X-Git-Url: https://git.libre-soc.org/?p=mesa.git;a=blobdiff_plain;f=src%2Ffreedreno%2Fvulkan%2Ftu_device.c;h=43ed5f6e4a8c8dc7e7092957e7b94f8c2dbaeaf1;hp=1635bfddb498ebec7b2007aac9cb3ce62a64d9b5;hb=aba8c579a9780f9d3ad2d677f9ef625df7726479;hpb=f1efc9a1c89e3b8f1b5faad4238035ebe8a8f217 diff --git a/src/freedreno/vulkan/tu_device.c b/src/freedreno/vulkan/tu_device.c index 1635bfddb49..43ed5f6e4a8 100644 --- a/src/freedreno/vulkan/tu_device.c +++ b/src/freedreno/vulkan/tu_device.c @@ -39,12 +39,19 @@ #include "compiler/glsl_types.h" #include "util/debug.h" #include "util/disk_cache.h" -#include "util/strtod.h" +#include "util/u_atomic.h" #include "vk_format.h" #include "vk_util.h" #include "drm-uapi/msm_drm.h" +/* for fd_get_driver/device_uuid() */ +#include "freedreno/common/freedreno_uuid.h" + +static void +tu_semaphore_remove_temp(struct tu_device *device, + struct tu_semaphore *sem); + static int tu_device_get_cache_uuid(uint16_t family, void *uuid) { @@ -61,19 +68,6 @@ tu_device_get_cache_uuid(uint16_t family, void *uuid) return 0; } -static void -tu_get_driver_uuid(void *uuid) -{ - memset(uuid, 0, VK_UUID_SIZE); - snprintf(uuid, VK_UUID_SIZE, "freedreno"); -} - -static void -tu_get_device_uuid(void *uuid) -{ - memset(uuid, 0, VK_UUID_SIZE); -} - static VkResult tu_bo_init(struct tu_device *dev, struct tu_bo *bo, @@ -216,6 +210,9 @@ tu_physical_device_init(struct tu_physical_device *device, return result; } + device->msm_major_version = version->version_major; + device->msm_minor_version = version->version_minor; + drmFreeVersion(version); if (instance->debug_flags & TU_DEBUG_STARTUP) @@ -253,13 +250,39 @@ tu_physical_device_init(struct tu_physical_device *device, goto fail; } + if (tu_drm_get_gmem_base(device, &device->gmem_base)) { + if (instance->debug_flags & TU_DEBUG_STARTUP) + tu_logi("Could not query the GMEM size"); + result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED, + "could not get GMEM size"); + goto fail; + } + memset(device->name, 0, sizeof(device->name)); sprintf(device->name, "FD%d", device->gpu_id); switch (device->gpu_id) { + case 618: + device->ccu_offset_gmem = 0x7c000; /* 0x7e000 in some cases? */ + device->ccu_offset_bypass = 0x10000; + device->tile_align_w = 64; + device->magic.PC_UNKNOWN_9805 = 0x0; + device->magic.SP_UNKNOWN_A0F8 = 0x0; + break; case 630: - device->tile_align_w = 32; - device->tile_align_h = 32; + case 640: + device->ccu_offset_gmem = 0xf8000; + device->ccu_offset_bypass = 0x20000; + device->tile_align_w = 64; + device->magic.PC_UNKNOWN_9805 = 0x1; + device->magic.SP_UNKNOWN_A0F8 = 0x1; + break; + case 650: + device->ccu_offset_gmem = 0x114000; + device->ccu_offset_bypass = 0x30000; + device->tile_align_w = 96; + device->magic.PC_UNKNOWN_9805 = 0x2; + device->magic.SP_UNKNOWN_A0F8 = 0x2; break; default: result = vk_errorf(instance, VK_ERROR_INITIALIZATION_FAILED, @@ -282,10 +305,10 @@ tu_physical_device_init(struct tu_physical_device *device, fprintf(stderr, "WARNING: tu is not a conformant vulkan implementation, " "testing use only.\n"); - tu_get_driver_uuid(&device->device_uuid); - tu_get_device_uuid(&device->device_uuid); + fd_get_driver_uuid(device->driver_uuid); + fd_get_device_uuid(device->device_uuid, device->gpu_id); - tu_fill_device_extension_table(device, &device->supported_extensions); + tu_physical_device_get_supported_extensions(device, &device->supported_extensions); if (result != VK_SUCCESS) { vk_error(instance, result); @@ -318,7 +341,7 @@ tu_physical_device_finish(struct tu_physical_device *device) close(device->master_fd); } -static void * +static VKAPI_ATTR void * default_alloc_func(void *pUserData, size_t size, size_t align, @@ -327,7 +350,7 @@ default_alloc_func(void *pUserData, return malloc(size); } -static void * +static VKAPI_ATTR void * default_realloc_func(void *pUserData, void *pOriginal, size_t size, @@ -337,7 +360,7 @@ default_realloc_func(void *pUserData, return realloc(pOriginal, size); } -static void +static VKAPI_ATTR void default_free_func(void *pUserData, void *pMemory) { free(pMemory); @@ -354,6 +377,10 @@ static const struct debug_control tu_debug_options[] = { { "startup", TU_DEBUG_STARTUP }, { "nir", TU_DEBUG_NIR }, { "ir3", TU_DEBUG_IR3 }, + { "nobin", TU_DEBUG_NOBIN }, + { "sysmem", TU_DEBUG_SYSMEM }, + { "forcebin", TU_DEBUG_FORCEBIN }, + { "noubwc", TU_DEBUG_NOUBWC }, { NULL, 0 } }; @@ -417,7 +444,7 @@ tu_CreateInstance(const VkInstanceCreateInfo *pCreateInfo, const char *ext_name = pCreateInfo->ppEnabledExtensionNames[i]; int index = tu_get_instance_extension_index(ext_name); - if (index < 0 || !tu_supported_instance_extensions.extensions[index]) { + if (index < 0 || !tu_instance_extensions_supported.extensions[index]) { vk_free2(&default_alloc, pAllocator, instance); return vk_error(instance, VK_ERROR_EXTENSION_NOT_PRESENT); } @@ -431,7 +458,6 @@ tu_CreateInstance(const VkInstanceCreateInfo *pCreateInfo, return vk_error(instance, result); } - _mesa_locale_init(); glsl_type_singleton_init_or_ref(); VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false)); @@ -457,7 +483,6 @@ tu_DestroyInstance(VkInstance _instance, VG(VALGRIND_DESTROY_MEMPOOL(instance)); glsl_type_singleton_decref(); - _mesa_locale_fini(); vk_debug_report_instance_destroy(&instance->debug_report_callbacks); @@ -563,30 +588,30 @@ tu_GetPhysicalDeviceFeatures(VkPhysicalDevice physicalDevice, memset(pFeatures, 0, sizeof(*pFeatures)); *pFeatures = (VkPhysicalDeviceFeatures) { - .robustBufferAccess = false, - .fullDrawIndexUint32 = false, - .imageCubeArray = false, - .independentBlend = false, - .geometryShader = false, - .tessellationShader = false, - .sampleRateShading = false, - .dualSrcBlend = false, - .logicOp = false, - .multiDrawIndirect = false, - .drawIndirectFirstInstance = false, - .depthClamp = false, + .robustBufferAccess = true, + .fullDrawIndexUint32 = true, + .imageCubeArray = true, + .independentBlend = true, + .geometryShader = true, + .tessellationShader = true, + .sampleRateShading = true, + .dualSrcBlend = true, + .logicOp = true, + .multiDrawIndirect = true, + .drawIndirectFirstInstance = true, + .depthClamp = true, .depthBiasClamp = false, - .fillModeNonSolid = false, - .depthBounds = false, + .fillModeNonSolid = true, + .depthBounds = true, .wideLines = false, .largePoints = false, - .alphaToOne = false, + .alphaToOne = true, .multiViewport = false, .samplerAnisotropy = true, .textureCompressionETC2 = true, .textureCompressionASTC_LDR = true, .textureCompressionBC = true, - .occlusionQueryPrecise = false, + .occlusionQueryPrecise = true, .pipelineStatisticsQuery = false, .vertexPipelineStoresAndAtomics = false, .fragmentStoresAndAtomics = false, @@ -618,6 +643,22 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice, vk_foreach_struct(ext, pFeatures->pNext) { switch (ext->sType) { + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VULKAN_1_1_FEATURES: { + VkPhysicalDeviceVulkan11Features *features = (void *) ext; + features->storageBuffer16BitAccess = false; + features->uniformAndStorageBuffer16BitAccess = false; + features->storagePushConstant16 = false; + features->storageInputOutput16 = false; + features->multiview = false; + features->multiviewGeometryShader = false; + features->multiviewTessellationShader = false; + features->variablePointersStorageBuffer = false; + features->variablePointers = false; + features->protectedMemory = false; + features->samplerYcbcrConversion = true; + features->shaderDrawParameters = true; + break; + } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: { VkPhysicalDeviceVariablePointersFeatures *features = (void *) ext; features->variablePointersStorageBuffer = false; @@ -635,7 +676,7 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice, case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: { VkPhysicalDeviceShaderDrawParametersFeatures *features = (VkPhysicalDeviceShaderDrawParametersFeatures *) ext; - features->shaderDrawParameters = false; + features->shaderDrawParameters = true; break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: { @@ -656,7 +697,7 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice, case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_YCBCR_CONVERSION_FEATURES: { VkPhysicalDeviceSamplerYcbcrConversionFeatures *features = (VkPhysicalDeviceSamplerYcbcrConversionFeatures *) ext; - features->samplerYcbcrConversion = false; + features->samplerYcbcrConversion = true; break; } case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT: { @@ -691,6 +732,26 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice, features->inheritedConditionalRendering = false; break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT: { + VkPhysicalDeviceTransformFeedbackFeaturesEXT *features = + (VkPhysicalDeviceTransformFeedbackFeaturesEXT *) ext; + features->transformFeedback = true; + features->geometryStreams = false; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INDEX_TYPE_UINT8_FEATURES_EXT: { + VkPhysicalDeviceIndexTypeUint8FeaturesEXT *features = + (VkPhysicalDeviceIndexTypeUint8FeaturesEXT *)ext; + features->indexTypeUint8 = true; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT: { + VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *features = + (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *)ext; + features->vertexAttributeInstanceRateDivisor = true; + features->vertexAttributeInstanceRateZeroDivisor = true; + break; + } default: break; } @@ -703,21 +764,16 @@ tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice, VkPhysicalDeviceProperties *pProperties) { TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice); - VkSampleCountFlags sample_counts = VK_SAMPLE_COUNT_1_BIT | - VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT | VK_SAMPLE_COUNT_8_BIT; - - /* make sure that the entire descriptor set is addressable with a signed - * 32-bit int. So the sum of all limits scaled by descriptor size has to - * be at most 2 GiB. the combined image & samples object count as one of - * both. This limit is for the pipeline layout, not for the set layout, but - * there is no set limit, so we just set a pipeline limit. I don't think - * any app is going to hit this soon. */ - size_t max_descriptor_set_size = - ((1ull << 31) - 16 * MAX_DYNAMIC_BUFFERS) / - (32 /* uniform buffer, 32 due to potential space wasted on alignment */ + - 32 /* storage buffer, 32 due to potential space wasted on alignment */ + - 32 /* sampler, largest when combined with image */ + - 64 /* sampled image */ + 64 /* storage image */); + VkSampleCountFlags sample_counts = + VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT; + + /* I have no idea what the maximum size is, but the hardware supports very + * large numbers of descriptors (at least 2^16). This limit is based on + * CP_LOAD_STATE6, which has a 28-bit field for the DWORD offset, so that + * we don't have to think about what to do if that overflows, but really + * nothing is likely to get close to this. + */ + const size_t max_descriptor_set_size = (1 << 28) / A6XX_TEX_CONST_DWORDS; VkPhysicalDeviceLimits limits = { .maxImageDimension1D = (1 << 14), @@ -726,8 +782,8 @@ tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice, .maxImageDimensionCube = (1 << 14), .maxImageArrayLayers = (1 << 11), .maxTexelBufferElements = 128 * 1024 * 1024, - .maxUniformBufferRange = UINT32_MAX, - .maxStorageBufferRange = UINT32_MAX, + .maxUniformBufferRange = MAX_UNIFORM_BUFFER_RANGE, + .maxStorageBufferRange = MAX_STORAGE_BUFFER_RANGE, .maxPushConstantsSize = MAX_PUSH_CONSTANTS_SIZE, .maxMemoryAllocationCount = UINT32_MAX, .maxSamplerAllocationCount = 64 * 1024, @@ -739,7 +795,7 @@ tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice, .maxPerStageDescriptorStorageBuffers = max_descriptor_set_size, .maxPerStageDescriptorSampledImages = max_descriptor_set_size, .maxPerStageDescriptorStorageImages = max_descriptor_set_size, - .maxPerStageDescriptorInputAttachments = max_descriptor_set_size, + .maxPerStageDescriptorInputAttachments = MAX_RTS, .maxPerStageResources = max_descriptor_set_size, .maxDescriptorSetSamplers = max_descriptor_set_size, .maxDescriptorSetUniformBuffers = max_descriptor_set_size, @@ -748,10 +804,10 @@ tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice, .maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_STORAGE_BUFFERS, .maxDescriptorSetSampledImages = max_descriptor_set_size, .maxDescriptorSetStorageImages = max_descriptor_set_size, - .maxDescriptorSetInputAttachments = max_descriptor_set_size, + .maxDescriptorSetInputAttachments = MAX_RTS, .maxVertexInputAttributes = 32, .maxVertexInputBindings = 32, - .maxVertexInputAttributeOffset = 2047, + .maxVertexInputAttributeOffset = 4095, .maxVertexInputBindingStride = 2048, .maxVertexOutputComponents = 128, .maxTessellationGenerationLevel = 64, @@ -762,12 +818,12 @@ tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice, .maxTessellationControlTotalOutputComponents = 4096, .maxTessellationEvaluationInputComponents = 128, .maxTessellationEvaluationOutputComponents = 128, - .maxGeometryShaderInvocations = 127, + .maxGeometryShaderInvocations = 32, .maxGeometryInputComponents = 64, .maxGeometryOutputComponents = 128, .maxGeometryOutputVertices = 256, .maxGeometryTotalOutputComponents = 1024, - .maxFragmentInputComponents = 128, + .maxFragmentInputComponents = 124, .maxFragmentOutputAttachments = 8, .maxFragmentDualSrcAttachments = 1, .maxFragmentCombinedOutputResources = 8, @@ -775,28 +831,28 @@ tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice, .maxComputeWorkGroupCount = { 65535, 65535, 65535 }, .maxComputeWorkGroupInvocations = 2048, .maxComputeWorkGroupSize = { 2048, 2048, 2048 }, - .subPixelPrecisionBits = 4 /* FIXME */, - .subTexelPrecisionBits = 4 /* FIXME */, - .mipmapPrecisionBits = 4 /* FIXME */, + .subPixelPrecisionBits = 8, + .subTexelPrecisionBits = 8, + .mipmapPrecisionBits = 8, .maxDrawIndexedIndexValue = UINT32_MAX, .maxDrawIndirectCount = UINT32_MAX, - .maxSamplerLodBias = 16, + .maxSamplerLodBias = 4095.0 / 256.0, /* [-16, 15.99609375] */ .maxSamplerAnisotropy = 16, .maxViewports = MAX_VIEWPORTS, .maxViewportDimensions = { (1 << 14), (1 << 14) }, .viewportBoundsRange = { INT16_MIN, INT16_MAX }, .viewportSubPixelBits = 8, .minMemoryMapAlignment = 4096, /* A page */ - .minTexelBufferOffsetAlignment = 1, - .minUniformBufferOffsetAlignment = 4, - .minStorageBufferOffsetAlignment = 4, - .minTexelOffset = -32, - .maxTexelOffset = 31, + .minTexelBufferOffsetAlignment = 64, + .minUniformBufferOffsetAlignment = 64, + .minStorageBufferOffsetAlignment = 64, + .minTexelOffset = -16, + .maxTexelOffset = 15, .minTexelGatherOffset = -32, .maxTexelGatherOffset = 31, - .minInterpolationOffset = -2, - .maxInterpolationOffset = 2, - .subPixelInterpolationOffsetBits = 8, + .minInterpolationOffset = -0.5, + .maxInterpolationOffset = 0.4375, + .subPixelInterpolationOffsetBits = 4, .maxFramebufferWidth = (1 << 14), .maxFramebufferHeight = (1 << 14), .maxFramebufferLayers = (1 << 10), @@ -812,7 +868,7 @@ tu_GetPhysicalDeviceProperties(VkPhysicalDevice physicalDevice, .storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT, .maxSampleMaskWords = 1, .timestampComputeAndGraphics = true, - .timestampPeriod = 1, + .timestampPeriod = 1000000000.0 / 19200000.0, /* CP_ALWAYS_ON_COUNTER is fixed 19.2MHz */ .maxClipDistances = 8, .maxCullDistances = 8, .maxCombinedClipAndCullDistances = 8, @@ -890,6 +946,60 @@ tu_GetPhysicalDeviceProperties2(VkPhysicalDevice physicalDevice, properties->maxMemoryAllocationSize = 0xFFFFFFFFull; break; } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT: { + VkPhysicalDeviceTransformFeedbackPropertiesEXT *properties = + (VkPhysicalDeviceTransformFeedbackPropertiesEXT *)ext; + + properties->maxTransformFeedbackStreams = IR3_MAX_SO_STREAMS; + properties->maxTransformFeedbackBuffers = IR3_MAX_SO_BUFFERS; + properties->maxTransformFeedbackBufferSize = UINT32_MAX; + properties->maxTransformFeedbackStreamDataSize = 512; + properties->maxTransformFeedbackBufferDataSize = 512; + properties->maxTransformFeedbackBufferDataStride = 512; + properties->transformFeedbackQueries = true; + properties->transformFeedbackStreamsLinesTriangles = false; + properties->transformFeedbackRasterizationStreamSelect = false; + properties->transformFeedbackDraw = true; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT: { + VkPhysicalDeviceSampleLocationsPropertiesEXT *properties = + (VkPhysicalDeviceSampleLocationsPropertiesEXT *)ext; + properties->sampleLocationSampleCounts = 0; + if (pdevice->supported_extensions.EXT_sample_locations) { + properties->sampleLocationSampleCounts = + VK_SAMPLE_COUNT_1_BIT | VK_SAMPLE_COUNT_2_BIT | VK_SAMPLE_COUNT_4_BIT; + } + properties->maxSampleLocationGridSize = (VkExtent2D) { 1 , 1 }; + properties->sampleLocationCoordinateRange[0] = 0.0f; + properties->sampleLocationCoordinateRange[1] = 0.9375f; + properties->sampleLocationSubPixelBits = 4; + properties->variableSampleLocations = true; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES: { + VkPhysicalDeviceSamplerFilterMinmaxProperties *properties = + (VkPhysicalDeviceSamplerFilterMinmaxProperties *)ext; + properties->filterMinmaxImageComponentMapping = true; + properties->filterMinmaxSingleComponentFormats = true; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SUBGROUP_PROPERTIES: { + VkPhysicalDeviceSubgroupProperties *properties = + (VkPhysicalDeviceSubgroupProperties *)ext; + properties->subgroupSize = 64; + properties->supportedStages = VK_SHADER_STAGE_COMPUTE_BIT; + properties->supportedOperations = VK_SUBGROUP_FEATURE_BASIC_BIT | + VK_SUBGROUP_FEATURE_VOTE_BIT; + properties->quadOperationsInAllStages = false; + break; + } + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_PROPERTIES_EXT: { + VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *props = + (VkPhysicalDeviceVertexAttributeDivisorPropertiesEXT *)ext; + props->maxVertexAttribDivisor = UINT32_MAX; + break; + } default: break; } @@ -900,7 +1010,7 @@ static const VkQueueFamilyProperties tu_queue_family_properties = { .queueFlags = VK_QUEUE_GRAPHICS_BIT | VK_QUEUE_COMPUTE_BIT | VK_QUEUE_TRANSFER_BIT, .queueCount = 1, - .timestampValidBits = 64, + .timestampValidBits = 48, .minImageTransferGranularity = { 1, 1, 1 }, }; @@ -1014,6 +1124,61 @@ tu_get_device_extension_index(const char *name) return -1; } +struct PACKED bcolor_entry { + uint32_t fp32[4]; + uint16_t ui16[4]; + int16_t si16[4]; + uint16_t fp16[4]; + uint16_t rgb565; + uint16_t rgb5a1; + uint16_t rgba4; + uint8_t __pad0[2]; + uint8_t ui8[4]; + int8_t si8[4]; + uint32_t rgb10a2; + uint32_t z24; /* also s8? */ + uint16_t srgb[4]; /* appears to duplicate fp16[], but clamped, used for srgb */ + uint8_t __pad1[56]; +} border_color[] = { + [VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = {}, + [VK_BORDER_COLOR_INT_TRANSPARENT_BLACK] = {}, + [VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = { + .fp32[3] = 0x3f800000, + .ui16[3] = 0xffff, + .si16[3] = 0x7fff, + .fp16[3] = 0x3c00, + .rgb5a1 = 0x8000, + .rgba4 = 0xf000, + .ui8[3] = 0xff, + .si8[3] = 0x7f, + .rgb10a2 = 0xc0000000, + .srgb[3] = 0x3c00, + }, + [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { + .fp32[3] = 1, + .fp16[3] = 1, + }, + [VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = { + .fp32[0 ... 3] = 0x3f800000, + .ui16[0 ... 3] = 0xffff, + .si16[0 ... 3] = 0x7fff, + .fp16[0 ... 3] = 0x3c00, + .rgb565 = 0xffff, + .rgb5a1 = 0xffff, + .rgba4 = 0xffff, + .ui8[0 ... 3] = 0xff, + .si8[0 ... 3] = 0x7f, + .rgb10a2 = 0xffffffff, + .z24 = 0xffffff, + .srgb[0 ... 3] = 0x3c00, + }, + [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { + .fp32[0 ... 3] = 1, + .fp16[0 ... 3] = 1, + }, +}; + + VkResult tu_CreateDevice(VkPhysicalDevice physicalDevice, const VkDeviceCreateInfo *pCreateInfo, @@ -1075,7 +1240,7 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice, 8, VK_SYSTEM_ALLOCATION_SCOPE_DEVICE); if (!device->queues[qfi]) { result = VK_ERROR_OUT_OF_HOST_MEMORY; - goto fail; + goto fail_queues; } memset(device->queues[qfi], 0, @@ -1087,13 +1252,38 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice, result = tu_queue_init(device, &device->queues[qfi][q], qfi, q, queue_create->flags); if (result != VK_SUCCESS) - goto fail; + goto fail_queues; } } device->compiler = ir3_compiler_create(NULL, physical_device->gpu_id); if (!device->compiler) - goto fail; + goto fail_queues; + +#define VSC_DRAW_STRM_SIZE(pitch) ((pitch) * 32 + 0x100) /* extra size to store VSC_SIZE */ +#define VSC_PRIM_STRM_SIZE(pitch) ((pitch) * 32) + + device->vsc_draw_strm_pitch = 0x440 * 4; + device->vsc_prim_strm_pitch = 0x1040 * 4; + + result = tu_bo_init_new(device, &device->vsc_draw_strm, VSC_DRAW_STRM_SIZE(device->vsc_draw_strm_pitch)); + if (result != VK_SUCCESS) + goto fail_vsc_data; + + result = tu_bo_init_new(device, &device->vsc_prim_strm, VSC_PRIM_STRM_SIZE(device->vsc_prim_strm_pitch)); + if (result != VK_SUCCESS) + goto fail_vsc_data2; + + STATIC_ASSERT(sizeof(struct bcolor_entry) == 128); + result = tu_bo_init_new(device, &device->border_color, sizeof(border_color)); + if (result != VK_SUCCESS) + goto fail_border_color; + + result = tu_bo_map(device, &device->border_color); + if (result != VK_SUCCESS) + goto fail_border_color_map; + + memcpy(device->border_color.map, border_color, sizeof(border_color)); VkPipelineCacheCreateInfo ci; ci.sType = VK_STRUCTURE_TYPE_PIPELINE_CACHE_CREATE_INFO; @@ -1105,14 +1295,30 @@ tu_CreateDevice(VkPhysicalDevice physicalDevice, result = tu_CreatePipelineCache(tu_device_to_handle(device), &ci, NULL, &pc); if (result != VK_SUCCESS) - goto fail; + goto fail_pipeline_cache; device->mem_cache = tu_pipeline_cache_from_handle(pc); + for (unsigned i = 0; i < ARRAY_SIZE(device->scratch_bos); i++) + mtx_init(&device->scratch_bos[i].construct_mtx, mtx_plain); + *pDevice = tu_device_to_handle(device); return VK_SUCCESS; -fail: +fail_pipeline_cache: +fail_border_color_map: + tu_bo_finish(device, &device->border_color); + +fail_border_color: + tu_bo_finish(device, &device->vsc_prim_strm); + +fail_vsc_data2: + tu_bo_finish(device, &device->vsc_draw_strm); + +fail_vsc_data: + ralloc_free(device->compiler); + +fail_queues: for (unsigned i = 0; i < TU_MAX_QUEUE_FAMILIES; i++) { for (unsigned q = 0; q < device->queue_count[i]; q++) tu_queue_finish(&device->queues[i][q]); @@ -1120,9 +1326,6 @@ fail: vk_free(&device->alloc, device->queues[i]); } - if (device->compiler) - ralloc_free(device->compiler); - vk_free(&device->alloc, device); return result; } @@ -1135,6 +1338,9 @@ tu_DestroyDevice(VkDevice _device, const VkAllocationCallbacks *pAllocator) if (!device) return; + tu_bo_finish(device, &device->vsc_draw_strm); + tu_bo_finish(device, &device->vsc_prim_strm); + for (unsigned i = 0; i < TU_MAX_QUEUE_FAMILIES; i++) { for (unsigned q = 0; q < device->queue_count[i]; q++) tu_queue_finish(&device->queues[i][q]); @@ -1142,8 +1348,12 @@ tu_DestroyDevice(VkDevice _device, const VkAllocationCallbacks *pAllocator) vk_free(&device->alloc, device->queues[i]); } - /* the compiler does not use pAllocator */ - ralloc_free(device->compiler); + for (unsigned i = 0; i < ARRAY_SIZE(device->scratch_bos); i++) { + if (device->scratch_bos[i].initialized) + tu_bo_finish(device, &device->scratch_bos[i].bo); + } + + ir3_compiler_destroy(device->compiler); VkPipelineCache pc = tu_pipeline_cache_to_handle(device->mem_cache); tu_DestroyPipelineCache(tu_device_to_handle(device), pc, NULL); @@ -1151,6 +1361,51 @@ tu_DestroyDevice(VkDevice _device, const VkAllocationCallbacks *pAllocator) vk_free(&device->alloc, device); } +VkResult +tu_get_scratch_bo(struct tu_device *dev, uint64_t size, struct tu_bo **bo) +{ + unsigned size_log2 = MAX2(util_logbase2_ceil64(size), MIN_SCRATCH_BO_SIZE_LOG2); + unsigned index = size_log2 - MIN_SCRATCH_BO_SIZE_LOG2; + assert(index < ARRAY_SIZE(dev->scratch_bos)); + + for (unsigned i = index; i < ARRAY_SIZE(dev->scratch_bos); i++) { + if (p_atomic_read(&dev->scratch_bos[i].initialized)) { + /* Fast path: just return the already-allocated BO. */ + *bo = &dev->scratch_bos[i].bo; + return VK_SUCCESS; + } + } + + /* Slow path: actually allocate the BO. We take a lock because the process + * of allocating it is slow, and we don't want to block the CPU while it + * finishes. + */ + mtx_lock(&dev->scratch_bos[index].construct_mtx); + + /* Another thread may have allocated it already while we were waiting on + * the lock. We need to check this in order to avoid double-allocating. + */ + if (dev->scratch_bos[index].initialized) { + mtx_unlock(&dev->scratch_bos[index].construct_mtx); + *bo = &dev->scratch_bos[index].bo; + return VK_SUCCESS; + } + + unsigned bo_size = 1ull << size_log2; + VkResult result = tu_bo_init_new(dev, &dev->scratch_bos[index].bo, bo_size); + if (result != VK_SUCCESS) { + mtx_unlock(&dev->scratch_bos[index].construct_mtx); + return result; + } + + p_atomic_set(&dev->scratch_bos[index].initialized, true); + + mtx_unlock(&dev->scratch_bos[index].construct_mtx); + + *bo = &dev->scratch_bos[index].bo; + return VK_SUCCESS; +} + VkResult tu_EnumerateInstanceLayerProperties(uint32_t *pPropertyCount, VkLayerProperties *pProperties) @@ -1208,6 +1463,65 @@ tu_GetDeviceQueue(VkDevice _device, tu_GetDeviceQueue2(_device, &info, pQueue); } +static VkResult +tu_get_semaphore_syncobjs(const VkSemaphore *sems, + uint32_t sem_count, + bool wait, + struct drm_msm_gem_submit_syncobj **out, + uint32_t *out_count) +{ + uint32_t syncobj_count = 0; + struct drm_msm_gem_submit_syncobj *syncobjs; + + for (uint32_t i = 0; i < sem_count; ++i) { + TU_FROM_HANDLE(tu_semaphore, sem, sems[i]); + + struct tu_semaphore_part *part = + sem->temporary.kind != TU_SEMAPHORE_NONE ? + &sem->temporary : &sem->permanent; + + if (part->kind == TU_SEMAPHORE_SYNCOBJ) + ++syncobj_count; + } + + *out = NULL; + *out_count = syncobj_count; + if (!syncobj_count) + return VK_SUCCESS; + + *out = syncobjs = calloc(syncobj_count, sizeof (*syncobjs)); + if (!syncobjs) + return VK_ERROR_OUT_OF_HOST_MEMORY; + + for (uint32_t i = 0, j = 0; i < sem_count; ++i) { + TU_FROM_HANDLE(tu_semaphore, sem, sems[i]); + + struct tu_semaphore_part *part = + sem->temporary.kind != TU_SEMAPHORE_NONE ? + &sem->temporary : &sem->permanent; + + if (part->kind == TU_SEMAPHORE_SYNCOBJ) { + syncobjs[j].handle = part->syncobj; + syncobjs[j].flags = wait ? MSM_SUBMIT_SYNCOBJ_RESET : 0; + ++j; + } + } + + return VK_SUCCESS; +} + + +static void +tu_semaphores_remove_temp(struct tu_device *device, + const VkSemaphore *sems, + uint32_t sem_count) +{ + for (uint32_t i = 0; i < sem_count; ++i) { + TU_FROM_HANDLE(tu_semaphore, sem, sems[i]); + tu_semaphore_remove_temp(device, sem); + } +} + VkResult tu_QueueSubmit(VkQueue _queue, uint32_t submitCount, @@ -1215,13 +1529,34 @@ tu_QueueSubmit(VkQueue _queue, VkFence _fence) { TU_FROM_HANDLE(tu_queue, queue, _queue); + VkResult result; for (uint32_t i = 0; i < submitCount; ++i) { const VkSubmitInfo *submit = pSubmits + i; const bool last_submit = (i == submitCount - 1); + struct drm_msm_gem_submit_syncobj *in_syncobjs = NULL, *out_syncobjs = NULL; + uint32_t nr_in_syncobjs, nr_out_syncobjs; struct tu_bo_list bo_list; tu_bo_list_init(&bo_list); + result = tu_get_semaphore_syncobjs(pSubmits[i].pWaitSemaphores, + pSubmits[i].waitSemaphoreCount, + false, &in_syncobjs, &nr_in_syncobjs); + if (result != VK_SUCCESS) { + /* TODO: emit VK_ERROR_DEVICE_LOST */ + fprintf(stderr, "failed to allocate space for semaphore submission\n"); + abort(); + } + + result = tu_get_semaphore_syncobjs(pSubmits[i].pSignalSemaphores, + pSubmits[i].signalSemaphoreCount, + false, &out_syncobjs, &nr_out_syncobjs); + if (result != VK_SUCCESS) { + /* TODO: emit VK_ERROR_DEVICE_LOST */ + fprintf(stderr, "failed to allocate space for semaphore submission\n"); + abort(); + } + uint32_t entry_count = 0; for (uint32_t j = 0; j < submit->commandBufferCount; ++j) { TU_FROM_HANDLE(tu_cmd_buffer, cmdbuf, submit->pCommandBuffers[j]); @@ -1249,6 +1584,13 @@ tu_QueueSubmit(VkQueue _queue, } uint32_t flags = MSM_PIPE_3D0; + if (nr_in_syncobjs) { + flags |= MSM_SUBMIT_SYNCOBJ_IN; + } + if (nr_out_syncobjs) { + flags |= MSM_SUBMIT_SYNCOBJ_OUT; + } + if (last_submit) { flags |= MSM_SUBMIT_FENCE_FD_OUT; } @@ -1260,6 +1602,11 @@ tu_QueueSubmit(VkQueue _queue, .nr_bos = bo_list.count, .cmds = (uint64_t)(uintptr_t)cmds, .nr_cmds = entry_count, + .in_syncobjs = (uint64_t)(uintptr_t)in_syncobjs, + .out_syncobjs = (uint64_t)(uintptr_t)out_syncobjs, + .nr_in_syncobjs = nr_in_syncobjs, + .nr_out_syncobjs = nr_out_syncobjs, + .syncobj_stride = sizeof(struct drm_msm_gem_submit_syncobj), }; int ret = drmCommandWriteRead(queue->device->physical_device->local_fd, @@ -1271,10 +1618,16 @@ tu_QueueSubmit(VkQueue _queue, } tu_bo_list_destroy(&bo_list); + free(in_syncobjs); + free(out_syncobjs); + tu_semaphores_remove_temp(queue->device, pSubmits[i].pWaitSemaphores, + pSubmits[i].waitSemaphoreCount); if (last_submit) { /* no need to merge fences as queue execution is serialized */ tu_fence_update_fd(&queue->submit_fence, req.fence_fd); + } else if (last_submit) { + close(req.fence_fd); } } @@ -1321,7 +1674,7 @@ tu_EnumerateInstanceExtensionProperties(const char *pLayerName, return vk_error(NULL, VK_ERROR_LAYER_NOT_PRESENT); for (int i = 0; i < TU_INSTANCE_EXTENSION_COUNT; i++) { - if (tu_supported_instance_extensions.extensions[i]) { + if (tu_instance_extensions_supported.extensions[i]) { vk_outarray_append(&out, prop) { *prop = tu_instance_extensions[i]; } } } @@ -1541,7 +1894,7 @@ tu_GetBufferMemoryRequirements(VkDevice _device, TU_FROM_HANDLE(tu_buffer, buffer, _buffer); pMemoryRequirements->memoryTypeBits = 1; - pMemoryRequirements->alignment = 16; + pMemoryRequirements->alignment = 64; pMemoryRequirements->size = align64(buffer->size, pMemoryRequirements->alignment); } @@ -1564,8 +1917,8 @@ tu_GetImageMemoryRequirements(VkDevice _device, TU_FROM_HANDLE(tu_image, image, _image); pMemoryRequirements->memoryTypeBits = 1; - pMemoryRequirements->size = image->size; - pMemoryRequirements->alignment = image->alignment; + pMemoryRequirements->size = image->layout.size; + pMemoryRequirements->alignment = image->layout.base_align; } void @@ -1688,6 +2041,30 @@ tu_QueueBindSparse(VkQueue _queue, // Queue semaphore functions + +static void +tu_semaphore_part_destroy(struct tu_device *device, + struct tu_semaphore_part *part) +{ + switch(part->kind) { + case TU_SEMAPHORE_NONE: + break; + case TU_SEMAPHORE_SYNCOBJ: + drmSyncobjDestroy(device->physical_device->local_fd, part->syncobj); + break; + } + part->kind = TU_SEMAPHORE_NONE; +} + +static void +tu_semaphore_remove_temp(struct tu_device *device, + struct tu_semaphore *sem) +{ + if (sem->temporary.kind != TU_SEMAPHORE_NONE) { + tu_semaphore_part_destroy(device, &sem->temporary); + } +} + VkResult tu_CreateSemaphore(VkDevice _device, const VkSemaphoreCreateInfo *pCreateInfo, @@ -1702,6 +2079,21 @@ tu_CreateSemaphore(VkDevice _device, if (!sem) return vk_error(device->instance, VK_ERROR_OUT_OF_HOST_MEMORY); + const VkExportSemaphoreCreateInfo *export = + vk_find_struct_const(pCreateInfo->pNext, EXPORT_SEMAPHORE_CREATE_INFO); + VkExternalSemaphoreHandleTypeFlags handleTypes = + export ? export->handleTypes : 0; + + sem->permanent.kind = TU_SEMAPHORE_NONE; + sem->temporary.kind = TU_SEMAPHORE_NONE; + + if (handleTypes) { + if (drmSyncobjCreate(device->physical_device->local_fd, 0, &sem->permanent.syncobj) < 0) { + vk_free2(&device->alloc, pAllocator, sem); + return VK_ERROR_OUT_OF_HOST_MEMORY; + } + sem->permanent.kind = TU_SEMAPHORE_SYNCOBJ; + } *pSemaphore = tu_semaphore_to_handle(sem); return VK_SUCCESS; } @@ -1716,6 +2108,9 @@ tu_DestroySemaphore(VkDevice _device, if (!_semaphore) return; + tu_semaphore_part_destroy(device, &sem->permanent); + tu_semaphore_part_destroy(device, &sem->temporary); + vk_free2(&device->alloc, pAllocator, sem); } @@ -1762,6 +2157,8 @@ tu_DestroyEvent(VkDevice _device, if (!event) return; + + tu_bo_finish(device, &event->bo); vk_free2(&device->alloc, pAllocator, event); } @@ -1832,14 +2229,6 @@ tu_DestroyBuffer(VkDevice _device, vk_free2(&device->alloc, pAllocator, buffer); } -static uint32_t -tu_surface_max_layer_count(struct tu_image_view *iview) -{ - return iview->type == VK_IMAGE_VIEW_TYPE_3D - ? iview->extent.depth - : (iview->base_layer + iview->layer_count); -} - VkResult tu_CreateFramebuffer(VkDevice _device, const VkFramebufferCreateInfo *pCreateInfo, @@ -1866,11 +2255,6 @@ tu_CreateFramebuffer(VkDevice _device, VkImageView _iview = pCreateInfo->pAttachments[i]; struct tu_image_view *iview = tu_image_view_from_handle(_iview); framebuffer->attachments[i].attachment = iview; - - framebuffer->width = MIN2(framebuffer->width, iview->extent.width); - framebuffer->height = MIN2(framebuffer->height, iview->extent.height); - framebuffer->layers = - MIN2(framebuffer->layers, tu_surface_max_layer_count(iview)); } *pFramebuffer = tu_framebuffer_to_handle(framebuffer); @@ -1890,77 +2274,63 @@ tu_DestroyFramebuffer(VkDevice _device, vk_free2(&device->alloc, pAllocator, fb); } -static enum a6xx_tex_clamp -tu6_tex_wrap(VkSamplerAddressMode address_mode, bool *needs_border) -{ - switch (address_mode) { - case VK_SAMPLER_ADDRESS_MODE_REPEAT: - return A6XX_TEX_REPEAT; - case VK_SAMPLER_ADDRESS_MODE_MIRRORED_REPEAT: - return A6XX_TEX_MIRROR_REPEAT; - case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE: - return A6XX_TEX_CLAMP_TO_EDGE; - case VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER: - *needs_border = true; - return A6XX_TEX_CLAMP_TO_BORDER; - case VK_SAMPLER_ADDRESS_MODE_MIRROR_CLAMP_TO_EDGE: - /* only works for PoT.. need to emulate otherwise! */ - return A6XX_TEX_MIRROR_CLAMP; - default: - unreachable("illegal tex wrap mode"); - break; - } -} - -static enum a6xx_tex_filter -tu6_tex_filter(VkFilter filter, unsigned aniso) -{ - switch (filter) { - case VK_FILTER_NEAREST: - return A6XX_TEX_NEAREST; - case VK_FILTER_LINEAR: - return aniso ? A6XX_TEX_ANISO : A6XX_TEX_LINEAR; - case VK_FILTER_CUBIC_IMG: - default: - unreachable("illegal texture filter"); - break; - } -} - static void tu_init_sampler(struct tu_device *device, struct tu_sampler *sampler, const VkSamplerCreateInfo *pCreateInfo) { + const struct VkSamplerReductionModeCreateInfo *reduction = + vk_find_struct_const(pCreateInfo->pNext, SAMPLER_REDUCTION_MODE_CREATE_INFO); + const struct VkSamplerYcbcrConversionInfo *ycbcr_conversion = + vk_find_struct_const(pCreateInfo->pNext, SAMPLER_YCBCR_CONVERSION_INFO); + unsigned aniso = pCreateInfo->anisotropyEnable ? util_last_bit(MIN2((uint32_t)pCreateInfo->maxAnisotropy >> 1, 8)) : 0; bool miplinear = (pCreateInfo->mipmapMode == VK_SAMPLER_MIPMAP_MODE_LINEAR); - bool needs_border = false; + float min_lod = CLAMP(pCreateInfo->minLod, 0.0f, 4095.0f / 256.0f); + float max_lod = CLAMP(pCreateInfo->maxLod, 0.0f, 4095.0f / 256.0f); - sampler->state[0] = + sampler->descriptor[0] = COND(miplinear, A6XX_TEX_SAMP_0_MIPFILTER_LINEAR_NEAR) | A6XX_TEX_SAMP_0_XY_MAG(tu6_tex_filter(pCreateInfo->magFilter, aniso)) | A6XX_TEX_SAMP_0_XY_MIN(tu6_tex_filter(pCreateInfo->minFilter, aniso)) | A6XX_TEX_SAMP_0_ANISO(aniso) | - A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo->addressModeU, &needs_border)) | - A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo->addressModeV, &needs_border)) | - A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo->addressModeW, &needs_border)) | + A6XX_TEX_SAMP_0_WRAP_S(tu6_tex_wrap(pCreateInfo->addressModeU)) | + A6XX_TEX_SAMP_0_WRAP_T(tu6_tex_wrap(pCreateInfo->addressModeV)) | + A6XX_TEX_SAMP_0_WRAP_R(tu6_tex_wrap(pCreateInfo->addressModeW)) | A6XX_TEX_SAMP_0_LOD_BIAS(pCreateInfo->mipLodBias); - sampler->state[1] = + sampler->descriptor[1] = /* COND(!cso->seamless_cube_map, A6XX_TEX_SAMP_1_CUBEMAPSEAMLESSFILTOFF) | */ COND(pCreateInfo->unnormalizedCoordinates, A6XX_TEX_SAMP_1_UNNORM_COORDS) | - A6XX_TEX_SAMP_1_MIN_LOD(pCreateInfo->minLod) | - A6XX_TEX_SAMP_1_MAX_LOD(pCreateInfo->maxLod) | - COND(pCreateInfo->compareEnable, A6XX_TEX_SAMP_1_COMPARE_FUNC(pCreateInfo->compareOp)); - sampler->state[2] = 0; - sampler->state[3] = 0; + A6XX_TEX_SAMP_1_MIN_LOD(min_lod) | + A6XX_TEX_SAMP_1_MAX_LOD(max_lod) | + COND(pCreateInfo->compareEnable, + A6XX_TEX_SAMP_1_COMPARE_FUNC(tu6_compare_func(pCreateInfo->compareOp))); + /* This is an offset into the border_color BO, which we fill with all the + * possible Vulkan border colors in the correct order, so we can just use + * the Vulkan enum with no translation necessary. + */ + sampler->descriptor[2] = + A6XX_TEX_SAMP_2_BCOLOR_OFFSET((unsigned) pCreateInfo->borderColor * + sizeof(struct bcolor_entry)); + sampler->descriptor[3] = 0; + + if (reduction) { + sampler->descriptor[2] |= A6XX_TEX_SAMP_2_REDUCTION_MODE( + tu6_reduction_mode(reduction->reductionMode)); + } + + sampler->ycbcr_sampler = ycbcr_conversion ? + tu_sampler_ycbcr_conversion_from_handle(ycbcr_conversion->conversion) : NULL; + + if (sampler->ycbcr_sampler && + sampler->ycbcr_sampler->chroma_filter == VK_FILTER_LINEAR) { + sampler->descriptor[2] |= A6XX_TEX_SAMP_2_CHROMA_LINEAR; + } /* TODO: * A6XX_TEX_SAMP_1_MIPFILTER_LINEAR_FAR disables mipmapping, but vk has no NONE mipfilter? - * border color */ - - sampler->needs_border = needs_border; } VkResult @@ -2078,15 +2448,132 @@ tu_GetMemoryFdPropertiesKHR(VkDevice _device, return VK_SUCCESS; } +VkResult +tu_ImportSemaphoreFdKHR(VkDevice _device, + const VkImportSemaphoreFdInfoKHR *pImportSemaphoreFdInfo) +{ + TU_FROM_HANDLE(tu_device, device, _device); + TU_FROM_HANDLE(tu_semaphore, sem, pImportSemaphoreFdInfo->semaphore); + int ret; + struct tu_semaphore_part *dst = NULL; + + if (pImportSemaphoreFdInfo->flags & VK_SEMAPHORE_IMPORT_TEMPORARY_BIT) { + dst = &sem->temporary; + } else { + dst = &sem->permanent; + } + + uint32_t syncobj = dst->kind == TU_SEMAPHORE_SYNCOBJ ? dst->syncobj : 0; + + switch(pImportSemaphoreFdInfo->handleType) { + case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT: { + uint32_t old_syncobj = syncobj; + ret = drmSyncobjFDToHandle(device->physical_device->local_fd, pImportSemaphoreFdInfo->fd, &syncobj); + if (ret == 0) { + close(pImportSemaphoreFdInfo->fd); + if (old_syncobj) + drmSyncobjDestroy(device->physical_device->local_fd, old_syncobj); + } + break; + } + case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT: { + if (!syncobj) { + ret = drmSyncobjCreate(device->physical_device->local_fd, 0, &syncobj); + if (ret) + break; + } + if (pImportSemaphoreFdInfo->fd == -1) { + ret = drmSyncobjSignal(device->physical_device->local_fd, &syncobj, 1); + } else { + ret = drmSyncobjImportSyncFile(device->physical_device->local_fd, syncobj, pImportSemaphoreFdInfo->fd); + } + if (!ret) + close(pImportSemaphoreFdInfo->fd); + break; + } + default: + unreachable("Unhandled semaphore handle type"); + } + + if (ret) { + return VK_ERROR_INVALID_EXTERNAL_HANDLE; + } + dst->syncobj = syncobj; + dst->kind = TU_SEMAPHORE_SYNCOBJ; + + return VK_SUCCESS; +} + +VkResult +tu_GetSemaphoreFdKHR(VkDevice _device, + const VkSemaphoreGetFdInfoKHR *pGetFdInfo, + int *pFd) +{ + TU_FROM_HANDLE(tu_device, device, _device); + TU_FROM_HANDLE(tu_semaphore, sem, pGetFdInfo->semaphore); + int ret; + uint32_t syncobj_handle; + + if (sem->temporary.kind != TU_SEMAPHORE_NONE) { + assert(sem->temporary.kind == TU_SEMAPHORE_SYNCOBJ); + syncobj_handle = sem->temporary.syncobj; + } else { + assert(sem->permanent.kind == TU_SEMAPHORE_SYNCOBJ); + syncobj_handle = sem->permanent.syncobj; + } + + switch(pGetFdInfo->handleType) { + case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT: + ret = drmSyncobjHandleToFD(device->physical_device->local_fd, syncobj_handle, pFd); + break; + case VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT: + ret = drmSyncobjExportSyncFile(device->physical_device->local_fd, syncobj_handle, pFd); + if (!ret) { + if (sem->temporary.kind != TU_SEMAPHORE_NONE) { + tu_semaphore_part_destroy(device, &sem->temporary); + } else { + drmSyncobjReset(device->physical_device->local_fd, &syncobj_handle, 1); + } + } + break; + default: + unreachable("Unhandled semaphore handle type"); + } + + if (ret) + return vk_error(device->instance, VK_ERROR_INVALID_EXTERNAL_HANDLE); + return VK_SUCCESS; +} + + +static bool tu_has_syncobj(struct tu_physical_device *pdev) +{ + uint64_t value; + if (drmGetCap(pdev->local_fd, DRM_CAP_SYNCOBJ, &value)) + return false; + return value && pdev->msm_major_version == 1 && pdev->msm_minor_version >= 6; +} + void tu_GetPhysicalDeviceExternalSemaphoreProperties( VkPhysicalDevice physicalDevice, const VkPhysicalDeviceExternalSemaphoreInfo *pExternalSemaphoreInfo, VkExternalSemaphoreProperties *pExternalSemaphoreProperties) { - pExternalSemaphoreProperties->exportFromImportedHandleTypes = 0; - pExternalSemaphoreProperties->compatibleHandleTypes = 0; - pExternalSemaphoreProperties->externalSemaphoreFeatures = 0; + TU_FROM_HANDLE(tu_physical_device, pdev, physicalDevice); + + if (tu_has_syncobj(pdev) && + (pExternalSemaphoreInfo->handleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT || + pExternalSemaphoreInfo->handleType == VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT)) { + pExternalSemaphoreProperties->exportFromImportedHandleTypes = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT | VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT; + pExternalSemaphoreProperties->compatibleHandleTypes = VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_OPAQUE_FD_BIT | VK_EXTERNAL_SEMAPHORE_HANDLE_TYPE_SYNC_FD_BIT; + pExternalSemaphoreProperties->externalSemaphoreFeatures = VK_EXTERNAL_SEMAPHORE_FEATURE_EXPORTABLE_BIT | + VK_EXTERNAL_SEMAPHORE_FEATURE_IMPORTABLE_BIT; + } else { + pExternalSemaphoreProperties->exportFromImportedHandleTypes = 0; + pExternalSemaphoreProperties->compatibleHandleTypes = 0; + pExternalSemaphoreProperties->externalSemaphoreFeatures = 0; + } } void @@ -2153,3 +2640,16 @@ tu_GetDeviceGroupPeerMemoryFeatures( VK_PEER_MEMORY_FEATURE_GENERIC_SRC_BIT | VK_PEER_MEMORY_FEATURE_GENERIC_DST_BIT; } + +void tu_GetPhysicalDeviceMultisamplePropertiesEXT( + VkPhysicalDevice physicalDevice, + VkSampleCountFlagBits samples, + VkMultisamplePropertiesEXT* pMultisampleProperties) +{ + TU_FROM_HANDLE(tu_physical_device, pdevice, physicalDevice); + + if (samples <= VK_SAMPLE_COUNT_4_BIT && pdevice->supported_extensions.EXT_sample_locations) + pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 1, 1 }; + else + pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 0, 0 }; +}