ci: Use arch-mingw-github-action v8

Fixes safe directory stuff giving invalid version info.

.github/workflows/artifacts.yml

@@ -15,11 +15,10 @@ jobs:
 
     - name: Build release
       id: build-release
-      uses: Joshua-Ashton/arch-mingw-github-action@v7
+      uses: Joshua-Ashton/arch-mingw-github-action@v8
       with:
         command: |
           export VERSION_NAME="${GITHUB_REF##*/}-${GITHUB_SHA##*/}"
-          git config --global --add safe.directory "$GITHUB_WORKSPACE"
           ./package-release.sh ${VERSION_NAME} build --no-package
           echo "VERSION_NAME=${VERSION_NAME}" >> $GITHUB_ENV
 

.github/workflows/test-build-linux.yml

@@ -18,64 +18,58 @@ jobs:
 
     - name: Build MinGW x86
       id: build-mingw-x86
-      uses: Joshua-Ashton/arch-mingw-github-action@v7
+      uses: Joshua-Ashton/arch-mingw-github-action@v8
       with:
         command: |
-          git config --global --add safe.directory "$GITHUB_WORKSPACE"
           meson -Denable_tests=True -Denable_extras=True --cross-file=build-win32.txt --buildtype release build-mingw-x86
           ninja -C build-mingw-x86
 
     - name: Build MinGW x64
       id: build-mingw-x64
-      uses: Joshua-Ashton/arch-mingw-github-action@v7
+      uses: Joshua-Ashton/arch-mingw-github-action@v8
       with:
         command: |
-          git config --global --add safe.directory "$GITHUB_WORKSPACE"
           meson -Denable_tests=True -Denable_extras=True --cross-file=build-win64.txt --buildtype release build-mingw-x64
           ninja -C build-mingw-x64
 
     - name: Build Native GCC x86
       id: build-native-gcc-x86
-      uses: Joshua-Ashton/arch-mingw-github-action@v7
+      uses: Joshua-Ashton/arch-mingw-github-action@v8
       with:
         command: |
           export CC="gcc -m32"
           export CXX="g++ -m32"
           export PKG_CONFIG_PATH="/usr/lib32/pkgconfig:/usr/lib/i386-linux-gnu/pkgconfig:/usr/lib/pkgconfig"
-          git config --global --add safe.directory "$GITHUB_WORKSPACE"
           meson -Denable_tests=True -Denable_extras=True --buildtype release build-native-gcc-x86
           ninja -C build-native-gcc-x86
 
     - name: Build Native GCC x64
       id: build-native-gcc-x64
-      uses: Joshua-Ashton/arch-mingw-github-action@v7
+      uses: Joshua-Ashton/arch-mingw-github-action@v8
       with:
         command: |
           export CC="gcc"
           export CXX="g++"
-          git config --global --add safe.directory "$GITHUB_WORKSPACE"
           meson -Denable_tests=True -Denable_extras=True --buildtype release build-native-gcc-x64
           ninja -C build-native-gcc-x64
 
     - name: Build Native Clang x86
       id: build-native-clang-x86
-      uses: Joshua-Ashton/arch-mingw-github-action@v7
+      uses: Joshua-Ashton/arch-mingw-github-action@v8
       with:
         command: |
           export CC="clang -m32"
           export CXX="clang++ -m32"
           export PKG_CONFIG_PATH="/usr/lib32/pkgconfig:/usr/lib/i386-linux-gnu/pkgconfig:/usr/lib/pkgconfig"
-          git config --global --add safe.directory "$GITHUB_WORKSPACE"
           meson -Denable_tests=True -Denable_extras=True --buildtype release build-native-clang-x86
           ninja -C build-native-clang-x86
 
     - name: Build Native Clang x64
       id: build-native-clang-x64
-      uses: Joshua-Ashton/arch-mingw-github-action@v7
+      uses: Joshua-Ashton/arch-mingw-github-action@v8
       with:
         command: |
           export CC="clang"
           export CXX="clang++"
-          git config --global --add safe.directory "$GITHUB_WORKSPACE"
           meson -Denable_tests=True -Denable_extras=True --buildtype release build-native-clang-x64
           ninja -C build-native-clang-x64

include/private/vkd3d_common.h

@@ -311,4 +311,19 @@ static inline uint64_t vkd3d_get_current_time_ns(void)
 #endif
 }
 
+#ifdef _MSC_VER
+#pragma intrinsic(__rdtsc)
+#endif
+
+static inline uint64_t vkd3d_get_current_time_ticks(void)
+{
+#ifdef _MSC_VER
+    return __rdtsc();
+#elif defined(__i386__) || defined(__x86_64__)
+    return __builtin_ia32_rdtsc();
+#else
+    return vkd3d_get_current_time_ns();
+#endif
+}
+
 #endif  /* __VKD3D_COMMON_H */

include/private/vkd3d_profiling.h

@@ -25,13 +25,6 @@
 
 #ifdef VKD3D_ENABLE_PROFILING
 
-#ifdef _WIN32
-#define WIN32_LEAN_AND_MEAN
-#include <windows.h>
-#else
-#include <time.h>
-#endif
-
 void vkd3d_init_profiling(void);
 bool vkd3d_uses_profiling(void);
 unsigned int vkd3d_profiling_register_region(const char *name, spinlock_t *lock, uint32_t *latch);
@@ -48,12 +41,12 @@ void vkd3d_profiling_notify_work(unsigned int index, uint64_t start_ticks, uint6
     do { \
         if (!(_vkd3d_region_index_##name = vkd3d_atomic_uint32_load_explicit(&_vkd3d_region_latch_##name, vkd3d_memory_order_acquire))) \
             _vkd3d_region_index_##name = vkd3d_profiling_register_region(#name, &_vkd3d_region_lock_##name, &_vkd3d_region_latch_##name); \
-        _vkd3d_region_begin_tick_##name = vkd3d_get_current_time_ns(); \
+        _vkd3d_region_begin_tick_##name = vkd3d_get_current_time_ticks(); \
     } while(0)
 
 #define VKD3D_REGION_END_ITERATIONS(name, iter) \
     do { \
-        _vkd3d_region_end_tick_##name = vkd3d_get_current_time_ns(); \
+        _vkd3d_region_end_tick_##name = vkd3d_get_current_time_ticks(); \
         vkd3d_profiling_notify_work(_vkd3d_region_index_##name, _vkd3d_region_begin_tick_##name, _vkd3d_region_end_tick_##name, iter); \
     } while(0)
 

include/private/vkd3d_string.h

@@ -30,6 +30,7 @@ WCHAR *vkd3d_dup_demangled_entry_point(const char *str);
 char *vkd3d_dup_demangled_entry_point_ascii(const char *str);
 
 bool vkd3d_export_strequal(const WCHAR *a, const WCHAR *b);
+bool vkd3d_export_strequal_mixed(const WCHAR *a, const char *b);
 bool vkd3d_export_strequal_substr(const WCHAR *a, size_t n, const WCHAR *b);
 
 char *vkd3d_strdup(const char *str);

include/shader-debug/GNUmakefile

@@ -6,6 +6,12 @@ COMP_SOURCES := $(wildcard $(M)/*.comp)
 TESC_SOURCES := $(wildcard $(M)/*.tesc)
 TESE_SOURCES := $(wildcard $(M)/*.tese)
 GEOM_SOURCES := $(wildcard $(M)/*.geom)
+RGEN_SOURCES := $(wildcard $(M)/*.rgen)
+RINT_SOURCES := $(wildcard $(M)/*.rint)
+RAHIT_SOURCES := $(wildcard $(M)/*.rahit)
+RCHIT_SOURCES := $(wildcard $(M)/*.rchit)
+RMISS_SOURCES := $(wildcard $(M)/*.rmiss)
+RCALL_SOURCES := $(wildcard $(M)/*.rcall)
 
 SPV_OBJECTS := \
 			   $(VERT_SOURCES:.vert=.spv) \
@@ -13,25 +19,49 @@ SPV_OBJECTS := \
 			   $(COMP_SOURCES:.comp=.spv) \
 			   $(TESC_SOURCES:.tesc=.spv) \
 			   $(TESE_SOURCES:.tese=.spv) \
-			   $(GEOM_SOURCES:.geom=.spv)
+			   $(GEOM_SOURCES:.geom=.spv) \
+			   $(RGEN_SOURCES:.rgen=.spv) \
+			   $(RINT_SOURCES:.rint=.spv) \
+			   $(RAHIT_SOURCES:.rahit=.spv) \
+			   $(RCHIT_SOURCES:.rchit=.spv) \
+			   $(RMISS_SOURCES:.rmiss=.spv) \
+			   $(RCALL_SOURCES:.rcall=.spv)
 
 %.spv: %.vert
-	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 $(GLSLC_FLAGS)
 
 %.spv: %.frag
-	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 -DDEBUG_CHANNEL_HELPER_LANES
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 -DDEBUG_CHANNEL_HELPER_LANES $(GLSLC_FLAGS)
 
 %.spv: %.comp
-	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 $(GLSLC_FLAGS)
 
 %.spv: %.geom
-	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 $(GLSLC_FLAGS)
 
 %.spv: %.tesc
-	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 $(GLSLC_FLAGS)
 
 %.spv: %.tese
-	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 $(GLSLC_FLAGS)
+
+%.spv: %.rgen
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 --target-spv=spv1.4 $(GLSLC_FLAGS)
+
+%.spv: %.rint
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 --target-spv=spv1.4 $(GLSLC_FLAGS)
+
+%.spv: %.rahit
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 --target-spv=spv1.4 $(GLSLC_FLAGS)
+
+%.spv: %.rchit
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 --target-spv=spv1.4 $(GLSLC_FLAGS)
+
+%.spv: %.rmiss
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 --target-spv=spv1.4 $(GLSLC_FLAGS)
+
+%.spv: %.rcall
+	glslc -o $@ $< -I$(INCLUDE_DIR) --target-env=vulkan1.1 --target-spv=spv1.4 $(GLSLC_FLAGS)
 
 all: $(SPV_OBJECTS)
 

include/shader-debug/debug_channel.h

@@ -97,6 +97,14 @@ void DEBUG_CHANNEL_INIT(uvec3 id)
 #endif
 }
 
+void DEBUG_CHANNEL_INIT_IMPLICIT_INSTANCE(uvec3 id, uint inst)
+{
+	if (!DEBUG_SHADER_RING_ACTIVE)
+		return;
+	DEBUG_CHANNEL_ID = id;
+	DEBUG_CHANNEL_INSTANCE_COUNTER = inst;
+}
+
 void DEBUG_CHANNEL_UNLOCK_MESSAGE(RingBuffer buf, uint offset, uint num_words)
 {
 	memoryBarrierBuffer();

include/vkd3d.h

@@ -89,6 +89,9 @@ extern "C" {
 #define VKD3D_CONFIG_FLAG_SHADER_CACHE_SYNC (1ull << 27)
 #define VKD3D_CONFIG_FLAG_FORCE_RAW_VA_CBV (1ull << 28)
 #define VKD3D_CONFIG_FLAG_ZERO_MEMORY_WORKAROUNDS_COMMITTED_BUFFER_UAV (1ull << 29)
+#define VKD3D_CONFIG_FLAG_ALLOW_SBT_COLLECTION (1ull << 30)
+#define VKD3D_CONFIG_FLAG_FORCE_NATIVE_FP16 (1ull << 31)
+#define VKD3D_CONFIG_FLAG_USE_HOST_IMPORT_FALLBACK (1ull << 32)
 
 typedef HRESULT (*PFN_vkd3d_signal_event)(HANDLE event);
 

include/vkd3d_d3d12.idl

@@ -3644,8 +3644,8 @@ interface ID3D12CommandQueue : ID3D12Pageable
             ID3D12Heap *heap,
             UINT range_count,
             const D3D12_TILE_RANGE_FLAGS *range_flags,
-            UINT *heap_range_offsets,
-            UINT *range_tile_counts,
+            const UINT *heap_range_offsets,
+            const UINT *range_tile_counts,
             D3D12_TILE_MAPPING_FLAGS flags);
 
     void CopyTileMappings(ID3D12Resource *dst_resource,

include/vkd3d_shader.h

@@ -660,6 +660,7 @@ struct vkd3d_shader_scan_info
     bool has_side_effects;
     bool needs_late_zs;
     bool discards;
+    bool has_uav_counter;
     unsigned int patch_vertex_count;
 };
 
@@ -753,7 +754,11 @@ int vkd3d_shader_compile_dxbc(const struct vkd3d_shader_code *dxbc,
 void vkd3d_shader_free_shader_code(struct vkd3d_shader_code *code);
 
 int vkd3d_shader_parse_root_signature(const struct vkd3d_shader_code *dxbc,
-        struct vkd3d_versioned_root_signature_desc *root_signature);
+        struct vkd3d_versioned_root_signature_desc *root_signature,
+        vkd3d_shader_hash_t *compatibility_hash);
+int vkd3d_shader_parse_root_signature_raw(const char *data, unsigned int data_size,
+        struct vkd3d_versioned_root_signature_desc *desc,
+        vkd3d_shader_hash_t *compatibility_hash);
 void vkd3d_shader_free_root_signature(struct vkd3d_versioned_root_signature_desc *root_signature);
 
 /* FIXME: Add support for returning error messages (ID3DBlob). */
@@ -779,19 +784,65 @@ void vkd3d_shader_free_shader_signature(struct vkd3d_shader_signature *signature
 struct vkd3d_shader_library_entry_point
 {
     unsigned int identifier;
+    VkShaderStageFlagBits stage;
     WCHAR *mangled_entry_point;
     WCHAR *plain_entry_point;
     char *real_entry_point;
-    VkShaderStageFlagBits stage;
 };
 
-int vkd3d_shader_dxil_append_library_entry_points(
+enum vkd3d_shader_subobject_kind
+{
+    /* Matches DXIL for simplicity. */
+    VKD3D_SHADER_SUBOBJECT_KIND_STATE_OBJECT_CONFIG = 0,
+    VKD3D_SHADER_SUBOBJECT_KIND_GLOBAL_ROOT_SIGNATURE = 1,
+    VKD3D_SHADER_SUBOBJECT_KIND_LOCAL_ROOT_SIGNATURE = 2,
+    VKD3D_SHADER_SUBOBJECT_KIND_SUBOBJECT_TO_EXPORTS_ASSOCIATION = 8,
+    VKD3D_SHADER_SUBOBJECT_KIND_RAYTRACING_SHADER_CONFIG = 9,
+    VKD3D_SHADER_SUBOBJECT_KIND_RAYTRACING_PIPELINE_CONFIG = 10,
+    VKD3D_SHADER_SUBOBJECT_KIND_HIT_GROUP = 11,
+    VKD3D_SHADER_SUBOBJECT_KIND_RAYTRACING_PIPELINE_CONFIG1 = 12,
+};
+
+struct vkd3d_shader_library_subobject
+{
+    enum vkd3d_shader_subobject_kind kind;
+    unsigned int dxil_identifier;
+
+    /* All const pointers here point directly to the DXBC blob,
+     * so they do not need to be freed.
+     * Fortunately for us, the C strings are zero-terminated in the blob itself. */
+
+    /* In the blob, ASCII is used as identifier, where API uses wide strings, sigh ... */
+    const char *name;
+
+    union
+    {
+        D3D12_RAYTRACING_PIPELINE_CONFIG1 pipeline_config;
+        D3D12_RAYTRACING_SHADER_CONFIG shader_config;
+        D3D12_STATE_OBJECT_CONFIG object_config;
+
+        /* Duped strings because API wants wide strings for no good reason. */
+        D3D12_HIT_GROUP_DESC hit_group;
+        D3D12_DXIL_SUBOBJECT_TO_EXPORTS_ASSOCIATION association;
+
+        struct
+        {
+            const void *data;
+            size_t size;
+        } payload;
+    } data;
+};
+
+int vkd3d_shader_dxil_append_library_entry_points_and_subobjects(
         const D3D12_DXIL_LIBRARY_DESC *library_desc,
         unsigned int identifier,
         struct vkd3d_shader_library_entry_point **entry_points,
-        size_t *entry_point_size, size_t *entry_point_count);
+        size_t *entry_point_size, size_t *entry_point_count,
+        struct vkd3d_shader_library_subobject **subobjects,
+        size_t *subobjects_size, size_t *subobjects_count);
 
 void vkd3d_shader_dxil_free_library_entry_points(struct vkd3d_shader_library_entry_point *entry_points, size_t count);
+void vkd3d_shader_dxil_free_library_subobjects(struct vkd3d_shader_library_subobject *subobjects, size_t count);
 
 int vkd3d_shader_compile_dxil_export(const struct vkd3d_shader_code *dxil,
         const char *export,
@@ -817,7 +868,8 @@ typedef int (*PFN_vkd3d_shader_compile_dxbc)(const struct vkd3d_shader_code *dxb
 typedef void (*PFN_vkd3d_shader_free_shader_code)(struct vkd3d_shader_code *code);
 
 typedef int (*PFN_vkd3d_shader_parse_root_signature)(const struct vkd3d_shader_code *dxbc,
-        struct vkd3d_versioned_root_signature_desc *root_signature);
+        struct vkd3d_versioned_root_signature_desc *root_signature,
+        vkd3d_shader_hash_t *compatibility_hash);
 typedef void (*PFN_vkd3d_shader_free_root_signature)(struct vkd3d_versioned_root_signature_desc *root_signature);
 
 typedef int (*PFN_vkd3d_shader_serialize_root_signature)(

libs/vkd3d-common/string.c

@@ -82,6 +82,21 @@ bool vkd3d_export_strequal(const WCHAR *a, const WCHAR *b)
     return *a == *b;
 }
 
+bool vkd3d_export_strequal_mixed(const WCHAR *a, const char *b)
+{
+    if (!a || !b)
+        return false;
+
+    while (*a != '\0' && *b != '\0')
+    {
+        if (*a != *b)
+            return false;
+        a++;
+        b++;
+    }
+    return *a == *b;
+}
+
 bool vkd3d_export_strequal_substr(const WCHAR *a, size_t expected_n, const WCHAR *b)
 {
     size_t n = 0;

libs/vkd3d-shader/dxbc.c

@@ -2755,8 +2755,9 @@ static int shader_parse_static_samplers(struct root_signature_parser_context *co
     return VKD3D_OK;
 }
 
-static int shader_parse_root_signature(const char *data, unsigned int data_size,
-        struct vkd3d_versioned_root_signature_desc *desc)
+int vkd3d_shader_parse_root_signature_raw(const char *data, unsigned int data_size,
+        struct vkd3d_versioned_root_signature_desc *desc,
+        vkd3d_shader_hash_t *compatibility_hash)
 {
     struct vkd3d_root_signature_desc *v_1_0 = &desc->v_1_0;
     struct root_signature_parser_context context;
@@ -2764,6 +2765,8 @@ static int shader_parse_root_signature(const char *data, unsigned int data_size,
     const char *ptr = data;
     int ret;
 
+    memset(desc, 0, sizeof(*desc));
+
     context.data = data;
     context.data_size = data_size;
 
@@ -2835,28 +2838,46 @@ static int shader_parse_root_signature(const char *data, unsigned int data_size,
     read_uint32(&ptr, &v_1_0->flags);
     TRACE("Flags %#x.\n", v_1_0->flags);
 
+    if (compatibility_hash)
+    {
+        struct vkd3d_shader_code code = { data, data_size };
+        *compatibility_hash = vkd3d_shader_hash(&code);
+    }
+
     return VKD3D_OK;
 }
 
 static int rts0_handler(const char *data, DWORD data_size, DWORD tag, void *context)
 {
-    struct vkd3d_versioned_root_signature_desc *desc = context;
+    struct vkd3d_shader_code *payload = context;
 
     if (tag != TAG_RTS0)
         return VKD3D_OK;
 
-    return shader_parse_root_signature(data, data_size, desc);
+    payload->code = data;
+    payload->size = data_size;
+    return VKD3D_OK;
 }
 
 int vkd3d_shader_parse_root_signature(const struct vkd3d_shader_code *dxbc,
-        struct vkd3d_versioned_root_signature_desc *root_signature)
+        struct vkd3d_versioned_root_signature_desc *root_signature,
+        vkd3d_shader_hash_t *compatibility_hash)
 {
+    struct vkd3d_shader_code raw_payload;
     int ret;
 
     TRACE("dxbc {%p, %zu}, root_signature %p.\n", dxbc->code, dxbc->size, root_signature);
 
-    memset(root_signature, 0, sizeof(*root_signature));
-    if ((ret = parse_dxbc(dxbc->code, dxbc->size, rts0_handler, root_signature)) < 0)
+    memset(&raw_payload, 0, sizeof(raw_payload));
+
+    if ((ret = parse_dxbc(dxbc->code, dxbc->size, rts0_handler, &raw_payload)) < 0)
+        return ret;
+
+    if (!raw_payload.code)
+        return VKD3D_ERROR;
+
+    if ((ret = vkd3d_shader_parse_root_signature_raw(raw_payload.code, raw_payload.size,
+            root_signature, compatibility_hash)) < 0)
     {
         vkd3d_shader_free_root_signature(root_signature);
         return ret;

libs/vkd3d-shader/dxil.c

@@ -1352,6 +1352,31 @@ void vkd3d_shader_dxil_free_library_entry_points(struct vkd3d_shader_library_ent
     vkd3d_free(entry_points);
 }
 
+void vkd3d_shader_dxil_free_library_subobjects(struct vkd3d_shader_library_subobject *subobjects, size_t count)
+{
+    size_t i, j;
+
+    for (i = 0; i < count; i++)
+    {
+        if (subobjects[i].kind == VKD3D_SHADER_SUBOBJECT_KIND_SUBOBJECT_TO_EXPORTS_ASSOCIATION)
+        {
+            for (j = 0; j < subobjects[i].data.association.NumExports; j++)
+                vkd3d_free((void*)subobjects[i].data.association.pExports[j]);
+            vkd3d_free((void*)subobjects[i].data.association.pExports);
+            vkd3d_free((void*)subobjects[i].data.association.SubobjectToAssociate);
+        }
+        else if (subobjects[i].kind == VKD3D_SHADER_SUBOBJECT_KIND_HIT_GROUP)
+        {
+            vkd3d_free((void*)subobjects[i].data.hit_group.HitGroupExport);
+            vkd3d_free((void*)subobjects[i].data.hit_group.AnyHitShaderImport);
+            vkd3d_free((void*)subobjects[i].data.hit_group.ClosestHitShaderImport);
+            vkd3d_free((void*)subobjects[i].data.hit_group.IntersectionShaderImport);
+        }
+    }
+
+    vkd3d_free(subobjects);
+}
+
 static VkShaderStageFlagBits convert_stage(dxil_spv_shader_stage stage)
 {
     /* Only interested in RT entry_points. There is no way yet to use lib_6_3+ for non-RT. */
@@ -1396,20 +1421,95 @@ static bool vkd3d_dxil_build_entry(struct vkd3d_shader_library_entry_point *entr
     return true;
 }
 
-int vkd3d_shader_dxil_append_library_entry_points(
+static void vkd3d_shader_dxil_copy_subobject(unsigned int identifier,
+        struct vkd3d_shader_library_subobject *subobject,
+        const dxil_spv_rdat_subobject *dxil_subobject)
+{
+    unsigned int i;
+
+    /* Reuse same enums as DXIL. */
+    subobject->kind = (enum vkd3d_shader_subobject_kind)dxil_subobject->kind;
+    subobject->name = dxil_subobject->subobject_name;
+    subobject->dxil_identifier = identifier;
+
+    switch (dxil_subobject->kind)
+    {
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_GLOBAL_ROOT_SIGNATURE:
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_LOCAL_ROOT_SIGNATURE:
+            subobject->data.payload.data = dxil_subobject->payload;
+            subobject->data.payload.size = dxil_subobject->payload_size;
+            break;
+
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_RAYTRACING_PIPELINE_CONFIG:
+            /* Normalize the kind. */
+            subobject->kind = VKD3D_SHADER_SUBOBJECT_KIND_RAYTRACING_PIPELINE_CONFIG1;
+            subobject->data.pipeline_config.MaxTraceRecursionDepth = dxil_subobject->args[0];
+            subobject->data.pipeline_config.Flags = 0;
+            break;
+
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_RAYTRACING_PIPELINE_CONFIG1:
+            subobject->kind = VKD3D_SHADER_SUBOBJECT_KIND_RAYTRACING_PIPELINE_CONFIG1;
+            subobject->data.pipeline_config.MaxTraceRecursionDepth = dxil_subobject->args[0];
+            subobject->data.pipeline_config.Flags = dxil_subobject->args[1];
+            break;
+
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_RAYTRACING_SHADER_CONFIG:
+            subobject->data.shader_config.MaxPayloadSizeInBytes = dxil_subobject->args[0];
+            subobject->data.shader_config.MaxAttributeSizeInBytes = dxil_subobject->args[1];
+            break;
+
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_HIT_GROUP:
+            /* Enum aliases. */
+            subobject->data.hit_group.Type = (D3D12_HIT_GROUP_TYPE)dxil_subobject->hit_group_type;
+            assert(dxil_subobject->num_exports == 3);
+            /* Implementation simplifies a lot if we can reuse the D3D12 type here. */
+            subobject->data.hit_group.HitGroupExport = vkd3d_dup_entry_point(dxil_subobject->subobject_name);
+            subobject->data.hit_group.AnyHitShaderImport = dxil_subobject->exports[0] && *dxil_subobject->exports[0] != '\0' ?
+                    vkd3d_dup_entry_point(dxil_subobject->exports[0]) : NULL;
+            subobject->data.hit_group.ClosestHitShaderImport = dxil_subobject->exports[1] && *dxil_subobject->exports[1] != '\0' ?
+                    vkd3d_dup_entry_point(dxil_subobject->exports[1]) : NULL;
+            subobject->data.hit_group.IntersectionShaderImport = dxil_subobject->exports[2] && *dxil_subobject->exports[2] != '\0' ?
+                    vkd3d_dup_entry_point(dxil_subobject->exports[2]) : NULL;
+            break;
+
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_STATE_OBJECT_CONFIG:
+            subobject->data.object_config.Flags = dxil_subobject->args[0];
+            break;
+
+        case DXIL_SPV_RDAT_SUBOBJECT_KIND_SUBOBJECT_TO_EXPORTS_ASSOCIATION:
+            assert(dxil_subobject->num_exports >= 1);
+            subobject->data.association.SubobjectToAssociate = vkd3d_dup_entry_point(dxil_subobject->exports[0]);
+            subobject->data.association.pExports = vkd3d_malloc((dxil_subobject->num_exports - 1) * sizeof(LPCWSTR));
+            subobject->data.association.NumExports = dxil_subobject->num_exports - 1;
+            for (i = 1; i < dxil_subobject->num_exports; i++)
+                subobject->data.association.pExports[i - 1] = vkd3d_dup_entry_point(dxil_subobject->exports[i]);
+            break;
+
+        default:
+            FIXME("Unrecognized RDAT subobject type: %u.\n", dxil_subobject->kind);
+            break;
+    }
+}
+
+int vkd3d_shader_dxil_append_library_entry_points_and_subobjects(
         const D3D12_DXIL_LIBRARY_DESC *library_desc,
         unsigned int identifier,
         struct vkd3d_shader_library_entry_point **entry_points,
-        size_t *entry_point_size, size_t *entry_point_count)
+        size_t *entry_point_size, size_t *entry_point_count,
+        struct vkd3d_shader_library_subobject **subobjects,
+        size_t *subobjects_size, size_t *subobjects_count)
 {
     struct vkd3d_shader_library_entry_point new_entry;
+    struct vkd3d_shader_library_subobject *subobject;
     dxil_spv_parsed_blob blob = NULL;
     struct vkd3d_shader_code code;
+    dxil_spv_rdat_subobject sub;
     dxil_spv_shader_stage stage;
     const char *mangled_entry;
     char *ascii_entry = NULL;
     vkd3d_shader_hash_t hash;
-    unsigned int count, i;
+    unsigned int count, i, j;
+    unsigned int rdat_count;
     int ret = VKD3D_OK;
 
     memset(&new_entry, 0, sizeof(new_entry));
@@ -1430,6 +1530,8 @@ int vkd3d_shader_dxil_append_library_entry_points(
         goto end;
     }
 
+    rdat_count = dxil_spv_parsed_blob_get_num_rdat_subobjects(blob);
+
     if (library_desc->NumExports)
     {
         for (i = 0; i < library_desc->NumExports; i++)
@@ -1439,24 +1541,44 @@ int vkd3d_shader_dxil_append_library_entry_points(
             else
                 ascii_entry = vkd3d_strdup_w_utf8(library_desc->pExports[i].Name, 0);
 
-            stage = dxil_spv_parsed_blob_get_shader_stage_for_entry(blob, ascii_entry);
-            if (stage == DXIL_SPV_STAGE_UNKNOWN)
+            /* An export can point to a subobject or an entry point. */
+            for (j = 0; j < rdat_count; j++)
             {
-                ret = VKD3D_ERROR_INVALID_ARGUMENT;
-                goto end;
+                dxil_spv_parsed_blob_get_rdat_subobject(blob, j, &sub);
+                /* Subobject names are not mangled. */
+                if (strcmp(sub.subobject_name, ascii_entry) == 0)
+                    break;
             }
 
-            new_entry.real_entry_point = ascii_entry;
-            new_entry.plain_entry_point = vkd3d_wstrdup(library_desc->pExports[i].Name);
-            new_entry.mangled_entry_point = NULL;
-            new_entry.identifier = identifier;
-            new_entry.stage = convert_stage(stage);
-            ascii_entry = NULL;
+            if (j < rdat_count)
+            {
+                vkd3d_array_reserve((void**)subobjects, subobjects_size,
+                        *subobjects_count + 1, sizeof(**subobjects));
+                subobject = &(*subobjects)[*subobjects_count];
+                vkd3d_shader_dxil_copy_subobject(identifier, subobject, &sub);
+                *subobjects_count += 1;
+            }
+            else
+            {
+                stage = dxil_spv_parsed_blob_get_shader_stage_for_entry(blob, ascii_entry);
+                if (stage == DXIL_SPV_STAGE_UNKNOWN)
+                {
+                    ret = VKD3D_ERROR_INVALID_ARGUMENT;
+                    goto end;
+                }
 
-            vkd3d_array_reserve((void**)entry_points, entry_point_size,
-                    *entry_point_count + 1, sizeof(new_entry));
-            (*entry_points)[(*entry_point_count)++] = new_entry;
-            memset(&new_entry, 0, sizeof(new_entry));
+                new_entry.real_entry_point = ascii_entry;
+                new_entry.plain_entry_point = vkd3d_wstrdup(library_desc->pExports[i].Name);
+                new_entry.mangled_entry_point = NULL;
+                new_entry.identifier = identifier;
+                new_entry.stage = convert_stage(stage);
+                ascii_entry = NULL;
+
+                vkd3d_array_reserve((void**)entry_points, entry_point_size,
+                        *entry_point_count + 1, sizeof(new_entry));
+                (*entry_points)[(*entry_point_count)++] = new_entry;
+                memset(&new_entry, 0, sizeof(new_entry));
+            }
         }
     }
     else
@@ -1489,6 +1611,21 @@ int vkd3d_shader_dxil_append_library_entry_points(
             (*entry_points)[(*entry_point_count)++] = new_entry;
             memset(&new_entry, 0, sizeof(new_entry));
         }
+
+        if (rdat_count)
+        {
+            /* All subobjects are also exported. */
+            vkd3d_array_reserve((void**)subobjects, subobjects_size,
+                    *subobjects_count + rdat_count, sizeof(**subobjects));
+
+            for (i = 0; i < rdat_count; i++)
+            {
+                dxil_spv_parsed_blob_get_rdat_subobject(blob, i, &sub);
+                subobject = &(*subobjects)[*subobjects_count];
+                vkd3d_shader_dxil_copy_subobject(identifier, subobject, &sub);
+                *subobjects_count += 1;
+            }
+        }
     }
 
 end:

libs/vkd3d-shader/spirv.c

@@ -1404,6 +1404,13 @@ static uint32_t vkd3d_spirv_build_op_logical_and(struct vkd3d_spirv_builder *bui
             SpvOpLogicalAnd, result_type, operand0, operand1);
 }
 
+static uint32_t vkd3d_spirv_build_op_any(struct vkd3d_spirv_builder *builder,
+        uint32_t result_type, uint32_t operand0)
+{
+    return vkd3d_spirv_build_op_tr1(builder, &builder->function_stream,
+            SpvOpAny, result_type, operand0);
+}
+
 static uint32_t vkd3d_spirv_build_op_iequal(struct vkd3d_spirv_builder *builder,
         uint32_t result_type, uint32_t operand0, uint32_t operand1)
 {
@@ -2323,6 +2330,8 @@ struct vkd3d_dxbc_compiler
     vkd3d_shader_hash_t descriptor_qa_shader_hash;
 #endif
 
+    uint32_t robust_physical_counter_func_id;
+
     int compiler_error;
 };
 
@@ -5522,31 +5531,22 @@ static const struct vkd3d_shader_global_binding *vkd3d_dxbc_compiler_get_global_
         {
             if (binding->flags & VKD3D_SHADER_BINDING_FLAG_RAW_VA)
             {
-                uint32_t counter_struct_id, pointer_struct_id, array_type_id;
-
-                counter_struct_id = vkd3d_spirv_get_type_id(builder, VKD3D_TYPE_UINT, 1);
-                counter_struct_id = vkd3d_spirv_build_op_type_struct(builder, &counter_struct_id, 1);
-
-                vkd3d_spirv_build_op_member_decorate1(builder, counter_struct_id, 0, SpvDecorationOffset, 0);
-                vkd3d_spirv_build_op_decorate(builder, counter_struct_id, SpvDecorationBlock, NULL, 0);
-                vkd3d_spirv_build_op_name(builder, counter_struct_id, "uav_ctr_t");
-
-                type_id = vkd3d_spirv_build_op_type_pointer(builder, SpvStorageClassPhysicalStorageBuffer, counter_struct_id);
+                uint32_t struct_id, array_type_id;
 
+                type_id = vkd3d_spirv_get_type_id(builder, VKD3D_TYPE_UINT, 2);
                 array_type_id = vkd3d_spirv_build_op_type_runtime_array(builder, type_id);
                 vkd3d_spirv_build_op_decorate1(builder, array_type_id, SpvDecorationArrayStride, sizeof(uint64_t));
+                struct_id = vkd3d_spirv_build_op_type_struct(builder, &array_type_id, 1);
 
-                pointer_struct_id = vkd3d_spirv_build_op_type_struct(builder, &array_type_id, 1);
-
-                vkd3d_spirv_build_op_member_decorate1(builder, pointer_struct_id, 0, SpvDecorationOffset, 0);
-                vkd3d_spirv_build_op_decorate(builder, pointer_struct_id, SpvDecorationBufferBlock, NULL, 0);
-                vkd3d_spirv_build_op_name(builder, pointer_struct_id, "uav_ctrs_t");
+                vkd3d_spirv_build_op_member_decorate1(builder, struct_id, 0, SpvDecorationOffset, 0);
+                vkd3d_spirv_build_op_member_decorate(builder, struct_id, 0, SpvDecorationNonWritable, NULL, 0);
+                vkd3d_spirv_build_op_decorate(builder, struct_id, SpvDecorationBufferBlock, NULL, 0);
+                vkd3d_spirv_build_op_name(builder, struct_id, "uav_ctrs_t");
 
                 var_id = vkd3d_spirv_build_op_variable(builder, &builder->global_stream,
-                        vkd3d_spirv_get_op_type_pointer(builder, storage_class, pointer_struct_id),
+                        vkd3d_spirv_get_op_type_pointer(builder, storage_class, struct_id),
                         storage_class, 0);
 
-                vkd3d_spirv_build_op_decorate(builder, var_id, SpvDecorationAliasedPointer, NULL, 0);
                 vkd3d_spirv_enable_capability(builder, SpvCapabilityPhysicalStorageBufferAddresses);
             }
             else
@@ -5719,10 +5719,116 @@ static const struct vkd3d_shader_buffer_reference_type *vkd3d_dxbc_compiler_get_
 static void vkd3d_dxbc_compiler_emit_descriptor_qa_checks(struct vkd3d_dxbc_compiler *compiler);
 #endif
 
+static void vkd3d_dxbc_compiler_emit_robust_physical_counter_func(struct vkd3d_dxbc_compiler *compiler)
+{
+    struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+    uint32_t not_equal_vec_id, not_equal_id;
+    uint32_t merge_label_id, body_label_id;
+    uint32_t ptr_type_id, ptr_id;
+    uint32_t parameter_types[3];
+    uint32_t parameter_ids[3];
+    uint32_t phi_arguments[4];
+    uint32_t atomic_args[4];
+    uint32_t func_type_id;
+    uint32_t phi_result_id;
+    uint32_t uvec2_type;
+    uint32_t bvec2_type;
+    uint32_t result_id;
+    uint32_t bool_type;
+    uint32_t u32_type;
+    uint32_t label_id;
+    uint32_t zero_id;
+    unsigned int i;
+
+    bool_type = vkd3d_spirv_get_type_id(builder, VKD3D_TYPE_BOOL, 1);
+    bvec2_type = vkd3d_spirv_get_type_id(builder, VKD3D_TYPE_BOOL, 2);
+    u32_type = vkd3d_spirv_get_type_id(builder, VKD3D_TYPE_UINT, 1);
+    uvec2_type = vkd3d_spirv_get_type_id(builder, VKD3D_TYPE_UINT, 2);
+
+    for (i = 0; i < ARRAY_SIZE(parameter_types); i++)
+        parameter_types[i] = i == 0 ? uvec2_type : u32_type;
+
+    func_type_id = vkd3d_spirv_get_op_type_function(builder, u32_type,
+            parameter_types, ARRAY_SIZE(parameter_types));
+    compiler->robust_physical_counter_func_id = vkd3d_spirv_alloc_id(builder);
+    vkd3d_spirv_build_op_name(builder, compiler->robust_physical_counter_func_id, "robust_physical_counter_op");
+    vkd3d_spirv_build_op_function(builder, u32_type, compiler->robust_physical_counter_func_id,
+            SpvFunctionControlMaskNone, func_type_id);
+
+    for (i = 0; i < ARRAY_SIZE(parameter_ids); i++)
+        parameter_ids[i] = vkd3d_spirv_build_op_function_parameter(builder, i == 0 ? uvec2_type : u32_type);
+
+    vkd3d_spirv_build_op_name(builder, parameter_ids[0], "bda");
+    vkd3d_spirv_build_op_name(builder, parameter_ids[1], "direction");
+    vkd3d_spirv_build_op_name(builder, parameter_ids[2], "fixup");
+
+    label_id = vkd3d_spirv_alloc_id(builder);
+    merge_label_id = vkd3d_spirv_alloc_id(builder);
+    body_label_id = vkd3d_spirv_alloc_id(builder);
+    zero_id = vkd3d_dxbc_compiler_get_constant_uint_vector(compiler, 0, 2);
+
+    vkd3d_spirv_build_op_label(builder, label_id);
+    not_equal_vec_id = vkd3d_spirv_build_op_inotequal(builder, bvec2_type,
+            parameter_ids[0], zero_id);
+    not_equal_id = vkd3d_spirv_build_op_any(builder, bool_type, not_equal_vec_id);
+
+    vkd3d_spirv_build_op_selection_merge(builder, merge_label_id, SpvSelectionControlMaskNone);
+    vkd3d_spirv_build_op_branch_conditional(builder, not_equal_id, body_label_id, merge_label_id);
+
+    phi_arguments[1] = body_label_id;
+    phi_arguments[2] = vkd3d_dxbc_compiler_get_constant_uint(compiler, 0);
+    phi_arguments[3] = label_id;
+
+    {
+        vkd3d_spirv_build_op_label(builder, body_label_id);
+        ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassPhysicalStorageBuffer, u32_type);
+        ptr_id = vkd3d_spirv_build_op_bitcast(builder, ptr_type_id, parameter_ids[0]);
+
+        atomic_args[0] = ptr_id;
+        atomic_args[1] = vkd3d_dxbc_compiler_get_constant_uint(compiler, SpvScopeDevice);
+        atomic_args[2] = vkd3d_dxbc_compiler_get_constant_uint(compiler, SpvMemoryAccessMaskNone);
+        atomic_args[3] = parameter_ids[1];
+
+        result_id = vkd3d_spirv_build_op_trv(builder, &builder->function_stream,
+                SpvOpAtomicIAdd, u32_type,
+                atomic_args, ARRAY_SIZE(atomic_args));
+        phi_arguments[0] = vkd3d_spirv_build_op_iadd(builder, u32_type,
+                result_id, parameter_ids[2]);
+
+        vkd3d_spirv_build_op_branch(builder, merge_label_id);
+    }
+
+    vkd3d_spirv_build_op_label(builder, merge_label_id);
+    phi_result_id = vkd3d_spirv_build_op_trv(builder, &builder->function_stream,
+            SpvOpPhi, u32_type,
+            phi_arguments, ARRAY_SIZE(phi_arguments));
+    vkd3d_spirv_build_op_return_value(builder, phi_result_id);
+    vkd3d_spirv_build_op_function_end(builder);
+    vkd3d_spirv_enable_capability(builder, SpvCapabilityPhysicalStorageBufferAddresses);
+}
+
+static uint32_t vkd3d_dxbc_compiler_emit_robust_physical_counter(struct vkd3d_dxbc_compiler *compiler,
+        uint32_t bda_id, bool increment)
+{
+    struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+    uint32_t u32_type;
+    uint32_t args[3];
+
+    u32_type = vkd3d_spirv_get_type_id(builder, VKD3D_TYPE_UINT, 1);
+    args[0] = bda_id;
+    args[1] = vkd3d_dxbc_compiler_get_constant_uint(compiler, increment ? 1u : -1u);
+    args[2] = vkd3d_dxbc_compiler_get_constant_uint(compiler, increment ? 0u : -1u);
+
+    return vkd3d_spirv_build_op_function_call(builder, u32_type,
+            compiler->robust_physical_counter_func_id,
+            args, ARRAY_SIZE(args));
+}
+
 static void vkd3d_dxbc_compiler_emit_initial_declarations(struct vkd3d_dxbc_compiler *compiler)
 {
     const struct vkd3d_shader_transform_feedback_info *xfb_info = compiler->shader_interface.xfb_info;
     struct vkd3d_spirv_builder *builder = &compiler->spirv_builder;
+    unsigned int i;
 
     switch (compiler->shader_type)
     {
@@ -5772,6 +5878,19 @@ static void vkd3d_dxbc_compiler_emit_initial_declarations(struct vkd3d_dxbc_comp
     vkd3d_dxbc_compiler_emit_descriptor_qa_checks(compiler);
 #endif
 
+    if (compiler->scan_info->has_uav_counter)
+    {
+        /* Check if we're expected to deal with RAW VAs. In this case we will enable BDA. */
+        for (i = 0; i < compiler->shader_interface.binding_count; i++)
+        {
+            if (compiler->shader_interface.bindings[i].flags & VKD3D_SHADER_BINDING_FLAG_RAW_VA)
+            {
+                vkd3d_dxbc_compiler_emit_robust_physical_counter_func(compiler);
+                break;
+            }
+        }
+    }
+
     if (compiler->shader_type != VKD3D_SHADER_TYPE_HULL)
     {
         vkd3d_spirv_builder_begin_main_function(builder);
@@ -10019,6 +10138,7 @@ static void vkd3d_dxbc_compiler_emit_uav_counter_instruction(struct vkd3d_dxbc_c
     const struct vkd3d_shader_resource_binding *binding;
     uint32_t type_id, result_id, pointer_id, zero_id;
     const struct vkd3d_symbol *resource_symbol;
+    bool check_post_decrement;
     uint32_t operands[3];
     SpvOp op;
 
@@ -10034,7 +10154,6 @@ static void vkd3d_dxbc_compiler_emit_uav_counter_instruction(struct vkd3d_dxbc_c
 
     if (binding && (binding->flags & VKD3D_SHADER_BINDING_FLAG_RAW_VA))
     {
-        uint32_t ctr_ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassPhysicalStorageBuffer, type_id);
         uint32_t buf_ptr_type_id = vkd3d_spirv_get_op_type_pointer(builder, SpvStorageClassUniform, resource_symbol->info.resource.uav_counter_type_id);
         uint32_t indices[2];
 
@@ -10049,8 +10168,10 @@ static void vkd3d_dxbc_compiler_emit_uav_counter_instruction(struct vkd3d_dxbc_c
                 resource_symbol->info.resource.uav_counter_type_id,
                 pointer_id, SpvMemoryAccessMaskNone);
 
-        pointer_id = vkd3d_spirv_build_op_access_chain1(builder,
-                ctr_ptr_type_id, pointer_id, zero_id);
+        result_id = vkd3d_dxbc_compiler_emit_robust_physical_counter(compiler, pointer_id,
+                instruction->handler_idx == VKD3DSIH_IMM_ATOMIC_ALLOC);
+
+        check_post_decrement = false;
     }
     else if (binding && (binding->flags & VKD3D_SHADER_BINDING_FLAG_BINDLESS))
     {
@@ -10070,6 +10191,8 @@ static void vkd3d_dxbc_compiler_emit_uav_counter_instruction(struct vkd3d_dxbc_c
         /* Need to mark the pointer argument itself as non-uniform. */
         if (src->reg.modifier == VKD3DSPRM_NONUNIFORM)
             vkd3d_dxbc_compiler_decorate_nonuniform(compiler, pointer_id);
+
+        check_post_decrement = true;
     }
     else
     {
@@ -10077,19 +10200,25 @@ static void vkd3d_dxbc_compiler_emit_uav_counter_instruction(struct vkd3d_dxbc_c
 
         pointer_id = vkd3d_spirv_build_op_image_texel_pointer(builder, ptr_type_id,
             resource_symbol->info.resource.uav_counter_id, zero_id, zero_id);
+
+        check_post_decrement = true;
     }
 
-    operands[0] = pointer_id;
-    operands[1] = vkd3d_dxbc_compiler_get_constant_uint(compiler, SpvScopeDevice);
-    operands[2] = vkd3d_dxbc_compiler_get_constant_uint(compiler, memory_semantics);
-    result_id = vkd3d_spirv_build_op_trv(builder, &builder->function_stream,
-            op, type_id, operands, ARRAY_SIZE(operands));
-    if (op == SpvOpAtomicIDecrement)
+    if (check_post_decrement)
     {
-        /* SpvOpAtomicIDecrement returns the original value. */
-        result_id = vkd3d_spirv_build_op_isub(builder, type_id, result_id,
-                vkd3d_dxbc_compiler_get_constant_uint(compiler, 1));
+        operands[0] = pointer_id;
+        operands[1] = vkd3d_dxbc_compiler_get_constant_uint(compiler, SpvScopeDevice);
+        operands[2] = vkd3d_dxbc_compiler_get_constant_uint(compiler, memory_semantics);
+        result_id = vkd3d_spirv_build_op_trv(builder, &builder->function_stream,
+                op, type_id, operands, ARRAY_SIZE(operands));
+        if (op == SpvOpAtomicIDecrement)
+        {
+            /* SpvOpAtomicIDecrement returns the original value. */
+            result_id = vkd3d_spirv_build_op_isub(builder, type_id, result_id,
+                    vkd3d_dxbc_compiler_get_constant_uint(compiler, 1));
+        }
     }
+
     vkd3d_dxbc_compiler_emit_store_dst(compiler, dst, result_id);
 }
 

libs/vkd3d-shader/vkd3d_shader_main.c

@@ -477,6 +477,7 @@ static void vkd3d_shader_scan_record_uav_counter(struct vkd3d_shader_scan_info *
         const struct vkd3d_shader_register *reg)
 {
     scan_info->has_side_effects = true;
+    scan_info->has_uav_counter = true;
     vkd3d_shader_scan_set_register_flags(scan_info, VKD3DSPR_UAV,
             reg->idx[0].offset, VKD3D_SHADER_UAV_FLAG_ATOMIC_COUNTER);
 }

libs/vkd3d/acceleration_structure.c

@@ -19,6 +19,8 @@
 #define VKD3D_DBG_CHANNEL VKD3D_DBG_CHANNEL_API
 #include "vkd3d_private.h"
 
+#define RT_TRACE TRACE
+
 void vkd3d_acceleration_structure_build_info_cleanup(
         struct vkd3d_acceleration_structure_build_info *info)
 {
@@ -74,19 +76,31 @@ bool vkd3d_acceleration_structure_convert_inputs(const struct d3d12_device *devi
     bool have_triangles, have_aabbs;
     unsigned int i;
 
+    RT_TRACE("Converting inputs.\n");
+    RT_TRACE("=====================\n");
+
     build_info = &info->build_info;
     memset(build_info, 0, sizeof(*build_info));
     build_info->sType = VK_STRUCTURE_TYPE_ACCELERATION_STRUCTURE_BUILD_GEOMETRY_INFO_KHR;
 
     if (desc->Type == D3D12_RAYTRACING_ACCELERATION_STRUCTURE_TYPE_TOP_LEVEL)
+    {
         build_info->type = VK_ACCELERATION_STRUCTURE_TYPE_TOP_LEVEL_KHR;
+        RT_TRACE("Top level build.\n");
+    }
     else
+    {
         build_info->type = VK_ACCELERATION_STRUCTURE_TYPE_BOTTOM_LEVEL_KHR;
+        RT_TRACE("Bottom level build.\n");
+    }
 
     build_info->flags = d3d12_build_flags_to_vk(desc->Flags);
 
     if (desc->Flags & D3D12_RAYTRACING_ACCELERATION_STRUCTURE_BUILD_FLAG_PERFORM_UPDATE)
+    {
+        RT_TRACE("BUILD_FLAG_PERFORM_UPDATE.\n");
         build_info->mode = VK_BUILD_ACCELERATION_STRUCTURE_MODE_UPDATE_KHR;
+    }
     else
         build_info->mode = VK_BUILD_ACCELERATION_STRUCTURE_MODE_BUILD_KHR;
 
@@ -109,6 +123,9 @@ bool vkd3d_acceleration_structure_convert_inputs(const struct d3d12_device *devi
         info->primitive_counts = info->primitive_counts_stack;
         info->primitive_counts[0] = desc->NumDescs;
         build_info->geometryCount = 1;
+        RT_TRACE("  ArrayOfPointers: %u.\n",
+                desc->DescsLayout == D3D12_ELEMENTS_LAYOUT_ARRAY_OF_POINTERS ? 1 : 0);
+        RT_TRACE("  NumDescs: %u.\n", info->primitive_counts[0]);
     }
     else
     {
@@ -132,13 +149,21 @@ bool vkd3d_acceleration_structure_convert_inputs(const struct d3d12_device *devi
         for (i = 0; i < desc->NumDescs; i++)
         {
             info->geometries[i].sType = VK_STRUCTURE_TYPE_ACCELERATION_STRUCTURE_GEOMETRY_KHR;
+            RT_TRACE(" Geom %u:\n", i);
 
             if (desc->DescsLayout == D3D12_ELEMENTS_LAYOUT_ARRAY_OF_POINTERS)
+            {
                 geom_desc = desc->ppGeometryDescs[i];
+                RT_TRACE("  ArrayOfPointers\n");
+            }
             else
+            {
                 geom_desc = &desc->pGeometryDescs[i];
+                RT_TRACE("  PointerToArray\n");
+            }
 
             info->geometries[i].flags = d3d12_geometry_flags_to_vk(geom_desc->Flags);
+            RT_TRACE("  Flags = #%x\n", geom_desc->Flags);
 
             switch (geom_desc->Type)
             {
@@ -155,17 +180,26 @@ bool vkd3d_acceleration_structure_convert_inputs(const struct d3d12_device *devi
                     triangles = &info->geometries[i].geometry.triangles;
                     triangles->sType = VK_STRUCTURE_TYPE_ACCELERATION_STRUCTURE_GEOMETRY_TRIANGLES_DATA_KHR;
                     triangles->indexData.deviceAddress = geom_desc->Triangles.IndexBuffer;
-                    if (geom_desc->Triangles.IndexBuffer)
+                    if (geom_desc->Triangles.IndexFormat != DXGI_FORMAT_UNKNOWN)
                     {
+                        if (!geom_desc->Triangles.IndexBuffer)
+                            WARN("Application is using IndexBuffer = 0 and IndexFormat != UNKNOWN. Likely application bug.\n");
+
                         triangles->indexType =
                                 geom_desc->Triangles.IndexFormat == DXGI_FORMAT_R16_UINT ?
                                         VK_INDEX_TYPE_UINT16 : VK_INDEX_TYPE_UINT32;
                         info->primitive_counts[i] = geom_desc->Triangles.IndexCount / 3;
+                        RT_TRACE("  Indexed : Index count = %u (%u bits)\n",
+                                geom_desc->Triangles.IndexCount,
+                                triangles->indexType == VK_INDEX_TYPE_UINT16 ? 16 : 32);
+                        RT_TRACE("  Vertex count: %u\n", geom_desc->Triangles.VertexCount);
+                        RT_TRACE("  IBO VA: %"PRIx64".\n", geom_desc->Triangles.IndexBuffer);
                     }
                     else
                     {
                         info->primitive_counts[i] = geom_desc->Triangles.VertexCount / 3;
                         triangles->indexType = VK_INDEX_TYPE_NONE_KHR;
+                        RT_TRACE("  Triangle list : Vertex count: %u\n", geom_desc->Triangles.VertexCount);
                     }
 
                     triangles->maxVertex = max(1, geom_desc->Triangles.VertexCount) - 1;
@@ -173,6 +207,11 @@ bool vkd3d_acceleration_structure_convert_inputs(const struct d3d12_device *devi
                     triangles->vertexFormat = vkd3d_internal_get_vk_format(device, geom_desc->Triangles.VertexFormat);
                     triangles->vertexData.deviceAddress = geom_desc->Triangles.VertexBuffer.StartAddress;
                     triangles->transformData.deviceAddress = geom_desc->Triangles.Transform3x4;
+
+                    RT_TRACE("  Transform3x4: %s\n", geom_desc->Triangles.Transform3x4 ? "on" : "off");
+                    RT_TRACE("  Vertex format: %s\n", debug_dxgi_format(geom_desc->Triangles.VertexFormat));
+                    RT_TRACE("  VBO VA: %"PRIx64"\n", geom_desc->Triangles.VertexBuffer.StartAddress);
+                    RT_TRACE("  Vertex stride: %"PRIu64" bytes\n", geom_desc->Triangles.VertexBuffer.StrideInBytes);
                     break;
 
                 case D3D12_RAYTRACING_GEOMETRY_TYPE_PROCEDURAL_PRIMITIVE_AABBS:
@@ -190,12 +229,15 @@ bool vkd3d_acceleration_structure_convert_inputs(const struct d3d12_device *devi
                     aabbs->stride = geom_desc->AABBs.AABBs.StrideInBytes;
                     aabbs->data.deviceAddress = geom_desc->AABBs.AABBs.StartAddress;
                     info->primitive_counts[i] = geom_desc->AABBs.AABBCount;
+                    RT_TRACE("  AABB stride: %"PRIu64" bytes\n", geom_desc->AABBs.AABBs.StrideInBytes);
                     break;
 
                 default:
                     FIXME("Unsupported geometry type %u.\n", geom_desc->Type);
                     return false;
             }
+
+            RT_TRACE("  Primitive count %u.\n", info->primitive_counts[i]);
         }
     }
 
@@ -209,6 +251,8 @@ bool vkd3d_acceleration_structure_convert_inputs(const struct d3d12_device *devi
     }
 
     build_info->pGeometries = info->geometries;
+
+    RT_TRACE("=====================\n");
     return true;
 }
 

libs/vkd3d/breadcrumbs.c

@@ -50,6 +50,8 @@ static const char *vkd3d_breadcrumb_command_type_to_str(enum vkd3d_breadcrumb_co
             return "dispatch";
         case VKD3D_BREADCRUMB_COMMAND_EXECUTE_INDIRECT:
             return "execute_indirect";
+        case VKD3D_BREADCRUMB_COMMAND_EXECUTE_INDIRECT_TEMPLATE:
+            return "execute_indirect_template";
         case VKD3D_BREADCRUMB_COMMAND_COPY:
             return "copy";
         case VKD3D_BREADCRUMB_COMMAND_RESOLVE:
@@ -82,6 +84,8 @@ static const char *vkd3d_breadcrumb_command_type_to_str(enum vkd3d_breadcrumb_co
             return "root_desc";
         case VKD3D_BREADCRUMB_COMMAND_ROOT_CONST:
             return "root_const";
+        case VKD3D_BREADCRUMB_COMMAND_TAG:
+            return "tag";
 
         default:
             return "?";
@@ -306,6 +310,10 @@ static void vkd3d_breadcrumb_tracer_report_command_list(
         {
             ERR(" Set arg: %"PRIu64" (#%"PRIx64")\n", cmd->word_64bit, cmd->word_64bit);
         }
+        else if (cmd->type == VKD3D_BREADCRUMB_COMMAND_TAG)
+        {
+            ERR("     Tag: %s\n", cmd->tag);
+        }
         else
         {
             ERR("  Command: %s\n", vkd3d_breadcrumb_command_type_to_str(cmd->type));

libs/vkd3d/debug_ring.c

@@ -61,12 +61,56 @@ void vkd3d_shader_debug_ring_init_spec_constant(struct d3d12_device *device,
 #define DEBUG_CHANNEL_WORD_COOKIE 0xdeadca70u
 #define DEBUG_CHANNEL_WORD_MASK 0xfffffff0u
 
+static const char *vkd3d_patch_command_token_str(enum vkd3d_patch_command_token token)
+{
+    switch (token)
+    {
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_CONST_U32: return "RootConst";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_VA_LO: return "IBO VA LO";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_VA_HI: return "IBO VA HI";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_SIZE: return "IBO Size";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_INDEX_FORMAT: return "IBO Type";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_VA_LO: return "VBO VA LO";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_VA_HI: return "VBO VA HI";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_SIZE: return "VBO Size";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_STRIDE: return "VBO Stride";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_ROOT_VA_LO: return "ROOT VA LO";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_ROOT_VA_HI: return "ROOT VA HI";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VERTEX_COUNT: return "Vertex Count";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_INDEX_COUNT: return "Index Count";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_INSTANCE_COUNT: return "Instance Count";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_FIRST_INDEX: return "First Index";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_FIRST_VERTEX: return "First Vertex";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_FIRST_INSTANCE: return "First Instance";
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VERTEX_OFFSET: return "Vertex Offset";
+        default: return "???";
+    }
+}
+
+static bool vkd3d_patch_command_token_is_hex(enum vkd3d_patch_command_token token)
+{
+    switch (token)
+    {
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_VA_LO:
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_VA_HI:
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_VA_LO:
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_VA_HI:
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_ROOT_VA_LO:
+        case VKD3D_PATCH_COMMAND_TOKEN_COPY_ROOT_VA_HI:
+            return true;
+
+        default:
+            return false;
+    }
+}
+
 static bool vkd3d_shader_debug_ring_print_message(struct vkd3d_shader_debug_ring *ring,
         uint32_t word_offset, uint32_t message_word_count)
 {
     uint32_t i, debug_instance, debug_thread_id[3], fmt;
     char message_buffer[4096];
     uint64_t shader_hash;
+    size_t len, avail;
 
     if (message_word_count < 8)
     {
@@ -80,52 +124,107 @@ static bool vkd3d_shader_debug_ring_print_message(struct vkd3d_shader_debug_ring
         debug_thread_id[i] = READ_RING_WORD(word_offset + 4 + i);
     fmt = READ_RING_WORD(word_offset + 7);
 
-    snprintf(message_buffer, sizeof(message_buffer), "Shader: %"PRIx64": Instance %u, ID (%u, %u, %u):",
-            shader_hash, debug_instance,
-            debug_thread_id[0], debug_thread_id[1], debug_thread_id[2]);
-
     word_offset += 8;
     message_word_count -= 8;
 
-    for (i = 0; i < message_word_count; i++)
+    if (shader_hash == 0)
     {
-        union
+        /* We got this from our internal debug shaders. Pretty-print.
+         * Make sure the log is sortable for easier debug.
+         * TODO: Might consider a callback system that listeners from different subsystems can listen to and print their own messages,
+         * but that is overengineering at this time ... */
+        snprintf(message_buffer, sizeof(message_buffer), "ExecuteIndirect: GlobalCommandIndex %010u, Debug tag %010u, DrawID %04u (ThreadID %04u): ",
+                debug_instance, debug_thread_id[0], debug_thread_id[1], debug_thread_id[2]);
+
+        if (message_word_count == 2)
         {
-            float f32;
-            uint32_t u32;
-            int32_t i32;
-        } u;
-        const char *delim;
-        size_t len, avail;
-        u.u32 = READ_RING_WORD(word_offset + i);
+            len = strlen(message_buffer);
+            avail = sizeof(message_buffer) - len;
+            snprintf(message_buffer + len, avail, "DrawCount %u, MaxDrawCount %u",
+                    READ_RING_WORD(word_offset + 0),
+                    READ_RING_WORD(word_offset + 1));
+        }
+        else if (message_word_count == 4)
+        {
+            union { uint32_t u32; float f32; int32_t s32; } value;
+            enum vkd3d_patch_command_token token;
+            uint32_t dst_offset;
+            uint32_t src_offset;
 
-        len = strlen(message_buffer);
-        if (len + 1 >= sizeof(message_buffer))
-            break;
-        avail = sizeof(message_buffer) - len;
+            len = strlen(message_buffer);
+            avail = sizeof(message_buffer) - len;
 
-        delim = i == 0 ? " " : ", ";
+            token = READ_RING_WORD(word_offset + 0);
+            dst_offset = READ_RING_WORD(word_offset + 1);
+            src_offset = READ_RING_WORD(word_offset + 2);
+            value.u32 = READ_RING_WORD(word_offset + 3);
+
+            if (vkd3d_patch_command_token_is_hex(token))
+            {
+                snprintf(message_buffer + len, avail, "%s <- #%08x",
+                        vkd3d_patch_command_token_str(token), value.u32);
+            }
+            else if (token == VKD3D_PATCH_COMMAND_TOKEN_COPY_CONST_U32)
+            {
+                snprintf(message_buffer + len, avail, "%s <- {hex #%08x, s32 %d, f32 %f}",
+                        vkd3d_patch_command_token_str(token), value.u32, value.s32, value.f32);
+            }
+            else
+            {
+                snprintf(message_buffer + len, avail, "%s <- %d",
+                        vkd3d_patch_command_token_str(token), value.s32);
+            }
+
+            len = strlen(message_buffer);
+            avail = sizeof(message_buffer) - len;
+            snprintf(message_buffer + len, avail, " (dst offset %u, src offset %u)", dst_offset, src_offset);
+        }
+    }
+    else
+    {
+        snprintf(message_buffer, sizeof(message_buffer), "Shader: %"PRIx64": Instance %010u, ID (%u, %u, %u):",
+                shader_hash, debug_instance,
+                debug_thread_id[0], debug_thread_id[1], debug_thread_id[2]);
+
+        for (i = 0; i < message_word_count; i++)
+        {
+            union
+            {
+                float f32;
+                uint32_t u32;
+                int32_t i32;
+            } u;
+            const char *delim;
+            u.u32 = READ_RING_WORD(word_offset + i);
+
+            len = strlen(message_buffer);
+            if (len + 1 >= sizeof(message_buffer))
+                break;
+            avail = sizeof(message_buffer) - len;
+
+            delim = i == 0 ? " " : ", ";
 
 #define VKD3D_DEBUG_CHANNEL_FMT_HEX 0u
 #define VKD3D_DEBUG_CHANNEL_FMT_I32 1u
 #define VKD3D_DEBUG_CHANNEL_FMT_F32 2u
-        switch ((fmt >> (2u * i)) & 3u)
-        {
-            case VKD3D_DEBUG_CHANNEL_FMT_HEX:
-                snprintf(message_buffer + len, avail, "%s#%x", delim, u.u32);
-                break;
+            switch ((fmt >> (2u * i)) & 3u)
+            {
+                case VKD3D_DEBUG_CHANNEL_FMT_HEX:
+                    snprintf(message_buffer + len, avail, "%s#%x", delim, u.u32);
+                    break;
 
-            case VKD3D_DEBUG_CHANNEL_FMT_I32:
-                snprintf(message_buffer + len, avail, "%s%d", delim, u.i32);
-                break;
+                case VKD3D_DEBUG_CHANNEL_FMT_I32:
+                    snprintf(message_buffer + len, avail, "%s%d", delim, u.i32);
+                    break;
 
-            case VKD3D_DEBUG_CHANNEL_FMT_F32:
-                snprintf(message_buffer + len, avail, "%s%f", delim, u.f32);
-                break;
+                case VKD3D_DEBUG_CHANNEL_FMT_F32:
+                    snprintf(message_buffer + len, avail, "%s%f", delim, u.f32);
+                    break;
 
-            default:
-                snprintf(message_buffer + len, avail, "%s????", delim);
-                break;
+                default:
+                    snprintf(message_buffer + len, avail, "%s????", delim);
+                    break;
+            }
         }
     }
 

libs/vkd3d/device.c

@@ -90,6 +90,10 @@ static const struct vkd3d_optional_extension_info optional_device_extensions[] =
     VK_EXTENSION(KHR_UNIFORM_BUFFER_STANDARD_LAYOUT, KHR_uniform_buffer_standard_layout),
     VK_EXTENSION(KHR_MAINTENANCE_4, KHR_maintenance4),
     VK_EXTENSION(KHR_FRAGMENT_SHADER_BARYCENTRIC, KHR_fragment_shader_barycentric),
+#ifdef _WIN32
+    VK_EXTENSION(KHR_EXTERNAL_MEMORY_WIN32, KHR_external_memory_win32),
+    VK_EXTENSION(KHR_EXTERNAL_SEMAPHORE_WIN32, KHR_external_semaphore_win32),
+#endif
     /* EXT extensions */
     VK_EXTENSION(EXT_CALIBRATED_TIMESTAMPS, EXT_calibrated_timestamps),
     VK_EXTENSION(EXT_CONDITIONAL_RENDERING, EXT_conditional_rendering),
@@ -483,6 +487,14 @@ static void vkd3d_init_debug_messenger_callback(struct vkd3d_instance *instance)
     instance->vk_debug_callback = callback;
 }
 
+/* Could be a flag style enum if needed. */
+enum vkd3d_application_feature_override
+{
+    VKD3D_APPLICATION_FEATURE_OVERRIDE_NONE = 0,
+    VKD3D_APPLICATION_FEATURE_OVERRIDE_PROMOTE_DXR_TO_ULTIMATE,
+};
+
+static enum vkd3d_application_feature_override vkd3d_application_feature_override;
 uint64_t vkd3d_config_flags;
 struct vkd3d_shader_quirk_info vkd3d_shader_quirk_info;
 
@@ -492,6 +504,7 @@ struct vkd3d_instance_application_meta
     const char *name;
     uint64_t global_flags_add;
     uint64_t global_flags_remove;
+    enum vkd3d_application_feature_override override;
 };
 static const struct vkd3d_instance_application_meta application_override[] = {
     /* MSVC fails to compile empty array. */
@@ -502,7 +515,8 @@ static const struct vkd3d_instance_application_meta application_override[] = {
      * This works okay with zerovram on first game boot, but not later, since this memory is guaranteed to be recycled.
      * Game also relies on indirectly modifying CBV root descriptors, which means we are forced to rely on RAW_VA_CBV. */
     { VKD3D_STRING_COMPARE_EXACT, "HaloInfinite.exe",
-            VKD3D_CONFIG_FLAG_ZERO_MEMORY_WORKAROUNDS_COMMITTED_BUFFER_UAV | VKD3D_CONFIG_FLAG_FORCE_RAW_VA_CBV, 0 },
+            VKD3D_CONFIG_FLAG_ZERO_MEMORY_WORKAROUNDS_COMMITTED_BUFFER_UAV | VKD3D_CONFIG_FLAG_FORCE_RAW_VA_CBV |
+            VKD3D_CONFIG_FLAG_USE_HOST_IMPORT_FALLBACK, 0 },
     /* Shadow of the Tomb Raider (750920).
      * Invariant workarounds actually cause more issues than they resolve on NV.
      * RADV already has workarounds by default.
@@ -518,6 +532,20 @@ static const struct vkd3d_instance_application_meta application_override[] = {
     /* Serious Sam 4 (257420).
      * Invariant workarounds cause graphical glitches when rendering foliage on NV. */
     { VKD3D_STRING_COMPARE_EXACT, "Sam4.exe", VKD3D_CONFIG_FLAG_FORCE_NO_INVARIANT_POSITION, 0 },
+    /* Cyberpunk 2077 (1091500). */
+    { VKD3D_STRING_COMPARE_EXACT, "Cyberpunk2077.exe", VKD3D_CONFIG_FLAG_ALLOW_SBT_COLLECTION, 0 },
+    /* Resident Evil: Village (1196590).
+     * Game relies on mesh + sampler feedback to be exposed to use DXR.
+     * Likely used as a proxy for Turing+ to avoid potential software fallbacks on Pascal. */
+    { VKD3D_STRING_COMPARE_EXACT, "re8.exe",
+            VKD3D_CONFIG_FLAG_FORCE_NATIVE_FP16, 0, VKD3D_APPLICATION_FEATURE_OVERRIDE_PROMOTE_DXR_TO_ULTIMATE },
+    /* Resident Evil 2 remake (883710). Same as RE: Village. */
+    { VKD3D_STRING_COMPARE_EXACT, "re2.exe",
+            VKD3D_CONFIG_FLAG_FORCE_NATIVE_FP16, 0, VKD3D_APPLICATION_FEATURE_OVERRIDE_PROMOTE_DXR_TO_ULTIMATE },
+    { VKD3D_STRING_COMPARE_EXACT, "re3.exe",
+            VKD3D_CONFIG_FLAG_FORCE_NATIVE_FP16, 0, VKD3D_APPLICATION_FEATURE_OVERRIDE_PROMOTE_DXR_TO_ULTIMATE },
+    { VKD3D_STRING_COMPARE_EXACT, "re7.exe",
+            VKD3D_CONFIG_FLAG_FORCE_NATIVE_FP16, 0, VKD3D_APPLICATION_FEATURE_OVERRIDE_PROMOTE_DXR_TO_ULTIMATE },
     { VKD3D_STRING_COMPARE_NEVER, NULL, 0, 0 }
 };
 
@@ -568,6 +596,7 @@ static void vkd3d_instance_apply_application_workarounds(void)
             vkd3d_config_flags &= ~application_override[i].global_flags_remove;
             INFO("Detected game %s, adding config 0x%"PRIx64", removing masks 0x%"PRIx64".\n",
                  app, application_override[i].global_flags_add, application_override[i].global_flags_remove);
+            vkd3d_application_feature_override = application_override[i].override;
             break;
         }
     }
@@ -664,6 +693,7 @@ static const struct vkd3d_debug_option vkd3d_config_options[] =
     {"pipeline_library_app_cache", VKD3D_CONFIG_FLAG_PIPELINE_LIBRARY_APP_CACHE_ONLY},
     {"shader_cache_sync", VKD3D_CONFIG_FLAG_SHADER_CACHE_SYNC},
     {"force_raw_va_cbv", VKD3D_CONFIG_FLAG_FORCE_RAW_VA_CBV},
+    {"allow_sbt_collection", VKD3D_CONFIG_FLAG_ALLOW_SBT_COLLECTION},
 };
 
 static void vkd3d_config_flags_init_once(void)
@@ -4375,19 +4405,263 @@ static HRESULT STDMETHODCALLTYPE d3d12_device_CreateSharedHandle(d3d12_device_if
         ID3D12DeviceChild *object, const SECURITY_ATTRIBUTES *attributes, DWORD access,
         const WCHAR *name, HANDLE *handle)
 {
-    FIXME("iface %p, object %p, attributes %p, access %#x, name %s, handle %p stub!\n",
+#ifdef _WIN32
+    struct d3d12_device *device = impl_from_ID3D12Device(iface);
+    const struct vkd3d_vk_device_procs *vk_procs;
+    struct DxvkSharedTextureMetadata metadata;
+    ID3D12Resource *resource_iface;
+    ID3D12Fence *fence_iface;
+
+    vk_procs = &device->vk_procs;
+
+    TRACE("iface %p, object %p, attributes %p, access %#x, name %s, handle %p\n",
             iface, object, attributes, access, debugstr_w(name), handle);
 
+    if (SUCCEEDED(ID3D12DeviceChild_QueryInterface(object, &IID_ID3D12Resource, (void**)&resource_iface)))
+    {
+        struct d3d12_resource *resource = impl_from_ID3D12Resource(resource_iface);
+        VkMemoryGetWin32HandleInfoKHR win32_handle_info;
+        VkResult vr;
+
+        if (!(resource->heap_flags & D3D12_HEAP_FLAG_SHARED))
+        {
+            ID3D12Resource_Release(resource_iface);
+            return DXGI_ERROR_INVALID_CALL;
+        }
+
+        if (attributes)
+            FIXME("attributes %p not handled.\n", attributes);
+        if (access)
+            FIXME("access %#x not handled.\n", access);
+        if (name)
+            FIXME("name %s not handled.\n", debugstr_w(name));
+
+        win32_handle_info.sType = VK_STRUCTURE_TYPE_MEMORY_GET_WIN32_HANDLE_INFO_KHR;
+        win32_handle_info.pNext = NULL;
+        win32_handle_info.memory = resource->mem.device_allocation.vk_memory;
+        win32_handle_info.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT;
+
+        vr = VK_CALL(vkGetMemoryWin32HandleKHR(device->vk_device, &win32_handle_info, handle));
+
+        if (vr == VK_SUCCESS)
+        {
+            if (resource->desc.Dimension != D3D12_RESOURCE_DIMENSION_TEXTURE2D)
+            {
+                FIXME("Shared texture metadata structure only supports 2D textures.");
+            }
+            else
+            {
+                metadata.Width = resource->desc.Width;
+                metadata.Height = resource->desc.Height;
+                metadata.MipLevels = resource->desc.MipLevels;
+                metadata.ArraySize = resource->desc.DepthOrArraySize;
+                metadata.Format = resource->desc.Format;
+                metadata.SampleDesc = resource->desc.SampleDesc;
+                metadata.Usage = D3D11_USAGE_DEFAULT;
+                metadata.BindFlags = D3D11_BIND_SHADER_RESOURCE;
+                metadata.CPUAccessFlags = 0;
+                metadata.MiscFlags = D3D11_RESOURCE_MISC_SHARED_NTHANDLE;
+
+                if (resource->desc.Flags & D3D12_RESOURCE_FLAG_ALLOW_RENDER_TARGET)
+                    metadata.BindFlags |= D3D11_BIND_RENDER_TARGET;
+                if (resource->desc.Flags & D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL)
+                    metadata.BindFlags |= D3D11_BIND_DEPTH_STENCIL;
+                if (resource->desc.Flags & D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS)
+                    metadata.BindFlags |= D3D11_BIND_UNORDERED_ACCESS;
+                if (resource->desc.Flags & D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE)
+                    metadata.BindFlags &= ~D3D11_BIND_SHADER_RESOURCE;
+
+                if (!vkd3d_set_shared_metadata(*handle, &metadata, sizeof(metadata)))
+                    ERR("Failed to set metadata for shared resource, importing created handle will fail.\n");
+            }
+        }
+
+        ID3D12Resource_Release(resource_iface);
+        return vr ? E_FAIL : S_OK;
+    }
+
+    if (SUCCEEDED(ID3D12DeviceChild_QueryInterface(object, &IID_ID3D12Fence, (void**)&fence_iface)))
+    {
+        VkSemaphoreGetWin32HandleInfoKHR win32_handle_info;
+        struct d3d12_shared_fence *fence;
+        VkResult vr;
+
+        if (!is_shared_ID3D12Fence(fence_iface))
+        {
+            ID3D12Fence_Release(fence_iface);
+            return DXGI_ERROR_INVALID_CALL;
+        }
+
+        fence = shared_impl_from_ID3D12Fence(fence_iface);
+
+        if (attributes)
+            FIXME("attributes %p not handled\n", attributes);
+        if (access)
+            FIXME("access %#x not handled\n", access);
+        if (name)
+            FIXME("name %s not handled\n", debugstr_w(name));
+
+        win32_handle_info.sType = VK_STRUCTURE_TYPE_SEMAPHORE_GET_WIN32_HANDLE_INFO_KHR;
+        win32_handle_info.pNext = NULL;
+        win32_handle_info.semaphore = fence->timeline_semaphore;
+        win32_handle_info.handleType = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE_BIT;
+
+        vr = VK_CALL(vkGetSemaphoreWin32HandleKHR(device->vk_device, &win32_handle_info, handle));
+
+        ID3D12Fence_Release(fence_iface);
+        return vr ? E_FAIL : S_OK;
+    }
+
+    FIXME("Creating shared handle for type of object %p unsupported.\n", object);
     return E_NOTIMPL;
+#else
+    FIXME("CreateSharedHandle can only be implemented in native Win32.\n");
+    return E_NOTIMPL;
+#endif
 }
 
+#ifdef _WIN32
+static inline bool handle_is_kmt_style(HANDLE handle)
+{
+    return ((ULONG_PTR)handle & 0x40000000) && ((ULONG_PTR)handle - 2) % 4 == 0;
+}
+#endif
+
 static HRESULT STDMETHODCALLTYPE d3d12_device_OpenSharedHandle(d3d12_device_iface *iface,
         HANDLE handle, REFIID riid, void **object)
 {
-    FIXME("iface %p, handle %p, riid %s, object %p stub!\n",
+#ifdef _WIN32
+    struct d3d12_device *device = impl_from_ID3D12Device(iface);
+    const struct vkd3d_vk_device_procs *vk_procs;
+    HRESULT hr;
+
+    vk_procs = &device->vk_procs;
+
+    TRACE("iface %p, handle %p, riid %s, object %p\n",
             iface, handle, debugstr_guid(riid), object);
 
+    if (IsEqualGUID(riid, &IID_ID3D12Resource))
+    {
+        struct DxvkSharedTextureMetadata metadata;
+        D3D12_HEAP_PROPERTIES heap_props;
+        struct d3d12_resource *resource;
+        D3D12_RESOURCE_DESC1 desc;
+        bool kmt_handle = false;
+
+        if (handle_is_kmt_style(handle))
+        {
+            handle = vkd3d_open_kmt_handle(handle);
+            kmt_handle = true;
+
+            if (handle == INVALID_HANDLE_VALUE)
+            {
+                WARN("Failed to open KMT-style ID3D12Resource shared handle.\n");
+                *object = NULL;
+                return E_INVALIDARG;
+            }
+        }
+
+        if (!vkd3d_get_shared_metadata(handle, &metadata, sizeof(metadata), NULL))
+        {
+            WARN("Failed to get ID3D12Resource shared handle metadata.\n");
+            if (kmt_handle)
+                CloseHandle(handle);
+
+            *object = NULL;
+            return E_INVALIDARG;
+        }
+
+        desc.Dimension = D3D12_RESOURCE_DIMENSION_TEXTURE2D;
+        desc.Alignment = 0;
+        desc.Width = metadata.Width;
+        desc.Height = metadata.Height;
+        desc.DepthOrArraySize = metadata.ArraySize;
+        desc.MipLevels = metadata.MipLevels;
+        desc.Format = metadata.Format;
+        desc.SampleDesc = metadata.SampleDesc;
+        switch (metadata.TextureLayout)
+        {
+            case D3D11_TEXTURE_LAYOUT_UNDEFINED: desc.Layout = D3D12_TEXTURE_LAYOUT_UNKNOWN; break;
+            case D3D11_TEXTURE_LAYOUT_ROW_MAJOR: desc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR; break;
+            case D3D11_TEXTURE_LAYOUT_64K_STANDARD_SWIZZLE: desc.Layout = D3D12_TEXTURE_LAYOUT_64KB_STANDARD_SWIZZLE; break;
+            default: desc.Layout = D3D12_TEXTURE_LAYOUT_UNKNOWN;
+        }
+        desc.Flags = D3D12_RESOURCE_FLAG_ALLOW_SIMULTANEOUS_ACCESS;
+        if (metadata.BindFlags & D3D11_BIND_RENDER_TARGET)
+            desc.Flags |= D3D12_RESOURCE_FLAG_ALLOW_RENDER_TARGET;
+        if (metadata.BindFlags & D3D11_BIND_DEPTH_STENCIL)
+            desc.Flags |= D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL;
+        if (metadata.BindFlags & D3D11_BIND_UNORDERED_ACCESS)
+            desc.Flags |= D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
+        if ((metadata.BindFlags & D3D11_BIND_DEPTH_STENCIL) && !(metadata.BindFlags & D3D11_BIND_SHADER_RESOURCE))
+            desc.Flags |= D3D12_RESOURCE_FLAG_DENY_SHADER_RESOURCE;
+        desc.SamplerFeedbackMipRegion.Width = 0;
+        desc.SamplerFeedbackMipRegion.Height = 0;
+        desc.SamplerFeedbackMipRegion.Depth = 0;
+
+        heap_props.Type = D3D12_HEAP_TYPE_DEFAULT;
+        heap_props.CPUPageProperty = D3D12_CPU_PAGE_PROPERTY_UNKNOWN;
+        heap_props.MemoryPoolPreference = D3D12_MEMORY_POOL_UNKNOWN;
+        heap_props.CreationNodeMask = 0;
+        heap_props.VisibleNodeMask = 0;
+
+        hr = d3d12_resource_create_committed(device, &desc, &heap_props,
+                D3D12_HEAP_FLAG_SHARED, D3D12_RESOURCE_STATE_COMMON, NULL, handle, &resource);
+        if (kmt_handle)
+            CloseHandle(handle);
+
+        if (FAILED(hr))
+        {
+            WARN("Failed to open shared ID3D12Resource, hr %#x.\n", hr);
+            *object = NULL;
+            return hr;
+        }
+
+        return return_interface(&resource->ID3D12Resource_iface, &IID_ID3D12Resource, riid, object);
+    }
+
+    if (IsEqualGUID(riid, &IID_ID3D12Fence))
+    {
+        VkImportSemaphoreWin32HandleInfoKHR import_info;
+        struct d3d12_shared_fence *fence;
+        VkResult vr;
+
+        hr = d3d12_shared_fence_create(device, 0, D3D12_FENCE_FLAG_SHARED, &fence);
+
+        if (FAILED(hr))
+        {
+            WARN("Failed to create object for imported ID3D12Fence, hr %#x.\n", hr);
+            *object = NULL;
+            return hr;
+        }
+
+        import_info.sType = VK_STRUCTURE_TYPE_IMPORT_SEMAPHORE_WIN32_HANDLE_INFO_KHR;
+        import_info.pNext = NULL;
+        import_info.semaphore = fence->timeline_semaphore;
+        import_info.flags = 0;
+        import_info.handleType = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_D3D12_FENCE_BIT;
+        import_info.handle = handle;
+        import_info.name = NULL;
+
+        vr = VK_CALL(vkImportSemaphoreWin32HandleKHR(device->vk_device, &import_info));
+
+        if (vr != VK_SUCCESS)
+        {
+            WARN("Failed to open shared ID3D12Fence, vr %d.\n", vr);
+            ID3D12Fence1_Release(&fence->ID3D12Fence_iface);
+            *object = NULL;
+            return E_FAIL;
+        }
+
+        return return_interface(&fence->ID3D12Fence_iface, &IID_ID3D12Fence, riid, object);
+    }
+
+    FIXME("Opening shared handle type %s unsupported\n", debugstr_guid(riid));
     return E_NOTIMPL;
+#else
+    FIXME("OpenSharedhandle can only be implemented in native Win32.\n");
+    return E_NOTIMPL;
+#endif
 }
 
 static HRESULT STDMETHODCALLTYPE d3d12_device_OpenSharedHandleByName(d3d12_device_iface *iface,
@@ -4421,12 +4695,24 @@ static HRESULT STDMETHODCALLTYPE d3d12_device_CreateFence(d3d12_device_iface *if
         UINT64 initial_value, D3D12_FENCE_FLAGS flags, REFIID riid, void **fence)
 {
     struct d3d12_device *device = impl_from_ID3D12Device(iface);
+    struct d3d12_shared_fence *shared_object;
     struct d3d12_fence *object;
     HRESULT hr;
 
     TRACE("iface %p, intial_value %#"PRIx64", flags %#x, riid %s, fence %p.\n",
             iface, initial_value, flags, debugstr_guid(riid), fence);
 
+    if (flags & D3D12_FENCE_FLAG_SHARED)
+    {
+        if (SUCCEEDED(hr = d3d12_shared_fence_create(device, initial_value, flags, &shared_object)))
+            return return_interface(&shared_object->ID3D12Fence_iface, &IID_ID3D12Fence, riid, fence);
+
+        if (hr != E_NOTIMPL)
+            return hr;
+
+        FIXME("Shared fences not supported by Vulkan host, returning regular fence.\n");
+    }
+
     if (FAILED(hr = d3d12_fence_create(device, initial_value, flags, &object)))
         return hr;
 
@@ -4935,7 +5221,7 @@ static HRESULT STDMETHODCALLTYPE d3d12_device_CreateStateObject(d3d12_device_ifa
     TRACE("iface %p, desc %p, iid %s, state_object %p!\n",
             iface, desc, debugstr_guid(iid), state_object);
 
-    if (FAILED(hr = d3d12_state_object_create(device, desc, &state)))
+    if (FAILED(hr = d3d12_state_object_create(device, desc, NULL, &state)))
         return hr;
 
     return return_interface(&state->ID3D12StateObject_iface, &IID_ID3D12StateObject, iid, state_object);
@@ -4979,6 +5265,10 @@ static void STDMETHODCALLTYPE d3d12_device_GetRaytracingAccelerationStructurePre
     info->ResultDataMaxSizeInBytes = size_info.accelerationStructureSize;
     info->ScratchDataSizeInBytes = size_info.buildScratchSize;
     info->UpdateScratchDataSizeInBytes = size_info.updateScratchSize;
+
+    TRACE("ResultDataMaxSizeInBytes: %"PRIu64".\n", info->ResultDataMaxSizeInBytes);
+    TRACE("ScratchDatSizeInBytes: %"PRIu64".\n", info->ScratchDataSizeInBytes);
+    TRACE("UpdateScratchDataSizeInBytes: %"PRIu64".\n", info->UpdateScratchDataSizeInBytes);
 }
 
 static D3D12_DRIVER_MATCHING_IDENTIFIER_STATUS STDMETHODCALLTYPE d3d12_device_CheckDriverMatchingIdentifier(d3d12_device_iface *iface,
@@ -5003,13 +5293,23 @@ static HRESULT STDMETHODCALLTYPE d3d12_device_SetBackgroundProcessingMode(d3d12_
     return E_NOTIMPL;
 }
 
-static HRESULT STDMETHODCALLTYPE d3d12_device_AddToStateObject(d3d12_device_iface *iface, const D3D12_STATE_OBJECT_DESC *addition,
-        ID3D12StateObject *state_object, REFIID riid, void **new_state_object)
+static HRESULT STDMETHODCALLTYPE d3d12_device_AddToStateObject(d3d12_device_iface *iface,
+        const D3D12_STATE_OBJECT_DESC *addition,
+        ID3D12StateObject *parent_state, REFIID riid, void **new_state_object)
 {
-    FIXME("iface %p, addition %p, state_object %p, riid %s, new_state_object %p stub!\n",
-            iface, addition, state_object, debugstr_guid(riid), new_state_object);
+    struct d3d12_device *device = impl_from_ID3D12Device(iface);
+    struct d3d12_state_object *parent;
+    struct d3d12_state_object *state;
+    HRESULT hr;
 
-    return E_NOTIMPL;
+    TRACE("iface %p, addition %p, state_object %p, riid %s, new_state_object %p stub!\n",
+            iface, addition, parent_state, debugstr_guid(riid), new_state_object);
+
+    parent = impl_from_ID3D12StateObject(parent_state);
+    if (FAILED(hr = d3d12_state_object_add(device, addition, parent, &state)))
+        return hr;
+
+    return return_interface(&state->ID3D12StateObject_iface, &IID_ID3D12StateObject, riid, new_state_object);
 }
 
 static HRESULT STDMETHODCALLTYPE d3d12_device_CreateProtectedResourceSession1(d3d12_device_iface *iface,
@@ -5114,7 +5414,7 @@ static HRESULT STDMETHODCALLTYPE d3d12_device_CreateCommittedResource2(d3d12_dev
         FIXME("Ignoring protected session %p.\n", protected_session);
 
     if (FAILED(hr = d3d12_resource_create_committed(device, desc, heap_properties,
-            heap_flags, initial_state, optimized_clear_value, &object)))
+            heap_flags, initial_state, optimized_clear_value, NULL, &object)))
     {
         *resource = NULL;
         return hr;
@@ -5539,16 +5839,14 @@ static D3D12_RESOURCE_HEAP_TIER d3d12_device_determine_heap_tier(struct d3d12_de
     const VkPhysicalDeviceLimits *limits = &device->device_info.properties2.properties.limits;
     const struct vkd3d_memory_info *mem_info = &device->memory_info;
     const struct vkd3d_memory_info_domain *non_cpu_domain;
-    const struct vkd3d_memory_info_domain *cpu_domain;
 
     non_cpu_domain = &mem_info->non_cpu_accessible_domain;
-    cpu_domain = &mem_info->cpu_accessible_domain;
 
-    // Heap Tier 2 requires us to be able to create a heap that supports all resource
-    // categories at the same time, except RT/DS textures on UPLOAD/READBACK heaps.
+    /* Heap Tier 2 requires us to be able to create a heap that supports all resource
+     * categories at the same time, except RT/DS textures on UPLOAD/READBACK heaps.
+     * Ignore CPU visible heaps since we only place buffers there. Textures are promoted to committed always. */
     if (limits->bufferImageGranularity > D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT ||
-            !(non_cpu_domain->buffer_type_mask & non_cpu_domain->sampled_type_mask & non_cpu_domain->rt_ds_type_mask) ||
-            !(cpu_domain->buffer_type_mask & cpu_domain->sampled_type_mask))
+            !(non_cpu_domain->buffer_type_mask & non_cpu_domain->sampled_type_mask & non_cpu_domain->rt_ds_type_mask))
         return D3D12_RESOURCE_HEAP_TIER_1;
 
     return D3D12_RESOURCE_HEAP_TIER_2;
@@ -5962,6 +6260,27 @@ static void d3d12_device_caps_init_shader_model(struct d3d12_device *device)
     }
 }
 
+static void d3d12_device_caps_override_application(struct d3d12_device *device)
+{
+    /* Some games rely on certain features to be exposed before they let the primary feature
+     * be exposed. */
+    switch (vkd3d_application_feature_override)
+    {
+        case VKD3D_APPLICATION_FEATURE_OVERRIDE_PROMOTE_DXR_TO_ULTIMATE:
+            if (device->d3d12_caps.options5.RaytracingTier >= D3D12_RAYTRACING_TIER_1_0)
+            {
+                device->d3d12_caps.options7.MeshShaderTier = D3D12_MESH_SHADER_TIER_1;
+                device->d3d12_caps.options7.SamplerFeedbackTier = D3D12_SAMPLER_FEEDBACK_TIER_1_0;
+                INFO("DXR enabled. Application also requires Mesh/Sampler feedback to be exposed (but unused). "
+                     "Enabling these features automatically.\n");
+            }
+            break;
+
+        default:
+            break;
+    }
+}
+
 static void d3d12_device_caps_override(struct d3d12_device *device)
 {
     D3D_FEATURE_LEVEL fl_override = (D3D_FEATURE_LEVEL)0;
@@ -6052,6 +6371,7 @@ static void d3d12_device_caps_init(struct d3d12_device *device)
     d3d12_device_caps_init_feature_level(device);
 
     d3d12_device_caps_override(device);
+    d3d12_device_caps_override_application(device);
 }
 
 static void vkd3d_init_shader_extensions(struct d3d12_device *device)
@@ -6105,7 +6425,8 @@ static void vkd3d_init_shader_extensions(struct d3d12_device *device)
     }
 
     if (device->d3d12_caps.options4.Native16BitShaderOpsSupported &&
-            device->device_info.driver_properties.driverID == VK_DRIVER_ID_MESA_RADV)
+            (device->device_info.driver_properties.driverID == VK_DRIVER_ID_MESA_RADV ||
+                    (vkd3d_config_flags & VKD3D_CONFIG_FLAG_FORCE_NATIVE_FP16)))
     {
         /* Native FP16 is buggy on NV for now. */
         device->vk_info.shader_extensions[device->vk_info.shader_extension_count++] =

libs/vkd3d/memory.c

@@ -327,23 +327,25 @@ static HRESULT vkd3d_import_host_memory(struct d3d12_device *device, void *host_
         void *pNext, struct vkd3d_device_memory_allocation *allocation)
 {
     VkImportMemoryHostPointerInfoEXT import_info;
-    HRESULT hr;
+    HRESULT hr = S_OK;
 
     import_info.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_HOST_POINTER_INFO_EXT;
     import_info.pNext = pNext;
     import_info.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT;
     import_info.pHostPointer = host_address;
 
-    if (FAILED(hr = vkd3d_try_allocate_device_memory(device, size,
+    if ((vkd3d_config_flags & VKD3D_CONFIG_FLAG_USE_HOST_IMPORT_FALLBACK) ||
+        FAILED(hr = vkd3d_try_allocate_device_memory(device, size,
             type_flags, type_mask, &import_info, allocation)))
     {
-        WARN("Failed to import host memory, hr %#x.\n", hr);
+        if (FAILED(hr))
+            WARN("Failed to import host memory, hr %#x.\n", hr);
         /* If we failed, fall back to a host-visible allocation. Generally
          * the app will access the memory thorugh the main host pointer,
          * so it's fine. */
         hr = vkd3d_try_allocate_device_memory(device, size,
                 VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT | VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
-                type_mask, &import_info, allocation);
+                type_mask, pNext, allocation);
     }
 
     return hr;
@@ -1155,6 +1157,7 @@ static HRESULT vkd3d_memory_allocator_flush_clears_locked(struct vkd3d_memory_al
         for (i = 0; i < queue_family->queue_count; i++)
         {
             vkd3d_queue_add_wait(queue_family->queues[i],
+                    NULL,
                     clear_queue->vk_semaphore,
                     clear_queue->next_signal_value);
         }
@@ -1477,6 +1480,7 @@ static bool vkd3d_heap_allocation_accept_deferred_resource_placements(struct d3d
 HRESULT vkd3d_allocate_heap_memory(struct d3d12_device *device, struct vkd3d_memory_allocator *allocator,
         const struct vkd3d_allocate_heap_memory_info *info, struct vkd3d_memory_allocation *allocation)
 {
+    struct vkd3d_allocate_heap_memory_info heap_info;
     struct vkd3d_allocate_memory_info alloc_info;
     HRESULT hr;
 
@@ -1492,6 +1496,27 @@ HRESULT vkd3d_allocate_heap_memory(struct d3d12_device *device, struct vkd3d_mem
     if (!(info->heap_desc.Flags & D3D12_HEAP_FLAG_DENY_BUFFERS))
         alloc_info.flags |= VKD3D_ALLOCATION_FLAG_GLOBAL_BUFFER;
 
+    if (is_cpu_accessible_heap(&info->heap_desc.Properties))
+    {
+        if (info->heap_desc.Flags & D3D12_HEAP_FLAG_DENY_BUFFERS)
+        {
+            /* If the heap was only designed to handle images, the heap is useless,
+             * and we can force everything to go through committed path. */
+            memset(allocation, 0, sizeof(*allocation));
+            return S_OK;
+        }
+        else
+        {
+            /* CPU visible textures are never placed on a heap directly,
+             * since LINEAR images have alignment / size requirements
+             * that are vastly different from OPTIMAL ones.
+             * We can place buffers however. */
+            heap_info = *info;
+            info = &heap_info;
+            heap_info.heap_desc.Flags |= D3D12_HEAP_FLAG_ALLOW_ONLY_BUFFERS;
+        }
+    }
+
     hr = vkd3d_allocate_memory(device, allocator, &alloc_info, allocation);
     if (hr == E_OUTOFMEMORY && vkd3d_heap_allocation_accept_deferred_resource_placements(device,
             &info->heap_desc.Properties, info->heap_desc.Flags))

libs/vkd3d/meson.build

@@ -28,6 +28,7 @@ vkd3d_shaders =[
   'shaders/vs_swapchain_fullscreen.vert',
   'shaders/fs_swapchain_fullscreen.frag',
   'shaders/cs_execute_indirect_patch.comp',
+  'shaders/cs_execute_indirect_patch_debug_ring.comp',
 ]
 
 vkd3d_src = [
@@ -66,6 +67,10 @@ if enable_breadcrumbs
   vkd3d_src += ['breadcrumbs.c']
 endif
 
+if vkd3d_platform == 'windows'
+  vkd3d_src += ['shared_metadata.c']
+endif
+
 if not enable_d3d12
   vkd3d_lib = shared_library('vkd3d-proton', vkd3d_src, glsl_generator.process(vkd3d_shaders), vkd3d_build, vkd3d_version,
     dependencies        : [ vkd3d_common_dep, vkd3d_shader_dep ] + vkd3d_extra_libs,

libs/vkd3d/meta.c

@@ -1244,14 +1244,24 @@ HRESULT vkd3d_execute_indirect_ops_init(struct vkd3d_execute_indirect_ops *meta_
     return S_OK;
 }
 
+struct vkd3d_meta_execute_indirect_spec_constant_data
+{
+    struct vkd3d_shader_debug_ring_spec_constants constants;
+    uint32_t workgroup_size_x;
+};
+
 HRESULT vkd3d_meta_get_execute_indirect_pipeline(struct vkd3d_meta_ops *meta_ops,
         uint32_t patch_command_count, struct vkd3d_execute_indirect_info *info)
 {
+    struct vkd3d_meta_execute_indirect_spec_constant_data execute_indirect_spec_constants;
+    VkSpecializationMapEntry map_entry[VKD3D_SHADER_DEBUG_RING_SPEC_INFO_MAP_ENTRIES + 1];
     struct vkd3d_execute_indirect_ops *meta_indirect_ops = &meta_ops->execute_indirect;
-    VkSpecializationMapEntry map_entry;
+    struct vkd3d_shader_debug_ring_spec_info debug_ring_info;
+
     VkSpecializationInfo spec;
     HRESULT hr = S_OK;
     VkResult vr;
+    bool debug;
     size_t i;
     int rc;
 
@@ -1271,14 +1281,41 @@ HRESULT vkd3d_meta_get_execute_indirect_pipeline(struct vkd3d_meta_ops *meta_ops
         }
     }
 
-    map_entry.constantID = 0;
-    map_entry.offset = 0;
-    map_entry.size = sizeof(patch_command_count);
+    debug = meta_ops->device->debug_ring.active;
 
-    spec.pMapEntries = &map_entry;
-    spec.pData = &patch_command_count;
-    spec.mapEntryCount = 1;
-    spec.dataSize = sizeof(patch_command_count);
+    /* If we have debug ring, we can dump indirect command buffer data to the ring as well.
+     * Vital for debugging broken execute indirect data with templates. */
+    if (debug)
+    {
+        vkd3d_shader_debug_ring_init_spec_constant(meta_ops->device, &debug_ring_info,
+                0 /* Reserve this hash for internal debug streams. */);
+
+        memset(&execute_indirect_spec_constants, 0, sizeof(execute_indirect_spec_constants));
+        execute_indirect_spec_constants.constants = debug_ring_info.constants;
+        execute_indirect_spec_constants.workgroup_size_x = patch_command_count;
+
+        memcpy(map_entry, debug_ring_info.map_entries, sizeof(debug_ring_info.map_entries));
+        map_entry[VKD3D_SHADER_DEBUG_RING_SPEC_INFO_MAP_ENTRIES].constantID = 4;
+        map_entry[VKD3D_SHADER_DEBUG_RING_SPEC_INFO_MAP_ENTRIES].offset =
+                offsetof(struct vkd3d_meta_execute_indirect_spec_constant_data, workgroup_size_x);
+        map_entry[VKD3D_SHADER_DEBUG_RING_SPEC_INFO_MAP_ENTRIES].size = sizeof(patch_command_count);
+
+        spec.pMapEntries = map_entry;
+        spec.pData = &execute_indirect_spec_constants;
+        spec.mapEntryCount = ARRAY_SIZE(map_entry);
+        spec.dataSize = sizeof(execute_indirect_spec_constants);
+    }
+    else
+    {
+        map_entry[0].constantID = 0;
+        map_entry[0].offset = 0;
+        map_entry[0].size = sizeof(patch_command_count);
+
+        spec.pMapEntries = map_entry;
+        spec.pData = &patch_command_count;
+        spec.mapEntryCount = 1;
+        spec.dataSize = sizeof(patch_command_count);
+    }
 
     vkd3d_array_reserve((void**)&meta_indirect_ops->pipelines, &meta_indirect_ops->pipelines_size,
             meta_indirect_ops->pipelines_count + 1, sizeof(*meta_indirect_ops->pipelines));
@@ -1286,7 +1323,8 @@ HRESULT vkd3d_meta_get_execute_indirect_pipeline(struct vkd3d_meta_ops *meta_ops
     meta_indirect_ops->pipelines[meta_indirect_ops->pipelines_count].workgroup_size_x = patch_command_count;
 
     vr = vkd3d_meta_create_compute_pipeline(meta_ops->device,
-            sizeof(cs_execute_indirect_patch), cs_execute_indirect_patch,
+            debug ? sizeof(cs_execute_indirect_patch_debug_ring) : sizeof(cs_execute_indirect_patch),
+            debug ? cs_execute_indirect_patch_debug_ring : cs_execute_indirect_patch,
             meta_indirect_ops->vk_pipeline_layout, &spec,
             &meta_indirect_ops->pipelines[meta_indirect_ops->pipelines_count].vk_pipeline);
 

libs/vkd3d/resource.c

@@ -488,6 +488,7 @@ static bool vkd3d_format_check_usage_support(struct d3d12_device *device, VkForm
 struct vkd3d_image_create_info
 {
     struct vkd3d_format_compatibility_list format_compat_list;
+    VkExternalMemoryImageCreateInfo external_info;
     VkImageFormatListCreateInfoKHR format_list;
     VkImageCreateInfo image_info;
 };
@@ -498,6 +499,7 @@ static HRESULT vkd3d_get_image_create_info(struct d3d12_device *device,
         struct vkd3d_image_create_info *create_info)
 {
     struct vkd3d_format_compatibility_list *compat_list = &create_info->format_compat_list;
+    VkExternalMemoryImageCreateInfo *external_info = &create_info->external_info;
     VkImageFormatListCreateInfoKHR *format_list = &create_info->format_list;
     const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs;
     VkImageCreateInfo *image_info = &create_info->image_info;
@@ -522,12 +524,22 @@ static HRESULT vkd3d_get_image_create_info(struct d3d12_device *device,
     image_info->sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO;
     image_info->pNext = NULL;
     image_info->flags = 0;
+
+    if (resource && (resource->heap_flags & D3D12_HEAP_FLAG_SHARED))
+    {
+        external_info->sType = VK_STRUCTURE_TYPE_EXTERNAL_MEMORY_IMAGE_CREATE_INFO;
+        external_info->pNext = NULL;
+        external_info->handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT;
+
+        image_info->pNext = external_info;
+    }
+
     if (!(desc->Flags & D3D12_RESOURCE_FLAG_ALLOW_DEPTH_STENCIL))
     {
         if (vkd3d_get_format_compatibility_list(device, desc, compat_list))
         {
             format_list->sType = VK_STRUCTURE_TYPE_IMAGE_FORMAT_LIST_CREATE_INFO_KHR;
-            format_list->pNext = NULL;
+            format_list->pNext = image_info->pNext;
             format_list->viewFormatCount = compat_list->format_count;
             format_list->pViewFormats = compat_list->vk_formats;
 
@@ -2701,7 +2713,7 @@ static HRESULT d3d12_resource_create(struct d3d12_device *device, uint32_t flags
 
 HRESULT d3d12_resource_create_committed(struct d3d12_device *device, const D3D12_RESOURCE_DESC1 *desc,
         const D3D12_HEAP_PROPERTIES *heap_properties, D3D12_HEAP_FLAGS heap_flags, D3D12_RESOURCE_STATES initial_state,
-        const D3D12_CLEAR_VALUE *optimized_clear_value, struct d3d12_resource **resource)
+        const D3D12_CLEAR_VALUE *optimized_clear_value, HANDLE shared_handle, struct d3d12_resource **resource)
 {
     const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs;
     struct d3d12_resource *object;
@@ -2722,6 +2734,11 @@ HRESULT d3d12_resource_create_committed(struct d3d12_device *device, const D3D12
         bool use_dedicated_allocation;
         VkResult vr;
 
+#ifdef _WIN32
+        VkImportMemoryWin32HandleInfoKHR import_info;
+        VkExportMemoryAllocateInfo export_info;
+#endif
+
         if (FAILED(hr = d3d12_resource_create_vk_resource(object, device)))
             goto fail;
 
@@ -2754,10 +2771,36 @@ HRESULT d3d12_resource_create_committed(struct d3d12_device *device, const D3D12
         else
             allocate_info.heap_flags |= D3D12_HEAP_FLAG_ALLOW_ONLY_NON_RT_DS_TEXTURES;
 
+        if (heap_flags & D3D12_HEAP_FLAG_SHARED)
+        {
+#ifdef _WIN32
+            use_dedicated_allocation = true;
+
+            if (shared_handle && shared_handle != INVALID_HANDLE_VALUE)
+            {
+                import_info.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_WIN32_HANDLE_INFO_KHR;
+                import_info.pNext = allocate_info.pNext;
+                import_info.handleType = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT;
+                import_info.handle = shared_handle;
+                import_info.name = NULL;
+                allocate_info.pNext = &import_info;
+            }
+            else
+            {
+                export_info.sType = VK_STRUCTURE_TYPE_EXPORT_MEMORY_ALLOCATE_INFO;
+                export_info.pNext = allocate_info.pNext;
+                export_info.handleTypes = VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_WIN32_BIT;
+                allocate_info.pNext = &export_info;
+            }
+#else
+            FIXME("D3D12_HEAP_FLAG_SHARED can only be implemented in native Win32.\n");
+#endif
+        }
+
         if (use_dedicated_allocation)
         {
             dedicated_info.sType = VK_STRUCTURE_TYPE_MEMORY_DEDICATED_ALLOCATE_INFO;
-            dedicated_info.pNext = NULL;
+            dedicated_info.pNext = allocate_info.pNext;
             dedicated_info.image = object->res.vk_image;
             dedicated_info.buffer = VK_NULL_HANDLE;
             allocate_info.pNext = &dedicated_info;
@@ -2863,20 +2906,30 @@ HRESULT d3d12_resource_create_placed(struct d3d12_device *device, const D3D12_RE
     VkMemoryRequirements memory_requirements;
     VkBindImageMemoryInfo bind_info;
     struct d3d12_resource *object;
+    bool force_committed;
     VkResult vr;
     HRESULT hr;
 
     if (FAILED(hr = d3d12_resource_validate_heap(desc, heap)))
         return hr;
 
-    if (heap->allocation.device_allocation.vk_memory == VK_NULL_HANDLE)
+    /* Placed linear textures are ... problematic
+     * since we have no way of signalling that they have different alignment and size requirements
+     * than optimal textures. GetResourceAllocationInfo() does not take heap property information
+     * and assumes that we are not modifying the tiling mode. */
+    force_committed = desc->Dimension != D3D12_RESOURCE_DIMENSION_BUFFER &&
+            is_cpu_accessible_heap(&heap->desc.Properties);
+
+    if (force_committed || heap->allocation.device_allocation.vk_memory == VK_NULL_HANDLE)
     {
-        WARN("Placing resource on heap with no memory backing it. Falling back to committed resource.\n");
+        if (!force_committed)
+            WARN("Placing resource on heap with no memory backing it. Falling back to committed resource.\n");
+
         if (FAILED(hr = d3d12_resource_create_committed(device, desc, &heap->desc.Properties,
                 heap->desc.Flags & ~(D3D12_HEAP_FLAG_DENY_BUFFERS |
                         D3D12_HEAP_FLAG_DENY_NON_RT_DS_TEXTURES |
                         D3D12_HEAP_FLAG_DENY_RT_DS_TEXTURES),
-                initial_state, optimized_clear_value, resource)))
+                initial_state, optimized_clear_value, NULL, resource)))
         {
             ERR("Failed to create fallback committed resource.\n");
         }

libs/vkd3d/shaders/cs_execute_indirect_patch.comp

@@ -11,12 +11,6 @@ struct Command
 	uint dst_offset;
 };
 
-const int COMMAND_TYPE_COPY_U32 = 0;
-const int COMMAND_TYPE_COPY_INDEX_TYPE = 1;
-const int DXGI_FORMAT_R32_UINT = 0x2a;
-const int VK_INDEX_TYPE_UINT16 = 0;
-const int VK_INDEX_TYPE_UINT32 = 1;
-
 layout(buffer_reference, std430, buffer_reference_align = 4) readonly buffer Commands
 {
 	Command commands[];
@@ -54,10 +48,11 @@ void main()
 	Command cmd = commands_va.commands[gl_LocalInvocationIndex];
 
 	uint draw_id = gl_WorkGroupID.x;
-	uint max_draws = ~0u;
+	uint max_draws = gl_NumWorkGroups.x;
+
 	if (any(notEqual(indirect_count_va, uvec2(0))))
 	{
-		max_draws = IndirectCount(indirect_count_va).count;
+		max_draws = min(max_draws, IndirectCount(indirect_count_va).count);
 		if (gl_WorkGroupID.x == 0u)
 			dst_indirect_count_va.count = max_draws;
 	}
@@ -66,11 +61,7 @@ void main()
 	{
 		uint src_offset = src_stride * draw_id + cmd.src_offset;
 		uint dst_offset = dst_stride * draw_id + cmd.dst_offset;
-
 		uint src_value = src_buffer_va.values[src_offset];
-		if (cmd.type == COMMAND_TYPE_COPY_INDEX_TYPE)
-			src_value = src_value == DXGI_FORMAT_R32_UINT ? VK_INDEX_TYPE_UINT32 : VK_INDEX_TYPE_UINT16;
-
 		dst_buffer_va.values[dst_offset] = src_value;
 	}
 }

libs/vkd3d/shaders/cs_execute_indirect_patch_debug_ring.comp

@@ -0,0 +1,83 @@
+#version 450
+#extension GL_EXT_buffer_reference : require
+#extension GL_EXT_buffer_reference_uvec2 : require
+#extension GL_GOOGLE_include_directive : require
+#include "../../../include/shader-debug/debug_channel.h"
+
+layout(local_size_x_id = 4) in;
+
+struct Command
+{
+	uint type;
+	uint src_offset;
+	uint dst_offset;
+};
+
+layout(buffer_reference, std430, buffer_reference_align = 4) readonly buffer Commands
+{
+	Command commands[];
+};
+
+layout(buffer_reference, std430, buffer_reference_align = 4) readonly buffer SrcBuffer {
+	uint values[];
+};
+
+layout(buffer_reference, std430, buffer_reference_align = 4) writeonly buffer DstBuffer {
+	uint values[];
+};
+
+layout(buffer_reference, std430, buffer_reference_align = 4) readonly buffer IndirectCount {
+	uint count;
+};
+
+layout(buffer_reference, std430, buffer_reference_align = 4) writeonly buffer IndirectCountWrite {
+	uint count;
+};
+
+layout(push_constant) uniform Registers
+{
+	Commands commands_va;
+	SrcBuffer src_buffer_va;
+	DstBuffer dst_buffer_va;
+	uvec2 indirect_count_va;
+	IndirectCountWrite dst_indirect_count_va;
+	uint src_stride;
+	uint dst_stride;
+
+	// Debug metadata here
+	uint debug_tag;
+    uint implicit_instance;
+};
+
+void main()
+{
+    if (debug_tag != 0u)
+        DEBUG_CHANNEL_INIT_IMPLICIT_INSTANCE(uvec3(debug_tag, gl_WorkGroupID.x, gl_LocalInvocationIndex), implicit_instance);
+
+    Command cmd = commands_va.commands[gl_LocalInvocationIndex];
+
+    uint draw_id = gl_WorkGroupID.x;
+    uint max_draws = gl_NumWorkGroups.x;
+    if (any(notEqual(indirect_count_va, uvec2(0))))
+    {
+        max_draws = min(max_draws, IndirectCount(indirect_count_va).count);
+        if (gl_WorkGroupID.x == 0u)
+            dst_indirect_count_va.count = max_draws;
+    }
+
+    if (debug_tag != 0u && gl_WorkGroupID.x == 0)
+        DEBUG_CHANNEL_MSG_UNIFORM(int(max_draws), int(gl_NumWorkGroups.x));
+
+    if (draw_id < max_draws)
+    {
+        uint src_offset = src_stride * draw_id + cmd.src_offset;
+        uint dst_offset = dst_stride * draw_id + cmd.dst_offset;
+
+        uint src_value = src_buffer_va.values[src_offset];
+
+        if (debug_tag != 0u)
+            DEBUG_CHANNEL_MSG(cmd.type, dst_offset, src_offset, src_value);
+
+        dst_buffer_va.values[dst_offset] = src_value;
+    }
+}

libs/vkd3d/shared_metadata.c

@@ -0,0 +1,68 @@
+/*
+ * Copyright 2021 Derek Lesho for Codeweavers
+ *
+ * This library is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * This library is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with this library; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301, USA
+ */
+
+#define VKD3D_DBG_CHANNEL VKD3D_DBG_CHANNEL_API
+#include "vkd3d_private.h"
+
+#include "winioctl.h"
+
+#define IOCTL_SHARED_GPU_RESOURCE_SET_METADATA           CTL_CODE(FILE_DEVICE_VIDEO, 4, METHOD_BUFFERED, FILE_WRITE_ACCESS)
+#define IOCTL_SHARED_GPU_RESOURCE_GET_METADATA           CTL_CODE(FILE_DEVICE_VIDEO, 5, METHOD_BUFFERED, FILE_READ_ACCESS)
+#define IOCTL_SHARED_GPU_RESOURCE_OPEN                   CTL_CODE(FILE_DEVICE_VIDEO, 1, METHOD_BUFFERED, FILE_WRITE_ACCESS)
+
+bool vkd3d_set_shared_metadata(HANDLE handle, void *buf, uint32_t buf_size)
+{
+    DWORD ret_size;
+
+    return DeviceIoControl(handle, IOCTL_SHARED_GPU_RESOURCE_SET_METADATA, buf, buf_size, NULL, 0, &ret_size, NULL);
+}
+
+bool vkd3d_get_shared_metadata(HANDLE handle, void *buf, uint32_t buf_size, uint32_t *metadata_size)
+{
+    DWORD ret_size;
+
+    bool ret = DeviceIoControl(handle, IOCTL_SHARED_GPU_RESOURCE_GET_METADATA, NULL, 0, buf, buf_size, &ret_size, NULL);
+
+    if (metadata_size)
+        *metadata_size = ret_size;
+
+    return ret;
+}
+
+HANDLE vkd3d_open_kmt_handle(HANDLE kmt_handle)
+{
+    struct
+    {
+        unsigned int kmt_handle;
+        /* the following parameter represents a larger sized string for a dynamically allocated struct for use when opening an object by name */
+        WCHAR name[1];
+    } shared_resource_open;
+
+    HANDLE nt_handle = CreateFileA("\\\\.\\SharedGpuResource", GENERIC_READ | GENERIC_WRITE, 0, NULL, OPEN_EXISTING, FILE_ATTRIBUTE_NORMAL, NULL);
+    if (nt_handle == INVALID_HANDLE_VALUE)
+        return nt_handle;
+
+    shared_resource_open.kmt_handle = (ULONG_PTR)kmt_handle;
+    shared_resource_open.name[0] = 0;
+    if (!DeviceIoControl(nt_handle, IOCTL_SHARED_GPU_RESOURCE_OPEN, &shared_resource_open, sizeof(shared_resource_open), NULL, 0, NULL, NULL))
+    {
+        CloseHandle(nt_handle);
+        return INVALID_HANDLE_VALUE;
+    }
+    return nt_handle;
+}

libs/vkd3d/state.c

@@ -1473,8 +1473,36 @@ fail:
     return hr;
 }
 
-HRESULT d3d12_root_signature_create(struct d3d12_device *device,
-        const void *bytecode, size_t bytecode_length, struct d3d12_root_signature **root_signature)
+HRESULT d3d12_root_signature_create_empty(struct d3d12_device *device,
+        struct d3d12_root_signature **root_signature)
+{
+    struct d3d12_root_signature *object;
+    D3D12_ROOT_SIGNATURE_DESC1 desc;
+    HRESULT hr;
+
+    if (!(object = vkd3d_malloc(sizeof(*object))))
+        return E_OUTOFMEMORY;
+
+    memset(&desc, 0, sizeof(desc));
+    hr = d3d12_root_signature_init(object, device, &desc);
+
+    /* For pipeline libraries, (and later DXR to some degree), we need a way to
+     * compare root signature objects. */
+    object->compatibility_hash = 0;
+
+    if (FAILED(hr))
+    {
+        vkd3d_free(object);
+        return hr;
+    }
+
+    *root_signature = object;
+    return S_OK;
+}
+
+static HRESULT d3d12_root_signature_create_from_blob(struct d3d12_device *device,
+        const void *bytecode, size_t bytecode_length, bool raw_payload,
+        struct d3d12_root_signature **root_signature)
 {
     const struct vkd3d_shader_code dxbc = {bytecode, bytecode_length};
     union
@@ -1482,14 +1510,26 @@ HRESULT d3d12_root_signature_create(struct d3d12_device *device,
         D3D12_VERSIONED_ROOT_SIGNATURE_DESC d3d12;
         struct vkd3d_versioned_root_signature_desc vkd3d;
     } root_signature_desc;
+    vkd3d_shader_hash_t compatibility_hash;
     struct d3d12_root_signature *object;
     HRESULT hr;
     int ret;
 
-    if ((ret = vkd3d_parse_root_signature_v_1_1(&dxbc, &root_signature_desc.vkd3d)) < 0)
+    if (raw_payload)
     {
-        WARN("Failed to parse root signature, vkd3d result %d.\n", ret);
-        return hresult_from_vkd3d_result(ret);
+        if ((ret = vkd3d_parse_root_signature_v_1_1_from_raw_payload(&dxbc, &root_signature_desc.vkd3d, &compatibility_hash)))
+        {
+            WARN("Failed to parse root signature, vkd3d result %d.\n", ret);
+            return hresult_from_vkd3d_result(ret);
+        }
+    }
+    else
+    {
+        if ((ret = vkd3d_parse_root_signature_v_1_1(&dxbc, &root_signature_desc.vkd3d, &compatibility_hash)) < 0)
+        {
+            WARN("Failed to parse root signature, vkd3d result %d.\n", ret);
+            return hresult_from_vkd3d_result(ret);
+        }
     }
 
     if (!(object = vkd3d_malloc(sizeof(*object))))
@@ -1502,7 +1542,7 @@ HRESULT d3d12_root_signature_create(struct d3d12_device *device,
 
     /* For pipeline libraries, (and later DXR to some degree), we need a way to
      * compare root signature objects. */
-    object->compatibility_hash = vkd3d_shader_hash(&dxbc);
+    object->compatibility_hash = compatibility_hash;
 
     vkd3d_shader_free_root_signature(&root_signature_desc.vkd3d);
     if (FAILED(hr))
@@ -1518,6 +1558,20 @@ HRESULT d3d12_root_signature_create(struct d3d12_device *device,
     return S_OK;
 }
 
+HRESULT d3d12_root_signature_create(struct d3d12_device *device,
+        const void *bytecode, size_t bytecode_length,
+        struct d3d12_root_signature **root_signature)
+{
+    return d3d12_root_signature_create_from_blob(device, bytecode, bytecode_length, false, root_signature);
+}
+
+HRESULT d3d12_root_signature_create_raw(struct d3d12_device *device,
+        const void *payload, size_t payload_length,
+        struct d3d12_root_signature **root_signature)
+{
+    return d3d12_root_signature_create_from_blob(device, payload, payload_length, true, root_signature);
+}
+
 unsigned int d3d12_root_signature_get_shader_interface_flags(const struct d3d12_root_signature *root_signature)
 {
     unsigned int flags = 0;

libs/vkd3d/swapchain.c

@@ -959,7 +959,7 @@ static HRESULT d3d12_swapchain_create_user_buffers(struct d3d12_swapchain *swapc
         {
             if (FAILED(hr = d3d12_resource_create_committed(d3d12_swapchain_device(swapchain),
                     &resource_desc, &heap_properties, D3D12_HEAP_FLAG_NONE,
-                    D3D12_RESOURCE_STATE_PRESENT, NULL, &object)))
+                    D3D12_RESOURCE_STATE_PRESENT, NULL, NULL, &object)))
             {
                 ERR("Failed to create image for swapchain buffer");
                 return hr;

libs/vkd3d/vkd3d_main.c

@@ -165,15 +165,28 @@ static CONST_VTBL struct ID3D12RootSignatureDeserializerVtbl d3d12_root_signatur
 
 static int vkd3d_parse_root_signature_for_version(const struct vkd3d_shader_code *dxbc,
         struct vkd3d_versioned_root_signature_desc *out_desc,
-        enum vkd3d_root_signature_version target_version)
+        enum vkd3d_root_signature_version target_version,
+        bool raw_payload,
+        vkd3d_shader_hash_t *compatibility_hash)
 {
     struct vkd3d_versioned_root_signature_desc desc, converted_desc;
     int ret;
 
-    if ((ret = vkd3d_shader_parse_root_signature(dxbc, &desc)) < 0)
+    if (raw_payload)
     {
-        WARN("Failed to parse root signature, vkd3d result %d.\n", ret);
-        return ret;
+        if ((ret = vkd3d_shader_parse_root_signature_raw(dxbc->code, dxbc->size, &desc, compatibility_hash)) < 0)
+        {
+            WARN("Failed to parse root signature, vkd3d result %d.\n", ret);
+            return ret;
+        }
+    }
+    else
+    {
+        if ((ret = vkd3d_shader_parse_root_signature(dxbc, &desc, compatibility_hash)) < 0)
+        {
+            WARN("Failed to parse root signature, vkd3d result %d.\n", ret);
+            return ret;
+        }
     }
 
     if (desc.version == target_version)
@@ -197,15 +210,27 @@ static int vkd3d_parse_root_signature_for_version(const struct vkd3d_shader_code
 }
 
 int vkd3d_parse_root_signature_v_1_0(const struct vkd3d_shader_code *dxbc,
-        struct vkd3d_versioned_root_signature_desc *out_desc)
+        struct vkd3d_versioned_root_signature_desc *out_desc,
+        vkd3d_shader_hash_t *compatibility_hash)
 {
-    return vkd3d_parse_root_signature_for_version(dxbc, out_desc, VKD3D_ROOT_SIGNATURE_VERSION_1_0);
+    return vkd3d_parse_root_signature_for_version(dxbc, out_desc, VKD3D_ROOT_SIGNATURE_VERSION_1_0, false,
+            compatibility_hash);
 }
 
 int vkd3d_parse_root_signature_v_1_1(const struct vkd3d_shader_code *dxbc,
-        struct vkd3d_versioned_root_signature_desc *out_desc)
+        struct vkd3d_versioned_root_signature_desc *out_desc,
+        vkd3d_shader_hash_t *compatibility_hash)
 {
-    return vkd3d_parse_root_signature_for_version(dxbc, out_desc, VKD3D_ROOT_SIGNATURE_VERSION_1_1);
+    return vkd3d_parse_root_signature_for_version(dxbc, out_desc, VKD3D_ROOT_SIGNATURE_VERSION_1_1, false,
+            compatibility_hash);
+}
+
+int vkd3d_parse_root_signature_v_1_1_from_raw_payload(const struct vkd3d_shader_code *dxbc,
+        struct vkd3d_versioned_root_signature_desc *out_desc,
+        vkd3d_shader_hash_t *compatibility_hash)
+{
+    return vkd3d_parse_root_signature_for_version(dxbc, out_desc, VKD3D_ROOT_SIGNATURE_VERSION_1_1, true,
+            compatibility_hash);
 }
 
 static HRESULT d3d12_root_signature_deserializer_init(struct d3d12_root_signature_deserializer *deserializer,
@@ -216,7 +241,7 @@ static HRESULT d3d12_root_signature_deserializer_init(struct d3d12_root_signatur
     deserializer->ID3D12RootSignatureDeserializer_iface.lpVtbl = &d3d12_root_signature_deserializer_vtbl;
     deserializer->refcount = 1;
 
-    if ((ret = vkd3d_parse_root_signature_v_1_0(dxbc, &deserializer->desc.vkd3d)) < 0)
+    if ((ret = vkd3d_parse_root_signature_v_1_0(dxbc, &deserializer->desc.vkd3d, NULL)) < 0)
         return hresult_from_vkd3d_result(ret);
 
     return S_OK;
@@ -394,7 +419,7 @@ static HRESULT d3d12_versioned_root_signature_deserializer_init(struct d3d12_ver
     deserializer->ID3D12VersionedRootSignatureDeserializer_iface.lpVtbl = &d3d12_versioned_root_signature_deserializer_vtbl;
     deserializer->refcount = 1;
 
-    if ((ret = vkd3d_shader_parse_root_signature(dxbc, &deserializer->desc.vkd3d)) < 0)
+    if ((ret = vkd3d_shader_parse_root_signature(dxbc, &deserializer->desc.vkd3d, NULL)) < 0)
     {
         WARN("Failed to parse root signature, vkd3d result %d.\n", ret);
         return hresult_from_vkd3d_result(ret);

libs/vkd3d/vkd3d_private.h

@@ -61,6 +61,8 @@
 
 #define VKD3D_TILE_SIZE 65536
 
+typedef ID3D12Fence1 d3d12_fence_iface;
+
 struct d3d12_command_list;
 struct d3d12_device;
 struct d3d12_resource;
@@ -135,6 +137,8 @@ struct vkd3d_vulkan_info
     bool KHR_maintenance4;
     bool KHR_ray_tracing_maintenance1;
     bool KHR_fragment_shader_barycentric;
+    bool KHR_external_memory_win32;
+    bool KHR_external_semaphore_win32;
     /* EXT device extensions */
     bool EXT_calibrated_timestamps;
     bool EXT_conditional_rendering;
@@ -226,8 +230,12 @@ HRESULT vkd3d_join_thread(struct vkd3d_instance *instance, union vkd3d_thread_ha
 
 struct vkd3d_waiting_fence
 {
-    struct d3d12_fence *fence;
+    d3d12_fence_iface *fence;
+    VkSemaphore submission_timeline;
     uint64_t value;
+    LONG **submission_counters;
+    size_t num_submission_counts;
+    bool signal;
 };
 
 struct vkd3d_fence_worker
@@ -508,8 +516,6 @@ static inline HRESULT vkd3d_set_private_data_interface(struct vkd3d_private_stor
 HRESULT STDMETHODCALLTYPE d3d12_object_SetName(ID3D12Object *iface, const WCHAR *name);
 
 /* ID3D12Fence */
-typedef ID3D12Fence1 d3d12_fence_iface;
-
 struct d3d12_fence_value
 {
     uint64_t virtual_value;
@@ -579,6 +585,52 @@ HRESULT d3d12_fence_create(struct d3d12_device *device,
 HRESULT d3d12_fence_set_event_on_completion(struct d3d12_fence *fence,
         UINT64 value, HANDLE event, enum vkd3d_waiting_event_type type);
 
+struct d3d12_shared_fence
+{
+    d3d12_fence_iface ID3D12Fence_iface;
+    LONG refcount_internal;
+    LONG refcount;
+
+    D3D12_FENCE_FLAGS d3d12_flags;
+
+    VkSemaphore timeline_semaphore;
+
+    struct d3d12_device *device;
+
+    struct vkd3d_private_store private_store;
+};
+
+static inline struct d3d12_shared_fence *shared_impl_from_ID3D12Fence1(ID3D12Fence1 *iface)
+{
+    extern CONST_VTBL struct ID3D12Fence1Vtbl d3d12_shared_fence_vtbl;
+    if (!iface)
+        return NULL;
+    assert(iface->lpVtbl == &d3d12_shared_fence_vtbl);
+    return CONTAINING_RECORD(iface, struct d3d12_shared_fence, ID3D12Fence_iface);
+}
+
+static inline struct d3d12_shared_fence *shared_impl_from_ID3D12Fence(ID3D12Fence *iface)
+{
+    return shared_impl_from_ID3D12Fence1((ID3D12Fence1 *)iface);
+}
+
+HRESULT d3d12_shared_fence_create(struct d3d12_device *device,
+        uint64_t initial_value, D3D12_FENCE_FLAGS flags, struct d3d12_shared_fence **fence);
+
+static inline bool is_shared_ID3D12Fence1(ID3D12Fence1 *iface)
+{
+    extern CONST_VTBL struct ID3D12Fence1Vtbl d3d12_shared_fence_vtbl;
+    extern CONST_VTBL struct ID3D12Fence1Vtbl d3d12_fence_vtbl;
+    assert(iface->lpVtbl ==  &d3d12_shared_fence_vtbl || iface->lpVtbl == &d3d12_fence_vtbl);
+
+    return iface->lpVtbl ==  &d3d12_shared_fence_vtbl;
+}
+
+static inline bool is_shared_ID3D12Fence(ID3D12Fence *iface)
+{
+    return is_shared_ID3D12Fence1((ID3D12Fence1 *)iface);
+}
+
 enum vkd3d_allocation_flag
 {
     VKD3D_ALLOCATION_FLAG_GLOBAL_BUFFER     = (1u << 0),
@@ -904,7 +956,7 @@ VkImageSubresource vk_image_subresource_from_d3d12(
 
 HRESULT d3d12_resource_create_committed(struct d3d12_device *device, const D3D12_RESOURCE_DESC1 *desc,
         const D3D12_HEAP_PROPERTIES *heap_properties, D3D12_HEAP_FLAGS heap_flags, D3D12_RESOURCE_STATES initial_state,
-        const D3D12_CLEAR_VALUE *optimized_clear_value, struct d3d12_resource **resource);
+        const D3D12_CLEAR_VALUE *optimized_clear_value, HANDLE shared_handle, struct d3d12_resource **resource);
 HRESULT d3d12_resource_create_placed(struct d3d12_device *device, const D3D12_RESOURCE_DESC1 *desc,
         struct d3d12_heap *heap, uint64_t heap_offset, D3D12_RESOURCE_STATES initial_state,
         const D3D12_CLEAR_VALUE *optimized_clear_value, struct d3d12_resource **resource);
@@ -1403,6 +1455,10 @@ struct d3d12_root_signature
 
 HRESULT d3d12_root_signature_create(struct d3d12_device *device, const void *bytecode,
         size_t bytecode_length, struct d3d12_root_signature **root_signature);
+HRESULT d3d12_root_signature_create_raw(struct d3d12_device *device, const void *payload,
+        size_t payload_size, struct d3d12_root_signature **root_signature);
+HRESULT d3d12_root_signature_create_empty(struct d3d12_device *device,
+        struct d3d12_root_signature **root_signature);
 /* Private ref counts, for pipeline library. */
 void d3d12_root_signature_inc_ref(struct d3d12_root_signature *state);
 void d3d12_root_signature_dec_ref(struct d3d12_root_signature *state);
@@ -1425,9 +1481,14 @@ HRESULT vkd3d_create_pipeline_layout(struct d3d12_device *device,
         VkPipelineLayout *pipeline_layout);
 
 int vkd3d_parse_root_signature_v_1_0(const struct vkd3d_shader_code *dxbc,
-        struct vkd3d_versioned_root_signature_desc *desc);
+        struct vkd3d_versioned_root_signature_desc *desc,
+        vkd3d_shader_hash_t *compatibility_hash);
 int vkd3d_parse_root_signature_v_1_1(const struct vkd3d_shader_code *dxbc,
-        struct vkd3d_versioned_root_signature_desc *desc);
+        struct vkd3d_versioned_root_signature_desc *desc,
+        vkd3d_shader_hash_t *compatibility_hash);
+int vkd3d_parse_root_signature_v_1_1_from_raw_payload(const struct vkd3d_shader_code *dxbc,
+        struct vkd3d_versioned_root_signature_desc *desc,
+        vkd3d_shader_hash_t *compatibility_hash);
 
 VkShaderStageFlags vkd3d_vk_stage_flags_from_visibility(D3D12_SHADER_VISIBILITY visibility);
 enum vkd3d_shader_visibility vkd3d_shader_visibility_from_d3d12(D3D12_SHADER_VISIBILITY visibility);
@@ -1457,10 +1518,11 @@ struct vkd3d_shader_debug_ring_spec_constants
     uint32_t ring_words;
 };
 
+#define VKD3D_SHADER_DEBUG_RING_SPEC_INFO_MAP_ENTRIES 4
 struct vkd3d_shader_debug_ring_spec_info
 {
     struct vkd3d_shader_debug_ring_spec_constants constants;
-    VkSpecializationMapEntry map_entries[4];
+    VkSpecializationMapEntry map_entries[VKD3D_SHADER_DEBUG_RING_SPEC_INFO_MAP_ENTRIES];
     VkSpecializationInfo spec_info;
 };
 
@@ -2277,6 +2339,8 @@ struct vkd3d_queue
     VkCommandPool barrier_pool;
     VkCommandBuffer barrier_command_buffer;
     VkSemaphore serializing_binary_semaphore;
+    VkSemaphore submission_timeline;
+    uint64_t submission_timeline_count;
 
     uint32_t vk_family_index;
     VkQueueFlags vk_queue_flags;
@@ -2289,6 +2353,8 @@ struct vkd3d_queue
     size_t wait_values_size;
     VkPipelineStageFlags *wait_stages;
     size_t wait_stages_size;
+    d3d12_fence_iface **wait_fences;
+    size_t wait_fences_size;
     uint32_t wait_count;
 };
 
@@ -2297,7 +2363,7 @@ HRESULT vkd3d_queue_create(struct d3d12_device *device, uint32_t family_index, u
         const VkQueueFamilyProperties *properties, struct vkd3d_queue **queue);
 void vkd3d_queue_destroy(struct vkd3d_queue *queue, struct d3d12_device *device);
 void vkd3d_queue_release(struct vkd3d_queue *queue);
-void vkd3d_queue_add_wait(struct vkd3d_queue *queue, VkSemaphore semaphore, uint64_t value);
+void vkd3d_queue_add_wait(struct vkd3d_queue *queue, d3d12_fence_iface *waiter, VkSemaphore semaphore, uint64_t value);
 
 enum vkd3d_submission_type
 {
@@ -2333,13 +2399,13 @@ struct vkd3d_sparse_memory_bind_range
 
 struct d3d12_command_queue_submission_wait
 {
-    struct d3d12_fence *fence;
+    d3d12_fence_iface *fence;
     UINT64 value;
 };
 
 struct d3d12_command_queue_submission_signal
 {
-    struct d3d12_fence *fence;
+    d3d12_fence_iface *fence;
     UINT64 value;
 };
 
@@ -2432,6 +2498,29 @@ struct vkd3d_execute_indirect_info
     VkPipeline vk_pipeline;
 };
 
+enum vkd3d_patch_command_token
+{
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_CONST_U32 = 0,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_VA_LO = 1,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_VA_HI = 2,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_IBO_SIZE = 3,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_INDEX_FORMAT = 4,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_VA_LO = 5,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_VA_HI = 6,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_SIZE = 7,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_VBO_STRIDE = 8,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_ROOT_VA_LO = 9,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_ROOT_VA_HI = 10,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_VERTEX_COUNT = 11,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_INDEX_COUNT = 12,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_INSTANCE_COUNT = 13,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_FIRST_INDEX = 14,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_FIRST_VERTEX = 15,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_FIRST_INSTANCE = 16,
+    VKD3D_PATCH_COMMAND_TOKEN_COPY_VERTEX_OFFSET = 17,
+    VKD3D_PATCH_COMMAND_INT_MAX = 0x7fffffff
+};
+
 /* ID3D12CommandSignature */
 struct d3d12_command_signature
 {
@@ -2536,6 +2625,7 @@ enum vkd3d_breadcrumb_command_type
     VKD3D_BREADCRUMB_COMMAND_DRAW_INDEXED,
     VKD3D_BREADCRUMB_COMMAND_DISPATCH,
     VKD3D_BREADCRUMB_COMMAND_EXECUTE_INDIRECT,
+    VKD3D_BREADCRUMB_COMMAND_EXECUTE_INDIRECT_TEMPLATE,
     VKD3D_BREADCRUMB_COMMAND_COPY,
     VKD3D_BREADCRUMB_COMMAND_RESOLVE,
     VKD3D_BREADCRUMB_COMMAND_WBI,
@@ -2552,6 +2642,7 @@ enum vkd3d_breadcrumb_command_type
     VKD3D_BREADCRUMB_COMMAND_IBO,
     VKD3D_BREADCRUMB_COMMAND_ROOT_DESC,
     VKD3D_BREADCRUMB_COMMAND_ROOT_CONST,
+    VKD3D_BREADCRUMB_COMMAND_TAG,
 };
 
 #ifdef VKD3D_ENABLE_BREADCRUMBS
@@ -2575,6 +2666,8 @@ struct vkd3d_breadcrumb_command
         uint32_t word_32bit;
         uint64_t word_64bit;
         uint32_t count;
+        /* Pointer must remain alive. */
+        const char *tag;
     };
 };
 
@@ -3041,6 +3134,10 @@ struct vkd3d_execute_indirect_args
     VkDeviceAddress dst_indirect_count_va;
     uint32_t api_buffer_word_stride;
     uint32_t device_generated_commands_word_stride;
+
+    /* Arbitrary tag used for debug version of state patcher. Debug messages from tag 0 are ignored. */
+    uint32_t debug_tag;
+    uint32_t implicit_instance;
 };
 
 struct vkd3d_execute_indirect_pipeline
@@ -3501,6 +3598,14 @@ struct d3d12_state_object_identifier
 
     /* The index into vkGetShaderStackSize and friends for pGroups[]. */
     uint32_t group_index;
+
+    /* For AddToStateObject(). We need to return the identifier pointer
+     * for the parent, not the child. This makes it easy to validate that
+     * we observe the same SBT handles as specified by DXR 1.1. */
+
+    /* If -1, ignore, otherwise, redirect. */
+    int inherited_collection_index;
+    uint32_t inherited_collection_export_index;
 };
 
 struct d3d12_state_object_stack_info
@@ -3513,6 +3618,15 @@ struct d3d12_state_object_stack_info
     uint32_t max_closest;
 };
 
+#ifdef VKD3D_ENABLE_BREADCRUMBS
+struct d3d12_state_object_breadcrumb_shader
+{
+    vkd3d_shader_hash_t hash;
+    VkShaderStageFlagBits stage;
+    char name[64];
+};
+#endif
+
 struct d3d12_state_object
 {
     d3d12_state_object_iface ID3D12StateObject_iface;
@@ -3534,7 +3648,14 @@ struct d3d12_state_object
     /* Normally stages_count == entry_points_count, but entry_points is the entry points we
      * export externally, and stages_count matches pStages[] size for purposes of index fixups. */
 
+    /* Can be bound. */
     VkPipeline pipeline;
+    /* Can be used as a library. */
+    VkPipeline pipeline_library;
+
+    /* Can be inherited by AddToStateObject(). */
+    D3D12_RAYTRACING_PIPELINE_CONFIG1 pipeline_config;
+    D3D12_RAYTRACING_SHADER_CONFIG shader_config;
 
     struct
     {
@@ -3543,6 +3664,8 @@ struct d3d12_state_object
         VkDescriptorSet desc_set;
         VkDescriptorPool desc_pool;
         uint32_t set_index;
+        uint64_t compatibility_hash;
+        bool owned_handles;
     } local_static_sampler;
 
     UINT64 pipeline_stack_size;
@@ -3551,10 +3674,23 @@ struct d3d12_state_object
     struct d3d12_state_object **collections;
     size_t collections_count;
 
+    struct d3d12_root_signature *global_root_signature;
+
+#ifdef VKD3D_ENABLE_BREADCRUMBS
+    /* For breadcrumbs. */
+    struct d3d12_state_object_breadcrumb_shader *breadcrumb_shaders;
+    size_t breadcrumb_shaders_size;
+    size_t breadcrumb_shaders_count;
+#endif
+
     struct vkd3d_private_store private_store;
 };
 
 HRESULT d3d12_state_object_create(struct d3d12_device *device, const D3D12_STATE_OBJECT_DESC *desc,
+        struct d3d12_state_object *parent,
+        struct d3d12_state_object **object);
+HRESULT d3d12_state_object_add(struct d3d12_device *device, const D3D12_STATE_OBJECT_DESC *desc,
+        struct d3d12_state_object *parent,
         struct d3d12_state_object **object);
 
 static inline struct d3d12_state_object *impl_from_ID3D12StateObject(ID3D12StateObject *iface)
@@ -3826,6 +3962,74 @@ void vkd3d_acceleration_structure_copy(
         D3D12_GPU_VIRTUAL_ADDRESS dst, D3D12_GPU_VIRTUAL_ADDRESS src,
         D3D12_RAYTRACING_ACCELERATION_STRUCTURE_COPY_MODE mode);
 
+typedef enum D3D11_USAGE
+{
+    D3D11_USAGE_DEFAULT,
+    D3D11_USAGE_IMMUTABLE,
+    D3D11_USAGE_DYNAMIC,
+    D3D11_USAGE_STAGING,
+} D3D11_USAGE;
+
+typedef enum D3D11_BIND_FLAG
+{
+    D3D11_BIND_VERTEX_BUFFER    = 0x0001,
+    D3D11_BIND_INDEX_BUFFER     = 0x0002,
+    D3D11_BIND_CONSTANT_BUFFER  = 0x0004,
+    D3D11_BIND_SHADER_RESOURCE  = 0x0008,
+    D3D11_BIND_STREAM_OUTPUT    = 0x0010,
+    D3D11_BIND_RENDER_TARGET    = 0x0020,
+    D3D11_BIND_DEPTH_STENCIL    = 0x0040,
+    D3D11_BIND_UNORDERED_ACCESS = 0x0080,
+    D3D11_BIND_DECODER          = 0x0200,
+    D3D11_BIND_VIDEO_ENCODER    = 0x0400
+} D3D11_BIND_FLAG;
+
+typedef enum D3D11_TEXTURE_LAYOUT
+{
+    D3D11_TEXTURE_LAYOUT_UNDEFINED = 0x0,
+    D3D11_TEXTURE_LAYOUT_ROW_MAJOR = 0x1,
+    D3D11_TEXTURE_LAYOUT_64K_STANDARD_SWIZZLE = 0x2,
+} D3D11_TEXTURE_LAYOUT;
+
+typedef enum D3D11_RESOURCE_MISC_FLAG
+{
+    D3D11_RESOURCE_MISC_GENERATE_MIPS                    = 0x1,
+    D3D11_RESOURCE_MISC_SHARED                           = 0x2,
+    D3D11_RESOURCE_MISC_TEXTURECUBE                      = 0x4,
+    D3D11_RESOURCE_MISC_DRAWINDIRECT_ARGS                = 0x10,
+    D3D11_RESOURCE_MISC_BUFFER_ALLOW_RAW_VIEWS           = 0x20,
+    D3D11_RESOURCE_MISC_BUFFER_STRUCTURED                = 0x40,
+    D3D11_RESOURCE_MISC_RESOURCE_CLAMP                   = 0x80,
+    D3D11_RESOURCE_MISC_SHARED_KEYEDMUTEX                = 0x100,
+    D3D11_RESOURCE_MISC_GDI_COMPATIBLE                   = 0x200,
+    D3D11_RESOURCE_MISC_SHARED_NTHANDLE                  = 0x800,
+    D3D11_RESOURCE_MISC_RESTRICTED_CONTENT               = 0x1000,
+    D3D11_RESOURCE_MISC_RESTRICT_SHARED_RESOURCE         = 0x2000,
+    D3D11_RESOURCE_MISC_RESTRICT_SHARED_RESOURCE_DRIVER  = 0x4000,
+    D3D11_RESOURCE_MISC_GUARDED                          = 0x8000,
+    D3D11_RESOURCE_MISC_TILE_POOL                        = 0x20000,
+    D3D11_RESOURCE_MISC_TILED                            = 0x40000,
+    D3D11_RESOURCE_MISC_HW_PROTECTED                     = 0x80000,
+} D3D11_RESOURCE_MISC_FLAG;
+
+struct DxvkSharedTextureMetadata {
+    UINT             Width;
+    UINT             Height;
+    UINT             MipLevels;
+    UINT             ArraySize;
+    DXGI_FORMAT      Format;
+    DXGI_SAMPLE_DESC SampleDesc;
+    D3D11_USAGE      Usage;
+    UINT             BindFlags;
+    UINT             CPUAccessFlags;
+    UINT             MiscFlags;
+    D3D11_TEXTURE_LAYOUT TextureLayout;
+};
+
+bool vkd3d_set_shared_metadata(HANDLE handle, void *buf, uint32_t buf_size);
+bool vkd3d_get_shared_metadata(HANDLE handle, void *buf, uint32_t buf_size, uint32_t *metadata_size);
+HANDLE vkd3d_open_kmt_handle(HANDLE kmt_handle);
+
 #define VKD3D_VENDOR_ID_NVIDIA 0x10DE
 #define VKD3D_VENDOR_ID_AMD 0x1002
 #define VKD3D_VENDOR_ID_INTEL 0x8086

libs/vkd3d/vkd3d_shaders.h

@@ -46,6 +46,7 @@ enum vkd3d_meta_copy_mode
 #include <cs_resolve_predicate.h>
 #include <cs_resolve_query.h>
 #include <cs_execute_indirect_patch.h>
+#include <cs_execute_indirect_patch_debug_ring.h>
 #include <vs_fullscreen_layer.h>
 #include <vs_fullscreen.h>
 #include <gs_fullscreen.h>

libs/vkd3d/vulkan_procs.h

@@ -49,6 +49,7 @@ VK_INSTANCE_PFN(vkGetPhysicalDeviceQueueFamilyProperties)
 VK_INSTANCE_PFN(vkGetPhysicalDeviceSparseImageFormatProperties)
 VK_INSTANCE_PFN(vkGetPhysicalDeviceFeatures2)
 VK_INSTANCE_PFN(vkGetPhysicalDeviceProperties2)
+VK_INSTANCE_PFN(vkGetPhysicalDeviceExternalSemaphoreProperties)
 
 /* VK_EXT_debug_utils */
 VK_INSTANCE_EXT_PFN(vkCreateDebugUtilsMessengerEXT)
@@ -223,6 +224,18 @@ VK_DEVICE_EXT_PFN(vkGetDeviceBufferMemoryRequirementsKHR)
 VK_DEVICE_EXT_PFN(vkGetDeviceImageMemoryRequirementsKHR)
 VK_DEVICE_EXT_PFN(vkGetDeviceImageSparseMemoryRequirementsKHR)
 
+#ifdef VK_KHR_external_memory_win32
+/* VK_KHR_external_memory_win32 */
+VK_DEVICE_EXT_PFN(vkGetMemoryWin32HandleKHR)
+VK_DEVICE_EXT_PFN(vkGetMemoryWin32HandlePropertiesKHR)
+#endif
+
+#ifdef VK_KHR_external_semaphore_win32
+/* VK_KHR_external_semaphore_win32 */
+VK_DEVICE_EXT_PFN(vkGetSemaphoreWin32HandleKHR)
+VK_DEVICE_EXT_PFN(vkImportSemaphoreWin32HandleKHR)
+#endif
+
 /* VK_EXT_calibrated_timestamps */
 VK_DEVICE_EXT_PFN(vkGetCalibratedTimestampsEXT)
 VK_INSTANCE_EXT_PFN(vkGetPhysicalDeviceCalibrateableTimeDomainsEXT)

meson.build

@@ -83,7 +83,7 @@ idl_generator = generator(idl_compiler,
   arguments : [ '-h', '-o', '@OUTPUT@', '@INPUT@' ])
 
 glsl_compiler  = find_program('glslangValidator')
-glsl_args = [ '-V', '--vn', '@BASENAME@', '@INPUT@', '-o', '@OUTPUT@' ]
+glsl_args = [ '-V', '--target-env', 'vulkan1.1', '--vn', '@BASENAME@', '@INPUT@', '-o', '@OUTPUT@' ]
 if run_command(glsl_compiler, [ '--quiet', '--version' ], check : false).returncode() == 0
     glsl_args += [ '--quiet' ]
 endif

programs/vkd3d-profile.py

@@ -75,17 +75,35 @@ def main():
     parser.add_argument('--per-iteration', action = 'store_true', help = 'Represent ticks in terms of ticks / iteration. Cannot be used with --divider.')
     parser.add_argument('--name', nargs = '+', type = str, help = 'Only display data for certain counters.')
     parser.add_argument('--sort', type = str, default = 'none', help = 'Sorts input data according to "iterations" or "ticks".')
+    parser.add_argument('--delta', type = str, help = 'Subtract iterations and timing from other profile blob.')
     parser.add_argument('profile', help = 'The profile binary blob.')
 
     args = parser.parse_args()
     if not args.profile:
         raise AssertionError('Need profile folder.')
 
+    delta_map = {}
+    if args.delta is not None:
+        with open(args.delta, 'rb') as f:
+            for block in iter(lambda: f.read(64), b''):
+                if is_valid_block(block):
+                    b = parse_block(block)
+                    delta_map[b.name] = b
+
     blocks = []
     with open(args.profile, 'rb') as f:
         for block in iter(lambda: f.read(64), b''):
             if is_valid_block(block):
-                blocks.append(parse_block(block))
+                b = parse_block(block)
+                if b.name in delta_map:
+                    d = delta_map[b.name]
+                    b = ProfileCase(ticks = b.ticks - d.ticks,
+                            iterations = b.iterations - d.iterations,
+                            name = b.name)
+                    if b.iterations < 0 or b.ticks < 0:
+                        raise AssertionError('After subtracting, iterations or ticks became negative.')
+                if b.iterations > 0:
+                    blocks.append(b)
 
     if args.divider is not None:
         if args.per_iteration:
@@ -114,11 +132,11 @@ def main():
                 print('    Iterations:', block.iterations)
 
             if args.divider is not None:
-                print('    Time spent per iteration of {}: {:.3f}'.format(args.divider, block.ticks / 1000.0), "us")
+                print('    Time spent per iteration of {}: {:.3f}'.format(args.divider, block.ticks / 1000.0), "Kcycles")
             elif args.per_iteration:
-                print('    Time spent per iteration: {:.3f}'.format(block.ticks / 1000.0), "us")
+                print('    Time spent per iteration: {:.3f}'.format(block.ticks / 1000.0), "Kcycles")
             else:
-                print('    Total time spent: {:.3f}'.format(block.ticks / 1000.0), "us")
+                print('    Total time spent: {:.3f}'.format(block.ticks / 1000.0), "Kcycles")
 
 if __name__ == '__main__':
     main()

subprojects/dxil-spirv

@@ -1 +1 @@
-Subproject commit b1afbf5fa8e6f10f6c226cea222a8a5d5518870f
+Subproject commit 9f2fd6356c14376ab5b88518d6dd4e6787084525

tests/d3d12_command.c

@@ -1449,6 +1449,721 @@ void test_vbv_stride_edge_cases(void)
     destroy_test_context(&context);
 }
 
+void test_execute_indirect_state(void)
+{
+    static const struct vec4 values = { 1000.0f, 2000.0f, 3000.0f, 4000.0f };
+    D3D12_INDIRECT_ARGUMENT_DESC indirect_argument_descs[2];
+    D3D12_COMMAND_SIGNATURE_DESC command_signature_desc;
+    D3D12_ROOT_SIGNATURE_DESC root_signature_desc;
+    D3D12_GRAPHICS_PIPELINE_STATE_DESC pso_desc;
+    ID3D12CommandSignature *command_signature;
+    D3D12_SO_DECLARATION_ENTRY so_entries[1];
+    ID3D12GraphicsCommandList *command_list;
+    D3D12_ROOT_PARAMETER root_parameters[4];
+    ID3D12RootSignature *root_signatures[2];
+    ID3D12Resource *argument_buffer_late;
+    D3D12_STREAM_OUTPUT_BUFFER_VIEW sov;
+    ID3D12Resource *streamout_buffer;
+    D3D12_VERTEX_BUFFER_VIEW vbvs[2];
+    ID3D12Resource *argument_buffer;
+    struct test_context_desc desc;
+    ID3D12PipelineState *psos[2];
+    struct test_context context;
+    struct resource_readback rb;
+    D3D12_INDEX_BUFFER_VIEW ibv;
+    ID3D12CommandQueue *queue;
+    const UINT so_stride = 16;
+    ID3D12Resource *vbo[3];
+    ID3D12Resource *ibo[2];
+    unsigned int i, j, k;
+    ID3D12Resource *cbv;
+    ID3D12Resource *srv;
+    ID3D12Resource *uav;
+    HRESULT hr;
+
+    static const D3D12_INPUT_ELEMENT_DESC layout_desc[] =
+    {
+        {"COLOR", 0, DXGI_FORMAT_R32_FLOAT, 0, 0, D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA, 0},
+        {"COLOR", 1, DXGI_FORMAT_R32_FLOAT, 1, 0, D3D12_INPUT_CLASSIFICATION_PER_VERTEX_DATA, 0},
+    };
+
+    struct test
+    {
+        const D3D12_INDIRECT_ARGUMENT_DESC *indirect_arguments;
+        uint32_t indirect_argument_count;
+        const void *argument_buffer_data;
+        size_t argument_buffer_size;
+        uint32_t api_max_count;
+        const struct vec4 *expected_output;
+        uint32_t expected_output_count;
+        uint32_t stride;
+        uint32_t pso_index;
+        bool needs_root_sig;
+    };
+
+    /* Modify root parameters. */
+    struct root_constant_data
+    {
+        float constants[2];
+        D3D12_DRAW_INDEXED_ARGUMENTS indexed;
+    };
+
+    static const D3D12_INDIRECT_ARGUMENT_DESC root_constant_sig[2] =
+    {
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_CONSTANT, .Constant = {
+            .RootParameterIndex = 0, .DestOffsetIn32BitValues = 1, .Num32BitValuesToSet = 2 }},
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED }
+    };
+
+    static const struct root_constant_data root_constant_data[] =
+    {
+        {
+            .constants = { 100.0f, 500.0f },
+            .indexed = { .IndexCountPerInstance = 2, .InstanceCount = 1 }
+        },
+        {
+            .constants = { 200.0f, 800.0f },
+            .indexed = { .IndexCountPerInstance = 1, .InstanceCount = 2,
+                         .StartIndexLocation = 1, .StartInstanceLocation = 100, }
+        },
+    };
+
+    static const struct vec4 root_constant_expected[] =
+    {
+        { 1000.0f, 64.0f + 100.0f, 500.0f, 4000.0f },
+        { 1001.0f, 65.0f + 100.0f, 500.0f, 4000.0f },
+        { 1001.0f, 65.0f + 200.0f, 800.0f, 4000.0f },
+        { 1001.0f, 65.0f + 200.0f, 800.0f, 4001.0f },
+    };
+
+    /* Modify root parameters, but very large root signature to test boundary conditions. */
+    static const D3D12_INDIRECT_ARGUMENT_DESC root_constant_spill_sig[2] =
+    {
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_CONSTANT, .Constant = {
+                .RootParameterIndex = 0, .DestOffsetIn32BitValues = 44 + 1, .Num32BitValuesToSet = 2 }},
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED }
+    };
+
+    static const struct root_constant_data root_constant_spill_data[] =
+    {
+        {
+            .constants = { 100.0f, 500.0f },
+            .indexed = { .IndexCountPerInstance = 2, .InstanceCount = 1 }
+        },
+        {
+            .constants = { 200.0f, 800.0f },
+            .indexed = { .IndexCountPerInstance = 1, .InstanceCount = 2,
+                    .StartIndexLocation = 1, .StartInstanceLocation = 100, }
+        },
+    };
+
+    static const struct vec4 root_constant_spill_expected[] =
+    {
+        { 1000.0f, 64.0f + 100.0f, 500.0f, 4000.0f },
+        { 1001.0f, 65.0f + 100.0f, 500.0f, 4000.0f },
+        { 1001.0f, 65.0f + 200.0f, 800.0f, 4000.0f },
+        { 1001.0f, 65.0f + 200.0f, 800.0f, 4001.0f },
+    };
+
+    /* Modify VBOs. */
+    struct indirect_vbo_data
+    {
+        D3D12_VERTEX_BUFFER_VIEW view[2];
+        D3D12_DRAW_INDEXED_ARGUMENTS indexed;
+    };
+
+    static const D3D12_INDIRECT_ARGUMENT_DESC indirect_vbo_sig[3] =
+    {
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_VERTEX_BUFFER_VIEW, .VertexBuffer = { .Slot = 0 }},
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_VERTEX_BUFFER_VIEW, .VertexBuffer = { .Slot = 1 }},
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED },
+    };
+
+    /* Fill buffer locations later. */
+    struct indirect_vbo_data indirect_vbo_data[] =
+    {
+        {
+            .view = { { 0, 64, 8 }, { 0, 64, 16 } },
+            .indexed = { .IndexCountPerInstance = 2, .InstanceCount = 2 }
+        },
+        {
+            /* Test indirectly binding NULL descriptor and 0 stride. */
+            .view = { { 0, 0, 0 }, { 0, 64, 0 } },
+            .indexed = { .IndexCountPerInstance = 2, .InstanceCount = 1 }
+        }
+    };
+
+    static const struct vec4 indirect_vbo_expected[] =
+    {
+        { 1064.0f, 2128.0f, 3000.0f, 4000.0f },
+        { 1066.0f, 2132.0f, 3000.0f, 4000.0f },
+        { 1064.0f, 2128.0f, 3000.0f, 4001.0f },
+        { 1066.0f, 2132.0f, 3000.0f, 4001.0f },
+        { 1000.0f, 2016.0f, 3000.0f, 4000.0f }, /* This is buggy on WARP and AMD. We seem to get null descriptor instead. */
+        { 1000.0f, 2016.0f, 3000.0f, 4000.0f }, /* This is buggy on WARP and AMD. */
+    };
+
+    /* Modify just one VBO. */
+    struct indirect_vbo_one_data
+    {
+        D3D12_VERTEX_BUFFER_VIEW view;
+        D3D12_DRAW_INDEXED_ARGUMENTS indexed;
+    };
+
+    static const D3D12_INDIRECT_ARGUMENT_DESC indirect_vbo_one_sig[2] =
+    {
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_VERTEX_BUFFER_VIEW, .VertexBuffer = { .Slot = 0 }},
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED },
+    };
+
+    /* Fill buffer locations later. */
+    struct indirect_vbo_one_data indirect_vbo_one_data[] =
+    {
+        {
+            .view = { 0, 64, 8 },
+            .indexed = { .IndexCountPerInstance = 2, .InstanceCount = 1 }
+        },
+        {
+            .indexed = { .IndexCountPerInstance = 1, .InstanceCount = 1 }
+        }
+    };
+
+    static const struct vec4 indirect_vbo_one_expected[] =
+    {
+        { 1128.0f, 2064.0f, 3000.0f, 4000.0f },
+        { 1130.0f, 2065.0f, 3000.0f, 4000.0f },
+        { 1000.0f, 2064.0f, 3000.0f, 4000.0f },
+    };
+
+    /* Indirect IBO */
+    struct indirect_ibo_data
+    {
+        D3D12_INDEX_BUFFER_VIEW view;
+        D3D12_DRAW_INDEXED_ARGUMENTS indexed;
+    };
+
+    static const D3D12_INDIRECT_ARGUMENT_DESC indirect_ibo_sig[2] =
+    {
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_INDEX_BUFFER_VIEW },
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED },
+    };
+
+    struct indirect_ibo_data indirect_ibo_data[] =
+    {
+        {
+            .view = { 0, 0, DXGI_FORMAT_R32_UINT },
+            .indexed = { .IndexCountPerInstance = 2, .InstanceCount = 1 }
+        },
+        {
+            .view = { 0, 64, DXGI_FORMAT_R16_UINT },
+            .indexed = { .IndexCountPerInstance = 4, .InstanceCount = 1 }
+        },
+    };
+
+    static const struct vec4 indirect_ibo_expected[] =
+    {
+        { 1000.0f, 2064.0f, 3000.0f, 4000.0f },
+        { 1000.0f, 2064.0f, 3000.0f, 4000.0f },
+        { 1016.0f, 2080.0f, 3000.0f, 4000.0f },
+        { 1000.0f, 2064.0f, 3000.0f, 4000.0f },
+        { 1017.0f, 2081.0f, 3000.0f, 4000.0f },
+        { 1000.0f, 2064.0f, 3000.0f, 4000.0f },
+    };
+
+    /* Indirect root arguments */
+    struct indirect_root_descriptor_data
+    {
+        D3D12_GPU_VIRTUAL_ADDRESS cbv;
+        D3D12_GPU_VIRTUAL_ADDRESS srv;
+        D3D12_GPU_VIRTUAL_ADDRESS uav;
+        D3D12_DRAW_ARGUMENTS array;
+    };
+
+    static const D3D12_INDIRECT_ARGUMENT_DESC indirect_root_descriptor_sig[4] =
+    {
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_CONSTANT_BUFFER_VIEW, .ConstantBufferView = { .RootParameterIndex = 1 } },
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_SHADER_RESOURCE_VIEW, .ShaderResourceView = { .RootParameterIndex = 2 } },
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_UNORDERED_ACCESS_VIEW, .UnorderedAccessView = { .RootParameterIndex = 3 } },
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW },
+    };
+
+    struct indirect_root_descriptor_data indirect_root_descriptor_data[] =
+    {
+        { .array = { .VertexCountPerInstance = 1, .InstanceCount = 1 } },
+        { .array = { .VertexCountPerInstance = 1, .InstanceCount = 1 } },
+    };
+
+    static const struct vec4 indirect_root_descriptor_expected[] =
+    {
+        { 1000.0f, 2064.0f, 3000.0f + 64.0f, 4000.0f + 2.0f },
+        { 1000.0f, 2064.0f, 3000.0f + 128.0f, 4000.0f + 3.0f },
+    };
+
+    /* Test packing rules.
+     * 64-bit aligned values are tightly packed with 32-bit alignment when they are in indirect command buffers. */
+    struct indirect_alignment_data
+    {
+        float value;
+        uint32_t cbv_va[2];
+        D3D12_DRAW_ARGUMENTS arrays;
+    };
+    static const D3D12_INDIRECT_ARGUMENT_DESC indirect_alignment_sig[3] =
+    {
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_CONSTANT, .Constant = {
+                .RootParameterIndex = 0, .DestOffsetIn32BitValues = 1, .Num32BitValuesToSet = 1 }},
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_CONSTANT_BUFFER_VIEW, .ConstantBufferView = { .RootParameterIndex = 1 }},
+        { .Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW },
+    };
+
+    struct indirect_alignment_data indirect_alignment_data[] =
+    {
+        {
+            .value = 5.0f,
+            .arrays = { .VertexCountPerInstance = 1, .InstanceCount = 1 }
+        },
+        {
+            .value = 6.0f,
+            .arrays = { .VertexCountPerInstance = 1, .InstanceCount = 1 }
+        },
+    };
+
+    static const struct vec4 indirect_alignment_expected[] =
+    {
+        { 1000.0f, 69.0f, 3064.0f, 4000.0f },
+        { 1000.0f, 70.0f, 3128.0f, 4000.0f },
+    };
+
+#define DECL_TEST(t, pso_index, needs_root_sig) { t##_sig, ARRAY_SIZE(t##_sig), t##_data, sizeof(t##_data), ARRAY_SIZE(t##_data), \
+        t##_expected, ARRAY_SIZE(t##_expected), sizeof(*(t##_data)), pso_index, needs_root_sig }
+    const struct test tests[] =
+    {
+        DECL_TEST(root_constant, 0, true),
+        DECL_TEST(indirect_vbo, 0, false),
+        DECL_TEST(indirect_vbo_one, 0, false),
+        DECL_TEST(indirect_ibo, 0, false),
+        DECL_TEST(indirect_root_descriptor, 0, true),
+        DECL_TEST(indirect_alignment, 0, true),
+        DECL_TEST(root_constant_spill, 1, true),
+        DECL_TEST(indirect_root_descriptor, 1, true),
+    };
+#undef DECL_TEST
+
+    uint32_t ibo_data[ARRAY_SIZE(ibo)][64];
+    float vbo_data[ARRAY_SIZE(vbo)][64];
+    float generic_data[4096];
+
+    static const DWORD vs_code_small_cbv[] =
+    {
+#if 0
+    cbuffer RootCBV : register(b0)
+    {
+            float a;
+    };
+
+    StructuredBuffer<float> RootSRV : register(t0);
+
+    cbuffer RootConstants : register(b0, space1)
+    {
+            float4 root;
+    };
+
+    float4 main(float c0 : COLOR0, float c1 : COLOR1, uint iid : SV_InstanceID) : SV_Position
+    {
+            return float4(c0, c1, a, RootSRV[0] + float(iid)) + root;
+    }
+#endif
+        0x43425844, 0x33b7b302, 0x34259b9b, 0x3e8568d9, 0x5a5e0c3e, 0x00000001, 0x00000268, 0x00000003,
+        0x0000002c, 0x00000098, 0x000000cc, 0x4e475349, 0x00000064, 0x00000003, 0x00000008, 0x00000050,
+        0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000101, 0x00000050, 0x00000001, 0x00000000,
+        0x00000003, 0x00000001, 0x00000101, 0x00000056, 0x00000000, 0x00000008, 0x00000001, 0x00000002,
+        0x00000101, 0x4f4c4f43, 0x56530052, 0x736e495f, 0x636e6174, 0x00444965, 0x4e47534f, 0x0000002c,
+        0x00000001, 0x00000008, 0x00000020, 0x00000000, 0x00000001, 0x00000003, 0x00000000, 0x0000000f,
+        0x505f5653, 0x7469736f, 0x006e6f69, 0x58454853, 0x00000194, 0x00010051, 0x00000065, 0x0100086a,
+        0x07000059, 0x00308e46, 0x00000000, 0x00000000, 0x00000000, 0x00000001, 0x00000000, 0x07000059,
+        0x00308e46, 0x00000001, 0x00000000, 0x00000000, 0x00000001, 0x00000001, 0x070000a2, 0x00307e46,
+        0x00000000, 0x00000000, 0x00000000, 0x00000004, 0x00000000, 0x0300005f, 0x00101012, 0x00000000,
+        0x0300005f, 0x00101012, 0x00000001, 0x04000060, 0x00101012, 0x00000002, 0x00000008, 0x04000067,
+        0x001020f2, 0x00000000, 0x00000001, 0x02000068, 0x00000001, 0x0a0000a7, 0x00100012, 0x00000000,
+        0x00004001, 0x00000000, 0x00004001, 0x00000000, 0x00207006, 0x00000000, 0x00000000, 0x05000056,
+        0x00100022, 0x00000000, 0x0010100a, 0x00000002, 0x07000000, 0x00100012, 0x00000000, 0x0010001a,
+        0x00000000, 0x0010000a, 0x00000000, 0x09000000, 0x00102012, 0x00000000, 0x0010100a, 0x00000000,
+        0x0030800a, 0x00000001, 0x00000000, 0x00000000, 0x09000000, 0x00102022, 0x00000000, 0x0010100a,
+        0x00000001, 0x0030801a, 0x00000001, 0x00000000, 0x00000000, 0x0b000000, 0x00102042, 0x00000000,
+        0x0030800a, 0x00000000, 0x00000000, 0x00000000, 0x0030802a, 0x00000001, 0x00000000, 0x00000000,
+        0x09000000, 0x00102082, 0x00000000, 0x0010000a, 0x00000000, 0x0030803a, 0x00000001, 0x00000000,
+        0x00000000, 0x0100003e,
+    };
+
+    static const DWORD vs_code_large_cbv[] =
+    {
+#if 0
+    cbuffer RootCBV : register(b0)
+    {
+        float a;
+    };
+
+    StructuredBuffer<float> RootSRV : register(t0);
+
+    cbuffer RootConstants : register(b0, space1)
+    {
+        // Cannot use arrays for root constants in D3D12.
+        float4 pad0, pad1, pad2, pad3, pad4, pad5, pad6, pad7, pad8, pad9, pad10;
+        float4 root;
+    };
+
+    float4 main(float c0 : COLOR0, float c1 : COLOR1, uint iid : SV_InstanceID) : SV_Position
+    {
+        return float4(c0, c1, a, RootSRV[0] + float(iid)) + root;
+    }
+#endif
+        0x43425844, 0x99a057e8, 0x20344569, 0x434f8a7a, 0xf9171e08, 0x00000001, 0x00000268, 0x00000003,
+        0x0000002c, 0x00000098, 0x000000cc, 0x4e475349, 0x00000064, 0x00000003, 0x00000008, 0x00000050,
+        0x00000000, 0x00000000, 0x00000003, 0x00000000, 0x00000101, 0x00000050, 0x00000001, 0x00000000,
+        0x00000003, 0x00000001, 0x00000101, 0x00000056, 0x00000000, 0x00000008, 0x00000001, 0x00000002,
+        0x00000101, 0x4f4c4f43, 0x56530052, 0x736e495f, 0x636e6174, 0x00444965, 0x4e47534f, 0x0000002c,
+        0x00000001, 0x00000008, 0x00000020, 0x00000000, 0x00000001, 0x00000003, 0x00000000, 0x0000000f,
+        0x505f5653, 0x7469736f, 0x006e6f69, 0x58454853, 0x00000194, 0x00010051, 0x00000065, 0x0100086a,
+        0x07000059, 0x00308e46, 0x00000000, 0x00000000, 0x00000000, 0x00000001, 0x00000000, 0x07000059,
+        0x00308e46, 0x00000001, 0x00000000, 0x00000000, 0x0000000c, 0x00000001, 0x070000a2, 0x00307e46,
+        0x00000000, 0x00000000, 0x00000000, 0x00000004, 0x00000000, 0x0300005f, 0x00101012, 0x00000000,
+        0x0300005f, 0x00101012, 0x00000001, 0x04000060, 0x00101012, 0x00000002, 0x00000008, 0x04000067,
+        0x001020f2, 0x00000000, 0x00000001, 0x02000068, 0x00000001, 0x0a0000a7, 0x00100012, 0x00000000,
+        0x00004001, 0x00000000, 0x00004001, 0x00000000, 0x00207006, 0x00000000, 0x00000000, 0x05000056,
+        0x00100022, 0x00000000, 0x0010100a, 0x00000002, 0x07000000, 0x00100012, 0x00000000, 0x0010001a,
+        0x00000000, 0x0010000a, 0x00000000, 0x09000000, 0x00102012, 0x00000000, 0x0010100a, 0x00000000,
+        0x0030800a, 0x00000001, 0x00000000, 0x0000000b, 0x09000000, 0x00102022, 0x00000000, 0x0010100a,
+        0x00000001, 0x0030801a, 0x00000001, 0x00000000, 0x0000000b, 0x0b000000, 0x00102042, 0x00000000,
+        0x0030800a, 0x00000000, 0x00000000, 0x00000000, 0x0030802a, 0x00000001, 0x00000000, 0x0000000b,
+        0x09000000, 0x00102082, 0x00000000, 0x0010000a, 0x00000000, 0x0030803a, 0x00000001, 0x00000000,
+        0x0000000b, 0x0100003e,
+    };
+
+    memset(&desc, 0, sizeof(desc));
+    desc.no_root_signature = true;
+    desc.no_pipeline = true;
+    if (!init_test_context(&context, &desc))
+        return;
+    command_list = context.list;
+    queue = context.queue;
+
+    for (j = 0; j < ARRAY_SIZE(ibo); j++)
+        for (i = 0; i < ARRAY_SIZE(ibo_data[j]); i++)
+            ibo_data[j][i] = j * 16 + i;
+
+    for (j = 0; j < ARRAY_SIZE(vbo); j++)
+        for (i = 0; i < ARRAY_SIZE(vbo_data[j]); i++)
+            vbo_data[j][i] = (float)(j * ARRAY_SIZE(vbo_data[j]) + i);
+
+    for (i = 0; i < ARRAY_SIZE(generic_data); i++)
+        generic_data[i] = (float)i;
+
+    for (i = 0; i < ARRAY_SIZE(ibo); i++)
+        ibo[i] = create_upload_buffer(context.device, sizeof(ibo_data[i]), ibo_data[i]);
+    for (i = 0; i < ARRAY_SIZE(vbo); i++)
+        vbo[i] = create_upload_buffer(context.device, sizeof(vbo_data[i]), vbo_data[i]);
+    cbv = create_upload_buffer(context.device, sizeof(generic_data), generic_data);
+    srv = create_upload_buffer(context.device, sizeof(generic_data), generic_data);
+    uav = create_default_buffer(context.device, sizeof(generic_data),
+            D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS,
+            D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
+
+    indirect_vbo_data[0].view[0].BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vbo[1]);
+    indirect_vbo_data[0].view[1].BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vbo[2]);
+    indirect_vbo_data[1].view[0].BufferLocation = 0;
+    indirect_vbo_data[1].view[1].BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vbo[0]) + 64;
+
+    indirect_vbo_one_data[0].view.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vbo[2]);
+    indirect_vbo_one_data[1].view.BufferLocation = 0;
+
+    indirect_ibo_data[1].view.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(ibo[1]);
+
+    indirect_root_descriptor_data[0].cbv = ID3D12Resource_GetGPUVirtualAddress(cbv) + 256;
+    indirect_root_descriptor_data[0].srv = ID3D12Resource_GetGPUVirtualAddress(srv) + 8;
+    indirect_root_descriptor_data[0].uav = ID3D12Resource_GetGPUVirtualAddress(uav) + 4;
+    indirect_root_descriptor_data[1].cbv = ID3D12Resource_GetGPUVirtualAddress(cbv) + 512;
+    indirect_root_descriptor_data[1].srv = ID3D12Resource_GetGPUVirtualAddress(srv) + 12;
+    indirect_root_descriptor_data[1].uav = ID3D12Resource_GetGPUVirtualAddress(uav) + 8;
+
+    memcpy(indirect_alignment_data[0].cbv_va, &indirect_root_descriptor_data[0].cbv, sizeof(D3D12_GPU_VIRTUAL_ADDRESS));
+    memcpy(indirect_alignment_data[1].cbv_va, &indirect_root_descriptor_data[1].cbv, sizeof(D3D12_GPU_VIRTUAL_ADDRESS));
+
+    memset(&root_signature_desc, 0, sizeof(root_signature_desc));
+    root_signature_desc.Flags = D3D12_ROOT_SIGNATURE_FLAG_ALLOW_INPUT_ASSEMBLER_INPUT_LAYOUT |
+            D3D12_ROOT_SIGNATURE_FLAG_ALLOW_STREAM_OUTPUT;
+
+    memset(root_parameters, 0, sizeof(root_parameters));
+    root_signature_desc.pParameters = root_parameters;
+    root_signature_desc.NumParameters = ARRAY_SIZE(root_parameters);
+    root_parameters[0].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
+    root_parameters[0].ParameterType = D3D12_ROOT_PARAMETER_TYPE_32BIT_CONSTANTS;
+    root_parameters[0].Constants.RegisterSpace = 1;
+    root_parameters[0].Constants.Num32BitValues = 4;
+    root_parameters[1].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
+    root_parameters[1].ParameterType = D3D12_ROOT_PARAMETER_TYPE_CBV;
+    root_parameters[2].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
+    root_parameters[2].ParameterType = D3D12_ROOT_PARAMETER_TYPE_SRV;
+    root_parameters[3].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
+    root_parameters[3].ParameterType = D3D12_ROOT_PARAMETER_TYPE_UAV;
+    hr = create_root_signature(context.device, &root_signature_desc, &root_signatures[0]);
+    ok(SUCCEEDED(hr), "Failed to create root signature, hr #%x.\n", hr);
+    root_parameters[0].Constants.Num32BitValues = 48;
+    hr = create_root_signature(context.device, &root_signature_desc, &root_signatures[1]);
+    ok(SUCCEEDED(hr), "Failed to create root signature, hr #%x.\n", hr);
+
+    memset(so_entries, 0, sizeof(so_entries));
+    so_entries[0].ComponentCount = 4;
+    so_entries[0].SemanticName = "SV_Position";
+
+    memset(&pso_desc, 0, sizeof(pso_desc));
+    pso_desc.VS.pShaderBytecode = vs_code_small_cbv;
+    pso_desc.VS.BytecodeLength = sizeof(vs_code_small_cbv);
+    pso_desc.StreamOutput.NumStrides = 1;
+    pso_desc.StreamOutput.pBufferStrides = &so_stride;
+    pso_desc.StreamOutput.pSODeclaration = so_entries;
+    pso_desc.StreamOutput.NumEntries = ARRAY_SIZE(so_entries);
+    pso_desc.StreamOutput.RasterizedStream = D3D12_SO_NO_RASTERIZED_STREAM;
+    pso_desc.pRootSignature = root_signatures[0];
+    pso_desc.SampleDesc.Count = 1;
+    pso_desc.PrimitiveTopologyType = D3D12_PRIMITIVE_TOPOLOGY_TYPE_POINT;
+    pso_desc.RasterizerState.FillMode = D3D12_FILL_MODE_SOLID;
+    pso_desc.RasterizerState.CullMode = D3D12_CULL_MODE_NONE;
+    pso_desc.InputLayout.NumElements = ARRAY_SIZE(layout_desc);
+    pso_desc.InputLayout.pInputElementDescs = layout_desc;
+    hr = ID3D12Device_CreateGraphicsPipelineState(context.device, &pso_desc, &IID_ID3D12PipelineState, (void**)&psos[0]);
+    ok(SUCCEEDED(hr), "Failed to create PSO, hr #%x.\n", hr);
+    pso_desc.VS.pShaderBytecode = vs_code_large_cbv;
+    pso_desc.VS.BytecodeLength = sizeof(vs_code_large_cbv);
+    pso_desc.pRootSignature = root_signatures[1];
+    hr = ID3D12Device_CreateGraphicsPipelineState(context.device, &pso_desc, &IID_ID3D12PipelineState, (void**)&psos[1]);
+    ok(SUCCEEDED(hr), "Failed to create PSO, hr #%x.\n", hr);
+
+    /* Verify sanity checks.
+     * As per validation layers, there must be exactly one command in the signature.
+     * It must come last. Verify that we check for this. */
+    memset(&command_signature_desc, 0, sizeof(command_signature_desc));
+    command_signature_desc.NumArgumentDescs = 1;
+    command_signature_desc.pArgumentDescs = indirect_argument_descs;
+    command_signature_desc.ByteStride = sizeof(D3D12_VERTEX_BUFFER_VIEW);
+    indirect_argument_descs[0].Type = D3D12_INDIRECT_ARGUMENT_TYPE_VERTEX_BUFFER_VIEW;
+    hr = ID3D12Device_CreateCommandSignature(context.device, &command_signature_desc, NULL,
+            &IID_ID3D12CommandSignature, (void**)&command_signature);
+    ok(hr == E_INVALIDARG, "Unexpected hr #%x.\n", hr);
+
+    command_signature_desc.NumArgumentDescs = 2;
+    command_signature_desc.pArgumentDescs = indirect_argument_descs;
+    command_signature_desc.ByteStride = sizeof(D3D12_DRAW_INDEXED_ARGUMENTS) + sizeof(D3D12_VERTEX_BUFFER_VIEW);
+    indirect_argument_descs[0].Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED;
+    indirect_argument_descs[1].Type = D3D12_INDIRECT_ARGUMENT_TYPE_VERTEX_BUFFER_VIEW;
+    hr = ID3D12Device_CreateCommandSignature(context.device, &command_signature_desc, NULL,
+            &IID_ID3D12CommandSignature, (void**)&command_signature);
+    ok(hr == E_INVALIDARG, "Unexpected hr #%x.\n", hr);
+
+    command_signature_desc.ByteStride = sizeof(D3D12_DRAW_INDEXED_ARGUMENTS) + sizeof(D3D12_DRAW_INDEXED_ARGUMENTS);
+    indirect_argument_descs[0].Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED;
+    indirect_argument_descs[1].Type = D3D12_INDIRECT_ARGUMENT_TYPE_DRAW_INDEXED;
+    hr = ID3D12Device_CreateCommandSignature(context.device, &command_signature_desc, NULL,
+            &IID_ID3D12CommandSignature, (void**)&command_signature);
+    ok(hr == E_INVALIDARG, "Unexpected hr #%x.\n", hr);
+
+    for (i = 0; i < ARRAY_SIZE(tests); i++)
+    {
+        struct vec4 expect_reset_state[2];
+        const struct vec4 *expect, *v;
+        uint32_t expected_output_size;
+        uint32_t clear_vbo_mask;
+        bool root_cbv;
+        uint32_t size;
+
+        vkd3d_test_set_context("Test %u", i);
+
+        command_signature_desc.ByteStride = tests[i].stride;
+        command_signature_desc.pArgumentDescs = tests[i].indirect_arguments;
+        command_signature_desc.NumArgumentDescs = tests[i].indirect_argument_count;
+        command_signature_desc.NodeMask = 0;
+        hr = ID3D12Device_CreateCommandSignature(context.device, &command_signature_desc,
+                tests[i].needs_root_sig ? root_signatures[tests[i].pso_index] : NULL,
+                &IID_ID3D12CommandSignature, (void**)&command_signature);
+
+        /* Updating root CBV requires push BDA path, which we don't enable on NV by default yet. */
+        root_cbv = false;
+        for (j = 0; j < tests[i].indirect_argument_count; j++)
+        {
+            if (tests[i].indirect_arguments[j].Type == D3D12_INDIRECT_ARGUMENT_TYPE_CONSTANT_BUFFER_VIEW)
+            {
+                root_cbv = true;
+                break;
+            }
+        }
+
+        if (FAILED(hr))
+        {
+            if (root_cbv && is_nvidia_device(context.device))
+                skip("Creating indirect root CBV update failed. If the GPU is NVIDIA, try VKD3D_CONFIG=force_raw_va_cbv.\n");
+            else
+                skip("Failed creating command signature, skipping test.\n");
+            continue;
+        }
+
+        argument_buffer = create_upload_buffer(context.device, 256 * 1024, NULL);
+        argument_buffer_late = create_default_buffer(context.device, 256 * 1024,
+                D3D12_RESOURCE_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_DEST);
+
+#define UNALIGNED_ARGUMENT_BUFFER_OFFSET (64 * 1024 + 4)
+#define UNALIGNED_COUNT_BUFFER_OFFSET (128 * 1024 + 4)
+#define ALIGNED_COUNT_BUFFER_OFFSET (128 * 1024 + 4 * 1024)
+        {
+            uint8_t *ptr;
+            ID3D12Resource_Map(argument_buffer, 0, NULL, (void**)&ptr);
+            memcpy(ptr, tests[i].argument_buffer_data, tests[i].argument_buffer_size);
+            memcpy(ptr + UNALIGNED_ARGUMENT_BUFFER_OFFSET, tests[i].argument_buffer_data, tests[i].argument_buffer_size);
+            memcpy(ptr + UNALIGNED_COUNT_BUFFER_OFFSET, &tests[i].api_max_count, sizeof(tests[i].api_max_count));
+            memcpy(ptr + ALIGNED_COUNT_BUFFER_OFFSET, &tests[i].api_max_count, sizeof(tests[i].api_max_count));
+            ID3D12Resource_Unmap(argument_buffer, 0, NULL);
+        }
+
+        streamout_buffer = create_default_buffer(context.device, 64 * 1024,
+                D3D12_RESOURCE_FLAG_NONE, D3D12_RESOURCE_STATE_STREAM_OUT);
+
+        ID3D12GraphicsCommandList_SetGraphicsRootSignature(command_list, root_signatures[tests[i].pso_index]);
+        ID3D12GraphicsCommandList_SetPipelineState(command_list, psos[tests[i].pso_index]);
+        sov.SizeInBytes = 64 * 1024 - sizeof(struct vec4);
+        sov.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(streamout_buffer) + sizeof(struct vec4);
+        sov.BufferFilledSizeLocation = ID3D12Resource_GetGPUVirtualAddress(streamout_buffer);
+        ID3D12GraphicsCommandList_SOSetTargets(command_list, 0, 1, &sov);
+
+        /* Set up default rendering state. */
+        ibv.BufferLocation = ID3D12Resource_GetGPUVirtualAddress(ibo[0]);
+        ibv.SizeInBytes = sizeof(ibo_data[0]);
+        ibv.Format = DXGI_FORMAT_R32_UINT;
+        vbvs[0].BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vbo[0]);
+        vbvs[0].SizeInBytes = sizeof(vbo_data[0]);
+        vbvs[0].StrideInBytes = 4;
+        vbvs[1].BufferLocation = ID3D12Resource_GetGPUVirtualAddress(vbo[1]);
+        vbvs[1].SizeInBytes = sizeof(vbo_data[1]);
+        vbvs[1].StrideInBytes = 4;
+
+        ID3D12GraphicsCommandList_IASetIndexBuffer(command_list, &ibv);
+        ID3D12GraphicsCommandList_IASetPrimitiveTopology(command_list, D3D_PRIMITIVE_TOPOLOGY_POINTLIST);
+        ID3D12GraphicsCommandList_IASetVertexBuffers(command_list, 0, 2, vbvs);
+
+        for (j = 0; j < (tests[i].pso_index ? 12 : 1); j++)
+            ID3D12GraphicsCommandList_SetGraphicsRoot32BitConstants(command_list, 0, 4, &values, 4 * j);
+
+        ID3D12GraphicsCommandList_SetGraphicsRootConstantBufferView(command_list, 1,
+                ID3D12Resource_GetGPUVirtualAddress(cbv));
+        ID3D12GraphicsCommandList_SetGraphicsRootShaderResourceView(command_list, 2,
+                ID3D12Resource_GetGPUVirtualAddress(srv));
+        ID3D12GraphicsCommandList_SetGraphicsRootUnorderedAccessView(command_list, 3,
+                ID3D12Resource_GetGPUVirtualAddress(uav));
+        ID3D12GraphicsCommandList_ExecuteIndirect(command_list, command_signature, tests[i].api_max_count,
+                argument_buffer, 0, NULL, 0);
+        /* Test equivalent call with indirect count. */
+        ID3D12GraphicsCommandList_ExecuteIndirect(command_list, command_signature, 1024,
+                argument_buffer, UNALIGNED_ARGUMENT_BUFFER_OFFSET,
+                argument_buffer, UNALIGNED_COUNT_BUFFER_OFFSET);
+        /* Test equivalent, but now with late transition to INDIRECT. */
+        ID3D12GraphicsCommandList_CopyResource(command_list, argument_buffer_late, argument_buffer);
+        transition_resource_state(command_list, argument_buffer_late, D3D12_RESOURCE_STATE_COPY_DEST,
+                D3D12_RESOURCE_STATE_INDIRECT_ARGUMENT);
+        ID3D12GraphicsCommandList_ExecuteIndirect(command_list, command_signature, 1024,
+                argument_buffer_late, 0, argument_buffer_late, ALIGNED_COUNT_BUFFER_OFFSET);
+
+        /* Root descriptors which are part of the state block are cleared to NULL. Recover them here
+         * since attempting to draw next test will crash GPU. */
+        ID3D12GraphicsCommandList_SetGraphicsRootConstantBufferView(command_list, 1,
+                ID3D12Resource_GetGPUVirtualAddress(cbv));
+        ID3D12GraphicsCommandList_SetGraphicsRootShaderResourceView(command_list, 2,
+                ID3D12Resource_GetGPUVirtualAddress(srv));
+        ID3D12GraphicsCommandList_SetGraphicsRootUnorderedAccessView(command_list, 3,
+                ID3D12Resource_GetGPUVirtualAddress(uav));
+
+        /* Other state is cleared to 0. */
+
+        ID3D12GraphicsCommandList_DrawInstanced(command_list, 2, 1, 0, 0);
+        transition_resource_state(command_list, streamout_buffer, D3D12_RESOURCE_STATE_STREAM_OUT, D3D12_RESOURCE_STATE_COPY_SOURCE);
+
+        get_buffer_readback_with_command_list(streamout_buffer, DXGI_FORMAT_R32G32B32A32_FLOAT, &rb, queue, command_list);
+        reset_command_list(command_list, context.allocator);
+
+        expected_output_size = (tests[i].expected_output_count * 3 + 2) * sizeof(struct vec4);
+        size = get_readback_uint(&rb, 0, 0, 0);
+        ok(size == expected_output_size, "Expected size %u, got %u.\n", expected_output_size, size);
+
+        for (j = 0; j < tests[i].expected_output_count; j++)
+        {
+            expect = &tests[i].expected_output[j];
+            v = get_readback_vec4(&rb, j + 1, 0);
+            ok(compare_vec4(v, expect, 0), "Element (direct count) %u failed: (%f, %f, %f, %f) != (%f, %f, %f, %f)\n",
+                    j, v->x, v->y, v->z, v->w, expect->x, expect->y, expect->z, expect->w);
+
+            v = get_readback_vec4(&rb, j + tests[i].expected_output_count + 1, 0);
+            ok(compare_vec4(v, expect, 0), "Element (indirect count) %u failed: (%f, %f, %f, %f) != (%f, %f, %f, %f)\n",
+                    j, v->x, v->y, v->z, v->w, expect->x, expect->y, expect->z, expect->w);
+
+            v = get_readback_vec4(&rb, j + 2 * tests[i].expected_output_count + 1, 0);
+            ok(compare_vec4(v, expect, 0), "Element (late latch) %u failed: (%f, %f, %f, %f) != (%f, %f, %f, %f)\n",
+                    j, v->x, v->y, v->z, v->w, expect->x, expect->y, expect->z, expect->w);
+        }
+
+        clear_vbo_mask = 0;
+        expect_reset_state[0] = values;
+
+        /* Root constant state is cleared to zero if it's part of the signature. */
+        for (j = 0; j < tests[i].indirect_argument_count; j++)
+        {
+            if (tests[i].indirect_arguments[j].Type == D3D12_INDIRECT_ARGUMENT_TYPE_CONSTANT)
+            {
+                for (k = 0; k < tests[i].indirect_arguments[j].Constant.Num32BitValuesToSet; k++)
+                    (&expect_reset_state[0].x)[(tests[i].indirect_arguments[j].Constant.DestOffsetIn32BitValues + k) % 4] = 0.0f;
+            }
+            else if (tests[i].indirect_arguments[j].Type == D3D12_INDIRECT_ARGUMENT_TYPE_VERTEX_BUFFER_VIEW)
+                clear_vbo_mask |= 1u << tests[i].indirect_arguments[j].VertexBuffer.Slot;
+        }
+
+        expect_reset_state[1] = expect_reset_state[0];
+
+        /* VBO/IBO state is cleared to zero if it's part of the signature.
+         * A NULL IBO should be seen as a IBO which only reads 0 index. */
+        if (!(clear_vbo_mask & (1u << 0)))
+            expect_reset_state[1].x += 1.0f;
+
+        if (!(clear_vbo_mask & (1u << 1)))
+        {
+            expect_reset_state[0].y += 64.0f;
+            expect_reset_state[1].y += 65.0f;
+        }
+
+        for (j = 0; j < 2; j++)
+        {
+            v = get_readback_vec4(&rb, j + 1 + 3 * tests[i].expected_output_count, 0);
+            expect = &expect_reset_state[j];
+            ok(compare_vec4(v, expect, 0), "Post-reset element %u failed: (%f, %f, %f, %f) != (%f, %f, %f, %f)\n",
+                    j, v->x, v->y, v->z, v->w, expect->x, expect->y, expect->z, expect->w);
+        }
+
+        ID3D12CommandSignature_Release(command_signature);
+        ID3D12Resource_Release(argument_buffer);
+        ID3D12Resource_Release(argument_buffer_late);
+        ID3D12Resource_Release(streamout_buffer);
+        release_resource_readback(&rb);
+    }
+    vkd3d_test_set_context(NULL);
+
+    for (i = 0; i < ARRAY_SIZE(psos); i++)
+        ID3D12PipelineState_Release(psos[i]);
+    for (i = 0; i < ARRAY_SIZE(root_signatures); i++)
+        ID3D12RootSignature_Release(root_signatures[i]);
+    for (i = 0; i < ARRAY_SIZE(vbo); i++)
+        ID3D12Resource_Release(vbo[i]);
+    for (i = 0; i < ARRAY_SIZE(ibo); i++)
+        ID3D12Resource_Release(ibo[i]);
+    ID3D12Resource_Release(cbv);
+    ID3D12Resource_Release(srv);
+    ID3D12Resource_Release(uav);
+
+    destroy_test_context(&context);
+}
+
 void test_execute_indirect(void)
 {
     ID3D12Resource *argument_buffer, *count_buffer, *uav;

tests/d3d12_descriptors.c

@@ -5293,3 +5293,227 @@ void test_root_descriptor_offset_sign(void)
     destroy_test_context(&context);
 #endif
 }
+
+static void test_uav_counters_null_behavior(bool use_dxil)
+{
+    D3D12_UNORDERED_ACCESS_VIEW_DESC uav_desc;
+    D3D12_ROOT_SIGNATURE_DESC rs_desc;
+    ID3D12DescriptorHeap *cpu_heap;
+    D3D12_ROOT_PARAMETER rs_param;
+    D3D12_DESCRIPTOR_RANGE range;
+    struct test_context context;
+    struct resource_readback rb;
+    ID3D12DescriptorHeap *heap;
+    ID3D12Resource *resource;
+    unsigned int i;
+
+#if 0
+    RWStructuredBuffer<uint> RWBuf[4] : register(u0);
+
+    [numthreads(1, 1, 1)]
+    void main(int wg : SV_GroupID)
+    {
+        RWBuf[wg >> 2][wg & 3] = RWBuf[wg >> 2].IncrementCounter() + 64;
+    }
+#endif
+    static const DWORD cs_code_dxbc[] =
+    {
+        0x43425844, 0xb5433247, 0x4cd30f6c, 0x58100e67, 0xc179ade1, 0x00000001, 0x00000134, 0x00000003,
+        0x0000002c, 0x0000003c, 0x0000004c, 0x4e475349, 0x00000008, 0x00000000, 0x00000008, 0x4e47534f,
+        0x00000008, 0x00000000, 0x00000008, 0x58454853, 0x000000e0, 0x00050051, 0x00000038, 0x0100086a,
+        0x0700009e, 0x0031ee46, 0x00000000, 0x00000000, 0x00000003, 0x00000004, 0x00000000, 0x0200005f,
+        0x00021012, 0x02000068, 0x00000002, 0x0400009b, 0x00000001, 0x00000001, 0x00000001, 0x0600002a,
+        0x00100012, 0x00000000, 0x0002100a, 0x00004001, 0x00000002, 0x06000001, 0x00100022, 0x00000000,
+        0x0002100a, 0x00004001, 0x00000003, 0x070000b2, 0x00100012, 0x00000001, 0x0421e000, 0x00000000,
+        0x0010000a, 0x00000000, 0x0700001e, 0x00100042, 0x00000000, 0x0010000a, 0x00000001, 0x00004001,
+        0x00000040, 0x0b0000a8, 0x0421e012, 0x00000000, 0x0010000a, 0x00000000, 0x0010001a, 0x00000000,
+        0x00004001, 0x00000000, 0x0010002a, 0x00000000, 0x0100003e,
+    };
+
+    static const BYTE cs_code_dxil[] =
+    {
+        0x44, 0x58, 0x42, 0x43, 0xc6, 0xfe, 0xe1, 0x77, 0xd8, 0x5c, 0x56, 0xc7, 0x6e, 0xf7, 0xe2, 0xf7, 0xb3, 0xb0, 0x40, 0xe0, 0x01, 0x00, 0x00, 0x00, 0x34, 0x06, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00,
+        0x38, 0x00, 0x00, 0x00, 0x48, 0x00, 0x00, 0x00, 0x58, 0x00, 0x00, 0x00, 0x68, 0x00, 0x00, 0x00, 0xd0, 0x00, 0x00, 0x00, 0xec, 0x00, 0x00, 0x00, 0x53, 0x46, 0x49, 0x30, 0x08, 0x00, 0x00, 0x00,
+        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x49, 0x53, 0x47, 0x31, 0x08, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x4f, 0x53, 0x47, 0x31, 0x08, 0x00, 0x00, 0x00,
+        0x00, 0x00, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x50, 0x53, 0x56, 0x30, 0x60, 0x00, 0x00, 0x00, 0x30, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xff, 0xff, 0xff, 0xff, 0x05, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00,
+        0x01, 0x00, 0x00, 0x00, 0x01, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x09, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00, 0x0c, 0x00, 0x00, 0x00,
+        0x00, 0x00, 0x00, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x48, 0x41, 0x53, 0x48, 0x14, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x2c, 0xc1, 0xf5, 0xe2,
+        0x29, 0x0a, 0x7c, 0x68, 0x4a, 0xfa, 0x15, 0xe9, 0x1a, 0x85, 0x63, 0x21, 0x44, 0x58, 0x49, 0x4c, 0x40, 0x05, 0x00, 0x00, 0x60, 0x00, 0x05, 0x00, 0x50, 0x01, 0x00, 0x00, 0x44, 0x58, 0x49, 0x4c,
+        0x00, 0x01, 0x00, 0x00, 0x10, 0x00, 0x00, 0x00, 0x28, 0x05, 0x00, 0x00, 0x42, 0x43, 0xc0, 0xde, 0x21, 0x0c, 0x00, 0x00, 0x47, 0x01, 0x00, 0x00, 0x0b, 0x82, 0x20, 0x00, 0x02, 0x00, 0x00, 0x00,
+        0x13, 0x00, 0x00, 0x00, 0x07, 0x81, 0x23, 0x91, 0x41, 0xc8, 0x04, 0x49, 0x06, 0x10, 0x32, 0x39, 0x92, 0x01, 0x84, 0x0c, 0x25, 0x05, 0x08, 0x19, 0x1e, 0x04, 0x8b, 0x62, 0x80, 0x14, 0x45, 0x02,
+        0x42, 0x92, 0x0b, 0x42, 0xa4, 0x10, 0x32, 0x14, 0x38, 0x08, 0x18, 0x4b, 0x0a, 0x32, 0x52, 0x88, 0x48, 0x90, 0x14, 0x20, 0x43, 0x46, 0x88, 0xa5, 0x00, 0x19, 0x32, 0x42, 0xe4, 0x48, 0x0e, 0x90,
+        0x91, 0x22, 0xc4, 0x50, 0x41, 0x51, 0x81, 0x8c, 0xe1, 0x83, 0xe5, 0x8a, 0x04, 0x29, 0x46, 0x06, 0x51, 0x18, 0x00, 0x00, 0x08, 0x00, 0x00, 0x00, 0x1b, 0x8c, 0xe0, 0xff, 0xff, 0xff, 0xff, 0x07,
+        0x40, 0x02, 0xa8, 0x0d, 0x84, 0xf0, 0xff, 0xff, 0xff, 0xff, 0x03, 0x20, 0x6d, 0x30, 0x86, 0xff, 0xff, 0xff, 0xff, 0x1f, 0x00, 0x09, 0xa8, 0x00, 0x49, 0x18, 0x00, 0x00, 0x03, 0x00, 0x00, 0x00,
+        0x13, 0x82, 0x60, 0x42, 0x20, 0x4c, 0x08, 0x06, 0x00, 0x00, 0x00, 0x00, 0x89, 0x20, 0x00, 0x00, 0x27, 0x00, 0x00, 0x00, 0x32, 0x22, 0x48, 0x09, 0x20, 0x64, 0x85, 0x04, 0x93, 0x22, 0xa4, 0x84,
+        0x04, 0x93, 0x22, 0xe3, 0x84, 0xa1, 0x90, 0x14, 0x12, 0x4c, 0x8a, 0x8c, 0x0b, 0x84, 0xa4, 0x4c, 0x10, 0x54, 0x23, 0x00, 0x25, 0x00, 0x14, 0xe6, 0x08, 0xc0, 0xa0, 0x0c, 0x63, 0x0c, 0x22, 0x73,
+        0x04, 0x08, 0x99, 0x7b, 0x86, 0xcb, 0x9f, 0xb0, 0x87, 0x90, 0xfc, 0x10, 0x68, 0x86, 0x85, 0x40, 0xc1, 0x29, 0xc4, 0x18, 0xc8, 0x50, 0x9a, 0x23, 0x08, 0x8a, 0x81, 0x86, 0x19, 0x63, 0x11, 0x2b,
+        0x0a, 0x18, 0x68, 0x8c, 0x31, 0xc6, 0x30, 0xe4, 0x06, 0x02, 0x66, 0x32, 0x83, 0x71, 0x60, 0x87, 0x70, 0x98, 0x87, 0x79, 0x70, 0x03, 0x59, 0xb8, 0x85, 0x59, 0xa0, 0x07, 0x79, 0xa8, 0x87, 0x71,
+        0xa0, 0x87, 0x7a, 0x90, 0x87, 0x72, 0x20, 0x07, 0x51, 0xa8, 0x07, 0x73, 0x30, 0x87, 0x72, 0x90, 0x07, 0x3e, 0xa8, 0x07, 0x77, 0x98, 0x87, 0x74, 0x38, 0x07, 0x77, 0x28, 0x07, 0x72, 0x00, 0x83,
+        0x74, 0x70, 0x07, 0x7a, 0xf0, 0x03, 0x14, 0x8c, 0x24, 0x88, 0x24, 0xe7, 0x08, 0x40, 0x01, 0x00, 0x13, 0x14, 0x72, 0xc0, 0x87, 0x74, 0x60, 0x87, 0x36, 0x68, 0x87, 0x79, 0x68, 0x03, 0x72, 0xc0,
+        0x87, 0x0d, 0xaf, 0x50, 0x0e, 0x6d, 0xd0, 0x0e, 0x7a, 0x50, 0x0e, 0x6d, 0x00, 0x0f, 0x7a, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x71, 0xa0, 0x07, 0x73, 0x20, 0x07,
+        0x6d, 0x90, 0x0e, 0x78, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90, 0x0e, 0x71, 0x60, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xd0, 0x06, 0xe9, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x90,
+        0x0e, 0x76, 0x40, 0x07, 0x7a, 0x60, 0x07, 0x74, 0xd0, 0x06, 0xe6, 0x10, 0x07, 0x76, 0xa0, 0x07, 0x73, 0x20, 0x07, 0x6d, 0x60, 0x0e, 0x73, 0x20, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xd0, 0x06, 0xe6,
+        0x60, 0x07, 0x74, 0xa0, 0x07, 0x76, 0x40, 0x07, 0x6d, 0xe0, 0x0e, 0x78, 0xa0, 0x07, 0x71, 0x60, 0x07, 0x7a, 0x30, 0x07, 0x72, 0xa0, 0x07, 0x76, 0x40, 0x07, 0x43, 0x9e, 0x00, 0x00, 0x00, 0x00,
+        0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x86, 0x3c, 0x04, 0x10, 0x00, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x0c, 0x79, 0x12, 0x20, 0x00, 0x04, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+        0x00, 0x18, 0xf2, 0x30, 0x40, 0x00, 0x0c, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x30, 0xe4, 0x71, 0x80, 0x00, 0x10, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x0b, 0x04, 0x00, 0x00,
+        0x09, 0x00, 0x00, 0x00, 0x32, 0x1e, 0x98, 0x10, 0x19, 0x11, 0x4c, 0x90, 0x8c, 0x09, 0x26, 0x47, 0xc6, 0x04, 0x43, 0x1a, 0x25, 0x30, 0x02, 0x50, 0x0c, 0x85, 0x50, 0x18, 0xb4, 0x46, 0x00, 0x6a,
+        0x80, 0x68, 0x81, 0xd0, 0x9c, 0x01, 0x00, 0x00, 0x79, 0x18, 0x00, 0x00, 0x3b, 0x00, 0x00, 0x00, 0x1a, 0x03, 0x4c, 0x90, 0x46, 0x02, 0x13, 0x44, 0x35, 0x18, 0x63, 0x0b, 0x73, 0x3b, 0x03, 0xb1,
+        0x2b, 0x93, 0x9b, 0x4b, 0x7b, 0x73, 0x03, 0x99, 0x71, 0xb9, 0x01, 0x41, 0xa1, 0x0b, 0x3b, 0x9b, 0x7b, 0x91, 0x2a, 0x62, 0x2a, 0x0a, 0x9a, 0x2a, 0xfa, 0x9a, 0xb9, 0x81, 0x79, 0x31, 0x4b, 0x73,
+        0x0b, 0x63, 0x4b, 0xd9, 0x10, 0x04, 0x13, 0x84, 0xa1, 0x98, 0x20, 0x0c, 0xc6, 0x06, 0x61, 0x20, 0x26, 0x08, 0xc3, 0xb1, 0x41, 0x18, 0x0c, 0x0a, 0x63, 0x73, 0x1b, 0x06, 0xc4, 0x20, 0x26, 0x08,
+        0x53, 0x43, 0x60, 0x82, 0x30, 0x20, 0x13, 0x84, 0x21, 0x99, 0x20, 0x2c, 0xca, 0x04, 0x61, 0x59, 0x36, 0x08, 0x03, 0xb3, 0x61, 0x21, 0x94, 0x85, 0x20, 0x98, 0xc6, 0x79, 0x1c, 0x68, 0x43, 0x10,
+        0x6d, 0x20, 0x00, 0x09, 0x00, 0x26, 0x08, 0x02, 0x40, 0xa2, 0x2d, 0x2c, 0xcd, 0x6d, 0x82, 0x40, 0x31, 0x1b, 0x86, 0x61, 0x18, 0x36, 0x10, 0x84, 0xc5, 0x5c, 0x1b, 0x0a, 0xaa, 0x02, 0x26, 0xac,
+        0x0a, 0x1b, 0x9b, 0x5d, 0x9b, 0x4b, 0x1a, 0x59, 0x99, 0x1b, 0xdd, 0x94, 0x20, 0xa8, 0x42, 0x86, 0xe7, 0x62, 0x57, 0x26, 0x37, 0x97, 0xf6, 0xe6, 0x36, 0x25, 0x20, 0x9a, 0x90, 0xe1, 0xb9, 0xd8,
+        0x85, 0xb1, 0xd9, 0x95, 0xc9, 0x4d, 0x09, 0x8c, 0x3a, 0x64, 0x78, 0x2e, 0x73, 0x68, 0x61, 0x64, 0x65, 0x72, 0x4d, 0x6f, 0x64, 0x65, 0x6c, 0x53, 0x02, 0xa4, 0x0c, 0x19, 0x9e, 0x8b, 0x5c, 0xd9,
+        0xdc, 0x5b, 0x9d, 0xdc, 0x58, 0xd9, 0xdc, 0x94, 0x40, 0xaa, 0x43, 0x86, 0xe7, 0x52, 0xe6, 0x46, 0x27, 0x97, 0x07, 0xf5, 0x96, 0xe6, 0x46, 0x37, 0x37, 0x25, 0xc0, 0x00, 0x79, 0x18, 0x00, 0x00,
+        0x4c, 0x00, 0x00, 0x00, 0x33, 0x08, 0x80, 0x1c, 0xc4, 0xe1, 0x1c, 0x66, 0x14, 0x01, 0x3d, 0x88, 0x43, 0x38, 0x84, 0xc3, 0x8c, 0x42, 0x80, 0x07, 0x79, 0x78, 0x07, 0x73, 0x98, 0x71, 0x0c, 0xe6,
+        0x00, 0x0f, 0xed, 0x10, 0x0e, 0xf4, 0x80, 0x0e, 0x33, 0x0c, 0x42, 0x1e, 0xc2, 0xc1, 0x1d, 0xce, 0xa1, 0x1c, 0x66, 0x30, 0x05, 0x3d, 0x88, 0x43, 0x38, 0x84, 0x83, 0x1b, 0xcc, 0x03, 0x3d, 0xc8,
+        0x43, 0x3d, 0x8c, 0x03, 0x3d, 0xcc, 0x78, 0x8c, 0x74, 0x70, 0x07, 0x7b, 0x08, 0x07, 0x79, 0x48, 0x87, 0x70, 0x70, 0x07, 0x7a, 0x70, 0x03, 0x76, 0x78, 0x87, 0x70, 0x20, 0x87, 0x19, 0xcc, 0x11,
+        0x0e, 0xec, 0x90, 0x0e, 0xe1, 0x30, 0x0f, 0x6e, 0x30, 0x0f, 0xe3, 0xf0, 0x0e, 0xf0, 0x50, 0x0e, 0x33, 0x10, 0xc4, 0x1d, 0xde, 0x21, 0x1c, 0xd8, 0x21, 0x1d, 0xc2, 0x61, 0x1e, 0x66, 0x30, 0x89,
+        0x3b, 0xbc, 0x83, 0x3b, 0xd0, 0x43, 0x39, 0xb4, 0x03, 0x3c, 0xbc, 0x83, 0x3c, 0x84, 0x03, 0x3b, 0xcc, 0xf0, 0x14, 0x76, 0x60, 0x07, 0x7b, 0x68, 0x07, 0x37, 0x68, 0x87, 0x72, 0x68, 0x07, 0x37,
+        0x80, 0x87, 0x70, 0x90, 0x87, 0x70, 0x60, 0x07, 0x76, 0x28, 0x07, 0x76, 0xf8, 0x05, 0x76, 0x78, 0x87, 0x77, 0x80, 0x87, 0x5f, 0x08, 0x87, 0x71, 0x18, 0x87, 0x72, 0x98, 0x87, 0x79, 0x98, 0x81,
+        0x2c, 0xee, 0xf0, 0x0e, 0xee, 0xe0, 0x0e, 0xf5, 0xc0, 0x0e, 0xec, 0x30, 0x03, 0x62, 0xc8, 0xa1, 0x1c, 0xe4, 0xa1, 0x1c, 0xcc, 0xa1, 0x1c, 0xe4, 0xa1, 0x1c, 0xdc, 0x61, 0x1c, 0xca, 0x21, 0x1c,
+        0xc4, 0x81, 0x1d, 0xca, 0x61, 0x06, 0xd6, 0x90, 0x43, 0x39, 0xc8, 0x43, 0x39, 0x98, 0x43, 0x39, 0xc8, 0x43, 0x39, 0xb8, 0xc3, 0x38, 0x94, 0x43, 0x38, 0x88, 0x03, 0x3b, 0x94, 0xc3, 0x2f, 0xbc,
+        0x83, 0x3c, 0xfc, 0x82, 0x3b, 0xd4, 0x03, 0x3b, 0xb0, 0xc3, 0x0c, 0xc4, 0x21, 0x07, 0x7c, 0x70, 0x03, 0x7a, 0x28, 0x87, 0x76, 0x80, 0x87, 0x19, 0xd1, 0x43, 0x0e, 0xf8, 0xe0, 0x06, 0xe4, 0x20,
+        0x0e, 0xe7, 0xe0, 0x06, 0xf6, 0x10, 0x0e, 0xf2, 0xc0, 0x0e, 0xe1, 0x90, 0x0f, 0xef, 0x50, 0x0f, 0xf4, 0x00, 0x00, 0x00, 0x71, 0x20, 0x00, 0x00, 0x13, 0x00, 0x00, 0x00, 0x46, 0x50, 0x0d, 0x97,
+        0xef, 0x3c, 0x7e, 0x40, 0x15, 0x05, 0x11, 0xb1, 0x93, 0x13, 0x11, 0x3e, 0x72, 0xdb, 0x26, 0x90, 0x0d, 0x97, 0xef, 0x3c, 0x7e, 0x40, 0x15, 0x05, 0x11, 0xb9, 0xcf, 0x00, 0x4c, 0x04, 0xe7, 0x50,
+        0xcd, 0x44, 0x44, 0x36, 0x20, 0x0d, 0x97, 0xef, 0x3c, 0xbe, 0x10, 0x11, 0xc0, 0x44, 0x84, 0x40, 0x33, 0x2c, 0x84, 0x05, 0x44, 0xc3, 0xe5, 0x3b, 0x8f, 0x6f, 0x44, 0x0e, 0xf5, 0x88, 0x83, 0x8f,
+        0xdc, 0xb6, 0x01, 0x10, 0x0c, 0x80, 0x34, 0x00, 0x61, 0x20, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x13, 0x04, 0x41, 0x2c, 0x10, 0x00, 0x00, 0x00, 0x06, 0x00, 0x00, 0x00, 0x34, 0x66, 0x00, 0x8a,
+        0x37, 0xa0, 0x08, 0x4a, 0xae, 0x18, 0x03, 0x0a, 0x30, 0xa0, 0x0c, 0x4a, 0x31, 0x80, 0x4c, 0x09, 0x00, 0x00, 0x00, 0x00, 0x23, 0x06, 0x06, 0x00, 0x82, 0x60, 0x40, 0x58, 0x48, 0x54, 0x01, 0x92,
+        0x15, 0x4c, 0x30, 0x62, 0x90, 0x00, 0x20, 0x08, 0x06, 0xcc, 0x95, 0x10, 0x54, 0x00, 0x8d, 0x18, 0x1c, 0x00, 0x08, 0x82, 0x41, 0x92, 0x25, 0x41, 0x51, 0x41, 0x02, 0x65, 0x24, 0x3a, 0x62, 0xd0,
+        0x00, 0x20, 0x08, 0x06, 0x8e, 0x96, 0x10, 0x01, 0x26, 0x40, 0x10, 0x84, 0x20, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+    };
+
+    static const D3D12_SHADER_BYTECODE cs_dxbc = SHADER_BYTECODE(cs_code_dxbc);
+    static const D3D12_SHADER_BYTECODE cs_dxil = SHADER_BYTECODE(cs_code_dxil);
+
+    if (!init_compute_test_context(&context))
+        return;
+
+    if (use_dxil && !context_supports_dxil(&context))
+    {
+        skip("Context does not support DXIL.\n");
+        destroy_test_context(&context);
+        return;
+    }
+
+    memset(&rs_desc, 0, sizeof(rs_desc));
+    memset(&rs_param, 0, sizeof(rs_param));
+    memset(&range, 0, sizeof(range));
+    rs_desc.NumParameters = 1;
+    rs_desc.pParameters = &rs_param;
+    rs_param.ParameterType = D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
+    rs_param.ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
+    rs_param.DescriptorTable.NumDescriptorRanges = 1;
+    rs_param.DescriptorTable.pDescriptorRanges = &range;
+    range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_UAV;
+    range.NumDescriptors = 8;
+    create_root_signature(context.device, &rs_desc, &context.root_signature);
+    context.pipeline_state = create_compute_pipeline_state(context.device, context.root_signature, use_dxil ? cs_dxil : cs_dxbc);
+
+    cpu_heap = create_cpu_descriptor_heap(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV, 8);
+    heap = create_gpu_descriptor_heap(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV, 8);
+    resource = create_default_buffer(context.device, D3D12_UAV_COUNTER_PLACEMENT_ALIGNMENT * 9,
+            D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
+
+    memset(&uav_desc, 0, sizeof(uav_desc));
+    uav_desc.ViewDimension = D3D12_UAV_DIMENSION_BUFFER;
+    for (i = 0; i < 8; i++)
+    {
+        D3D12_CPU_DESCRIPTOR_HANDLE cpu_h, gpu_h;
+        cpu_h = ID3D12DescriptorHeap_GetCPUDescriptorHandleForHeapStart(cpu_heap);
+        gpu_h = ID3D12DescriptorHeap_GetCPUDescriptorHandleForHeapStart(heap);
+        cpu_h.ptr += ID3D12Device_GetDescriptorHandleIncrementSize(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV) * i;
+        gpu_h.ptr += ID3D12Device_GetDescriptorHandleIncrementSize(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV) * i;
+
+        uav_desc.Buffer.NumElements = 4;
+        uav_desc.Buffer.FirstElement = 4 * i;
+        uav_desc.Buffer.StructureByteStride = 4;
+        uav_desc.Buffer.CounterOffsetInBytes = D3D12_UAV_COUNTER_PLACEMENT_ALIGNMENT * (i + 1);
+
+        /* AMD drivers don't seem to clear the UAV counter if we pass in NULL, so
+         * test a path which does not do that. */
+        if (i < 4)
+        {
+            ID3D12Device_CreateUnorderedAccessView(context.device, resource, resource, &uav_desc, cpu_h);
+            ID3D12Device_CreateUnorderedAccessView(context.device, resource, resource, &uav_desc, gpu_h);
+        }
+
+        uav_desc.Buffer.CounterOffsetInBytes = 0;
+
+        /* Test writing NULL UAV counter after a non-NULL UAV counter. Makes sure that we are indeed supposed
+         * to clear out UAV counters to NULL every time. */
+        if ((i & 3) == 3)
+        {
+            ID3D12Device_CreateUnorderedAccessView(context.device, NULL, NULL, &uav_desc, cpu_h);
+            ID3D12Device_CreateUnorderedAccessView(context.device, NULL, NULL, &uav_desc, gpu_h);
+        }
+        else if ((i & 3) >= 1)
+        {
+            ID3D12Device_CreateUnorderedAccessView(context.device, resource, NULL, &uav_desc, cpu_h);
+            ID3D12Device_CreateUnorderedAccessView(context.device, resource, NULL, &uav_desc, gpu_h);
+        }
+        else
+        {
+            /* Test copy behavior. Make sure we correctly copy NULL counters as well. */
+            ID3D12Device_CreateUnorderedAccessView(context.device, resource, NULL, &uav_desc, cpu_h);
+            ID3D12Device_CopyDescriptorsSimple(context.device, 1,
+                       gpu_h, cpu_h, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
+        }
+    }
+
+    ID3D12GraphicsCommandList_SetDescriptorHeaps(context.list, 1, &heap);
+    ID3D12GraphicsCommandList_SetComputeRootSignature(context.list, context.root_signature);
+    ID3D12GraphicsCommandList_SetPipelineState(context.list, context.pipeline_state);
+    ID3D12GraphicsCommandList_SetComputeRootDescriptorTable(context.list, 0,
+              ID3D12DescriptorHeap_GetGPUDescriptorHandleForHeapStart(heap));
+    ID3D12GraphicsCommandList_Dispatch(context.list, 8 * 4, 1, 1);
+    transition_resource_state(context.list, resource, D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
+    get_buffer_readback_with_command_list(resource, DXGI_FORMAT_R32_UINT, &rb, context.queue, context.list);
+
+    for (i = 0; i < 8 * 4; i++)
+    {
+        /* Possible behavior is very varied here:
+         * NV: If UAV counter is NULL, NV makes the main descriptor robust.
+         * AMD: Writing NULL uav counter does not update the counter descriptor, the atomic update will still go through.
+         * Intel: Behaves as you would expect. Atomic op returns 0, writes to main descriptor behaves as you'd expect. */
+        uint32_t value = get_readback_uint(&rb, i, 0, 0);
+        ok(value == 0 || (value >= 64 && value < (64 + 4)), "Unexpected value %u = %u\n", i, value);
+    }
+
+    for (i = 0; i < 8; i++)
+    {
+        uint32_t value = get_readback_uint(&rb, (i + 1) * (D3D12_UAV_COUNTER_PLACEMENT_ALIGNMENT / 4), 0, 0);
+        if (i < 4)
+        {
+            /* AMD behavior: Passing NULL does not necessarily clear out UAV counter.
+             * It is undefined to access UAV counter like this.
+             * https://docs.microsoft.com/en-us/windows/win32/direct3d12/uav-counters
+             * "If a shader attempts to access the counter of a UAV that does not have an associated counter,
+             *  then the debug layer will issue a warning,
+             *  and a GPU page fault will occur causing the apps’s device to be removed." */
+            ok(value == 0 || value == 4, "Unexpected counter %u = %u.\n", i, value);
+        }
+        else
+        {
+            /* Technically undefined, but all drivers behave robustly here, we should too. */
+            ok(value == 0, "Unexpected counter %u = %u.\n", i, value);
+        }
+    }
+
+    release_resource_readback(&rb);
+    ID3D12DescriptorHeap_Release(heap);
+    ID3D12DescriptorHeap_Release(cpu_heap);
+    ID3D12Resource_Release(resource);
+    destroy_test_context(&context);
+}
+
+void test_uav_counter_null_behavior_dxbc(void)
+{
+    test_uav_counters_null_behavior(false);
+}
+
+void test_uav_counter_null_behavior_dxil(void)
+{
+    test_uav_counters_null_behavior(true);
+}

tests/d3d12_root_signature.c

@@ -1467,3 +1467,36 @@ void test_missing_bindings_root_signature(void)
     destroy_test_context(&context);
 }
 
+void test_root_signature_empty_blob(void)
+{
+    ID3D12RootSignature *root_signature;
+    struct test_context context;
+    HRESULT hr;
+
+    static const DWORD cs_code[] =
+    {
+#if 0
+    RWStructuredBuffer<uint> RWBuf;
+
+    [numthreads(1, 1, 1)]
+    void main(int wg : SV_GroupID)
+    {
+            RWBuf[wg] = wg;
+    }
+#endif
+        0x43425844, 0x81a88c98, 0x1ab24abd, 0xfdb8fb1f, 0x7e9cb035, 0x00000001, 0x000000a8, 0x00000003,
+        0x0000002c, 0x0000003c, 0x0000004c, 0x4e475349, 0x00000008, 0x00000000, 0x00000008, 0x4e47534f,
+        0x00000008, 0x00000000, 0x00000008, 0x58454853, 0x00000054, 0x00050050, 0x00000015, 0x0100086a,
+        0x0400009e, 0x0011e000, 0x00000000, 0x00000004, 0x0200005f, 0x00021012, 0x0400009b, 0x00000001,
+        0x00000001, 0x00000001, 0x070000a8, 0x0011e012, 0x00000000, 0x0002100a, 0x00004001, 0x00000000,
+        0x0002100a, 0x0100003e,
+    };
+
+    if (!init_compute_test_context(&context))
+        return;
+
+    hr = ID3D12Device_CreateRootSignature(context.device, 0, cs_code, sizeof(cs_code), &IID_ID3D12RootSignature, (void **)&root_signature);
+    /* Has to be E_FAIL, not E_INVALIDARG, oddly enough. */
+    ok(hr == E_FAIL, "Unexpected hr #%x.\n", hr);
+    destroy_test_context(&context);
+}

tests/d3d12_sparse.c

@@ -3383,3 +3383,248 @@ void test_texture_feedback_instructions_dxil(void)
     test_texture_feedback_instructions(true);
 }
 
+void test_sparse_buffer_memory_lifetime(void)
+{
+    /* Attempt to bind sparse memory, then free the underlying heap, but keep the sparse resource
+     * alive. This should confuse drivers that attempt to track BO lifetimes. */
+    D3D12_UNORDERED_ACCESS_VIEW_DESC uav_desc;
+    D3D12_SHADER_RESOURCE_VIEW_DESC srv_desc;
+    D3D12_FEATURE_DATA_D3D12_OPTIONS options;
+    const UINT values[] = { 42, 42, 42, 42 };
+    D3D12_ROOT_PARAMETER root_parameters[2];
+    D3D12_TILE_REGION_SIZE region_size;
+    D3D12_GPU_DESCRIPTOR_HANDLE h_gpu;
+    D3D12_CPU_DESCRIPTOR_HANDLE h_cpu;
+    D3D12_ROOT_SIGNATURE_DESC rs_desc;
+    D3D12_DESCRIPTOR_RANGE desc_range;
+    struct test_context context;
+    struct resource_readback rb;
+    ID3D12DescriptorHeap *cpu;
+    ID3D12DescriptorHeap *gpu;
+    D3D12_HEAP_DESC heap_desc;
+    D3D12_RESOURCE_DESC desc;
+    ID3D12Resource *sparse;
+    ID3D12Resource *buffer;
+    ID3D12Heap *heap_live;
+    ID3D12Heap *heap;
+    unsigned int i;
+    HRESULT hr;
+
+    static const DWORD cs_sparse_query_dxbc[] =
+    {
+#if 0
+    RWStructuredBuffer<uint> RWBuf : register(u0);
+    Buffer<uint> Buf : register(t0);
+
+    [numthreads(1, 1, 1)]
+    void main(uint thr : SV_DispatchThreadID)
+    {
+            uint code;
+
+            // Sample mapped, but freed memory. See what CheckAccessFullyMapped returns.
+            uint data = Buf.Load(thr, code);
+            uint value = CheckAccessFullyMapped(code) ? (1u << 16) : 0u;
+            value |= data & 0xffffu;
+            RWBuf[2 * thr + 0] = value;
+
+            // Sample not yet mapped memory. See what CheckAccessFullyMapped returns.
+            data = Buf.Load(thr + 1024 * 1024, code);
+            value = CheckAccessFullyMapped(code) ? (1u << 16) : 0u;
+            value |= data & 0xffffu;
+
+            RWBuf[2 * thr + 1] = value;
+    }
+#endif
+        0x43425844, 0x8c2a40af, 0x2a9b20a6, 0xa99f0977, 0x37daacf5, 0x00000001, 0x00000280, 0x00000004,
+        0x00000030, 0x00000040, 0x00000050, 0x00000270, 0x4e475349, 0x00000008, 0x00000000, 0x00000008,
+        0x4e47534f, 0x00000008, 0x00000000, 0x00000008, 0x58454853, 0x00000218, 0x00050050, 0x00000086,
+        0x0100086a, 0x04000858, 0x00107000, 0x00000000, 0x00004444, 0x0400009e, 0x0011e000, 0x00000000,
+        0x00000004, 0x0200005f, 0x00020012, 0x02000068, 0x00000002, 0x0400009b, 0x00000001, 0x00000001,
+        0x00000001, 0x8a0000df, 0x80000042, 0x00111103, 0x00100012, 0x00000000, 0x00100012, 0x00000001,
+        0x00020006, 0x00107e46, 0x00000000, 0x050000ea, 0x00100022, 0x00000000, 0x0010000a, 0x00000001,
+        0x09000037, 0x00100022, 0x00000000, 0x0010001a, 0x00000000, 0x00004001, 0x00010000, 0x00004001,
+        0x00000000, 0x0b00008c, 0x00100012, 0x00000000, 0x00004001, 0x00000010, 0x00004001, 0x00000000,
+        0x0010000a, 0x00000000, 0x0010001a, 0x00000000, 0x06000029, 0x00100022, 0x00000000, 0x0002000a,
+        0x00004001, 0x00000001, 0x090000a8, 0x0011e012, 0x00000000, 0x0010001a, 0x00000000, 0x00004001,
+        0x00000000, 0x0010000a, 0x00000000, 0x1300008c, 0x00100052, 0x00000000, 0x00004002, 0x00000014,
+        0x00000000, 0x0000001f, 0x00000000, 0x00004002, 0x00000000, 0x00000000, 0x00000001, 0x00000000,
+        0x00020006, 0x00004002, 0x00100000, 0x00000000, 0x00000001, 0x00000000, 0x8b0000df, 0x80000042,
+        0x00111103, 0x00100012, 0x00000000, 0x00100012, 0x00000001, 0x00100006, 0x00000000, 0x00107e46,
+        0x00000000, 0x050000ea, 0x00100082, 0x00000000, 0x0010000a, 0x00000001, 0x09000037, 0x00100082,
+        0x00000000, 0x0010003a, 0x00000000, 0x00004001, 0x00010000, 0x00004001, 0x00000000, 0x0b00008c,
+        0x00100012, 0x00000000, 0x00004001, 0x00000010, 0x00004001, 0x00000000, 0x0010000a, 0x00000000,
+        0x0010003a, 0x00000000, 0x090000a8, 0x0011e012, 0x00000000, 0x0010002a, 0x00000000, 0x00004001,
+        0x00000000, 0x0010000a, 0x00000000, 0x0100003e, 0x30494653, 0x00000008, 0x00000100, 0x00000000,
+    };
+    static const D3D12_SHADER_BYTECODE cs_sparse_query = SHADER_BYTECODE(cs_sparse_query_dxbc);
+
+    if (!init_compute_test_context(&context))
+        return;
+
+    hr = ID3D12Device_CheckFeatureSupport(context.device, D3D12_FEATURE_D3D12_OPTIONS, &options, sizeof(options));
+    ok(hr == S_OK, "Failed to check feature support, hr %#x.\n", hr);
+
+    if (options.TiledResourcesTier < D3D12_TILED_RESOURCES_TIER_2)
+    {
+        skip("Tiled resources Tier 2 not supported by device.\n");
+        destroy_test_context(&context);
+        return;
+    }
+
+    memset(&rs_desc, 0, sizeof(rs_desc));
+    memset(root_parameters, 0, sizeof(root_parameters));
+    memset(&desc_range, 0, sizeof(desc_range));
+    rs_desc.NumParameters = ARRAY_SIZE(root_parameters);
+    rs_desc.pParameters = root_parameters;
+    root_parameters[0].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
+    root_parameters[0].ParameterType = D3D12_ROOT_PARAMETER_TYPE_UAV;
+    root_parameters[1].ShaderVisibility = D3D12_SHADER_VISIBILITY_ALL;
+    root_parameters[1].ParameterType = D3D12_ROOT_PARAMETER_TYPE_DESCRIPTOR_TABLE;
+    root_parameters[1].DescriptorTable.NumDescriptorRanges = 1;
+    root_parameters[1].DescriptorTable.pDescriptorRanges = &desc_range;
+    desc_range.NumDescriptors = 1;
+    desc_range.RangeType = D3D12_DESCRIPTOR_RANGE_TYPE_SRV;
+    create_root_signature(context.device, &rs_desc, &context.root_signature);
+    context.pipeline_state = create_compute_pipeline_state(context.device, context.root_signature, cs_sparse_query);
+
+    memset(&heap_desc, 0, sizeof(heap_desc));
+    heap_desc.SizeInBytes = 4 * 1024 * 1024;
+    heap_desc.Properties.Type = D3D12_HEAP_TYPE_DEFAULT;
+    heap_desc.Alignment = D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT;
+    heap_desc.Flags = D3D12_HEAP_FLAG_ALLOW_ONLY_BUFFERS;
+    hr = ID3D12Device_CreateHeap(context.device, &heap_desc, &IID_ID3D12Heap, (void**)&heap);
+    ok(SUCCEEDED(hr), "Failed to create heap, hr #%x.\n", hr);
+    hr = ID3D12Device_CreateHeap(context.device, &heap_desc, &IID_ID3D12Heap, (void**)&heap_live);
+    ok(SUCCEEDED(hr), "Failed to create heap, hr #%x.\n", hr);
+
+    memset(&desc, 0, sizeof(desc));
+    desc.Width = 64 * 1024 * 1024;
+    desc.Height = 1;
+    desc.DepthOrArraySize = 1;
+    desc.SampleDesc.Count = 1;
+    desc.Format = DXGI_FORMAT_UNKNOWN;
+    desc.Dimension = D3D12_RESOURCE_DIMENSION_BUFFER;
+    desc.MipLevels = 1;
+    desc.Alignment = D3D12_DEFAULT_RESOURCE_PLACEMENT_ALIGNMENT;
+    desc.Layout = D3D12_TEXTURE_LAYOUT_ROW_MAJOR;
+    desc.Flags = D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS;
+    hr = ID3D12Device_CreateReservedResource(context.device, &desc, D3D12_RESOURCE_STATE_UNORDERED_ACCESS,
+            NULL, &IID_ID3D12Resource, (void**)&sparse);
+    ok(SUCCEEDED(hr), "Failed to create reserved resource, hr #%x.\n", hr);
+
+    {
+        const D3D12_TILED_RESOURCE_COORDINATE region_start_coordinate = { 0 };
+        const D3D12_TILE_RANGE_FLAGS range_flag = D3D12_TILE_RANGE_FLAG_NULL;
+        const UINT offset = 0;
+        const UINT count = desc.Width / (64 * 1024);
+        region_size.UseBox = FALSE;
+        region_size.NumTiles = desc.Width / (64 * 1024);
+        ID3D12CommandQueue_UpdateTileMappings(context.queue, sparse, 1, &region_start_coordinate, &region_size,
+                NULL, 1, &range_flag, &offset, &count, D3D12_TILE_MAPPING_FLAG_NONE);
+    }
+
+    region_size.UseBox = FALSE;
+    region_size.NumTiles = 1;
+
+    for (i = 0; i < 2; i++)
+    {
+        const D3D12_TILED_RESOURCE_COORDINATE region_start_coordinate = { i, 0, 0, 0 };
+        const D3D12_TILE_RANGE_FLAGS range_flag = D3D12_TILE_RANGE_FLAG_NONE;
+        const UINT offset = i;
+        const UINT count = 1;
+
+        ID3D12CommandQueue_UpdateTileMappings(context.queue, sparse, 1, &region_start_coordinate, &region_size,
+                i == 0 ? heap : heap_live, 1, &range_flag, &offset, &count, D3D12_TILE_MAPPING_FLAG_NONE);
+    }
+    wait_queue_idle(context.device, context.queue);
+
+    buffer = create_default_buffer(context.device, 128 * 1024,
+            D3D12_RESOURCE_FLAG_ALLOW_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_DEST);
+    cpu = create_cpu_descriptor_heap(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV, 1);
+    gpu = create_gpu_descriptor_heap(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV, 2);
+    memset(&uav_desc, 0, sizeof(uav_desc));
+    uav_desc.ViewDimension = D3D12_UAV_DIMENSION_BUFFER;
+    uav_desc.Format = DXGI_FORMAT_R32_UINT;
+    uav_desc.Buffer.NumElements = 128 * 1024 / 4;
+    uav_desc.Buffer.FirstElement = 0;
+    ID3D12Device_CreateUnorderedAccessView(context.device, sparse, NULL, &uav_desc,
+            ID3D12DescriptorHeap_GetCPUDescriptorHandleForHeapStart(cpu));
+    ID3D12Device_CreateUnorderedAccessView(context.device, sparse, NULL, &uav_desc,
+            ID3D12DescriptorHeap_GetCPUDescriptorHandleForHeapStart(gpu));
+
+    memset(&srv_desc, 0, sizeof(srv_desc));
+    srv_desc.Buffer.FirstElement = 0;
+    srv_desc.Buffer.NumElements = 2 * 1024 * 1024;
+    srv_desc.ViewDimension = D3D12_SRV_DIMENSION_BUFFER;
+    srv_desc.Format = DXGI_FORMAT_R32_UINT;
+    srv_desc.Shader4ComponentMapping = D3D12_DEFAULT_SHADER_4_COMPONENT_MAPPING;
+
+    h_cpu = ID3D12DescriptorHeap_GetCPUDescriptorHandleForHeapStart(gpu);
+    h_cpu.ptr += ID3D12Device_GetDescriptorHandleIncrementSize(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
+    ID3D12Device_CreateShaderResourceView(context.device, sparse, &srv_desc, h_cpu);
+
+    ID3D12GraphicsCommandList_SetDescriptorHeaps(context.list, 1, &gpu);
+    ID3D12GraphicsCommandList_ClearUnorderedAccessViewUint(context.list,
+            ID3D12DescriptorHeap_GetGPUDescriptorHandleForHeapStart(gpu),
+            ID3D12DescriptorHeap_GetCPUDescriptorHandleForHeapStart(cpu), sparse, values, 0, NULL);
+    transition_resource_state(context.list, sparse,
+            D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
+    ID3D12GraphicsCommandList_CopyBufferRegion(context.list, buffer, 0, sparse, 0, 128 * 1024);
+    transition_resource_state(context.list, buffer,
+            D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_COPY_SOURCE);
+    get_buffer_readback_with_command_list(buffer, DXGI_FORMAT_R32_UINT,
+            &rb, context.queue, context.list);
+    reset_command_list(context.list, context.allocator);
+    ok(get_readback_uint(&rb, 0, 0, 0) == 42, "Got #%x, expected 42.\n", get_readback_uint(&rb, 0, 0, 0));
+    ok(get_readback_uint(&rb, 64 * 1024 / 4, 0, 0) == 42, "Got #%x, expected 42.\n", get_readback_uint(&rb, 64 * 1024 / 4, 0, 0));
+    release_resource_readback(&rb);
+
+    ID3D12Heap_Release(heap);
+
+    /* Access a resource where we can hypothetically access the freed heap memory. */
+    /* On AMD Windows native at least, if we read the freed region, we read garbage, which proves it's not required to unbind explicitly.
+     * We'd read 0 in that case. */
+    ID3D12GraphicsCommandList_CopyBufferRegion(context.list, buffer, 0, sparse, 64 * 1024, 64 * 1024);
+
+#define EXPLORE_UNDEFINED_BEHAVIOR 0
+
+#if EXPLORE_UNDEFINED_BEHAVIOR
+    /* This reads unmapped memory. */
+    ID3D12GraphicsCommandList_CopyBufferRegion(context.list, buffer, 1024, sparse, 1024, 1024);
+#endif
+
+    transition_resource_state(context.list, buffer, D3D12_RESOURCE_STATE_COPY_DEST, D3D12_RESOURCE_STATE_UNORDERED_ACCESS);
+
+    h_gpu = ID3D12DescriptorHeap_GetGPUDescriptorHandleForHeapStart(gpu);
+    h_gpu.ptr += ID3D12Device_GetDescriptorHandleIncrementSize(context.device, D3D12_DESCRIPTOR_HEAP_TYPE_CBV_SRV_UAV);
+    ID3D12GraphicsCommandList_SetDescriptorHeaps(context.list, 1, &gpu);
+    ID3D12GraphicsCommandList_SetComputeRootSignature(context.list, context.root_signature);
+    ID3D12GraphicsCommandList_SetPipelineState(context.list, context.pipeline_state);
+    ID3D12GraphicsCommandList_SetComputeRootUnorderedAccessView(context.list, 0, ID3D12Resource_GetGPUVirtualAddress(buffer));
+    ID3D12GraphicsCommandList_SetComputeRootDescriptorTable(context.list, 1, h_gpu);
+#if EXPLORE_UNDEFINED_BEHAVIOR
+    ID3D12GraphicsCommandList_Dispatch(context.list, 1, 1, 1);
+#endif
+
+    transition_resource_state(context.list, buffer, D3D12_RESOURCE_STATE_UNORDERED_ACCESS, D3D12_RESOURCE_STATE_COPY_SOURCE);
+
+    get_buffer_readback_with_command_list(buffer, DXGI_FORMAT_R32_UINT,
+            &rb, context.queue, context.list);
+
+#if EXPLORE_UNDEFINED_BEHAVIOR
+    skip("Reading undefined value #%x.\n", get_readback_uint(&rb, 0, 0, 0));
+    skip("Reading value #%x (expect 0).\n", get_readback_uint(&rb, 1, 0, 0));
+    skip("Reading undefined value #%x.\n", get_readback_uint(&rb, 1024 / 4, 0, 0));
+#endif
+    ok(get_readback_uint(&rb, 2048 / 4, 0, 0) == 42, "Got #%x, expected 42.\n", get_readback_uint(&rb, 2048 / 4, 0, 0));
+    ok(get_readback_uint(&rb, 64 * 1024 / 4, 0, 0) == 42, "Got #%x, expected 42.\n", get_readback_uint(&rb, 64 * 1024 / 4, 0, 0));
+    release_resource_readback(&rb);
+
+    ID3D12Resource_Release(buffer);
+    ID3D12Resource_Release(sparse);
+    ID3D12DescriptorHeap_Release(cpu);
+    ID3D12DescriptorHeap_Release(gpu);
+    ID3D12Heap_Release(heap_live);
+    destroy_test_context(&context);
+}
+

tests/d3d12_sync.c

@@ -1187,3 +1187,247 @@ void test_create_fence(void)
     ok(!refcount, "ID3D12Device has %u references left.\n", (unsigned int)refcount);
 }
 
+void test_fence_wait_robustness_inner(bool shared_handles)
+{
+    VKD3D_UNUSED HANDLE shared_signal = NULL;
+    VKD3D_UNUSED HANDLE shared_drain = NULL;
+    VKD3D_UNUSED HANDLE shared_wait = NULL;
+    ID3D12CommandAllocator *allocator[2];
+    ID3D12Fence *signal_fence_dup = NULL;
+    D3D12_COMMAND_QUEUE_DESC queue_desc;
+    ID3D12Fence *drain_fence_dup = NULL;
+    ID3D12Fence *wait_fence_dup = NULL;
+    ID3D12GraphicsCommandList *list[2];
+    ID3D12CommandQueue *compute_queue;
+    struct test_context context;
+    ID3D12Fence *signal_fence;
+    ID3D12Fence *drain_fence;
+    ID3D12Fence *wait_fence;
+    ID3D12Resource *src;
+    ID3D12Resource *dst;
+    unsigned int i;
+    HANDLE event;
+    UINT value;
+    HRESULT hr;
+
+    if (!init_compute_test_context(&context))
+        return;
+
+    hr = ID3D12Device_CreateFence(context.device, 0,
+            shared_handles ? D3D12_FENCE_FLAG_SHARED : D3D12_FENCE_FLAG_NONE,
+            &IID_ID3D12Fence, (void**)&signal_fence);
+    todo_if(shared_handles) ok(SUCCEEDED(hr), "Failed to create fence, hr #%x.\n", hr);
+
+    if (FAILED(hr))
+    {
+        skip("Failed to create fence, skipping test ...\n");
+        destroy_test_context(&context);
+        return;
+    }
+
+    hr = ID3D12Device_CreateFence(context.device, 0,
+            shared_handles ? D3D12_FENCE_FLAG_SHARED : D3D12_FENCE_FLAG_NONE,
+            &IID_ID3D12Fence, (void**)&wait_fence);
+    ok(SUCCEEDED(hr), "Failed to create fence, hr #%x.\n", hr);
+
+    if (FAILED(hr))
+    {
+        skip("Failed to create fence, skipping test ...\n");
+        ID3D12Fence_Release(signal_fence);
+        destroy_test_context(&context);
+        return;
+    }
+
+    hr = ID3D12Device_CreateFence(context.device, 0,
+            shared_handles ? D3D12_FENCE_FLAG_SHARED : D3D12_FENCE_FLAG_NONE,
+            &IID_ID3D12Fence, (void**)&drain_fence);
+    ok(SUCCEEDED(hr), "Failed to create fence, hr #%x.\n", hr);
+
+    if (FAILED(hr))
+    {
+        skip("Failed to create fence, skipping test ...\n");
+        ID3D12Fence_Release(signal_fence);
+        ID3D12Fence_Release(wait_fence);
+        destroy_test_context(&context);
+        return;
+    }
+
+#ifdef _WIN32
+    if (shared_handles)
+    {
+        hr = ID3D12Device_CreateSharedHandle(context.device, (ID3D12DeviceChild*)signal_fence,
+                NULL, GENERIC_ALL, NULL, &shared_signal);
+        ok(SUCCEEDED(hr), "Failed to create shared handle, hr #%x.\n", hr);
+        hr = ID3D12Device_CreateSharedHandle(context.device, (ID3D12DeviceChild*)wait_fence,
+                NULL, GENERIC_ALL, NULL, &shared_wait);
+        ok(SUCCEEDED(hr), "Failed to create shared handle, hr #%x.\n", hr);
+        hr = ID3D12Device_CreateSharedHandle(context.device, (ID3D12DeviceChild*)drain_fence,
+                NULL, GENERIC_ALL, NULL, &shared_drain);
+        ok(SUCCEEDED(hr), "Failed to create shared handle, hr #%x.\n", hr);
+
+        ID3D12Fence_Release(signal_fence);
+        ID3D12Fence_Release(wait_fence);
+        ID3D12Fence_Release(drain_fence);
+
+        hr = ID3D12Device_OpenSharedHandle(context.device, shared_signal, &IID_ID3D12Fence, (void**)&signal_fence);
+        ok(SUCCEEDED(hr), "Failed to open shared handle, hr #%x.\n", hr);
+        hr = ID3D12Device_OpenSharedHandle(context.device, shared_wait, &IID_ID3D12Fence, (void**)&wait_fence);
+        ok(SUCCEEDED(hr), "Failed to open shared handle, hr #%x.\n", hr);
+        hr = ID3D12Device_OpenSharedHandle(context.device, shared_drain, &IID_ID3D12Fence, (void**)&drain_fence);
+        ok(SUCCEEDED(hr), "Failed to open shared handle, hr #%x.\n", hr);
+
+        /* OpenSharedHandle takes a kernel level reference on the HANDLE. */
+        hr = ID3D12Device_OpenSharedHandle(context.device, shared_signal, &IID_ID3D12Fence, (void**)&signal_fence_dup);
+        ok(SUCCEEDED(hr), "Failed to open shared handle, hr #%x.\n", hr);
+        hr = ID3D12Device_OpenSharedHandle(context.device, shared_wait, &IID_ID3D12Fence, (void**)&wait_fence_dup);
+        ok(SUCCEEDED(hr), "Failed to open shared handle, hr #%x.\n", hr);
+        hr = ID3D12Device_OpenSharedHandle(context.device, shared_drain, &IID_ID3D12Fence, (void**)&drain_fence_dup);
+        ok(SUCCEEDED(hr), "Failed to open shared handle, hr #%x.\n", hr);
+
+        /* Observed behavior: Closing the last reference to the kernel HANDLE object unblocks all waiters.
+         * This isn't really implementable in Wine as it stands since applications are free to share
+         * the HANDLE and Dupe it arbitrarily.
+         * For now, assume this is not a thing, we can report TDR-like situations if this comes up in practice. */
+        if (shared_signal)
+            CloseHandle(shared_signal);
+        if (shared_wait)
+            CloseHandle(shared_wait);
+        if (shared_drain)
+            CloseHandle(shared_drain);
+    }
+#endif
+
+    memset(&queue_desc, 0, sizeof(queue_desc));
+    queue_desc.Flags = D3D12_COMMAND_QUEUE_FLAG_NONE;
+    queue_desc.Priority = D3D12_COMMAND_QUEUE_PRIORITY_NORMAL;
+    queue_desc.Type = D3D12_COMMAND_LIST_TYPE_COMPUTE;
+
+    src = create_default_buffer(context.device, 256 * 1024 * 1024, D3D12_RESOURCE_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_SOURCE);
+    dst = create_default_buffer(context.device, 256 * 1024 * 1024, D3D12_RESOURCE_FLAG_NONE, D3D12_RESOURCE_STATE_COPY_DEST);
+
+    ID3D12Device_CreateCommandQueue(context.device, &queue_desc, &IID_ID3D12CommandQueue, (void**)&compute_queue);
+
+    for (i = 0; i < 2; i++)
+    {
+        ID3D12Device_CreateCommandAllocator(context.device, D3D12_COMMAND_LIST_TYPE_COMPUTE,
+                &IID_ID3D12CommandAllocator, (void**)&allocator[i]);
+        ID3D12Device_CreateCommandList(context.device, 0, D3D12_COMMAND_LIST_TYPE_COMPUTE, allocator[i], NULL,
+                &IID_ID3D12GraphicsCommandList, (void**)&list[i]);
+    }
+
+    /* Heavy copy action. */
+    for (i = 0; i < 128; i++)
+    {
+        ID3D12GraphicsCommandList_CopyResource(list[0], dst, src);
+        ID3D12GraphicsCommandList_CopyResource(list[1], src, dst);
+    }
+
+    ID3D12GraphicsCommandList_Close(list[0]);
+    ID3D12GraphicsCommandList_Close(list[1]);
+
+    /* Note on ref-count checks: The debug layers can take transient public ref-counts it seems. */
+
+    ID3D12CommandQueue_ExecuteCommandLists(context.queue, 1, (ID3D12CommandList * const *)&list[0]);
+    ID3D12CommandQueue_Signal(context.queue, signal_fence, 1);
+    /* Validate that signal/wait does not take public ref-counts. */
+    value = get_refcount(signal_fence);
+    ok(value == 1, "Unexpected ref-count %u\n", value);
+
+    /* The GPU copy is 32 GB worth of BW. There is literally zero chance it would have completed in this amount of time. */
+    value = (UINT)ID3D12Fence_GetCompletedValue(signal_fence);
+    ok(value == 0, "Unexpected signal event %u.\n", value);
+
+    /* Try waiting for a signal that never comes. We'll be able to unblock this wait
+     * when we fully release the fence. */
+    ID3D12CommandQueue_Wait(compute_queue, signal_fence, UINT64_MAX);
+    value = get_refcount(signal_fence);
+    ok(value == 1, "Unexpected ref-count %u\n", value);
+
+    ID3D12CommandQueue_Signal(compute_queue, wait_fence, 1);
+    value = get_refcount(wait_fence);
+    ok(value == 1, "Unexpected ref-count %u\n", value);
+
+    /* The GPU copy is 32 GB worth of BW. There is literally zero chance it would have completed in this amount of time. */
+    value = (UINT)ID3D12Fence_GetCompletedValue(wait_fence);
+    ok(value == 0, "Unexpected signal event %u.\n", value);
+    value = (UINT)ID3D12Fence_GetCompletedValue(signal_fence);
+    ok(value == 0, "Unexpected signal event %u.\n", value);
+
+    ID3D12CommandQueue_Wait(compute_queue, wait_fence, 1);
+    value = get_refcount(wait_fence);
+    ok(value == 1, "Unexpected ref-count %u\n", value);
+
+    /* Check that we can queue up event completion.
+     * Again, verify that releasing the fence unblocks all waiters ... */
+    event = create_event();
+    ID3D12Fence_SetEventOnCompletion(signal_fence, UINT64_MAX, event);
+
+    if (signal_fence_dup)
+        ID3D12Fence_Release(signal_fence_dup);
+    if (wait_fence_dup)
+        ID3D12Fence_Release(wait_fence_dup);
+
+    /* The GPU copy is 32 GB worth of BW. There is literally zero chance it would have completed in this amount of time.
+     * Makes sure that the fences aren't signalled when we try to free them.
+     * (Sure, there is a theoretical race condition if GPU completes between this check and the release, but seriously ...). */
+    value = (UINT)ID3D12Fence_GetCompletedValue(signal_fence);
+    ok(value == 0, "Unexpected signal event %u.\n", value);
+    value = (UINT)ID3D12Fence_GetCompletedValue(wait_fence);
+    ok(value == 0, "Unexpected signal event %u.\n", value);
+
+    /* Test that it's valid to release fence while it's in flight.
+     * If we don't cause device lost and drain_fence is waited on successfully we pass the test. */
+    value = ID3D12Fence_Release(signal_fence);
+    ok(value == 0, "Unexpected fence ref-count %u.\n", value);
+    value = ID3D12Fence_Release(wait_fence);
+    ok(value == 0, "Unexpected fence ref-count %u.\n", value);
+
+    ID3D12CommandQueue_ExecuteCommandLists(compute_queue, 1, (ID3D12CommandList * const *)&list[1]);
+    ID3D12CommandQueue_Signal(compute_queue, drain_fence, 1);
+
+    wait_event(event, INFINITE);
+    destroy_event(event);
+    ID3D12Fence_SetEventOnCompletion(drain_fence, 1, NULL);
+    value = (UINT)ID3D12Fence_GetCompletedValue(drain_fence);
+    ok(value == 1, "Expected fence wait value 1, but got %u.\n", value);
+
+    if (drain_fence_dup)
+    {
+        /* Check we observe the counter in sibling fences as well. */
+        value = (UINT)ID3D12Fence_GetCompletedValue(drain_fence_dup);
+        ok(value == 1, "Expected fence wait value 1, but got %u.\n", value);
+        ID3D12Fence_Release(drain_fence_dup);
+    }
+
+    value = ID3D12Fence_Release(drain_fence);
+    ok(value == 0, "Unexpected fence ref-count %u.\n", value);
+
+    /* Early freeing of fences might signal the drain fence too early, causing GPU hang. */
+    wait_queue_idle(context.device, context.queue);
+    wait_queue_idle(context.device, compute_queue);
+
+    ID3D12CommandQueue_Release(compute_queue);
+    for (i = 0; i < 2; i++)
+    {
+        ID3D12CommandAllocator_Release(allocator[i]);
+        ID3D12GraphicsCommandList_Release(list[i]);
+    }
+    ID3D12Resource_Release(dst);
+    ID3D12Resource_Release(src);
+
+    destroy_test_context(&context);
+}
+
+void test_fence_wait_robustness(void)
+{
+    test_fence_wait_robustness_inner(false);
+}
+
+void test_fence_wait_robustness_shared(void)
+{
+#ifdef _WIN32
+    test_fence_wait_robustness_inner(true);
+#else
+    skip("Shared fences not supported on native Linux build.\n");
+#endif
+}

tests/d3d12_test_utils.c

@@ -27,6 +27,10 @@ PFN_D3D12_GET_DEBUG_INTERFACE pfn_D3D12GetDebugInterface;
 const char *vkd3d_test_platform = "other";
 struct vkd3d_test_state_context vkd3d_test_state;
 
+#ifdef _WIN32
+RENDERDOC_API_1_0_0 *renderdoc_api;
+#endif
+
 bool compare_float(float f, float g, int ulps)
 {
     int x, y;
@@ -861,6 +865,7 @@ ID3D12CommandSignature *create_command_signature_(unsigned int line,
 
 bool init_compute_test_context_(unsigned int line, struct test_context *context)
 {
+    D3D12_COMMAND_LIST_TYPE command_list_type = D3D12_COMMAND_LIST_TYPE_COMPUTE;
     ID3D12Device *device;
     HRESULT hr;
 
@@ -873,14 +878,21 @@ bool init_compute_test_context_(unsigned int line, struct test_context *context)
     }
     device = context->device;
 
-    context->queue = create_command_queue_(line, device,
-            D3D12_COMMAND_LIST_TYPE_COMPUTE, D3D12_COMMAND_QUEUE_PRIORITY_NORMAL);
+#ifdef _WIN32
+    begin_renderdoc_capturing(device);
+    /* Workaround RenderDoc bug. It expects a DIRECT command queue to exist. */
+    if (renderdoc_api)
+        command_list_type = D3D12_COMMAND_LIST_TYPE_DIRECT;
+#endif
 
-    hr = ID3D12Device_CreateCommandAllocator(device, D3D12_COMMAND_LIST_TYPE_COMPUTE,
+    context->queue = create_command_queue_(line, device,
+            command_list_type, D3D12_COMMAND_QUEUE_PRIORITY_NORMAL);
+
+    hr = ID3D12Device_CreateCommandAllocator(device, command_list_type,
             &IID_ID3D12CommandAllocator, (void **)&context->allocator);
     ok_(line)(hr == S_OK, "Failed to create command allocator, hr %#x.\n", hr);
 
-    hr = ID3D12Device_CreateCommandList(device, 0, D3D12_COMMAND_LIST_TYPE_COMPUTE,
+    hr = ID3D12Device_CreateCommandList(device, 0, command_list_type,
             context->allocator, NULL, &IID_ID3D12GraphicsCommandList, (void **)&context->list);
     ok_(line)(hr == S_OK, "Failed to create command list, hr %#x.\n", hr);
 

tests/d3d12_test_utils.h

@@ -19,6 +19,10 @@
 #ifndef __VKD3D_D3D12_TEST_UTILS_H
 #define __VKD3D_D3D12_TEST_UTILS_H
 
+#ifdef _WIN32
+#include "renderdoc_app.h"
+#endif
+
 #define SHADER_BYTECODE(code) {code,sizeof(code)}
 
 #define wait_queue_idle(a, b) wait_queue_idle_(__LINE__, a, b)
@@ -1049,6 +1053,45 @@ static inline void create_render_target_(unsigned int line, struct test_context
         ID3D12Device_CreateRenderTargetView(context->device, *render_target, NULL, *rtv);
 }
 
+/* Utility code for capturing native D3D12 tests, which is why this only covers Win32.
+ * Launch the d3d12.exe test binary from RenderDoc UI.
+ * For Vulkan capturing, use VKD3D_AUTO_CAPTURE_COUNTS and friends instead. */
+#ifdef _WIN32
+extern RENDERDOC_API_1_0_0 *renderdoc_api;
+
+static inline void begin_renderdoc_capturing(ID3D12Device *device)
+{
+    pRENDERDOC_GetAPI get_api;
+    HANDLE renderdoc;
+    FARPROC fn_ptr;
+
+    if (!renderdoc_api)
+    {
+        renderdoc = GetModuleHandleA("renderdoc.dll");
+        if (renderdoc)
+        {
+            fn_ptr = GetProcAddress(renderdoc, "RENDERDOC_GetAPI");
+            if (fn_ptr)
+            {
+                /* Workaround compiler warnings about casting to function pointer. */
+                memcpy(&get_api, &fn_ptr, sizeof(fn_ptr));
+                if (!get_api(eRENDERDOC_API_Version_1_0_0, (void **)&renderdoc_api))
+                    renderdoc_api = NULL;
+            }
+        }
+    }
+
+    if (renderdoc_api)
+        renderdoc_api->StartFrameCapture(device, NULL);
+}
+
+static inline void end_renderdoc_capturing(ID3D12Device *device)
+{
+    if (renderdoc_api)
+        renderdoc_api->EndFrameCapture(device, NULL);
+}
+#endif
+
 #define init_test_context(context, desc) init_test_context_(__LINE__, context, desc)
 static inline bool init_test_context_(unsigned int line, struct test_context *context,
         const struct test_context_desc *desc)
@@ -1066,6 +1109,10 @@ static inline bool init_test_context_(unsigned int line, struct test_context *co
     }
     device = context->device;
 
+#ifdef _WIN32
+    begin_renderdoc_capturing(device);
+#endif
+
     context->queue = create_command_queue_(line, device, D3D12_COMMAND_LIST_TYPE_DIRECT, D3D12_COMMAND_QUEUE_PRIORITY_NORMAL);
 
     hr = ID3D12Device_CreateCommandAllocator(device, D3D12_COMMAND_LIST_TYPE_DIRECT,
@@ -1117,6 +1164,10 @@ static inline void destroy_test_context_(unsigned int line, struct test_context
 {
     ULONG refcount;
 
+#ifdef _WIN32
+    end_renderdoc_capturing(context->device);
+#endif
+
     if (context->pipeline_state)
         ID3D12PipelineState_Release(context->pipeline_state);
     if (context->root_signature)

tests/d3d12_tests.h

@@ -123,6 +123,8 @@ decl_test(test_tgsm);
 decl_test(test_uav_load);
 decl_test(test_cs_uav_store);
 decl_test(test_uav_counters);
+decl_test(test_uav_counter_null_behavior_dxbc);
+decl_test(test_uav_counter_null_behavior_dxil);
 decl_test(test_decrement_uav_counter);
 decl_test(test_atomic_instructions_dxbc);
 decl_test(test_atomic_instructions_dxil);
@@ -135,6 +137,7 @@ decl_test(test_resolve_non_issued_query_data);
 decl_test(test_resolve_query_data_in_different_command_list);
 decl_test(test_resolve_query_data_in_reordered_command_list);
 decl_test(test_execute_indirect);
+decl_test(test_execute_indirect_state);
 decl_test(test_dispatch_zero_thread_groups);
 decl_test(test_unaligned_vertex_stride);
 decl_test(test_zero_vertex_stride);
@@ -294,6 +297,7 @@ decl_test(test_integer_blending_pipeline_state);
 decl_test(test_discard_resource_uav);
 decl_test(test_unbound_rtv_rendering);
 decl_test(test_raytracing_local_rs_static_sampler);
+decl_test(test_raytracing_local_rs_static_sampler_collection);
 decl_test(test_rayquery);
 decl_test(test_typed_srv_uav_cast);
 decl_test(test_typed_srv_cast_clear);
@@ -307,3 +311,13 @@ decl_test(test_shader_waveop_maximal_convergence);
 decl_test(test_uav_3d_sliced_view);
 decl_test(test_pipeline_no_ps_nonzero_rts);
 decl_test(test_root_descriptor_offset_sign);
+decl_test(test_raytracing_no_global_root_signature);
+decl_test(test_raytracing_missing_required_objects);
+decl_test(test_raytracing_reject_duplicate_objects);
+decl_test(test_raytracing_embedded_subobjects);
+decl_test(test_raytracing_default_association_tiebreak);
+decl_test(test_raytracing_collection_identifiers);
+decl_test(test_fence_wait_robustness);
+decl_test(test_fence_wait_robustness_shared);
+decl_test(test_root_signature_empty_blob);
+decl_test(test_sparse_buffer_memory_lifetime);