#include "util/timespec.h"
#include "util/u_atomic.h"
#include "compiler/glsl_types.h"
-#include "util/xmlpool.h"
+#include "util/driconf.h"
static struct radv_timeline_point *
radv_timeline_find_point_at_least_locked(struct radv_device *device,
unsigned type_count = 0;
- if (device->rad_info.has_dedicated_vram) {
- if (vram_index >= 0) {
- device->memory_domains[type_count] = RADEON_DOMAIN_VRAM;
- device->memory_flags[type_count] = RADEON_FLAG_NO_CPU_ACCESS;
- device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
- .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
- .heapIndex = vram_index,
- };
- }
- } else {
- if (visible_vram_index >= 0) {
- device->memory_domains[type_count] = RADEON_DOMAIN_VRAM;
- device->memory_flags[type_count] = RADEON_FLAG_NO_CPU_ACCESS;
- device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
- .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
- .heapIndex = visible_vram_index,
- };
- }
+ if (vram_index >= 0 || visible_vram_index >= 0) {
+ device->memory_domains[type_count] = RADEON_DOMAIN_VRAM;
+ device->memory_flags[type_count] = RADEON_FLAG_NO_CPU_ACCESS;
+ device->memory_properties.memoryTypes[type_count++] = (VkMemoryType) {
+ .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT,
+ .heapIndex = vram_index >= 0 ? vram_index : visible_vram_index,
+ };
}
if (gart_index >= 0) {
static const char *
radv_get_compiler_string(struct radv_physical_device *pdevice)
{
- if (pdevice->use_aco) {
+ if (!pdevice->use_llvm) {
/* Some games like SotTR apply shader workarounds if the LLVM
* version is too old or if the LLVM version string is
* missing. This gives 2-5% performance with SotTR and ACO.
device->local_fd = fd;
device->ws->query_info(device->ws, &device->rad_info);
- device->use_aco = instance->perftest_flags & RADV_PERFTEST_ACO;
+ device->use_llvm = instance->debug_flags & RADV_DEBUG_LLVM;
snprintf(device->name, sizeof(device->name),
"AMD RADV %s (%s)",
}
/* These flags affect shader compilation. */
- uint64_t shader_env_flags = (device->use_aco ? 0x2 : 0);
+ uint64_t shader_env_flags = (device->use_llvm ? 0 : 0x2);
/* The gpu id is already embedded in the uuid so we just pass "radv"
* when creating the cache.
disk_cache_format_hex_id(buf, device->cache_uuid, VK_UUID_SIZE * 2);
device->disk_cache = disk_cache_create(device->name, buf, shader_env_flags);
- if (device->rad_info.chip_class < GFX8)
+ if (device->rad_info.chip_class < GFX8 || !device->use_llvm)
fprintf(stderr, "WARNING: radv is not a conformant vulkan implementation, testing use only.\n");
radv_get_driver_uuid(&device->driver_uuid);
device->dcc_msaa_allowed =
(device->instance->perftest_flags & RADV_PERFTEST_DCC_MSAA);
- device->use_shader_ballot = (device->use_aco && device->rad_info.chip_class >= GFX8) ||
- (device->instance->perftest_flags & RADV_PERFTEST_SHADER_BALLOT);
-
device->use_ngg = device->rad_info.chip_class >= GFX10 &&
device->rad_info.family != CHIP_NAVI14 &&
!(device->instance->debug_flags & RADV_DEBUG_NO_NGG);
/* TODO: Implement NGG GS with ACO. */
- device->use_ngg_gs = device->use_ngg && !device->use_aco;
+ device->use_ngg_gs = device->use_ngg && device->use_llvm;
device->use_ngg_streamout = false;
/* Determine the number of threads per wave for all stages. */
fail_alloc:
vk_free(&instance->alloc, device);
fail_fd:
- close(fd);
+ if (fd != -1)
+ close(fd);
if (master_fd != -1)
close(master_fd);
return result;
{"checkir", RADV_DEBUG_CHECKIR},
{"nothreadllvm", RADV_DEBUG_NOTHREADLLVM},
{"nobinning", RADV_DEBUG_NOBINNING},
- {"noloadstoreopt", RADV_DEBUG_NO_LOAD_STORE_OPT},
{"nongg", RADV_DEBUG_NO_NGG},
- {"noshaderballot", RADV_DEBUG_NO_SHADER_BALLOT},
{"allentrypoints", RADV_DEBUG_ALL_ENTRYPOINTS},
{"metashaders", RADV_DEBUG_DUMP_META_SHADERS},
{"nomemorycache", RADV_DEBUG_NO_MEMORY_CACHE},
+ {"llvm", RADV_DEBUG_LLVM},
{NULL, 0}
};
{"localbos", RADV_PERFTEST_LOCAL_BOS},
{"dccmsaa", RADV_PERFTEST_DCC_MSAA},
{"bolist", RADV_PERFTEST_BO_LIST},
- {"shader_ballot", RADV_PERFTEST_SHADER_BALLOT},
{"tccompatcmask", RADV_PERFTEST_TC_COMPAT_CMASK},
{"cswave32", RADV_PERFTEST_CS_WAVE_32},
{"pswave32", RADV_PERFTEST_PS_WAVE_32},
{"gewave32", RADV_PERFTEST_GE_WAVE_32},
{"dfsm", RADV_PERFTEST_DFSM},
- {"aco", RADV_PERFTEST_ACO},
{NULL, 0}
};
const VkApplicationInfo *info)
{
const char *name = info ? info->pApplicationName : NULL;
-
- if (!name)
- return;
-
- if (!strcmp(name, "DOOM_VFR")) {
- /* Work around a Doom VFR game bug */
- instance->debug_flags |= RADV_DEBUG_NO_DYNAMIC_BOUNDS;
- } else if (!strcmp(name, "MonsterHunterWorld.exe")) {
- /* Workaround for a WaW hazard when LLVM moves/merges
- * load/store memory operations.
- * See https://reviews.llvm.org/D61313
- */
- if (LLVM_VERSION_MAJOR < 9)
- instance->debug_flags |= RADV_DEBUG_NO_LOAD_STORE_OPT;
- } else if (!strcmp(name, "Wolfenstein: Youngblood")) {
- if (!(instance->debug_flags & RADV_DEBUG_NO_SHADER_BALLOT) &&
- !(instance->perftest_flags & RADV_PERFTEST_ACO)) {
- /* Force enable VK_AMD_shader_ballot because it looks
- * safe and it gives a nice boost (+20% on Vega 56 at
- * this time). It also prevents corruption on LLVM.
+ const char *engine_name = info ? info->pEngineName : NULL;
+
+ if (name) {
+ if (!strcmp(name, "DOOM_VFR")) {
+ /* Work around a Doom VFR game bug */
+ instance->debug_flags |= RADV_DEBUG_NO_DYNAMIC_BOUNDS;
+ } else if (!strcmp(name, "Fledge")) {
+ /*
+ * Zero VRAM for "The Surge 2"
+ *
+ * This avoid a hang when when rendering any level. Likely
+ * uninitialized data in an indirect draw.
+ */
+ instance->debug_flags |= RADV_DEBUG_ZERO_VRAM;
+ } else if (!strcmp(name, "No Man's Sky")) {
+ /* Work around a NMS game bug */
+ instance->debug_flags |= RADV_DEBUG_DISCARD_TO_DEMOTE;
+ } else if (!strcmp(name, "DOOMEternal")) {
+ /* Zero VRAM for Doom Eternal to fix rendering issues. */
+ instance->debug_flags |= RADV_DEBUG_ZERO_VRAM;
+ } else if (!strcmp(name, "Red Dead Redemption 2")) {
+ /* Work around a RDR2 game bug */
+ instance->debug_flags |= RADV_DEBUG_DISCARD_TO_DEMOTE;
+ }
+ }
+
+ if (engine_name) {
+ if (!strcmp(engine_name, "vkd3d")) {
+ /* Zero VRAM for all VKD3D (DX12->VK) games to fix
+ * rendering issues.
*/
- instance->perftest_flags |= RADV_PERFTEST_SHADER_BALLOT;
+ instance->debug_flags |= RADV_DEBUG_ZERO_VRAM;
+ } else if (!strcmp(engine_name, "Quantic Dream Engine")) {
+ /* Fix various artifacts in Detroit: Become Human */
+ instance->debug_flags |= RADV_DEBUG_ZERO_VRAM |
+ RADV_DEBUG_DISCARD_TO_DEMOTE;
}
- } else if (!strcmp(name, "Fledge")) {
- /*
- * Zero VRAM for "The Surge 2"
- *
- * This avoid a hang when when rendering any level. Likely
- * uninitialized data in an indirect draw.
- */
- instance->debug_flags |= RADV_DEBUG_ZERO_VRAM;
- } else if (!strcmp(name, "No Man's Sky")) {
- /* Work around a NMS game bug */
- instance->debug_flags |= RADV_DEBUG_DISCARD_TO_DEMOTE;
}
+
+ instance->enable_mrt_output_nan_fixup =
+ driQueryOptionb(&instance->dri_options,
+ "radv_enable_mrt_output_nan_fixup");
+
+ if (driQueryOptionb(&instance->dri_options, "radv_no_dynamic_bounds"))
+ instance->debug_flags |= RADV_DEBUG_NO_DYNAMIC_BOUNDS;
}
static const char radv_dri_options_xml[] =
DRI_CONF_ADAPTIVE_SYNC("true")
DRI_CONF_VK_X11_OVERRIDE_MIN_IMAGE_COUNT(0)
DRI_CONF_VK_X11_STRICT_IMAGE_COUNT("false")
+ DRI_CONF_VK_X11_ENSURE_MIN_IMAGE_COUNT("false")
DRI_CONF_RADV_REPORT_LLVM9_VERSION_STRING("false")
+ DRI_CONF_RADV_ENABLE_MRT_OUTPUT_NAN_FIXUP("false")
+ DRI_CONF_RADV_NO_DYNAMIC_BOUNDS("false")
DRI_CONF_SECTION_END
DRI_CONF_SECTION_DEBUG
if (instance->apiVersion == 0)
instance->apiVersion = VK_API_VERSION_1_0;
- /* Get secure compile thread count. NOTE: We cap this at 32 */
-#define MAX_SC_PROCS 32
- char *num_sc_threads = getenv("RADV_SECURE_COMPILE_THREADS");
- if (num_sc_threads)
- instance->num_sc_threads = MIN2(strtoul(num_sc_threads, NULL, 10), MAX_SC_PROCS);
-
instance->debug_flags = parse_debug_string(getenv("RADV_DEBUG"),
radv_debug_options);
- /* Disable memory cache when secure compile is set */
- if (radv_device_use_secure_compile(instance))
- instance->debug_flags |= RADV_DEBUG_NO_MEMORY_CACHE;
-
instance->perftest_flags = parse_debug_string(getenv("RADV_PERFTEST"),
radv_perftest_options);
- if (instance->perftest_flags & RADV_PERFTEST_ACO)
- fprintf(stderr, "WARNING: Experimental compiler backend enabled. Here be dragons! Incorrect rendering, GPU hangs and/or resets are likely\n");
-
if (instance->debug_flags & RADV_DEBUG_STARTUP)
radv_logi("Created an instance");
if (idx >= RADV_INSTANCE_EXTENSION_COUNT ||
!radv_instance_extensions_supported.extensions[idx]) {
+ vk_object_base_finish(&instance->base);
vk_free2(&default_alloc, pAllocator, instance);
return vk_error(instance, VK_ERROR_EXTENSION_NOT_PRESENT);
}
result = vk_debug_report_instance_init(&instance->debug_report_callbacks);
if (result != VK_SUCCESS) {
+ vk_object_base_finish(&instance->base);
vk_free2(&default_alloc, pAllocator, instance);
return vk_error(instance, result);
}
.shaderCullDistance = true,
.shaderFloat64 = true,
.shaderInt64 = true,
- .shaderInt16 = !pdevice->use_aco || pdevice->rad_info.chip_class >= GFX8,
+ .shaderInt16 = true,
.sparseBinding = true,
.variableMultisampleRate = true,
.shaderResourceMinLod = true,
};
}
+static void
+radv_get_physical_device_features_1_1(struct radv_physical_device *pdevice,
+ VkPhysicalDeviceVulkan11Features *f)
+{
+ assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES);
+
+ f->storageBuffer16BitAccess = true;
+ f->uniformAndStorageBuffer16BitAccess = true;
+ f->storagePushConstant16 = true;
+ f->storageInputOutput16 = pdevice->rad_info.has_packed_math_16bit && (LLVM_VERSION_MAJOR >= 9 || !pdevice->use_llvm);
+ f->multiview = true;
+ f->multiviewGeometryShader = true;
+ f->multiviewTessellationShader = true;
+ f->variablePointersStorageBuffer = true;
+ f->variablePointers = true;
+ f->protectedMemory = false;
+ f->samplerYcbcrConversion = true;
+ f->shaderDrawParameters = true;
+}
+
+static void
+radv_get_physical_device_features_1_2(struct radv_physical_device *pdevice,
+ VkPhysicalDeviceVulkan12Features *f)
+{
+ assert(f->sType == VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES);
+
+ f->samplerMirrorClampToEdge = true;
+ f->drawIndirectCount = true;
+ f->storageBuffer8BitAccess = true;
+ f->uniformAndStorageBuffer8BitAccess = true;
+ f->storagePushConstant8 = true;
+ f->shaderBufferInt64Atomics = LLVM_VERSION_MAJOR >= 9 || !pdevice->use_llvm;
+ f->shaderSharedInt64Atomics = LLVM_VERSION_MAJOR >= 9 || !pdevice->use_llvm;
+ f->shaderFloat16 = pdevice->rad_info.has_packed_math_16bit;
+ f->shaderInt8 = true;
+
+ f->descriptorIndexing = true;
+ f->shaderInputAttachmentArrayDynamicIndexing = true;
+ f->shaderUniformTexelBufferArrayDynamicIndexing = true;
+ f->shaderStorageTexelBufferArrayDynamicIndexing = true;
+ f->shaderUniformBufferArrayNonUniformIndexing = true;
+ f->shaderSampledImageArrayNonUniformIndexing = true;
+ f->shaderStorageBufferArrayNonUniformIndexing = true;
+ f->shaderStorageImageArrayNonUniformIndexing = true;
+ f->shaderInputAttachmentArrayNonUniformIndexing = true;
+ f->shaderUniformTexelBufferArrayNonUniformIndexing = true;
+ f->shaderStorageTexelBufferArrayNonUniformIndexing = true;
+ f->descriptorBindingUniformBufferUpdateAfterBind = true;
+ f->descriptorBindingSampledImageUpdateAfterBind = true;
+ f->descriptorBindingStorageImageUpdateAfterBind = true;
+ f->descriptorBindingStorageBufferUpdateAfterBind = true;
+ f->descriptorBindingUniformTexelBufferUpdateAfterBind = true;
+ f->descriptorBindingStorageTexelBufferUpdateAfterBind = true;
+ f->descriptorBindingUpdateUnusedWhilePending = true;
+ f->descriptorBindingPartiallyBound = true;
+ f->descriptorBindingVariableDescriptorCount = true;
+ f->runtimeDescriptorArray = true;
+
+ f->samplerFilterMinmax = true;
+ f->scalarBlockLayout = pdevice->rad_info.chip_class >= GFX7;
+ f->imagelessFramebuffer = true;
+ f->uniformBufferStandardLayout = true;
+ f->shaderSubgroupExtendedTypes = true;
+ f->separateDepthStencilLayouts = true;
+ f->hostQueryReset = true;
+ f->timelineSemaphore = pdevice->rad_info.has_syncobj_wait_for_submit;
+ f->bufferDeviceAddress = true;
+ f->bufferDeviceAddressCaptureReplay = false;
+ f->bufferDeviceAddressMultiDevice = false;
+ f->vulkanMemoryModel = false;
+ f->vulkanMemoryModelDeviceScope = false;
+ f->vulkanMemoryModelAvailabilityVisibilityChains = false;
+ f->shaderOutputViewportIndex = true;
+ f->shaderOutputLayer = true;
+ f->subgroupBroadcastDynamicId = true;
+}
+
void radv_GetPhysicalDeviceFeatures2(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceFeatures2 *pFeatures)
{
RADV_FROM_HANDLE(radv_physical_device, pdevice, physicalDevice);
+ radv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
+
+ VkPhysicalDeviceVulkan11Features core_1_1 = {
+ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES,
+ };
+ radv_get_physical_device_features_1_1(pdevice, &core_1_1);
+
+ VkPhysicalDeviceVulkan12Features core_1_2 = {
+ .sType = VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES,
+ };
+ radv_get_physical_device_features_1_2(pdevice, &core_1_2);
+
+#define CORE_FEATURE(major, minor, feature) \
+ features->feature = core_##major##_##minor.feature
+
vk_foreach_struct(ext, pFeatures->pNext) {
switch (ext->sType) {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: {
VkPhysicalDeviceVariablePointersFeatures *features = (void *)ext;
- features->variablePointersStorageBuffer = true;
- features->variablePointers = true;
+ CORE_FEATURE(1, 1, variablePointersStorageBuffer);
+ CORE_FEATURE(1, 1, variablePointers);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
VkPhysicalDeviceMultiviewFeatures *features = (VkPhysicalDeviceMultiviewFeatures*)ext;
- features->multiview = true;
- features->multiviewGeometryShader = true;
- features->multiviewTessellationShader = true;
+ CORE_FEATURE(1, 1, multiview);
+ CORE_FEATURE(1, 1, multiviewGeometryShader);
+ CORE_FEATURE(1, 1, multiviewTessellationShader);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: {
VkPhysicalDeviceShaderDrawParametersFeatures *features =
(VkPhysicalDeviceShaderDrawParametersFeatures*)ext;
- features->shaderDrawParameters = true;
+ CORE_FEATURE(1, 1, shaderDrawParameters);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
VkPhysicalDeviceProtectedMemoryFeatures *features =
(VkPhysicalDeviceProtectedMemoryFeatures*)ext;
- features->protectedMemory = false;
+ CORE_FEATURE(1, 1, protectedMemory);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
VkPhysicalDevice16BitStorageFeatures *features =
(VkPhysicalDevice16BitStorageFeatures*)ext;
- bool enable = !pdevice->use_aco || pdevice->rad_info.chip_class >= GFX8;
- features->storageBuffer16BitAccess = enable;
- features->uniformAndStorageBuffer16BitAccess = enable;
- features->storagePushConstant16 = enable;
- features->storageInputOutput16 = pdevice->rad_info.has_double_rate_fp16 && (LLVM_VERSION_MAJOR >= 9 || pdevice->use_aco);
+ CORE_FEATURE(1, 1, storageBuffer16BitAccess);
+ CORE_FEATURE(1, 1, uniformAndStorageBuffer16BitAccess);
+ CORE_FEATURE(1, 1, storagePushConstant16);
+ CORE_FEATURE(1, 1, storageInputOutput16);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: {
VkPhysicalDeviceSamplerYcbcrConversionFeatures *features =
(VkPhysicalDeviceSamplerYcbcrConversionFeatures*)ext;
- features->samplerYcbcrConversion = true;
+ CORE_FEATURE(1, 1, samplerYcbcrConversion);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES: {
VkPhysicalDeviceDescriptorIndexingFeatures *features =
(VkPhysicalDeviceDescriptorIndexingFeatures*)ext;
- features->shaderInputAttachmentArrayDynamicIndexing = true;
- features->shaderUniformTexelBufferArrayDynamicIndexing = true;
- features->shaderStorageTexelBufferArrayDynamicIndexing = true;
- features->shaderUniformBufferArrayNonUniformIndexing = true;
- features->shaderSampledImageArrayNonUniformIndexing = true;
- features->shaderStorageBufferArrayNonUniformIndexing = true;
- features->shaderStorageImageArrayNonUniformIndexing = true;
- features->shaderInputAttachmentArrayNonUniformIndexing = true;
- features->shaderUniformTexelBufferArrayNonUniformIndexing = true;
- features->shaderStorageTexelBufferArrayNonUniformIndexing = true;
- features->descriptorBindingUniformBufferUpdateAfterBind = true;
- features->descriptorBindingSampledImageUpdateAfterBind = true;
- features->descriptorBindingStorageImageUpdateAfterBind = true;
- features->descriptorBindingStorageBufferUpdateAfterBind = true;
- features->descriptorBindingUniformTexelBufferUpdateAfterBind = true;
- features->descriptorBindingStorageTexelBufferUpdateAfterBind = true;
- features->descriptorBindingUpdateUnusedWhilePending = true;
- features->descriptorBindingPartiallyBound = true;
- features->descriptorBindingVariableDescriptorCount = true;
- features->runtimeDescriptorArray = true;
+ CORE_FEATURE(1, 2, shaderInputAttachmentArrayDynamicIndexing);
+ CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayDynamicIndexing);
+ CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayDynamicIndexing);
+ CORE_FEATURE(1, 2, shaderUniformBufferArrayNonUniformIndexing);
+ CORE_FEATURE(1, 2, shaderSampledImageArrayNonUniformIndexing);
+ CORE_FEATURE(1, 2, shaderStorageBufferArrayNonUniformIndexing);
+ CORE_FEATURE(1, 2, shaderStorageImageArrayNonUniformIndexing);
+ CORE_FEATURE(1, 2, shaderInputAttachmentArrayNonUniformIndexing);
+ CORE_FEATURE(1, 2, shaderUniformTexelBufferArrayNonUniformIndexing);
+ CORE_FEATURE(1, 2, shaderStorageTexelBufferArrayNonUniformIndexing);
+ CORE_FEATURE(1, 2, descriptorBindingUniformBufferUpdateAfterBind);
+ CORE_FEATURE(1, 2, descriptorBindingSampledImageUpdateAfterBind);
+ CORE_FEATURE(1, 2, descriptorBindingStorageImageUpdateAfterBind);
+ CORE_FEATURE(1, 2, descriptorBindingStorageBufferUpdateAfterBind);
+ CORE_FEATURE(1, 2, descriptorBindingUniformTexelBufferUpdateAfterBind);
+ CORE_FEATURE(1, 2, descriptorBindingStorageTexelBufferUpdateAfterBind);
+ CORE_FEATURE(1, 2, descriptorBindingUpdateUnusedWhilePending);
+ CORE_FEATURE(1, 2, descriptorBindingPartiallyBound);
+ CORE_FEATURE(1, 2, descriptorBindingVariableDescriptorCount);
+ CORE_FEATURE(1, 2, runtimeDescriptorArray);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SCALAR_BLOCK_LAYOUT_FEATURES: {
VkPhysicalDeviceScalarBlockLayoutFeatures *features =
(VkPhysicalDeviceScalarBlockLayoutFeatures *)ext;
- features->scalarBlockLayout = pdevice->rad_info.chip_class >= GFX7;
+ CORE_FEATURE(1, 2, scalarBlockLayout);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_PRIORITY_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES: {
VkPhysicalDeviceBufferDeviceAddressFeatures *features =
(VkPhysicalDeviceBufferDeviceAddressFeatures *)ext;
- features->bufferDeviceAddress = true;
- features->bufferDeviceAddressCaptureReplay = false;
- features->bufferDeviceAddressMultiDevice = false;
+ CORE_FEATURE(1, 2, bufferDeviceAddress);
+ CORE_FEATURE(1, 2, bufferDeviceAddressCaptureReplay);
+ CORE_FEATURE(1, 2, bufferDeviceAddressMultiDevice);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES: {
VkPhysicalDeviceHostQueryResetFeatures *features =
(VkPhysicalDeviceHostQueryResetFeatures *)ext;
- features->hostQueryReset = true;
+ CORE_FEATURE(1, 2, hostQueryReset);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES: {
VkPhysicalDevice8BitStorageFeatures *features =
(VkPhysicalDevice8BitStorageFeatures *)ext;
- bool enable = !pdevice->use_aco || pdevice->rad_info.chip_class >= GFX8;
- features->storageBuffer8BitAccess = enable;
- features->uniformAndStorageBuffer8BitAccess = enable;
- features->storagePushConstant8 = enable;
+ CORE_FEATURE(1, 2, storageBuffer8BitAccess);
+ CORE_FEATURE(1, 2, uniformAndStorageBuffer8BitAccess);
+ CORE_FEATURE(1, 2, storagePushConstant8);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_FLOAT16_INT8_FEATURES: {
VkPhysicalDeviceShaderFloat16Int8Features *features =
(VkPhysicalDeviceShaderFloat16Int8Features*)ext;
- features->shaderFloat16 = pdevice->rad_info.has_double_rate_fp16 && !pdevice->use_aco;
- features->shaderInt8 = !pdevice->use_aco || pdevice->rad_info.chip_class >= GFX8;
+ CORE_FEATURE(1, 2, shaderFloat16);
+ CORE_FEATURE(1, 2, shaderInt8);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES: {
VkPhysicalDeviceShaderAtomicInt64Features *features =
(VkPhysicalDeviceShaderAtomicInt64Features *)ext;
- features->shaderBufferInt64Atomics = LLVM_VERSION_MAJOR >= 9;
- features->shaderSharedInt64Atomics = LLVM_VERSION_MAJOR >= 9;
+ CORE_FEATURE(1, 2, shaderBufferInt64Atomics);
+ CORE_FEATURE(1, 2, shaderSharedInt64Atomics);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DEMOTE_TO_HELPER_INVOCATION_FEATURES_EXT: {
VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT *features =
(VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT *)ext;
- features->shaderDemoteToHelperInvocation = pdevice->use_aco;
+ features->shaderDemoteToHelperInvocation = LLVM_VERSION_MAJOR >= 9 || !pdevice->use_llvm;
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES: {
VkPhysicalDeviceUniformBufferStandardLayoutFeatures *features =
(VkPhysicalDeviceUniformBufferStandardLayoutFeatures *)ext;
- features->uniformBufferStandardLayout = true;
+ CORE_FEATURE(1, 2, uniformBufferStandardLayout);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGELESS_FRAMEBUFFER_FEATURES: {
VkPhysicalDeviceImagelessFramebufferFeatures *features =
(VkPhysicalDeviceImagelessFramebufferFeatures *)ext;
- features->imagelessFramebuffer = true;
+ CORE_FEATURE(1, 2, imagelessFramebuffer);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PIPELINE_EXECUTABLE_PROPERTIES_FEATURES_KHR: {
VkPhysicalDeviceShaderClockFeaturesKHR *features =
(VkPhysicalDeviceShaderClockFeaturesKHR *)ext;
features->shaderSubgroupClock = true;
- features->shaderDeviceClock = false;
+ features->shaderDeviceClock = pdevice->rad_info.chip_class >= GFX8;
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TIMELINE_SEMAPHORE_FEATURES: {
VkPhysicalDeviceTimelineSemaphoreFeatures *features =
(VkPhysicalDeviceTimelineSemaphoreFeatures *) ext;
- features->timelineSemaphore = true;
+ CORE_FEATURE(1, 2, timelineSemaphore);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_SIZE_CONTROL_FEATURES_EXT: {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_SUBGROUP_EXTENDED_TYPES_FEATURES: {
VkPhysicalDeviceShaderSubgroupExtendedTypesFeatures *features =
(VkPhysicalDeviceShaderSubgroupExtendedTypesFeatures *)ext;
- features->shaderSubgroupExtendedTypes = !pdevice->use_aco;
+ CORE_FEATURE(1, 2, shaderSubgroupExtendedTypes);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SEPARATE_DEPTH_STENCIL_LAYOUTS_FEATURES_KHR: {
VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *features =
(VkPhysicalDeviceSeparateDepthStencilLayoutsFeaturesKHR *)ext;
- features->separateDepthStencilLayouts = true;
+ CORE_FEATURE(1, 2, separateDepthStencilLayouts);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES: {
- VkPhysicalDeviceVulkan11Features *features =
- (VkPhysicalDeviceVulkan11Features *)ext;
- bool storage16_enable = !pdevice->use_aco || pdevice->rad_info.chip_class >= GFX8;
- features->storageBuffer16BitAccess = storage16_enable;
- features->uniformAndStorageBuffer16BitAccess = storage16_enable;
- features->storagePushConstant16 = storage16_enable;
- features->storageInputOutput16 = pdevice->rad_info.has_double_rate_fp16 && (LLVM_VERSION_MAJOR >= 9 || pdevice->use_aco);
- features->multiview = true;
- features->multiviewGeometryShader = true;
- features->multiviewTessellationShader = true;
- features->variablePointersStorageBuffer = true;
- features->variablePointers = true;
- features->protectedMemory = false;
- features->samplerYcbcrConversion = true;
- features->shaderDrawParameters = true;
+ radv_get_physical_device_features_1_1(pdevice, (void *)ext);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_2_FEATURES: {
- VkPhysicalDeviceVulkan12Features *features =
- (VkPhysicalDeviceVulkan12Features *)ext;
- bool int8_enable = !pdevice->use_aco || pdevice->rad_info.chip_class >= GFX8;
- features->samplerMirrorClampToEdge = true;
- features->drawIndirectCount = true;
- features->storageBuffer8BitAccess = int8_enable;
- features->uniformAndStorageBuffer8BitAccess = int8_enable;
- features->storagePushConstant8 = int8_enable;
- features->shaderBufferInt64Atomics = LLVM_VERSION_MAJOR >= 9;
- features->shaderSharedInt64Atomics = LLVM_VERSION_MAJOR >= 9;
- features->shaderFloat16 = pdevice->rad_info.has_double_rate_fp16 && !pdevice->use_aco;
- features->shaderInt8 = int8_enable;
- features->descriptorIndexing = true;
- features->shaderInputAttachmentArrayDynamicIndexing = true;
- features->shaderUniformTexelBufferArrayDynamicIndexing = true;
- features->shaderStorageTexelBufferArrayDynamicIndexing = true;
- features->shaderUniformBufferArrayNonUniformIndexing = true;
- features->shaderSampledImageArrayNonUniformIndexing = true;
- features->shaderStorageBufferArrayNonUniformIndexing = true;
- features->shaderStorageImageArrayNonUniformIndexing = true;
- features->shaderInputAttachmentArrayNonUniformIndexing = true;
- features->shaderUniformTexelBufferArrayNonUniformIndexing = true;
- features->shaderStorageTexelBufferArrayNonUniformIndexing = true;
- features->descriptorBindingUniformBufferUpdateAfterBind = true;
- features->descriptorBindingSampledImageUpdateAfterBind = true;
- features->descriptorBindingStorageImageUpdateAfterBind = true;
- features->descriptorBindingStorageBufferUpdateAfterBind = true;
- features->descriptorBindingUniformTexelBufferUpdateAfterBind = true;
- features->descriptorBindingStorageTexelBufferUpdateAfterBind = true;
- features->descriptorBindingUpdateUnusedWhilePending = true;
- features->descriptorBindingPartiallyBound = true;
- features->descriptorBindingVariableDescriptorCount = true;
- features->runtimeDescriptorArray = true;
- features->samplerFilterMinmax = true;
- features->scalarBlockLayout = pdevice->rad_info.chip_class >= GFX7;
- features->imagelessFramebuffer = true;
- features->uniformBufferStandardLayout = true;
- features->shaderSubgroupExtendedTypes = !pdevice->use_aco;
- features->separateDepthStencilLayouts = true;
- features->hostQueryReset = true;
- features->timelineSemaphore = pdevice->rad_info.has_syncobj_wait_for_submit;
- features->bufferDeviceAddress = true;
- features->bufferDeviceAddressCaptureReplay = false;
- features->bufferDeviceAddressMultiDevice = false;
- features->vulkanMemoryModel = false;
- features->vulkanMemoryModelDeviceScope = false;
- features->vulkanMemoryModelAvailabilityVisibilityChains = false;
- features->shaderOutputViewportIndex = true;
- features->shaderOutputLayer = true;
- features->subgroupBroadcastDynamicId = true;
+ radv_get_physical_device_features_1_2(pdevice, (void *)ext);
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_LINE_RASTERIZATION_FEATURES_EXT: {
features-> pipelineCreationCacheControl = true;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTENDED_DYNAMIC_STATE_FEATURES_EXT: {
+ VkPhysicalDeviceExtendedDynamicStateFeaturesEXT *features =
+ (VkPhysicalDeviceExtendedDynamicStateFeaturesEXT *) ext;
+ features->extendedDynamicState = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_IMAGE_ROBUSTNESS_FEATURES_EXT: {
+ VkPhysicalDeviceImageRobustnessFeaturesEXT *features =
+ (VkPhysicalDeviceImageRobustnessFeaturesEXT *)ext;
+ features->robustImageAccess = true;
+ break;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_FLOAT_FEATURES_EXT: {
+ VkPhysicalDeviceShaderAtomicFloatFeaturesEXT *features =
+ (VkPhysicalDeviceShaderAtomicFloatFeaturesEXT *)ext;
+ features->shaderBufferFloat32Atomics = true;
+ features->shaderBufferFloat32AtomicAdd = false;
+ features->shaderBufferFloat64Atomics = true;
+ features->shaderBufferFloat64AtomicAdd = false;
+ features->shaderSharedFloat32Atomics = true;
+ features->shaderSharedFloat32AtomicAdd = pdevice->rad_info.chip_class >= GFX8 &&
+ (!pdevice->use_llvm || LLVM_VERSION_MAJOR >= 10);
+ features->shaderSharedFloat64Atomics = true;
+ features->shaderSharedFloat64AtomicAdd = false;
+ features->shaderImageFloat32Atomics = true;
+ features->shaderImageFloat32AtomicAdd = false;
+ features->sparseImageFloat32Atomics = false;
+ features->sparseImageFloat32AtomicAdd = false;
+ break;
+ }
default:
break;
}
}
- radv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
+#undef CORE_FEATURE
}
static size_t
.maxCullDistances = 8,
.maxCombinedClipAndCullDistances = 8,
.discreteQueuePriorities = 2,
- .pointSizeRange = { 0.0, 8192.0 },
- .lineWidthRange = { 0.0, 8192.0 },
+ .pointSizeRange = { 0.0, 8191.875 },
+ .lineWidthRange = { 0.0, 8191.875 },
.pointSizeGranularity = (1.0 / 8.0),
.lineWidthGranularity = (1.0 / 8.0),
.strictLines = false, /* FINISHME */
VK_SUBGROUP_FEATURE_ARITHMETIC_BIT |
VK_SUBGROUP_FEATURE_BALLOT_BIT |
VK_SUBGROUP_FEATURE_CLUSTERED_BIT |
- VK_SUBGROUP_FEATURE_QUAD_BIT;
-
- if (((pdevice->rad_info.chip_class == GFX6 ||
- pdevice->rad_info.chip_class == GFX7) && !pdevice->use_aco) ||
- pdevice->rad_info.chip_class >= GFX8) {
- p->subgroupSupportedOperations |= VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
- VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT;
- }
+ VK_SUBGROUP_FEATURE_QUAD_BIT |
+ VK_SUBGROUP_FEATURE_SHUFFLE_BIT |
+ VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT;
p->subgroupQuadOperationsInAllStages = true;
p->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES;
/* On AMD hardware, denormals and rounding modes for fp16/fp64 are
* controlled by the same config register.
*/
- if (pdevice->rad_info.has_double_rate_fp16) {
+ if (pdevice->rad_info.has_packed_math_16bit) {
p->denormBehaviorIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY_KHR;
p->roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_32_BIT_ONLY_KHR;
} else {
p->roundingModeIndependence = VK_SHADER_FLOAT_CONTROLS_INDEPENDENCE_ALL_KHR;
}
- /* Do not allow both preserving and flushing denorms because different
- * shaders in the same pipeline can have different settings and this
- * won't work for merged shaders. To make it work, this requires LLVM
+ /* With LLVM, do not allow both preserving and flushing denorms because
+ * different shaders in the same pipeline can have different settings and
+ * this won't work for merged shaders. To make it work, this requires LLVM
* support for changing the register. The same logic applies for the
* rounding modes because they are configured with the same config
- * register. TODO: we can enable a lot of these for ACO when it
- * supports all stages.
+ * register.
*/
p->shaderDenormFlushToZeroFloat32 = true;
- p->shaderDenormPreserveFloat32 = false;
+ p->shaderDenormPreserveFloat32 = !pdevice->use_llvm;
p->shaderRoundingModeRTEFloat32 = true;
- p->shaderRoundingModeRTZFloat32 = false;
+ p->shaderRoundingModeRTZFloat32 = !pdevice->use_llvm;
p->shaderSignedZeroInfNanPreserveFloat32 = true;
- p->shaderDenormFlushToZeroFloat16 = false;
- p->shaderDenormPreserveFloat16 = pdevice->rad_info.has_double_rate_fp16;
- p->shaderRoundingModeRTEFloat16 = pdevice->rad_info.has_double_rate_fp16;
- p->shaderRoundingModeRTZFloat16 = false;
- p->shaderSignedZeroInfNanPreserveFloat16 = pdevice->rad_info.has_double_rate_fp16;
+ p->shaderDenormFlushToZeroFloat16 = pdevice->rad_info.has_packed_math_16bit && !pdevice->use_llvm;
+ p->shaderDenormPreserveFloat16 = pdevice->rad_info.has_packed_math_16bit;
+ p->shaderRoundingModeRTEFloat16 = pdevice->rad_info.has_packed_math_16bit;
+ p->shaderRoundingModeRTZFloat16 = pdevice->rad_info.has_packed_math_16bit && !pdevice->use_llvm;
+ p->shaderSignedZeroInfNanPreserveFloat16 = pdevice->rad_info.has_packed_math_16bit;
- p->shaderDenormFlushToZeroFloat64 = false;
+ p->shaderDenormFlushToZeroFloat64 = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_llvm;
p->shaderDenormPreserveFloat64 = pdevice->rad_info.chip_class >= GFX8;
p->shaderRoundingModeRTEFloat64 = pdevice->rad_info.chip_class >= GFX8;
- p->shaderRoundingModeRTZFloat64 = false;
+ p->shaderRoundingModeRTZFloat64 = pdevice->rad_info.chip_class >= GFX8 && !pdevice->use_llvm;
p->shaderSignedZeroInfNanPreserveFloat64 = pdevice->rad_info.chip_class >= GFX8;
p->maxUpdateAfterBindDescriptorsInAllPools = UINT32_MAX / 64;
properties->shaderArraysPerEngineCount =
pdevice->rad_info.max_sh_per_se;
properties->computeUnitsPerShaderArray =
- pdevice->rad_info.num_good_cu_per_sh;
+ pdevice->rad_info.min_good_cu_per_sa;
properties->simdPerComputeUnit =
pdevice->rad_info.num_simd_per_compute_unit;
properties->wavefrontsPerSimd =
return result;
}
-static int install_seccomp_filter() {
-
- struct sock_filter filter[] = {
- /* Check arch is 64bit x86 */
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, arch))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, AUDIT_ARCH_X86_64, 0, 12),
-
- /* Futex is required for mutex locks */
- #if defined __NR__newselect
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR__newselect, 11, 0),
- #elif defined __NR_select
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR_select, 11, 0),
- #else
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR_pselect6, 11, 0),
- #endif
-
- /* Allow system exit calls for the forked process */
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR_exit_group, 9, 0),
-
- /* Allow system read calls */
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR_read, 7, 0),
-
- /* Allow system write calls */
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR_write, 5, 0),
-
- /* Allow system brk calls (we need this for malloc) */
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR_brk, 3, 0),
-
- /* Futex is required for mutex locks */
- BPF_STMT(BPF_LD + BPF_W + BPF_ABS, (offsetof(struct seccomp_data, nr))),
- BPF_JUMP(BPF_JMP + BPF_JEQ + BPF_K, __NR_futex, 1, 0),
-
- /* Return error if we hit a system call not on the whitelist */
- BPF_STMT(BPF_RET + BPF_K, SECCOMP_RET_ERRNO | (EPERM & SECCOMP_RET_DATA)),
-
- /* Allow whitelisted system calls */
- BPF_STMT(BPF_RET + BPF_K, SECCOMP_RET_ALLOW),
- };
-
- struct sock_fprog prog = {
- .len = (unsigned short)(sizeof(filter) / sizeof(filter[0])),
- .filter = filter,
- };
-
- if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0))
- return -1;
-
- if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog))
- return -1;
-
- return 0;
-}
-
-/* Helper function with timeout support for reading from the pipe between
- * processes used for secure compile.
- */
-bool radv_sc_read(int fd, void *buf, size_t size, bool timeout)
-{
- fd_set fds;
- struct timeval tv;
-
- FD_ZERO(&fds);
- FD_SET(fd, &fds);
-
- while (true) {
- /* We can't rely on the value of tv after calling select() so
- * we must reset it on each iteration of the loop.
- */
- tv.tv_sec = 5;
- tv.tv_usec = 0;
-
- int rval = select(fd + 1, &fds, NULL, NULL, timeout ? &tv : NULL);
-
- if (rval == -1) {
- /* select error */
- return false;
- } else if (rval) {
- ssize_t bytes_read = read(fd, buf, size);
- if (bytes_read < 0)
- return false;
-
- buf += bytes_read;
- size -= bytes_read;
- if (size == 0)
- return true;
- } else {
- /* select timeout */
- return false;
- }
- }
-}
-
-static bool radv_close_all_fds(const int *keep_fds, int keep_fd_count)
-{
- DIR *d;
- struct dirent *dir;
- d = opendir("/proc/self/fd");
- if (!d)
- return false;
- int dir_fd = dirfd(d);
-
- while ((dir = readdir(d)) != NULL) {
- if (dir->d_name[0] == '.')
- continue;
-
- int fd = atoi(dir->d_name);
- if (fd == dir_fd)
- continue;
-
- bool keep = false;
- for (int i = 0; !keep && i < keep_fd_count; ++i)
- if (keep_fds[i] == fd)
- keep = true;
-
- if (keep)
- continue;
-
- close(fd);
- }
- closedir(d);
- return true;
-}
-
-static bool secure_compile_open_fifo_fds(struct radv_secure_compile_state *sc,
- int *fd_server, int *fd_client,
- unsigned process, bool make_fifo)
-{
- bool result = false;
- char *fifo_server_path = NULL;
- char *fifo_client_path = NULL;
-
- if (asprintf(&fifo_server_path, "/tmp/radv_server_%s_%u", sc->uid, process) == -1)
- goto open_fifo_exit;
-
- if (asprintf(&fifo_client_path, "/tmp/radv_client_%s_%u", sc->uid, process) == -1)
- goto open_fifo_exit;
-
- if (make_fifo) {
- int file1 = mkfifo(fifo_server_path, 0666);
- if(file1 < 0)
- goto open_fifo_exit;
-
- int file2 = mkfifo(fifo_client_path, 0666);
- if(file2 < 0)
- goto open_fifo_exit;
- }
-
- *fd_server = open(fifo_server_path, O_RDWR);
- if(*fd_server < 1)
- goto open_fifo_exit;
-
- *fd_client = open(fifo_client_path, O_RDWR);
- if(*fd_client < 1) {
- close(*fd_server);
- goto open_fifo_exit;
- }
-
- result = true;
-
-open_fifo_exit:
- free(fifo_server_path);
- free(fifo_client_path);
-
- return result;
-}
-
-static void run_secure_compile_device(struct radv_device *device, unsigned process,
- int fd_idle_device_output)
-{
- int fd_secure_input;
- int fd_secure_output;
- bool fifo_result = secure_compile_open_fifo_fds(device->sc_state,
- &fd_secure_input,
- &fd_secure_output,
- process, false);
-
- enum radv_secure_compile_type sc_type;
-
- const int needed_fds[] = {
- fd_secure_input,
- fd_secure_output,
- fd_idle_device_output,
- };
-
- if (!fifo_result || !radv_close_all_fds(needed_fds, ARRAY_SIZE(needed_fds)) ||
- install_seccomp_filter() == -1) {
- sc_type = RADV_SC_TYPE_INIT_FAILURE;
- } else {
- sc_type = RADV_SC_TYPE_INIT_SUCCESS;
- device->sc_state->secure_compile_processes[process].fd_secure_input = fd_secure_input;
- device->sc_state->secure_compile_processes[process].fd_secure_output = fd_secure_output;
- }
-
- write(fd_idle_device_output, &sc_type, sizeof(sc_type));
-
- if (sc_type == RADV_SC_TYPE_INIT_FAILURE)
- goto secure_compile_exit;
-
- while (true) {
- radv_sc_read(fd_secure_input, &sc_type, sizeof(sc_type), false);
-
- if (sc_type == RADV_SC_TYPE_COMPILE_PIPELINE) {
- struct radv_pipeline *pipeline;
- bool sc_read = true;
-
- pipeline = vk_zalloc2(&device->vk.alloc, NULL, sizeof(*pipeline), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
-
- pipeline->device = device;
-
- /* Read pipeline layout */
- struct radv_pipeline_layout layout;
- sc_read = radv_sc_read(fd_secure_input, &layout, sizeof(struct radv_pipeline_layout), true);
- sc_read &= radv_sc_read(fd_secure_input, &layout.num_sets, sizeof(uint32_t), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- for (uint32_t set = 0; set < layout.num_sets; set++) {
- uint32_t layout_size;
- sc_read &= radv_sc_read(fd_secure_input, &layout_size, sizeof(uint32_t), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- layout.set[set].layout = malloc(layout_size);
- layout.set[set].layout->layout_size = layout_size;
- sc_read &= radv_sc_read(fd_secure_input, layout.set[set].layout,
- layout.set[set].layout->layout_size, true);
- }
-
- pipeline->layout = &layout;
-
- /* Read pipeline key */
- struct radv_pipeline_key key;
- sc_read &= radv_sc_read(fd_secure_input, &key, sizeof(struct radv_pipeline_key), true);
-
- /* Read pipeline create flags */
- VkPipelineCreateFlags flags;
- sc_read &= radv_sc_read(fd_secure_input, &flags, sizeof(VkPipelineCreateFlags), true);
-
- /* Read stage and shader information */
- uint32_t num_stages;
- const VkPipelineShaderStageCreateInfo *pStages[MESA_SHADER_STAGES] = { 0, };
- sc_read &= radv_sc_read(fd_secure_input, &num_stages, sizeof(uint32_t), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- for (uint32_t i = 0; i < num_stages; i++) {
-
- /* Read stage */
- gl_shader_stage stage;
- sc_read &= radv_sc_read(fd_secure_input, &stage, sizeof(gl_shader_stage), true);
-
- VkPipelineShaderStageCreateInfo *pStage = calloc(1, sizeof(VkPipelineShaderStageCreateInfo));
-
- /* Read entry point name */
- size_t name_size;
- sc_read &= radv_sc_read(fd_secure_input, &name_size, sizeof(size_t), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- char *ep_name = malloc(name_size);
- sc_read &= radv_sc_read(fd_secure_input, ep_name, name_size, true);
- pStage->pName = ep_name;
-
- /* Read shader module */
- size_t module_size;
- sc_read &= radv_sc_read(fd_secure_input, &module_size, sizeof(size_t), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- struct radv_shader_module *module = malloc(module_size);
- sc_read &= radv_sc_read(fd_secure_input, module, module_size, true);
- pStage->module = radv_shader_module_to_handle(module);
-
- /* Read specialization info */
- bool has_spec_info;
- sc_read &= radv_sc_read(fd_secure_input, &has_spec_info, sizeof(bool), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- if (has_spec_info) {
- VkSpecializationInfo *specInfo = malloc(sizeof(VkSpecializationInfo));
- pStage->pSpecializationInfo = specInfo;
-
- sc_read &= radv_sc_read(fd_secure_input, &specInfo->dataSize, sizeof(size_t), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- void *si_data = malloc(specInfo->dataSize);
- sc_read &= radv_sc_read(fd_secure_input, si_data, specInfo->dataSize, true);
- specInfo->pData = si_data;
-
- sc_read &= radv_sc_read(fd_secure_input, &specInfo->mapEntryCount, sizeof(uint32_t), true);
- if (!sc_read)
- goto secure_compile_exit;
-
- VkSpecializationMapEntry *mapEntries = malloc(sizeof(VkSpecializationMapEntry) * specInfo->mapEntryCount);
- for (uint32_t j = 0; j < specInfo->mapEntryCount; j++) {
- sc_read &= radv_sc_read(fd_secure_input, &mapEntries[j], sizeof(VkSpecializationMapEntry), true);
- if (!sc_read)
- goto secure_compile_exit;
- }
-
- specInfo->pMapEntries = mapEntries;
- }
-
- pStages[stage] = pStage;
- }
-
- /* Compile the shaders */
- VkPipelineCreationFeedbackEXT *stage_feedbacks[MESA_SHADER_STAGES] = { 0 };
-
- /* Not fully to spec but if we're doing sandboxed compilations already this doesn't matter. */
- flags &= ~VK_PIPELINE_CREATE_FAIL_ON_PIPELINE_COMPILE_REQUIRED_BIT_EXT;
-
- radv_create_shaders(pipeline, device, NULL, &key, pStages, flags, NULL, stage_feedbacks);
-
- /* free memory allocated above */
- for (uint32_t set = 0; set < layout.num_sets; set++)
- free(layout.set[set].layout);
-
- for (uint32_t i = 0; i < MESA_SHADER_STAGES; i++) {
- if (!pStages[i])
- continue;
-
- free((void *) pStages[i]->pName);
- free(radv_shader_module_from_handle(pStages[i]->module));
- if (pStages[i]->pSpecializationInfo) {
- free((void *) pStages[i]->pSpecializationInfo->pData);
- free((void *) pStages[i]->pSpecializationInfo->pMapEntries);
- free((void *) pStages[i]->pSpecializationInfo);
- }
- free((void *) pStages[i]);
- }
-
- vk_free(&device->vk.alloc, pipeline);
-
- sc_type = RADV_SC_TYPE_COMPILE_PIPELINE_FINISHED;
- write(fd_secure_output, &sc_type, sizeof(sc_type));
-
- } else if (sc_type == RADV_SC_TYPE_DESTROY_DEVICE) {
- goto secure_compile_exit;
- }
- }
-
-secure_compile_exit:
- close(fd_secure_input);
- close(fd_secure_output);
- close(fd_idle_device_output);
- _exit(0);
-}
-
-static enum radv_secure_compile_type fork_secure_compile_device(struct radv_device *device, unsigned process)
-{
- int fd_secure_input[2];
- int fd_secure_output[2];
-
- /* create pipe descriptors (used to communicate between processes) */
- if (pipe(fd_secure_input) == -1 || pipe(fd_secure_output) == -1)
- return RADV_SC_TYPE_INIT_FAILURE;
-
-
- int sc_pid;
- if ((sc_pid = fork()) == 0) {
- device->sc_state->secure_compile_thread_counter = process;
- run_secure_compile_device(device, process, fd_secure_output[1]);
- } else {
- if (sc_pid == -1)
- return RADV_SC_TYPE_INIT_FAILURE;
-
- /* Read the init result returned from the secure process */
- enum radv_secure_compile_type sc_type;
- bool sc_read = radv_sc_read(fd_secure_output[0], &sc_type, sizeof(sc_type), true);
-
- if (sc_type == RADV_SC_TYPE_INIT_FAILURE || !sc_read) {
- close(fd_secure_input[0]);
- close(fd_secure_input[1]);
- close(fd_secure_output[1]);
- close(fd_secure_output[0]);
- int status;
- waitpid(sc_pid, &status, 0);
-
- return RADV_SC_TYPE_INIT_FAILURE;
- } else {
- assert(sc_type == RADV_SC_TYPE_INIT_SUCCESS);
- write(device->sc_state->secure_compile_processes[process].fd_secure_output, &sc_type, sizeof(sc_type));
-
- close(fd_secure_input[0]);
- close(fd_secure_input[1]);
- close(fd_secure_output[1]);
- close(fd_secure_output[0]);
-
- int status;
- waitpid(sc_pid, &status, 0);
- }
- }
-
- return RADV_SC_TYPE_INIT_SUCCESS;
-}
-
-/* Run a bare bones fork of a device that was forked right after its creation.
- * This device will have low overhead when it is forked again before each
- * pipeline compilation. This device sits idle and its only job is to fork
- * itself.
- */
-static void run_secure_compile_idle_device(struct radv_device *device, unsigned process,
- int fd_secure_input, int fd_secure_output)
-{
- enum radv_secure_compile_type sc_type = RADV_SC_TYPE_INIT_SUCCESS;
- device->sc_state->secure_compile_processes[process].fd_secure_input = fd_secure_input;
- device->sc_state->secure_compile_processes[process].fd_secure_output = fd_secure_output;
-
- write(fd_secure_output, &sc_type, sizeof(sc_type));
-
- while (true) {
- radv_sc_read(fd_secure_input, &sc_type, sizeof(sc_type), false);
-
- if (sc_type == RADV_SC_TYPE_FORK_DEVICE) {
- sc_type = fork_secure_compile_device(device, process);
-
- if (sc_type == RADV_SC_TYPE_INIT_FAILURE)
- goto secure_compile_exit;
-
- } else if (sc_type == RADV_SC_TYPE_DESTROY_DEVICE) {
- goto secure_compile_exit;
- }
- }
-
-secure_compile_exit:
- close(fd_secure_input);
- close(fd_secure_output);
- _exit(0);
-}
-
-static void destroy_secure_compile_device(struct radv_device *device, unsigned process)
-{
- int fd_secure_input = device->sc_state->secure_compile_processes[process].fd_secure_input;
-
- enum radv_secure_compile_type sc_type = RADV_SC_TYPE_DESTROY_DEVICE;
- write(fd_secure_input, &sc_type, sizeof(sc_type));
-
- close(device->sc_state->secure_compile_processes[process].fd_secure_input);
- close(device->sc_state->secure_compile_processes[process].fd_secure_output);
-
- int status;
- waitpid(device->sc_state->secure_compile_processes[process].sc_pid, &status, 0);
-}
-
-static VkResult fork_secure_compile_idle_device(struct radv_device *device)
-{
- device->sc_state = vk_zalloc(&device->vk.alloc,
- sizeof(struct radv_secure_compile_state),
- 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
-
- mtx_init(&device->sc_state->secure_compile_mutex, mtx_plain);
-
- pid_t upid = getpid();
- time_t seconds = time(NULL);
-
- char *uid;
- if (asprintf(&uid, "%ld_%ld", (long) upid, (long) seconds) == -1)
- return VK_ERROR_INITIALIZATION_FAILED;
-
- device->sc_state->uid = uid;
-
- uint8_t sc_threads = device->instance->num_sc_threads;
- int fd_secure_input[MAX_SC_PROCS][2];
- int fd_secure_output[MAX_SC_PROCS][2];
-
- /* create pipe descriptors (used to communicate between processes) */
- for (unsigned i = 0; i < sc_threads; i++) {
- if (pipe(fd_secure_input[i]) == -1 ||
- pipe(fd_secure_output[i]) == -1) {
- return VK_ERROR_INITIALIZATION_FAILED;
- }
- }
-
- device->sc_state->secure_compile_processes = vk_zalloc(&device->vk.alloc,
- sizeof(struct radv_secure_compile_process) * sc_threads, 8,
- VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
-
- for (unsigned process = 0; process < sc_threads; process++) {
- if ((device->sc_state->secure_compile_processes[process].sc_pid = fork()) == 0) {
- device->sc_state->secure_compile_thread_counter = process;
- run_secure_compile_idle_device(device, process, fd_secure_input[process][0], fd_secure_output[process][1]);
- } else {
- if (device->sc_state->secure_compile_processes[process].sc_pid == -1)
- return VK_ERROR_INITIALIZATION_FAILED;
-
- /* Read the init result returned from the secure process */
- enum radv_secure_compile_type sc_type;
- bool sc_read = radv_sc_read(fd_secure_output[process][0], &sc_type, sizeof(sc_type), true);
-
- bool fifo_result;
- if (sc_read && sc_type == RADV_SC_TYPE_INIT_SUCCESS) {
- fifo_result = secure_compile_open_fifo_fds(device->sc_state,
- &device->sc_state->secure_compile_processes[process].fd_server,
- &device->sc_state->secure_compile_processes[process].fd_client,
- process, true);
-
- device->sc_state->secure_compile_processes[process].fd_secure_input = fd_secure_input[process][1];
- device->sc_state->secure_compile_processes[process].fd_secure_output = fd_secure_output[process][0];
- }
-
- if (sc_type == RADV_SC_TYPE_INIT_FAILURE || !sc_read || !fifo_result) {
- close(fd_secure_input[process][0]);
- close(fd_secure_input[process][1]);
- close(fd_secure_output[process][1]);
- close(fd_secure_output[process][0]);
- int status;
- waitpid(device->sc_state->secure_compile_processes[process].sc_pid, &status, 0);
-
- /* Destroy any forks that were created sucessfully */
- for (unsigned i = 0; i < process; i++) {
- destroy_secure_compile_device(device, i);
- }
-
- return VK_ERROR_INITIALIZATION_FAILED;
- }
- }
- }
- return VK_SUCCESS;
-}
-
static void
radv_device_init_dispatch(struct radv_device *device)
{
device->border_color_data.colors_gpu_ptr =
device->ws->buffer_map(device->border_color_data.bo);
+ if (!device->border_color_data.colors_gpu_ptr)
+ return vk_error(device->physical_device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
pthread_mutex_init(&device->border_color_data.mutex, NULL);
return VK_SUCCESS;
goto fail;
}
- /* Temporarily disable secure compile while we create meta shaders, etc */
- uint8_t sc_threads = device->instance->num_sc_threads;
- if (sc_threads)
- device->instance->num_sc_threads = 0;
-
device->keep_shader_info = keep_shader_info;
result = radv_device_init_meta(device);
if (result != VK_SUCCESS)
for (int family = 0; family < RADV_MAX_QUEUE_FAMILIES; ++family) {
device->empty_cs[family] = device->ws->cs_create(device->ws, family);
+ if (!device->empty_cs[family])
+ goto fail;
+
switch (family) {
case RADV_QUEUE_GENERAL:
radeon_emit(device->empty_cs[family], PKT3(PKT3_CONTEXT_CONTROL, 1, 0));
radeon_emit(device->empty_cs[family], 0);
break;
}
- device->ws->cs_finalize(device->empty_cs[family]);
+
+ result = device->ws->cs_finalize(device->empty_cs[family]);
+ if (result != VK_SUCCESS)
+ goto fail;
}
if (device->physical_device->rad_info.chip_class >= GFX7)
1 << util_logbase2(device->force_aniso));
}
- /* Fork device for secure compile as required */
- device->instance->num_sc_threads = sc_threads;
- if (radv_device_use_secure_compile(device->instance)) {
-
- result = fork_secure_compile_idle_device(device);
- if (result != VK_SUCCESS)
- goto fail_meta;
- }
-
*pDevice = radv_device_to_handle(device);
return VK_SUCCESS;
radv_thread_trace_finish(device);
- if (radv_device_use_secure_compile(device->instance)) {
- for (unsigned i = 0; i < device->instance->num_sc_threads; i++ ) {
- destroy_secure_compile_device(device, i);
- }
- }
-
- if (device->sc_state) {
- free(device->sc_state->uid);
- vk_free(&device->vk.alloc, device->sc_state->secure_compile_processes);
- }
- vk_free(&device->vk.alloc, device->sc_state);
vk_free(&device->vk.alloc, device);
}
}
*max_offchip_buffers_p = max_offchip_buffers;
- if (device->physical_device->rad_info.chip_class >= GFX7) {
+ if (device->physical_device->rad_info.chip_class >= GFX10_3) {
+ hs_offchip_param = S_03093C_OFFCHIP_BUFFERING_GFX103(max_offchip_buffers - 1) |
+ S_03093C_OFFCHIP_GRANULARITY_GFX103(offchip_granularity);
+ } else if (device->physical_device->rad_info.chip_class >= GFX7) {
if (device->physical_device->rad_info.chip_class >= GFX8)
--max_offchip_buffers;
hs_offchip_param =
if (descriptor_bo != queue->descriptor_bo) {
uint32_t *map = (uint32_t*)queue->device->ws->buffer_map(descriptor_bo);
+ if (!map)
+ goto fail;
if (scratch_bo) {
uint64_t scratch_va = radv_buffer_get_va(scratch_bo);
if (queue->device->trace_bo)
radv_cs_add_buffer(queue->device->ws, cs, queue->device->trace_bo);
+ if (queue->device->border_color_data.bo)
+ radv_cs_add_buffer(queue->device->ws, cs,
+ queue->device->border_color_data.bo);
+
if (i == 0) {
si_cs_emit_cache_flush(cs,
queue->device->physical_device->rad_info.chip_class,
RADV_CMD_FLAG_START_PIPELINE_STATS, 0);
}
- if (!queue->device->ws->cs_finalize(cs))
+ if (queue->device->ws->cs_finalize(cs) != VK_SUCCESS)
goto fail;
}
VkFence _fence,
bool is_signal)
{
- int syncobj_idx = 0, sem_idx = 0;
+ int syncobj_idx = 0, non_reset_idx = 0, sem_idx = 0;
if (num_sems == 0 && _fence == VK_NULL_HANDLE)
return VK_SUCCESS;
switch(sems[i]->kind) {
case RADV_SEMAPHORE_SYNCOBJ:
counts->syncobj_count++;
+ counts->syncobj_reset_count++;
break;
case RADV_SEMAPHORE_WINSYS:
counts->sem_count++;
if (_fence != VK_NULL_HANDLE) {
RADV_FROM_HANDLE(radv_fence, fence, _fence);
- if (fence->temp_syncobj || fence->syncobj)
+
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
+ if (part->kind == RADV_FENCE_SYNCOBJ)
counts->syncobj_count++;
}
}
}
+ non_reset_idx = counts->syncobj_reset_count;
+
for (uint32_t i = 0; i < num_sems; i++) {
switch(sems[i]->kind) {
case RADV_SEMAPHORE_NONE:
pthread_mutex_unlock(&sems[i]->timeline.mutex);
if (point) {
- counts->syncobj[syncobj_idx++] = point->syncobj;
+ counts->syncobj[non_reset_idx++] = point->syncobj;
} else {
/* Explicitly remove the semaphore so we might not find
* a point later post-submit. */
if (_fence != VK_NULL_HANDLE) {
RADV_FROM_HANDLE(radv_fence, fence, _fence);
- if (fence->temp_syncobj)
- counts->syncobj[syncobj_idx++] = fence->temp_syncobj;
- else if (fence->syncobj)
- counts->syncobj[syncobj_idx++] = fence->syncobj;
+
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
+ if (part->kind == RADV_FENCE_SYNCOBJ)
+ counts->syncobj[non_reset_idx++] = part->syncobj;
}
- assert(syncobj_idx <= counts->syncobj_count);
- counts->syncobj_count = syncobj_idx;
+ assert(MAX2(syncobj_idx, non_reset_idx) <= counts->syncobj_count);
+ counts->syncobj_count = MAX2(syncobj_idx, non_reset_idx);
return VK_SUCCESS;
}
}
}
-static void
+static VkResult
radv_sparse_buffer_bind_memory(struct radv_device *device,
const VkSparseBufferMemoryBindInfo *bind)
{
RADV_FROM_HANDLE(radv_buffer, buffer, bind->buffer);
+ VkResult result;
for (uint32_t i = 0; i < bind->bindCount; ++i) {
struct radv_device_memory *mem = NULL;
if (bind->pBinds[i].memory != VK_NULL_HANDLE)
mem = radv_device_memory_from_handle(bind->pBinds[i].memory);
- device->ws->buffer_virtual_bind(buffer->bo,
- bind->pBinds[i].resourceOffset,
- bind->pBinds[i].size,
- mem ? mem->bo : NULL,
- bind->pBinds[i].memoryOffset);
+ result = device->ws->buffer_virtual_bind(buffer->bo,
+ bind->pBinds[i].resourceOffset,
+ bind->pBinds[i].size,
+ mem ? mem->bo : NULL,
+ bind->pBinds[i].memoryOffset);
+ if (result != VK_SUCCESS)
+ return result;
}
+
+ return VK_SUCCESS;
}
-static void
+static VkResult
radv_sparse_image_opaque_bind_memory(struct radv_device *device,
const VkSparseImageOpaqueMemoryBindInfo *bind)
{
RADV_FROM_HANDLE(radv_image, image, bind->image);
+ VkResult result;
for (uint32_t i = 0; i < bind->bindCount; ++i) {
struct radv_device_memory *mem = NULL;
if (bind->pBinds[i].memory != VK_NULL_HANDLE)
mem = radv_device_memory_from_handle(bind->pBinds[i].memory);
- device->ws->buffer_virtual_bind(image->bo,
- bind->pBinds[i].resourceOffset,
- bind->pBinds[i].size,
- mem ? mem->bo : NULL,
- bind->pBinds[i].memoryOffset);
+ result = device->ws->buffer_virtual_bind(image->bo,
+ bind->pBinds[i].resourceOffset,
+ bind->pBinds[i].size,
+ mem ? mem->bo : NULL,
+ bind->pBinds[i].memoryOffset);
+ if (result != VK_SUCCESS)
+ return result;
}
+
+ return VK_SUCCESS;
}
static VkResult
struct radv_queue *queue = submission->queue;
struct radeon_winsys_ctx *ctx = queue->hw_ctx;
uint32_t max_cs_submission = queue->device->trace_bo ? 1 : RADV_MAX_IBS_PER_SUBMIT;
- struct radeon_winsys_fence *base_fence = fence ? fence->fence : NULL;
+ struct radeon_winsys_fence *base_fence = NULL;
bool do_flush = submission->flush_caches || submission->wait_dst_stage_mask;
bool can_patch = true;
uint32_t advance;
struct radv_winsys_sem_info sem_info;
VkResult result;
- int ret;
struct radeon_cmdbuf *initial_preamble_cs = NULL;
struct radeon_cmdbuf *initial_flush_preamble_cs = NULL;
struct radeon_cmdbuf *continue_preamble_cs = NULL;
+ if (fence) {
+ /* Under most circumstances, out fences won't be temporary.
+ * However, the spec does allow it for opaque_fd.
+ *
+ * From the Vulkan 1.0.53 spec:
+ *
+ * "If the import is temporary, the implementation must
+ * restore the semaphore to its prior permanent state after
+ * submitting the next semaphore wait operation."
+ */
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
+ if (part->kind == RADV_FENCE_WINSYS)
+ base_fence = part->fence;
+ }
+
result = radv_get_preambles(queue, submission->cmd_buffers,
submission->cmd_buffer_count,
&initial_preamble_cs,
goto fail;
for (uint32_t i = 0; i < submission->buffer_bind_count; ++i) {
- radv_sparse_buffer_bind_memory(queue->device,
- submission->buffer_binds + i);
+ result = radv_sparse_buffer_bind_memory(queue->device,
+ submission->buffer_binds + i);
+ if (result != VK_SUCCESS)
+ goto fail;
}
for (uint32_t i = 0; i < submission->image_opaque_bind_count; ++i) {
- radv_sparse_image_opaque_bind_memory(queue->device,
- submission->image_opaque_binds + i);
+ result = radv_sparse_image_opaque_bind_memory(queue->device,
+ submission->image_opaque_binds + i);
+ if (result != VK_SUCCESS)
+ goto fail;
}
if (!submission->cmd_buffer_count) {
- ret = queue->device->ws->cs_submit(ctx, queue->queue_idx,
- &queue->device->empty_cs[queue->queue_family_index],
- 1, NULL, NULL,
- &sem_info, NULL,
- false, base_fence);
- if (ret) {
- radv_loge("failed to submit CS\n");
- abort();
- }
-
- goto success;
+ result = queue->device->ws->cs_submit(ctx, queue->queue_idx,
+ &queue->device->empty_cs[queue->queue_family_index],
+ 1, NULL, NULL,
+ &sem_info, NULL,
+ false, base_fence);
+ if (result != VK_SUCCESS)
+ goto fail;
} else {
struct radeon_cmdbuf **cs_array = malloc(sizeof(struct radeon_cmdbuf *) *
(submission->cmd_buffer_count));
bo_list = &queue->device->bo_list.list;
}
- ret = queue->device->ws->cs_submit(ctx, queue->queue_idx, cs_array + j,
- advance, initial_preamble, continue_preamble_cs,
- &sem_info, bo_list,
- can_patch, base_fence);
+ result = queue->device->ws->cs_submit(ctx, queue->queue_idx, cs_array + j,
+ advance, initial_preamble, continue_preamble_cs,
+ &sem_info, bo_list,
+ can_patch, base_fence);
if (unlikely(queue->device->use_global_bo_list))
pthread_mutex_unlock(&queue->device->bo_list.mutex);
- if (ret) {
- radv_loge("failed to submit CS\n");
- abort();
- }
+ if (result != VK_SUCCESS)
+ goto fail;
+
if (queue->device->trace_bo) {
radv_check_gpu_hangs(queue, cs_array[j]);
}
free(cs_array);
}
-success:
radv_free_temp_syncobjs(queue->device,
submission->temporary_semaphore_part_count,
submission->temporary_semaphore_parts);
return VK_SUCCESS;
fail:
+ if (result != VK_SUCCESS && result != VK_ERROR_DEVICE_LOST) {
+ /* When something bad happened during the submission, such as
+ * an out of memory issue, it might be hard to recover from
+ * this inconsistent state. To avoid this sort of problem, we
+ * assume that we are in a really bad situation and return
+ * VK_ERROR_DEVICE_LOST to ensure the clients do not attempt
+ * to submit the same job again to this device.
+ */
+ result = VK_ERROR_DEVICE_LOST;
+ }
+
radv_free_temp_syncobjs(queue->device,
submission->temporary_semaphore_part_count,
submission->temporary_semaphore_parts);
free(submission);
- return VK_ERROR_DEVICE_LOST;
+ return result;
}
static VkResult
struct radeon_winsys_ctx *ctx = queue->hw_ctx;
struct radv_winsys_sem_info sem_info;
VkResult result;
- int ret;
result = radv_alloc_sem_info(queue->device, &sem_info, 0, NULL, 0, 0,
0, NULL, VK_NULL_HANDLE);
if (result != VK_SUCCESS)
return false;
- ret = queue->device->ws->cs_submit(ctx, queue->queue_idx, &cs, 1, NULL,
- NULL, &sem_info, NULL, false, NULL);
+ result = queue->device->ws->cs_submit(ctx, queue->queue_idx, &cs, 1,
+ NULL, NULL, &sem_info, NULL,
+ false, NULL);
radv_free_sem_info(&sem_info);
- return !ret;
+ if (result != VK_SUCCESS)
+ return false;
+
+ return true;
+
}
/* Signals fence as soon as all the work currently put on queue is done. */
}
pthread_mutex_unlock(&queue->pending_mutex);
- queue->device->ws->ctx_wait_idle(queue->hw_ctx,
- radv_queue_family_to_ring(queue->queue_family_index),
- queue->queue_idx);
+ if (!queue->device->ws->ctx_wait_idle(queue->hw_ctx,
+ radv_queue_family_to_ring(queue->queue_family_index),
+ queue->queue_idx))
+ return VK_ERROR_DEVICE_LOST;
+
return VK_SUCCESS;
}
for (unsigned i = 0; i < RADV_MAX_QUEUE_FAMILIES; i++) {
for (unsigned q = 0; q < device->queue_count[i]; q++) {
- radv_QueueWaitIdle(radv_queue_to_handle(&device->queues[i][q]));
+ VkResult result =
+ radv_QueueWaitIdle(radv_queue_to_handle(&device->queues[i][q]));
+
+ if (result != VK_SUCCESS)
+ return result;
}
}
return VK_SUCCESS;
}
-static void radv_free_memory(struct radv_device *device,
- const VkAllocationCallbacks* pAllocator,
- struct radv_device_memory *mem)
+void
+radv_free_memory(struct radv_device *device,
+ const VkAllocationCallbacks* pAllocator,
+ struct radv_device_memory *mem)
{
if (mem == NULL)
return;
return VK_SUCCESS;
}
+static void
+radv_destroy_fence_part(struct radv_device *device,
+ struct radv_fence_part *part)
+{
+ switch (part->kind) {
+ case RADV_FENCE_NONE:
+ break;
+ case RADV_FENCE_WINSYS:
+ device->ws->destroy_fence(part->fence);
+ break;
+ case RADV_FENCE_SYNCOBJ:
+ device->ws->destroy_syncobj(device->ws, part->syncobj);
+ break;
+ case RADV_FENCE_WSI:
+ part->fence_wsi->destroy(part->fence_wsi);
+ break;
+ default:
+ unreachable("Invalid fence type");
+ }
+
+ part->kind = RADV_FENCE_NONE;
+}
+
+static void
+radv_destroy_fence(struct radv_device *device,
+ const VkAllocationCallbacks *pAllocator,
+ struct radv_fence *fence)
+{
+ radv_destroy_fence_part(device, &fence->temporary);
+ radv_destroy_fence_part(device, &fence->permanent);
+
+ vk_object_base_finish(&fence->base);
+ vk_free2(&device->vk.alloc, pAllocator, fence);
+}
+
VkResult radv_CreateFence(
VkDevice _device,
const VkFenceCreateInfo* pCreateInfo,
vk_find_struct_const(pCreateInfo->pNext, EXPORT_FENCE_CREATE_INFO);
VkExternalFenceHandleTypeFlags handleTypes =
export ? export->handleTypes : 0;
+ struct radv_fence *fence;
- struct radv_fence *fence = vk_alloc2(&device->vk.alloc, pAllocator,
- sizeof(*fence), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
-
+ fence = vk_zalloc2(&device->vk.alloc, pAllocator, sizeof(*fence), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (!fence)
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
vk_object_base_init(&device->vk, &fence->base, VK_OBJECT_TYPE_FENCE);
- fence->fence_wsi = NULL;
- fence->temp_syncobj = 0;
if (device->always_use_syncobj || handleTypes) {
- int ret = device->ws->create_syncobj(device->ws, &fence->syncobj);
+ fence->permanent.kind = RADV_FENCE_SYNCOBJ;
+
+ bool create_signaled = false;
+ if (pCreateInfo->flags & VK_FENCE_CREATE_SIGNALED_BIT)
+ create_signaled = true;
+
+ int ret = device->ws->create_syncobj(device->ws, create_signaled,
+ &fence->permanent.syncobj);
if (ret) {
- vk_free2(&device->vk.alloc, pAllocator, fence);
+ radv_destroy_fence(device, pAllocator, fence);
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
}
- if (pCreateInfo->flags & VK_FENCE_CREATE_SIGNALED_BIT) {
- device->ws->signal_syncobj(device->ws, fence->syncobj);
- }
- fence->fence = NULL;
} else {
- fence->fence = device->ws->create_fence();
- if (!fence->fence) {
+ fence->permanent.kind = RADV_FENCE_WINSYS;
+
+ fence->permanent.fence = device->ws->create_fence();
+ if (!fence->permanent.fence) {
vk_free2(&device->vk.alloc, pAllocator, fence);
+ radv_destroy_fence(device, pAllocator, fence);
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
}
- fence->syncobj = 0;
if (pCreateInfo->flags & VK_FENCE_CREATE_SIGNALED_BIT)
- device->ws->signal_fence(fence->fence);
+ device->ws->signal_fence(fence->permanent.fence);
}
*pFence = radv_fence_to_handle(fence);
return VK_SUCCESS;
}
+
void radv_DestroyFence(
VkDevice _device,
VkFence _fence,
if (!fence)
return;
- if (fence->temp_syncobj)
- device->ws->destroy_syncobj(device->ws, fence->temp_syncobj);
- if (fence->syncobj)
- device->ws->destroy_syncobj(device->ws, fence->syncobj);
- if (fence->fence)
- device->ws->destroy_fence(fence->fence);
- if (fence->fence_wsi)
- fence->fence_wsi->destroy(fence->fence_wsi);
-
- vk_object_base_finish(&fence->base);
- vk_free2(&device->vk.alloc, pAllocator, fence);
+ radv_destroy_fence(device, pAllocator, fence);
}
{
for (uint32_t i = 0; i < fenceCount; ++i) {
RADV_FROM_HANDLE(radv_fence, fence, pFences[i]);
- if (fence->fence == NULL || fence->syncobj ||
- fence->temp_syncobj || fence->fence_wsi ||
- (!device->ws->is_fence_waitable(fence->fence)))
+
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
+ if (part->kind != RADV_FENCE_WINSYS ||
+ !device->ws->is_fence_waitable(part->fence))
return false;
}
return true;
{
for (uint32_t i = 0; i < fenceCount; ++i) {
RADV_FROM_HANDLE(radv_fence, fence, pFences[i]);
- if (fence->syncobj == 0 && fence->temp_syncobj == 0)
+
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
+ if (part->kind != RADV_FENCE_SYNCOBJ)
return false;
}
return true;
for (uint32_t i = 0; i < fenceCount; ++i) {
RADV_FROM_HANDLE(radv_fence, fence, pFences[i]);
- handles[i] = fence->temp_syncobj ? fence->temp_syncobj : fence->syncobj;
+
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
+
+ assert(part->kind == RADV_FENCE_SYNCOBJ);
+ handles[i] = part->syncobj;
}
bool success = device->ws->wait_syncobj(device->ws, handles, fenceCount, waitAll, timeout);
for (uint32_t i = 0; i < fenceCount; ++i) {
RADV_FROM_HANDLE(radv_fence, fence, pFences[i]);
- if (device->ws->fence_wait(device->ws, fence->fence, false, 0)) {
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
+ assert(part->kind == RADV_FENCE_WINSYS);
+
+ if (device->ws->fence_wait(device->ws, part->fence, false, 0)) {
free(fences);
return VK_SUCCESS;
}
- fences[wait_count++] = fence->fence;
+ fences[wait_count++] = part->fence;
}
bool success = device->ws->fences_wait(device->ws, fences, wait_count,
RADV_FROM_HANDLE(radv_fence, fence, pFences[i]);
bool expired = false;
- if (fence->temp_syncobj) {
- if (!device->ws->wait_syncobj(device->ws, &fence->temp_syncobj, 1, true, timeout))
- return VK_TIMEOUT;
- continue;
- }
-
- if (fence->syncobj) {
- if (!device->ws->wait_syncobj(device->ws, &fence->syncobj, 1, true, timeout))
- return VK_TIMEOUT;
- continue;
- }
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
- if (fence->fence) {
- if (!device->ws->is_fence_waitable(fence->fence)) {
- while(!device->ws->is_fence_waitable(fence->fence) &&
+ switch (part->kind) {
+ case RADV_FENCE_NONE:
+ break;
+ case RADV_FENCE_WINSYS:
+ if (!device->ws->is_fence_waitable(part->fence)) {
+ while (!device->ws->is_fence_waitable(part->fence) &&
radv_get_current_time() <= timeout)
/* Do nothing */;
}
expired = device->ws->fence_wait(device->ws,
- fence->fence,
+ part->fence,
true, timeout);
if (!expired)
return VK_TIMEOUT;
- }
-
- if (fence->fence_wsi) {
- VkResult result = fence->fence_wsi->wait(fence->fence_wsi, timeout);
+ break;
+ case RADV_FENCE_SYNCOBJ:
+ if (!device->ws->wait_syncobj(device->ws,
+ &part->syncobj, 1, true,
+ timeout))
+ return VK_TIMEOUT;
+ break;
+ case RADV_FENCE_WSI: {
+ VkResult result = part->fence_wsi->wait(part->fence_wsi, timeout);
if (result != VK_SUCCESS)
return result;
+ break;
+ }
+ default:
+ unreachable("Invalid fence type");
}
}
for (unsigned i = 0; i < fenceCount; ++i) {
RADV_FROM_HANDLE(radv_fence, fence, pFences[i]);
- if (fence->fence)
- device->ws->reset_fence(fence->fence);
- /* Per spec, we first restore the permanent payload, and then reset, so
- * having a temp syncobj should not skip resetting the permanent syncobj. */
- if (fence->temp_syncobj) {
- device->ws->destroy_syncobj(device->ws, fence->temp_syncobj);
- fence->temp_syncobj = 0;
- }
+ /* From the Vulkan 1.0.53 spec:
+ *
+ * "If any member of pFences currently has its payload
+ * imported with temporary permanence, that fence’s prior
+ * permanent payload is irst restored. The remaining
+ * operations described therefore operate on the restored
+ * payload."
+ */
+ if (fence->temporary.kind != RADV_FENCE_NONE)
+ radv_destroy_fence_part(device, &fence->temporary);
- if (fence->syncobj) {
- device->ws->reset_syncobj(device->ws, fence->syncobj);
+ struct radv_fence_part *part = &fence->permanent;
+
+ switch (part->kind) {
+ case RADV_FENCE_WSI:
+ device->ws->reset_fence(part->fence);
+ break;
+ case RADV_FENCE_SYNCOBJ:
+ device->ws->reset_syncobj(device->ws, part->syncobj);
+ break;
+ default:
+ unreachable("Invalid fence type");
}
}
RADV_FROM_HANDLE(radv_device, device, _device);
RADV_FROM_HANDLE(radv_fence, fence, _fence);
- if (fence->temp_syncobj) {
- bool success = device->ws->wait_syncobj(device->ws, &fence->temp_syncobj, 1, true, 0);
- return success ? VK_SUCCESS : VK_NOT_READY;
- }
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
- if (fence->syncobj) {
- bool success = device->ws->wait_syncobj(device->ws, &fence->syncobj, 1, true, 0);
- return success ? VK_SUCCESS : VK_NOT_READY;
- }
-
- if (fence->fence) {
- if (!device->ws->fence_wait(device->ws, fence->fence, false, 0))
+ switch (part->kind) {
+ case RADV_FENCE_NONE:
+ break;
+ case RADV_FENCE_WINSYS:
+ if (!device->ws->fence_wait(device->ws, part->fence, false, 0))
+ return VK_NOT_READY;
+ break;
+ case RADV_FENCE_SYNCOBJ: {
+ bool success = device->ws->wait_syncobj(device->ws,
+ &part->syncobj, 1, true, 0);
+ if (!success)
return VK_NOT_READY;
+ break;
}
- if (fence->fence_wsi) {
- VkResult result = fence->fence_wsi->wait(fence->fence_wsi, 0);
-
+ case RADV_FENCE_WSI: {
+ VkResult result = part->fence_wsi->wait(part->fence_wsi, 0);
if (result != VK_SUCCESS) {
if (result == VK_TIMEOUT)
return VK_NOT_READY;
return result;
}
+ break;
}
+ default:
+ unreachable("Invalid fence type");
+ }
+
return VK_SUCCESS;
}
struct radv_timeline_point *ret = NULL;
struct radv_timeline_point *prev = NULL;
+ int r;
if (p <= timeline->highest_signaled)
return NULL;
if (list_is_empty(&timeline->free_points)) {
ret = malloc(sizeof(struct radv_timeline_point));
- device->ws->create_syncobj(device->ws, &ret->syncobj);
+ r = device->ws->create_syncobj(device->ws, false, &ret->syncobj);
+ if (r) {
+ free(ret);
+ return NULL;
+ }
} else {
ret = list_first_entry(&timeline->free_points, struct radv_timeline_point, list);
list_del(&ret->list);
return type_info->semaphoreType;
}
+static void
+radv_destroy_semaphore(struct radv_device *device,
+ const VkAllocationCallbacks *pAllocator,
+ struct radv_semaphore *sem)
+{
+ radv_destroy_semaphore_part(device, &sem->temporary);
+ radv_destroy_semaphore_part(device, &sem->permanent);
+ vk_object_base_finish(&sem->base);
+ vk_free2(&device->vk.alloc, pAllocator, sem);
+}
+
VkResult radv_CreateSemaphore(
VkDevice _device,
const VkSemaphoreCreateInfo* pCreateInfo,
sem->permanent.kind = RADV_SEMAPHORE_TIMELINE;
} else if (device->always_use_syncobj || handleTypes) {
assert (device->physical_device->rad_info.has_syncobj);
- int ret = device->ws->create_syncobj(device->ws, &sem->permanent.syncobj);
+ int ret = device->ws->create_syncobj(device->ws, false,
+ &sem->permanent.syncobj);
if (ret) {
- vk_free2(&device->vk.alloc, pAllocator, sem);
+ radv_destroy_semaphore(device, pAllocator, sem);
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
}
sem->permanent.kind = RADV_SEMAPHORE_SYNCOBJ;
} else {
sem->permanent.ws_sem = device->ws->create_sem(device->ws);
if (!sem->permanent.ws_sem) {
- vk_free2(&device->vk.alloc, pAllocator, sem);
+ radv_destroy_semaphore(device, pAllocator, sem);
return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
}
sem->permanent.kind = RADV_SEMAPHORE_WINSYS;
if (!_semaphore)
return;
- radv_destroy_semaphore_part(device, &sem->temporary);
- radv_destroy_semaphore_part(device, &sem->permanent);
- vk_object_base_finish(&sem->base);
- vk_free2(&device->vk.alloc, pAllocator, sem);
+ radv_destroy_semaphore(device, pAllocator, sem);
}
VkResult
return VK_SUCCESS;
}
+static void radv_destroy_event(struct radv_device *device,
+ const VkAllocationCallbacks* pAllocator,
+ struct radv_event *event)
+{
+ if (event->bo)
+ device->ws->buffer_destroy(event->bo);
+ vk_object_base_finish(&event->base);
+ vk_free2(&device->vk.alloc, pAllocator, event);
+}
VkResult radv_CreateEvent(
VkDevice _device,
RADEON_FLAG_VA_UNCACHED | RADEON_FLAG_CPU_ACCESS | RADEON_FLAG_NO_INTERPROCESS_SHARING,
RADV_BO_PRIORITY_FENCE);
if (!event->bo) {
- vk_free2(&device->vk.alloc, pAllocator, event);
+ radv_destroy_event(device, pAllocator, event);
return vk_error(device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
}
event->map = (uint64_t*)device->ws->buffer_map(event->bo);
+ if (!event->map) {
+ radv_destroy_event(device, pAllocator, event);
+ return vk_error(device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
+ }
*pEvent = radv_event_to_handle(event);
if (!event)
return;
- device->ws->buffer_destroy(event->bo);
- vk_object_base_finish(&event->base);
- vk_free2(&device->vk.alloc, pAllocator, event);
+
+ radv_destroy_event(device, pAllocator, event);
}
VkResult radv_GetEventStatus(
return VK_SUCCESS;
}
+static void
+radv_destroy_buffer(struct radv_device *device,
+ const VkAllocationCallbacks *pAllocator,
+ struct radv_buffer *buffer)
+{
+ if ((buffer->flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT) && buffer->bo)
+ device->ws->buffer_destroy(buffer->bo);
+
+ vk_object_base_finish(&buffer->base);
+ vk_free2(&device->vk.alloc, pAllocator, buffer);
+}
+
VkResult radv_CreateBuffer(
VkDevice _device,
const VkBufferCreateInfo* pCreateInfo,
4096, 0, RADEON_FLAG_VIRTUAL,
RADV_BO_PRIORITY_VIRTUAL);
if (!buffer->bo) {
- vk_free2(&device->vk.alloc, pAllocator, buffer);
+ radv_destroy_buffer(device, pAllocator, buffer);
return vk_error(device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
}
}
if (!buffer)
return;
- if (buffer->flags & VK_BUFFER_CREATE_SPARSE_BINDING_BIT)
- device->ws->buffer_destroy(buffer->bo);
-
- vk_object_base_finish(&buffer->base);
- vk_free2(&device->vk.alloc, pAllocator, buffer);
+ radv_destroy_buffer(device, pAllocator, buffer);
}
VkDeviceAddress radv_GetBufferDeviceAddress(
.pipe_aligned = 1,
};
- if (iview->image->dcc_offset)
+ if (surf->dcc_offset)
meta = surf->u.gfx9.dcc;
cb->cb_color_attrib |= S_028C74_COLOR_SW_MODE(surf->u.gfx9.surf.swizzle_mode) |
/* CMASK variables */
va = radv_buffer_get_va(iview->bo) + iview->image->offset;
- va += iview->image->cmask_offset;
+ va += surf->cmask_offset;
cb->cb_color_cmask = va >> 8;
va = radv_buffer_get_va(iview->bo) + iview->image->offset;
- va += iview->image->dcc_offset;
+ va += surf->dcc_offset;
if (radv_dcc_enabled(iview->image, iview->base_mip) &&
device->physical_device->rad_info.chip_class <= GFX8)
}
if (radv_image_has_fmask(iview->image)) {
- va = radv_buffer_get_va(iview->bo) + iview->image->offset + iview->image->fmask_offset;
+ va = radv_buffer_get_va(iview->bo) + iview->image->offset + surf->fmask_offset;
cb->cb_color_fmask = va >> 8;
cb->cb_color_fmask |= surf->fmask_tile_swizzle;
} else {
/* Use all of the htile_buffer for depth if there's no stencil. */
ds->db_stencil_info |= S_02803C_TILE_STENCIL_DISABLE(1);
va = radv_buffer_get_va(iview->bo) + iview->image->offset +
- iview->image->htile_offset;
+ surf->htile_offset;
ds->db_htile_data_base = va >> 8;
ds->db_htile_surface = S_028ABC_FULL_CACHE(1) |
S_028ABC_PIPE_ALIGNED(1);
ds->db_stencil_info |= S_028044_TILE_STENCIL_DISABLE(1);
va = radv_buffer_get_va(iview->bo) + iview->image->offset +
- iview->image->htile_offset;
+ surf->htile_offset;
ds->db_htile_data_base = va >> 8;
ds->db_htile_surface = S_028ABC_FULL_CACHE(1);
* leave a syncobj in an undetermined state in the fence. */
uint32_t syncobj_handle = *syncobj;
if (!syncobj_handle) {
- int ret = device->ws->create_syncobj(device->ws, &syncobj_handle);
+ bool create_signaled = fd == -1 ? true : false;
+
+ int ret = device->ws->create_syncobj(device->ws, create_signaled,
+ &syncobj_handle);
if (ret) {
- return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
+ return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY);
}
+ } else {
+ if (fd == -1)
+ device->ws->signal_syncobj(device->ws, syncobj_handle);
}
- if (fd == -1) {
- device->ws->signal_syncobj(device->ws, syncobj_handle);
- } else {
+ if (fd != -1) {
int ret = device->ws->import_syncobj_from_sync_file(device->ws, syncobj_handle, fd);
- if (ret != 0)
- return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
+ if (ret)
+ return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
+ close(fd);
}
*syncobj = syncobj_handle;
- if (fd != -1)
- close(fd);
return VK_SUCCESS;
}
switch(pGetFdInfo->handleType) {
case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT:
ret = device->ws->export_syncobj(device->ws, syncobj_handle, pFd);
+ if (ret)
+ return vk_error(device->instance, VK_ERROR_TOO_MANY_OBJECTS);
break;
case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT:
ret = device->ws->export_syncobj_to_sync_file(device->ws, syncobj_handle, pFd);
- if (!ret) {
- if (sem->temporary.kind != RADV_SEMAPHORE_NONE) {
- radv_destroy_semaphore_part(device, &sem->temporary);
- } else {
- device->ws->reset_syncobj(device->ws, syncobj_handle);
- }
+ if (ret)
+ return vk_error(device->instance, VK_ERROR_TOO_MANY_OBJECTS);
+
+ if (sem->temporary.kind != RADV_SEMAPHORE_NONE) {
+ radv_destroy_semaphore_part(device, &sem->temporary);
+ } else {
+ device->ws->reset_syncobj(device->ws, syncobj_handle);
}
break;
default:
unreachable("Unhandled semaphore handle type");
}
- if (ret)
- return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
return VK_SUCCESS;
}
{
RADV_FROM_HANDLE(radv_device, device, _device);
RADV_FROM_HANDLE(radv_fence, fence, pImportFenceFdInfo->fence);
- uint32_t *syncobj_dst = NULL;
-
+ struct radv_fence_part *dst = NULL;
+ VkResult result;
if (pImportFenceFdInfo->flags & VK_FENCE_IMPORT_TEMPORARY_BIT) {
- syncobj_dst = &fence->temp_syncobj;
+ dst = &fence->temporary;
} else {
- syncobj_dst = &fence->syncobj;
+ dst = &fence->permanent;
}
+ uint32_t syncobj = dst->kind == RADV_FENCE_SYNCOBJ ? dst->syncobj : 0;
+
switch(pImportFenceFdInfo->handleType) {
case VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT:
- return radv_import_opaque_fd(device, pImportFenceFdInfo->fd, syncobj_dst);
+ result = radv_import_opaque_fd(device, pImportFenceFdInfo->fd, &syncobj);
+ break;
case VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT:
- return radv_import_sync_fd(device, pImportFenceFdInfo->fd, syncobj_dst);
+ result = radv_import_sync_fd(device, pImportFenceFdInfo->fd, &syncobj);
+ break;
default:
unreachable("Unhandled fence handle type");
}
+
+ if (result == VK_SUCCESS) {
+ dst->syncobj = syncobj;
+ dst->kind = RADV_FENCE_SYNCOBJ;
+ }
+
+ return result;
}
VkResult radv_GetFenceFdKHR(VkDevice _device,
RADV_FROM_HANDLE(radv_device, device, _device);
RADV_FROM_HANDLE(radv_fence, fence, pGetFdInfo->fence);
int ret;
- uint32_t syncobj_handle;
- if (fence->temp_syncobj)
- syncobj_handle = fence->temp_syncobj;
- else
- syncobj_handle = fence->syncobj;
+ struct radv_fence_part *part =
+ fence->temporary.kind != RADV_FENCE_NONE ?
+ &fence->temporary : &fence->permanent;
switch(pGetFdInfo->handleType) {
case VK_EXTERNAL_FENCE_HANDLE_TYPE_OPAQUE_FD_BIT:
- ret = device->ws->export_syncobj(device->ws, syncobj_handle, pFd);
+ ret = device->ws->export_syncobj(device->ws, part->syncobj, pFd);
+ if (ret)
+ return vk_error(device->instance, VK_ERROR_TOO_MANY_OBJECTS);
break;
case VK_EXTERNAL_FENCE_HANDLE_TYPE_SYNC_FD_BIT:
- ret = device->ws->export_syncobj_to_sync_file(device->ws, syncobj_handle, pFd);
- if (!ret) {
- if (fence->temp_syncobj) {
- close (fence->temp_syncobj);
- fence->temp_syncobj = 0;
- } else {
- device->ws->reset_syncobj(device->ws, syncobj_handle);
- }
+ ret = device->ws->export_syncobj_to_sync_file(device->ws,
+ part->syncobj, pFd);
+ if (ret)
+ return vk_error(device->instance, VK_ERROR_TOO_MANY_OBJECTS);
+
+ if (part == &fence->temporary) {
+ radv_destroy_fence_part(device, part);
+ } else {
+ device->ws->reset_syncobj(device->ws, part->syncobj);
}
break;
default:
unreachable("Unhandled fence handle type");
}
- if (ret)
- return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE);
return VK_SUCCESS;
}
projects / mesa.git / blobdiff