#include "anv_private.h"
#include "mesa/main/git_sha1.h"
#include "util/strtod.h"
+#include "util/debug.h"
#include "gen7_pack.h"
fprintf(stderr, "WARNING: Haswell Vulkan support is incomplete\n");
} else if (device->info->gen == 7 && !device->info->is_baytrail) {
fprintf(stderr, "WARNING: Ivy Bridge Vulkan support is incomplete\n");
- } else if (device->info->gen == 9) {
- fprintf(stderr, "WARNING: Skylake Vulkan support is incomplete\n");
- } else if (device->info->gen == 8 && !device->info->is_cherryview) {
- /* Broadwell is as fully supported as anything */
+ } else if (device->info->gen == 7 && device->info->is_baytrail) {
+ fprintf(stderr, "WARNING: Bay Trail Vulkan support is incomplete\n");
+ } else if (device->info->gen >= 8) {
+ /* Broadwell, Cherryview, Skylake, Broxton, Kabylake is as fully
+ * supported as anything */
} else {
- result = vk_errorf(VK_UNSUPPORTED,
+ result = vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER,
"Vulkan not yet supported on %s", device->name);
goto fail;
}
goto fail;
}
- if (!anv_gem_get_param(fd, I915_PARAM_HAS_LLC)) {
+ if (!device->info->has_llc &&
+ anv_gem_get_param(fd, I915_PARAM_MMAP_VERSION) < 1) {
result = vk_errorf(VK_ERROR_INITIALIZATION_FAILED,
- "non-llc gpu");
+ "kernel missing wc mmap");
goto fail;
}
+ bool swizzled = anv_gem_get_bit6_swizzle(fd, I915_TILING_X);
+
close(fd);
brw_process_intel_debug_variable();
device->compiler->shader_debug_log = compiler_debug_log;
device->compiler->shader_perf_log = compiler_perf_log;
- isl_device_init(&device->isl_dev, device->info);
+ /* XXX: Actually detect bit6 swizzling */
+ isl_device_init(&device->isl_dev, device->info, swizzled);
return VK_SUCCESS;
static const VkExtensionProperties global_extensions[] = {
{
- .extensionName = VK_EXT_KHR_SWAPCHAIN_EXTENSION_NAME,
- .specVersion = 17,
+ .extensionName = VK_KHR_SURFACE_EXTENSION_NAME,
+ .specVersion = 25,
+ },
+ {
+ .extensionName = VK_KHR_XCB_SURFACE_EXTENSION_NAME,
+ .specVersion = 5,
+ },
+#ifdef HAVE_WAYLAND_PLATFORM
+ {
+ .extensionName = VK_KHR_WAYLAND_SURFACE_EXTENSION_NAME,
+ .specVersion = 4,
},
+#endif
};
static const VkExtensionProperties device_extensions[] = {
{
- .extensionName = VK_EXT_KHR_DEVICE_SWAPCHAIN_EXTENSION_NAME,
- .specVersion = 53,
+ .extensionName = VK_KHR_SWAPCHAIN_EXTENSION_NAME,
+ .specVersion = 67,
},
};
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_INSTANCE_CREATE_INFO);
- if (pCreateInfo->pApplicationInfo->apiVersion != VK_MAKE_VERSION(0, 170, 2))
- return vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
+ uint32_t client_version = pCreateInfo->pApplicationInfo ?
+ pCreateInfo->pApplicationInfo->apiVersion :
+ VK_MAKE_VERSION(1, 0, 0);
+ if (VK_MAKE_VERSION(1, 0, 0) > client_version ||
+ client_version > VK_MAKE_VERSION(1, 0, 3)) {
+ return vk_errorf(VK_ERROR_INCOMPATIBLE_DRIVER,
+ "Client requested version %d.%d.%d",
+ VK_VERSION_MAJOR(client_version),
+ VK_VERSION_MINOR(client_version),
+ VK_VERSION_PATCH(client_version));
+ }
- for (uint32_t i = 0; i < pCreateInfo->enabledExtensionNameCount; i++) {
+ for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
bool found = false;
for (uint32_t j = 0; j < ARRAY_SIZE(global_extensions); j++) {
if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
}
instance = anv_alloc2(&default_alloc, pAllocator, sizeof(*instance), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
if (!instance)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
else
instance->alloc = default_alloc;
- instance->apiVersion = pCreateInfo->pApplicationInfo->apiVersion;
+ instance->apiVersion = client_version;
instance->physicalDeviceCount = -1;
_mesa_locale_init();
if (instance->physicalDeviceCount < 0) {
result = anv_physical_device_init(&instance->physicalDevice,
instance, "/dev/dri/renderD128");
- if (result == VK_UNSUPPORTED) {
+ if (result == VK_ERROR_INCOMPATIBLE_DRIVER) {
instance->physicalDeviceCount = 0;
} else if (result == VK_SUCCESS) {
instance->physicalDeviceCount = 1;
anv_finishme("Get correct values for PhysicalDeviceFeatures");
*pFeatures = (VkPhysicalDeviceFeatures) {
- .robustBufferAccess = false,
+ .robustBufferAccess = true,
.fullDrawIndexUint32 = false,
.imageCubeArray = false,
.independentBlend = false,
.geometryShader = true,
.tessellationShader = false,
.sampleRateShading = false,
- .dualSourceBlend = true,
+ .dualSrcBlend = true,
.logicOp = true,
- .multiDrawIndirect = true,
- .depthClip = false,
+ .multiDrawIndirect = false,
+ .drawIndirectFirstInstance = false,
+ .depthClamp = false,
.depthBiasClamp = false,
.fillModeNonSolid = true,
.depthBounds = false,
.wideLines = true,
.largePoints = true,
+ .alphaToOne = true,
+ .multiViewport = true,
+ .samplerAnisotropy = false, /* FINISHME */
.textureCompressionETC2 = true,
.textureCompressionASTC_LDR = true,
.textureCompressionBC = true,
- .occlusionQueryNonConservative = false, /* FINISHME */
+ .occlusionQueryPrecise = false, /* FINISHME */
.pipelineStatisticsQuery = true,
- .vertexSideEffects = false,
- .tessellationSideEffects = false,
- .geometrySideEffects = false,
- .fragmentSideEffects = false,
- .shaderTessellationPointSize = false,
- .shaderGeometryPointSize = true,
+ .vertexPipelineStoresAndAtomics = false,
+ .fragmentStoresAndAtomics = true,
+ .shaderTessellationAndGeometryPointSize = true,
.shaderImageGatherExtended = true,
.shaderStorageImageExtendedFormats = false,
.shaderStorageImageMultisample = false,
.shaderSampledImageArrayDynamicIndexing = false,
.shaderStorageBufferArrayDynamicIndexing = false,
.shaderStorageImageArrayDynamicIndexing = false,
+ .shaderStorageImageReadWithoutFormat = false,
+ .shaderStorageImageWriteWithoutFormat = true,
.shaderClipDistance = false,
.shaderCullDistance = false,
.shaderFloat64 = false,
.shaderInt64 = false,
.shaderInt16 = false,
.alphaToOne = true,
+ .variableMultisampleRate = false,
+ .inheritedQueries = false,
};
}
+void
+anv_device_get_cache_uuid(void *uuid)
+{
+ memset(uuid, 0, VK_UUID_SIZE);
+ snprintf(uuid, VK_UUID_SIZE, "anv-%s", MESA_GIT_SHA1 + 4);
+}
+
void anv_GetPhysicalDeviceProperties(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceProperties* pProperties)
anv_finishme("Get correct values for VkPhysicalDeviceLimits");
+ const float time_stamp_base = devinfo->gen >= 9 ? 83.333 : 80.0;
+
+ VkSampleCountFlags sample_counts =
+ isl_device_get_sample_counts(&pdevice->isl_dev);
+
VkPhysicalDeviceLimits limits = {
.maxImageDimension1D = (1 << 14),
.maxImageDimension2D = (1 << 14),
.maxImageDimension3D = (1 << 10),
.maxImageDimensionCube = (1 << 14),
.maxImageArrayLayers = (1 << 10),
-
- /* Broadwell supports 1, 2, 4, and 8 samples. */
- .sampleCounts = 4,
-
- .maxTexelBufferSize = (1 << 14),
- .maxUniformBufferSize = UINT32_MAX,
- .maxStorageBufferSize = UINT32_MAX,
+ .maxTexelBufferElements = 128 * 1024 * 1024,
+ .maxUniformBufferRange = UINT32_MAX,
+ .maxStorageBufferRange = UINT32_MAX,
.maxPushConstantsSize = MAX_PUSH_CONSTANTS_SIZE,
.maxMemoryAllocationCount = UINT32_MAX,
+ .maxSamplerAllocationCount = 64 * 1024,
.bufferImageGranularity = 64, /* A cache line */
.sparseAddressSpaceSize = 0,
.maxBoundDescriptorSets = MAX_SETS,
- .maxDescriptorSets = UINT32_MAX,
.maxPerStageDescriptorSamplers = 64,
.maxPerStageDescriptorUniformBuffers = 64,
.maxPerStageDescriptorStorageBuffers = 64,
.maxPerStageDescriptorSampledImages = 64,
.maxPerStageDescriptorStorageImages = 64,
+ .maxPerStageDescriptorInputAttachments = 64,
+ .maxPerStageResources = 128,
.maxDescriptorSetSamplers = 256,
.maxDescriptorSetUniformBuffers = 256,
.maxDescriptorSetUniformBuffersDynamic = 256,
.maxDescriptorSetStorageBuffersDynamic = 256,
.maxDescriptorSetSampledImages = 256,
.maxDescriptorSetStorageImages = 256,
+ .maxDescriptorSetInputAttachments = 256,
.maxVertexInputAttributes = 32,
.maxVertexInputBindings = 32,
- .maxVertexInputAttributeOffset = 256,
- .maxVertexInputBindingStride = 256,
- .maxVertexOutputComponents = 32,
- .maxTessGenLevel = 0,
- .maxTessPatchSize = 0,
- .maxTessControlPerVertexInputComponents = 0,
- .maxTessControlPerVertexOutputComponents = 0,
- .maxTessControlPerPatchOutputComponents = 0,
- .maxTessControlTotalOutputComponents = 0,
- .maxTessEvaluationInputComponents = 0,
- .maxTessEvaluationOutputComponents = 0,
- .maxGeometryShaderInvocations = 6,
- .maxGeometryInputComponents = 16,
- .maxGeometryOutputComponents = 16,
- .maxGeometryOutputVertices = 16,
- .maxGeometryTotalOutputComponents = 16,
- .maxFragmentInputComponents = 16,
- .maxFragmentOutputBuffers = 8,
- .maxFragmentDualSourceBuffers = 2,
+ .maxVertexInputAttributeOffset = 2047,
+ .maxVertexInputBindingStride = 2048,
+ .maxVertexOutputComponents = 128,
+ .maxTessellationGenerationLevel = 0,
+ .maxTessellationPatchSize = 0,
+ .maxTessellationControlPerVertexInputComponents = 0,
+ .maxTessellationControlPerVertexOutputComponents = 0,
+ .maxTessellationControlPerPatchOutputComponents = 0,
+ .maxTessellationControlTotalOutputComponents = 0,
+ .maxTessellationEvaluationInputComponents = 0,
+ .maxTessellationEvaluationOutputComponents = 0,
+ .maxGeometryShaderInvocations = 32,
+ .maxGeometryInputComponents = 64,
+ .maxGeometryOutputComponents = 128,
+ .maxGeometryOutputVertices = 256,
+ .maxGeometryTotalOutputComponents = 1024,
+ .maxFragmentInputComponents = 128,
+ .maxFragmentOutputAttachments = 8,
+ .maxFragmentDualSrcAttachments = 2,
.maxFragmentCombinedOutputResources = 8,
- .maxComputeSharedMemorySize = 1024,
- .maxComputeWorkGroupCount = {
- 16 * devinfo->max_cs_threads,
- 16 * devinfo->max_cs_threads,
- 16 * devinfo->max_cs_threads,
- },
+ .maxComputeSharedMemorySize = 32768,
+ .maxComputeWorkGroupCount = { 65535, 65535, 65535 },
.maxComputeWorkGroupInvocations = 16 * devinfo->max_cs_threads,
.maxComputeWorkGroupSize = {
16 * devinfo->max_cs_threads,
.subTexelPrecisionBits = 4 /* FIXME */,
.mipmapPrecisionBits = 4 /* FIXME */,
.maxDrawIndexedIndexValue = UINT32_MAX,
- .maxDrawIndirectInstanceCount = UINT32_MAX,
- .primitiveRestartForPatches = UINT32_MAX,
+ .maxDrawIndirectCount = UINT32_MAX,
.maxSamplerLodBias = 16,
.maxSamplerAnisotropy = 16,
.maxViewports = MAX_VIEWPORTS,
.maxViewportDimensions = { (1 << 14), (1 << 14) },
- .viewportBoundsRange = { -1.0, 1.0 }, /* FIXME */
+ .viewportBoundsRange = { -16384.0, 16384.0 },
.viewportSubPixelBits = 13, /* We take a float? */
- .minMemoryMapAlignment = 64, /* A cache line */
+ .minMemoryMapAlignment = 4096, /* A page */
.minTexelBufferOffsetAlignment = 1,
.minUniformBufferOffsetAlignment = 1,
.minStorageBufferOffsetAlignment = 1,
- .minTexelOffset = 0, /* FIXME */
- .maxTexelOffset = 0, /* FIXME */
- .minTexelGatherOffset = 0, /* FIXME */
- .maxTexelGatherOffset = 0, /* FIXME */
+ .minTexelOffset = -8,
+ .maxTexelOffset = 7,
+ .minTexelGatherOffset = -8,
+ .maxTexelGatherOffset = 7,
.minInterpolationOffset = 0, /* FIXME */
.maxInterpolationOffset = 0, /* FIXME */
.subPixelInterpolationOffsetBits = 0, /* FIXME */
.maxFramebufferWidth = (1 << 14),
.maxFramebufferHeight = (1 << 14),
.maxFramebufferLayers = (1 << 10),
- .maxFramebufferColorSamples = 8,
- .maxFramebufferDepthSamples = 8,
- .maxFramebufferStencilSamples = 8,
+ .framebufferColorSampleCounts = sample_counts,
+ .framebufferDepthSampleCounts = sample_counts,
+ .framebufferStencilSampleCounts = sample_counts,
+ .framebufferNoAttachmentsSampleCounts = sample_counts,
.maxColorAttachments = MAX_RTS,
- .maxSampledImageColorSamples = 8,
- .maxSampledImageDepthSamples = 8,
- .maxSampledImageIntegerSamples = 1,
- .maxStorageImageSamples = 1,
+ .sampledImageColorSampleCounts = sample_counts,
+ .sampledImageIntegerSampleCounts = VK_SAMPLE_COUNT_1_BIT,
+ .sampledImageDepthSampleCounts = sample_counts,
+ .sampledImageStencilSampleCounts = sample_counts,
+ .storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
.maxSampleMaskWords = 1,
- .timestampFrequency = 1000 * 1000 * 1000 / 80,
+ .timestampComputeAndGraphics = false,
+ .timestampPeriod = time_stamp_base / (1000 * 1000 * 1000),
.maxClipDistances = 0 /* FIXME */,
.maxCullDistances = 0 /* FIXME */,
.maxCombinedClipAndCullDistances = 0 /* FIXME */,
+ .discreteQueuePriorities = 1,
.pointSizeRange = { 0.125, 255.875 },
.lineWidthRange = { 0.0, 7.9921875 },
.pointSizeGranularity = (1.0 / 8.0),
.lineWidthGranularity = (1.0 / 128.0),
+ .strictLines = false, /* FINISHME */
+ .standardSampleLocations = true,
+ .optimalBufferCopyOffsetAlignment = 128,
+ .optimalBufferCopyRowPitchAlignment = 128,
+ .nonCoherentAtomSize = 64,
};
*pProperties = (VkPhysicalDeviceProperties) {
- .apiVersion = VK_MAKE_VERSION(0, 170, 2),
+ .apiVersion = VK_MAKE_VERSION(1, 0, 2),
.driverVersion = 1,
.vendorID = 0x8086,
.deviceID = pdevice->chipset_id,
};
strcpy(pProperties->deviceName, pdevice->name);
- snprintf((char *)pProperties->pipelineCacheUUID, VK_UUID_SIZE,
- "anv-%s", MESA_GIT_SHA1 + 4);
+ anv_device_get_cache_uuid(pProperties->pipelineCacheUUID);
}
void anv_GetPhysicalDeviceQueueFamilyProperties(
VK_QUEUE_COMPUTE_BIT |
VK_QUEUE_TRANSFER_BIT,
.queueCount = 1,
- .supportsTimestamps = true,
+ .timestampValidBits = 36, /* XXX: Real value here */
+ .minImageTransferGranularity = (VkExtent3D) { 1, 1, 1 },
};
}
*/
heap_size = 3 * physical_device->aperture_size / 4;
- /* The property flags below are valid only for llc platforms. */
- pMemoryProperties->memoryTypeCount = 1;
- pMemoryProperties->memoryTypes[0] = (VkMemoryType) {
- .propertyFlags = VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT,
- .heapIndex = 1,
- };
+ if (physical_device->info->has_llc) {
+ /* Big core GPUs share LLC with the CPU and thus one memory type can be
+ * both cached and coherent at the same time.
+ */
+ pMemoryProperties->memoryTypeCount = 1;
+ pMemoryProperties->memoryTypes[0] = (VkMemoryType) {
+ .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
+ VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+ VK_MEMORY_PROPERTY_HOST_COHERENT_BIT |
+ VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
+ .heapIndex = 0,
+ };
+ } else {
+ /* The spec requires that we expose a host-visible, coherent memory
+ * type, but Atom GPUs don't share LLC. Thus we offer two memory types
+ * to give the application a choice between cached, but not coherent and
+ * coherent but uncached (WC though).
+ */
+ pMemoryProperties->memoryTypeCount = 2;
+ pMemoryProperties->memoryTypes[0] = (VkMemoryType) {
+ .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
+ VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+ VK_MEMORY_PROPERTY_HOST_COHERENT_BIT,
+ .heapIndex = 0,
+ };
+ pMemoryProperties->memoryTypes[1] = (VkMemoryType) {
+ .propertyFlags = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT |
+ VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT |
+ VK_MEMORY_PROPERTY_HOST_CACHED_BIT,
+ .heapIndex = 0,
+ };
+ }
pMemoryProperties->memoryHeapCount = 1;
pMemoryProperties->memoryHeaps[0] = (VkMemoryHeap) {
.size = heap_size,
- .flags = VK_MEMORY_HEAP_HOST_LOCAL_BIT,
+ .flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
};
}
return anv_lookup_entrypoint(pName);
}
+/* The loader wants us to expose a second GetInstanceProcAddr function
+ * to work around certain LD_PRELOAD issues seen in apps.
+ */
+VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(
+ VkInstance instance,
+ const char* pName);
+
+VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vk_icdGetInstanceProcAddr(
+ VkInstance instance,
+ const char* pName)
+{
+ return anv_GetInstanceProcAddr(instance, pName);
+}
+
PFN_vkVoidFunction anv_GetDeviceProcAddr(
VkDevice device,
const char* pName)
{
}
+static struct anv_state
+anv_state_pool_emit_data(struct anv_state_pool *pool, size_t size, size_t align, const void *p)
+{
+ struct anv_state state;
+
+ state = anv_state_pool_alloc(pool, size, align);
+ memcpy(state.map, p, size);
+
+ if (!pool->block_pool->device->info.has_llc)
+ anv_state_clflush(state);
+
+ return state;
+}
+
+struct gen8_border_color {
+ union {
+ float float32[4];
+ uint32_t uint32[4];
+ };
+ /* Pad out to 64 bytes */
+ uint32_t _pad[12];
+};
+
static void
anv_device_init_border_colors(struct anv_device *device)
{
- static const VkClearColorValue border_colors[] = {
+ static const struct gen8_border_color border_colors[] = {
[VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 0.0 } },
[VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK] = { .float32 = { 0.0, 0.0, 0.0, 1.0 } },
[VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE] = { .float32 = { 1.0, 1.0, 1.0, 1.0 } },
[VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
};
- device->border_colors =
- anv_state_pool_alloc(&device->dynamic_state_pool,
- sizeof(border_colors), 32);
- memcpy(device->border_colors.map, border_colors, sizeof(border_colors));
+ device->border_colors = anv_state_pool_emit_data(&device->dynamic_state_pool,
+ sizeof(border_colors), 64,
+ border_colors);
+}
+
+VkResult
+anv_device_submit_simple_batch(struct anv_device *device,
+ struct anv_batch *batch)
+{
+ struct drm_i915_gem_execbuffer2 execbuf;
+ struct drm_i915_gem_exec_object2 exec2_objects[1];
+ struct anv_bo bo;
+ VkResult result = VK_SUCCESS;
+ uint32_t size;
+ int64_t timeout;
+ int ret;
+
+ /* Kernel driver requires 8 byte aligned batch length */
+ size = align_u32(batch->next - batch->start, 8);
+ assert(size < device->batch_bo_pool.bo_size);
+ result = anv_bo_pool_alloc(&device->batch_bo_pool, &bo);
+ if (result != VK_SUCCESS)
+ return result;
+
+ memcpy(bo.map, batch->start, size);
+ if (!device->info.has_llc)
+ anv_clflush_range(bo.map, size);
+
+ exec2_objects[0].handle = bo.gem_handle;
+ exec2_objects[0].relocation_count = 0;
+ exec2_objects[0].relocs_ptr = 0;
+ exec2_objects[0].alignment = 0;
+ exec2_objects[0].offset = bo.offset;
+ exec2_objects[0].flags = 0;
+ exec2_objects[0].rsvd1 = 0;
+ exec2_objects[0].rsvd2 = 0;
+
+ execbuf.buffers_ptr = (uintptr_t) exec2_objects;
+ execbuf.buffer_count = 1;
+ execbuf.batch_start_offset = 0;
+ execbuf.batch_len = size;
+ execbuf.cliprects_ptr = 0;
+ execbuf.num_cliprects = 0;
+ execbuf.DR1 = 0;
+ execbuf.DR4 = 0;
+
+ execbuf.flags =
+ I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
+ execbuf.rsvd1 = device->context_id;
+ execbuf.rsvd2 = 0;
+
+ ret = anv_gem_execbuffer(device, &execbuf);
+ if (ret != 0) {
+ /* We don't know the real error. */
+ result = vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY, "execbuf2 failed: %m");
+ goto fail;
+ }
+
+ timeout = INT64_MAX;
+ ret = anv_gem_wait(device, bo.gem_handle, &timeout);
+ if (ret != 0) {
+ /* We don't know the real error. */
+ result = vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY, "execbuf2 failed: %m");
+ goto fail;
+ }
+
+ fail:
+ anv_bo_pool_free(&device->batch_bo_pool, &bo);
+
+ return result;
}
VkResult anv_CreateDevice(
VkDevice* pDevice)
{
ANV_FROM_HANDLE(anv_physical_device, physical_device, physicalDevice);
+ VkResult result;
struct anv_device *device;
assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_DEVICE_CREATE_INFO);
- for (uint32_t i = 0; i < pCreateInfo->enabledExtensionNameCount; i++) {
+ for (uint32_t i = 0; i < pCreateInfo->enabledExtensionCount; i++) {
bool found = false;
for (uint32_t j = 0; j < ARRAY_SIZE(device_extensions); j++) {
if (strcmp(pCreateInfo->ppEnabledExtensionNames[i],
device = anv_alloc2(&physical_device->instance->alloc, pAllocator,
sizeof(*device), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
if (!device)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
device->instance = physical_device->instance;
+ device->chipset_id = physical_device->chipset_id;
if (pAllocator)
device->alloc = *pAllocator;
/* XXX(chadv): Can we dup() physicalDevice->fd here? */
device->fd = open(physical_device->path, O_RDWR | O_CLOEXEC);
- if (device->fd == -1)
+ if (device->fd == -1) {
+ result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
goto fail_device;
+ }
device->context_id = anv_gem_create_context(device);
- if (device->context_id == -1)
+ if (device->context_id == -1) {
+ result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
goto fail_fd;
+ }
+
+ device->info = *physical_device->info;
+ device->isl_dev = physical_device->isl_dev;
pthread_mutex_init(&device->mutex, NULL);
anv_bo_pool_init(&device->batch_bo_pool, device, ANV_CMD_BUFFER_BATCH_SIZE);
- anv_block_pool_init(&device->dynamic_state_block_pool, device, 2048);
+ anv_block_pool_init(&device->dynamic_state_block_pool, device, 16384);
anv_state_pool_init(&device->dynamic_state_pool,
&device->dynamic_state_block_pool);
- anv_block_pool_init(&device->instruction_block_pool, device, 4096);
+ anv_block_pool_init(&device->instruction_block_pool, device, 128 * 1024);
+ anv_pipeline_cache_init(&device->default_pipeline_cache, device);
+
anv_block_pool_init(&device->surface_state_block_pool, device, 4096);
anv_state_pool_init(&device->surface_state_pool,
anv_block_pool_init(&device->scratch_block_pool, device, 0x10000);
- device->info = *physical_device->info;
- device->isl_dev = physical_device->isl_dev;
-
anv_queue_init(device, &device->queue);
- anv_device_init_meta(device);
+ switch (device->info.gen) {
+ case 7:
+ if (!device->info.is_haswell)
+ result = gen7_init_device_state(device);
+ else
+ result = gen75_init_device_state(device);
+ break;
+ case 8:
+ result = gen8_init_device_state(device);
+ break;
+ case 9:
+ result = gen9_init_device_state(device);
+ break;
+ default:
+ /* Shouldn't get here as we don't create physical devices for any other
+ * gens. */
+ unreachable("unhandled gen");
+ }
+ if (result != VK_SUCCESS)
+ goto fail_fd;
+
+ result = anv_device_init_meta(device);
+ if (result != VK_SUCCESS)
+ goto fail_fd;
anv_device_init_border_colors(device);
fail_device:
anv_free(&device->alloc, device);
- return vk_error(VK_ERROR_INITIALIZATION_FAILED);
+ return result;
}
void anv_DestroyDevice(
close(device->fd);
+ pthread_mutex_destroy(&device->mutex);
+
anv_free(&device->alloc, device);
}
VkResult anv_QueueSubmit(
VkQueue _queue,
- uint32_t commandBufferCount,
- const VkCommandBuffer* pCommandBuffers,
+ uint32_t submitCount,
+ const VkSubmitInfo* pSubmits,
VkFence _fence)
{
ANV_FROM_HANDLE(anv_queue, queue, _queue);
struct anv_device *device = queue->device;
int ret;
- for (uint32_t i = 0; i < commandBufferCount; i++) {
- ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer, pCommandBuffers[i]);
-
- assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
-
- ret = anv_gem_execbuffer(device, &cmd_buffer->execbuf2.execbuf);
- if (ret != 0) {
- /* We don't know the real error. */
- return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY,
- "execbuf2 failed: %m");
- }
+ for (uint32_t i = 0; i < submitCount; i++) {
+ for (uint32_t j = 0; j < pSubmits[i].commandBufferCount; j++) {
+ ANV_FROM_HANDLE(anv_cmd_buffer, cmd_buffer,
+ pSubmits[i].pCommandBuffers[j]);
+ assert(cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY);
- if (fence) {
- ret = anv_gem_execbuffer(device, &fence->execbuf);
+ ret = anv_gem_execbuffer(device, &cmd_buffer->execbuf2.execbuf);
if (ret != 0) {
/* We don't know the real error. */
return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY,
"execbuf2 failed: %m");
}
+
+ for (uint32_t k = 0; k < cmd_buffer->execbuf2.bo_count; k++)
+ cmd_buffer->execbuf2.bos[k]->offset = cmd_buffer->execbuf2.objects[k].offset;
}
+ }
- for (uint32_t i = 0; i < cmd_buffer->execbuf2.bo_count; i++)
- cmd_buffer->execbuf2.bos[i]->offset = cmd_buffer->execbuf2.objects[i].offset;
+ if (fence) {
+ ret = anv_gem_execbuffer(device, &fence->execbuf);
+ if (ret != 0) {
+ /* We don't know the real error. */
+ return vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY,
+ "execbuf2 failed: %m");
+ }
}
return VK_SUCCESS;
VkDevice _device)
{
ANV_FROM_HANDLE(anv_device, device, _device);
- struct anv_state state;
struct anv_batch batch;
- struct drm_i915_gem_execbuffer2 execbuf;
- struct drm_i915_gem_exec_object2 exec2_objects[1];
- struct anv_bo *bo = NULL;
- VkResult result;
- int64_t timeout;
- int ret;
- state = anv_state_pool_alloc(&device->dynamic_state_pool, 32, 32);
- bo = &device->dynamic_state_pool.block_pool->bo;
- batch.start = batch.next = state.map;
- batch.end = state.map + 32;
+ uint32_t cmds[8];
+ batch.start = batch.next = cmds;
+ batch.end = (void *) cmds + sizeof(cmds);
+
anv_batch_emit(&batch, GEN7_MI_BATCH_BUFFER_END);
anv_batch_emit(&batch, GEN7_MI_NOOP);
- exec2_objects[0].handle = bo->gem_handle;
- exec2_objects[0].relocation_count = 0;
- exec2_objects[0].relocs_ptr = 0;
- exec2_objects[0].alignment = 0;
- exec2_objects[0].offset = bo->offset;
- exec2_objects[0].flags = 0;
- exec2_objects[0].rsvd1 = 0;
- exec2_objects[0].rsvd2 = 0;
-
- execbuf.buffers_ptr = (uintptr_t) exec2_objects;
- execbuf.buffer_count = 1;
- execbuf.batch_start_offset = state.offset;
- execbuf.batch_len = batch.next - state.map;
- execbuf.cliprects_ptr = 0;
- execbuf.num_cliprects = 0;
- execbuf.DR1 = 0;
- execbuf.DR4 = 0;
-
- execbuf.flags =
- I915_EXEC_HANDLE_LUT | I915_EXEC_NO_RELOC | I915_EXEC_RENDER;
- execbuf.rsvd1 = device->context_id;
- execbuf.rsvd2 = 0;
-
- ret = anv_gem_execbuffer(device, &execbuf);
- if (ret != 0) {
- /* We don't know the real error. */
- result = vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY, "execbuf2 failed: %m");
- goto fail;
- }
-
- timeout = INT64_MAX;
- ret = anv_gem_wait(device, bo->gem_handle, &timeout);
- if (ret != 0) {
- /* We don't know the real error. */
- result = vk_errorf(VK_ERROR_OUT_OF_DEVICE_MEMORY, "execbuf2 failed: %m");
- goto fail;
- }
-
- anv_state_pool_free(&device->dynamic_state_pool, state);
-
- return VK_SUCCESS;
-
- fail:
- anv_state_pool_free(&device->dynamic_state_pool, state);
-
- return result;
+ return anv_device_submit_simple_batch(device, &batch);
}
VkResult
assert(pAllocateInfo->sType == VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO);
+ if (pAllocateInfo->allocationSize == 0) {
+ /* Apparently, this is allowed */
+ *pMem = VK_NULL_HANDLE;
+ return VK_SUCCESS;
+ }
+
/* We support exactly one memory heap. */
- assert(pAllocateInfo->memoryTypeIndex == 0);
+ assert(pAllocateInfo->memoryTypeIndex == 0 ||
+ (!device->info.has_llc && pAllocateInfo->memoryTypeIndex < 2));
/* FINISHME: Fail if allocation request exceeds heap size. */
if (mem == NULL)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
- result = anv_bo_init_new(&mem->bo, device, pAllocateInfo->allocationSize);
+ /* The kernel is going to give us whole pages anyway */
+ uint64_t alloc_size = align_u64(pAllocateInfo->allocationSize, 4096);
+
+ result = anv_bo_init_new(&mem->bo, device, alloc_size);
if (result != VK_SUCCESS)
goto fail;
+ mem->type_index = pAllocateInfo->memoryTypeIndex;
+
*pMem = anv_device_memory_to_handle(mem);
return VK_SUCCESS;
ANV_FROM_HANDLE(anv_device, device, _device);
ANV_FROM_HANDLE(anv_device_memory, mem, _mem);
+ if (mem == NULL)
+ return;
+
if (mem->bo.map)
anv_gem_munmap(mem->bo.map, mem->bo.size);
ANV_FROM_HANDLE(anv_device, device, _device);
ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
+ if (mem == NULL) {
+ *ppData = NULL;
+ return VK_SUCCESS;
+ }
+
+ if (size == VK_WHOLE_SIZE)
+ size = mem->bo.size - offset;
+
/* FIXME: Is this supposed to be thread safe? Since vkUnmapMemory() only
* takes a VkDeviceMemory pointer, it seems like only one map of the memory
* at a time is valid. We could just mmap up front and return an offset
* pointer here, but that may exhaust virtual memory on 32 bit
* userspace. */
- mem->map = anv_gem_mmap(device, mem->bo.gem_handle, offset, size);
- mem->map_size = size;
+ uint32_t gem_flags = 0;
+ if (!device->info.has_llc && mem->type_index == 0)
+ gem_flags |= I915_MMAP_WC;
+
+ /* GEM will fail to map if the offset isn't 4k-aligned. Round down. */
+ uint64_t map_offset = offset & ~4095ull;
+ assert(offset >= map_offset);
+ uint64_t map_size = (offset + size) - map_offset;
- *ppData = mem->map;
+ /* Let's map whole pages */
+ map_size = align_u64(map_size, 4096);
+
+ mem->map = anv_gem_mmap(device, mem->bo.gem_handle,
+ map_offset, map_size, gem_flags);
+ mem->map_size = map_size;
+
+ *ppData = mem->map + (offset - map_offset);
return VK_SUCCESS;
}
{
ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
+ if (mem == NULL)
+ return;
+
anv_gem_munmap(mem->map, mem->map_size);
}
+static void
+clflush_mapped_ranges(struct anv_device *device,
+ uint32_t count,
+ const VkMappedMemoryRange *ranges)
+{
+ for (uint32_t i = 0; i < count; i++) {
+ ANV_FROM_HANDLE(anv_device_memory, mem, ranges[i].memory);
+ void *p = mem->map + (ranges[i].offset & ~CACHELINE_MASK);
+ void *end;
+
+ if (ranges[i].offset + ranges[i].size > mem->map_size)
+ end = mem->map + mem->map_size;
+ else
+ end = mem->map + ranges[i].offset + ranges[i].size;
+
+ while (p < end) {
+ __builtin_ia32_clflush(p);
+ p += CACHELINE_SIZE;
+ }
+ }
+}
+
VkResult anv_FlushMappedMemoryRanges(
- VkDevice device,
+ VkDevice _device,
uint32_t memoryRangeCount,
const VkMappedMemoryRange* pMemoryRanges)
{
- /* clflush here for !llc platforms */
+ ANV_FROM_HANDLE(anv_device, device, _device);
+
+ if (device->info.has_llc)
+ return VK_SUCCESS;
+
+ /* Make sure the writes we're flushing have landed. */
+ __builtin_ia32_mfence();
+
+ clflush_mapped_ranges(device, memoryRangeCount, pMemoryRanges);
return VK_SUCCESS;
}
VkResult anv_InvalidateMappedMemoryRanges(
- VkDevice device,
+ VkDevice _device,
uint32_t memoryRangeCount,
const VkMappedMemoryRange* pMemoryRanges)
{
- return anv_FlushMappedMemoryRanges(device, memoryRangeCount, pMemoryRanges);
+ ANV_FROM_HANDLE(anv_device, device, _device);
+
+ if (device->info.has_llc)
+ return VK_SUCCESS;
+
+ clflush_mapped_ranges(device, memoryRangeCount, pMemoryRanges);
+
+ /* Make sure no reads get moved up above the invalidate. */
+ __builtin_ia32_mfence();
+
+ return VK_SUCCESS;
}
void anv_GetBufferMemoryRequirements(
ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
ANV_FROM_HANDLE(anv_buffer, buffer, _buffer);
- buffer->bo = &mem->bo;
- buffer->offset = memoryOffset;
+ if (mem) {
+ buffer->bo = &mem->bo;
+ buffer->offset = memoryOffset;
+ } else {
+ buffer->bo = NULL;
+ buffer->offset = 0;
+ }
return VK_SUCCESS;
}
ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
ANV_FROM_HANDLE(anv_image, image, _image);
- image->bo = &mem->bo;
- image->offset = memoryOffset;
+ if (mem) {
+ image->bo = &mem->bo;
+ image->offset = memoryOffset;
+ } else {
+ image->bo = NULL;
+ image->offset = 0;
+ }
return VK_SUCCESS;
}
const VkBindSparseInfo* pBindInfo,
VkFence fence)
{
- stub_return(VK_UNSUPPORTED);
+ stub_return(VK_ERROR_INCOMPATIBLE_DRIVER);
}
VkResult anv_CreateFence(
goto fail;
fence->bo.map =
- anv_gem_mmap(device, fence->bo.gem_handle, 0, fence->bo.size);
+ anv_gem_mmap(device, fence->bo.gem_handle, 0, fence->bo.size, 0);
batch.next = batch.start = fence->bo.map;
batch.end = fence->bo.map + fence->bo.size;
anv_batch_emit(&batch, GEN7_MI_BATCH_BUFFER_END);
anv_batch_emit(&batch, GEN7_MI_NOOP);
+ if (!device->info.has_llc) {
+ assert(((uintptr_t) fence->bo.map & CACHELINE_MASK) == 0);
+ assert(batch.next - fence->bo.map <= CACHELINE_SIZE);
+ __builtin_ia32_mfence();
+ __builtin_ia32_clflush(fence->bo.map);
+ }
+
fence->exec2_objects[0].handle = fence->bo.gem_handle;
fence->exec2_objects[0].relocation_count = 0;
fence->exec2_objects[0].relocs_ptr = 0;
fence->execbuf.rsvd1 = device->context_id;
fence->execbuf.rsvd2 = 0;
+ fence->ready = false;
+
*pFence = anv_fence_to_handle(fence);
return VK_SUCCESS;
const VkAllocationCallbacks* pAllocator,
VkSemaphore* pSemaphore)
{
+ /* The DRM execbuffer ioctl always execute in-oder, even between different
+ * rings. As such, there's nothing to do for the user space semaphore.
+ */
+
*pSemaphore = (VkSemaphore)1;
- stub_return(VK_SUCCESS);
+
+ return VK_SUCCESS;
}
void anv_DestroySemaphore(
VkSemaphore semaphore,
const VkAllocationCallbacks* pAllocator)
{
- stub();
}
// Event functions
VkResult anv_CreateEvent(
- VkDevice device,
+ VkDevice _device,
const VkEventCreateInfo* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkEvent* pEvent)
{
- stub_return(VK_UNSUPPORTED);
+ ANV_FROM_HANDLE(anv_device, device, _device);
+ struct anv_state state;
+ struct anv_event *event;
+
+ assert(pCreateInfo->sType == VK_STRUCTURE_TYPE_EVENT_CREATE_INFO);
+
+ state = anv_state_pool_alloc(&device->dynamic_state_pool,
+ sizeof(*event), 8);
+ event = state.map;
+ event->state = state;
+ event->semaphore = VK_EVENT_RESET;
+
+ if (!device->info.has_llc) {
+ /* Make sure the writes we're flushing have landed. */
+ __builtin_ia32_mfence();
+ __builtin_ia32_clflush(event);
+ }
+
+ *pEvent = anv_event_to_handle(event);
+
+ return VK_SUCCESS;
}
void anv_DestroyEvent(
- VkDevice device,
- VkEvent event,
+ VkDevice _device,
+ VkEvent _event,
const VkAllocationCallbacks* pAllocator)
{
- stub();
+ ANV_FROM_HANDLE(anv_device, device, _device);
+ ANV_FROM_HANDLE(anv_event, event, _event);
+
+ anv_state_pool_free(&device->dynamic_state_pool, event->state);
}
VkResult anv_GetEventStatus(
- VkDevice device,
- VkEvent event)
+ VkDevice _device,
+ VkEvent _event)
{
- stub_return(VK_UNSUPPORTED);
+ ANV_FROM_HANDLE(anv_device, device, _device);
+ ANV_FROM_HANDLE(anv_event, event, _event);
+
+ if (!device->info.has_llc) {
+ /* Invalidate read cache before reading event written by GPU. */
+ __builtin_ia32_clflush(event);
+ __builtin_ia32_mfence();
+
+ }
+
+ return event->semaphore;
}
VkResult anv_SetEvent(
- VkDevice device,
- VkEvent event)
+ VkDevice _device,
+ VkEvent _event)
{
- stub_return(VK_UNSUPPORTED);
+ ANV_FROM_HANDLE(anv_device, device, _device);
+ ANV_FROM_HANDLE(anv_event, event, _event);
+
+ event->semaphore = VK_EVENT_SET;
+
+ if (!device->info.has_llc) {
+ /* Make sure the writes we're flushing have landed. */
+ __builtin_ia32_mfence();
+ __builtin_ia32_clflush(event);
+ }
+
+ return VK_SUCCESS;
}
VkResult anv_ResetEvent(
- VkDevice device,
- VkEvent event)
+ VkDevice _device,
+ VkEvent _event)
{
- stub_return(VK_UNSUPPORTED);
+ ANV_FROM_HANDLE(anv_device, device, _device);
+ ANV_FROM_HANDLE(anv_event, event, _event);
+
+ event->semaphore = VK_EVENT_RESET;
+
+ if (!device->info.has_llc) {
+ /* Make sure the writes we're flushing have landed. */
+ __builtin_ia32_mfence();
+ __builtin_ia32_clflush(event);
+ }
+
+ return VK_SUCCESS;
}
// Buffer functions
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
buffer->size = pCreateInfo->size;
+ buffer->usage = pCreateInfo->usage;
buffer->bo = NULL;
buffer->offset = 0;
}
void
-anv_fill_buffer_surface_state(struct anv_device *device, void *state,
- const struct anv_format *format,
+anv_fill_buffer_surface_state(struct anv_device *device, struct anv_state state,
+ enum isl_format format,
uint32_t offset, uint32_t range, uint32_t stride)
{
switch (device->info.gen) {
case 7:
if (device->info.is_haswell)
- gen75_fill_buffer_surface_state(state, format, offset, range, stride);
+ gen75_fill_buffer_surface_state(state.map, format, offset, range,
+ stride);
else
- gen7_fill_buffer_surface_state(state, format, offset, range, stride);
+ gen7_fill_buffer_surface_state(state.map, format, offset, range,
+ stride);
break;
case 8:
- gen8_fill_buffer_surface_state(state, format, offset, range, stride);
+ gen8_fill_buffer_surface_state(state.map, format, offset, range, stride);
break;
case 9:
- gen9_fill_buffer_surface_state(state, format, offset, range, stride);
+ gen9_fill_buffer_surface_state(state.map, format, offset, range, stride);
break;
default:
unreachable("unsupported gen\n");
}
-}
-VkResult anv_CreateBufferView(
- VkDevice _device,
- const VkBufferViewCreateInfo* pCreateInfo,
- const VkAllocationCallbacks* pAllocator,
- VkBufferView* pView)
-{
- stub_return(VK_UNSUPPORTED);
-}
-
-void anv_DestroyBufferView(
- VkDevice _device,
- VkBufferView _bview,
- const VkAllocationCallbacks* pAllocator)
-{
- stub();
+ if (!device->info.has_llc)
+ anv_state_clflush(state);
}
void anv_DestroySampler(