#include <unistd.h>
#include <fcntl.h>
#include <xf86drm.h>
-#include <drm_fourcc.h>
+#include "drm-uapi/drm_fourcc.h"
#include "anv_private.h"
#include "util/strtod.h"
#include "util/build_id.h"
#include "util/disk_cache.h"
#include "util/mesa-sha1.h"
+#include "util/os_file.h"
+#include "util/u_atomic.h"
+#include "util/u_string.h"
+#include "git_sha1.h"
#include "vk_util.h"
#include "common/gen_defines.h"
+#include "compiler/glsl_types.h"
#include "genxml/gen7_pack.h"
+/* This is probably far to big but it reflects the max size used for messages
+ * in OpenGLs KHR_debug.
+ */
+#define MAX_DEBUG_MESSAGE_LENGTH 4096
+
static void
compiler_debug_log(void *data, const char *fmt, ...)
-{ }
+{
+ char str[MAX_DEBUG_MESSAGE_LENGTH];
+ struct anv_device *device = (struct anv_device *)data;
+
+ if (list_empty(&device->instance->debug_report_callbacks.callbacks))
+ return;
+
+ va_list args;
+ va_start(args, fmt);
+ (void) vsnprintf(str, MAX_DEBUG_MESSAGE_LENGTH, fmt, args);
+ va_end(args);
+
+ vk_debug_report(&device->instance->debug_report_callbacks,
+ VK_DEBUG_REPORT_DEBUG_BIT_EXT,
+ VK_DEBUG_REPORT_OBJECT_TYPE_UNKNOWN_EXT,
+ 0, 0, 0, "anv", str);
+}
static void
compiler_perf_log(void *data, const char *fmt, ...)
va_end(args);
}
-static VkResult
-anv_compute_heap_size(int fd, uint64_t gtt_size, uint64_t *heap_size)
+static uint64_t
+anv_compute_heap_size(int fd, uint64_t gtt_size)
{
/* Query the total ram from the system */
struct sysinfo info;
*/
uint64_t available_gtt = gtt_size * 3 / 4;
- *heap_size = MIN2(available_ram, available_gtt);
-
- return VK_SUCCESS;
+ return MIN2(available_ram, available_gtt);
}
static VkResult
device->supports_48bit_addresses = (device->info.gen >= 8) &&
gtt_size > (4ULL << 30 /* GiB */);
- uint64_t heap_size = 0;
- VkResult result = anv_compute_heap_size(fd, gtt_size, &heap_size);
- if (result != VK_SUCCESS)
- return result;
+ uint64_t heap_size = anv_compute_heap_size(fd, gtt_size);
if (heap_size > (2ull << 30) && !device->supports_48bit_addresses) {
/* When running with an overridden PCI ID, we may get a GTT size from
*/
device->memory.heap_count = 1;
device->memory.heaps[0] = (struct anv_memory_heap) {
+ .vma_start = LOW_HEAP_MIN_ADDRESS,
+ .vma_size = LOW_HEAP_SIZE,
.size = heap_size,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
.supports_48bit_addresses = false,
device->memory.heap_count = 2;
device->memory.heaps[0] = (struct anv_memory_heap) {
+ .vma_start = HIGH_HEAP_MIN_ADDRESS,
+ /* Leave the last 4GiB out of the high vma range, so that no state
+ * base address + size can overflow 48 bits. For more information see
+ * the comment about Wa32bitGeneralStateOffset in anv_allocator.c
+ */
+ .vma_size = gtt_size - (1ull << 32) - HIGH_HEAP_MIN_ADDRESS,
.size = heap_size_48bit,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
.supports_48bit_addresses = true,
};
device->memory.heaps[1] = (struct anv_memory_heap) {
+ .vma_start = LOW_HEAP_MIN_ADDRESS,
+ .vma_size = LOW_HEAP_SIZE,
.size = heap_size_32bit,
.flags = VK_MEMORY_HEAP_DEVICE_LOCAL_BIT,
.supports_48bit_addresses = false,
_mesa_sha1_update(&sha1_ctx, build_id_data(note), build_id_len);
_mesa_sha1_update(&sha1_ctx, &device->chipset_id,
sizeof(device->chipset_id));
+ _mesa_sha1_update(&sha1_ctx, &device->always_use_bindless,
+ sizeof(device->always_use_bindless));
+ _mesa_sha1_update(&sha1_ctx, &device->has_a64_buffer_access,
+ sizeof(device->has_a64_buffer_access));
+ _mesa_sha1_update(&sha1_ctx, &device->has_bindless_images,
+ sizeof(device->has_bindless_images));
+ _mesa_sha1_update(&sha1_ctx, &device->has_bindless_samplers,
+ sizeof(device->has_bindless_samplers));
_mesa_sha1_final(&sha1_ctx, sha1);
memcpy(device->pipeline_cache_uuid, sha1, VK_UUID_SIZE);
char timestamp[41];
_mesa_sha1_format(timestamp, device->driver_build_sha1);
- const uint64_t driver_flags = INTEL_DEBUG & DEBUG_DISK_CACHE_MASK;
+ const uint64_t driver_flags =
+ brw_get_compiler_config_value(device->compiler);
device->disk_cache = disk_cache_create(renderer, timestamp, driver_flags);
#else
device->disk_cache = NULL;
#endif
}
+static uint64_t
+get_available_system_memory()
+{
+ char *meminfo = os_read_file("/proc/meminfo");
+ if (!meminfo)
+ return 0;
+
+ char *str = strstr(meminfo, "MemAvailable:");
+ if (!str) {
+ free(meminfo);
+ return 0;
+ }
+
+ uint64_t kb_mem_available;
+ if (sscanf(str, "MemAvailable: %" PRIx64, &kb_mem_available) == 1) {
+ free(meminfo);
+ return kb_mem_available << 10;
+ }
+
+ free(meminfo);
+ return 0;
+}
+
static VkResult
anv_physical_device_init(struct anv_physical_device *device,
struct anv_instance *instance,
- const char *primary_path,
- const char *path)
+ drmDevicePtr drm_device)
{
+ const char *primary_path = drm_device->nodes[DRM_NODE_PRIMARY];
+ const char *path = drm_device->nodes[DRM_NODE_RENDER];
VkResult result;
int fd;
int master_fd = -1;
device->instance = instance;
assert(strlen(path) < ARRAY_SIZE(device->path));
- strncpy(device->path, path, ARRAY_SIZE(device->path));
+ snprintf(device->path, ARRAY_SIZE(device->path), "%s", path);
device->no_hw = getenv("INTEL_NO_HW") != NULL;
device->no_hw = true;
}
+ device->pci_info.domain = drm_device->businfo.pci->domain;
+ device->pci_info.bus = drm_device->businfo.pci->bus;
+ device->pci_info.device = drm_device->businfo.pci->dev;
+ device->pci_info.function = drm_device->businfo.pci->func;
+
device->name = gen_get_device_name(device->chipset_id);
if (!gen_get_device_info(device->chipset_id, &device->info)) {
result = vk_error(VK_ERROR_INCOMPATIBLE_DRIVER);
intel_logw("Ivy Bridge Vulkan support is incomplete");
} else if (device->info.gen == 7 && device->info.is_baytrail) {
intel_logw("Bay Trail Vulkan support is incomplete");
- } else if (device->info.gen >= 8 && device->info.gen <= 10) {
- /* Gen8-10 fully supported */
- } else if (device->info.gen == 11) {
- intel_logw("Vulkan is not yet fully supported on gen11.");
+ } else if (device->info.gen >= 8 && device->info.gen <= 11) {
+ /* Gen8-11 fully supported */
} else {
result = vk_errorf(device->instance, device,
VK_ERROR_INCOMPATIBLE_DRIVER,
device->has_context_isolation =
anv_gem_get_param(fd, I915_PARAM_HAS_CONTEXT_ISOLATION);
- bool swizzled = anv_gem_get_bit6_swizzle(fd, I915_TILING_X);
+ device->always_use_bindless =
+ env_var_as_boolean("ANV_ALWAYS_BINDLESS", false);
+
+ /* We first got the A64 messages on broadwell and we can only use them if
+ * we can pass addresses directly into the shader which requires softpin.
+ */
+ device->has_a64_buffer_access = device->info.gen >= 8 &&
+ device->use_softpin;
+
+ /* We first get bindless image access on Skylake and we can only really do
+ * it if we don't have any relocations so we need softpin.
+ */
+ device->has_bindless_images = device->info.gen >= 9 &&
+ device->use_softpin;
+
+ /* We've had bindless samplers since Ivy Bridge (forever in Vulkan terms)
+ * because it's just a matter of setting the sampler address in the sample
+ * message header. However, we've not bothered to wire it up for vec4 so
+ * we leave it disabled on gen7.
+ */
+ device->has_bindless_samplers = device->info.gen >= 8;
+
+ device->has_mem_available = get_available_system_memory() != 0;
/* Starting with Gen10, the timestamp frequency of the command streamer may
* vary from one part to another. We can query the value from the kernel.
device->info.gen < 8 || !device->has_context_isolation;
device->compiler->supports_shader_constants = true;
+ /* Broadwell PRM says:
+ *
+ * "Before Gen8, there was a historical configuration control field to
+ * swizzle address bit[6] for in X/Y tiling modes. This was set in three
+ * different places: TILECTL[1:0], ARB_MODE[5:4], and
+ * DISP_ARB_CTL[14:13].
+ *
+ * For Gen8 and subsequent generations, the swizzle fields are all
+ * reserved, and the CPU's memory controller performs all address
+ * swizzling modifications."
+ */
+ bool swizzled =
+ device->info.gen < 8 && anv_gem_get_bit6_swizzle(fd, I915_TILING_X);
+
isl_device_init(&device->isl_dev, &device->info, swizzled);
result = anv_physical_device_init_uuids(device);
else
instance->alloc = default_alloc;
- if (pCreateInfo->pApplicationInfo &&
- pCreateInfo->pApplicationInfo->apiVersion != 0) {
- instance->apiVersion = pCreateInfo->pApplicationInfo->apiVersion;
- } else {
- anv_EnumerateInstanceVersion(&instance->apiVersion);
+ instance->app_info = (struct anv_app_info) { .api_version = 0 };
+ if (pCreateInfo->pApplicationInfo) {
+ const VkApplicationInfo *app = pCreateInfo->pApplicationInfo;
+
+ instance->app_info.app_name =
+ vk_strdup(&instance->alloc, app->pApplicationName,
+ VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
+ instance->app_info.app_version = app->applicationVersion;
+
+ instance->app_info.engine_name =
+ vk_strdup(&instance->alloc, app->pEngineName,
+ VK_SYSTEM_ALLOCATION_SCOPE_INSTANCE);
+ instance->app_info.engine_version = app->engineVersion;
+
+ instance->app_info.api_version = app->apiVersion;
}
+ if (instance->app_info.api_version == 0)
+ instance->app_info.api_version = VK_API_VERSION_1_0;
+
instance->enabled_extensions = enabled_extensions;
for (unsigned i = 0; i < ARRAY_SIZE(instance->dispatch.entrypoints); i++) {
/* Vulkan requires that entrypoints for extensions which have not been
* enabled must not be advertised.
*/
- if (!anv_entrypoint_is_enabled(i, instance->apiVersion,
- &instance->enabled_extensions, NULL)) {
+ if (!anv_instance_entrypoint_is_enabled(i, instance->app_info.api_version,
+ &instance->enabled_extensions)) {
instance->dispatch.entrypoints[i] = NULL;
- } else if (anv_dispatch_table.entrypoints[i] != NULL) {
- instance->dispatch.entrypoints[i] = anv_dispatch_table.entrypoints[i];
} else {
instance->dispatch.entrypoints[i] =
- anv_tramp_dispatch_table.entrypoints[i];
+ anv_instance_dispatch_table.entrypoints[i];
+ }
+ }
+
+ for (unsigned i = 0; i < ARRAY_SIZE(instance->device_dispatch.entrypoints); i++) {
+ /* Vulkan requires that entrypoints for extensions which have not been
+ * enabled must not be advertised.
+ */
+ if (!anv_device_entrypoint_is_enabled(i, instance->app_info.api_version,
+ &instance->enabled_extensions, NULL)) {
+ instance->device_dispatch.entrypoints[i] = NULL;
+ } else {
+ instance->device_dispatch.entrypoints[i] =
+ anv_device_dispatch_table.entrypoints[i];
}
}
env_var_as_boolean("ANV_ENABLE_PIPELINE_CACHE", true);
_mesa_locale_init();
+ glsl_type_singleton_init_or_ref();
VG(VALGRIND_CREATE_MEMPOOL(instance, 0, false));
anv_physical_device_finish(&instance->physicalDevice);
}
+ vk_free(&instance->alloc, (char *)instance->app_info.app_name);
+ vk_free(&instance->alloc, (char *)instance->app_info.engine_name);
+
VG(VALGRIND_DESTROY_MEMPOOL(instance));
vk_debug_report_instance_destroy(&instance->debug_report_callbacks);
+ glsl_type_singleton_decref();
_mesa_locale_fini();
vk_free(&instance->alloc, instance);
devices[i]->deviceinfo.pci->vendor_id == 0x8086) {
result = anv_physical_device_init(&instance->physicalDevice,
- instance,
- devices[i]->nodes[DRM_NODE_PRIMARY],
- devices[i]->nodes[DRM_NODE_RENDER]);
+ instance, devices[i]);
if (result != VK_ERROR_INCOMPATIBLE_DRIVER)
break;
}
memset(p->physicalDevices, 0, sizeof(p->physicalDevices));
p->physicalDevices[0] =
anv_physical_device_to_handle(&instance->physicalDevice);
- p->subsetAllocation = VK_FALSE;
+ p->subsetAllocation = false;
vk_foreach_struct(ext, p->pNext)
anv_debug_ignored_stype(ext->sType);
.shaderInt64 = pdevice->info.gen >= 8 &&
pdevice->info.has_64bit_types,
.shaderInt16 = pdevice->info.gen >= 8,
- .shaderResourceMinLod = false,
+ .shaderResourceMinLod = pdevice->info.gen >= 9,
.variableMultisampleRate = true,
.inheritedQueries = true,
};
pFeatures->vertexPipelineStoresAndAtomics =
pdevice->compiler->scalar_stage[MESA_SHADER_VERTEX] &&
pdevice->compiler->scalar_stage[MESA_SHADER_GEOMETRY];
+
+ struct anv_app_info *app_info = &pdevice->instance->app_info;
+
+ /* The new DOOM and Wolfenstein games require depthBounds without
+ * checking for it. They seem to run fine without it so just claim it's
+ * there and accept the consequences.
+ */
+ if (app_info->engine_name && strcmp(app_info->engine_name, "idTech") == 0)
+ pFeatures->depthBounds = true;
}
void anv_GetPhysicalDeviceFeatures2(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceFeatures2* pFeatures)
{
+ ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
anv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
vk_foreach_struct(ext, pFeatures->pNext) {
switch (ext->sType) {
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
- VkPhysicalDeviceProtectedMemoryFeatures *features = (void *)ext;
- features->protectedMemory = VK_FALSE;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR: {
+ VkPhysicalDevice8BitStorageFeaturesKHR *features =
+ (VkPhysicalDevice8BitStorageFeaturesKHR *)ext;
+ features->storageBuffer8BitAccess = pdevice->info.gen >= 8;
+ features->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8;
+ features->storagePushConstant8 = pdevice->info.gen >= 8;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
+ VkPhysicalDevice16BitStorageFeatures *features =
+ (VkPhysicalDevice16BitStorageFeatures *)ext;
+ features->storageBuffer16BitAccess = pdevice->info.gen >= 8;
+ features->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8;
+ features->storagePushConstant16 = pdevice->info.gen >= 8;
+ features->storageInputOutput16 = false;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_BUFFER_DEVICE_ADDRESS_FEATURES_EXT: {
+ VkPhysicalDeviceBufferDeviceAddressFeaturesEXT *features = (void *)ext;
+ features->bufferDeviceAddress = pdevice->has_a64_buffer_access;
+ features->bufferDeviceAddressCaptureReplay = false;
+ features->bufferDeviceAddressMultiDevice = false;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_COMPUTE_SHADER_DERIVATIVES_FEATURES_NV: {
+ VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *features =
+ (VkPhysicalDeviceComputeShaderDerivativesFeaturesNV *)ext;
+ features->computeDerivativeGroupQuads = true;
+ features->computeDerivativeGroupLinear = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CONDITIONAL_RENDERING_FEATURES_EXT: {
+ VkPhysicalDeviceConditionalRenderingFeaturesEXT *features =
+ (VkPhysicalDeviceConditionalRenderingFeaturesEXT*)ext;
+ features->conditionalRendering = pdevice->info.gen >= 8 ||
+ pdevice->info.is_haswell;
+ features->inheritedConditionalRendering = pdevice->info.gen >= 8 ||
+ pdevice->info.is_haswell;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_CLIP_ENABLE_FEATURES_EXT: {
+ VkPhysicalDeviceDepthClipEnableFeaturesEXT *features =
+ (VkPhysicalDeviceDepthClipEnableFeaturesEXT *)ext;
+ features->depthClipEnable = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FLOAT16_INT8_FEATURES_KHR: {
+ VkPhysicalDeviceFloat16Int8FeaturesKHR *features = (void *)ext;
+ features->shaderFloat16 = pdevice->info.gen >= 8;
+ features->shaderInt8 = pdevice->info.gen >= 8;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_FRAGMENT_SHADER_INTERLOCK_FEATURES_EXT: {
+ VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT *features =
+ (VkPhysicalDeviceFragmentShaderInterlockFeaturesEXT *)ext;
+ features->fragmentShaderSampleInterlock = pdevice->info.gen >= 9;
+ features->fragmentShaderPixelInterlock = pdevice->info.gen >= 9;
+ features->fragmentShaderShadingRateInterlock = false;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_HOST_QUERY_RESET_FEATURES_EXT: {
+ VkPhysicalDeviceHostQueryResetFeaturesEXT *features =
+ (VkPhysicalDeviceHostQueryResetFeaturesEXT *)ext;
+ features->hostQueryReset = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_FEATURES_EXT: {
+ VkPhysicalDeviceDescriptorIndexingFeaturesEXT *features =
+ (VkPhysicalDeviceDescriptorIndexingFeaturesEXT *)ext;
+ features->shaderInputAttachmentArrayDynamicIndexing = false;
+ features->shaderUniformTexelBufferArrayDynamicIndexing = true;
+ features->shaderStorageTexelBufferArrayDynamicIndexing = true;
+ features->shaderUniformBufferArrayNonUniformIndexing = false;
+ features->shaderSampledImageArrayNonUniformIndexing = true;
+ features->shaderStorageBufferArrayNonUniformIndexing = true;
+ features->shaderStorageImageArrayNonUniformIndexing = true;
+ features->shaderInputAttachmentArrayNonUniformIndexing = false;
+ features->shaderUniformTexelBufferArrayNonUniformIndexing = true;
+ features->shaderStorageTexelBufferArrayNonUniformIndexing = true;
+ features->descriptorBindingUniformBufferUpdateAfterBind = false;
+ features->descriptorBindingSampledImageUpdateAfterBind = true;
+ features->descriptorBindingStorageImageUpdateAfterBind = true;
+ features->descriptorBindingStorageBufferUpdateAfterBind = true;
+ features->descriptorBindingUniformTexelBufferUpdateAfterBind = true;
+ features->descriptorBindingStorageTexelBufferUpdateAfterBind = true;
+ features->descriptorBindingUpdateUnusedWhilePending = true;
+ features->descriptorBindingPartiallyBound = true;
+ features->descriptorBindingVariableDescriptorCount = false;
+ features->runtimeDescriptorArray = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_FEATURES_EXT: {
+ VkPhysicalDeviceInlineUniformBlockFeaturesEXT *features =
+ (VkPhysicalDeviceInlineUniformBlockFeaturesEXT *)ext;
+ features->inlineUniformBlock = true;
+ features->descriptorBindingInlineUniformBlockUpdateAfterBind = true;
break;
}
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTER_FEATURES: {
- VkPhysicalDeviceVariablePointerFeatures *features = (void *)ext;
- features->variablePointersStorageBuffer = true;
- features->variablePointers = true;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
+ VkPhysicalDeviceProtectedMemoryFeatures *features = (void *)ext;
+ features->protectedMemory = false;
break;
}
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETER_FEATURES: {
- VkPhysicalDeviceShaderDrawParameterFeatures *features = (void *)ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SCALAR_BLOCK_LAYOUT_FEATURES_EXT: {
+ VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *features =
+ (VkPhysicalDeviceScalarBlockLayoutFeaturesEXT *)ext;
+ features->scalarBlockLayout = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_ATOMIC_INT64_FEATURES_KHR: {
+ VkPhysicalDeviceShaderAtomicInt64FeaturesKHR *features = (void *)ext;
+ features->shaderBufferInt64Atomics =
+ pdevice->info.gen >= 9 && pdevice->use_softpin;
+ features->shaderSharedInt64Atomics = VK_FALSE;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DEMOTE_TO_HELPER_INVOCATION_FEATURES_EXT: {
+ VkPhysicalDeviceShaderDemoteToHelperInvocationFeaturesEXT *features = (void *)ext;
+ features->shaderDemoteToHelperInvocation = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SHADER_DRAW_PARAMETERS_FEATURES: {
+ VkPhysicalDeviceShaderDrawParametersFeatures *features = (void *)ext;
features->shaderDrawParameters = true;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES_KHR: {
- VkPhysicalDevice16BitStorageFeaturesKHR *features =
- (VkPhysicalDevice16BitStorageFeaturesKHR *)ext;
- ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_FEATURES_EXT: {
+ VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT *features =
+ (VkPhysicalDeviceTexelBufferAlignmentFeaturesEXT *)ext;
+ features->texelBufferAlignment = true;
+ break;
+ }
- features->storageBuffer16BitAccess = pdevice->info.gen >= 8;
- features->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8;
- features->storagePushConstant16 = pdevice->info.gen >= 8;
- features->storageInputOutput16 = false;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTERS_FEATURES: {
+ VkPhysicalDeviceVariablePointersFeatures *features = (void *)ext;
+ features->variablePointersStorageBuffer = true;
+ features->variablePointers = true;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR: {
- VkPhysicalDevice8BitStorageFeaturesKHR *features =
- (VkPhysicalDevice8BitStorageFeaturesKHR *)ext;
- ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_FEATURES_EXT: {
+ VkPhysicalDeviceTransformFeedbackFeaturesEXT *features =
+ (VkPhysicalDeviceTransformFeedbackFeaturesEXT *)ext;
+ features->transformFeedback = true;
+ features->geometryStreams = true;
+ break;
+ }
- features->storageBuffer8BitAccess = pdevice->info.gen >= 8;
- features->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8;
- features->storagePushConstant8 = pdevice->info.gen >= 8;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_UNIFORM_BUFFER_STANDARD_LAYOUT_FEATURES_KHR: {
+ VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *features =
+ (VkPhysicalDeviceUniformBufferStandardLayoutFeaturesKHR *)ext;
+ features->uniformBufferStandardLayout = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT: {
+ VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *features =
+ (VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *)ext;
+ features->vertexAttributeInstanceRateDivisor = true;
+ features->vertexAttributeInstanceRateZeroDivisor = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_YCBCR_IMAGE_ARRAYS_FEATURES_EXT: {
+ VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *features =
+ (VkPhysicalDeviceYcbcrImageArraysFeaturesEXT *)ext;
+ features->ycbcrImageArrays = true;
break;
}
}
}
+#define MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS 64
+
+#define MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS 64
+#define MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS 256
+
void anv_GetPhysicalDeviceProperties(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceProperties* pProperties)
const uint32_t max_raw_buffer_sz = devinfo->gen >= 7 ?
(1ul << 30) : (1ul << 27);
- const uint32_t max_samplers = (devinfo->gen >= 8 || devinfo->is_haswell) ?
- 128 : 16;
+ const uint32_t max_ssbos = pdevice->has_a64_buffer_access ? UINT16_MAX : 64;
+ const uint32_t max_textures =
+ pdevice->has_bindless_images ? UINT16_MAX : 128;
+ const uint32_t max_samplers =
+ pdevice->has_bindless_samplers ? UINT16_MAX :
+ (devinfo->gen >= 8 || devinfo->is_haswell) ? 128 : 16;
+ const uint32_t max_images =
+ pdevice->has_bindless_images ? UINT16_MAX : MAX_IMAGES;
+
+ /* The moment we have anything bindless, claim a high per-stage limit */
+ const uint32_t max_per_stage =
+ pdevice->has_a64_buffer_access ? UINT32_MAX :
+ MAX_BINDING_TABLE_SIZE - MAX_RTS;
VkSampleCountFlags sample_counts =
isl_device_get_sample_counts(&pdevice->isl_dev);
+
VkPhysicalDeviceLimits limits = {
.maxImageDimension1D = (1 << 14),
.maxImageDimension2D = (1 << 14),
.sparseAddressSpaceSize = 0,
.maxBoundDescriptorSets = MAX_SETS,
.maxPerStageDescriptorSamplers = max_samplers,
- .maxPerStageDescriptorUniformBuffers = 64,
- .maxPerStageDescriptorStorageBuffers = 64,
- .maxPerStageDescriptorSampledImages = max_samplers,
- .maxPerStageDescriptorStorageImages = 64,
- .maxPerStageDescriptorInputAttachments = 64,
- .maxPerStageResources = 250,
+ .maxPerStageDescriptorUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS,
+ .maxPerStageDescriptorStorageBuffers = max_ssbos,
+ .maxPerStageDescriptorSampledImages = max_textures,
+ .maxPerStageDescriptorStorageImages = max_images,
+ .maxPerStageDescriptorInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS,
+ .maxPerStageResources = max_per_stage,
.maxDescriptorSetSamplers = 6 * max_samplers, /* number of stages * maxPerStageDescriptorSamplers */
- .maxDescriptorSetUniformBuffers = 6 * 64, /* number of stages * maxPerStageDescriptorUniformBuffers */
+ .maxDescriptorSetUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS, /* number of stages * maxPerStageDescriptorUniformBuffers */
.maxDescriptorSetUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
- .maxDescriptorSetStorageBuffers = 6 * 64, /* number of stages * maxPerStageDescriptorStorageBuffers */
+ .maxDescriptorSetStorageBuffers = 6 * max_ssbos, /* number of stages * maxPerStageDescriptorStorageBuffers */
.maxDescriptorSetStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2,
- .maxDescriptorSetSampledImages = 6 * max_samplers, /* number of stages * maxPerStageDescriptorSampledImages */
- .maxDescriptorSetStorageImages = 6 * 64, /* number of stages * maxPerStageDescriptorStorageImages */
- .maxDescriptorSetInputAttachments = 256,
+ .maxDescriptorSetSampledImages = 6 * max_textures, /* number of stages * maxPerStageDescriptorSampledImages */
+ .maxDescriptorSetStorageImages = 6 * max_images, /* number of stages * maxPerStageDescriptorStorageImages */
+ .maxDescriptorSetInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS,
.maxVertexInputAttributes = MAX_VBS,
.maxVertexInputBindings = MAX_VBS,
.maxVertexInputAttributeOffset = 2047,
.maxGeometryOutputComponents = 128,
.maxGeometryOutputVertices = 256,
.maxGeometryTotalOutputComponents = 1024,
- .maxFragmentInputComponents = 112, /* 128 components - (POS, PSIZ, CLIP_DIST0, CLIP_DIST1) */
+ .maxFragmentInputComponents = 116, /* 128 components - (PSIZ, CLIP_DIST0, CLIP_DIST1) */
.maxFragmentOutputAttachments = 8,
.maxFragmentDualSrcAttachments = 1,
.maxFragmentCombinedOutputResources = 8,
- .maxComputeSharedMemorySize = 32768,
+ .maxComputeSharedMemorySize = 64 * 1024,
.maxComputeWorkGroupCount = { 65535, 65535, 65535 },
- .maxComputeWorkGroupInvocations = 16 * devinfo->max_cs_threads,
+ .maxComputeWorkGroupInvocations = 32 * devinfo->max_cs_threads,
.maxComputeWorkGroupSize = {
16 * devinfo->max_cs_threads,
16 * devinfo->max_cs_threads,
16 * devinfo->max_cs_threads,
},
- .subPixelPrecisionBits = 4 /* FIXME */,
- .subTexelPrecisionBits = 4 /* FIXME */,
- .mipmapPrecisionBits = 4 /* FIXME */,
+ .subPixelPrecisionBits = 8,
+ .subTexelPrecisionBits = 8,
+ .mipmapPrecisionBits = 8,
.maxDrawIndexedIndexValue = UINT32_MAX,
.maxDrawIndirectCount = UINT32_MAX,
.maxSamplerLodBias = 16,
.viewportBoundsRange = { INT16_MIN, INT16_MAX },
.viewportSubPixelBits = 13, /* We take a float? */
.minMemoryMapAlignment = 4096, /* A page */
- .minTexelBufferOffsetAlignment = 1,
+ /* The dataport requires texel alignment so we need to assume a worst
+ * case of R32G32B32A32 which is 16 bytes.
+ */
+ .minTexelBufferOffsetAlignment = 16,
/* We need 16 for UBO block reads to work and 32 for push UBOs */
.minUniformBufferOffsetAlignment = 32,
.minStorageBufferOffsetAlignment = 4,
.sampledImageStencilSampleCounts = sample_counts,
.storageImageSampleCounts = VK_SAMPLE_COUNT_1_BIT,
.maxSampleMaskWords = 1,
- .timestampComputeAndGraphics = false,
+ .timestampComputeAndGraphics = true,
.timestampPeriod = 1000000000.0 / devinfo->timestamp_frequency,
.maxClipDistances = 8,
.maxCullDistances = 8,
.maxCombinedClipAndCullDistances = 8,
- .discreteQueuePriorities = 1,
+ .discreteQueuePriorities = 2,
.pointSizeRange = { 0.125, 255.875 },
.lineWidthRange = { 0.0, 7.9921875 },
.pointSizeGranularity = (1.0 / 8.0),
vk_foreach_struct(ext, pProperties->pNext) {
switch (ext->sType) {
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: {
- VkPhysicalDevicePushDescriptorPropertiesKHR *properties =
- (VkPhysicalDevicePushDescriptorPropertiesKHR *) ext;
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DEPTH_STENCIL_RESOLVE_PROPERTIES_KHR: {
+ VkPhysicalDeviceDepthStencilResolvePropertiesKHR *props =
+ (VkPhysicalDeviceDepthStencilResolvePropertiesKHR *)ext;
+
+ /* We support all of the depth resolve modes */
+ props->supportedDepthResolveModes =
+ VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR |
+ VK_RESOLVE_MODE_AVERAGE_BIT_KHR |
+ VK_RESOLVE_MODE_MIN_BIT_KHR |
+ VK_RESOLVE_MODE_MAX_BIT_KHR;
+
+ /* Average doesn't make sense for stencil so we don't support that */
+ props->supportedStencilResolveModes =
+ VK_RESOLVE_MODE_SAMPLE_ZERO_BIT_KHR;
+ if (pdevice->info.gen >= 8) {
+ /* The advanced stencil resolve modes currently require stencil
+ * sampling be supported by the hardware.
+ */
+ props->supportedStencilResolveModes |=
+ VK_RESOLVE_MODE_MIN_BIT_KHR |
+ VK_RESOLVE_MODE_MAX_BIT_KHR;
+ }
- properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
+ props->independentResolveNone = true;
+ props->independentResolve = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DESCRIPTOR_INDEXING_PROPERTIES_EXT: {
+ VkPhysicalDeviceDescriptorIndexingPropertiesEXT *props =
+ (VkPhysicalDeviceDescriptorIndexingPropertiesEXT *)ext;
+
+ /* It's a bit hard to exactly map our implementation to the limits
+ * described here. The bindless surface handle in the extended
+ * message descriptors is 20 bits and it's an index into the table of
+ * RENDER_SURFACE_STATE structs that starts at bindless surface base
+ * address. Given that most things consume two surface states per
+ * view (general/sampled for textures and write-only/read-write for
+ * images), we claim 2^19 things.
+ *
+ * For SSBOs, we just use A64 messages so there is no real limit
+ * there beyond the limit on the total size of a descriptor set.
+ */
+ const unsigned max_bindless_views = 1 << 19;
+
+ props->maxUpdateAfterBindDescriptorsInAllPools = max_bindless_views;
+ props->shaderUniformBufferArrayNonUniformIndexingNative = false;
+ props->shaderSampledImageArrayNonUniformIndexingNative = false;
+ props->shaderStorageBufferArrayNonUniformIndexingNative = true;
+ props->shaderStorageImageArrayNonUniformIndexingNative = false;
+ props->shaderInputAttachmentArrayNonUniformIndexingNative = false;
+ props->robustBufferAccessUpdateAfterBind = true;
+ props->quadDivergentImplicitLod = false;
+ props->maxPerStageDescriptorUpdateAfterBindSamplers = max_bindless_views;
+ props->maxPerStageDescriptorUpdateAfterBindUniformBuffers = MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS;
+ props->maxPerStageDescriptorUpdateAfterBindStorageBuffers = UINT32_MAX;
+ props->maxPerStageDescriptorUpdateAfterBindSampledImages = max_bindless_views;
+ props->maxPerStageDescriptorUpdateAfterBindStorageImages = max_bindless_views;
+ props->maxPerStageDescriptorUpdateAfterBindInputAttachments = MAX_PER_STAGE_DESCRIPTOR_INPUT_ATTACHMENTS;
+ props->maxPerStageUpdateAfterBindResources = UINT32_MAX;
+ props->maxDescriptorSetUpdateAfterBindSamplers = max_bindless_views;
+ props->maxDescriptorSetUpdateAfterBindUniformBuffers = 6 * MAX_PER_STAGE_DESCRIPTOR_UNIFORM_BUFFERS;
+ props->maxDescriptorSetUpdateAfterBindUniformBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2;
+ props->maxDescriptorSetUpdateAfterBindStorageBuffers = UINT32_MAX;
+ props->maxDescriptorSetUpdateAfterBindStorageBuffersDynamic = MAX_DYNAMIC_BUFFERS / 2;
+ props->maxDescriptorSetUpdateAfterBindSampledImages = max_bindless_views;
+ props->maxDescriptorSetUpdateAfterBindStorageImages = max_bindless_views;
+ props->maxDescriptorSetUpdateAfterBindInputAttachments = MAX_DESCRIPTOR_SET_INPUT_ATTACHMENTS;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_DRIVER_PROPERTIES_KHR: {
+ VkPhysicalDeviceDriverPropertiesKHR *driver_props =
+ (VkPhysicalDeviceDriverPropertiesKHR *) ext;
+
+ driver_props->driverID = VK_DRIVER_ID_INTEL_OPEN_SOURCE_MESA_KHR;
+ snprintf(driver_props->driverName, VK_MAX_DRIVER_NAME_SIZE_KHR,
+ "Intel open-source Mesa driver");
+
+ snprintf(driver_props->driverInfo, VK_MAX_DRIVER_INFO_SIZE_KHR,
+ "Mesa " PACKAGE_VERSION MESA_GIT_SHA1);
+
+ driver_props->conformanceVersion = (VkConformanceVersionKHR) {
+ .major = 1,
+ .minor = 1,
+ .subminor = 2,
+ .patch = 0,
+ };
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_EXTERNAL_MEMORY_HOST_PROPERTIES_EXT: {
+ VkPhysicalDeviceExternalMemoryHostPropertiesEXT *props =
+ (VkPhysicalDeviceExternalMemoryHostPropertiesEXT *) ext;
+ /* Userptr needs page aligned memory. */
+ props->minImportedHostPointerAlignment = 4096;
break;
}
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_PROPERTIES_EXT: {
+ VkPhysicalDeviceInlineUniformBlockPropertiesEXT *props =
+ (VkPhysicalDeviceInlineUniformBlockPropertiesEXT *)ext;
+ props->maxInlineUniformBlockSize = MAX_INLINE_UNIFORM_BLOCK_SIZE;
+ props->maxPerStageDescriptorInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ props->maxPerStageDescriptorUpdateAfterBindInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ props->maxDescriptorSetInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ props->maxDescriptorSetUpdateAfterBindInlineUniformBlocks =
+ MAX_INLINE_UNIFORM_BLOCK_DESCRIPTORS;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MAINTENANCE_3_PROPERTIES: {
VkPhysicalDeviceMaintenance3Properties *props =
(VkPhysicalDeviceMaintenance3Properties *)ext;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PCI_BUS_INFO_PROPERTIES_EXT: {
+ VkPhysicalDevicePCIBusInfoPropertiesEXT *properties =
+ (VkPhysicalDevicePCIBusInfoPropertiesEXT *)ext;
+ properties->pciDomain = pdevice->pci_info.domain;
+ properties->pciBus = pdevice->pci_info.bus;
+ properties->pciDevice = pdevice->pci_info.device;
+ properties->pciFunction = pdevice->pci_info.function;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_POINT_CLIPPING_PROPERTIES: {
VkPhysicalDevicePointClippingProperties *properties =
(VkPhysicalDevicePointClippingProperties *) ext;
- properties->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_ALL_CLIP_PLANES;
- anv_finishme("Implement pop-free point clipping");
+ properties->pointClippingBehavior = VK_POINT_CLIPPING_BEHAVIOR_USER_CLIP_PLANES_ONLY;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_PROPERTIES: {
+ VkPhysicalDeviceProtectedMemoryProperties *props =
+ (VkPhysicalDeviceProtectedMemoryProperties *)ext;
+ props->protectedNoFault = false;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PUSH_DESCRIPTOR_PROPERTIES_KHR: {
+ VkPhysicalDevicePushDescriptorPropertiesKHR *properties =
+ (VkPhysicalDevicePushDescriptorPropertiesKHR *) ext;
+
+ properties->maxPushDescriptors = MAX_PUSH_DESCRIPTORS;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLER_FILTER_MINMAX_PROPERTIES_EXT: {
+ VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *properties =
+ (VkPhysicalDeviceSamplerFilterMinmaxPropertiesEXT *)ext;
+ properties->filterMinmaxImageComponentMapping = pdevice->info.gen >= 9;
+ properties->filterMinmaxSingleComponentFormats = true;
break;
}
VK_SUBGROUP_FEATURE_SHUFFLE_RELATIVE_BIT |
VK_SUBGROUP_FEATURE_CLUSTERED_BIT |
VK_SUBGROUP_FEATURE_QUAD_BIT;
- properties->quadOperationsInAllStages = VK_TRUE;
+ properties->quadOperationsInAllStages = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TEXEL_BUFFER_ALIGNMENT_PROPERTIES_EXT: {
+ VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT *props =
+ (VkPhysicalDeviceTexelBufferAlignmentPropertiesEXT *)ext;
+
+ /* From the SKL PRM Vol. 2d, docs for RENDER_SURFACE_STATE::Surface
+ * Base Address:
+ *
+ * "For SURFTYPE_BUFFER non-rendertarget surfaces, this field
+ * specifies the base address of the first element of the surface,
+ * computed in software by adding the surface base address to the
+ * byte offset of the element in the buffer. The base address must
+ * be aligned to element size."
+ *
+ * The typed dataport messages require that things be texel aligned.
+ * Otherwise, we may just load/store the wrong data or, in the worst
+ * case, there may be hangs.
+ */
+ props->storageTexelBufferOffsetAlignmentBytes = 16;
+ props->storageTexelBufferOffsetSingleTexelAlignment = true;
+
+ /* The sampler, however, is much more forgiving and it can handle
+ * arbitrary byte alignment for linear and buffer surfaces. It's
+ * hard to find a good PRM citation for this but years of empirical
+ * experience demonstrate that this is true.
+ */
+ props->uniformTexelBufferOffsetAlignmentBytes = 1;
+ props->uniformTexelBufferOffsetSingleTexelAlignment = false;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_TRANSFORM_FEEDBACK_PROPERTIES_EXT: {
+ VkPhysicalDeviceTransformFeedbackPropertiesEXT *props =
+ (VkPhysicalDeviceTransformFeedbackPropertiesEXT *)ext;
+
+ props->maxTransformFeedbackStreams = MAX_XFB_STREAMS;
+ props->maxTransformFeedbackBuffers = MAX_XFB_BUFFERS;
+ props->maxTransformFeedbackBufferSize = (1ull << 32);
+ props->maxTransformFeedbackStreamDataSize = 128 * 4;
+ props->maxTransformFeedbackBufferDataSize = 128 * 4;
+ props->maxTransformFeedbackBufferDataStride = 2048;
+ props->transformFeedbackQueries = true;
+ props->transformFeedbackStreamsLinesTriangles = false;
+ props->transformFeedbackRasterizationStreamSelect = false;
+ props->transformFeedbackDraw = true;
break;
}
}
}
+static void
+anv_get_memory_budget(VkPhysicalDevice physicalDevice,
+ VkPhysicalDeviceMemoryBudgetPropertiesEXT *memoryBudget)
+{
+ ANV_FROM_HANDLE(anv_physical_device, device, physicalDevice);
+ uint64_t sys_available = get_available_system_memory();
+ assert(sys_available > 0);
+
+ VkDeviceSize total_heaps_size = 0;
+ for (size_t i = 0; i < device->memory.heap_count; i++)
+ total_heaps_size += device->memory.heaps[i].size;
+
+ for (size_t i = 0; i < device->memory.heap_count; i++) {
+ VkDeviceSize heap_size = device->memory.heaps[i].size;
+ VkDeviceSize heap_used = device->memory.heaps[i].used;
+ VkDeviceSize heap_budget;
+
+ double heap_proportion = (double) heap_size / total_heaps_size;
+ VkDeviceSize sys_available_prop = sys_available * heap_proportion;
+
+ /*
+ * Let's not incite the app to starve the system: report at most 90% of
+ * available system memory.
+ */
+ uint64_t heap_available = sys_available_prop * 9 / 10;
+ heap_budget = MIN2(heap_size, heap_used + heap_available);
+
+ /*
+ * Round down to the nearest MB
+ */
+ heap_budget &= ~((1ull << 20) - 1);
+
+ /*
+ * The heapBudget value must be non-zero for array elements less than
+ * VkPhysicalDeviceMemoryProperties::memoryHeapCount. The heapBudget
+ * value must be less than or equal to VkMemoryHeap::size for each heap.
+ */
+ assert(0 < heap_budget && heap_budget <= heap_size);
+
+ memoryBudget->heapUsage[i] = heap_used;
+ memoryBudget->heapBudget[i] = heap_budget;
+ }
+
+ /* The heapBudget and heapUsage values must be zero for array elements
+ * greater than or equal to VkPhysicalDeviceMemoryProperties::memoryHeapCount
+ */
+ for (uint32_t i = device->memory.heap_count; i < VK_MAX_MEMORY_HEAPS; i++) {
+ memoryBudget->heapBudget[i] = 0;
+ memoryBudget->heapUsage[i] = 0;
+ }
+}
+
void anv_GetPhysicalDeviceMemoryProperties2(
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceMemoryProperties2* pMemoryProperties)
vk_foreach_struct(ext, pMemoryProperties->pNext) {
switch (ext->sType) {
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MEMORY_BUDGET_PROPERTIES_EXT:
+ anv_get_memory_budget(physicalDevice, (void*)ext);
+ break;
default:
anv_debug_ignored_stype(ext->sType);
break;
if (instance == NULL)
return NULL;
- int idx = anv_get_entrypoint_index(pName);
- if (idx < 0)
- return NULL;
+ int idx = anv_get_instance_entrypoint_index(pName);
+ if (idx >= 0)
+ return instance->dispatch.entrypoints[idx];
+
+ idx = anv_get_device_entrypoint_index(pName);
+ if (idx >= 0)
+ return instance->device_dispatch.entrypoints[idx];
- return instance->dispatch.entrypoints[idx];
+ return NULL;
}
/* With version 1+ of the loader interface the ICD should expose
if (!device || !pName)
return NULL;
- int idx = anv_get_entrypoint_index(pName);
+ int idx = anv_get_device_entrypoint_index(pName);
if (idx < 0)
return NULL;
state = anv_state_pool_alloc(pool, size, align);
memcpy(state.map, p, size);
- anv_state_flush(pool->block_pool.device, state);
-
return state;
}
+/* Haswell border color is a bit of a disaster. Float and unorm formats use a
+ * straightforward 32-bit float color in the first 64 bytes. Instead of using
+ * a nice float/integer union like Gen8+, Haswell specifies the integer border
+ * color as a separate entry /after/ the float color. The layout of this entry
+ * also depends on the format's bpp (with extra hacks for RG32), and overlaps.
+ *
+ * Since we don't know the format/bpp, we can't make any of the border colors
+ * containing '1' work for all formats, as it would be in the wrong place for
+ * some of them. We opt to make 32-bit integers work as this seems like the
+ * most common option. Fortunately, transparent black works regardless, as
+ * all zeroes is the same in every bit-size.
+ */
+struct hsw_border_color {
+ float float32[4];
+ uint32_t _pad0[12];
+ uint32_t uint32[4];
+ uint32_t _pad1[108];
+};
+
struct gen8_border_color {
union {
float float32[4];
static void
anv_device_init_border_colors(struct anv_device *device)
{
- 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_TRANSPARENT_BLACK] = { .uint32 = { 0, 0, 0, 0 } },
- [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { .uint32 = { 0, 0, 0, 1 } },
- [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
- };
+ if (device->info.is_haswell) {
+ static const struct hsw_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_TRANSPARENT_BLACK] = { .uint32 = { 0, 0, 0, 0 } },
+ [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { .uint32 = { 0, 0, 0, 1 } },
+ [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
+ };
+
+ device->border_colors =
+ anv_state_pool_emit_data(&device->dynamic_state_pool,
+ sizeof(border_colors), 512, border_colors);
+ } else {
+ 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_TRANSPARENT_BLACK] = { .uint32 = { 0, 0, 0, 0 } },
+ [VK_BORDER_COLOR_INT_OPAQUE_BLACK] = { .uint32 = { 0, 0, 0, 1 } },
+ [VK_BORDER_COLOR_INT_OPAQUE_WHITE] = { .uint32 = { 1, 1, 1, 1 } },
+ };
- device->border_colors = anv_state_pool_emit_data(&device->dynamic_state_pool,
- sizeof(border_colors), 64,
- border_colors);
+ device->border_colors =
+ anv_state_pool_emit_data(&device->dynamic_state_pool,
+ sizeof(border_colors), 64, border_colors);
+ }
}
static void
{
ANV_FROM_HANDLE(anv_physical_device, device, physicalDevice);
VK_OUTARRAY_MAKE(out, pProperties, pPropertyCount);
- (void)device;
for (int i = 0; i < ANV_DEVICE_EXTENSION_COUNT; i++) {
if (device->supported_extensions.extensions[i]) {
static void
anv_device_init_dispatch(struct anv_device *device)
{
- const struct anv_dispatch_table *genX_table;
+ const struct anv_device_dispatch_table *genX_table;
switch (device->info.gen) {
case 11:
- genX_table = &gen11_dispatch_table;
+ genX_table = &gen11_device_dispatch_table;
break;
case 10:
- genX_table = &gen10_dispatch_table;
+ genX_table = &gen10_device_dispatch_table;
break;
case 9:
- genX_table = &gen9_dispatch_table;
+ genX_table = &gen9_device_dispatch_table;
break;
case 8:
- genX_table = &gen8_dispatch_table;
+ genX_table = &gen8_device_dispatch_table;
break;
case 7:
if (device->info.is_haswell)
- genX_table = &gen75_dispatch_table;
+ genX_table = &gen75_device_dispatch_table;
else
- genX_table = &gen7_dispatch_table;
+ genX_table = &gen7_device_dispatch_table;
break;
default:
unreachable("unsupported gen\n");
/* Vulkan requires that entrypoints for extensions which have not been
* enabled must not be advertised.
*/
- if (!anv_entrypoint_is_enabled(i, device->instance->apiVersion,
- &device->instance->enabled_extensions,
- &device->enabled_extensions)) {
+ if (!anv_device_entrypoint_is_enabled(i, device->instance->app_info.api_version,
+ &device->instance->enabled_extensions,
+ &device->enabled_extensions)) {
device->dispatch.entrypoints[i] = NULL;
} else if (genX_table->entrypoints[i]) {
device->dispatch.entrypoints[i] = genX_table->entrypoints[i];
} else {
- device->dispatch.entrypoints[i] = anv_dispatch_table.entrypoints[i];
+ device->dispatch.entrypoints[i] =
+ anv_device_dispatch_table.entrypoints[i];
}
}
}
}
static void
-anv_device_init_hiz_clear_batch(struct anv_device *device)
+anv_device_init_hiz_clear_value_bo(struct anv_device *device)
{
anv_bo_init_new(&device->hiz_clear_bo, device, 4096);
+
+ if (device->instance->physicalDevice.has_exec_async)
+ device->hiz_clear_bo.flags |= EXEC_OBJECT_ASYNC;
+
+ if (device->instance->physicalDevice.use_softpin)
+ device->hiz_clear_bo.flags |= EXEC_OBJECT_PINNED;
+
+ anv_vma_alloc(device, &device->hiz_clear_bo);
+
uint32_t *map = anv_gem_mmap(device, device->hiz_clear_bo.gem_handle,
0, 4096, 0);
anv_gem_munmap(map, device->hiz_clear_bo.size);
}
+static bool
+get_bo_from_pool(struct gen_batch_decode_bo *ret,
+ struct anv_block_pool *pool,
+ uint64_t address)
+{
+ for (uint32_t i = 0; i < pool->nbos; i++) {
+ uint64_t bo_address = pool->bos[i].offset & (~0ull >> 16);
+ uint32_t bo_size = pool->bos[i].size;
+ if (address >= bo_address && address < (bo_address + bo_size)) {
+ *ret = (struct gen_batch_decode_bo) {
+ .addr = bo_address,
+ .size = bo_size,
+ .map = pool->bos[i].map,
+ };
+ return true;
+ }
+ }
+ return false;
+}
+
+/* Finding a buffer for batch decoding */
+static struct gen_batch_decode_bo
+decode_get_bo(void *v_batch, bool ppgtt, uint64_t address)
+{
+ struct anv_device *device = v_batch;
+ struct gen_batch_decode_bo ret_bo = {};
+
+ assert(ppgtt);
+
+ if (get_bo_from_pool(&ret_bo, &device->dynamic_state_pool.block_pool, address))
+ return ret_bo;
+ if (get_bo_from_pool(&ret_bo, &device->instruction_state_pool.block_pool, address))
+ return ret_bo;
+ if (get_bo_from_pool(&ret_bo, &device->binding_table_pool.block_pool, address))
+ return ret_bo;
+ if (get_bo_from_pool(&ret_bo, &device->surface_state_pool.block_pool, address))
+ return ret_bo;
+
+ if (!device->cmd_buffer_being_decoded)
+ return (struct gen_batch_decode_bo) { };
+
+ struct anv_batch_bo **bo;
+
+ u_vector_foreach(bo, &device->cmd_buffer_being_decoded->seen_bbos) {
+ /* The decoder zeroes out the top 16 bits, so we need to as well */
+ uint64_t bo_address = (*bo)->bo.offset & (~0ull >> 16);
+
+ if (address >= bo_address && address < bo_address + (*bo)->bo.size) {
+ return (struct gen_batch_decode_bo) {
+ .addr = bo_address,
+ .size = (*bo)->bo.size,
+ .map = (*bo)->bo.map,
+ };
+ }
+ }
+
+ return (struct gen_batch_decode_bo) { };
+}
+
VkResult anv_CreateDevice(
VkPhysicalDevice physicalDevice,
const VkDeviceCreateInfo* pCreateInfo,
if (!device)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
+ if (INTEL_DEBUG & DEBUG_BATCH) {
+ const unsigned decode_flags =
+ GEN_BATCH_DECODE_FULL |
+ ((INTEL_DEBUG & DEBUG_COLOR) ? GEN_BATCH_DECODE_IN_COLOR : 0) |
+ GEN_BATCH_DECODE_OFFSETS |
+ GEN_BATCH_DECODE_FLOATS;
+
+ gen_batch_decode_ctx_init(&device->decoder_ctx,
+ &physical_device->info,
+ stderr, decode_flags, NULL,
+ decode_get_bo, NULL, device);
+ }
+
device->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
device->instance = physical_device->instance;
device->chipset_id = physical_device->chipset_id;
device->no_hw = physical_device->no_hw;
- device->lost = false;
+ device->_lost = false;
if (pAllocator)
device->alloc = *pAllocator;
}
/* keep the page with address zero out of the allocator */
- util_vma_heap_init(&device->vma_lo, LOW_HEAP_MIN_ADDRESS, LOW_HEAP_SIZE);
- device->vma_lo_available =
- physical_device->memory.heaps[physical_device->memory.heap_count - 1].size;
-
- /* Leave the last 4GiB out of the high vma range, so that no state base
- * address + size can overflow 48 bits. For more information see the
- * comment about Wa32bitGeneralStateOffset in anv_allocator.c
- */
- util_vma_heap_init(&device->vma_hi, HIGH_HEAP_MIN_ADDRESS,
- HIGH_HEAP_SIZE);
+ struct anv_memory_heap *low_heap =
+ &physical_device->memory.heaps[physical_device->memory.heap_count - 1];
+ util_vma_heap_init(&device->vma_lo, low_heap->vma_start, low_heap->vma_size);
+ device->vma_lo_available = low_heap->size;
+
+ struct anv_memory_heap *high_heap =
+ &physical_device->memory.heaps[0];
+ util_vma_heap_init(&device->vma_hi, high_heap->vma_start, high_heap->vma_size);
device->vma_hi_available = physical_device->memory.heap_count == 1 ? 0 :
- physical_device->memory.heaps[0].size;
+ high_heap->size;
}
+ list_inithead(&device->memory_objects);
+
/* As per spec, the driver implementation may deny requests to acquire
* a priority above the default priority (MEDIUM) if the caller does not
* have sufficient privileges. In this scenario VK_ERROR_NOT_PERMITTED_EXT
result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
goto fail_mutex;
}
- if (pthread_cond_init(&device->queue_submit, NULL) != 0) {
+ if (pthread_cond_init(&device->queue_submit, &condattr) != 0) {
pthread_condattr_destroy(&condattr);
result = vk_error(VK_ERROR_INITIALIZATION_FAILED);
goto fail_mutex;
goto fail_surface_state_pool;
}
- result = anv_bo_init_new(&device->workaround_bo, device, 1024);
+ result = anv_bo_init_new(&device->workaround_bo, device, 4096);
if (result != VK_SUCCESS)
goto fail_binding_table_pool;
anv_device_init_trivial_batch(device);
if (device->info.gen >= 10)
- anv_device_init_hiz_clear_batch(device);
+ anv_device_init_hiz_clear_value_bo(device);
anv_scratch_pool_init(device, &device->scratch_pool);
const VkAllocationCallbacks* pAllocator)
{
ANV_FROM_HANDLE(anv_device, device, _device);
- struct anv_physical_device *physical_device = &device->instance->physicalDevice;
+ struct anv_physical_device *physical_device;
if (!device)
return;
+ physical_device = &device->instance->physicalDevice;
+
anv_device_finish_blorp(device);
anv_pipeline_cache_finish(&device->default_pipeline_cache);
anv_gem_destroy_context(device, device->context_id);
+ if (INTEL_DEBUG & DEBUG_BATCH)
+ gen_batch_decode_ctx_finish(&device->decoder_ctx);
+
close(device->fd);
vk_free(&device->alloc, device);
*pQueue = NULL;
}
+VkResult
+_anv_device_set_lost(struct anv_device *device,
+ const char *file, int line,
+ const char *msg, ...)
+{
+ VkResult err;
+ va_list ap;
+
+ device->_lost = true;
+
+ va_start(ap, msg);
+ err = __vk_errorv(device->instance, device,
+ VK_DEBUG_REPORT_OBJECT_TYPE_DEVICE_EXT,
+ VK_ERROR_DEVICE_LOST, file, line, msg, ap);
+ va_end(ap);
+
+ if (env_var_as_boolean("ANV_ABORT_ON_DEVICE_LOSS", false))
+ abort();
+
+ return err;
+}
+
VkResult
anv_device_query_status(struct anv_device *device)
{
* for it. However, it doesn't hurt to check and it potentially lets us
* avoid an ioctl.
*/
- if (unlikely(device->lost))
+ if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
uint32_t active, pending;
int ret = anv_gem_gpu_get_reset_stats(device, &active, &pending);
if (ret == -1) {
/* We don't know the real error. */
- device->lost = true;
- return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
- "get_reset_stats failed: %m");
+ return anv_device_set_lost(device, "get_reset_stats failed: %m");
}
if (active) {
- device->lost = true;
- return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
- "GPU hung on one of our command buffers");
+ return anv_device_set_lost(device, "GPU hung on one of our command buffers");
} else if (pending) {
- device->lost = true;
- return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
- "GPU hung with commands in-flight");
+ return anv_device_set_lost(device, "GPU hung with commands in-flight");
}
return VK_SUCCESS;
return VK_NOT_READY;
} else if (ret == -1) {
/* We don't know the real error. */
- device->lost = true;
- return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
- "gem wait failed: %m");
+ return anv_device_set_lost(device, "gem wait failed: %m");
}
/* Query for device status after the busy call. If the BO we're checking
return VK_TIMEOUT;
} else if (ret == -1) {
/* We don't know the real error. */
- device->lost = true;
- return vk_errorf(device->instance, device, VK_ERROR_DEVICE_LOST,
- "gem wait failed: %m");
+ return anv_device_set_lost(device, "gem wait failed: %m");
}
/* Query for device status after the wait. If the BO we're waiting on got
VkDevice _device)
{
ANV_FROM_HANDLE(anv_device, device, _device);
- if (unlikely(device->lost))
+ if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
struct anv_batch batch;
util_vma_heap_free(&device->vma_lo, addr_48b, bo->size);
device->vma_lo_available += bo->size;
} else {
- assert(addr_48b >= HIGH_HEAP_MIN_ADDRESS &&
- addr_48b <= HIGH_HEAP_MAX_ADDRESS);
+ MAYBE_UNUSED const struct anv_physical_device *physical_device =
+ &device->instance->physicalDevice;
+ assert(addr_48b >= physical_device->memory.heaps[0].vma_start &&
+ addr_48b < (physical_device->memory.heaps[0].vma_start +
+ physical_device->memory.heaps[0].vma_size));
util_vma_heap_free(&device->vma_hi, addr_48b, bo->size);
device->vma_hi_available += bo->size;
}
mem->type = &pdevice->memory.types[pAllocateInfo->memoryTypeIndex];
mem->map = NULL;
mem->map_size = 0;
+ mem->ahw = NULL;
+ mem->host_ptr = NULL;
uint64_t bo_flags = 0;
if (pdevice->use_softpin)
bo_flags |= EXEC_OBJECT_PINNED;
+ const VkExportMemoryAllocateInfo *export_info =
+ vk_find_struct_const(pAllocateInfo->pNext, EXPORT_MEMORY_ALLOCATE_INFO);
+
+ /* Check if we need to support Android HW buffer export. If so,
+ * create AHardwareBuffer and import memory from it.
+ */
+ bool android_export = false;
+ if (export_info && export_info->handleTypes &
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_ANDROID_HARDWARE_BUFFER_BIT_ANDROID)
+ android_export = true;
+
+ /* Android memory import. */
+ const struct VkImportAndroidHardwareBufferInfoANDROID *ahw_import_info =
+ vk_find_struct_const(pAllocateInfo->pNext,
+ IMPORT_ANDROID_HARDWARE_BUFFER_INFO_ANDROID);
+
+ if (ahw_import_info) {
+ result = anv_import_ahw_memory(_device, mem, ahw_import_info);
+ if (result != VK_SUCCESS)
+ goto fail;
+
+ goto success;
+ } else if (android_export) {
+ result = anv_create_ahw_memory(_device, mem, pAllocateInfo);
+ if (result != VK_SUCCESS)
+ goto fail;
+
+ const struct VkImportAndroidHardwareBufferInfoANDROID import_info = {
+ .buffer = mem->ahw,
+ };
+ result = anv_import_ahw_memory(_device, mem, &import_info);
+ if (result != VK_SUCCESS)
+ goto fail;
+
+ goto success;
+ }
+
const VkImportMemoryFdInfoKHR *fd_info =
vk_find_struct_const(pAllocateInfo->pNext, IMPORT_MEMORY_FD_INFO_KHR);
fd_info->handleType ==
VK_EXTERNAL_MEMORY_HANDLE_TYPE_DMA_BUF_BIT_EXT);
- result = anv_bo_cache_import(device, &device->bo_cache,
- fd_info->fd, bo_flags, &mem->bo);
+ result = anv_bo_cache_import(device, &device->bo_cache, fd_info->fd,
+ bo_flags | ANV_BO_EXTERNAL, &mem->bo);
if (result != VK_SUCCESS)
goto fail;
*/
if (mem->bo->size < aligned_alloc_size) {
result = vk_errorf(device->instance, device,
- VK_ERROR_INVALID_EXTERNAL_HANDLE_KHR,
+ VK_ERROR_INVALID_EXTERNAL_HANDLE,
"aligned allocationSize too large for "
- "VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT_KHR: "
+ "VK_EXTERNAL_MEMORY_HANDLE_TYPE_OPAQUE_FD_BIT: "
"%"PRIu64"B > %"PRIu64"B",
aligned_alloc_size, mem->bo->size);
anv_bo_cache_release(device, &device->bo_cache, mem->bo);
* If the import fails, we leave the file descriptor open.
*/
close(fd_info->fd);
- } else {
- result = anv_bo_cache_alloc(device, &device->bo_cache,
- pAllocateInfo->allocationSize, bo_flags,
- &mem->bo);
+ goto success;
+ }
+
+ const VkImportMemoryHostPointerInfoEXT *host_ptr_info =
+ vk_find_struct_const(pAllocateInfo->pNext,
+ IMPORT_MEMORY_HOST_POINTER_INFO_EXT);
+ if (host_ptr_info && host_ptr_info->handleType) {
+ if (host_ptr_info->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_MAPPED_FOREIGN_MEMORY_BIT_EXT) {
+ result = vk_error(VK_ERROR_INVALID_EXTERNAL_HANDLE);
+ goto fail;
+ }
+
+ assert(host_ptr_info->handleType ==
+ VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT);
+
+ result = anv_bo_cache_import_host_ptr(
+ device, &device->bo_cache, host_ptr_info->pHostPointer,
+ pAllocateInfo->allocationSize, bo_flags, &mem->bo);
+
if (result != VK_SUCCESS)
goto fail;
- const VkMemoryDedicatedAllocateInfoKHR *dedicated_info =
- vk_find_struct_const(pAllocateInfo->pNext, MEMORY_DEDICATED_ALLOCATE_INFO_KHR);
- if (dedicated_info && dedicated_info->image != VK_NULL_HANDLE) {
- ANV_FROM_HANDLE(anv_image, image, dedicated_info->image);
+ mem->host_ptr = host_ptr_info->pHostPointer;
+ goto success;
+ }
- /* Some legacy (non-modifiers) consumers need the tiling to be set on
- * the BO. In this case, we have a dedicated allocation.
- */
- if (image->needs_set_tiling) {
- const uint32_t i915_tiling =
- isl_tiling_to_i915_tiling(image->planes[0].surface.isl.tiling);
- int ret = anv_gem_set_tiling(device, mem->bo->gem_handle,
- image->planes[0].surface.isl.row_pitch,
- i915_tiling);
- if (ret) {
- anv_bo_cache_release(device, &device->bo_cache, mem->bo);
- return vk_errorf(device->instance, NULL,
- VK_ERROR_OUT_OF_DEVICE_MEMORY,
- "failed to set BO tiling: %m");
- }
+ /* Regular allocate (not importing memory). */
+
+ if (export_info && export_info->handleTypes)
+ bo_flags |= ANV_BO_EXTERNAL;
+
+ result = anv_bo_cache_alloc(device, &device->bo_cache,
+ pAllocateInfo->allocationSize, bo_flags,
+ &mem->bo);
+ if (result != VK_SUCCESS)
+ goto fail;
+
+ const VkMemoryDedicatedAllocateInfo *dedicated_info =
+ vk_find_struct_const(pAllocateInfo->pNext, MEMORY_DEDICATED_ALLOCATE_INFO);
+ if (dedicated_info && dedicated_info->image != VK_NULL_HANDLE) {
+ ANV_FROM_HANDLE(anv_image, image, dedicated_info->image);
+
+ /* Some legacy (non-modifiers) consumers need the tiling to be set on
+ * the BO. In this case, we have a dedicated allocation.
+ */
+ if (image->needs_set_tiling) {
+ const uint32_t i915_tiling =
+ isl_tiling_to_i915_tiling(image->planes[0].surface.isl.tiling);
+ int ret = anv_gem_set_tiling(device, mem->bo->gem_handle,
+ image->planes[0].surface.isl.row_pitch_B,
+ i915_tiling);
+ if (ret) {
+ anv_bo_cache_release(device, &device->bo_cache, mem->bo);
+ return vk_errorf(device->instance, NULL,
+ VK_ERROR_OUT_OF_DEVICE_MEMORY,
+ "failed to set BO tiling: %m");
}
}
}
+ success:
+ pthread_mutex_lock(&device->mutex);
+ list_addtail(&mem->link, &device->memory_objects);
+ pthread_mutex_unlock(&device->mutex);
+
*pMem = anv_device_memory_to_handle(mem);
+ p_atomic_add(&pdevice->memory.heaps[mem->type->heapIndex].used,
+ mem->bo->size);
+
return VK_SUCCESS;
fail:
VkResult anv_GetMemoryFdPropertiesKHR(
VkDevice _device,
- VkExternalMemoryHandleTypeFlagBitsKHR handleType,
+ VkExternalMemoryHandleTypeFlagBits handleType,
int fd,
VkMemoryFdPropertiesKHR* pMemoryFdProperties)
{
}
}
+VkResult anv_GetMemoryHostPointerPropertiesEXT(
+ VkDevice _device,
+ VkExternalMemoryHandleTypeFlagBits handleType,
+ const void* pHostPointer,
+ VkMemoryHostPointerPropertiesEXT* pMemoryHostPointerProperties)
+{
+ ANV_FROM_HANDLE(anv_device, device, _device);
+
+ assert(pMemoryHostPointerProperties->sType ==
+ VK_STRUCTURE_TYPE_MEMORY_HOST_POINTER_PROPERTIES_EXT);
+
+ switch (handleType) {
+ case VK_EXTERNAL_MEMORY_HANDLE_TYPE_HOST_ALLOCATION_BIT_EXT: {
+ struct anv_physical_device *pdevice = &device->instance->physicalDevice;
+
+ /* Host memory can be imported as any memory type. */
+ pMemoryHostPointerProperties->memoryTypeBits =
+ (1ull << pdevice->memory.type_count) - 1;
+
+ return VK_SUCCESS;
+ }
+ default:
+ return VK_ERROR_INVALID_EXTERNAL_HANDLE;
+ }
+}
+
void anv_FreeMemory(
VkDevice _device,
VkDeviceMemory _mem,
{
ANV_FROM_HANDLE(anv_device, device, _device);
ANV_FROM_HANDLE(anv_device_memory, mem, _mem);
+ struct anv_physical_device *pdevice = &device->instance->physicalDevice;
if (mem == NULL)
return;
+ pthread_mutex_lock(&device->mutex);
+ list_del(&mem->link);
+ pthread_mutex_unlock(&device->mutex);
+
if (mem->map)
anv_UnmapMemory(_device, _mem);
+ p_atomic_add(&pdevice->memory.heaps[mem->type->heapIndex].used,
+ -mem->bo->size);
+
anv_bo_cache_release(device, &device->bo_cache, mem->bo);
+#if defined(ANDROID) && ANDROID_API_LEVEL >= 26
+ if (mem->ahw)
+ AHardwareBuffer_release(mem->ahw);
+#endif
+
vk_free2(&device->alloc, pAllocator, mem);
}
return VK_SUCCESS;
}
+ if (mem->host_ptr) {
+ *ppData = mem->host_ptr + offset;
+ return VK_SUCCESS;
+ }
+
if (size == VK_WHOLE_SIZE)
size = mem->bo->size - offset;
{
ANV_FROM_HANDLE(anv_device_memory, mem, _memory);
- if (mem == NULL)
+ if (mem == NULL || mem->host_ptr)
return;
anv_gem_munmap(mem->map, mem->map_size);
switch (ext->sType) {
case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS: {
VkMemoryDedicatedRequirements *requirements = (void *)ext;
- requirements->prefersDedicatedAllocation = VK_FALSE;
- requirements->requiresDedicatedAllocation = VK_FALSE;
+ requirements->prefersDedicatedAllocation = false;
+ requirements->requiresDedicatedAllocation = false;
break;
}
*/
uint32_t memory_types = (1ull << pdevice->memory.type_count) - 1;
+ /* We must have image allocated or imported at this point. According to the
+ * specification, external images must have been bound to memory before
+ * calling GetImageMemoryRequirements.
+ */
+ assert(image->size > 0);
+
pMemoryRequirements->size = image->size;
pMemoryRequirements->alignment = image->alignment;
pMemoryRequirements->memoryTypeBits = memory_types;
switch (ext->sType) {
case VK_STRUCTURE_TYPE_IMAGE_PLANE_MEMORY_REQUIREMENTS_INFO: {
struct anv_physical_device *pdevice = &device->instance->physicalDevice;
- const VkImagePlaneMemoryRequirementsInfoKHR *plane_reqs =
- (const VkImagePlaneMemoryRequirementsInfoKHR *) ext;
+ const VkImagePlaneMemoryRequirementsInfo *plane_reqs =
+ (const VkImagePlaneMemoryRequirementsInfo *) ext;
uint32_t plane = anv_image_aspect_to_plane(image->aspects,
plane_reqs->planeAspect);
pMemoryRequirements->memoryRequirements.memoryTypeBits =
(1ull << pdevice->memory.type_count) - 1;
+ /* We must have image allocated or imported at this point. According to the
+ * specification, external images must have been bound to memory before
+ * calling GetImageMemoryRequirements.
+ */
+ assert(image->planes[plane].size > 0);
+
pMemoryRequirements->memoryRequirements.size = image->planes[plane].size;
pMemoryRequirements->memoryRequirements.alignment =
image->planes[plane].alignment;
switch (ext->sType) {
case VK_STRUCTURE_TYPE_MEMORY_DEDICATED_REQUIREMENTS: {
VkMemoryDedicatedRequirements *requirements = (void *)ext;
- if (image->needs_set_tiling) {
+ if (image->needs_set_tiling || image->external_format) {
/* If we need to set the tiling for external consumers, we need a
* dedicated allocation.
*
* See also anv_AllocateMemory.
*/
- requirements->prefersDedicatedAllocation = VK_TRUE;
- requirements->requiresDedicatedAllocation = VK_TRUE;
+ requirements->prefersDedicatedAllocation = true;
+ requirements->requiresDedicatedAllocation = true;
} else {
- requirements->prefersDedicatedAllocation = VK_FALSE;
- requirements->requiresDedicatedAllocation = VK_FALSE;
+ requirements->prefersDedicatedAllocation = false;
+ requirements->requiresDedicatedAllocation = false;
}
break;
}
VkFence fence)
{
ANV_FROM_HANDLE(anv_queue, queue, _queue);
- if (unlikely(queue->device->lost))
+ if (anv_device_is_lost(queue->device))
return VK_ERROR_DEVICE_LOST;
return vk_error(VK_ERROR_FEATURE_NOT_PRESENT);
ANV_FROM_HANDLE(anv_device, device, _device);
ANV_FROM_HANDLE(anv_event, event, _event);
- if (unlikely(device->lost))
+ if (anv_device_is_lost(device))
return VK_ERROR_DEVICE_LOST;
if (!device->info.has_llc) {
vk_free2(&device->alloc, pAllocator, buffer);
}
+VkDeviceAddress anv_GetBufferDeviceAddressEXT(
+ VkDevice device,
+ const VkBufferDeviceAddressInfoEXT* pInfo)
+{
+ ANV_FROM_HANDLE(anv_buffer, buffer, pInfo->buffer);
+
+ assert(buffer->address.bo->flags & EXEC_OBJECT_PINNED);
+
+ return anv_address_physical(buffer->address);
+}
+
void
anv_fill_buffer_surface_state(struct anv_device *device, struct anv_state state,
enum isl_format format,
isl_buffer_fill_state(&device->isl_dev, state.map,
.address = anv_address_physical(address),
.mocs = device->default_mocs,
- .size = range,
+ .size_B = range,
.format = format,
- .stride = stride);
-
- anv_state_flush(device, state);
+ .swizzle = ISL_SWIZZLE_IDENTITY,
+ .stride_B = stride);
}
void anv_DestroySampler(
if (!sampler)
return;
+ if (sampler->bindless_state.map) {
+ anv_state_pool_free(&device->dynamic_state_pool,
+ sampler->bindless_state);
+ }
+
vk_free2(&device->alloc, pAllocator, sampler);
}
vk_free2(&device->alloc, pAllocator, fb);
}
+static const VkTimeDomainEXT anv_time_domains[] = {
+ VK_TIME_DOMAIN_DEVICE_EXT,
+ VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT,
+ VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT,
+};
+
+VkResult anv_GetPhysicalDeviceCalibrateableTimeDomainsEXT(
+ VkPhysicalDevice physicalDevice,
+ uint32_t *pTimeDomainCount,
+ VkTimeDomainEXT *pTimeDomains)
+{
+ int d;
+ VK_OUTARRAY_MAKE(out, pTimeDomains, pTimeDomainCount);
+
+ for (d = 0; d < ARRAY_SIZE(anv_time_domains); d++) {
+ vk_outarray_append(&out, i) {
+ *i = anv_time_domains[d];
+ }
+ }
+
+ return vk_outarray_status(&out);
+}
+
+static uint64_t
+anv_clock_gettime(clockid_t clock_id)
+{
+ struct timespec current;
+ int ret;
+
+ ret = clock_gettime(clock_id, ¤t);
+ if (ret < 0 && clock_id == CLOCK_MONOTONIC_RAW)
+ ret = clock_gettime(CLOCK_MONOTONIC, ¤t);
+ if (ret < 0)
+ return 0;
+
+ return (uint64_t) current.tv_sec * 1000000000ULL + current.tv_nsec;
+}
+
+#define TIMESTAMP 0x2358
+
+VkResult anv_GetCalibratedTimestampsEXT(
+ VkDevice _device,
+ uint32_t timestampCount,
+ const VkCalibratedTimestampInfoEXT *pTimestampInfos,
+ uint64_t *pTimestamps,
+ uint64_t *pMaxDeviation)
+{
+ ANV_FROM_HANDLE(anv_device, device, _device);
+ uint64_t timestamp_frequency = device->info.timestamp_frequency;
+ int ret;
+ int d;
+ uint64_t begin, end;
+ uint64_t max_clock_period = 0;
+
+ begin = anv_clock_gettime(CLOCK_MONOTONIC_RAW);
+
+ for (d = 0; d < timestampCount; d++) {
+ switch (pTimestampInfos[d].timeDomain) {
+ case VK_TIME_DOMAIN_DEVICE_EXT:
+ ret = anv_gem_reg_read(device, TIMESTAMP | 1,
+ &pTimestamps[d]);
+
+ if (ret != 0) {
+ return anv_device_set_lost(device, "Failed to read the TIMESTAMP "
+ "register: %m");
+ }
+ uint64_t device_period = DIV_ROUND_UP(1000000000, timestamp_frequency);
+ max_clock_period = MAX2(max_clock_period, device_period);
+ break;
+ case VK_TIME_DOMAIN_CLOCK_MONOTONIC_EXT:
+ pTimestamps[d] = anv_clock_gettime(CLOCK_MONOTONIC);
+ max_clock_period = MAX2(max_clock_period, 1);
+ break;
+
+ case VK_TIME_DOMAIN_CLOCK_MONOTONIC_RAW_EXT:
+ pTimestamps[d] = begin;
+ break;
+ default:
+ pTimestamps[d] = 0;
+ break;
+ }
+ }
+
+ end = anv_clock_gettime(CLOCK_MONOTONIC_RAW);
+
+ /*
+ * The maximum deviation is the sum of the interval over which we
+ * perform the sampling and the maximum period of any sampled
+ * clock. That's because the maximum skew between any two sampled
+ * clock edges is when the sampled clock with the largest period is
+ * sampled at the end of that period but right at the beginning of the
+ * sampling interval and some other clock is sampled right at the
+ * begining of its sampling period and right at the end of the
+ * sampling interval. Let's assume the GPU has the longest clock
+ * period and that the application is sampling GPU and monotonic:
+ *
+ * s e
+ * w x y z 0 1 2 3 4 5 6 7 8 9 a b c d e f
+ * Raw -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
+ *
+ * g
+ * 0 1 2 3
+ * GPU -----_____-----_____-----_____-----_____
+ *
+ * m
+ * x y z 0 1 2 3 4 5 6 7 8 9 a b c
+ * Monotonic -_-_-_-_-_-_-_-_-_-_-_-_-_-_-_-
+ *
+ * Interval <----------------->
+ * Deviation <-------------------------->
+ *
+ * s = read(raw) 2
+ * g = read(GPU) 1
+ * m = read(monotonic) 2
+ * e = read(raw) b
+ *
+ * We round the sample interval up by one tick to cover sampling error
+ * in the interval clock
+ */
+
+ uint64_t sample_interval = end - begin + 1;
+
+ *pMaxDeviation = sample_interval + max_clock_period;
+
+ return VK_SUCCESS;
+}
+
/* vk_icd.h does not declare this function, so we declare it here to
* suppress Wmissing-prototypes.
*/
projects / mesa.git / blobdiff