gem_handle);
+int anv_gem_fd_to_handle(struct anv_device *device, int fd);
+int anv_gem_userptr(struct anv_device *device, void *mem, size_t size);
+
+VkResult anv_bo_init_new(struct anv_bo *bo, struct anv_device *device, uint64_t size);
+
+/* TODO: Remove hardcoded reloc limit. */
+#define ANV_BATCH_MAX_RELOCS 256
+
+struct anv_reloc_list {
+   size_t                                       num_relocs;
+   struct drm_i915_gem_relocation_entry         relocs[ANV_BATCH_MAX_RELOCS];
+   struct anv_bo *                              reloc_bos[ANV_BATCH_MAX_RELOCS];
+};
+
+struct anv_batch {
+   struct anv_bo                                bo;
+   void *                                       next;
+   struct anv_reloc_list                        cmd_relocs;
+   struct anv_reloc_list                        surf_relocs;
+};
+
+VkResult anv_batch_init(struct anv_batch *batch, struct anv_device *device);
+void anv_batch_finish(struct anv_batch *batch, struct anv_device *device);
+void anv_batch_reset(struct anv_batch *batch);
+void *anv_batch_emit_dwords(struct anv_batch *batch, int num_dwords);
+void anv_batch_emit_batch(struct anv_batch *batch, struct anv_batch *other);
+uint64_t anv_batch_emit_reloc(struct anv_batch *batch,
+                              void *location, struct anv_bo *bo, uint32_t offset);
+
+struct anv_address {
+   struct anv_bo *bo;
+   uint32_t offset;
+};
+
+#define __gen_address_type struct anv_address
+#define __gen_user_data struct anv_batch
+
+static inline uint64_t
+__gen_combine_address(struct anv_batch *batch, void *location,
+                      const struct anv_address address, uint32_t delta)
+{   
+   if (address.bo == NULL) {
+      return delta;
+   } else {
+      assert(batch->bo.map <= location &&
+             (char *) location < (char *) batch->bo.map + batch->bo.size);
+
+      return anv_batch_emit_reloc(batch, location, address.bo, address.offset + delta);
+   }
+}
+   
+#undef GEN8_3DSTATE_MULTISAMPLE
+#include "gen8_pack.h"
+
+#define anv_batch_emit(batch, cmd, ...) do {                            \
+      struct cmd __template = {                                         \
+         cmd ## _header,                                                \
+         __VA_ARGS__                                                    \
+      };                                                                \
+      void *__dst = anv_batch_emit_dwords(batch, cmd ## _length);       \
+      cmd ## _pack(batch, __dst, &__template);                          \
+   } while (0)
+
+#define anv_batch_emitn(batch, n, cmd, ...) ({          \
+      struct cmd __template = {                         \
+         cmd ## _header,                                \
+        .DwordLength = n - cmd ## _length_bias,         \
+         __VA_ARGS__                                    \
+      };                                                \
+      void *__dst = anv_batch_emit_dwords(batch, n);    \
+      cmd ## _pack(batch, __dst, &__template);          \
+      __dst;                                            \
+   })
+
+struct anv_device_memory {
+   struct anv_bo bo;
+   VkDeviceSize map_size;
+   void *map;
+};
+
+struct anv_dynamic_vp_state {
+   struct anv_state sf_clip_vp;
+   struct anv_state cc_vp;
+   struct anv_state scissor;
+};
+
+struct anv_dynamic_rs_state {
+   uint32_t state_sf[GEN8_3DSTATE_SF_length];
+};
+
+struct anv_dynamic_cb_state {
+   uint32_t blend_offset;
+};
+
+struct anv_descriptor_set_layout {
+   uint32_t total; /* total number of entries in all stages */
+   uint32_t count;
+   struct {
+      VkDescriptorType type;
+      uint32_t mask;
+   } bindings[0];
+};
+
+struct anv_descriptor_set {
+   void *descriptors[0];
+};
+
+struct anv_pipeline_layout_entry {
+   VkDescriptorType type;
+   uint32_t set;
+   uint32_t index;
+};
+
+struct anv_pipeline_layout {
+   struct {
+      uint32_t count;
+      struct anv_pipeline_layout_entry *entries;
+   } stage[VK_NUM_SHADER_STAGE];
+   
+   struct anv_pipeline_layout_entry entries[0];
+};
+
+struct anv_buffer {
+   struct anv_device *                          device;
+   VkDeviceSize                                 size;
+
+   /* Set when bound */
+   struct anv_device_memory *                   mem;
+   VkDeviceSize                                 offset;   
+};
+
+#define MAX_VBS   32
+#define MAX_SETS   8
+#define MAX_RTS    8
+
+#define ANV_CMD_BUFFER_PIPELINE_DIRTY           (1 << 0)
+#define ANV_CMD_BUFFER_DESCRIPTOR_SET_DIRTY     (1 << 1)
+#define ANV_CMD_BUFFER_RS_DIRTY                 (1 << 2)
+   
+struct anv_cmd_buffer {
+   struct anv_device *                          device;
+
+   struct drm_i915_gem_execbuffer2              execbuf;
+   struct drm_i915_gem_exec_object2 *           exec2_objects;
+   struct anv_bo **                             exec2_bos;
+   bool                                         need_reloc;
+   uint32_t                                     serial;
+
+   uint32_t                                     bo_count;
+   struct anv_batch                             batch;
+   struct anv_state_stream                      surface_state_stream;
+
+   /* State required while building cmd buffer */
+   struct {
+      struct anv_buffer *buffer;
+      VkDeviceSize offset;
+   }                                            vb[MAX_VBS];
+   uint32_t                                     vb_dirty;
+   uint32_t                                     num_descriptor_sets;
+   struct anv_descriptor_set *                  descriptor_sets[MAX_SETS];
+   uint32_t                                     dirty;
+   struct anv_pipeline *                        pipeline;
+   struct anv_framebuffer *                     framebuffer;
+   struct anv_dynamic_rs_state *                rs_state;
+};
+
+void anv_cmd_buffer_dump(struct anv_cmd_buffer *cmd_buffer);
+void anv_aub_writer_destroy(struct anv_aub_writer *writer);
+
+struct anv_shader {
+   uint32_t size;
+   char data[0];
+};
+
+struct anv_pipeline {
+   struct anv_device *                          device;
+   struct anv_batch                             batch;
+   struct anv_shader *                          shaders[VK_NUM_SHADER_STAGE];
+   struct anv_pipeline_layout *                 layout;
+   bool                                         use_repclear;
+
+   struct brw_vs_prog_data                      vs_prog_data;
+   struct brw_wm_prog_data                      wm_prog_data;
+   struct brw_gs_prog_data                      gs_prog_data;
+   struct brw_stage_prog_data *                 prog_data[VK_NUM_SHADER_STAGE];
+   struct {
+      uint32_t                                  vs_start;
+      uint32_t                                  vs_size;
+      uint32_t                                  nr_vs_entries;
+      uint32_t                                  gs_start;
+      uint32_t                                  gs_size;
+      uint32_t                                  nr_gs_entries;
+   } urb;
+
+   struct anv_bo                                vs_scratch_bo;
+   struct anv_bo                                ps_scratch_bo;
+   struct anv_bo                                gs_scratch_bo;
+
+   uint32_t                                     active_stages;
+   uint32_t                                     program_block;
+   uint32_t                                     program_next;
+   uint32_t                                     vs_simd8;
+   uint32_t                                     ps_simd8;
+   uint32_t                                     ps_simd16;
+   uint32_t                                     gs_vec4;
+   uint32_t                                     gs_vertex_count;
+
+   uint32_t                                     binding_stride[MAX_VBS];
+
+   uint32_t                                     state_sf[GEN8_3DSTATE_SF_length];
+   uint32_t                                     state_raster[GEN8_3DSTATE_RASTER_length];
+};
+
+VkResult anv_pipeline_destroy(struct anv_pipeline *pipeline);
+
+struct anv_compiler *anv_compiler_create(int fd);
+void anv_compiler_destroy(struct anv_compiler *compiler);
+int anv_compiler_run(struct anv_compiler *compiler, struct anv_pipeline *pipeline);
+void anv_compiler_free(struct anv_pipeline *pipeline);
+
+struct anv_format {
+   uint32_t                                     format;
+   int32_t                                      cpp;
+   int32_t                                      channels;
+};
+
+const struct anv_format *
+anv_format_for_vk_format(VkFormat format);
+
+struct anv_image {
+   VkImageType                                  type;
+   VkExtent3D                                   extent;
+   uint32_t                                     tile_mode;
+   VkDeviceSize                                 size;
+   uint32_t                                     alignment;
+   int32_t                                      stride;
+
+   /* Set when bound */
+   struct anv_device_memory *                   mem;
+   VkDeviceSize                                 offset;
+};
+
+struct anv_buffer_view {
+   struct anv_buffer *                          buffer;
+   struct anv_state                             surface_state;
+   uint32_t                                     offset;
+};
+
+struct anv_color_attachment_view {
+   struct anv_image *                           image;
+   struct anv_state                             surface_state;
+};
+
+struct anv_image_view {
+   struct anv_image *                           image;
+   struct anv_state                             surface_state;
+};
+
+struct anv_depth_stencil_view {
+};
+
+struct anv_framebuffer {
+   uint32_t                                     color_attachment_count;
+   struct anv_color_attachment_view *           color_attachments[MAX_RTS];
+   struct anv_depth_stencil_view *              depth_stencil;
+
+   uint32_t                                     sample_count;
+   uint32_t                                     width;
+   uint32_t                                     height;
+   uint32_t                                     layers;
+};
+
+struct anv_render_pass {
+   VkRect                                       render_area;
+};
+
+
+#ifdef __cplusplus
+}
+#endif

src/vulkan/util.c

@@ -0,0 +1,99 @@
+/*
+ * 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"
+
+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.c

@@ -0,0 +1,723 @@
+#include <stdlib.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <string.h>
+
+#define VK_PROTOTYPES
+#include <vulkan/vulkan.h>
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <stdint.h>
+#include <stdarg.h>
+#include <poll.h>
+#include <libpng16/png.h>
+
+static void
+fail_if(int cond, const char *format, ...)
+{
+   va_list args;
+
+   if (!cond)
+      return;
+
+   va_start(args, format);
+   vfprintf(stderr, format, args);
+   va_end(args);
+
+   exit(1);
+}
+
+static void
+write_png(char *path, int32_t width, int32_t height, int32_t stride, void *pixels)
+{
+   FILE *f = NULL;
+   png_structp png_writer = NULL;
+   png_infop png_info = NULL;
+
+   uint8_t *rows[height];
+
+   for (int32_t y = 0; y < height; y++)
+      rows[y] = pixels + y * stride;
+
+   f = fopen(path, "wb");
+   fail_if(!f, "failed to open file for writing: %s", path);
+
+   png_writer = png_create_write_struct(PNG_LIBPNG_VER_STRING,
+                                        NULL, NULL, NULL);
+   fail_if (!png_writer, "failed to create png writer");
+
+   png_info = png_create_info_struct(png_writer);
+   fail_if(!png_info, "failed to create png writer info");
+
+   png_init_io(png_writer, f);
+   png_set_IHDR(png_writer, png_info,
+                width, height,
+                8, PNG_COLOR_TYPE_RGBA,
+                PNG_INTERLACE_NONE, PNG_COMPRESSION_TYPE_DEFAULT,
+                PNG_FILTER_TYPE_DEFAULT);
+   png_write_info(png_writer, png_info);
+   png_set_rows(png_writer, png_info, rows);
+   png_write_png(png_writer, png_info, PNG_TRANSFORM_IDENTITY, NULL);
+
+   png_destroy_write_struct(&png_writer, &png_info);
+
+   fclose(f);
+}
+
+static void *
+test_alloc(void*                                       pUserData,
+           size_t                                      size,
+           size_t                                      alignment,
+           VkSystemAllocType                           allocType)
+{
+   return malloc(size);
+}
+
+static void
+test_free(void*                                       pUserData,
+          void*                                       pMem)
+{
+   free(pMem);
+}
+
+#define GLSL(src) "#version 330\n" #src
+
+static void
+create_pipeline(VkDevice device, VkPipeline *pipeline,
+                VkPipelineLayout pipeline_layout)
+{
+   VkPipelineIaStateCreateInfo ia_create_info = {
+      .sType = VK_STRUCTURE_TYPE_PIPELINE_IA_STATE_CREATE_INFO,
+      .topology = VK_PRIMITIVE_TOPOLOGY_TRIANGLE_STRIP,
+      .disableVertexReuse = false,
+      .primitiveRestartEnable = false,
+      .primitiveRestartIndex = 0
+   };
+
+   static const char vs_source[] = GLSL(
+      layout(location = 0) in vec4 a_position;
+      layout(location = 1) in vec4 a_color;
+      layout(set = 0, index = 0) uniform block1 {
+         vec4 color;
+      } u1;
+      layout(set = 0, index = 1) uniform block2 {
+         vec4 color;
+      } u2;
+      layout(set = 1, index = 0) uniform block3 {
+         vec4 color;
+      } u3;
+      out vec4 v_color;
+      void main()
+      {
+        gl_Position = a_position;
+        v_color = a_color + u1.color + u2.color + u3.color;
+      });
+
+   static const char fs_source[] = GLSL(
+      out vec4 f_color;
+      in vec4 v_color;
+      layout(set = 0, index = 0) uniform sampler2D tex;
+      void main()
+      {
+         f_color = v_color + texture2D(tex, vec2(0.1, 0.1));
+      });
+
+   VkShader vs;
+   vkCreateShader(device,
+                  &(VkShaderCreateInfo) {
+                     .sType = VK_STRUCTURE_TYPE_SHADER_CREATE_INFO,
+                     .codeSize = sizeof(vs_source),
+                     .pCode = vs_source,
+                     .flags = 0
+                  },
+                  &vs);
+
+   VkShader fs;
+   vkCreateShader(device,
+                  &(VkShaderCreateInfo) {
+                     .sType = VK_STRUCTURE_TYPE_SHADER_CREATE_INFO,
+                     .codeSize = sizeof(fs_source),
+                     .pCode = fs_source,
+                     .flags = 0
+                  },
+                  &fs);
+
+   VkPipelineShaderStageCreateInfo vs_create_info = {
+      .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
+      .pNext = &ia_create_info,
+      .shader = {
+         .stage = VK_SHADER_STAGE_VERTEX,
+         .shader = vs,
+         .linkConstBufferCount = 0,
+         .pLinkConstBufferInfo = NULL,
+         .pSpecializationInfo = NULL
+      }
+   };
+
+   VkPipelineShaderStageCreateInfo fs_create_info = {
+      .sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
+      .pNext = &vs_create_info,
+      .shader = {
+         .stage = VK_SHADER_STAGE_FRAGMENT,
+         .shader = fs,
+         .linkConstBufferCount = 0,
+         .pLinkConstBufferInfo = NULL,
+         .pSpecializationInfo = NULL
+      }
+   };
+
+   VkPipelineVertexInputCreateInfo vi_create_info = {
+      .sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_CREATE_INFO,
+      .pNext = &fs_create_info,
+      .bindingCount = 2,
+      .pVertexBindingDescriptions = (VkVertexInputBindingDescription[]) {
+         {
+            .binding = 0,
+            .strideInBytes = 16,
+            .stepRate = VK_VERTEX_INPUT_STEP_RATE_VERTEX
+         },
+         {
+            .binding = 1,
+            .strideInBytes = 0,
+            .stepRate = VK_VERTEX_INPUT_STEP_RATE_VERTEX
+         }
+      },
+      .attributeCount = 2,
+      .pVertexAttributeDescriptions = (VkVertexInputAttributeDescription[]) {
+         {
+            .location = 0,
+            .binding = 0,
+            .format = VK_FORMAT_R32G32B32A32_SFLOAT,
+            .offsetInBytes = 0
+         },
+         {
+            .location = 1,
+            .binding = 1,
+            .format = VK_FORMAT_R32G32B32A32_SFLOAT,
+            .offsetInBytes = 0
+         }
+      }      
+   };
+
+   VkPipelineRsStateCreateInfo rs_create_info = {
+      .sType = VK_STRUCTURE_TYPE_PIPELINE_RS_STATE_CREATE_INFO,
+      .pNext = &vi_create_info,
+
+      .depthClipEnable = true,
+      .rasterizerDiscardEnable = false,
+      .fillMode = VK_FILL_MODE_SOLID,
+      .cullMode = VK_CULL_MODE_NONE,
+      .frontFace = VK_FRONT_FACE_CCW
+   };
+
+   vkCreateGraphicsPipeline(device,
+                            &(VkGraphicsPipelineCreateInfo) {
+                               .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
+                               .pNext = &rs_create_info,
+                               .flags = 0,
+                               .layout = pipeline_layout
+                            },
+                            pipeline);
+
+
+   vkDestroyObject(device, VK_OBJECT_TYPE_SHADER, fs);
+   vkDestroyObject(device, VK_OBJECT_TYPE_SHADER, vs);
+}
+
+int main(int argc, char *argv[])
+{
+   VkInstance instance;
+   vkCreateInstance(&(VkInstanceCreateInfo) {
+         .sType = VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO,
+         .pAllocCb = &(VkAllocCallbacks) {
+            .pUserData = NULL,
+            .pfnAlloc = test_alloc,
+            .pfnFree = test_free
+         },
+         .pAppInfo = &(VkApplicationInfo) {
+            .sType = VK_STRUCTURE_TYPE_APPLICATION_INFO,
+            .pAppName = "vk",
+            .apiVersion = 1
+         }
+      },
+      &instance);
+
+   uint32_t count = 1;
+   VkPhysicalDevice physicalDevices[1];
+   vkEnumeratePhysicalDevices(instance, &count, physicalDevices);
+   printf("%d physical devices\n", count);
+
+   VkPhysicalDeviceProperties properties;
+   size_t size = sizeof(properties);
+   vkGetPhysicalDeviceInfo(physicalDevices[0],
+                           VK_PHYSICAL_DEVICE_INFO_TYPE_PROPERTIES,
+                           &size, &properties);
+   printf("vendor id %04x, device name %s\n",
+          properties.vendorId, properties.deviceName);
+
+   VkDevice device;
+   vkCreateDevice(physicalDevices[0],
+                  &(VkDeviceCreateInfo) {
+                     .sType = VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO,
+                     .queueRecordCount = 1,
+                     .pRequestedQueues = &(VkDeviceQueueCreateInfo) {
+                        .queueNodeIndex = 0,
+                        .queueCount = 1                       
+                     }
+                  },
+                  &device);
+
+   VkQueue queue;
+   vkGetDeviceQueue(device, 0, 0, &queue);
+
+   VkCmdBuffer cmdBuffer;
+   vkCreateCommandBuffer(device,
+                         &(VkCmdBufferCreateInfo) {
+                            .sType = VK_STRUCTURE_TYPE_CMD_BUFFER_CREATE_INFO,
+                            .queueNodeIndex = 0,
+                            .flags = 0
+                         },
+                         &cmdBuffer);
+
+
+   VkDescriptorSetLayout set_layout[2];
+   vkCreateDescriptorSetLayout(device,
+                               &(VkDescriptorSetLayoutCreateInfo) {
+                                  .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+                                  .count = 2,
+                                  .pBinding = (VkDescriptorSetLayoutBinding[]) {
+                                     {
+                                        .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
+                                        .count = 2,
+                                        .stageFlags = VK_SHADER_STAGE_VERTEX_BIT,
+                                        .pImmutableSamplers = NULL
+                                     },
+                                     {
+                                        .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
+                                        .count = 1,
+                                        .stageFlags = VK_SHADER_STAGE_FRAGMENT_BIT,
+                                        .pImmutableSamplers = NULL
+                                     }
+                                  }
+                               },
+                               &set_layout[0]);
+
+   vkCreateDescriptorSetLayout(device,
+                               &(VkDescriptorSetLayoutCreateInfo) {
+                                  .sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
+                                  .count = 1,
+                                  .pBinding = (VkDescriptorSetLayoutBinding[]) {
+                                     {
+                                        .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
+                                        .count = 1,
+                                        .stageFlags = VK_SHADER_STAGE_VERTEX_BIT,
+                                        .pImmutableSamplers = NULL
+                                     }
+                                  }
+                               },
+                               &set_layout[1]);
+
+   VkPipelineLayout pipeline_layout;
+   vkCreatePipelineLayout(device,
+                          &(VkPipelineLayoutCreateInfo) {
+                             .sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
+                             .descriptorSetCount = 2,
+                             .pSetLayouts = set_layout,
+                          },
+                          &pipeline_layout);
+
+   VkPipeline pipeline;
+   create_pipeline(device, &pipeline, pipeline_layout);
+
+   VkDescriptorSet set[2];
+   vkAllocDescriptorSets(device, 0 /* pool */,
+                         VK_DESCRIPTOR_SET_USAGE_STATIC,
+                         2, set_layout, set, &count);
+
+   VkBuffer buffer;
+   vkCreateBuffer(device,
+                  &(VkBufferCreateInfo) {
+                     .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+                     .size = 1024,
+                     .usage = VK_BUFFER_USAGE_GENERAL,
+                     .flags = 0
+                  },
+                  &buffer);
+
+   VkMemoryRequirements buffer_requirements;
+   size = sizeof(buffer_requirements);
+   vkGetObjectInfo(device, VK_OBJECT_TYPE_BUFFER, buffer,
+                   VK_OBJECT_INFO_TYPE_MEMORY_REQUIREMENTS,
+                   &size, &buffer_requirements);
+
+   int32_t width = 256, height = 256;
+
+   VkImage rt;
+   vkCreateImage(device,
+                 &(VkImageCreateInfo) {
+                    .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+                    .imageType = VK_IMAGE_TYPE_2D,
+                    .format = VK_FORMAT_R8G8B8A8_UNORM,
+                    .extent = { .width = width, .height = height, .depth = 1 },
+                    .mipLevels = 1,
+                    .arraySize = 1,
+                    .samples = 1,
+                    .tiling = VK_IMAGE_TILING_LINEAR,
+                    .usage = VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT,
+                    .flags = 0,
+                 },
+                 &rt);
+
+   VkMemoryRequirements rt_requirements;
+   size = sizeof(rt_requirements);
+   vkGetObjectInfo(device, VK_OBJECT_TYPE_IMAGE, rt,
+                   VK_OBJECT_INFO_TYPE_MEMORY_REQUIREMENTS,
+                   &size, &rt_requirements);
+
+   VkBuffer vertex_buffer;
+   vkCreateBuffer(device,
+                  &(VkBufferCreateInfo) {
+                     .sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
+                     .size = 1024,
+                     .usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT,
+                     .flags = 0
+                  },
+                  &vertex_buffer);
+
+   VkMemoryRequirements vb_requirements;
+   size = sizeof(vb_requirements);
+   vkGetObjectInfo(device, VK_OBJECT_TYPE_BUFFER, vertex_buffer,
+                   VK_OBJECT_INFO_TYPE_MEMORY_REQUIREMENTS,
+                   &size, &vb_requirements);
+
+   printf("buffer size: %lu, buffer alignment: %lu\n",
+          buffer_requirements.size, buffer_requirements.alignment);
+   printf("rt size: %lu, rt alignment: %lu\n",
+          rt_requirements.size, rt_requirements.alignment);
+   printf("vb size: %lu vb alignment: %lu\n",
+          vb_requirements.size, vb_requirements.alignment);
+
+   size_t mem_size = rt_requirements.size + 2048 + 16 * 16 * 4;
+   VkDeviceMemory mem;
+   vkAllocMemory(device,
+                 &(VkMemoryAllocInfo) {
+                    .sType = VK_STRUCTURE_TYPE_MEMORY_ALLOC_INFO,
+                    .allocationSize = mem_size,
+                    .memProps = VK_MEMORY_PROPERTY_HOST_DEVICE_COHERENT_BIT,
+                    .memPriority = VK_MEMORY_PRIORITY_NORMAL
+                 },
+                 &mem);
+
+   void *map;
+   vkMapMemory(device, mem, 0, mem_size, 0, &map);
+   memset(map, 192, mem_size);
+
+   vkQueueBindObjectMemory(queue, VK_OBJECT_TYPE_BUFFER,
+                           buffer,
+                           0, /* allocation index; for objects which need to bind to multiple mems */
+                           mem, 128);
+
+   float color[12] = {
+      0.0, 0.2, 0.0, 0.0,
+      0.0, 0.0, 0.5, 0.0,
+      0.0, 0.0, 0.5, 0.5
+   };
+   memcpy(map + 128 + 16, color, sizeof(color));
+   VkBufferView buffer_view[3];
+   vkCreateBufferView(device,
+                      &(VkBufferViewCreateInfo) {
+                         .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
+                         .buffer = buffer,
+                         .viewType = VK_BUFFER_VIEW_TYPE_RAW,
+                         .format = VK_FORMAT_R32G32B32A32_SFLOAT,
+                         .offset = 16,
+                         .range = 64
+                      },
+                      &buffer_view[0]);
+
+   vkCreateBufferView(device,
+                      &(VkBufferViewCreateInfo) {
+                         .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
+                         .buffer = buffer,
+                         .viewType = VK_BUFFER_VIEW_TYPE_RAW,
+                         .format = VK_FORMAT_R32G32B32A32_SFLOAT,
+                         .offset = 32,
+                         .range = 64
+                      },
+                      &buffer_view[1]);
+
+   vkCreateBufferView(device,
+                      &(VkBufferViewCreateInfo) {
+                         .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO,
+                         .buffer = buffer,
+                         .viewType = VK_BUFFER_VIEW_TYPE_RAW,
+                         .format = VK_FORMAT_R32G32B32A32_SFLOAT,
+                         .offset = 48,
+                         .range = 64
+                      },
+                      &buffer_view[2]);
+
+   vkQueueBindObjectMemory(queue, VK_OBJECT_TYPE_BUFFER,
+                           vertex_buffer,
+                           0, /* allocation index; for objects which need to bind to multiple mems */
+                           mem, 1024);
+   static const float vertex_data[] = {
+      /* Triangle coordinates */
+      -0.5, -0.5, 0.0, 1.0,
+       0.5, -0.5, 0.0, 1.0,
+       0.0,  0.5, 0.0, 1.0,
+      /* Color */
+       1.0,  0.0, 0.0, 0.2,
+   };
+   memcpy(map + 1024, vertex_data, sizeof(vertex_data));
+
+   VkDynamicVpState vp_state;
+   vkCreateDynamicViewportState(device,
+                                &(VkDynamicVpStateCreateInfo) {
+                                   .sType = VK_STRUCTURE_TYPE_DYNAMIC_VP_STATE_CREATE_INFO,
+                                   .viewportAndScissorCount = 2,
+                                   .pViewports = (VkViewport[]) {
+                                      {
+                                         .originX = 0,
+                                         .originY = 0,
+                                         .width = width,
+                                         .height = height,
+                                         .minDepth = 0,
+                                         .maxDepth = 1
+                                      },
+                                      {
+                                         .originX = -10,
+                                         .originY = -10,
+                                         .width = 20,
+                                         .height = 20,
+                                         .minDepth = -1,
+                                         .maxDepth = 1
+                                      },
+                                   },
+                                   .pScissors = (VkRect[]) {
+                                      { {  0,  0 }, { width, height } },
+                                      { { 10, 10 }, { 236, 236 } }
+                                   }
+                                },
+                                &vp_state);
+
+   VkDynamicRsState rs_state;
+   vkCreateDynamicRasterState(device,
+                              &(VkDynamicRsStateCreateInfo) {
+                                 .sType = VK_STRUCTURE_TYPE_DYNAMIC_RS_STATE_CREATE_INFO,
+                              },
+                              &rs_state);
+   
+   /* FIXME: Need to query memory info before binding to memory */
+   vkQueueBindObjectMemory(queue, VK_OBJECT_TYPE_IMAGE,
+                           rt,
+                           0, /* allocation index; for objects which need to bind to multiple mems */
+                           mem, 2048);
+
+   const uint32_t texture_width = 16, texture_height = 16;
+   VkImage texture;
+   vkCreateImage(device,
+                 &(VkImageCreateInfo) {
+                    .sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
+                    .imageType = VK_IMAGE_TYPE_2D,
+                    .format = VK_FORMAT_R8G8B8A8_UNORM,
+                    .extent = { .width = texture_width, .height = texture_height, .depth = 1 },
+                    .mipLevels = 1,
+                    .arraySize = 1,
+                    .samples = 1,
+                    .tiling = VK_IMAGE_TILING_LINEAR,
+                    .usage = VK_IMAGE_USAGE_SAMPLED_BIT,
+                    .flags = 0,
+                 },
+                 &texture);
+
+   VkImageView image_view;
+   vkCreateImageView(device,
+                     &(VkImageViewCreateInfo) {
+                        .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
+                        .image = texture,
+                        .viewType = VK_IMAGE_VIEW_TYPE_2D,
+                        .format = VK_FORMAT_R8G8B8A8_UNORM,
+                        .channels = {
+                           VK_CHANNEL_SWIZZLE_R,
+                           VK_CHANNEL_SWIZZLE_G,
+                           VK_CHANNEL_SWIZZLE_B,
+                           VK_CHANNEL_SWIZZLE_A
+                        },
+                        .subresourceRange = {
+                           .aspect = VK_IMAGE_ASPECT_COLOR,
+                           .baseMipLevel = 0,
+                           .mipLevels = 1,
+                           .baseArraySlice = 0,
+                           .arraySize = 1
+                        },
+                        .minLod = 0
+                      },
+                      &image_view);
+
+   vkQueueBindObjectMemory(queue, VK_OBJECT_TYPE_IMAGE,
+                           texture,
+                           0, /* allocation index; for objects which need to bind to multiple mems */
+                           mem, 2048 + 256 * 256 * 4);
+
+   vkUpdateDescriptors(device, set[0], 2,
+                       (const void * []) {
+                          &(VkUpdateBuffers) {
+                             .sType = VK_STRUCTURE_TYPE_UPDATE_BUFFERS,
+                             .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
+                             .arrayIndex = 0,
+                             .binding = 0,
+                             .count = 2,
+                             .pBufferViews = (VkBufferViewAttachInfo[]) {
+                                {
+                                   .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_ATTACH_INFO,
+                                   .view = buffer_view[0]
+                                },
+                                {
+                                   .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_ATTACH_INFO,
+                                   .view = buffer_view[1]
+                                }
+                             }
+                          },
+                          &(VkUpdateImages) {
+                             .sType = VK_STRUCTURE_TYPE_UPDATE_IMAGES,
+                             .descriptorType = VK_DESCRIPTOR_TYPE_SAMPLED_IMAGE,
+                             .binding = 2,
+                             .count = 1,
+                             .pImageViews = (VkImageViewAttachInfo[]) {
+                                {
+                                   .sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_ATTACH_INFO,
+                                   .view = image_view,
+                                   .layout = VK_IMAGE_LAYOUT_GENERAL,
+                                }
+                             }
+                          }
+                       });
+
+   vkUpdateDescriptors(device, set[1], 1,
+                       (const void * []) {
+                          &(VkUpdateBuffers) {
+                             .sType = VK_STRUCTURE_TYPE_UPDATE_BUFFERS,
+                             .descriptorType = VK_DESCRIPTOR_TYPE_UNIFORM_BUFFER,
+                             .arrayIndex = 0,
+                             .count = 1,
+                             .pBufferViews = (VkBufferViewAttachInfo[]) {
+                                {
+                                   .sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_ATTACH_INFO,
+                                   .view = buffer_view[2]
+                                }
+                             }
+                          }
+                       });
+
+   VkColorAttachmentView view;
+   vkCreateColorAttachmentView(device,
+                               &(VkColorAttachmentViewCreateInfo) {
+                                  .sType = VK_STRUCTURE_TYPE_COLOR_ATTACHMENT_VIEW_CREATE_INFO,
+                                  .image = rt,
+                                  .format = VK_FORMAT_R8G8B8A8_UNORM,
+                                  .mipLevel = 0,
+                                  .baseArraySlice = 0,
+                                  .arraySize = 1,
+                                  .msaaResolveImage = 0,
+                                  .msaaResolveSubResource = { 0, }
+                               },
+                               &view);
+   
+   VkFramebuffer framebuffer;
+   vkCreateFramebuffer(device,
+                       &(VkFramebufferCreateInfo) {
+                          .sType = VK_STRUCTURE_TYPE_FRAMEBUFFER_CREATE_INFO,
+                          .colorAttachmentCount = 1,
+                          .pColorAttachments = (VkColorAttachmentBindInfo[]) {
+                             {
+                                .view = view,
+                                .layout = VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL
+                             }
+                          },
+                          .pDepthStencilAttachment = NULL,
+                          .sampleCount = 1,
+                          .width = width,
+                          .height = height,
+                          .layers = 1
+                       },
+                       &framebuffer);
+
+   VkRenderPass pass;
+   vkCreateRenderPass(device,
+                      &(VkRenderPassCreateInfo) {
+                         .sType = VK_STRUCTURE_TYPE_RENDER_PASS_CREATE_INFO,
+                         .renderArea = { { 0, 0 }, { width, height } },
+                         .colorAttachmentCount = 1,
+                         .extent = { },
+                         .sampleCount = 1,
+                         .layers = 1,
+                         .pColorFormats = (VkFormat[]) { VK_FORMAT_R8G8B8A8_UNORM },
+                         .pColorLayouts = (VkImageLayout[]) { VK_IMAGE_LAYOUT_COLOR_ATTACHMENT_OPTIMAL },
+                         .pColorLoadOps = (VkAttachmentLoadOp[]) { VK_ATTACHMENT_LOAD_OP_CLEAR },
+                         .pColorStoreOps = (VkAttachmentStoreOp[]) { VK_ATTACHMENT_STORE_OP_STORE },
+                         .pColorLoadClearValues = (VkClearColor[]) {
+                            { .color = { .floatColor = { 1.0, 0.0, 0.0, 1.0 } }, .useRawValue = false }
+                         },
+                         .depthStencilFormat = VK_FORMAT_UNDEFINED,
+                      },
+                      &pass);
+
+   vkBeginCommandBuffer(cmdBuffer,
+                        &(VkCmdBufferBeginInfo) {
+                           .sType = VK_STRUCTURE_TYPE_CMD_BUFFER_BEGIN_INFO,
+                           .flags = 0
+                        });
+
+   vkCmdBeginRenderPass(cmdBuffer,
+                        &(VkRenderPassBegin) {
+                           .renderPass = pass,
+                           .framebuffer = framebuffer
+                        });
+
+   vkCmdBindVertexBuffers(cmdBuffer, 0, 2,
+                          (VkBuffer[]) { vertex_buffer, vertex_buffer },
+                          (VkDeviceSize[]) { 0, 3 * 4 * sizeof(float) });
+
+   vkCmdBindPipeline(cmdBuffer, VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
+
+   vkCmdBindDescriptorSets(cmdBuffer,
+                           VK_PIPELINE_BIND_POINT_GRAPHICS, 0, 1,
+                           &set[0], 0, NULL);
+   vkCmdBindDescriptorSets(cmdBuffer,
+                           VK_PIPELINE_BIND_POINT_GRAPHICS, 1, 1,
+                           &set[1], 0, NULL);
+
+   vkCmdBindDynamicStateObject(cmdBuffer,
+                               VK_STATE_BIND_POINT_VIEWPORT, vp_state);
+   vkCmdBindDynamicStateObject(cmdBuffer,
+                               VK_STATE_BIND_POINT_RASTER, rs_state);
+
+   vkCmdWriteTimestamp(cmdBuffer, VK_TIMESTAMP_TYPE_TOP, buffer, 0);
+   vkCmdWriteTimestamp(cmdBuffer, VK_TIMESTAMP_TYPE_BOTTOM, buffer, 8);
+
+   vkCmdDraw(cmdBuffer, 0, 3, 0, 1);
+
+   vkCmdEndRenderPass(cmdBuffer, pass);
+
+   vkEndCommandBuffer(cmdBuffer);
+
+   vkQueueSubmit(queue, 1, &cmdBuffer, 0);
+
+   vkQueueWaitIdle(queue);
+
+   write_png("vk.png", width, height, 1024, map + 2048);
+
+   vkDestroyObject(device, VK_OBJECT_TYPE_IMAGE, texture);
+   vkDestroyObject(device, VK_OBJECT_TYPE_IMAGE, rt);
+   vkDestroyObject(device, VK_OBJECT_TYPE_BUFFER, buffer);
+   vkDestroyObject(device, VK_OBJECT_TYPE_COMMAND_BUFFER, cmdBuffer);
+   vkDestroyObject(device, VK_OBJECT_TYPE_PIPELINE, pipeline);
+
+   vkDestroyDevice(device);
+
+   vkDestroyInstance(instance);
+
+   return 0;
+}