Mirrors/yuzu
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Merge pull request #3600 from ReinUsesLisp/no-pointer-buf-cache

Browse source 
buffer_cache: Return handles instead of pointer to handles
This commit is contained in:
Fernando Sahmkow 2020-04-16 19:58:13 -04:00 • committed by GitHub
commit c81f256111
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
14 changed files with 90 additions and 228 deletions
Download patch file Download diff file Expand all files Collapse all files

12
src/video_core/renderer_vulkan/vk_buffer_cache.cpp
View file

@ -74,18 +74,18 @@ Buffer VKBufferCache::CreateBlock(VAddr cpu_addr, std::size_t size) {
return std::make_shared<CachedBufferBlock>(device, memory_manager, cpu_addr, size);
}
const VkBuffer* VKBufferCache::ToHandle(const Buffer& buffer) {
VkBuffer VKBufferCache::ToHandle(const Buffer& buffer) {
return buffer->GetHandle();
}
const VkBuffer* VKBufferCache::GetEmptyBuffer(std::size_t size) {
VkBuffer VKBufferCache::GetEmptyBuffer(std::size_t size) {
size = std::max(size, std::size_t(4));
const auto& empty = staging_pool.GetUnusedBuffer(size, false);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([size, buffer = *empty.handle](vk::CommandBuffer cmdbuf) {
cmdbuf.FillBuffer(buffer, 0, size, 0);
});
return empty.handle.address();
return *empty.handle;
}
void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
@ -94,7 +94,7 @@ void VKBufferCache::UploadBlockData(const Buffer& buffer, std::size_t offset, st
std::memcpy(staging.commit->Map(size), data, size);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([staging = *staging.handle, buffer = *buffer->GetHandle(), offset,
scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
size](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(staging, buffer, VkBufferCopy{0, offset, size});
@ -117,7 +117,7 @@ void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
u8* data) {
const auto& staging = staging_pool.GetUnusedBuffer(size, true);
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([staging = *staging.handle, buffer = *buffer->GetHandle(), offset,
scheduler.Record([staging = *staging.handle, buffer = buffer->GetHandle(), offset,
size](vk::CommandBuffer cmdbuf) {
VkBufferMemoryBarrier barrier;
barrier.sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER;
@ -144,7 +144,7 @@ void VKBufferCache::DownloadBlockData(const Buffer& buffer, std::size_t offset,
void VKBufferCache::CopyBlock(const Buffer& src, const Buffer& dst, std::size_t src_offset,
std::size_t dst_offset, std::size_t size) {
scheduler.RequestOutsideRenderPassOperationContext();
scheduler.Record([src_buffer = *src->GetHandle(), dst_buffer = *dst->GetHandle(), src_offset,
scheduler.Record([src_buffer = src->GetHandle(), dst_buffer = dst->GetHandle(), src_offset,
dst_offset, size](vk::CommandBuffer cmdbuf) {
cmdbuf.CopyBuffer(src_buffer, dst_buffer, VkBufferCopy{src_offset, dst_offset, size});

10
src/video_core/renderer_vulkan/vk_buffer_cache.h
View file

@ -33,8 +33,8 @@ public:
VAddr cpu_addr, std::size_t size);
~CachedBufferBlock();
const VkBuffer* GetHandle() const {
return buffer.handle.address();
VkBuffer GetHandle() const {
return *buffer.handle;
}
private:
@ -50,15 +50,15 @@ public:
VKScheduler& scheduler, VKStagingBufferPool& staging_pool);
~VKBufferCache();
const VkBuffer* GetEmptyBuffer(std::size_t size) override;
VkBuffer GetEmptyBuffer(std::size_t size) override;
protected:
VkBuffer ToHandle(const Buffer& buffer) override;
void WriteBarrier() override {}
Buffer CreateBlock(VAddr cpu_addr, std::size_t size) override;
const VkBuffer* ToHandle(const Buffer& buffer) override;
void UploadBlockData(const Buffer& buffer, std::size_t offset, std::size_t size,
const u8* data) override;

16
src/video_core/renderer_vulkan/vk_compute_pass.cpp
View file

@ -343,13 +343,13 @@ QuadArrayPass::QuadArrayPass(const VKDevice& device, VKScheduler& scheduler,
QuadArrayPass::~QuadArrayPass() = default;
std::pair<const VkBuffer*, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
std::pair<VkBuffer, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertices, u32 first) {
const u32 num_triangle_vertices = num_vertices * 6 / 4;
const std::size_t staging_size = num_triangle_vertices * sizeof(u32);
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(buffer.handle.address(), 0, staging_size);
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
scheduler.RequestOutsideRenderPassOperationContext();
@ -377,7 +377,7 @@ std::pair<const VkBuffer*, VkDeviceSize> QuadArrayPass::Assemble(u32 num_vertice
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, {barrier}, {});
});
return {buffer.handle.address(), 0};
return {*buffer.handle, 0};
}
Uint8Pass::Uint8Pass(const VKDevice& device, VKScheduler& scheduler,
@ -391,14 +391,14 @@ Uint8Pass::Uint8Pass(const VKDevice& device, VKScheduler& scheduler,
Uint8Pass::~Uint8Pass() = default;
std::pair<const VkBuffer*, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
u64 src_offset) {
std::pair<VkBuffer, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer src_buffer,
u64 src_offset) {
const auto staging_size = static_cast<u32>(num_vertices * sizeof(u16));
auto& buffer = staging_buffer_pool.GetUnusedBuffer(staging_size, false);
update_descriptor_queue.Acquire();
update_descriptor_queue.AddBuffer(&src_buffer, src_offset, num_vertices);
update_descriptor_queue.AddBuffer(buffer.handle.address(), 0, staging_size);
update_descriptor_queue.AddBuffer(src_buffer, src_offset, num_vertices);
update_descriptor_queue.AddBuffer(*buffer.handle, 0, staging_size);
const auto set = CommitDescriptorSet(update_descriptor_queue, scheduler.GetFence());
scheduler.RequestOutsideRenderPassOperationContext();
@ -422,7 +422,7 @@ std::pair<const VkBuffer*, u64> Uint8Pass::Assemble(u32 num_vertices, VkBuffer s
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT, 0, {}, barrier, {});
});
return {buffer.handle.address(), 0};
return {*buffer.handle, 0};
}
} // namespace Vulkan

4
src/video_core/renderer_vulkan/vk_compute_pass.h
View file

@ -50,7 +50,7 @@ public:
VKUpdateDescriptorQueue& update_descriptor_queue);
~QuadArrayPass();
std::pair<const VkBuffer*, VkDeviceSize> Assemble(u32 num_vertices, u32 first);
std::pair<VkBuffer, VkDeviceSize> Assemble(u32 num_vertices, u32 first);
private:
VKScheduler& scheduler;
@ -65,7 +65,7 @@ public:
VKUpdateDescriptorQueue& update_descriptor_queue);
~Uint8Pass();
std::pair<const VkBuffer*, u64> Assemble(u32 num_vertices, VkBuffer src_buffer, u64 src_offset);
std::pair<VkBuffer, u64> Assemble(u32 num_vertices, VkBuffer src_buffer, u64 src_offset);
private:
VKScheduler& scheduler;

30
src/video_core/renderer_vulkan/vk_rasterizer.cpp
View file

@ -137,13 +137,13 @@ Tegra::Texture::FullTextureInfo GetTextureInfo(const Engine& engine, const Entry
class BufferBindings final {
public:
void AddVertexBinding(const VkBuffer* buffer, VkDeviceSize offset) {
vertex.buffer_ptrs[vertex.num_buffers] = buffer;
void AddVertexBinding(VkBuffer buffer, VkDeviceSize offset) {
vertex.buffers[vertex.num_buffers] = buffer;
vertex.offsets[vertex.num_buffers] = offset;
++vertex.num_buffers;
}
void SetIndexBinding(const VkBuffer* buffer, VkDeviceSize offset, VkIndexType type) {
void SetIndexBinding(VkBuffer buffer, VkDeviceSize offset, VkIndexType type) {
index.buffer = buffer;
index.offset = offset;
index.type = type;
@ -227,19 +227,19 @@ private:
// Some of these fields are intentionally left uninitialized to avoid initializing them twice.
struct {
std::size_t num_buffers = 0;
std::array<const VkBuffer*, Maxwell::NumVertexArrays> buffer_ptrs;
std::array<VkBuffer, Maxwell::NumVertexArrays> buffers;
std::array<VkDeviceSize, Maxwell::NumVertexArrays> offsets;
} vertex;
struct {
const VkBuffer* buffer = nullptr;
VkBuffer buffer = nullptr;
VkDeviceSize offset;
VkIndexType type;
} index;
template <std::size_t N>
void BindStatic(VKScheduler& scheduler) const {
if (index.buffer != nullptr) {
if (index.buffer) {
BindStatic<N, true>(scheduler);
} else {
BindStatic<N, false>(scheduler);
@ -254,18 +254,14 @@ private:
}
std::array<VkBuffer, N> buffers;
std::transform(vertex.buffer_ptrs.begin(), vertex.buffer_ptrs.begin() + N, buffers.begin(),
[](const auto ptr) { return *ptr; });
std::array<VkDeviceSize, N> offsets;
std::copy(vertex.buffers.begin(), vertex.buffers.begin() + N, buffers.begin());
std::copy(vertex.offsets.begin(), vertex.offsets.begin() + N, offsets.begin());
if constexpr (is_indexed) {
// Indexed draw
scheduler.Record([buffers, offsets, index_buffer = *index.buffer,
index_offset = index.offset,
index_type = index.type](vk::CommandBuffer cmdbuf) {
cmdbuf.BindIndexBuffer(index_buffer, index_offset, index_type);
scheduler.Record([buffers, offsets, index = index](vk::CommandBuffer cmdbuf) {
cmdbuf.BindIndexBuffer(index.buffer, index.offset, index.type);
cmdbuf.BindVertexBuffers(0, static_cast<u32>(N), buffers.data(), offsets.data());
});
} else {
@ -790,7 +786,7 @@ void RasterizerVulkan::BeginTransformFeedback() {
const std::size_t size = binding.buffer_size;
const auto [buffer, offset] = buffer_cache.UploadMemory(gpu_addr, size, 4, true);
scheduler.Record([buffer = *buffer, offset = offset, size](vk::CommandBuffer cmdbuf) {
scheduler.Record([buffer = buffer, offset = offset, size](vk::CommandBuffer cmdbuf) {
cmdbuf.BindTransformFeedbackBuffersEXT(0, 1, &buffer, &offset, &size);
cmdbuf.BeginTransformFeedbackEXT(0, 0, nullptr, nullptr);
});
@ -870,7 +866,7 @@ void RasterizerVulkan::SetupIndexBuffer(BufferBindings& buffer_bindings, DrawPar
auto format = regs.index_array.format;
const bool is_uint8 = format == Maxwell::IndexFormat::UnsignedByte;
if (is_uint8 && !device.IsExtIndexTypeUint8Supported()) {
std::tie(buffer, offset) = uint8_pass.Assemble(params.num_vertices, *buffer, offset);
std::tie(buffer, offset) = uint8_pass.Assemble(params.num_vertices, buffer, offset);
format = Maxwell::IndexFormat::UnsignedShort;
}
@ -1007,8 +1003,8 @@ void RasterizerVulkan::SetupGlobalBuffer(const GlobalBufferEntry& entry, GPUVAdd
const auto size = memory_manager.Read<u32>(address + 8);
if (size == 0) {
// Sometimes global memory pointers don't have a proper size. Upload a dummy entry because
// Vulkan doesn't like empty buffers.
// Sometimes global memory pointers don't have a proper size. Upload a dummy entry
// because Vulkan doesn't like empty buffers.
constexpr std::size_t dummy_size = 4;
const auto buffer = buffer_cache.GetEmptyBuffer(dummy_size);
update_descriptor_queue.AddBuffer(buffer, 0, dummy_size);

5
src/video_core/renderer_vulkan/vk_update_descriptor.cpp
View file

@ -35,12 +35,13 @@ void VKUpdateDescriptorQueue::Send(VkDescriptorUpdateTemplateKHR update_template
payload.clear();
}
// TODO(Rodrigo): Rework to write the payload directly
const auto payload_start = payload.data() + payload.size();
for (const auto& entry : entries) {
if (const auto image = std::get_if<VkDescriptorImageInfo>(&entry)) {
payload.push_back(*image);
} else if (const auto buffer = std::get_if<Buffer>(&entry)) {
payload.emplace_back(*buffer->buffer, buffer->offset, buffer->size);
} else if (const auto buffer = std::get_if<VkDescriptorBufferInfo>(&entry)) {
payload.push_back(*buffer);
} else if (const auto texel = std::get_if<VkBufferView>(&entry)) {
payload.push_back(*texel);
} else {

16
src/video_core/renderer_vulkan/vk_update_descriptor.h
View file

@ -18,12 +18,11 @@ class VKScheduler;
class DescriptorUpdateEntry {
public:
explicit DescriptorUpdateEntry() : image{} {}
explicit DescriptorUpdateEntry() {}
DescriptorUpdateEntry(VkDescriptorImageInfo image) : image{image} {}
DescriptorUpdateEntry(VkBuffer buffer, VkDeviceSize offset, VkDeviceSize size)
: buffer{buffer, offset, size} {}
DescriptorUpdateEntry(VkDescriptorBufferInfo buffer) : buffer{buffer} {}
DescriptorUpdateEntry(VkBufferView texel_buffer) : texel_buffer{texel_buffer} {}
@ -54,8 +53,8 @@ public:
entries.emplace_back(VkDescriptorImageInfo{{}, image_view, {}});
}
void AddBuffer(const VkBuffer* buffer, u64 offset, std::size_t size) {
entries.push_back(Buffer{buffer, offset, size});
void AddBuffer(VkBuffer buffer, u64 offset, std::size_t size) {
entries.emplace_back(VkDescriptorBufferInfo{buffer, offset, size});
}
void AddTexelBuffer(VkBufferView texel_buffer) {
@ -67,12 +66,7 @@ public:
}
private:
struct Buffer {
const VkBuffer* buffer = nullptr;
u64 offset = 0;
std::size_t size = 0;
};
using Variant = std::variant<VkDescriptorImageInfo, Buffer, VkBufferView>;
using Variant = std::variant<VkDescriptorImageInfo, VkDescriptorBufferInfo, VkBufferView>;
const VKDevice& device;
VKScheduler& scheduler;