ShuriZma
/
xenia-canary
mirror of https://github.com/xenia-canary/xenia-canary.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

BufferCache: Use a CircularBuffer as the transient buffer.

This commit is contained in:
Dr. Chat 2016-04-09 21:27:32 -05:00
parent b7f2c93d73
commit 4811ebc2ce
4 changed files with 67 additions and 179 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

201
src/xenia/gpu/vulkan/buffer_cache.cc
View File

@ -30,90 +30,14 @@ constexpr VkDeviceSize kConstantRegisterUniformRange =
BufferCache::BufferCache(RegisterFile* register_file,
ui::vulkan::VulkanDevice* device, size_t capacity)
: register_file_(register_file),
device_(*device),
transient_capacity_(capacity) {
// Uniform buffer.
VkBufferCreateInfo uniform_buffer_info;
uniform_buffer_info.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
uniform_buffer_info.pNext = nullptr;
uniform_buffer_info.flags = 0;
uniform_buffer_info.size = transient_capacity_;
uniform_buffer_info.usage = VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT;
uniform_buffer_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
uniform_buffer_info.queueFamilyIndexCount = 0;
uniform_buffer_info.pQueueFamilyIndices = nullptr;
auto err = vkCreateBuffer(device_, &uniform_buffer_info, nullptr,
&transient_uniform_buffer_);
CheckResult(err, "vkCreateBuffer");
// Index buffer.
VkBufferCreateInfo index_buffer_info;
index_buffer_info.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
index_buffer_info.pNext = nullptr;
index_buffer_info.flags = 0;
index_buffer_info.size = transient_capacity_;
index_buffer_info.usage = VK_BUFFER_USAGE_INDEX_BUFFER_BIT;
index_buffer_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
index_buffer_info.queueFamilyIndexCount = 0;
index_buffer_info.pQueueFamilyIndices = nullptr;
err = vkCreateBuffer(device_, &index_buffer_info, nullptr,
&transient_index_buffer_);
CheckResult(err, "vkCreateBuffer");
// Vertex buffer.
VkBufferCreateInfo vertex_buffer_info;
vertex_buffer_info.sType = VK_STRUCTURE_TYPE_BUFFER_CREATE_INFO;
vertex_buffer_info.pNext = nullptr;
vertex_buffer_info.flags = 0;
vertex_buffer_info.size = transient_capacity_;
vertex_buffer_info.usage = VK_BUFFER_USAGE_VERTEX_BUFFER_BIT;
vertex_buffer_info.sharingMode = VK_SHARING_MODE_EXCLUSIVE;
vertex_buffer_info.queueFamilyIndexCount = 0;
vertex_buffer_info.pQueueFamilyIndices = nullptr;
err = vkCreateBuffer(*device, &vertex_buffer_info, nullptr,
&transient_vertex_buffer_);
CheckResult(err, "vkCreateBuffer");
// Allocate the underlying buffer we use for all storage.
// We query all types and take the max alignment.
VkMemoryRequirements uniform_buffer_requirements;
VkMemoryRequirements index_buffer_requirements;
VkMemoryRequirements vertex_buffer_requirements;
vkGetBufferMemoryRequirements(device_, transient_uniform_buffer_,
&uniform_buffer_requirements);
vkGetBufferMemoryRequirements(device_, transient_index_buffer_,
&index_buffer_requirements);
vkGetBufferMemoryRequirements(device_, transient_vertex_buffer_,
&vertex_buffer_requirements);
uniform_buffer_alignment_ = uniform_buffer_requirements.alignment;
index_buffer_alignment_ = index_buffer_requirements.alignment;
vertex_buffer_alignment_ = vertex_buffer_requirements.alignment;
VkMemoryRequirements buffer_requirements;
buffer_requirements.size = transient_capacity_;
buffer_requirements.alignment =
std::max(uniform_buffer_requirements.alignment,
std::max(index_buffer_requirements.alignment,
vertex_buffer_requirements.alignment));
buffer_requirements.memoryTypeBits =
uniform_buffer_requirements.memoryTypeBits |
index_buffer_requirements.memoryTypeBits |
vertex_buffer_requirements.memoryTypeBits;
transient_buffer_memory_ = device->AllocateMemory(
buffer_requirements, VK_MEMORY_PROPERTY_HOST_VISIBLE_BIT);
// Alias all buffers to our memory.
vkBindBufferMemory(device_, transient_uniform_buffer_,
transient_buffer_memory_, 0);
vkBindBufferMemory(device_, transient_index_buffer_, transient_buffer_memory_,
0);
vkBindBufferMemory(device_, transient_vertex_buffer_,
transient_buffer_memory_, 0);
// Map memory and keep it mapped while we use it.
err = vkMapMemory(device_, transient_buffer_memory_, 0, VK_WHOLE_SIZE, 0,
&transient_buffer_data_);
CheckResult(err, "vkMapMemory");
: register_file_(register_file), device_(*device) {
transient_buffer_ = std::make_unique<ui::vulkan::CircularBuffer>(device);
if (!transient_buffer_->Initialize(capacity,
VK_BUFFER_USAGE_UNIFORM_BUFFER_BIT |
VK_BUFFER_USAGE_INDEX_BUFFER_BIT |
VK_BUFFER_USAGE_VERTEX_BUFFER_BIT)) {
assert_always();
}
// Descriptor pool used for all of our cached descriptors.
// In the steady state we don't allocate anything, so these are all manually
@ -129,8 +53,8 @@ BufferCache::BufferCache(RegisterFile* register_file,
pool_sizes[0].descriptorCount = 2;
descriptor_pool_info.poolSizeCount = 1;
descriptor_pool_info.pPoolSizes = pool_sizes;
err = vkCreateDescriptorPool(device_, &descriptor_pool_info, nullptr,
&descriptor_pool_);
auto err = vkCreateDescriptorPool(device_, &descriptor_pool_info, nullptr,
&descriptor_pool_);
CheckResult(err, "vkCreateDescriptorPool");
// Create the descriptor set layout used for our uniform buffer.
@ -180,7 +104,7 @@ BufferCache::BufferCache(RegisterFile* register_file,
// Initialize descriptor set with our buffers.
VkDescriptorBufferInfo buffer_info;
buffer_info.buffer = transient_uniform_buffer_;
buffer_info.buffer = transient_buffer_->gpu_buffer();
buffer_info.offset = 0;
buffer_info.range = kConstantRegisterUniformRange;
VkWriteDescriptorSet descriptor_writes[2];
@ -212,25 +136,20 @@ BufferCache::~BufferCache() {
&transient_descriptor_set_);
vkDestroyDescriptorSetLayout(device_, descriptor_set_layout_, nullptr);
vkDestroyDescriptorPool(device_, descriptor_pool_, nullptr);
vkUnmapMemory(device_, transient_buffer_memory_);
vkFreeMemory(device_, transient_buffer_memory_, nullptr);
vkDestroyBuffer(device_, transient_uniform_buffer_, nullptr);
vkDestroyBuffer(device_, transient_index_buffer_, nullptr);
vkDestroyBuffer(device_, transient_vertex_buffer_, nullptr);
transient_buffer_->Shutdown();
}
std::pair<VkDeviceSize, VkDeviceSize> BufferCache::UploadConstantRegisters(
const Shader::ConstantRegisterMap& vertex_constant_register_map,
const Shader::ConstantRegisterMap& pixel_constant_register_map) {
const Shader::ConstantRegisterMap& pixel_constant_register_map,
std::shared_ptr<ui::vulkan::Fence> fence) {
// Fat struct, including all registers:
// struct {
// vec4 float[512];
// uint bool[8];
// uint loop[32];
// };
size_t total_size =
xe::round_up(kConstantRegisterUniformRange, uniform_buffer_alignment_);
auto offset = AllocateTransientData(uniform_buffer_alignment_, total_size);
auto offset = AllocateTransientData(kConstantRegisterUniformRange, fence);
if (offset == VK_WHOLE_SIZE) {
// OOM.
return {VK_WHOLE_SIZE, VK_WHOLE_SIZE};
@ -238,8 +157,7 @@ std::pair<VkDeviceSize, VkDeviceSize> BufferCache::UploadConstantRegisters(
// Copy over all the registers.
const auto& values = register_file_->values;
uint8_t* dest_ptr =
reinterpret_cast<uint8_t*>(transient_buffer_data_) + offset;
uint8_t* dest_ptr = transient_buffer_->host_base() + offset;
std::memcpy(dest_ptr, &values[XE_GPU_REG_SHADER_CONSTANT_000_X].f32,
(512 * 4 * 4));
dest_ptr += 512 * 4 * 4;
@ -258,8 +176,8 @@ std::pair<VkDeviceSize, VkDeviceSize> BufferCache::UploadConstantRegisters(
// constant indexing.
#if 0
// Allocate space in the buffer for our data.
auto offset = AllocateTransientData(uniform_buffer_alignment_,
constant_register_map.packed_byte_length);
auto offset =
AllocateTransientData(constant_register_map.packed_byte_length, fence);
if (offset == VK_WHOLE_SIZE) {
// OOM.
return VK_WHOLE_SIZE;
@ -304,11 +222,12 @@ std::pair<VkDeviceSize, VkDeviceSize> BufferCache::UploadConstantRegisters(
}
std::pair<VkBuffer, VkDeviceSize> BufferCache::UploadIndexBuffer(
const void* source_ptr, size_t source_length, IndexFormat format) {
const void* source_ptr, size_t source_length, IndexFormat format,
std::shared_ptr<ui::vulkan::Fence> fence) {
// TODO(benvanik): check cache.
// Allocate space in the buffer for our data.
auto offset = AllocateTransientData(index_buffer_alignment_, source_length);
auto offset = AllocateTransientData(source_length, fence);
if (offset == VK_WHOLE_SIZE) {
// OOM.
return {nullptr, VK_WHOLE_SIZE};
@ -319,25 +238,24 @@ std::pair<VkBuffer, VkDeviceSize> BufferCache::UploadIndexBuffer(
// TODO(benvanik): memcpy then use compute shaders to swap?
if (format == IndexFormat::kInt16) {
// Endian::k8in16, swap half-words.
xe::copy_and_swap_16_aligned(
reinterpret_cast<uint8_t*>(transient_buffer_data_) + offset, source_ptr,
source_length / 2);
xe::copy_and_swap_16_aligned(transient_buffer_->host_base() + offset,
source_ptr, source_length / 2);
} else if (format == IndexFormat::kInt32) {
// Endian::k8in32, swap words.
xe::copy_and_swap_32_aligned(
reinterpret_cast<uint8_t*>(transient_buffer_data_) + offset, source_ptr,
source_length / 4);
xe::copy_and_swap_32_aligned(transient_buffer_->host_base() + offset,
source_ptr, source_length / 4);
}
return {transient_index_buffer_, offset};
return {transient_buffer_->gpu_buffer(), offset};
}
std::pair<VkBuffer, VkDeviceSize> BufferCache::UploadVertexBuffer(
const void* source_ptr, size_t source_length) {
const void* source_ptr, size_t source_length,
std::shared_ptr<ui::vulkan::Fence> fence) {
// TODO(benvanik): check cache.
// Allocate space in the buffer for our data.
auto offset = AllocateTransientData(vertex_buffer_alignment_, source_length);
auto offset = AllocateTransientData(source_length, fence);
if (offset == VK_WHOLE_SIZE) {
// OOM.
return {nullptr, VK_WHOLE_SIZE};
@ -346,59 +264,34 @@ std::pair<VkBuffer, VkDeviceSize> BufferCache::UploadVertexBuffer(
// Copy data into the buffer.
// TODO(benvanik): memcpy then use compute shaders to swap?
// Endian::k8in32, swap words.
xe::copy_and_swap_32_aligned(
reinterpret_cast<uint8_t*>(transient_buffer_data_) + offset, source_ptr,
source_length / 4);
xe::copy_and_swap_32_aligned(transient_buffer_->host_base() + offset,
source_ptr, source_length / 4);
return {transient_vertex_buffer_, offset};
return {transient_buffer_->gpu_buffer(), offset};
}
VkDeviceSize BufferCache::AllocateTransientData(VkDeviceSize alignment,
VkDeviceSize length) {
VkDeviceSize BufferCache::AllocateTransientData(
VkDeviceSize length, std::shared_ptr<ui::vulkan::Fence> fence) {
// Try fast path (if we have space).
VkDeviceSize offset = TryAllocateTransientData(alignment, length);
VkDeviceSize offset = TryAllocateTransientData(length, fence);
if (offset != VK_WHOLE_SIZE) {
return offset;
}
// Ran out of easy allocations.
// Try consuming fences before we panic.
assert_always("Reclamation not yet implemented");
transient_buffer_->Scavenge();
// Try again. It may still fail if we didn't get enough space back.
return TryAllocateTransientData(alignment, length);
offset = TryAllocateTransientData(length, fence);
return offset;
}
VkDeviceSize BufferCache::TryAllocateTransientData(VkDeviceSize alignment,
VkDeviceSize length) {
if (transient_tail_offset_ >= transient_head_offset_) {
// Tail follows head, so things are easy:
// | H----T |
if (xe::round_up(transient_tail_offset_, alignment) + length <=
transient_capacity_) {
// Allocation fits from tail to end of buffer, so grow.
// | H----**T |
VkDeviceSize offset = xe::round_up(transient_tail_offset_, alignment);
transient_tail_offset_ = offset + length;
return offset;
} else if (length + kDeadZone <= transient_head_offset_) {
// Can't fit at the end, but can fit if we wrap around.
// |**T H----....|
VkDeviceSize offset = 0;
transient_tail_offset_ = length;
return offset;
}
} else {
// Head follows tail, so we're reversed:
// |----T H---|
if (xe::round_up(transient_tail_offset_, alignment) + length + kDeadZone <=
transient_head_offset_) {
// Fits from tail to head.
// |----***T H---|
VkDeviceSize offset = xe::round_up(transient_tail_offset_, alignment);
transient_tail_offset_ = offset + length;
return offset;
}
VkDeviceSize BufferCache::TryAllocateTransientData(
VkDeviceSize length, std::shared_ptr<ui::vulkan::Fence> fence) {
auto alloc = transient_buffer_->Acquire(length, fence);
if (alloc) {
return alloc->offset;
}
// No more space.
@ -420,9 +313,9 @@ void BufferCache::Flush(VkCommandBuffer command_buffer) {
VkMappedMemoryRange dirty_range;
dirty_range.sType = VK_STRUCTURE_TYPE_MAPPED_MEMORY_RANGE;
dirty_range.pNext = nullptr;
dirty_range.memory = transient_buffer_memory_;
dirty_range.memory = transient_buffer_->gpu_memory();
dirty_range.offset = 0;
dirty_range.size = transient_capacity_;
dirty_range.size = transient_buffer_->capacity();
vkFlushMappedMemoryRanges(device_, 1, &dirty_range);
}
@ -432,10 +325,10 @@ void BufferCache::InvalidateCache() {
void BufferCache::ClearCache() {
// TODO(benvanik): caching.
// Temporary clear.
transient_tail_offset_ = transient_head_offset_;
}
void BufferCache::Scavenge() { transient_buffer_->Scavenge(); }
} // namespace vulkan
} // namespace gpu
} // namespace xe

41
src/xenia/gpu/vulkan/buffer_cache.h
View File

@ -13,6 +13,7 @@
#include "xenia/gpu/register_file.h"
#include "xenia/gpu/shader.h"
#include "xenia/gpu/xenos.h"
#include "xenia/ui/vulkan/circular_buffer.h"
#include "xenia/ui/vulkan/vulkan.h"
#include "xenia/ui/vulkan/vulkan_device.h"
@ -50,22 +51,24 @@ class BufferCache {
// The returned offsets may alias.
std::pair<VkDeviceSize, VkDeviceSize> UploadConstantRegisters(
const Shader::ConstantRegisterMap& vertex_constant_register_map,
const Shader::ConstantRegisterMap& pixel_constant_register_map);
const Shader::ConstantRegisterMap& pixel_constant_register_map,
std::shared_ptr<ui::vulkan::Fence> fence);
// Uploads index buffer data from guest memory, possibly eliding with
// recently uploaded data or cached copies.
// Returns a buffer and offset that can be used with vkCmdBindIndexBuffer.
// Size will be VK_WHOLE_SIZE if the data could not be uploaded (OOM).
std::pair<VkBuffer, VkDeviceSize> UploadIndexBuffer(const void* source_ptr,
size_t source_length,
IndexFormat format);
std::pair<VkBuffer, VkDeviceSize> UploadIndexBuffer(
const void* source_ptr, size_t source_length, IndexFormat format,
std::shared_ptr<ui::vulkan::Fence> fence);
// Uploads vertex buffer data from guest memory, possibly eliding with
// recently uploaded data or cached copies.
// Returns a buffer and offset that can be used with vkCmdBindVertexBuffers.
// Size will be VK_WHOLE_SIZE if the data could not be uploaded (OOM).
std::pair<VkBuffer, VkDeviceSize> UploadVertexBuffer(const void* source_ptr,
size_t source_length);
std::pair<VkBuffer, VkDeviceSize> UploadVertexBuffer(
const void* source_ptr, size_t source_length,
std::shared_ptr<ui::vulkan::Fence> fence);
// Flushes all pending data to the GPU.
// Until this is called the GPU is not guaranteed to see any data.
@ -81,36 +84,26 @@ class BufferCache {
// Clears all cached content and prevents future elision with pending data.
void ClearCache();
// Wipes all data no longer needed.
void Scavenge();
private:
// Allocates a block of memory in the transient buffer.
// When memory is not available fences are checked and space is reclaimed.
// Returns VK_WHOLE_SIZE if requested amount of memory is not available.
VkDeviceSize AllocateTransientData(VkDeviceSize alignment,
VkDeviceSize length);
VkDeviceSize AllocateTransientData(VkDeviceSize length,
std::shared_ptr<ui::vulkan::Fence> fence);
// Tries to allocate a block of memory in the transient buffer.
// Returns VK_WHOLE_SIZE if requested amount of memory is not available.
VkDeviceSize TryAllocateTransientData(VkDeviceSize alignment,
VkDeviceSize length);
VkDeviceSize TryAllocateTransientData(
VkDeviceSize length, std::shared_ptr<ui::vulkan::Fence> fence);
RegisterFile* register_file_ = nullptr;
VkDevice device_ = nullptr;
// Staging ringbuffer we cycle through fast. Used for data we don't
// plan on keeping past the current frame.
size_t transient_capacity_ = 0;
VkBuffer transient_uniform_buffer_ = nullptr;
VkBuffer transient_index_buffer_ = nullptr;
VkBuffer transient_vertex_buffer_ = nullptr;
VkDeviceMemory transient_buffer_memory_ = nullptr;
void* transient_buffer_data_ = nullptr;
VkDeviceSize transient_head_offset_ = 0;
VkDeviceSize transient_tail_offset_ = 0;
// Required alignments for our various types.
// All allocations must start at the appropriate alignment.
VkDeviceSize uniform_buffer_alignment_ = 0;
VkDeviceSize index_buffer_alignment_ = 0;
VkDeviceSize vertex_buffer_alignment_ = 0;
std::unique_ptr<ui::vulkan::CircularBuffer> transient_buffer_ = nullptr;
VkDescriptorPool descriptor_pool_ = nullptr;
VkDescriptorSetLayout descriptor_set_layout_ = nullptr;

3
src/xenia/ui/vulkan/circular_buffer.cc
View File

@ -134,12 +134,13 @@ CircularBuffer::Allocation* CircularBuffer::Acquire(
}
VkDeviceSize aligned_length = xe::round_up(length, alignment_);
assert_true(write_head_ % alignment_ == 0);
if (allocations_.empty()) {
// Entire buffer available.
assert(read_head_ == write_head_);
assert(capacity_ > aligned_length);
write_head_ = length;
write_head_ = aligned_length;
auto alloc = new Allocation();
alloc->host_ptr = host_base_ + 0;

1
src/xenia/ui/vulkan/circular_buffer.h
View File

@ -46,6 +46,7 @@ class CircularBuffer {
VkDeviceSize alignment = 256);
void Shutdown();
VkDeviceSize alignment() const { return alignment_; }
VkDeviceSize capacity() const { return capacity_; }
VkBuffer gpu_buffer() const { return gpu_buffer_; }
VkDeviceMemory gpu_memory() const { return gpu_memory_; }