TextureCache: Simplify blitting of D24S8 formats and fix bugs.

This commit is contained in:
Fernando Sahmkow 2021-11-22 00:00:01 +01:00
parent d7f4434bd5
commit 8532849439
10 changed files with 73 additions and 195 deletions

View file

@ -11,13 +11,9 @@ set(SHADER_FILES
block_linear_unswizzle_2d.comp
block_linear_unswizzle_3d.comp
convert_abgr8_to_d24s8.frag
convert_b10g11r11_to_d24s8.frag
convert_d24s8_to_abgr8.frag
convert_d24s8_to_b10g11r11.frag
convert_d24s8_to_r16g16.frag
convert_depth_to_float.frag
convert_float_to_depth.frag
convert_r16g16_to_d24s8.frag
full_screen_triangle.vert
fxaa.frag
fxaa.vert

View file

@ -1,27 +0,0 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#version 450
#extension GL_ARB_shader_stencil_export : require
layout(binding = 0) uniform sampler2D color_texture;
uint conv_from_float(float value_f, uint mantissa_bits) {
uint value = floatBitsToInt(value_f);
uint exp = (value >> 23) & 0x1Fu;
uint mantissa_shift = 32u - mantissa_bits;
uint mantissa = (value << 9u) >> mantissa_shift;
return (exp << mantissa_bits) | mantissa;
}
void main() {
ivec2 coord = ivec2(gl_FragCoord.xy);
vec4 color = texelFetch(color_texture, coord, 0).rgba;
uint depth_stencil_unorm = (conv_from_float(color.r, 6u) << 21)
| (conv_from_float(color.g, 6u) << 10)
| conv_from_float(color.b, 5u);
gl_FragDepth = float(depth_stencil_unorm & 0x00FFFFFFu) / (exp2(24.0) - 1.0f);
gl_FragStencilRefARB = int(depth_stencil_unorm >> 24);
}

View file

@ -1,32 +0,0 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#version 450
layout(binding = 0) uniform sampler2D depth_tex;
layout(binding = 1) uniform isampler2D stencil_tex;
layout(location = 0) out vec4 color;
float conv_to_float(uint value, uint mantissa_bits) {
uint exp = (value >> mantissa_bits) & 0x1Fu;
uint mantissa_shift = 32u - mantissa_bits;
uint mantissa = (value << mantissa_shift) >> mantissa_shift;
return uintBitsToFloat((exp << 23) | (mantissa << (23 - mantissa_bits)));
}
void main() {
ivec2 coord = ivec2(gl_FragCoord.xy);
uint depth = uint(textureLod(depth_tex, coord, 0).r * (exp2(32.0) - 1.0f));
uint stencil = uint(textureLod(stencil_tex, coord, 0).r);
uint depth_stencil = (stencil << 24) | (depth >> 8);
uint red_int = (depth_stencil >> 21) & 0x07FF;
uint green_int = (depth_stencil >> 10) & 0x07FF;
uint blue_int = depth_stencil & 0x03FF;
color.r = conv_to_float(red_int, 6u);
color.g = conv_to_float(green_int, 6u);
color.b = conv_to_float(blue_int, 5u);
color.a = 1.0f;
}

View file

@ -1,22 +0,0 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#version 450
layout(binding = 0) uniform sampler2D depth_tex;
layout(binding = 1) uniform isampler2D stencil_tex;
layout(location = 0) out vec4 color;
void main() {
ivec2 coord = ivec2(gl_FragCoord.xy);
uint depth = uint(textureLod(depth_tex, coord, 0).r * (exp2(32.0) - 1.0f));
uint stencil = uint(textureLod(stencil_tex, coord, 0).r);
uint depth_stencil = (stencil << 24) | (depth >> 8);
color.r = float(depth_stencil & 0x0000FFFFu) / (exp2(16) - 1.0);
color.g = float(depth_stencil >> 16) / (exp2(16) - 1.0);
color.b = 0.0f;
color.a = 1.0f;
}

View file

@ -1,19 +0,0 @@
// Copyright 2021 yuzu Emulator Project
// Licensed under GPLv2 or any later version
// Refer to the license.txt file included.
#version 450
#extension GL_ARB_shader_stencil_export : require
layout(binding = 0) uniform sampler2D color_texture;
void main() {
ivec2 coord = ivec2(gl_FragCoord.xy);
vec4 color = texelFetch(color_texture, coord, 0).rgba;
uvec2 bytes = uvec2(color.rg * (exp2(16) - 1.0f)) << uvec2(0, 16);
uint depth_stencil_unorm =
uint(color.r * (exp2(16) - 1.0f)) | (uint(color.g * (exp2(16) - 1.0f)) << 16);
gl_FragDepth = float(depth_stencil_unorm & 0x00FFFFFFu) / (exp2(24.0) - 1.0f);
gl_FragStencilRefARB = int(depth_stencil_unorm >> 24);
}

View file

@ -5,13 +5,9 @@
#include <algorithm>
#include "video_core/host_shaders/convert_abgr8_to_d24s8_frag_spv.h"
#include "video_core/host_shaders/convert_b10g11r11_to_d24s8_frag_spv.h"
#include "video_core/host_shaders/convert_d24s8_to_abgr8_frag_spv.h"
#include "video_core/host_shaders/convert_d24s8_to_b10g11r11_frag_spv.h"
#include "video_core/host_shaders/convert_d24s8_to_r16g16_frag_spv.h"
#include "video_core/host_shaders/convert_depth_to_float_frag_spv.h"
#include "video_core/host_shaders/convert_float_to_depth_frag_spv.h"
#include "video_core/host_shaders/convert_r16g16_to_d24s8_frag_spv.h"
#include "video_core/host_shaders/full_screen_triangle_vert_spv.h"
#include "video_core/host_shaders/vulkan_blit_color_float_frag_spv.h"
#include "video_core/host_shaders/vulkan_blit_depth_stencil_frag_spv.h"
@ -361,11 +357,7 @@ BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_,
convert_depth_to_float_frag(BuildShader(device, CONVERT_DEPTH_TO_FLOAT_FRAG_SPV)),
convert_float_to_depth_frag(BuildShader(device, CONVERT_FLOAT_TO_DEPTH_FRAG_SPV)),
convert_abgr8_to_d24s8_frag(BuildShader(device, CONVERT_ABGR8_TO_D24S8_FRAG_SPV)),
convert_b10g11r11_to_d24s8_frag(BuildShader(device, CONVERT_B10G11R11_TO_D24S8_FRAG_SPV)),
convert_r16g16_to_d24s8_frag(BuildShader(device, CONVERT_R16G16_TO_D24S8_FRAG_SPV)),
convert_d24s8_to_abgr8_frag(BuildShader(device, CONVERT_D24S8_TO_ABGR8_FRAG_SPV)),
convert_d24s8_to_b10g11r11_frag(BuildShader(device, CONVERT_D24S8_TO_B10G11R11_FRAG_SPV)),
convert_d24s8_to_r16g16_frag(BuildShader(device, CONVERT_D24S8_TO_R16G16_FRAG_SPV)),
linear_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_LINEAR>)),
nearest_sampler(device.GetLogical().CreateSampler(SAMPLER_CREATE_INFO<VK_FILTER_NEAREST>)) {
if (device.IsExtShaderStencilExportSupported()) {
@ -461,30 +453,11 @@ void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer,
}
void BlitImageHelper::ConvertABGR8ToD24S8(const Framebuffer* dst_framebuffer,
const ImageView& src_image_view, u32 up_scale,
u32 down_shift) {
ImageView& src_image_view, u32 up_scale, u32 down_shift) {
ConvertPipelineDepthTargetEx(convert_abgr8_to_d24s8_pipeline, dst_framebuffer->RenderPass(),
convert_abgr8_to_d24s8_frag, true);
Convert(*convert_abgr8_to_d24s8_pipeline, dst_framebuffer, src_image_view, up_scale,
down_shift);
}
void BlitImageHelper::ConvertB10G11R11ToD24S8(const Framebuffer* dst_framebuffer,
const ImageView& src_image_view, u32 up_scale,
u32 down_shift) {
ConvertPipelineDepthTargetEx(convert_b10g11r11_to_d24s8_pipeline, dst_framebuffer->RenderPass(),
convert_b10g11r11_to_d24s8_frag, true);
Convert(*convert_b10g11r11_to_d24s8_pipeline, dst_framebuffer, src_image_view, up_scale,
down_shift);
}
void BlitImageHelper::ConvertR16G16ToD24S8(const Framebuffer* dst_framebuffer,
const ImageView& src_image_view, u32 up_scale,
u32 down_shift) {
ConvertPipelineDepthTargetEx(convert_r16g16_to_d24s8_pipeline, dst_framebuffer->RenderPass(),
convert_r16g16_to_d24s8_frag, true);
Convert(*convert_r16g16_to_d24s8_pipeline, dst_framebuffer, src_image_view, up_scale,
down_shift);
ConvertColor(*convert_abgr8_to_d24s8_pipeline, dst_framebuffer, src_image_view, up_scale,
down_shift);
}
void BlitImageHelper::ConvertD24S8ToABGR8(const Framebuffer* dst_framebuffer,
@ -495,24 +468,6 @@ void BlitImageHelper::ConvertD24S8ToABGR8(const Framebuffer* dst_framebuffer,
down_shift);
}
void BlitImageHelper::ConvertD24S8ToB10G11R11(const Framebuffer* dst_framebuffer,
ImageView& src_image_view, u32 up_scale,
u32 down_shift) {
ConvertPipelineColorTargetEx(convert_d24s8_to_b10g11r11_pipeline, dst_framebuffer->RenderPass(),
convert_d24s8_to_b10g11r11_frag, false);
ConvertDepthStencil(*convert_d24s8_to_b10g11r11_pipeline, dst_framebuffer, src_image_view,
up_scale, down_shift);
}
void BlitImageHelper::ConvertD24S8ToR16G16(const Framebuffer* dst_framebuffer,
ImageView& src_image_view, u32 up_scale,
u32 down_shift) {
ConvertPipelineColorTargetEx(convert_d24s8_to_r16g16_pipeline, dst_framebuffer->RenderPass(),
convert_d24s8_to_r16g16_frag, false);
ConvertDepthStencil(*convert_d24s8_to_r16g16_pipeline, dst_framebuffer, src_image_view,
up_scale, down_shift);
}
void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
const ImageView& src_image_view, u32 up_scale, u32 down_shift) {
const VkPipelineLayout layout = *one_texture_pipeline_layout;
@ -560,6 +515,53 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb
scheduler.InvalidateState();
}
void BlitImageHelper::ConvertColor(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
ImageView& src_image_view, u32 up_scale, u32 down_shift) {
const VkPipelineLayout layout = *one_texture_pipeline_layout;
const VkImageView src_view = src_image_view.ColorView();
const VkSampler sampler = *nearest_sampler;
const VkExtent2D extent{
.width = std::max((src_image_view.size.width * up_scale) >> down_shift, 1U),
.height = std::max((src_image_view.size.height * up_scale) >> down_shift, 1U),
};
scheduler.RequestRenderpass(dst_framebuffer);
scheduler.Record([pipeline, layout, sampler, src_view, extent, up_scale, down_shift,
this](vk::CommandBuffer cmdbuf) {
const VkOffset2D offset{
.x = 0,
.y = 0,
};
const VkViewport viewport{
.x = 0.0f,
.y = 0.0f,
.width = static_cast<float>(extent.width),
.height = static_cast<float>(extent.height),
.minDepth = 0.0f,
.maxDepth = 0.0f,
};
const VkRect2D scissor{
.offset = offset,
.extent = extent,
};
const PushConstants push_constants{
.tex_scale = {viewport.width, viewport.height},
.tex_offset = {0.0f, 0.0f},
};
const VkDescriptorSet descriptor_set = one_texture_descriptor_allocator.Commit();
UpdateOneTextureDescriptorSet(device, descriptor_set, sampler, src_view);
// TODO: Barriers
cmdbuf.BindPipeline(VK_PIPELINE_BIND_POINT_GRAPHICS, pipeline);
cmdbuf.BindDescriptorSets(VK_PIPELINE_BIND_POINT_GRAPHICS, layout, 0, descriptor_set,
nullptr);
cmdbuf.SetViewport(0, viewport);
cmdbuf.SetScissor(0, scissor);
cmdbuf.PushConstants(layout, VK_SHADER_STAGE_VERTEX_BIT, push_constants);
cmdbuf.Draw(3, 1, 0, 0);
});
scheduler.InvalidateState();
}
void BlitImageHelper::ConvertDepthStencil(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
ImageView& src_image_view, u32 up_scale, u32 down_shift) {
const VkPipelineLayout layout = *two_textures_pipeline_layout;

View file

@ -56,28 +56,19 @@ public:
void ConvertR16ToD16(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
u32 up_scale, u32 down_shift);
void ConvertABGR8ToD24S8(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
void ConvertABGR8ToD24S8(const Framebuffer* dst_framebuffer, ImageView& src_image_view,
u32 up_scale, u32 down_shift);
void ConvertB10G11R11ToD24S8(const Framebuffer* dst_framebuffer,
const ImageView& src_image_view, u32 up_scale, u32 down_shift);
void ConvertR16G16ToD24S8(const Framebuffer* dst_framebuffer, const ImageView& src_image_view,
u32 up_scale, u32 down_shift);
void ConvertD24S8ToABGR8(const Framebuffer* dst_framebuffer, ImageView& src_image_view,
u32 up_scale, u32 down_shift);
void ConvertD24S8ToB10G11R11(const Framebuffer* dst_framebuffer, ImageView& src_image_view,
u32 up_scale, u32 down_shift);
void ConvertD24S8ToR16G16(const Framebuffer* dst_framebuffer, ImageView& src_image_view,
u32 up_scale, u32 down_shift);
private:
void Convert(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
const ImageView& src_image_view, u32 up_scale, u32 down_shift);
void ConvertColor(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
ImageView& src_image_view, u32 up_scale, u32 down_shift);
void ConvertDepthStencil(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
ImageView& src_image_view, u32 up_scale, u32 down_shift);
@ -111,11 +102,7 @@ private:
vk::ShaderModule convert_depth_to_float_frag;
vk::ShaderModule convert_float_to_depth_frag;
vk::ShaderModule convert_abgr8_to_d24s8_frag;
vk::ShaderModule convert_b10g11r11_to_d24s8_frag;
vk::ShaderModule convert_r16g16_to_d24s8_frag;
vk::ShaderModule convert_d24s8_to_abgr8_frag;
vk::ShaderModule convert_d24s8_to_b10g11r11_frag;
vk::ShaderModule convert_d24s8_to_r16g16_frag;
vk::Sampler linear_sampler;
vk::Sampler nearest_sampler;
@ -128,11 +115,7 @@ private:
vk::Pipeline convert_d16_to_r16_pipeline;
vk::Pipeline convert_r16_to_d16_pipeline;
vk::Pipeline convert_abgr8_to_d24s8_pipeline;
vk::Pipeline convert_b10g11r11_to_d24s8_pipeline;
vk::Pipeline convert_r16g16_to_d24s8_pipeline;
vk::Pipeline convert_d24s8_to_abgr8_pipeline;
vk::Pipeline convert_d24s8_to_b10g11r11_pipeline;
vk::Pipeline convert_d24s8_to_r16g16_pipeline;
};
} // namespace Vulkan

View file

@ -1063,21 +1063,10 @@ void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, Im
}
break;
case PixelFormat::A8B8G8R8_UNORM:
case PixelFormat::B8G8R8A8_UNORM:
if (src_view.format == PixelFormat::S8_UINT_D24_UNORM) {
return blit_image_helper.ConvertD24S8ToABGR8(dst, src_view, up_scale, down_shift);
}
break;
case PixelFormat::B10G11R11_FLOAT:
if (src_view.format == PixelFormat::S8_UINT_D24_UNORM) {
return blit_image_helper.ConvertD24S8ToB10G11R11(dst, src_view, up_scale, down_shift);
}
break;
case PixelFormat::R16G16_UNORM:
if (src_view.format == PixelFormat::S8_UINT_D24_UNORM) {
return blit_image_helper.ConvertD24S8ToR16G16(dst, src_view, up_scale, down_shift);
}
break;
case PixelFormat::R32_FLOAT:
if (src_view.format == PixelFormat::D32_FLOAT) {
return blit_image_helper.ConvertD32ToR32(dst, src_view, up_scale, down_shift);
@ -1089,16 +1078,7 @@ void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, Im
}
break;
case PixelFormat::S8_UINT_D24_UNORM:
if (src_view.format == PixelFormat::A8B8G8R8_UNORM ||
src_view.format == PixelFormat::B8G8R8A8_UNORM) {
return blit_image_helper.ConvertABGR8ToD24S8(dst, src_view, up_scale, down_shift);
}
if (src_view.format == PixelFormat::B10G11R11_FLOAT) {
return blit_image_helper.ConvertB10G11R11ToD24S8(dst, src_view, up_scale, down_shift);
}
if (src_view.format == PixelFormat::R16G16_UNORM) {
return blit_image_helper.ConvertR16G16ToD24S8(dst, src_view, up_scale, down_shift);
}
return blit_image_helper.ConvertABGR8ToD24S8(dst, src_view, up_scale, down_shift);
break;
case PixelFormat::D32_FLOAT:
if (src_view.format == PixelFormat::R32_FLOAT) {
@ -1595,6 +1575,14 @@ VkImageView ImageView::StencilView() {
return *stencil_view;
}
VkImageView ImageView::ColorView() {
if (color_view) {
return *color_view;
}
color_view = MakeView(VK_FORMAT_R8G8B8A8_UNORM, VK_IMAGE_ASPECT_COLOR_BIT);
return *color_view;
}
VkImageView ImageView::StorageView(Shader::TextureType texture_type,
Shader::ImageFormat image_format) {
if (image_format == Shader::ImageFormat::Typeless) {

View file

@ -184,6 +184,8 @@ public:
[[nodiscard]] VkImageView StencilView();
[[nodiscard]] VkImageView ColorView();
[[nodiscard]] VkImageView StorageView(Shader::TextureType texture_type,
Shader::ImageFormat image_format);
@ -224,6 +226,7 @@ private:
std::unique_ptr<StorageViews> storage_views;
vk::ImageView depth_view;
vk::ImageView stencil_view;
vk::ImageView color_view;
VkImage image_handle = VK_NULL_HANDLE;
VkImageView render_target = VK_NULL_HANDLE;
VkSampleCountFlagBits samples = VK_SAMPLE_COUNT_1_BIT;

View file

@ -1781,7 +1781,13 @@ void TextureCache<P>::CopyImage(ImageId dst_id, ImageId src_id, std::vector<Imag
const SubresourceExtent src_extent{.levels = 1, .layers = 1};
const SubresourceRange dst_range{.base = dst_base, .extent = dst_extent};
const SubresourceRange src_range{.base = src_base, .extent = src_extent};
const ImageViewInfo dst_view_info(ImageViewType::e2D, dst.info.format, dst_range);
PixelFormat dst_format = dst.info.format;
if (GetFormatType(src.info.format) == SurfaceType::DepthStencil &&
GetFormatType(dst_format) == SurfaceType::ColorTexture &&
BytesPerBlock(dst_format) == 4) {
dst_format = PixelFormat::A8B8G8R8_UNORM;
}
const ImageViewInfo dst_view_info(ImageViewType::e2D, dst_format, dst_range);
const ImageViewInfo src_view_info(ImageViewType::e2D, src.info.format, src_range);
const auto [dst_framebuffer_id, dst_view_id] = RenderTargetFromImage(dst_id, dst_view_info);
Framebuffer* const dst_framebuffer = &slot_framebuffers[dst_framebuffer_id];