#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 "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_final(&sha1_ctx, sha1);
memcpy(device->pipeline_cache_uuid, sha1, VK_UUID_SIZE);
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;
/* 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);
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));
vk_debug_report_instance_destroy(&instance->debug_report_callbacks);
+ glsl_type_singleton_decref();
_mesa_locale_fini();
vk_free(&instance->alloc, instance);
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);
VkPhysicalDevice physicalDevice,
VkPhysicalDeviceFeatures2* pFeatures)
{
+ ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
anv_GetPhysicalDeviceFeatures(physicalDevice, &pFeatures->features);
vk_foreach_struct(ext, pFeatures->pNext) {
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_8BIT_STORAGE_FEATURES_KHR: {
VkPhysicalDevice8BitStorageFeaturesKHR *features =
(VkPhysicalDevice8BitStorageFeaturesKHR *)ext;
- ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
-
features->storageBuffer8BitAccess = pdevice->info.gen >= 8;
features->uniformAndStorageBuffer8BitAccess = pdevice->info.gen >= 8;
features->storagePushConstant8 = pdevice->info.gen >= 8;
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_16BIT_STORAGE_FEATURES: {
VkPhysicalDevice16BitStorageFeatures *features =
(VkPhysicalDevice16BitStorageFeatures *)ext;
- ANV_FROM_HANDLE(anv_physical_device, pdevice, physicalDevice);
-
features->storageBuffer16BitAccess = pdevice->info.gen >= 8;
features->uniformAndStorageBuffer16BitAccess = pdevice->info.gen >= 8;
features->storagePushConstant16 = pdevice->info.gen >= 8;
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_HOST_QUERY_RESET_FEATURES_EXT: {
+ VkPhysicalDeviceHostQueryResetFeaturesEXT *features =
+ (VkPhysicalDeviceHostQueryResetFeaturesEXT *)ext;
+ features->hostQueryReset = true;
+ break;
+ }
+
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_INLINE_UNIFORM_BLOCK_FEATURES_EXT: {
+ VkPhysicalDeviceInlineUniformBlockFeaturesEXT *features =
+ (VkPhysicalDeviceInlineUniformBlockFeaturesEXT *)ext;
+ features->inlineUniformBlock = true;
+ features->descriptorBindingInlineUniformBlockUpdateAfterBind = false;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_MULTIVIEW_FEATURES: {
VkPhysicalDeviceMultiviewFeatures *features =
(VkPhysicalDeviceMultiviewFeatures *)ext;
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PROTECTED_MEMORY_FEATURES: {
VkPhysicalDeviceProtectedMemoryFeatures *features = (void *)ext;
- features->protectedMemory = VK_FALSE;
+ 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_SHADER_DRAW_PARAMETERS_FEATURES: {
+ VkPhysicalDeviceShaderDrawParametersFeatures *features = (void *)ext;
features->shaderDrawParameters = true;
break;
}
- case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VARIABLE_POINTER_FEATURES: {
- VkPhysicalDeviceVariablePointerFeatures *features = (void *)ext;
+ 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_TRANSFORM_FEEDBACK_FEATURES_EXT: {
+ VkPhysicalDeviceTransformFeedbackFeaturesEXT *features =
+ (VkPhysicalDeviceTransformFeedbackFeaturesEXT *)ext;
+ features->transformFeedback = true;
+ features->geometryStreams = true;
+ break;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_VERTEX_ATTRIBUTE_DIVISOR_FEATURES_EXT: {
VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *features =
(VkPhysicalDeviceVertexAttributeDivisorFeaturesEXT *)ext;
- features->vertexAttributeInstanceRateDivisor = VK_TRUE;
- features->vertexAttributeInstanceRateZeroDivisor = VK_TRUE;
+ 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;
}
const uint32_t max_raw_buffer_sz = devinfo->gen >= 7 ?
(1ul << 30) : (1ul << 27);
+ const uint32_t max_ssbos = pdevice->has_a64_buffer_access ? UINT16_MAX : 64;
const uint32_t max_samplers = (devinfo->gen >= 8 || devinfo->is_haswell) ?
128 : 16;
- const uint32_t max_images = devinfo->gen < 9 ? MAX_GEN8_IMAGES : 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);
.maxBoundDescriptorSets = MAX_SETS,
.maxPerStageDescriptorSamplers = max_samplers,
.maxPerStageDescriptorUniformBuffers = 64,
- .maxPerStageDescriptorStorageBuffers = 64,
+ .maxPerStageDescriptorStorageBuffers = max_ssbos,
.maxPerStageDescriptorSampledImages = max_samplers,
- .maxPerStageDescriptorStorageImages = max_images,
+ .maxPerStageDescriptorStorageImages = MAX_IMAGES,
.maxPerStageDescriptorInputAttachments = 64,
- .maxPerStageResources = 250,
+ .maxPerStageResources = max_per_stage,
.maxDescriptorSetSamplers = 6 * max_samplers, /* number of stages * maxPerStageDescriptorSamplers */
.maxDescriptorSetUniformBuffers = 6 * 64, /* 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 * max_images, /* number of stages * maxPerStageDescriptorStorageImages */
+ .maxDescriptorSetStorageImages = 6 * MAX_IMAGES, /* number of stages * maxPerStageDescriptorStorageImages */
.maxDescriptorSetInputAttachments = 256,
.maxVertexInputAttributes = MAX_VBS,
.maxVertexInputBindings = MAX_VBS,
.maxFragmentCombinedOutputResources = 8,
.maxComputeSharedMemorySize = 32768,
.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,
VK_RESOLVE_MODE_MAX_BIT_KHR;
}
- props->independentResolveNone = VK_TRUE;
- props->independentResolve = VK_TRUE;
+ props->independentResolveNone = true;
+ props->independentResolve = true;
break;
}
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;
+ }
+
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_ID_PROPERTIES: {
VkPhysicalDeviceIDProperties *id_props =
(VkPhysicalDeviceIDProperties *)ext;
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;
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_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;
}
state = anv_state_pool_alloc(pool, size, align);
memcpy(state.map, p, size);
- anv_state_flush(pool->block_pool.device, state);
-
return state;
}
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);
+ 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;
}
/* 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;
anv_gem_destroy_context(device, device->context_id);
+ gen_batch_decode_ctx_finish(&device->decoder_ctx);
+
close(device->fd);
vk_free(&device->alloc, device);
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->map = NULL;
mem->map_size = 0;
mem->ahw = NULL;
+ mem->host_ptr = NULL;
uint64_t bo_flags = 0;
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;
+
+ mem->host_ptr = host_ptr_info->pHostPointer;
+ goto success;
+ }
+
/* Regular allocate (not importing memory). */
if (export_info && export_info->handleTypes)
}
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);
return VK_SUCCESS;
}
}
+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,
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);
anv_bo_cache_release(device, &device->bo_cache, mem->bo);
-#ifdef ANDROID
+#if defined(ANDROID) && ANDROID_API_LEVEL >= 26
if (mem->ahw)
AHardwareBuffer_release(mem->ahw);
#endif
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;
}
*
* 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;
}
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,
.mocs = device->default_mocs,
.size_B = range,
.format = format,
+ .swizzle = ISL_SWIZZLE_IDENTITY,
.stride_B = stride);
-
- anv_state_flush(device, state);
}
void anv_DestroySampler(