hdkv/yuzu
1
Fork 0
Code Issues Pull Requests Releases 1 Activity

early-access version 1457

Browse Source
This commit is contained in:
pineappleEA
2021-02-14 03:46:35 +01:00
parent b6510b3d37
commit 68b5ea0637
24 changed files with 2102 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/video_core/renderer_vulkan/maxwell_to_vk.cpp
View File

@@ -166,7 +166,7 @@ struct FormatTuple {
{VK_FORMAT_R16G16_SINT, Attachable | Storage}, // R16G16_SINT
{VK_FORMAT_R16G16_SNORM, Attachable | Storage}, // R16G16_SNORM
{VK_FORMAT_UNDEFINED}, // R32G32B32_FLOAT
{VK_FORMAT_R8G8B8A8_SRGB, Attachable}, // A8B8G8R8_SRGB
{VK_FORMAT_A8B8G8R8_SRGB_PACK32, Attachable}, // A8B8G8R8_SRGB
{VK_FORMAT_R8G8_UNORM, Attachable | Storage}, // R8G8_UNORM
{VK_FORMAT_R8G8_SNORM, Attachable | Storage}, // R8G8_SNORM
{VK_FORMAT_R8G8_SINT, Attachable | Storage}, // R8G8_SINT

337
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View File

@@ -11,18 +11,39 @@
#include "common/assert.h"
#include "common/common_types.h"
#include "common/div_ceil.h"
#include "video_core/host_shaders/astc_decoder_comp_spv.h"
#include "video_core/host_shaders/vulkan_quad_indexed_comp_spv.h"
#include "video_core/host_shaders/vulkan_uint8_comp_spv.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
#include "video_core/renderer_vulkan/vk_texture_cache.h"
#include "video_core/renderer_vulkan/vk_update_descriptor.h"
#include "video_core/texture_cache/accelerated_swizzle.h"
#include "video_core/texture_cache/types.h"
#include "video_core/textures/astc.h"
#include "video_core/textures/decoders.h"
#include "video_core/vulkan_common/vulkan_device.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
using Tegra::Texture::SWIZZLE_TABLE;
using Tegra::Texture::ASTC::EncodingsValues;
using namespace Tegra::Texture::ASTC;
namespace {
constexpr u32 ASTC_BINDING_INPUT_BUFFER = 0;
constexpr u32 ASTC_BINDING_ENC_BUFFER = 1;
constexpr u32 ASTC_BINDING_6_TO_8_BUFFER = 2;
constexpr u32 ASTC_BINDING_7_TO_8_BUFFER = 3;
constexpr u32 ASTC_BINDING_8_TO_8_BUFFER = 4;
constexpr u32 ASTC_BINDING_BYTE_TO_16_BUFFER = 5;
constexpr u32 ASTC_BINDING_SWIZZLE_BUFFER = 6;
constexpr u32 ASTC_BINDING_OUTPUT_IMAGE = 7;
VkPushConstantRange BuildComputePushConstantRange(std::size_t size) {
return {
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
@@ -50,6 +71,67 @@ std::array<VkDescriptorSetLayoutBinding, 2> BuildInputOutputDescriptorSetBinding
}};
}
std::array<VkDescriptorSetLayoutBinding, 8> BuildASTCDescriptorSetBindings() {
return {{
{
.binding = ASTC_BINDING_INPUT_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_ENC_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_6_TO_8_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_7_TO_8_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_8_TO_8_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_BYTE_TO_16_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_SWIZZLE_BUFFER,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
{
.binding = ASTC_BINDING_OUTPUT_IMAGE,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = nullptr,
},
}};
}
VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
return {
.dstBinding = 0,
@@ -61,6 +143,95 @@ VkDescriptorUpdateTemplateEntryKHR BuildInputOutputDescriptorUpdateTemplate() {
};
}
std::array<VkDescriptorUpdateTemplateEntryKHR, 8> BuildASTCPassDescriptorUpdateTemplateEntry() {
return {{
{
.dstBinding = ASTC_BINDING_INPUT_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_INPUT_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_ENC_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_ENC_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_6_TO_8_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_6_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_7_TO_8_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_7_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_8_TO_8_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_8_TO_8_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_BYTE_TO_16_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_BYTE_TO_16_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_SWIZZLE_BUFFER,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_BUFFER,
.offset = ASTC_BINDING_SWIZZLE_BUFFER * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
{
.dstBinding = ASTC_BINDING_OUTPUT_IMAGE,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.offset = ASTC_BINDING_OUTPUT_IMAGE * sizeof(DescriptorUpdateEntry),
.stride = sizeof(DescriptorUpdateEntry),
},
}};
}
struct AstcPushConstants {
std::array<u32, 2> num_image_blocks;
std::array<u32, 2> blocks_dims;
u32 bytes_per_block_log2;
u32 layer_stride;
u32 block_size;
u32 x_shift;
u32 block_height;
u32 block_height_mask;
};
struct AstcBufferData {
decltype(SWIZZLE_TABLE) swizzle_table_buffer = SWIZZLE_TABLE;
decltype(EncodingsValues) encoding_values = EncodingsValues;
decltype(REPLICATE_6_BIT_TO_8_TABLE) replicate_6_to_8 = REPLICATE_6_BIT_TO_8_TABLE;
decltype(REPLICATE_7_BIT_TO_8_TABLE) replicate_7_to_8 = REPLICATE_7_BIT_TO_8_TABLE;
decltype(REPLICATE_8_BIT_TO_8_TABLE) replicate_8_to_8 = REPLICATE_8_BIT_TO_8_TABLE;
decltype(REPLICATE_BYTE_TO_16_TABLE) replicate_byte_to_16 = REPLICATE_BYTE_TO_16_TABLE;
} constexpr ASTC_BUFFER_DATA;
} // Anonymous namespace
VKComputePass::VKComputePass(const Device& device, VKDescriptorPool& descriptor_pool,
@@ -238,4 +409,170 @@ std::pair<VkBuffer, VkDeviceSize> QuadIndexedPass::Assemble(
return {staging.buffer, staging.offset};
}
ASTCDecoderPass::ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
MemoryAllocator& memory_allocator_)
: VKComputePass(device_, descriptor_pool_, BuildASTCDescriptorSetBindings(),
BuildASTCPassDescriptorUpdateTemplateEntry(),
BuildComputePushConstantRange(sizeof(AstcPushConstants)),
ASTC_DECODER_COMP_SPV),
device{device_}, scheduler{scheduler_}, staging_buffer_pool{staging_buffer_pool_},
update_descriptor_queue{update_descriptor_queue_}, memory_allocator{memory_allocator_} {}
ASTCDecoderPass::~ASTCDecoderPass() = default;
void ASTCDecoderPass::MakeDataBuffer() {
constexpr auto TOTAL_BUFFER_SIZE = sizeof(ASTC_BUFFER_DATA) + sizeof(SWIZZLE_TABLE);
data_buffer = device.GetLogical().CreateBuffer(VkBufferCreateInfo{
.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.size = TOTAL_BUFFER_SIZE,
.usage = VK_BUFFER_USAGE_STORAGE_BUFFER_BIT | VK_BUFFER_USAGE_TRANSFER_DST_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
});
data_buffer_commit = memory_allocator.Commit(data_buffer, MemoryUsage::Upload);
const auto staging_ref = staging_buffer_pool.Request(TOTAL_BUFFER_SIZE, MemoryUsage::Upload);
std::memcpy(staging_ref.mapped_span.data(), &ASTC_BUFFER_DATA, sizeof(ASTC_BUFFER_DATA));
// Tack on the swizzle table at the end of the buffer
std::memcpy(staging_ref.mapped_span.data() + sizeof(ASTC_BUFFER_DATA), &SWIZZLE_TABLE,
sizeof(SWIZZLE_TABLE));
scheduler.Record([src = staging_ref.buffer, offset = staging_ref.offset, dst = *data_buffer,
TOTAL_BUFFER_SIZE](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(src, dst,
VkBufferCopy{
.srcOffset = offset,
.dstOffset = 0,
.size = TOTAL_BUFFER_SIZE,
});
cmdbuf.PipelineBarrier(
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, 0,
VkMemoryBarrier{
.sType = VK_STRUCTURE_TYPE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = VK_ACCESS_TRANSFER_READ_BIT | VK_ACCESS_TRANSFER_WRITE_BIT,
},
{}, {});
});
}
void ASTCDecoderPass::Assemble(Image& image, const StagingBufferRef& map,
std::span<const VideoCommon::SwizzleParameters> swizzles) {
using namespace VideoCommon::Accelerated;
const VideoCommon::Extent2D tile_size{
.width = VideoCore::Surface::DefaultBlockWidth(image.info.format),
.height = VideoCore::Surface::DefaultBlockHeight(image.info.format),
};
scheduler.RequestOutsideRenderPassOperationContext();
if (!data_buffer) {
MakeDataBuffer();
}
const VkImageAspectFlags aspect_mask = image.AspectMask();
const VkImage vk_image = image.Handle();
const bool is_initialized = image.ExchangeInitialization();
scheduler.Record([vk_image, aspect_mask, is_initialized](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier image_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
.oldLayout = is_initialized ? VK_IMAGE_LAYOUT_GENERAL : VK_IMAGE_LAYOUT_UNDEFINED,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = vk_image,
.subresourceRange{
.aspectMask = aspect_mask,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
cmdbuf.PipelineBarrier(0, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
});
const std::array<u32, 2> block_dims{tile_size.width, tile_size.height};
for (const VideoCommon::SwizzleParameters& swizzle : swizzles) {
const size_t input_offset = swizzle.buffer_offset + map.offset;
const u32 num_dispatches_x = Common::DivCeil(swizzle.num_tiles.width, 32U);
const u32 num_dispatches_y = Common::DivCeil(swizzle.num_tiles.height, 32U);
const u32 num_dispatches_z = image.info.resources.layers;
const std::array num_image_blocks{swizzle.num_tiles.width, swizzle.num_tiles.height};
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(map.buffer, input_offset,
image.guest_size_bytes - swizzle.buffer_offset);
update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, encoding_values),
sizeof(AstcBufferData::encoding_values));
update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, replicate_6_to_8),
sizeof(AstcBufferData::replicate_6_to_8));
update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, replicate_7_to_8),
sizeof(AstcBufferData::replicate_7_to_8));
update_descriptor_queue.AddBuffer(*data_buffer, offsetof(AstcBufferData, replicate_8_to_8),
sizeof(AstcBufferData::replicate_8_to_8));
update_descriptor_queue.AddBuffer(*data_buffer,
offsetof(AstcBufferData, replicate_byte_to_16),
sizeof(AstcBufferData::replicate_byte_to_16));
update_descriptor_queue.AddBuffer(*data_buffer, sizeof(AstcBufferData),
sizeof(SWIZZLE_TABLE));
update_descriptor_queue.AddImage(image.StorageImageView(swizzle.level));
const VkDescriptorSet set = CommitDescriptorSet(update_descriptor_queue);
const VkPipelineLayout vk_layout = *layout;
const VkPipeline vk_pipeline = *pipeline;
// To unswizzle the ASTC data
const auto params = MakeBlockLinearSwizzle2DParams(swizzle, image.info);
ASSERT(params.origin == (std::array<u32, 3>{0, 0, 0}));
ASSERT(params.destination == (std::array<s32, 3>{0, 0, 0}));
scheduler.Record([vk_layout, vk_pipeline, num_dispatches_x, num_dispatches_y,
num_dispatches_z, num_image_blocks, block_dims, params,
set](vk::CommandBuffer cmdbuf) {
const AstcPushConstants uniforms{
.num_image_blocks = num_image_blocks,
.blocks_dims = block_dims,
.bytes_per_block_log2 = params.bytes_per_block_log2,
.layer_stride = params.layer_stride,
.block_size = params.block_size,
.x_shift = params.x_shift,
.block_height = params.block_height,
.block_height_mask = params.block_height_mask,
};
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, vk_pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, vk_layout, 0, set, {});
cmdbuf.PushConstants(vk_layout, VK_SHADER_STAGE_COMPUTE_BIT, uniforms);
cmdbuf.Dispatch(num_dispatches_x, num_dispatches_y, num_dispatches_z);
});
}
scheduler.Record([vk_image, aspect_mask](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier image_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT,
.dstAccessMask = VK_ACCESS_SHADER_READ_BIT | VK_ACCESS_SHADER_WRITE_BIT,
.oldLayout = VK_IMAGE_LAYOUT_GENERAL,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = vk_image,
.subresourceRange{
.aspectMask = aspect_mask,
.baseMipLevel = 0,
.levelCount = VK_REMAINING_MIP_LEVELS,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
};
cmdbuf.PipelineBarrier(0, VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, image_barrier);
});
}
} // namespace Vulkan

32
src/video_core/renderer_vulkan/vk_compute_pass.h
View File

@@ -11,14 +11,21 @@
#include "common/common_types.h"
#include "video_core/engines/maxwell_3d.h"
#include "video_core/renderer_vulkan/vk_descriptor_pool.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace VideoCommon {
struct SwizzleParameters;
}
namespace Vulkan {
class Device;
class StagingBufferPool;
class VKScheduler;
class VKUpdateDescriptorQueue;
class Image;
struct StagingBufferRef;
class VKComputePass {
public:
@@ -77,4 +84,29 @@ private:
VKUpdateDescriptorQueue& update_descriptor_queue;
};
class ASTCDecoderPass final : public VKComputePass {
public:
explicit ASTCDecoderPass(const Device& device_, VKScheduler& scheduler_,
VKDescriptorPool& descriptor_pool_,
StagingBufferPool& staging_buffer_pool_,
VKUpdateDescriptorQueue& update_descriptor_queue_,
MemoryAllocator& memory_allocator_);
~ASTCDecoderPass();
void Assemble(Image& image, const StagingBufferRef& map,
std::span<const VideoCommon::SwizzleParameters> swizzles);
private:
void MakeDataBuffer();
const Device& device;
VKScheduler& scheduler;
StagingBufferPool& staging_buffer_pool;
VKUpdateDescriptorQueue& update_descriptor_queue;
MemoryAllocator& memory_allocator;
vk::Buffer data_buffer;
MemoryCommit data_buffer_commit;
};
} // namespace Vulkan

5
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View File

@@ -241,7 +241,10 @@ RasterizerVulkan::RasterizerVulkan(Core::Frontend::EmuWindow& emu_window_, Tegra
staging_pool(device, memory_allocator, scheduler), descriptor_pool(device, scheduler),
update_descriptor_queue(device, scheduler),
blit_image(device, scheduler, state_tracker, descriptor_pool),
texture_cache_runtime{device, scheduler, memory_allocator, staging_pool, blit_image},
astc_decoder_pass(device, scheduler, descriptor_pool, staging_pool, update_descriptor_queue,
memory_allocator),
texture_cache_runtime{device, scheduler, memory_allocator,
staging_pool, blit_image, astc_decoder_pass},
texture_cache(texture_cache_runtime, *this, maxwell3d, kepler_compute, gpu_memory),
buffer_cache_runtime(device, memory_allocator, scheduler, staging_pool,
update_descriptor_queue, descriptor_pool),

1
src/video_core/renderer_vulkan/vk_rasterizer.h
View File

@@ -174,6 +174,7 @@ private:
VKDescriptorPool descriptor_pool;
VKUpdateDescriptorQueue update_descriptor_queue;
BlitImageHelper blit_image;
ASTCDecoderPass astc_decoder_pass;
TextureCacheRuntime texture_cache_runtime;
TextureCache texture_cache;

14
src/video_core/renderer_vulkan/vk_resource_pool.cpp
View File

@@ -17,21 +17,21 @@ ResourcePool::~ResourcePool() = default;
size_t ResourcePool::CommitResource() {
// Refresh semaphore to query updated results
master_semaphore.Refresh();
const auto search = [this](size_t begin, size_t end) -> std::optional<size_t> {
const u64 gpu_tick = master_semaphore.KnownGpuTick();
const auto search = [this, gpu_tick](size_t begin, size_t end) -> std::optional<size_t> {
for (size_t iterator = begin; iterator < end; ++iterator) {
if (master_semaphore.IsFree(ticks[iterator])) {
if (gpu_tick >= ticks[iterator]) {
ticks[iterator] = master_semaphore.CurrentTick();
return iterator;
}
}
return {};
return std::nullopt;
};
// Try to find a free resource from the hinted position to the end.
auto found = search(free_iterator, ticks.size());
std::optional<size_t> found = search(hint_iterator, ticks.size());
if (!found) {
// Search from beginning to the hinted position.
found = search(0, free_iterator);
found = search(0, hint_iterator);
if (!found) {
// Both searches failed, the pool is full; handle it.
const size_t free_resource = ManageOverflow();
@@ -41,7 +41,7 @@ size_t ResourcePool::CommitResource() {
}
}
// Free iterator is hinted to the resource after the one that's been commited.
free_iterator = (*found + 1) % ticks.size();
hint_iterator = (*found + 1) % ticks.size();
return *found;
}

2
src/video_core/renderer_vulkan/vk_resource_pool.h
View File

@@ -36,7 +36,7 @@ private:
MasterSemaphore& master_semaphore;
size_t grow_step = 0; ///< Number of new resources created after an overflow
size_t free_iterator = 0; ///< Hint to where the next free resources is likely to be found
size_t hint_iterator = 0; ///< Hint to where the next free resources is likely to be found
std::vector<u64> ticks; ///< Ticks for each resource
};

49
src/video_core/renderer_vulkan/vk_texture_cache.cpp
View File

@@ -10,6 +10,7 @@
#include "video_core/engines/fermi_2d.h"
#include "video_core/renderer_vulkan/blit_image.h"
#include "video_core/renderer_vulkan/maxwell_to_vk.h"
#include "video_core/renderer_vulkan/vk_compute_pass.h"
#include "video_core/renderer_vulkan/vk_rasterizer.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_staging_buffer_pool.h"
@@ -807,7 +808,7 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
commit = runtime.memory_allocator.Commit(buffer, MemoryUsage::DeviceLocal);
}
if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
flags |= VideoCommon::ImageFlagBits::Converted;
flags |= VideoCommon::ImageFlagBits::AcceleratedUpload;
}
if (runtime.device.HasDebuggingToolAttached()) {
if (image) {
@@ -816,6 +817,38 @@ Image::Image(TextureCacheRuntime& runtime, const ImageInfo& info_, GPUVAddr gpu_
buffer.SetObjectNameEXT(VideoCommon::Name(*this).c_str());
}
}
static constexpr VkImageViewUsageCreateInfo storage_image_view_usage_create_info{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
.pNext = nullptr,
.usage = VK_IMAGE_USAGE_STORAGE_BIT,
};
if (IsPixelFormatASTC(info.format) && !runtime.device.IsOptimalAstcSupported()) {
const auto& device = runtime.device.GetLogical();
storage_image_views.reserve(info.resources.levels);
for (s32 level = 0; level < info.resources.levels; ++level) {
storage_image_views.push_back(device.CreateImageView(VkImageViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.pNext = &storage_image_view_usage_create_info,
.flags = 0,
.image = *image,
.viewType = VK_IMAGE_VIEW_TYPE_2D_ARRAY,
.format = VK_FORMAT_A8B8G8R8_UNORM_PACK32,
.components{
.r = VK_COMPONENT_SWIZZLE_IDENTITY,
.g = VK_COMPONENT_SWIZZLE_IDENTITY,
.b = VK_COMPONENT_SWIZZLE_IDENTITY,
.a = VK_COMPONENT_SWIZZLE_IDENTITY,
},
.subresourceRange{
.aspectMask = aspect_mask,
.baseMipLevel = static_cast<u32>(level),
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = VK_REMAINING_ARRAY_LAYERS,
},
}));
}
}
}
void Image::UploadMemory(const StagingBufferRef& map, std::span<const BufferImageCopy> copies) {
@@ -918,7 +951,6 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
}
}
const auto format_info = MaxwellToVK::SurfaceFormat(*device, FormatType::Optimal, true, format);
const VkFormat vk_format = format_info.format;
const VkImageViewUsageCreateInfo image_view_usage{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_USAGE_CREATE_INFO,
.pNext = nullptr,
@@ -930,7 +962,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
.flags = 0,
.image = image.Handle(),
.viewType = VkImageViewType{},
.format = vk_format,
.format = format_info.format,
.components{
.r = ComponentSwizzle(swizzle[0]),
.g = ComponentSwizzle(swizzle[1]),
@@ -982,7 +1014,7 @@ ImageView::ImageView(TextureCacheRuntime& runtime, const VideoCommon::ImageViewI
.pNext = nullptr,
.flags = 0,
.buffer = image.Buffer(),
.format = vk_format,
.format = format_info.format,
.offset = 0, // TODO: Redesign buffer cache to support this
.range = image.guest_size_bytes,
});
@@ -1167,4 +1199,13 @@ Framebuffer::Framebuffer(TextureCacheRuntime& runtime, std::span<ImageView*, NUM
}
}
void TextureCacheRuntime::AccelerateImageUpload(
Image& image, const StagingBufferRef& map,
std::span<const VideoCommon::SwizzleParameters> swizzles) {
if (IsPixelFormatASTC(image.info.format)) {
return astc_decoder_pass.Assemble(image, map, swizzles);
}
UNREACHABLE();
}
} // namespace Vulkan

19
src/video_core/renderer_vulkan/vk_texture_cache.h
View File

@@ -20,6 +20,7 @@ using VideoCommon::Offset2D;
using VideoCommon::RenderTargets;
using VideoCore::Surface::PixelFormat;
class ASTCDecoderPass;
class BlitImageHelper;
class Device;
class Image;
@@ -60,6 +61,7 @@ struct TextureCacheRuntime {
MemoryAllocator& memory_allocator;
StagingBufferPool& staging_buffer_pool;
BlitImageHelper& blit_image_helper;
ASTCDecoderPass& astc_decoder_pass;
std::unordered_map<RenderPassKey, vk::RenderPass> renderpass_cache{};
void Finish();
@@ -83,9 +85,7 @@ struct TextureCacheRuntime {
}
void AccelerateImageUpload(Image&, const StagingBufferRef&,
std::span<const VideoCommon::SwizzleParameters>) {
UNREACHABLE();
}
std::span<const VideoCommon::SwizzleParameters>);
void InsertUploadMemoryBarrier() {}
@@ -116,15 +116,26 @@ public:
return *buffer;
}
[[nodiscard]] VkImageCreateFlags AspectMask() const noexcept {
[[nodiscard]] VkImageAspectFlags AspectMask() const noexcept {
return aspect_mask;
}
[[nodiscard]] VkImageView StorageImageView(s32 level) const noexcept {
return *storage_image_views[level];
}
/// Returns true when the image is already initialized and mark it as initialized
[[nodiscard]] bool ExchangeInitialization() noexcept {
return std::exchange(initialized, true);
}
private:
VKScheduler* scheduler;
vk::Image image;
vk::Buffer buffer;
MemoryCommit commit;
vk::ImageView image_view;
std::vector<vk::ImageView> storage_image_views;
VkImageAspectFlags aspect_mask = 0;
bool initialized = false;
};