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

Presentation: Only use FP16 in scaling shaders on supported devices in Vulkan

Browse source
This commit is contained in:
Marshall Mohror 2021-10-22 23:09:29 -05:00 • committed by Fernando Sahmkow
parent a39e867c73
commit dcc5b4f6b0
15 changed files with 199 additions and 115 deletions
Download patch file Download diff file Expand all files Collapse all files

17
src/video_core/renderer_vulkan/vk_blit_screen.cpp
View file

@ -21,8 +21,9 @@
#include "video_core/host_shaders/fxaa_vert_spv.h"
#include "video_core/host_shaders/present_bicubic_frag_spv.h"
#include "video_core/host_shaders/present_gaussian_frag_spv.h"
#include "video_core/host_shaders/present_scaleforce_frag_spv.h"
#include "video_core/host_shaders/vulkan_present_frag_spv.h"
#include "video_core/host_shaders/vulkan_present_scaleforce_fp16_frag_spv.h"
#include "video_core/host_shaders/vulkan_present_scaleforce_fp32_frag_spv.h"
#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"
@ -328,7 +329,7 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
blit_read_barrier.srcAccessMask = VK_ACCESS_COLOR_ATTACHMENT_WRITE_BIT;
blit_read_barrier.dstAccessMask = VK_ACCESS_SHADER_READ_BIT;
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT ,
cmdbuf.PipelineBarrier(VK_PIPELINE_STAGE_COLOR_ATTACHMENT_OUTPUT_BIT,
VK_PIPELINE_STAGE_FRAGMENT_SHADER_BIT, 0, blit_read_barrier);
}
});
@ -344,8 +345,12 @@ VkSemaphore VKBlitScreen::Draw(const Tegra::FramebufferConfig& framebuffer,
crop_rect.bottom = framebuffer.height;
}
crop_rect = crop_rect.Scale(Settings::values.resolution_info.up_factor);
VkExtent2D fsr_input_size{
.width = Settings::values.resolution_info.ScaleUp(framebuffer.width),
.height = Settings::values.resolution_info.ScaleUp(framebuffer.height),
};
VkImageView fsr_image_view =
fsr->Draw(scheduler, image_index, source_image_view, crop_rect);
fsr->Draw(scheduler, image_index, source_image_view, fsr_input_size, crop_rect);
UpdateDescriptorSet(image_index, fsr_image_view, true);
} else {
const bool is_nn =
@ -500,7 +505,11 @@ void VKBlitScreen::CreateShaders() {
bilinear_fragment_shader = BuildShader(device, VULKAN_PRESENT_FRAG_SPV);
bicubic_fragment_shader = BuildShader(device, PRESENT_BICUBIC_FRAG_SPV);
gaussian_fragment_shader = BuildShader(device, PRESENT_GAUSSIAN_FRAG_SPV);
scaleforce_fragment_shader = BuildShader(device, PRESENT_SCALEFORCE_FRAG_SPV);
if (device.IsFloat16Supported()) {
scaleforce_fragment_shader = BuildShader(device, VULKAN_PRESENT_SCALEFORCE_FP16_FRAG_SPV);
} else {
scaleforce_fragment_shader = BuildShader(device, VULKAN_PRESENT_SCALEFORCE_FP32_FRAG_SPV);
}
}
void VKBlitScreen::CreateSemaphores() {

46
src/video_core/renderer_vulkan/vk_fsr.cpp
View file

@ -4,13 +4,19 @@
#include "common/common_types.h"
#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/host_shaders/vulkan_fidelityfx_fsr_easu_fp16_comp_spv.h"
#include "video_core/host_shaders/vulkan_fidelityfx_fsr_easu_fp32_comp_spv.h"
#include "video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp16_comp_spv.h"
#include "video_core/host_shaders/vulkan_fidelityfx_fsr_rcas_fp32_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"
#define A_CPU
#include <ffx_a.h>
#include <ffx_fsr1.h>
namespace Vulkan {
FSR::FSR(const Device& device_, MemoryAllocator& memory_allocator_, size_t image_count_,
@ -29,11 +35,11 @@ FSR::FSR(const Device& device_, MemoryAllocator& memory_allocator_, size_t image
}
VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView image_view,
const Common::Rectangle<int>& crop_rect) {
VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect) {
UpdateDescriptorSet(image_index, image_view);
scheduler.Record([this, image_index, crop_rect](vk::CommandBuffer cmdbuf) {
scheduler.Record([this, image_index, input_image_extent, crop_rect](vk::CommandBuffer cmdbuf) {
const VkImageMemoryBarrier base_barrier{
.sType = VK_STRUCTURE_TYPE_IMAGE_MEMORY_BARRIER,
.pNext = nullptr,
@ -54,13 +60,18 @@ VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView im
},
};
// 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()),
});
std::array<AU1, 4 * 4> push_constants;
FsrEasuConOffset(
push_constants.data() + 0, push_constants.data() + 4, push_constants.data() + 8,
push_constants.data() + 12,
static_cast<AF1>(crop_rect.GetWidth()), static_cast<AF1>(crop_rect.GetHeight()),
static_cast<AF1>(input_image_extent.width), static_cast<AF1>(input_image_extent.height),
static_cast<AF1>(output_size.width), static_cast<AF1>(output_size.height),
static_cast<AF1>(crop_rect.left), static_cast<AF1>(crop_rect.top));
cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, push_constants);
{
VkImageMemoryBarrier fsr_write_barrier = base_barrier;
@ -77,7 +88,9 @@ VkImageView FSR::Draw(VKScheduler& scheduler, size_t image_index, VkImageView im
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);
FsrRcasCon(push_constants.data(), 0.25f);
cmdbuf.PushConstants(*pipeline_layout, VK_SHADER_STAGE_COMPUTE_BIT, push_constants);
{
std::array<VkImageMemoryBarrier, 2> barriers;
@ -247,7 +260,7 @@ void FSR::CreatePipelineLayout() {
VkPushConstantRange push_const{
.stageFlags = VK_SHADER_STAGE_COMPUTE_BIT,
.offset = 0,
.size = sizeof(std::array<u32, 2>),
.size = sizeof(std::array<u32, 4 * 4>),
};
VkPipelineLayoutCreateInfo ci{
.sType = VK_STRUCTURE_TYPE_PIPELINE_LAYOUT_CREATE_INFO,
@ -344,8 +357,13 @@ void FSR::CreateSampler() {
}
void FSR::CreateShaders() {
easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_COMP_SPV);
rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_RCAS_COMP_SPV);
if (device.IsFloat16Supported()) {
easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_FP16_COMP_SPV);
rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_RCAS_FP16_COMP_SPV);
} else {
easu_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_EASU_FP32_COMP_SPV);
rcas_shader = BuildShader(device, VULKAN_FIDELITYFX_FSR_RCAS_FP32_COMP_SPV);
}
}
void FSR::CreatePipeline() {

2
src/video_core/renderer_vulkan/vk_fsr.h
View file

@ -18,7 +18,7 @@ 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);
VkExtent2D input_image_extent, const Common::Rectangle<int>& crop_rect);
private:
void CreateDescriptorPool();