blob: 0c23d7d448ed48ae6d6d6ae0cd8604b05ed035f3 [file] [log] [blame]
// Copyright 2020 The Fuchsia Authors. All rights reserved.
// Use of this source code is governed by a BSD-style license that can be
// found in the LICENSE file.
#include <fuchsia/sysmem/cpp/fidl.h>
#include <lib/fdio/directory.h>
#include <vk_dispatch_table_helper.h>
#include <vk_layer_data.h>
#include <vk_layer_extension_utils.h>
#include <vk_layer_utils_minimal.h>
#include <unordered_map>
#include <vector>
#include <fbl/auto_call.h>
#include <sdk/lib/syslog/cpp/macros.h>
#include <vulkan/vk_layer.h>
#define VK_LAYER_API_VERSION VK_MAKE_VERSION(1, 1, VK_HEADER_VERSION)
namespace compact_image {
constexpr VkExtensionProperties device_extensions[] = {{
.extensionName = "VK_FUCHSIA_compact_image",
.specVersion = 1,
}};
constexpr VkLayerProperties compact_image_layer = {
"VK_LAYER_FUCHSIA_compact_image",
VK_LAYER_API_VERSION,
1,
"Compact Image",
};
//
// Shaders
//
#include "pack.spv.h"
#include "unpack.spv.h"
// Push constant block used by all shaders.
struct PushConstantBlock {
VkDeviceAddress image_address;
VkDeviceAddress scratch_address;
VkDeviceAddress aux_address;
uint32_t body_offset;
uint32_t block_count;
};
//
// AFBC constants
//
constexpr uint32_t kAfbcBodyAlignment = 4096u;
constexpr uint32_t kAfbcWidthAlignment = 128u;
constexpr uint32_t kAfbcHeightAlignment = 128u;
constexpr uint32_t kAfbcTilePixelWidth = 16u;
constexpr uint32_t kAfbcTilePixelHeight = 16u;
constexpr uint32_t kTileBytesPerPixel = 4u;
constexpr uint32_t kTileNumPixels = kAfbcTilePixelWidth * kAfbcTilePixelHeight;
constexpr uint32_t kTileNumBytes = kTileNumPixels * kTileBytesPerPixel;
constexpr uint32_t kAfbcBytesPerTileHeader = 16u;
constexpr uint32_t kAfbcSuperblockTileWidth = 8u;
constexpr uint32_t kAfbcSuperblockTileHeight = 8u;
constexpr uint32_t kAfbcSuperblockPixelWidth = kAfbcSuperblockTileWidth * kAfbcTilePixelWidth;
constexpr uint32_t kAfbcSuperblockPixelHeight = kAfbcSuperblockTileHeight * kAfbcTilePixelHeight;
constexpr uint32_t kAfbcSuperblockTileCount = kAfbcSuperblockTileWidth * kAfbcSuperblockTileHeight;
constexpr uint32_t kAfbcBytesPerSuperblockHeader =
kAfbcSuperblockTileCount * kAfbcBytesPerTileHeader;
//
// Vulkan utility functions
//
struct vk_struct_common {
VkStructureType sType;
struct vk_struct_common* pNext;
};
#define vk_foreach_struct(__iter, __start) \
for (struct vk_struct_common* __iter = (struct vk_struct_common*)(__start); __iter; \
__iter = __iter->pNext)
#define vk_foreach_struct_const(__iter, __start) \
for (const struct vk_struct_common* __iter = (const struct vk_struct_common*)(__start); __iter; \
__iter = __iter->pNext)
inline void* __vk_find_struct(void* start, VkStructureType sType) {
vk_foreach_struct(s, start) {
if (s->sType == sType)
return s;
}
return nullptr;
}
#define vk_find_struct(__start, __sType) __vk_find_struct((__start), VK_STRUCTURE_TYPE_##__sType)
#define vk_find_struct_const(__start, __sType) \
(const void*)__vk_find_struct((void*)(__start), VK_STRUCTURE_TYPE_##__sType)
//
// AFBC image compactor
//
// This class implements packing and unpacking of AFBC images
// when transitioning to/from image layouts that support packed
// mode.
//
// Compute shaders are used to pack/unpack images and dedicated
// memory allocations are required to control the exact tiling
// format used for images.
//
class ImageCompactor {
public:
ImageCompactor(VkDevice device, VkLayerDispatchTable* dispatch)
: device_(device), dispatch_(dispatch) {}
// Returns true if vendor ID and device ID are supported.
static bool IsSupportedGPU(uint32_t vendor_id, uint32_t device_id) {
switch (vendor_id) {
case 0x13b5: // ARM
switch (device_id) {
case 0x70930000: // BIFROST4
case 0x72120000: // BIFROST8
return true;
default:
break;
}
default:
break;
}
return false;
}
// Returns compact image format properties if supported.
static VkResult GetImageFormatProperties(const VkPhysicalDeviceProperties& gpu_properties,
VkFormat format, VkImageType type, VkImageTiling tiling,
VkImageUsageFlags usage, VkImageCreateFlags flags,
VkImageFormatProperties* pImageFormatProperties) {
if (!IsSupportedGPU(gpu_properties.vendorID, gpu_properties.deviceID)) {
return VK_ERROR_FORMAT_NOT_SUPPORTED;
}
if (!IsSupportedFormat(format) || !IsSupportedUsage(usage)) {
return VK_ERROR_FORMAT_NOT_SUPPORTED;
}
// 2D image type is supported.
if (type != VK_IMAGE_TYPE_2D) {
return VK_ERROR_FORMAT_NOT_SUPPORTED;
}
// Compaction is only supported for optimal tiling images.
if (tiling != VK_IMAGE_TILING_OPTIMAL) {
return VK_ERROR_FORMAT_NOT_SUPPORTED;
}
// Mutable formats are not supported.
if (flags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT) {
return VK_ERROR_FORMAT_NOT_SUPPORTED;
}
// Return compact image properties.
*pImageFormatProperties = VkImageFormatProperties{
.maxExtent = VkExtent3D{.width = 8192, .height = 8192, .depth = 1},
.maxMipLevels = 1,
.maxArrayLayers = 1,
.sampleCounts = 1,
.maxResourceSize = 0x80000000,
};
return VK_SUCCESS;
}
// Initialize image compactor. Connects to sysmem and creates
// compute pipelines needed for packing and unpacking.
VkResult Init(PFN_vkGetDeviceProcAddr fpGetDeviceProcAddr,
const VkAllocationCallbacks* pAllocator) {
// Connect to sysmem allocator service.
zx_status_t status = fdio_service_connect(
"/svc/fuchsia.sysmem.Allocator", sysmem_allocator_.NewRequest().TakeChannel().release());
if (status != ZX_OK) {
fprintf(stderr, "Couldn't connect to sysmem service\n");
return VK_ERROR_INITIALIZATION_FAILED;
}
auto cleanup = fbl::MakeAutoCall([this, pAllocator] { Cleanup(pAllocator); });
// Create pipeline layout that will be used by all shaders.
VkResult result =
CreatePipelineLayout(sizeof(PushConstantBlock), pAllocator, &pipeline_layout_);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
// Create compute pipeline used to pack AFBC images.
result = CreateComputePipeline(pack_comp_spv, sizeof(pack_comp_spv), pipeline_layout_,
pAllocator, &pack_pipeline_);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
// Create compute pipeline used to unpack AFBC images.
result = CreateComputePipeline(unpack_comp_spv, sizeof(unpack_comp_spv), pipeline_layout_,
pAllocator, &unpack_pipeline_);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
result = CreateBuffer(kTileNumBytes * kAfbcSuperblockTileCount, nullptr, pAllocator,
&scratch_.buffer);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
result = AllocateAndBindBufferMemory(scratch_.buffer, nullptr, pAllocator, &scratch_.memory,
&scratch_.device_address);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
cleanup.cancel();
return VK_SUCCESS;
}
void Cleanup(const VkAllocationCallbacks* pAllocator) {
if (scratch_.buffer != VK_NULL_HANDLE) {
dispatch_->DestroyBuffer(device_, scratch_.buffer, pAllocator);
}
if (scratch_.memory != VK_NULL_HANDLE) {
dispatch_->FreeMemory(device_, scratch_.memory, pAllocator);
}
if (pack_pipeline_ != VK_NULL_HANDLE) {
dispatch_->DestroyPipeline(device_, pack_pipeline_, pAllocator);
}
if (unpack_pipeline_ != VK_NULL_HANDLE) {
dispatch_->DestroyPipeline(device_, unpack_pipeline_, pAllocator);
}
if (pipeline_layout_ != VK_NULL_HANDLE) {
dispatch_->DestroyPipelineLayout(device_, pipeline_layout_, pAllocator);
}
}
VkResult CreateImage(const VkImageCreateInfo* pCreateInfo,
const VkAllocationCallbacks* pAllocator, VkImage* pImage) {
// Early out if this is a regular image.
if (!(pCreateInfo->flags & VK_IMAGE_CREATE_COMPACT_BIT_FUCHSIA)) {
return dispatch_->CreateImage(device_, pCreateInfo, pAllocator, pImage);
}
FX_CHECK(IsSupportedFormat(pCreateInfo->format));
FX_CHECK(IsSupportedUsage(pCreateInfo->usage));
FX_CHECK(pCreateInfo->imageType == VK_IMAGE_TYPE_2D);
FX_CHECK(pCreateInfo->tiling == VK_IMAGE_TILING_OPTIMAL);
FX_CHECK(!(pCreateInfo->flags & VK_IMAGE_CREATE_MUTABLE_FORMAT_BIT));
FX_CHECK(pCreateInfo->mipLevels == 1);
FX_CHECK(pCreateInfo->arrayLayers == 1);
FX_CHECK(pCreateInfo->samples == VK_SAMPLE_COUNT_1_BIT);
// Calculate superblock dimensions for image.
uint32_t width_in_superblocks =
RoundUp(pCreateInfo->extent.width, kAfbcWidthAlignment) / kAfbcSuperblockPixelWidth;
uint32_t height_in_superblocks =
RoundUp(pCreateInfo->extent.height, kAfbcHeightAlignment) / kAfbcSuperblockPixelHeight;
uint32_t num_superblocks = width_in_superblocks * height_in_superblocks;
uint32_t body_offset =
RoundUp(num_superblocks * kAfbcBytesPerSuperblockHeader, kAfbcBodyAlignment);
// Create single buffer collection for image.
fuchsia::sysmem::BufferCollectionTokenSyncPtr local_token;
zx_status_t status = sysmem_allocator_->AllocateSharedCollection(local_token.NewRequest());
if (status != ZX_OK) {
return VK_ERROR_INITIALIZATION_FAILED;
}
fuchsia::sysmem::BufferCollectionTokenSyncPtr vulkan_image_token;
status = local_token->Duplicate(std::numeric_limits<uint32_t>::max(),
vulkan_image_token.NewRequest());
if (status != ZX_OK) {
return VK_ERROR_INITIALIZATION_FAILED;
}
fuchsia::sysmem::BufferCollectionTokenSyncPtr vulkan_buffer_token;
status = local_token->Duplicate(std::numeric_limits<uint32_t>::max(),
vulkan_buffer_token.NewRequest());
if (status != ZX_OK) {
return VK_ERROR_INITIALIZATION_FAILED;
}
status = local_token->Sync();
if (status != ZX_OK) {
return VK_ERROR_INITIALIZATION_FAILED;
}
VkBufferCollectionCreateInfoFUCHSIA image_collection_create_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_COLLECTION_CREATE_INFO_FUCHSIA,
.pNext = nullptr,
.collectionToken = vulkan_image_token.Unbind().TakeChannel().release(),
};
VkBufferCollectionFUCHSIA collection;
VkResult result = dispatch_->CreateBufferCollectionFUCHSIA(
device_, &image_collection_create_info, nullptr, &collection);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
auto cleanup_collection = fbl::MakeAutoCall([this, collection, pAllocator] {
dispatch_->DestroyBufferCollectionFUCHSIA(device_, collection, pAllocator);
});
VkSysmemColorSpaceFUCHSIA color_space = {
.colorSpace = static_cast<uint32_t>(fuchsia::sysmem::ColorSpaceType::SRGB),
};
VkImageFormatConstraintsInfoFUCHSIA image_format_constraints_info = {
.sType = VK_STRUCTURE_TYPE_IMAGE_CONSTRAINTS_INFO_FUCHSIA,
.pNext = nullptr,
.requiredFormatFeatures = 0,
.flags = 0,
.sysmemFormat = static_cast<uint32_t>(fuchsia::sysmem::PixelFormatType::R8G8B8A8),
.colorSpaceCount = 1,
.pColorSpaces = &color_space,
};
VkImageConstraintsInfoFUCHSIA image_constraints_info = {
.sType = VK_STRUCTURE_TYPE_IMAGE_CONSTRAINTS_INFO_FUCHSIA,
.pNext = nullptr,
.createInfoCount = 1,
.pCreateInfos = pCreateInfo,
.pFormatConstraints = &image_format_constraints_info,
.minBufferCount = 1,
.maxBufferCount = 1,
.minBufferCountForDedicatedSlack = 0,
.minBufferCountForSharedSlack = 0,
};
result = dispatch_->SetBufferCollectionImageConstraintsFUCHSIA(device_, collection,
&image_constraints_info);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
VkBufferCollectionCreateInfoFUCHSIA buffer_collection_create_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_COLLECTION_CREATE_INFO_FUCHSIA,
.pNext = nullptr,
.collectionToken = vulkan_buffer_token.Unbind().TakeChannel().release(),
};
VkBufferCollectionFUCHSIA collection_for_buffer;
result = dispatch_->CreateBufferCollectionFUCHSIA(device_, &buffer_collection_create_info,
nullptr, &collection_for_buffer);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
auto cleanup_collection_for_buffer =
fbl::MakeAutoCall([this, collection_for_buffer, pAllocator] {
dispatch_->DestroyBufferCollectionFUCHSIA(device_, collection_for_buffer, pAllocator);
});
VkBufferCreateInfo buffer_create_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = body_offset + num_superblocks * kAfbcSuperblockTileCount * kTileNumBytes,
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
};
VkBufferConstraintsInfoFUCHSIA buffer_constraints_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_CONSTRAINTS_INFO_FUCHSIA,
.pNext = nullptr,
.pBufferCreateInfo = &buffer_create_info,
.requiredFormatFeatures = 0,
.minCount = 1,
};
result = dispatch_->SetBufferCollectionBufferConstraintsFUCHSIA(device_, collection_for_buffer,
&buffer_constraints_info);
if (result != VK_SUCCESS) {
return VK_ERROR_INITIALIZATION_FAILED;
}
fuchsia::sysmem::BufferCollectionSyncPtr buffer_collection;
status = sysmem_allocator_->BindSharedCollection(std::move(local_token),
buffer_collection.NewRequest());
if (status != ZX_OK) {
return VK_ERROR_INITIALIZATION_FAILED;
}
fuchsia::sysmem::BufferCollectionConstraints constraints;
constraints.min_buffer_count = 1;
constraints.usage.vulkan = 0;
if (pCreateInfo->usage & VK_IMAGE_USAGE_TRANSFER_SRC_BIT) {
constraints.usage.vulkan |= fuchsia::sysmem::VULKAN_IMAGE_USAGE_TRANSFER_SRC;
}
if (pCreateInfo->usage & VK_IMAGE_USAGE_TRANSFER_DST_BIT) {
constraints.usage.vulkan |= fuchsia::sysmem::VULKAN_IMAGE_USAGE_TRANSFER_DST;
}
if (pCreateInfo->usage & VK_IMAGE_USAGE_SAMPLED_BIT) {
constraints.usage.vulkan |= fuchsia::sysmem::VULKAN_IMAGE_USAGE_SAMPLED;
}
constraints.has_buffer_memory_constraints = true;
constraints.buffer_memory_constraints.ram_domain_supported = true;
constraints.buffer_memory_constraints.cpu_domain_supported = true;
constraints.buffer_memory_constraints.inaccessible_domain_supported = true;
constraints.image_format_constraints_count = 1;
fuchsia::sysmem::ImageFormatConstraints& image_constraints =
constraints.image_format_constraints[0];
image_constraints = fuchsia::sysmem::ImageFormatConstraints();
image_constraints.min_coded_width = pCreateInfo->extent.width;
image_constraints.min_coded_height = pCreateInfo->extent.height;
image_constraints.max_coded_width = pCreateInfo->extent.width;
image_constraints.max_coded_height = pCreateInfo->extent.height;
image_constraints.min_bytes_per_row = 0;
image_constraints.pixel_format.type = fuchsia::sysmem::PixelFormatType::R8G8B8A8;
image_constraints.color_spaces_count = 1;
image_constraints.color_space[0].type = fuchsia::sysmem::ColorSpaceType::SRGB;
image_constraints.pixel_format.has_format_modifier = true;
image_constraints.pixel_format.format_modifier.value =
fuchsia::sysmem::FORMAT_MODIFIER_ARM_AFBC_16X16_YUV_TILED_HEADER;
status = buffer_collection->SetConstraints(true, constraints);
FX_CHECK(status == ZX_OK);
zx_status_t allocation_status;
fuchsia::sysmem::BufferCollectionInfo_2 buffer_collection_info = {};
status =
buffer_collection->WaitForBuffersAllocated(&allocation_status, &buffer_collection_info);
FX_CHECK(status == ZX_OK);
FX_CHECK(allocation_status == ZX_OK);
FX_CHECK(buffer_collection_info.settings.image_format_constraints.pixel_format.type ==
image_constraints.pixel_format.type);
VkBufferCollectionBufferCreateInfoFUCHSIA collection_buffer_create_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_COLLECTION_BUFFER_CREATE_INFO_FUCHSIA,
.pNext = nullptr,
.collection = collection_for_buffer,
.index = 0,
};
buffer_create_info.pNext = &collection_buffer_create_info;
VkBuffer buffer;
result = dispatch_->CreateBuffer(device_, &buffer_create_info, pAllocator, &buffer);
if (result != VK_SUCCESS) {
return result;
}
auto cleanup_buffer = fbl::MakeAutoCall(
[this, buffer, pAllocator] { dispatch_->DestroyBuffer(device_, buffer, pAllocator); });
// Create memory allocation by importing the sysmem collection.
VkImportMemoryBufferCollectionFUCHSIA import_info = {
.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_BUFFER_COLLECTION_FUCHSIA,
.pNext = nullptr,
.collection = collection_for_buffer,
.index = 0,
};
VkDeviceMemory buffer_memory;
VkDeviceAddress buffer_device_address;
result = AllocateAndBindBufferMemory(buffer, &import_info, pAllocator, &buffer_memory,
&buffer_device_address);
if (result != VK_SUCCESS) {
return result;
}
auto cleanup_buffer_memory = fbl::MakeAutoCall([this, buffer_memory, pAllocator] {
dispatch_->FreeMemory(device_, buffer_memory, pAllocator);
});
// Create 4 byte auxiliary buffer.
VkBuffer aux_buffer;
result = CreateBuffer(4, nullptr, pAllocator, &aux_buffer);
if (result != VK_SUCCESS) {
return result;
}
auto cleanup_aux_buffer = fbl::MakeAutoCall([this, aux_buffer, pAllocator] {
dispatch_->DestroyBuffer(device_, aux_buffer, pAllocator);
});
VkDeviceMemory aux_buffer_memory;
VkDeviceAddress aux_buffer_device_address;
result = AllocateAndBindBufferMemory(aux_buffer, nullptr, pAllocator, &aux_buffer_memory,
&aux_buffer_device_address);
if (result != VK_SUCCESS) {
return result;
}
auto cleanup_aux_buffer_memory = fbl::MakeAutoCall([this, aux_buffer_memory, pAllocator] {
dispatch_->FreeMemory(device_, aux_buffer_memory, pAllocator);
});
// Create image after successfully initializing extra state.
VkBufferCollectionImageCreateInfoFUCHSIA collection_image_create_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_COLLECTION_IMAGE_CREATE_INFO_FUCHSIA,
.pNext = nullptr,
.collection = collection,
.index = 0,
};
VkImageCreateInfo image_create_info = {
.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
.pNext = &collection_image_create_info,
.flags = pCreateInfo->flags & ~VK_IMAGE_CREATE_COMPACT_BIT_FUCHSIA,
.imageType = pCreateInfo->imageType,
.format = pCreateInfo->format,
.extent = pCreateInfo->extent,
.mipLevels = pCreateInfo->mipLevels,
.arrayLayers = pCreateInfo->arrayLayers,
.samples = pCreateInfo->samples,
.tiling = pCreateInfo->tiling,
.usage = pCreateInfo->usage,
.sharingMode = pCreateInfo->sharingMode,
.initialLayout = pCreateInfo->initialLayout,
};
result = dispatch_->CreateImage(device_, &image_create_info, pAllocator, pImage);
if (result != VK_SUCCESS) {
return result;
}
// Buffer collection can be closed as vulkan handle is enough.
buffer_collection->Close();
// Reset cleanup handlers.
cleanup_collection.cancel();
cleanup_buffer.cancel();
cleanup_buffer_memory.cancel();
cleanup_aux_buffer.cancel();
cleanup_aux_buffer_memory.cancel();
compact_images_.insert(
{*pImage, CompactImage{
.collection = collection,
.buffer = Buffer{.buffer = buffer,
.memory = buffer_memory,
.device_address = buffer_device_address},
.aux = Buffer{.buffer = aux_buffer,
.memory = aux_buffer_memory,
.device_address = aux_buffer_device_address},
.allocation_size = buffer_collection_info.settings.buffer_settings.size_bytes,
.width_in_superblocks = width_in_superblocks,
.height_in_superblocks = height_in_superblocks,
.compact_memory_bound = false,
}});
return result;
}
void DestroyImage(VkImage image, const VkAllocationCallbacks* pAllocator) {
auto it = compact_images_.find(image);
if (it != compact_images_.end()) {
dispatch_->DestroyBuffer(device_, it->second.buffer.buffer, pAllocator);
dispatch_->FreeMemory(device_, it->second.buffer.memory, pAllocator);
dispatch_->DestroyBuffer(device_, it->second.aux.buffer, pAllocator);
dispatch_->FreeMemory(device_, it->second.aux.memory, pAllocator);
dispatch_->DestroyBufferCollectionFUCHSIA(device_, it->second.collection, pAllocator);
compact_images_.erase(it);
}
dispatch_->DestroyImage(device_, image, pAllocator);
}
void GetImageMemoryRequirements2(const VkImageMemoryRequirementsInfo2* pInfo,
VkMemoryRequirements2* pMemoryRequirements) {
dispatch_->GetImageMemoryRequirements2(device_, pInfo, pMemoryRequirements);
auto it = compact_images_.find(pInfo->image);
if (it != compact_images_.end()) {
VkBufferCollectionPropertiesFUCHSIA properties;
dispatch_->GetBufferCollectionPropertiesFUCHSIA(device_, it->second.collection, &properties);
pMemoryRequirements->memoryRequirements.memoryTypeBits &= properties.memoryTypeBits;
auto dedicated_requirements = static_cast<VkMemoryDedicatedRequirements*>(
vk_find_struct(pMemoryRequirements, MEMORY_DEDICATED_REQUIREMENTS));
// Add dedicated allocation preference as required for compact images.
if (dedicated_requirements) {
dedicated_requirements->prefersDedicatedAllocation = VK_TRUE;
}
}
}
VkResult AllocateMemory(const VkMemoryAllocateInfo* pAllocateInfo,
const VkAllocationCallbacks* pAllocator, VkDeviceMemory* pMemory) {
auto dedicated_alloc_info = static_cast<const VkMemoryDedicatedAllocateInfo*>(
vk_find_struct_const(pAllocateInfo, MEMORY_DEDICATED_ALLOCATE_INFO));
// Early out if not a dedicated allocation.
if (!dedicated_alloc_info) {
return dispatch_->AllocateMemory(device_, pAllocateInfo, pAllocator, pMemory);
}
// Early out if dedicated image is not compact.
auto it = compact_images_.find(dedicated_alloc_info->image);
if (it == compact_images_.end()) {
return dispatch_->AllocateMemory(device_, pAllocateInfo, pAllocator, pMemory);
}
// Create memory allocation by importing the sysmem collection.
VkImportMemoryBufferCollectionFUCHSIA import_info = {
.sType = VK_STRUCTURE_TYPE_IMPORT_MEMORY_BUFFER_COLLECTION_FUCHSIA,
.pNext = nullptr,
.collection = it->second.collection,
.index = 0,
};
VkMemoryAllocateInfo allocation_info = {
.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
.pNext = &import_info,
.allocationSize = pAllocateInfo->allocationSize,
.memoryTypeIndex = pAllocateInfo->memoryTypeIndex,
};
VkResult result = dispatch_->AllocateMemory(device_, &allocation_info, pAllocator, pMemory);
if (result != VK_SUCCESS) {
return result;
}
dedicated_image_memory_.insert({*pMemory, dedicated_alloc_info->image});
return VK_SUCCESS;
}
void FreeMemory(VkDeviceMemory memory, const VkAllocationCallbacks* pAllocator) {
dispatch_->FreeMemory(device_, memory, pAllocator);
dedicated_image_memory_.erase(memory);
}
VkResult BindImageMemory(VkImage image, VkDeviceMemory memory, VkDeviceSize memoryOffset) {
VkResult result = dispatch_->BindImageMemory(device_, image, memory, memoryOffset);
if (result != VK_SUCCESS) {
return result;
}
// Activate packing for compact image if bound to buffer collection
// backed memory.
auto it = compact_images_.find(image);
if (it != compact_images_.end()) {
it->second.compact_memory_bound = IsDedicatedImageMemory(memory, image);
}
return VK_SUCCESS;
}
VkResult BindImageMemory2(uint32_t bindInfoCount, const VkBindImageMemoryInfo* pBindInfos) {
VkResult result = dispatch_->BindImageMemory2(device_, bindInfoCount, pBindInfos);
if (result != VK_SUCCESS) {
return result;
}
for (uint32_t i = 0; i < bindInfoCount; ++i) {
// Activate packing for compact image if bound to buffer collection
// backed memory.
auto it = compact_images_.find(pBindInfos[i].image);
if (it != compact_images_.end()) {
it->second.compact_memory_bound =
IsDedicatedImageMemory(pBindInfos[i].memory, pBindInfos[i].image);
}
}
return VK_SUCCESS;
}
VkResult BeginCommandBuffer(VkCommandBuffer commandBuffer,
const VkCommandBufferBeginInfo* pBeginInfo) {
command_buffer_state_[commandBuffer] = CommandBufferState();
return dispatch_->BeginCommandBuffer(commandBuffer, pBeginInfo);
}
VkResult EndCommandBuffer(VkCommandBuffer commandBuffer) {
command_buffer_state_.erase(commandBuffer);
return dispatch_->EndCommandBuffer(commandBuffer);
}
void CmdBindPipeline(VkCommandBuffer commandBuffer, VkPipelineBindPoint pipelineBindPoint,
VkPipeline pipeline) {
if (pipelineBindPoint == VK_PIPELINE_BIND_POINT_COMPUTE) {
command_buffer_state_[commandBuffer].pipeline = pipeline;
}
dispatch_->CmdBindPipeline(commandBuffer, pipelineBindPoint, pipeline);
}
void CmdPushConstants(VkCommandBuffer commandBuffer, VkPipelineLayout layout,
VkShaderStageFlags stageFlags, uint32_t offset, uint32_t size,
const void* pValues) {
auto& state = command_buffer_state_[commandBuffer];
state.pipeline_layout = layout;
state.stage_flags |= stageFlags;
state.push_constants.resize(offset + size, 0);
memcpy(state.push_constants.data() + offset, pValues, size);
dispatch_->CmdPushConstants(commandBuffer, layout, stageFlags, offset, size, pValues);
}
void CmdPipelineBarrier(VkCommandBuffer commandBuffer, VkPipelineStageFlags srcStageMask,
VkPipelineStageFlags dstStageMask, VkDependencyFlags dependencyFlags,
uint32_t memoryBarrierCount, const VkMemoryBarrier* pMemoryBarriers,
uint32_t bufferMemoryBarrierCount,
const VkBufferMemoryBarrier* pBufferMemoryBarriers,
uint32_t imageMemoryBarrierCount,
const VkImageMemoryBarrier* pImageMemoryBarriers) {
std::vector<VkImageMemoryBarrier> pack_image_memory_barriers;
std::vector<VkImageMemoryBarrier> unpack_image_memory_barriers;
std::vector<VkImageMemoryBarrier> other_image_memory_barriers;
// Iterate over image barriers and extract barriers that required packing.
for (uint32_t i = 0; i < imageMemoryBarrierCount; i++) {
bool oldLayoutIsCompact = IsCompactLayout(pImageMemoryBarriers[i].oldLayout);
bool newLayoutIsCompact = IsCompactLayout(pImageMemoryBarriers[i].newLayout);
if (oldLayoutIsCompact != newLayoutIsCompact &&
IsCompactImage(pImageMemoryBarriers[i].image)) {
if (newLayoutIsCompact) {
pack_image_memory_barriers.push_back(pImageMemoryBarriers[i]);
} else {
unpack_image_memory_barriers.push_back(pImageMemoryBarriers[i]);
}
} else {
other_image_memory_barriers.push_back(pImageMemoryBarriers[i]);
}
}
// Forward barriers that don't require packing.
dispatch_->CmdPipelineBarrier(commandBuffer, srcStageMask, dstStageMask, dependencyFlags,
memoryBarrierCount, pMemoryBarriers, bufferMemoryBarrierCount,
pBufferMemoryBarriers,
static_cast<uint32_t>(other_image_memory_barriers.size()),
other_image_memory_barriers.data());
// Check if we have at least one image barrier that require packing.
if (!pack_image_memory_barriers.empty() || !pack_image_memory_barriers.empty()) {
auto& state = command_buffer_state_[commandBuffer];
// Emit commands for image barriers that use pack shader.
for (auto& barrier : pack_image_memory_barriers) {
PackingPipelineBarrier(commandBuffer, srcStageMask, dstStageMask, dependencyFlags, barrier,
pack_pipeline_);
}
// Emit commands for image barriers that use unpack shader.
for (auto& barrier : unpack_image_memory_barriers) {
PackingPipelineBarrier(commandBuffer, srcStageMask, dstStageMask, dependencyFlags, barrier,
unpack_pipeline_);
}
// Restore compute bind point if needed.
if (state.pipeline != VK_NULL_HANDLE) {
dispatch_->CmdBindPipeline(commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, state.pipeline);
}
// Restore push constants if needed.
if (state.pipeline_layout != VK_NULL_HANDLE) {
dispatch_->CmdPushConstants(commandBuffer, state.pipeline_layout, state.stage_flags, 0,
static_cast<uint32_t>(state.push_constants.size()),
state.push_constants.data());
}
}
}
void CmdWriteCompactImageMemorySize(VkCommandBuffer commandBuffer, VkImage image,
VkImageLayout imageLayout, VkBuffer buffer,
VkDeviceSize bufferOffset,
const VkImageSubresourceLayers* subresourceLayers) {
FX_CHECK(compact_images_.find(image) != compact_images_.end());
auto& compact_image = compact_images_[image];
// Compact image support is limited to single layer 2D images.
FX_CHECK(subresourceLayers->aspectMask == VK_IMAGE_ASPECT_COLOR_BIT);
FX_CHECK(subresourceLayers->mipLevel == 0);
FX_CHECK(subresourceLayers->baseArrayLayer == 0);
FX_CHECK(subresourceLayers->layerCount == 1);
if (IsCompactLayout(imageLayout) && compact_image.compact_memory_bound) {
VkBufferCopy region = {
.srcOffset = 0,
.dstOffset = bufferOffset,
.size = 4,
};
dispatch_->CmdCopyBuffer(commandBuffer, compact_image.aux.buffer, buffer, 1, &region);
} else {
dispatch_->CmdFillBuffer(commandBuffer, buffer, bufferOffset, 4,
static_cast<uint32_t>(compact_image.allocation_size));
}
}
private:
struct Buffer {
VkBuffer buffer = VK_NULL_HANDLE;
VkDeviceMemory memory = VK_NULL_HANDLE;
VkDeviceAddress device_address;
};
struct CompactImage {
VkBufferCollectionFUCHSIA collection = VK_NULL_HANDLE;
Buffer buffer;
Buffer aux;
VkDeviceSize allocation_size;
uint32_t width_in_superblocks;
uint32_t height_in_superblocks;
bool compact_memory_bound;
};
struct CommandBufferState {
VkPipeline pipeline = VK_NULL_HANDLE;
VkPipelineLayout pipeline_layout = VK_NULL_HANDLE;
VkShaderStageFlags stage_flags = 0;
std::vector<uint8_t> push_constants;
};
static bool IsSupportedUsage(VkImageUsageFlags usage) {
// TODO(reveman): Add VK_IMAGE_USAGE_COLOR_ATTACHMENT_BIT
// after implementing render pass support.
constexpr VkImageUsageFlags kSupportedUsage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT |
VK_IMAGE_USAGE_TRANSFER_DST_BIT |
VK_IMAGE_USAGE_SAMPLED_BIT;
return !(usage & ~kSupportedUsage);
}
static bool IsSupportedFormat(VkFormat format) {
constexpr VkFormat kSupportedFormats[] = {
VK_FORMAT_R8G8B8A8_UNORM,
VK_FORMAT_R8G8B8A8_SRGB,
};
return std::find(std::begin(kSupportedFormats), std::end(kSupportedFormats), format) !=
std::end(kSupportedFormats);
}
static bool IsCompactLayout(VkImageLayout layout) {
constexpr VkImageLayout compact_layouts[] = {
VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL,
VK_IMAGE_LAYOUT_SHADER_READ_ONLY_OPTIMAL,
};
return std::find(std::begin(compact_layouts), std::end(compact_layouts), layout) !=
std::end(compact_layouts);
}
static uint32_t RoundUp(uint32_t value, uint32_t multiple) {
return ((value + (multiple - 1)) / multiple) * multiple;
}
bool IsCompactImage(VkImage image) const {
auto it = compact_images_.find(image);
return it != compact_images_.end() && it->second.compact_memory_bound;
}
bool IsDedicatedImageMemory(VkDeviceMemory memory, VkImage image) const {
auto it = dedicated_image_memory_.find(memory);
if (it == dedicated_image_memory_.end()) {
return false;
}
return it->second == image;
}
VkResult CreatePipelineLayout(uint32_t push_constant_block_size,
const VkAllocationCallbacks* pAllocator, VkPipelineLayout* layout) {
VkPushConstantRange push_constant_range = {
.stageFlags = 0,
.offset = 0,
.size = push_constant_block_size,
};
VkPipelineLayoutCreateInfo pipeline_layout_info = {
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.setLayoutCount = 0,
.pSetLayouts = nullptr,
.pushConstantRangeCount = 1,
.pPushConstantRanges = &push_constant_range,
};
return dispatch_->CreatePipelineLayout(device_, &pipeline_layout_info, pAllocator, layout);
}
VkResult CreateComputePipeline(const uint32_t* spv, unsigned int spv_len,
VkPipelineLayout pipeline_layout,
const VkAllocationCallbacks* pAllocator, VkPipeline* pipeline) {
VkShaderModule module;
VkShaderModuleCreateInfo module_info = {
.sType = VK_STRUCTURE_TYPE_SHADER_MODULE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.codeSize = spv_len,
.pCode = spv,
};
VkResult result = dispatch_->CreateShaderModule(device_, &module_info, pAllocator, &module);
if (result != VK_SUCCESS) {
return result;
}
VkPipelineShaderStageCreateInfo stage_info = {
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.module = module,
.pName = "main",
.pSpecializationInfo = nullptr,
};
VkComputePipelineCreateInfo pipeline_info = {
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = stage_info,
.layout = pipeline_layout,
.basePipelineHandle = VK_NULL_HANDLE,
.basePipelineIndex = 0,
};
result = dispatch_->CreateComputePipelines(device_, VK_NULL_HANDLE, 1, &pipeline_info,
pAllocator, pipeline);
dispatch_->DestroyShaderModule(device_, module, pAllocator);
return result;
}
VkResult CreateBuffer(uint32_t size, const void* pNext, const VkAllocationCallbacks* pAllocator,
VkBuffer* pBuffer) {
VkBufferCreateInfo buffer_create_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = pNext,
.flags = 0,
.size = size,
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_SHADER_DEVICE_ADDRESS_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
};
return dispatch_->CreateBuffer(device_, &buffer_create_info, pAllocator, pBuffer);
}
VkResult AllocateAndBindBufferMemory(VkBuffer buffer, const void* pNext,
const VkAllocationCallbacks* pAllocator,
VkDeviceMemory* pMemory, VkDeviceAddress* pDeviceAddress) {
VkMemoryRequirements memory_requirements;
dispatch_->GetBufferMemoryRequirements(device_, buffer, &memory_requirements);
uint32_t memory_type_index = __builtin_ctz(memory_requirements.memoryTypeBits);
VkMemoryAllocateInfo allocate_info = {
.sType = VK_STRUCTURE_TYPE_MEMORY_ALLOCATE_INFO,
.pNext = pNext,
.allocationSize = memory_requirements.size,
.memoryTypeIndex = memory_type_index,
};
VkResult result = dispatch_->AllocateMemory(device_, &allocate_info, pAllocator, pMemory);
if (result != VK_SUCCESS) {
return result;
}
auto cleanup_memory = fbl::MakeAutoCall(
[this, pMemory, pAllocator] { dispatch_->FreeMemory(device_, *pMemory, pAllocator); });
result = dispatch_->BindBufferMemory(device_, buffer, *pMemory, 0);
if (result != VK_SUCCESS) {
return result;
}
VkBufferDeviceAddressInfo device_address_info = {
.sType = VK_STRUCTURE_TYPE_BUFFER_DEVICE_ADDRESS_INFO,
.pNext = nullptr,
.buffer = buffer,
};
*pDeviceAddress = dispatch_->GetBufferDeviceAddress(device_, &device_address_info);
cleanup_memory.cancel();
return VK_SUCCESS;
}
void PackingPipelineBarrier(VkCommandBuffer commandBuffer, VkPipelineStageFlags srcStageMask,
VkPipelineStageFlags dstStageMask, VkDependencyFlags dependencyFlags,
const VkImageMemoryBarrier& barrier, VkPipeline pipeline) {
auto& compact_image = compact_images_[barrier.image];
// Image barrier to ensure that image memory is available to
// compute shader.
VkImageMemoryBarrier pre_image_barrier = {
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = barrier.srcAccessMask,
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
.oldLayout = barrier.oldLayout,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.image = barrier.image,
.subresourceRange = barrier.subresourceRange,
};
dispatch_->CmdPipelineBarrier(commandBuffer, srcStageMask, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
dependencyFlags | VK_DEPENDENCY_BY_REGION_BIT, 0, nullptr, 0,
nullptr, 1, &pre_image_barrier);
uint32_t num_superblocks =
compact_image.width_in_superblocks * compact_image.height_in_superblocks;
uint32_t body_offset =
RoundUp(num_superblocks * kAfbcBytesPerSuperblockHeader, kAfbcBodyAlignment);
// Bind pipeline used for this layout transition.
dispatch_->CmdBindPipeline(commandBuffer, VK_PIPELINE_BIND_POINT_COMPUTE, pipeline);
// Update push constants.
PushConstantBlock push_constants = {
.image_address = compact_image.buffer.device_address,
.scratch_address = scratch_.device_address,
.aux_address = compact_image.aux.device_address,
.body_offset = body_offset,
.block_count = num_superblocks,
};
dispatch_->CmdPushConstants(commandBuffer, pipeline_layout_, VK_SHADER_STAGE_COMPUTE_BIT, 0,
sizeof(push_constants), &push_constants);
// Dispatch compute shader that performs the layout transition.
// TODO(reveman): Use multiple workgroups for improved performance.
dispatch_->CmdDispatch(commandBuffer, 1, 1, 1);
// Image and buffer barriers to ensure that memory written by
// compute shader is visible to destination stage.
VkImageMemoryBarrier post_image_barrier = {
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = barrier.dstAccessMask,
.oldLayout = VK_IMAGE_LAYOUT_GENERAL,
.newLayout = barrier.newLayout,
.image = barrier.image,
.subresourceRange = barrier.subresourceRange,
};
VkBufferMemoryBarrier post_buffer_barrier = {
.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.buffer = compact_image.aux.buffer,
.offset = 0,
.size = 4,
};
dispatch_->CmdPipelineBarrier(commandBuffer, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, dstStageMask,
dependencyFlags | VK_DEPENDENCY_BY_REGION_BIT, 0, nullptr, 1,
&post_buffer_barrier, 1, &post_image_barrier);
}
VkDevice device_;
VkLayerDispatchTable* dispatch_;
fuchsia::sysmem::AllocatorSyncPtr sysmem_allocator_;
Buffer scratch_;
VkPipelineLayout pipeline_layout_ = VK_NULL_HANDLE;
VkPipeline pack_pipeline_ = VK_NULL_HANDLE;
VkPipeline unpack_pipeline_ = VK_NULL_HANDLE;
std::unordered_map<VkImage, CompactImage> compact_images_;
std::unordered_map<VkDeviceMemory, VkImage> dedicated_image_memory_;
std::unordered_map<VkCommandBuffer, CommandBufferState> command_buffer_state_;
};
struct LayerData {
VkInstance instance;
std::unique_ptr<VkLayerDispatchTable> device_dispatch_table;
std::unique_ptr<VkLayerInstanceDispatchTable> instance_dispatch_table;
std::unique_ptr<ImageCompactor> compactor;
};
small_unordered_map<void*, LayerData*, 2> layer_data_map;
VKAPI_ATTR VkResult VKAPI_CALL CreateInstance(const VkInstanceCreateInfo* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkInstance* pInstance) {
VkLayerInstanceCreateInfo* chain_info = get_chain_info(pCreateInfo, VK_LAYER_LINK_INFO);
FX_CHECK(chain_info->u.pLayerInfo);
PFN_vkGetInstanceProcAddr fpGetInstanceProcAddr =
chain_info->u.pLayerInfo->pfnNextGetInstanceProcAddr;
PFN_vkCreateInstance fpCreateInstance =
reinterpret_cast<PFN_vkCreateInstance>(fpGetInstanceProcAddr(nullptr, "vkCreateInstance"));
if (!fpCreateInstance) {
return VK_ERROR_INITIALIZATION_FAILED;
}
// Advance the link info for the next element on the chain.
chain_info->u.pLayerInfo = chain_info->u.pLayerInfo->pNext;
VkResult result = fpCreateInstance(pCreateInfo, pAllocator, pInstance);
if (result != VK_SUCCESS) {
return result;
}
LayerData* instance_layer_data = GetLayerDataPtr(get_dispatch_key(*pInstance), layer_data_map);
instance_layer_data->instance = *pInstance;
instance_layer_data->instance_dispatch_table = std::make_unique<VkLayerInstanceDispatchTable>();
layer_init_instance_dispatch_table(*pInstance, instance_layer_data->instance_dispatch_table.get(),
fpGetInstanceProcAddr);
return result;
}
VKAPI_ATTR void VKAPI_CALL DestroyInstance(VkInstance instance,
const VkAllocationCallbacks* pAllocator) {
dispatch_key instance_key = get_dispatch_key(instance);
LayerData* instance_layer_data = GetLayerDataPtr(instance_key, layer_data_map);
instance_layer_data->instance_dispatch_table->DestroyInstance(instance, pAllocator);
// Remove from |layer_data_map| and free LayerData struct.
FreeLayerDataPtr(instance_key, layer_data_map);
}
VKAPI_ATTR VkResult VKAPI_CALL GetPhysicalDeviceImageFormatProperties(
VkPhysicalDevice gpu, VkFormat format, VkImageType type, VkImageTiling tiling,
VkImageUsageFlags usage, VkImageCreateFlags flags,
VkImageFormatProperties* pImageFormatProperties) {
void* gpu_key = get_dispatch_key(gpu);
LayerData* gpu_layer_data = GetLayerDataPtr(gpu_key, layer_data_map);
if (flags & VK_IMAGE_CREATE_COMPACT_BIT_FUCHSIA) {
VkPhysicalDeviceProperties gpu_properties;
gpu_layer_data->instance_dispatch_table->GetPhysicalDeviceProperties(gpu, &gpu_properties);
return ImageCompactor::GetImageFormatProperties(gpu_properties, format, type, tiling, usage,
flags, pImageFormatProperties);
}
return gpu_layer_data->instance_dispatch_table->GetPhysicalDeviceImageFormatProperties(
gpu, format, type, tiling, usage, flags, pImageFormatProperties);
}
VKAPI_ATTR VkResult VKAPI_CALL GetPhysicalDeviceImageFormatProperties2(
VkPhysicalDevice gpu, const VkPhysicalDeviceImageFormatInfo2* pImageFormatInfo,
VkImageFormatProperties2* pImageFormatProperties) {
void* gpu_key = get_dispatch_key(gpu);
LayerData* gpu_layer_data = GetLayerDataPtr(gpu_key, layer_data_map);
if (pImageFormatInfo->flags & VK_IMAGE_CREATE_COMPACT_BIT_FUCHSIA) {
VkPhysicalDeviceProperties gpu_properties;
gpu_layer_data->instance_dispatch_table->GetPhysicalDeviceProperties(gpu, &gpu_properties);
return ImageCompactor::GetImageFormatProperties(
gpu_properties, pImageFormatInfo->format, pImageFormatInfo->type, pImageFormatInfo->tiling,
pImageFormatInfo->usage, pImageFormatInfo->flags,
&pImageFormatProperties->imageFormatProperties);
}
return gpu_layer_data->instance_dispatch_table->GetPhysicalDeviceImageFormatProperties2(
gpu, pImageFormatInfo, pImageFormatProperties);
}
VKAPI_ATTR VkResult VKAPI_CALL CreateDevice(VkPhysicalDevice gpu,
const VkDeviceCreateInfo* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkDevice* pDevice) {
void* gpu_key = get_dispatch_key(gpu);
LayerData* gpu_layer_data = GetLayerDataPtr(gpu_key, layer_data_map);
bool khr_buffer_device_address = false;
bool fuchsia_buffer_collection_extension_available = false;
uint32_t device_extension_count;
VkResult result = gpu_layer_data->instance_dispatch_table->EnumerateDeviceExtensionProperties(
gpu, nullptr, &device_extension_count, nullptr);
if (result == VK_SUCCESS && device_extension_count > 0) {
std::vector<VkExtensionProperties> device_extensions(device_extension_count);
result = gpu_layer_data->instance_dispatch_table->EnumerateDeviceExtensionProperties(
gpu, nullptr, &device_extension_count, device_extensions.data());
if (result == VK_SUCCESS) {
for (uint32_t i = 0; i < device_extension_count; i++) {
if (!strcmp(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME,
device_extensions[i].extensionName)) {
khr_buffer_device_address = true;
}
if (!strcmp(VK_FUCHSIA_BUFFER_COLLECTION_EXTENSION_NAME,
device_extensions[i].extensionName)) {
fuchsia_buffer_collection_extension_available = true;
}
}
}
}
if (!khr_buffer_device_address) {
fprintf(stderr, "Device extension not available: %s\n",
VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME);
}
if (!fuchsia_buffer_collection_extension_available) {
fprintf(stderr, "Device extension not available: %s\n",
VK_FUCHSIA_BUFFER_COLLECTION_EXTENSION_NAME);
}
if (!khr_buffer_device_address || !fuchsia_buffer_collection_extension_available) {
return VK_ERROR_INITIALIZATION_FAILED;
}
VkPhysicalDeviceProperties gpu_properties;
gpu_layer_data->instance_dispatch_table->GetPhysicalDeviceProperties(gpu, &gpu_properties);
VkDeviceCreateInfo create_info = *pCreateInfo;
std::vector<const char*> enabled_extensions;
for (uint32_t i = 0; i < create_info.enabledExtensionCount; i++) {
enabled_extensions.push_back(create_info.ppEnabledExtensionNames[i]);
}
enabled_extensions.push_back(VK_KHR_BUFFER_DEVICE_ADDRESS_EXTENSION_NAME);
enabled_extensions.push_back(VK_FUCHSIA_BUFFER_COLLECTION_EXTENSION_NAME);
create_info.enabledExtensionCount = static_cast<uint32_t>(enabled_extensions.size());
create_info.ppEnabledExtensionNames = enabled_extensions.data();
VkLayerDeviceCreateInfo* chain_info = get_chain_info(pCreateInfo, VK_LAYER_LINK_INFO);
FX_CHECK(chain_info->u.pLayerInfo);
PFN_vkGetInstanceProcAddr fpGetInstanceProcAddr =
chain_info->u.pLayerInfo->pfnNextGetInstanceProcAddr;
PFN_vkGetDeviceProcAddr fpGetDeviceProcAddr = chain_info->u.pLayerInfo->pfnNextGetDeviceProcAddr;
PFN_vkCreateDevice fpCreateDevice = reinterpret_cast<PFN_vkCreateDevice>(
fpGetInstanceProcAddr(gpu_layer_data->instance, "vkCreateDevice"));
if (!fpCreateDevice) {
return VK_ERROR_INITIALIZATION_FAILED;
}
// Advance the link info for the next element on the chain.
chain_info->u.pLayerInfo = chain_info->u.pLayerInfo->pNext;
result = fpCreateDevice(gpu, &create_info, pAllocator, pDevice);
if (result != VK_SUCCESS) {
return result;
}
LayerData* device_layer_data = GetLayerDataPtr(get_dispatch_key(*pDevice), layer_data_map);
// Setup device dispatch table.
device_layer_data->device_dispatch_table = std::make_unique<VkLayerDispatchTable>();
device_layer_data->instance = gpu_layer_data->instance;
layer_init_device_dispatch_table(*pDevice, device_layer_data->device_dispatch_table.get(),
fpGetDeviceProcAddr);
// Create image compactor if GPU is supported.
if (ImageCompactor::IsSupportedGPU(gpu_properties.vendorID, gpu_properties.deviceID)) {
device_layer_data->compactor =
std::make_unique<ImageCompactor>(*pDevice, device_layer_data->device_dispatch_table.get());
return device_layer_data->compactor->Init(fpGetDeviceProcAddr, pAllocator);
}
return VK_SUCCESS;
}
VKAPI_ATTR void VKAPI_CALL DestroyDevice(VkDevice device, const VkAllocationCallbacks* pAllocator) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
device_layer_data->device_dispatch_table->DestroyDevice(device, pAllocator);
// Remove from |layer_data_map| and free LayerData struct.
FreeLayerDataPtr(device_key, layer_data_map);
}
VKAPI_ATTR VkResult VKAPI_CALL CreateImage(VkDevice device, const VkImageCreateInfo* pCreateInfo,
const VkAllocationCallbacks* pAllocator,
VkImage* pImage) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->CreateImage(pCreateInfo, pAllocator, pImage);
}
VKAPI_ATTR void VKAPI_CALL DestroyImage(VkDevice device, VkImage image,
const VkAllocationCallbacks* pAllocator) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
compactor->DestroyImage(image, pAllocator);
}
VKAPI_ATTR void VKAPI_CALL GetImageMemoryRequirements2(VkDevice device,
const VkImageMemoryRequirementsInfo2* pInfo,
VkMemoryRequirements2* pMemoryRequirements) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
compactor->GetImageMemoryRequirements2(pInfo, pMemoryRequirements);
}
VKAPI_ATTR VkResult VKAPI_CALL AllocateMemory(VkDevice device,
const VkMemoryAllocateInfo* pAllocateInfo,
const VkAllocationCallbacks* pAllocator,
VkDeviceMemory* pMemory) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->AllocateMemory(pAllocateInfo, pAllocator, pMemory);
}
VKAPI_ATTR void VKAPI_CALL FreeMemory(VkDevice device, VkDeviceMemory memory,
const VkAllocationCallbacks* pAllocator) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
compactor->FreeMemory(memory, pAllocator);
}
VKAPI_ATTR VkResult VKAPI_CALL BindImageMemory(VkDevice device, VkImage image,
VkDeviceMemory memory, VkDeviceSize memoryOffset) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->BindImageMemory(image, memory, memoryOffset);
}
VKAPI_ATTR VkResult VKAPI_CALL BindImageMemory2(VkDevice device, uint32_t bindInfoCount,
const VkBindImageMemoryInfo* pBindInfos) {
dispatch_key device_key = get_dispatch_key(device);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->BindImageMemory2(bindInfoCount, pBindInfos);
}
VKAPI_ATTR VkResult VKAPI_CALL BeginCommandBuffer(VkCommandBuffer commandBuffer,
const VkCommandBufferBeginInfo* pBeginInfo) {
dispatch_key device_key = get_dispatch_key(commandBuffer);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->BeginCommandBuffer(commandBuffer, pBeginInfo);
}
VKAPI_ATTR VkResult VKAPI_CALL EndCommandBuffer(VkCommandBuffer commandBuffer) {
dispatch_key device_key = get_dispatch_key(commandBuffer);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->EndCommandBuffer(commandBuffer);
}
VKAPI_ATTR void VKAPI_CALL CmdBindPipeline(VkCommandBuffer commandBuffer,
VkPipelineBindPoint pipelineBindPoint,
VkPipeline pipeline) {
dispatch_key device_key = get_dispatch_key(commandBuffer);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->CmdBindPipeline(commandBuffer, pipelineBindPoint, pipeline);
}
VKAPI_ATTR void VKAPI_CALL CmdPushConstants(VkCommandBuffer commandBuffer, VkPipelineLayout layout,
VkShaderStageFlags stageFlags, uint32_t offset,
uint32_t size, const void* pValues) {
dispatch_key device_key = get_dispatch_key(commandBuffer);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
return compactor->CmdPushConstants(commandBuffer, layout, stageFlags, offset, size, pValues);
}
VKAPI_ATTR void VKAPI_CALL CmdPipelineBarrier(
VkCommandBuffer commandBuffer, VkPipelineStageFlags srcStageMask,
VkPipelineStageFlags dstStageMask, VkDependencyFlags dependencyFlags,
uint32_t memoryBarrierCount, const VkMemoryBarrier* pMemoryBarriers,
uint32_t bufferMemoryBarrierCount, const VkBufferMemoryBarrier* pBufferMemoryBarriers,
uint32_t imageMemoryBarrierCount, const VkImageMemoryBarrier* pImageMemoryBarriers) {
dispatch_key device_key = get_dispatch_key(commandBuffer);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
ImageCompactor* compactor = device_layer_data->compactor.get();
FX_CHECK(compactor);
compactor->CmdPipelineBarrier(commandBuffer, srcStageMask, dstStageMask, dependencyFlags,
memoryBarrierCount, pMemoryBarriers, bufferMemoryBarrierCount,
pBufferMemoryBarriers, imageMemoryBarrierCount,
pImageMemoryBarriers);
}
VKAPI_ATTR void VKAPI_CALL CmdWriteCompactImageMemorySizeFUCHSIA(
VkCommandBuffer commandBuffer, VkImage image, VkImageLayout imageLayout, VkBuffer buffer,
VkDeviceSize bufferOffset, const VkImageSubresourceLayers* subresourceLayers) {
dispatch_key device_key = get_dispatch_key(commandBuffer);
LayerData* device_layer_data = GetLayerDataPtr(device_key, layer_data_map);
device_layer_data->compactor->CmdWriteCompactImageMemorySize(
commandBuffer, image, imageLayout, buffer, bufferOffset, subresourceLayers);
}
VKAPI_ATTR VkResult VKAPI_CALL EnumerateInstanceLayerProperties(uint32_t* pCount,
VkLayerProperties* pProperties) {
return util_GetLayerProperties(1, &compact_image_layer, pCount, pProperties);
}
VKAPI_ATTR VkResult VKAPI_CALL EnumerateDeviceLayerProperties(VkPhysicalDevice physicalDevice,
uint32_t* pCount,
VkLayerProperties* pProperties) {
return util_GetLayerProperties(1, &compact_image_layer, pCount, pProperties);
}
VKAPI_ATTR VkResult VKAPI_CALL EnumerateInstanceExtensionProperties(
const char* pLayerName, uint32_t* pCount, VkExtensionProperties* pProperties) {
if (pLayerName && !strcmp(pLayerName, compact_image_layer.layerName)) {
return util_GetExtensionProperties(0, nullptr, pCount, pProperties);
}
return VK_ERROR_LAYER_NOT_PRESENT;
}
VKAPI_ATTR VkResult VKAPI_CALL
EnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice, const char* pLayerName,
uint32_t* pCount, VkExtensionProperties* pProperties) {
if (pLayerName && !strcmp(pLayerName, compact_image_layer.layerName)) {
return util_GetExtensionProperties(ARRAY_SIZE(device_extensions), device_extensions, pCount,
pProperties);
}
FX_CHECK(physicalDevice);
dispatch_key key = get_dispatch_key(physicalDevice);
LayerData* data = GetLayerDataPtr(key, layer_data_map);
return data->instance_dispatch_table->EnumerateDeviceExtensionProperties(physicalDevice, nullptr,
pCount, pProperties);
}
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL GetDeviceProcAddr(VkDevice device, const char* funcName);
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL GetInstanceProcAddr(VkInstance instance,
const char* funcName);
static inline PFN_vkVoidFunction layer_intercept_proc(const char* name) {
if (!name || name[0] != 'v' || name[1] != 'k') {
return nullptr;
}
name += 2;
if (!strcmp(name, "GetDeviceProcAddr")) {
return reinterpret_cast<PFN_vkVoidFunction>(GetDeviceProcAddr);
}
if (!strcmp(name, "CreateInstance")) {
return reinterpret_cast<PFN_vkVoidFunction>(CreateInstance);
}
if (!strcmp(name, "DestroyInstance")) {
return reinterpret_cast<PFN_vkVoidFunction>(DestroyInstance);
}
if (!strcmp(name, "CreateDevice")) {
return reinterpret_cast<PFN_vkVoidFunction>(CreateDevice);
}
if (!strcmp(name, "DestroyDevice")) {
return reinterpret_cast<PFN_vkVoidFunction>(DestroyDevice);
}
if (!strcmp(name, "CreateImage")) {
return reinterpret_cast<PFN_vkVoidFunction>(CreateImage);
}
if (!strcmp(name, "DestroyImage")) {
return reinterpret_cast<PFN_vkVoidFunction>(DestroyImage);
}
if (!strcmp(name, "GetImageMemoryRequirements2")) {
return reinterpret_cast<PFN_vkVoidFunction>(GetImageMemoryRequirements2);
}
if (!strcmp(name, "AllocateMemory")) {
return reinterpret_cast<PFN_vkVoidFunction>(AllocateMemory);
}
if (!strcmp(name, "FreeMemory")) {
return reinterpret_cast<PFN_vkVoidFunction>(FreeMemory);
}
if (!strcmp(name, "BindImageMemory")) {
return reinterpret_cast<PFN_vkVoidFunction>(BindImageMemory);
}
if (!strcmp(name, "BindImageMemory2")) {
return reinterpret_cast<PFN_vkVoidFunction>(BindImageMemory2);
}
if (!strcmp(name, "BeginCommandBuffer")) {
return reinterpret_cast<PFN_vkVoidFunction>(BeginCommandBuffer);
}
if (!strcmp(name, "EndCommandBuffer")) {
return reinterpret_cast<PFN_vkVoidFunction>(EndCommandBuffer);
}
if (!strcmp(name, "CmdPipelineBarrier")) {
return reinterpret_cast<PFN_vkVoidFunction>(CmdPipelineBarrier);
}
if (!strcmp(name, "CmdWriteCompactImageMemorySizeFUCHSIA")) {
return reinterpret_cast<PFN_vkVoidFunction>(CmdWriteCompactImageMemorySizeFUCHSIA);
}
if (!strcmp("EnumerateDeviceExtensionProperties", name)) {
return reinterpret_cast<PFN_vkVoidFunction>(EnumerateDeviceExtensionProperties);
}
if (!strcmp("EnumerateInstanceExtensionProperties", name)) {
return reinterpret_cast<PFN_vkVoidFunction>(EnumerateInstanceExtensionProperties);
}
if (!strcmp("EnumerateDeviceLayerProperties", name)) {
return reinterpret_cast<PFN_vkVoidFunction>(EnumerateDeviceLayerProperties);
}
if (!strcmp("EnumerateInstanceLayerProperties", name)) {
return reinterpret_cast<PFN_vkVoidFunction>(EnumerateInstanceLayerProperties);
}
return nullptr;
}
static inline PFN_vkVoidFunction layer_intercept_instance_proc(const char* name) {
if (!name || name[0] != 'v' || name[1] != 'k') {
return nullptr;
}
name += 2;
if (!strcmp(name, "GetInstanceProcAddr")) {
return reinterpret_cast<PFN_vkVoidFunction>(GetInstanceProcAddr);
}
if (!strcmp(name, "CreateInstance")) {
return reinterpret_cast<PFN_vkVoidFunction>(CreateInstance);
}
if (!strcmp(name, "DestroyInstance")) {
return reinterpret_cast<PFN_vkVoidFunction>(DestroyInstance);
}
if (!strcmp("GetPhysicalDeviceImageFormatProperties", name)) {
return reinterpret_cast<PFN_vkVoidFunction>(GetPhysicalDeviceImageFormatProperties);
}
if (!strcmp("GetPhysicalDeviceImageFormatProperties2", name)) {
return reinterpret_cast<PFN_vkVoidFunction>(GetPhysicalDeviceImageFormatProperties2);
}
return nullptr;
}
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL GetDeviceProcAddr(VkDevice device, const char* funcName) {
PFN_vkVoidFunction addr;
LayerData* data;
FX_CHECK(device);
data = GetLayerDataPtr(get_dispatch_key(device), layer_data_map);
if (data->compactor) {
addr = layer_intercept_proc(funcName);
if (addr) {
return addr;
}
}
VkLayerDispatchTable* pTable = data->device_dispatch_table.get();
if (!pTable->GetDeviceProcAddr) {
return nullptr;
}
return pTable->GetDeviceProcAddr(device, funcName);
}
VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL GetInstanceProcAddr(VkInstance instance,
const char* funcName) {
PFN_vkVoidFunction addr;
LayerData* data;
addr = layer_intercept_instance_proc(funcName);
if (!addr) {
addr = layer_intercept_proc(funcName);
}
if (addr) {
return addr;
}
if (!instance) {
return nullptr;
}
data = GetLayerDataPtr(get_dispatch_key(instance), layer_data_map);
VkLayerInstanceDispatchTable* pTable = data->instance_dispatch_table.get();
if (!pTable->GetInstanceProcAddr) {
return nullptr;
}
addr = pTable->GetInstanceProcAddr(instance, funcName);
return addr;
}
} // namespace compact_image
VK_LAYER_EXPORT VKAPI_ATTR VkResult VKAPI_CALL vkEnumerateInstanceExtensionProperties(
const char* pLayerName, uint32_t* pCount, VkExtensionProperties* pProperties) {
return compact_image::EnumerateInstanceExtensionProperties(pLayerName, pCount, pProperties);
}
VK_LAYER_EXPORT VKAPI_ATTR VkResult VKAPI_CALL
vkEnumerateInstanceLayerProperties(uint32_t* pCount, VkLayerProperties* pProperties) {
return compact_image::EnumerateInstanceLayerProperties(pCount, pProperties);
}
VK_LAYER_EXPORT VKAPI_ATTR VkResult VKAPI_CALL vkEnumerateDeviceLayerProperties(
VkPhysicalDevice physicalDevice, uint32_t* pCount, VkLayerProperties* pProperties) {
FX_CHECK(physicalDevice == VK_NULL_HANDLE);
return compact_image::EnumerateDeviceLayerProperties(VK_NULL_HANDLE, pCount, pProperties);
}
VK_LAYER_EXPORT VKAPI_ATTR VkResult VKAPI_CALL
vkEnumerateDeviceExtensionProperties(VkPhysicalDevice physicalDevice, const char* pLayerName,
uint32_t* pCount, VkExtensionProperties* pProperties) {
FX_CHECK(physicalDevice == VK_NULL_HANDLE);
return compact_image::EnumerateDeviceExtensionProperties(VK_NULL_HANDLE, pLayerName, pCount,
pProperties);
}
VK_LAYER_EXPORT VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL vkGetDeviceProcAddr(VkDevice dev,
const char* funcName) {
return compact_image::GetDeviceProcAddr(dev, funcName);
}
VK_LAYER_EXPORT VKAPI_ATTR PFN_vkVoidFunction VKAPI_CALL
vkGetInstanceProcAddr(VkInstance instance, const char* funcName) {
return compact_image::GetInstanceProcAddr(instance, funcName);
}