early-access version 2215

README.md

@@ -1,7 +1,7 @@
 yuzu emulator early access
 =============
 
-This is the source code for early-access 2214.
+This is the source code for early-access 2215.
 
 ## Legal Notice
 

src/video_core/host_shaders/CMakeLists.txt

@@ -10,6 +10,8 @@ set(SHADER_FILES
     astc_decoder.comp
     block_linear_unswizzle_2d.comp
     block_linear_unswizzle_3d.comp
+    convert_abgr8_to_d24s8.frag
+    convert_d24s8_to_abgr8.frag
     convert_depth_to_float.frag
     convert_float_to_depth.frag
     full_screen_triangle.vert

src/video_core/host_shaders/convert_abgr8_to_d24s8.frag

@@ -0,0 +1,17 @@
+// 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);
+    uvec4 color = uvec4(texelFetch(color_texture, coord, 0).rgba * (exp2(8) - 1.0f));
+    uint depth_unorm = (color.r << 16) | (color.g << 8) | color.b;
+
+    gl_FragDepth = float(depth_unorm) / (exp2(24.0) - 1.0f);
+    // gl_FragStencilRefARB = int(color.a);
+}

src/video_core/host_shaders/convert_d24s8_to_abgr8.frag

@@ -0,0 +1,21 @@
+// 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(24.0) - 1.0f));
+    uint stencil = uint(textureLod(stencil_tex, coord, 0).r);
+
+    color.r = float(depth >> 16) / (exp2(8) - 1.0);
+    color.g = float((depth >> 8) & 0x00FF) / (exp2(8) - 1.0);
+    color.b = float(depth & 0x00FF) / (exp2(8) - 1.0);
+    color.a = float(stencil) / (exp2(8) - 1.0);
+}

src/video_core/renderer_vulkan/blit_image.cpp

@@ -4,6 +4,8 @@
 
 #include <algorithm>
 
+#include "video_core/host_shaders/convert_abgr8_to_d24s8_frag_spv.h"
+#include "video_core/host_shaders/convert_d24s8_to_abgr8_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/full_screen_triangle_vert_spv.h"
@@ -354,6 +356,8 @@ BlitImageHelper::BlitImageHelper(const Device& device_, VKScheduler& scheduler_,
       blit_color_to_color_frag(BuildShader(device, VULKAN_BLIT_COLOR_FLOAT_FRAG_SPV)),
       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_d24s8_to_abgr8_frag(BuildShader(device, CONVERT_D24S8_TO_ABGR8_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()) {
@@ -448,6 +452,23 @@ void BlitImageHelper::ConvertR16ToD16(const Framebuffer* dst_framebuffer,
     Convert(*convert_r16_to_d16_pipeline, dst_framebuffer, src_image_view, up_scale, down_shift);
 }
 
+void BlitImageHelper::ConvertABGR8ToD24S8(const Framebuffer* dst_framebuffer,
+                                          const ImageView& src_image_view, u32 up_scale,
+                                          u32 down_shift) {
+    ConvertPipelineEx(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::ConvertD24S8ToABGR8(const Framebuffer* dst_framebuffer,
+                                          ImageView& src_image_view, u32 up_scale, u32 down_shift) {
+    ConvertPipelineEx(convert_d24s8_to_abgr8_pipeline, dst_framebuffer->RenderPass(),
+                      convert_d24s8_to_abgr8_frag, false);
+    ConvertDepthStencil(*convert_d24s8_to_abgr8_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;
@@ -495,6 +516,54 @@ void BlitImageHelper::Convert(VkPipeline pipeline, const Framebuffer* dst_frameb
     scheduler.InvalidateState();
 }
 
+void BlitImageHelper::ConvertDepthStencil(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_depth_view = src_image_view.DepthView();
+    const VkImageView src_stencil_view = src_image_view.StencilView();
+    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_depth_view, src_stencil_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 = two_textures_descriptor_allocator.Commit();
+        UpdateTwoTexturesDescriptorSet(device, descriptor_set, sampler, src_depth_view,
+                                       src_stencil_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();
+}
+
 VkPipeline BlitImageHelper::FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key) {
     const auto it = std::ranges::find(blit_color_keys, key);
     if (it != blit_color_keys.end()) {
@@ -636,4 +705,33 @@ void BlitImageHelper::ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRend
     });
 }
 
+void BlitImageHelper::ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+                                        vk::ShaderModule& module, bool single_texture) {
+    if (pipeline) {
+        return;
+    }
+    const std::array stages = MakeStages(*full_screen_vert, *module);
+    pipeline = device.GetLogical().CreateGraphicsPipeline({
+        .sType = VK_STRUCTURE_TYPE_GRAPHICS_PIPELINE_CREATE_INFO,
+        .pNext = nullptr,
+        .flags = 0,
+        .stageCount = static_cast<u32>(stages.size()),
+        .pStages = stages.data(),
+        .pVertexInputState = &PIPELINE_VERTEX_INPUT_STATE_CREATE_INFO,
+        .pInputAssemblyState = &PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO,
+        .pTessellationState = nullptr,
+        .pViewportState = &PIPELINE_VIEWPORT_STATE_CREATE_INFO,
+        .pRasterizationState = &PIPELINE_RASTERIZATION_STATE_CREATE_INFO,
+        .pMultisampleState = &PIPELINE_MULTISAMPLE_STATE_CREATE_INFO,
+        .pDepthStencilState = &PIPELINE_DEPTH_STENCIL_STATE_CREATE_INFO,
+        .pColorBlendState = &PIPELINE_COLOR_BLEND_STATE_EMPTY_CREATE_INFO,
+        .pDynamicState = &PIPELINE_DYNAMIC_STATE_CREATE_INFO,
+        .layout = single_texture ? *one_texture_pipeline_layout : *two_textures_pipeline_layout,
+        .renderPass = renderpass,
+        .subpass = 0,
+        .basePipelineHandle = VK_NULL_HANDLE,
+        .basePipelineIndex = 0,
+    });
+}
+
 } // namespace Vulkan

src/video_core/renderer_vulkan/blit_image.h

@@ -56,10 +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,
+                             u32 up_scale, u32 down_shift);
+
+    void ConvertD24S8ToABGR8(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 ConvertDepthStencil(VkPipeline pipeline, const Framebuffer* dst_framebuffer,
+                             ImageView& src_image_view, u32 up_scale, u32 down_shift);
+
     [[nodiscard]] VkPipeline FindOrEmplaceColorPipeline(const BlitImagePipelineKey& key);
 
     [[nodiscard]] VkPipeline FindOrEmplaceDepthStencilPipeline(const BlitImagePipelineKey& key);
@@ -68,6 +77,9 @@ private:
 
     void ConvertColorToDepthPipeline(vk::Pipeline& pipeline, VkRenderPass renderpass);
 
+    void ConvertPipelineEx(vk::Pipeline& pipeline, VkRenderPass renderpass,
+                           vk::ShaderModule& module, bool single_texture);
+
     const Device& device;
     VKScheduler& scheduler;
     StateTracker& state_tracker;
@@ -83,6 +95,8 @@ private:
     vk::ShaderModule blit_depth_stencil_frag;
     vk::ShaderModule convert_depth_to_float_frag;
     vk::ShaderModule convert_float_to_depth_frag;
+    vk::ShaderModule convert_abgr8_to_d24s8_frag;
+    vk::ShaderModule convert_d24s8_to_abgr8_frag;
     vk::Sampler linear_sampler;
     vk::Sampler nearest_sampler;
 
@@ -94,6 +108,8 @@ private:
     vk::Pipeline convert_r32_to_d32_pipeline;
     vk::Pipeline convert_d16_to_r16_pipeline;
     vk::Pipeline convert_r16_to_d16_pipeline;
+    vk::Pipeline convert_abgr8_to_d24s8_pipeline;
+    vk::Pipeline convert_d24s8_to_abgr8_pipeline;
 };
 
 } // namespace Vulkan

src/video_core/renderer_vulkan/vk_texture_cache.cpp

@@ -886,6 +886,12 @@ void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, Im
             return blit_image_helper.ConvertD16ToR16(dst, src_view, up_scale, down_shift);
         }
         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::R32_FLOAT:
         if (src_view.format == PixelFormat::D32_FLOAT) {
             return blit_image_helper.ConvertD32ToR32(dst, src_view, up_scale, down_shift);
@@ -896,6 +902,12 @@ void TextureCacheRuntime::ConvertImage(Framebuffer* dst, ImageView& dst_view, Im
             return blit_image_helper.ConvertR16ToD16(dst, src_view, up_scale, down_shift);
         }
         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);
+        }
+        break;
     case PixelFormat::D32_FLOAT:
         if (src_view.format == PixelFormat::R32_FLOAT) {
             return blit_image_helper.ConvertR32ToD32(dst, src_view, up_scale, down_shift);

src/video_core/texture_cache/texture_cache.h

@@ -475,6 +475,7 @@ void TextureCache<P>::BlitImage(const Tegra::Engines::Fermi2D::Surface& dst,
     const BlitImages images = GetBlitImages(dst, src);
     const ImageId dst_id = images.dst_id;
     const ImageId src_id = images.src_id;
+
     PrepareImage(src_id, false, false);
     PrepareImage(dst_id, true, false);
 
@@ -1094,12 +1095,8 @@ typename TextureCache<P>::BlitImages TextureCache<P>::GetBlitImages(
         if (GetFormatType(dst_info.format) != GetFormatType(src_info.format)) {
             continue;
         }
-        if (!dst_id) {
+        src_id = FindOrInsertImage(src_info, src_addr);
-            dst_id = InsertImage(dst_info, dst_addr, RelaxedOptions{});
+        dst_id = FindOrInsertImage(dst_info, dst_addr);
-        }
-        if (!src_id) {
-            src_id = InsertImage(src_info, src_addr, RelaxedOptions{});
-        }
     } while (has_deleted_images);
     return BlitImages{
         .dst_id = dst_id,

src/video_core/texture_cache/util.cpp

@@ -1151,18 +1151,38 @@ bool IsSubresource(const ImageInfo& candidate, const ImageBase& image, GPUVAddr
 
 void DeduceBlitImages(ImageInfo& dst_info, ImageInfo& src_info, const ImageBase* dst,
                       const ImageBase* src) {
-    if (src && GetFormatType(src->info.format) != SurfaceType::ColorTexture) {
+    if (src) {
         src_info.format = src->info.format;
+        src_info.num_samples = src->info.num_samples;
+        src_info.size = src->info.size;
     }
-    if (dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) {
+    if (dst) {
         dst_info.format = dst->info.format;
+        dst_info.num_samples = dst->info.num_samples;
+        dst_info.size = dst->info.size;
     }
     if (src && GetFormatType(src->info.format) != SurfaceType::ColorTexture) {
+        if (dst) {
+            src_info.format = dst_info.format;
+        } else {
             dst_info.format = src->info.format;
         }
+    }
     if (dst && GetFormatType(dst->info.format) != SurfaceType::ColorTexture) {
+        if (src) {
+            if (GetFormatType(src->info.format) == SurfaceType::ColorTexture) {
+                dst_info.format = src->info.format;
+            }
+        } else {
             src_info.format = dst->info.format;
         }
+    }
+    if (src_info.num_samples > 1) {
+        dst_info.format = src_info.format;
+    }
+    if (dst_info.num_samples > 1) {
+        src_info.format = dst_info.format;
+    }
 }
 
 u32 MapSizeBytes(const ImageBase& image) {