INSANEMODE/yuzu-android
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu-android.git synced 2025-06-22 10:37:52 -05:00
Code Issues Packages Projects Releases Wiki Activity

vulkan: Implement FidelityFX Super Resolution

Browse Source
This commit is contained in:
Marshall Mohror
2021-10-16 20:33:58 -05:00
committed by Fernando Sahmkow
parent d4f5193bd3
commit 37cb0377ae
11 changed files with 643 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

62
src/video_core/renderer_vulkan/vk_blit_screen.cpp
View File

@ -23,6 +23,7 @@
#include "video_core/host_shaders/vulkan_present_vert_spv.h"
#include "video_core/renderer_vulkan/renderer_vulkan.h"
#include "video_core/renderer_vulkan/vk_blit_screen.h"
#include "video_core/renderer_vulkan/vk_fsr.h"
#include "video_core/renderer_vulkan/vk_master_semaphore.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
@ -147,8 +148,12 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
scheduler.Wait(resource_ticks[image_index]);
resource_ticks[image_index] = scheduler.CurrentTick();
UpdateDescriptorSet(image_index,
use_accelerated ? screen_info.image_view : *raw_image_views[image_index]);
const VkImageView source_image_view =
use_accelerated ? screen_info.image_view : *raw_image_views[image_index];
if (!fsr) {
UpdateDescriptorSet(image_index, source_image_view);
}
BufferData data;
SetUniformData(data, layout);
@ -225,9 +230,26 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
read_barrier);
cmdbuf.CopyBufferToImage(*buffer, image, VK_IMAGE_LAYOUT_GENERAL, copy);
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_TRANSFER_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, write_barrier);
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT |
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
0, write_barrier);
});
}
if (fsr) {
auto crop_rect = framebuffer.crop_rect;
if (crop_rect.GetWidth() == 0) {
crop_rect.right = framebuffer.width;
}
if (crop_rect.GetHeight() == 0) {
crop_rect.bottom = framebuffer.height;
}
crop_rect = crop_rect.Scale(Settings::values.resolution_info.up_factor);
VkImageView fsr_image_view =
fsr->Draw(scheduler, image_index, source_image_view, crop_rect);
UpdateDescriptorSet(image_index, fsr_image_view);
}
scheduler.Record(
[this, host_framebuffer, image_index, size = render_area](vk::CommandBuffer cmdbuf) {
const f32 bg_red = Settings::values.bg_red.GetValue() / 255.0f;
@ -325,6 +347,13 @@ void VKBlitScreen::CreateDynamicResources() {
CreateRenderPass();
CreateFramebuffers();
CreateGraphicsPipeline();
fsr.reset();
if (Settings::values.scaling_filter.GetValue() == Settings::ScalingFilter::Fsr) {
const auto& layout = render_window.GetFramebufferLayout();
fsr = std::make_unique<FSR>(
device, memory_allocator, image_count,
VkExtent2D{.width = layout.screen.GetWidth(), .height = layout.screen.GetHeight()});
}
}
void VKBlitScreen::RefreshResources(const Tegra::FramebufferConfig& framebuffer) {
@ -716,13 +745,14 @@ void VKBlitScreen::CreateGraphicsPipeline() {
}
void VKBlitScreen::CreateSampler() {
bool linear = Settings::values.scaling_filter.GetValue() != Settings::ScalingFilter::Fsr;
const VkSamplerCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.magFilter = VK_FILTER_LINEAR,
.minFilter = VK_FILTER_LINEAR,
.mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR,
.magFilter = linear ? VK_FILTER_LINEAR : VK_FILTER_NEAREST,
.minFilter = linear ? VK_FILTER_LINEAR : VK_FILTER_NEAREST,
.mipmapMode = VK_SAMPLER_MIPMAP_MODE_NEAREST,
.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER,
.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER,
.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER,
@ -905,17 +935,19 @@ void VKBlitScreen::SetVertexData(BufferData& data, const Tegra::FramebufferConfi
UNIMPLEMENTED_IF(framebuffer_crop_rect.top != 0);
UNIMPLEMENTED_IF(framebuffer_crop_rect.left != 0);
// Scale the output by the crop width/height. This is commonly used with 1280x720 rendering
// (e.g. handheld mode) on a 1920x1080 framebuffer.
f32 scale_u = 1.0f;
f32 scale_v = 1.0f;
if (framebuffer_crop_rect.GetWidth() > 0) {
scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) /
static_cast<f32>(screen_info.width);
}
if (framebuffer_crop_rect.GetHeight() > 0) {
scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) /
static_cast<f32>(screen_info.height);
// Scale the output by the crop width/height. This is commonly used with 1280x720 rendering
// (e.g. handheld mode) on a 1920x1080 framebuffer.
if (!fsr) {
if (framebuffer_crop_rect.GetWidth() > 0) {
scale_u = static_cast<f32>(framebuffer_crop_rect.GetWidth()) /
static_cast<f32>(screen_info.width);
}
if (framebuffer_crop_rect.GetHeight() > 0) {
scale_v = static_cast<f32>(framebuffer_crop_rect.GetHeight()) /
static_cast<f32>(screen_info.height);
}
}
const auto& screen = layout.screen;

4
src/video_core/renderer_vulkan/vk_blit_screen.h
View File

@ -38,6 +38,8 @@ class RasterizerVulkan;
class VKScheduler;
class VKSwapchain;
class FSR;
struct VKScreenInfo {
VkImageView image_view{};
u32 width{};
@ -132,6 +134,8 @@ private:
std::vector<MemoryCommit> raw_buffer_commits;
u32 raw_width = 0;
u32 raw_height = 0;
std::unique_ptr<FSR> fsr;
};
} // namespace Vulkan

375
src/video_core/renderer_vulkan/vk_fsr.cpp Normal file
View File

@ -0,0 +1,375 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#include "common/div_ceil.h"
#include "video_core/host_shaders/vulkan_fidelityfx_fsr_easu_comp_spv.h"
#include "video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_comp_spv.h"
#include "video_core/renderer_vulkan/vk_fsr.h"
#include "video_core/renderer_vulkan/vk_scheduler.h"
#include "video_core/renderer_vulkan/vk_shader_util.h"
#include "video_core/vulkan_common/vulkan_device.h"
namespace Vulkan {
FSR::FSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count,
VkExtent2D output_size)
: device{device}, memory_allocator{memory_allocator}, image_count{image_count},
output_size{output_size} {
CreateImages();
CreateSampler();
CreateShaders();
CreateDescriptorPool();
CreateDescriptorSetLayout();
CreateDescriptorSets();
CreatePipelineLayout();
CreatePipeline();
}
VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view,
const Common::Rectangle<int>& crop_rect) {
UpdateDescriptorSet(image_index, image_view);
scheduler.Record([this, image_index, crop_rect](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier base_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
.srcAccessMask = 0,
.dstAccessMask = 0,
.oldLayout = VK_IMAGE_LAYOUT_GENERAL,
.newLayout = VK_IMAGE_LAYOUT_GENERAL,
.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED,
.image = {},
.subresourceRange =
{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = 1,
},
};
// TODO: Support clear color
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *easu_pipeline);
cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT,
VkExtent2D{
.width = static_cast<u32>(crop_rect.GetWidth()),
.height = static_cast<u32>(crop_rect.GetHeight()),
});
{
VkImageMemoryBarrier fsr_write_barrier = base_barrier;
fsr_write_barrier.image = *images[image_index],
fsr_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_ALL_COMMANDS_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, fsr_write_barrier);
}
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0,
descriptor_sets[image_index * 2], {});
cmdbuf.Dispatch(Common::DivCeil(output_size.width, 16u),
Common::DivCeil(output_size.height, 16u), 1);
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_COMPUTE, *rcas_pipeline);
cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, output_size);
{
std::array<VkImageMemoryBarrier, 2> barriers;
auto& fsr_read_barrier = barriers[0];
auto& blit_write_barrier = barriers[1];
fsr_read_barrier = base_barrier;
fsr_read_barrier.image = *images[image_index];
fsr_read_barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
fsr_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
blit_write_barrier = base_barrier;
blit_write_barrier.image = *images[image_count + image_index];
blit_write_barrier.oldLayout = VK_IMAGE_LAYOUT_UNDEFINED;
blit_write_barrier.newLayout = VK_IMAGE_LAYOUT_GENERAL;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT, 0, {}, {}, barriers);
}
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_COMPUTE, *pipeline_layout, 0,
descriptor_sets[image_index * 2 + 1], {});
cmdbuf.Dispatch(Common::DivCeil(output_size.width, 16u),
Common::DivCeil(output_size.height, 16u), 1);
{
std::array<VkImageMemoryBarrier, 1> barriers;
auto& blit_read_barrier = barriers[0];
blit_read_barrier = base_barrier;
blit_read_barrier.image = *images[image_count + image_index];
blit_read_barrier.srcAccessMask = VK_ACCESS_SHADER_WRITE_BIT;
blit_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COMPUTE_SHADER_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, {}, {}, barriers);
}
});
return *image_views[image_count + image_index];
}
void FSR::CreateDescriptorPool() {
const std::array<VkDescriptorPoolSize, 2> pool_sizes{{
{
.type = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
.descriptorCount = static_cast<u32>(image_count * 2),
},
{
.type = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = static_cast<u32>(image_count * 2),
},
}};
const VkDescriptorPoolCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_POOL_CREATE_INFO,
.pNext = nullptr,
.flags = VK_DESCRIPTOR_POOL_CREATE_FREE_DESCRIPTOR_SET_BIT,
.maxSets = static_cast<u32>(image_count * 2),
.poolSizeCount = static_cast<u32>(pool_sizes.size()),
.pPoolSizes = pool_sizes.data(),
};
descriptor_pool = device.GetLogical().CreateDescriptorPool(ci);
}
void FSR::CreateDescriptorSetLayout() {
const std::array<VkDescriptorSetLayoutBinding, 2> layout_bindings{{
{
.binding = 0,
.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = sampler.address(),
},
{
.binding = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.descriptorCount = 1,
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.pImmutableSamplers = sampler.address(),
},
}};
const VkDescriptorSetLayoutCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.bindingCount = static_cast<u32>(layout_bindings.size()),
.pBindings = layout_bindings.data(),
};
descriptor_set_layout = device.GetLogical().CreateDescriptorSetLayout(ci);
}
void FSR::CreateDescriptorSets() {
const u32 sets = static_cast<u32>(image_count * 2);
const std::vector layouts(sets, *descriptor_set_layout);
const VkDescriptorSetAllocateInfo ai{
.sType = VK_STRUCTURE_TYPE_DESCRIPTOR_SET_ALLOCATE_INFO,
.pNext = nullptr,
.descriptorPool = *descriptor_pool,
.descriptorSetCount = sets,
.pSetLayouts = layouts.data(),
};
descriptor_sets = descriptor_pool.Allocate(ai);
}
void FSR::CreateImages() {
images.resize(image_count * 2);
image_views.resize(image_count * 2);
buffer_commits.resize(image_count * 2);
for (size_t i = 0; i < image_count * 2; ++i) {
images[i] = device.GetLogical().CreateImage(VkImageCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.imageType = VK_IMAGE_TYPE_2D,
.format = VK_FORMAT_R16G16B16A16_SFLOAT,
.extent =
{
.width = output_size.width,
.height = output_size.height,
.depth = 1,
},
.mipLevels = 1,
.arrayLayers = 1,
.samples = VK_SAMPLE_COUNT_1_BIT,
.tiling = VK_IMAGE_TILING_OPTIMAL,
.usage = VK_IMAGE_USAGE_TRANSFER_SRC_BIT | VK_IMAGE_USAGE_STORAGE_BIT |
VK_IMAGE_USAGE_SAMPLED_BIT,
.sharingMode = VK_SHARING_MODE_EXCLUSIVE,
.queueFamilyIndexCount = 0,
.pQueueFamilyIndices = nullptr,
.initialLayout = VK_IMAGE_LAYOUT_UNDEFINED,
});
buffer_commits[i] = memory_allocator.Commit(images[i], MemoryUsage::DeviceLocal);
image_views[i] = device.GetLogical().CreateImageView(VkImageViewCreateInfo{
.sType = VK_STRUCTURE_TYPE_IMAGE_VIEW_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.image = *images[i],
.viewType = VK_IMAGE_VIEW_TYPE_2D,
.format = VK_FORMAT_R16G16B16A16_SFLOAT,
.components =
{
.r = VK_COMPONENT_SWIZZLE_IDENTITY,
.g = VK_COMPONENT_SWIZZLE_IDENTITY,
.b = VK_COMPONENT_SWIZZLE_IDENTITY,
.a = VK_COMPONENT_SWIZZLE_IDENTITY,
},
.subresourceRange =
{
.aspectMask = VK_IMAGE_ASPECT_COLOR_BIT,
.baseMipLevel = 0,
.levelCount = 1,
.baseArrayLayer = 0,
.layerCount = 1,
},
});
}
}
void FSR::CreatePipelineLayout() {
VkPushConstantRange push_const{
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.offset = 0,
.size = sizeof(std::array<u32, 2>),
};
VkPipelineLayoutCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.setLayoutCount = 1,
.pSetLayouts = descriptor_set_layout.address(),
.pushConstantRangeCount = 1,
.pPushConstantRanges = &push_const,
};
pipeline_layout = device.GetLogical().CreatePipelineLayout(ci);
}
void FSR::UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const {
const auto fsr_image_view = *image_views[image_index];
const auto blit_image_view = *image_views[image_count + image_index];
const VkDescriptorImageInfo image_info{
.imageView = image_view,
.imageLayout = VK_IMAGE_LAYOUT_GENERAL,
};
const VkDescriptorImageInfo fsr_image_info{
.imageView = fsr_image_view,
.imageLayout = VK_IMAGE_LAYOUT_GENERAL,
};
const VkDescriptorImageInfo blit_image_info{
.imageView = blit_image_view,
.imageLayout = VK_IMAGE_LAYOUT_GENERAL,
};
VkWriteDescriptorSet sampler_write{
.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.pNext = nullptr,
.dstSet = descriptor_sets[image_index * 2],
.dstBinding = 0,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_COMBINED_IMAGE_SAMPLER,
.pImageInfo = &image_info,
.pBufferInfo = nullptr,
.pTexelBufferView = nullptr,
};
VkWriteDescriptorSet output_write{
.sType = VK_STRUCTURE_TYPE_WRITE_DESCRIPTOR_SET,
.pNext = nullptr,
.dstSet = descriptor_sets[image_index * 2],
.dstBinding = 1,
.dstArrayElement = 0,
.descriptorCount = 1,
.descriptorType = VK_DESCRIPTOR_TYPE_STORAGE_IMAGE,
.pImageInfo = &fsr_image_info,
.pBufferInfo = nullptr,
.pTexelBufferView = nullptr,
};
device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, output_write}, {});
sampler_write.dstSet = descriptor_sets[image_index * 2 + 1];
sampler_write.pImageInfo = &fsr_image_info;
output_write.dstSet = descriptor_sets[image_index * 2 + 1];
output_write.pImageInfo = &blit_image_info;
device.GetLogical().UpdateDescriptorSets(std::array{sampler_write, output_write}, {});
}
void FSR::CreateSampler() {
const VkSamplerCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_SAMPLER_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.magFilter = VK_FILTER_LINEAR,
.minFilter = VK_FILTER_LINEAR,
.mipmapMode = VK_SAMPLER_MIPMAP_MODE_LINEAR,
.addressModeU = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE,
.addressModeV = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE,
.addressModeW = VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_EDGE,
.mipLodBias = 0.0f,
.anisotropyEnable = VK_FALSE,
.maxAnisotropy = 0.0f,
.compareEnable = VK_FALSE,
.compareOp = VK_COMPARE_OP_NEVER,
.minLod = 0.0f,
.maxLod = 0.0f,
.borderColor = VK_BORDER_COLOR_FLOAT_OPAQUE_BLACK,
.unnormalizedCoordinates = VK_FALSE,
};
sampler = device.GetLogical().CreateSampler(ci);
}
void FSR::CreateShaders() {
easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_COMP_SPV);
rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_COMP_SPV);
}
void FSR::CreatePipeline() {
VkPipelineShaderStageCreateInfo shader_stage{
.sType = VK_STRUCTURE_TYPE_PIPELINE_SHADER_STAGE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.stage = VK_SHADER_STAGE_COMPUTE_BIT,
.pName = "main",
.pSpecializationInfo = nullptr,
};
VkComputePipelineCreateInfo pipeline_ci{
.sType = VK_STRUCTURE_TYPE_COMPUTE_PIPELINE_CREATE_INFO,
.pNext = nullptr,
.flags = 0,
.layout = *pipeline_layout,
.basePipelineIndex = 0,
};
shader_stage.module = *easu_shader;
pipeline_ci.stage = shader_stage;
easu_pipeline = device.GetLogical().CreateComputePipeline(pipeline_ci);
shader_stage.module = *rcas_shader;
pipeline_ci.stage = shader_stage;
rcas_pipeline = device.GetLogical().CreateComputePipeline(pipeline_ci);
}
} // namespace Vulkan

54
src/video_core/renderer_vulkan/vk_fsr.h Normal file
View File

@ -0,0 +1,54 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#pragma once
#include "common/math_util.h"
#include "video_core/vulkan_common/vulkan_memory_allocator.h"
#include "video_core/vulkan_common/vulkan_wrapper.h"
namespace Vulkan {
class Device;
class VKScheduler;
class FSR {
public:
explicit FSR(const Device& device, MemoryAllocator& memory_allocator, size_t image_count,
VkExtent2D output_size);
VkImageView Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view,
const Common::Rectangle<int>& crop_rect);
private:
void CreateDescriptorPool();
void CreateDescriptorSetLayout();
void CreateDescriptorSets();
void CreateImages();
void CreateSampler();
void CreateShaders();
void CreatePipeline();
void CreatePipelineLayout();
void UpdateDescriptorSet(std::size_t image_index, VkImageView image_view) const;
const Device& device;
MemoryAllocator& memory_allocator;
size_t image_count;
VkExtent2D output_size;
vk::DescriptorPool descriptor_pool;
vk::DescriptorSetLayout descriptor_set_layout;
vk::DescriptorSets descriptor_sets;
vk::PipelineLayout pipeline_layout;
vk::ShaderModule easu_shader;
vk::ShaderModule rcas_shader;
vk::Pipeline easu_pipeline;
vk::Pipeline rcas_pipeline;
vk::Sampler sampler;
std::vector<vk::Image> images;
std::vector<vk::ImageView> image_views;
std::vector<MemoryCommit> buffer_commits;
};
} // namespace Vulkan