texture_cache: Fermi2D reform and implement View Mirage

This also does some fixes on compressed textures reinterpret and on the
Fermi2D engine in general.

src/video_core/engines/fermi_2d.cpp

@@ -4,7 +4,6 @@
 
 #include "common/assert.h"
 #include "common/logging/log.h"
-#include "common/math_util.h"
 #include "video_core/engines/fermi_2d.h"
 #include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
@@ -35,7 +34,7 @@ void Fermi2D::HandleSurfaceCopy() {
                 static_cast<u32>(regs.operation));
 
     // TODO(Subv): Only raw copies are implemented.
-    ASSERT(regs.operation == Regs::Operation::SrcCopy);
+    ASSERT(regs.operation == Operation::SrcCopy);
 
     const u32 src_blit_x1{static_cast<u32>(regs.blit_src_x >> 32)};
     const u32 src_blit_y1{static_cast<u32>(regs.blit_src_y >> 32)};
@@ -48,8 +47,13 @@ void Fermi2D::HandleSurfaceCopy() {
     const Common::Rectangle<u32> dst_rect{regs.blit_dst_x, regs.blit_dst_y,
                                           regs.blit_dst_x + regs.blit_dst_width,
                                           regs.blit_dst_y + regs.blit_dst_height};
+    Config copy_config;
+    copy_config.operation = regs.operation;
+    copy_config.filter = regs.blit_control.filter;
+    copy_config.src_rect = src_rect;
+    copy_config.dst_rect = dst_rect;
 
-    if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst, src_rect, dst_rect)) {
+    if (!rasterizer.AccelerateSurfaceCopy(regs.src, regs.dst, copy_config)) {
         UNIMPLEMENTED();
     }
 }

src/video_core/engines/fermi_2d.h

@@ -9,6 +9,7 @@
 #include "common/bit_field.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "common/math_util.h"
 #include "video_core/gpu.h"
 
 namespace Tegra {
@@ -38,6 +39,26 @@ public:
     /// Write the value to the register identified by method.
     void CallMethod(const GPU::MethodCall& method_call);
 
+    enum class Origin : u32 {
+        Center = 0,
+        Corner = 1,
+    };
+
+    enum class Filter : u32 {
+        PointSample = 0, // Nearest
+        Linear = 1,
+    };
+
+    enum class Operation : u32 {
+        SrcCopyAnd = 0,
+        ROPAnd = 1,
+        Blend = 2,
+        SrcCopy = 3,
+        ROP = 4,
+        SrcCopyPremult = 5,
+        BlendPremult = 6,
+    };
+
     struct Regs {
         static constexpr std::size_t NUM_REGS = 0x258;
 
@@ -76,16 +97,6 @@ public:
         };
         static_assert(sizeof(Surface) == 0x28, "Surface has incorrect size");
 
-        enum class Operation : u32 {
-            SrcCopyAnd = 0,
-            ROPAnd = 1,
-            Blend = 2,
-            SrcCopy = 3,
-            ROP = 4,
-            SrcCopyPremult = 5,
-            BlendPremult = 6,
-        };
-
         union {
             struct {
                 INSERT_PADDING_WORDS(0x80);
@@ -102,7 +113,11 @@ public:
 
                 INSERT_PADDING_WORDS(0x177);
 
-                u32 blit_control;
+                union {
+                    u32 raw;
+                    BitField<0, 1, Origin> origin;
+                    BitField<4, 1, Filter> filter;
+                } blit_control;
 
                 INSERT_PADDING_WORDS(0x8);
 
@@ -121,6 +136,13 @@ public:
         };
     } regs{};
 
+    struct Config {
+        Operation operation;
+        Filter filter;
+        Common::Rectangle<u32> src_rect;
+        Common::Rectangle<u32> dst_rect;
+    };
+
 private:
     VideoCore::RasterizerInterface& rasterizer;
     MemoryManager& memory_manager;

src/video_core/rasterizer_interface.h

@@ -52,8 +52,7 @@ public:
     /// Attempt to use a faster method to perform a surface copy
     virtual bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                        const Tegra::Engines::Fermi2D::Regs::Surface& dst,
-                                       const Common::Rectangle<u32>& src_rect,
+                                       const Tegra::Engines::Fermi2D::Config& copy_config) {
-                                       const Common::Rectangle<u32>& dst_rect) {
         return false;
     }
 

src/video_core/renderer_opengl/gl_framebuffer_cache.cpp

@@ -37,7 +37,7 @@ OGLFramebuffer FramebufferCacheOpenGL::CreateFramebuffer(const FramebufferCacheK
 
     if (key.is_single_buffer) {
         if (key.color_attachments[0] != GL_NONE && key.colors[0]) {
-            key.colors[0]->Attach(key.color_attachments[0]);
+            key.colors[0]->Attach(key.color_attachments[0], GL_DRAW_FRAMEBUFFER);
             glDrawBuffer(key.color_attachments[0]);
         } else {
             glDrawBuffer(GL_NONE);
@@ -45,14 +45,16 @@ OGLFramebuffer FramebufferCacheOpenGL::CreateFramebuffer(const FramebufferCacheK
     } else {
         for (std::size_t index = 0; index < Maxwell::NumRenderTargets; ++index) {
             if (key.colors[index]) {
-                key.colors[index]->Attach(GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index));
+                key.colors[index]->Attach(GL_COLOR_ATTACHMENT0 + static_cast<GLenum>(index),
+                                          GL_DRAW_FRAMEBUFFER);
             }
         }
         glDrawBuffers(key.colors_count, key.color_attachments.data());
     }
 
     if (key.zeta) {
-        key.zeta->Attach(key.stencil_enable ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT);
+        key.zeta->Attach(key.stencil_enable ? GL_DEPTH_STENCIL_ATTACHMENT : GL_DEPTH_ATTACHMENT,
+                         GL_DRAW_FRAMEBUFFER);
     }
 
     return framebuffer;

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -696,10 +696,9 @@ void RasterizerOpenGL::FlushAndInvalidateRegion(CacheAddr addr, u64 size) {
 
 bool RasterizerOpenGL::AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                              const Tegra::Engines::Fermi2D::Regs::Surface& dst,
-                                             const Common::Rectangle<u32>& src_rect,
+                                             const Tegra::Engines::Fermi2D::Config& copy_config) {
-                                             const Common::Rectangle<u32>& dst_rect) {
     MICROPROFILE_SCOPE(OpenGL_Blits);
-    texture_cache.DoFermiCopy(src, dst, src_rect, dst_rect);
+    texture_cache.DoFermiCopy(src, dst, copy_config);
     return true;
 }
 

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -67,8 +67,7 @@ public:
     void FlushAndInvalidateRegion(CacheAddr addr, u64 size) override;
     bool AccelerateSurfaceCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src,
                                const Tegra::Engines::Fermi2D::Regs::Surface& dst,
-                               const Common::Rectangle<u32>& src_rect,
+                               const Tegra::Engines::Fermi2D::Config& copy_config) override;
-                               const Common::Rectangle<u32>& dst_rect) override;
     bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
                            u32 pixel_stride) override;
     bool AccelerateDrawBatch(bool is_indexed) override;

src/video_core/renderer_opengl/gl_texture_cache.cpp

@@ -378,26 +378,26 @@ CachedSurfaceView::CachedSurfaceView(CachedSurface& surface, const ViewParams& p
 
 CachedSurfaceView::~CachedSurfaceView() = default;
 
-void CachedSurfaceView::Attach(GLenum attachment) const {
+void CachedSurfaceView::Attach(GLenum attachment, GLenum target) const {
     ASSERT(params.num_layers == 1 && params.num_levels == 1);
 
     const auto& owner_params = surface.GetSurfaceParams();
 
     switch (owner_params.target) {
     case SurfaceTarget::Texture1D:
-        glFramebufferTexture1D(GL_DRAW_FRAMEBUFFER, attachment, surface.GetTarget(),
+        glFramebufferTexture1D(target, attachment, surface.GetTarget(), surface.GetTexture(),
-                               surface.GetTexture(), params.base_level);
+                               params.base_level);
         break;
     case SurfaceTarget::Texture2D:
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, attachment, surface.GetTarget(),
+        glFramebufferTexture2D(target, attachment, surface.GetTarget(), surface.GetTexture(),
-                               surface.GetTexture(), params.base_level);
+                               params.base_level);
         break;
     case SurfaceTarget::Texture1DArray:
     case SurfaceTarget::Texture2DArray:
     case SurfaceTarget::TextureCubemap:
     case SurfaceTarget::TextureCubeArray:
-        glFramebufferTextureLayer(GL_DRAW_FRAMEBUFFER, attachment, surface.GetTexture(),
+        glFramebufferTextureLayer(target, attachment, surface.GetTexture(), params.base_level,
-                                  params.base_level, params.base_layer);
+                                  params.base_layer);
         break;
     default:
         UNIMPLEMENTED();
@@ -460,11 +460,10 @@ void TextureCacheOpenGL::ImageCopy(Surface src_surface, Surface dst_surface,
                        copy_params.depth);
 }
 
-void TextureCacheOpenGL::ImageBlit(Surface src_surface, Surface dst_surface,
+void TextureCacheOpenGL::ImageBlit(View src_view, View dst_view,
-                                   const Common::Rectangle<u32>& src_rect,
+                                   const Tegra::Engines::Fermi2D::Config& copy_config) {
-                                   const Common::Rectangle<u32>& dst_rect) {
+    const auto& src_params{src_view->GetSurfaceParams()};
-    const auto& src_params{src_surface->GetSurfaceParams()};
+    const auto& dst_params{dst_view->GetSurfaceParams()};
-    const auto& dst_params{dst_surface->GetSurfaceParams()};
 
     OpenGLState prev_state{OpenGLState::GetCurState()};
     SCOPE_EXIT({ prev_state.Apply(); });
@@ -476,51 +475,46 @@ void TextureCacheOpenGL::ImageBlit(Surface src_surface, Surface dst_surface,
 
     u32 buffers{};
 
-    UNIMPLEMENTED_IF(src_params.target != SurfaceTarget::Texture2D);
+    UNIMPLEMENTED_IF(src_params.target == SurfaceTarget::Texture3D);
-    UNIMPLEMENTED_IF(dst_params.target != SurfaceTarget::Texture2D);
+    UNIMPLEMENTED_IF(dst_params.target == SurfaceTarget::Texture3D);
-
-    const GLuint src_texture{src_surface->GetTexture()};
-    const GLuint dst_texture{dst_surface->GetTexture()};
 
     if (src_params.type == SurfaceType::ColorTexture) {
-        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D,
+        src_view->Attach(GL_COLOR_ATTACHMENT0, GL_READ_FRAMEBUFFER);
-                               src_texture, 0);
         glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
                                0);
 
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D,
+        dst_view->Attach(GL_COLOR_ATTACHMENT0, GL_DRAW_FRAMEBUFFER);
-                               dst_texture, 0);
         glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0,
                                0);
 
         buffers = GL_COLOR_BUFFER_BIT;
     } else if (src_params.type == SurfaceType::Depth) {
         glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
-        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, src_texture,
+        src_view->Attach(GL_DEPTH_ATTACHMENT, GL_READ_FRAMEBUFFER);
-                               0);
         glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
 
         glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_ATTACHMENT, GL_TEXTURE_2D, dst_texture,
+        dst_view->Attach(GL_DEPTH_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
-                               0);
         glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_STENCIL_ATTACHMENT, GL_TEXTURE_2D, 0, 0);
 
         buffers = GL_DEPTH_BUFFER_BIT;
     } else if (src_params.type == SurfaceType::DepthStencil) {
         glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
-        glFramebufferTexture2D(GL_READ_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+        src_view->Attach(GL_DEPTH_STENCIL_ATTACHMENT, GL_READ_FRAMEBUFFER);
-                               src_texture, 0);
 
         glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_COLOR_ATTACHMENT0, GL_TEXTURE_2D, 0, 0);
-        glFramebufferTexture2D(GL_DRAW_FRAMEBUFFER, GL_DEPTH_STENCIL_ATTACHMENT, GL_TEXTURE_2D,
+        dst_view->Attach(GL_DEPTH_STENCIL_ATTACHMENT, GL_DRAW_FRAMEBUFFER);
-                               dst_texture, 0);
 
         buffers = GL_DEPTH_BUFFER_BIT | GL_STENCIL_BUFFER_BIT;
     }
 
+    const Common::Rectangle<u32>& src_rect = copy_config.src_rect;
+    const Common::Rectangle<u32>& dst_rect = copy_config.dst_rect;
+    const bool is_linear = copy_config.filter == Tegra::Engines::Fermi2D::Filter::Linear;
+
     glBlitFramebuffer(src_rect.left, src_rect.top, src_rect.right, src_rect.bottom, dst_rect.left,
                       dst_rect.top, dst_rect.right, dst_rect.bottom, buffers,
-                      buffers == GL_COLOR_BUFFER_BIT ? GL_LINEAR : GL_NEAREST);
+                      is_linear ? GL_LINEAR : GL_NEAREST);
 }
 
 } // namespace OpenGL

src/video_core/renderer_opengl/gl_texture_cache.h

@@ -73,7 +73,7 @@ public:
     ~CachedSurfaceView();
 
     /// Attaches this texture view to the current bound GL_DRAW_FRAMEBUFFER
-    void Attach(GLenum attachment) const;
+    void Attach(GLenum attachment, GLenum target) const;
 
     GLuint GetTexture() {
         if (is_proxy) {
@@ -138,8 +138,8 @@ protected:
     void ImageCopy(Surface src_surface, Surface dst_surface,
                    const VideoCommon::CopyParams& copy_params) override;
 
-    void ImageBlit(Surface src_surface, Surface dst_surface, const Common::Rectangle<u32>& src_rect,
+    void ImageBlit(View src_view, View dst_view,
-                   const Common::Rectangle<u32>& dst_rect) override;
+                   const Tegra::Engines::Fermi2D::Config& copy_config) override;
 
 private:
     OGLFramebuffer src_framebuffer;

src/video_core/texture_cache/surface_base.h

@@ -126,14 +126,19 @@ public:
             return MatchStructureResult::None;
         }
         // Tiled surface
-        if (std::tie(params.height, params.depth, params.block_width, params.block_height,
+        if (std::tie(params.depth, params.block_width, params.block_height, params.block_depth,
-                     params.block_depth, params.tile_width_spacing, params.num_levels) ==
+                     params.tile_width_spacing, params.num_levels) ==
-            std::tie(rhs.height, rhs.depth, rhs.block_width, rhs.block_height, rhs.block_depth,
+            std::tie(rhs.depth, rhs.block_width, rhs.block_height, rhs.block_depth,
                      rhs.tile_width_spacing, rhs.num_levels)) {
-            if (params.width == rhs.width) {
+            if (std::tie(params.width, params.height) == std::tie(rhs.width, rhs.height)) {
                 return MatchStructureResult::FullMatch;
             }
-            if (params.GetBlockAlignedWidth() == rhs.GetBlockAlignedWidth()) {
+            const u32 ws = SurfaceParams::ConvertWidth(rhs.GetBlockAlignedWidth(),
+                                                       params.pixel_format, rhs.pixel_format);
+            const u32 hs =
+                SurfaceParams::ConvertHeight(rhs.height, params.pixel_format, rhs.pixel_format);
+            const u32 w1 = params.GetBlockAlignedWidth();
+            if (std::tie(w1, params.height) == std::tie(ws, hs)) {
                 return MatchStructureResult::SemiMatch;
             }
         }

src/video_core/texture_cache/surface_params.h

@@ -126,6 +126,20 @@ public:
     /// Returns the size of a layer in bytes in host memory for a given mipmap level.
     std::size_t GetHostLayerSize(u32 level) const;
 
+    static u32 ConvertWidth(u32 width, VideoCore::Surface::PixelFormat pixel_format_from,
+                            VideoCore::Surface::PixelFormat pixel_format_to) {
+        const u32 bw1 = VideoCore::Surface::GetDefaultBlockWidth(pixel_format_from);
+        const u32 bw2 = VideoCore::Surface::GetDefaultBlockWidth(pixel_format_to);
+        return (width * bw2 + bw1 - 1) / bw1;
+    }
+
+    static u32 ConvertHeight(u32 height, VideoCore::Surface::PixelFormat pixel_format_from,
+                             VideoCore::Surface::PixelFormat pixel_format_to) {
+        const u32 bh1 = VideoCore::Surface::GetDefaultBlockHeight(pixel_format_from);
+        const u32 bh2 = VideoCore::Surface::GetDefaultBlockHeight(pixel_format_to);
+        return (height * bh2 + bh1 - 1) / bh1;
+    }
+
     /// Returns the default block width.
     u32 GetDefaultBlockWidth() const {
         return VideoCore::Surface::GetDefaultBlockWidth(pixel_format);

src/video_core/texture_cache/texture_cache.h

@@ -141,11 +141,6 @@ public:
             return {};
         }
 
-        if (regs.color_mask[index].raw == 0) {
-            SetEmptyColorBuffer(index);
-            return {};
-        }
-
         const auto& config{regs.rt[index]};
         const auto gpu_addr{config.Address()};
         if (!gpu_addr) {
@@ -192,11 +187,11 @@ public:
 
     void DoFermiCopy(const Tegra::Engines::Fermi2D::Regs::Surface& src_config,
                      const Tegra::Engines::Fermi2D::Regs::Surface& dst_config,
-                     const Common::Rectangle<u32>& src_rect,
+                     const Tegra::Engines::Fermi2D::Config& copy_config) {
-                     const Common::Rectangle<u32>& dst_rect) {
+        std::pair<TSurface, TView> dst_surface = GetFermiSurface(dst_config);
-        TSurface dst_surface = GetFermiSurface(dst_config);
+        std::pair<TSurface, TView> src_surface = GetFermiSurface(src_config);
-        ImageBlit(GetFermiSurface(src_config), dst_surface, src_rect, dst_rect);
+        ImageBlit(src_surface.second, dst_surface.second, copy_config);
-        dst_surface->MarkAsModified(true, Tick());
+        dst_surface.first->MarkAsModified(true, Tick());
     }
 
     TSurface TryFindFramebufferSurface(const u8* host_ptr) {
@@ -234,8 +229,8 @@ protected:
     virtual void ImageCopy(TSurface src_surface, TSurface dst_surface,
                            const CopyParams& copy_params) = 0;
 
-    virtual void ImageBlit(TSurface src, TSurface dst, const Common::Rectangle<u32>& src_rect,
+    virtual void ImageBlit(TView src_view, TView dst_view,
-                           const Common::Rectangle<u32>& dst_rect) = 0;
+                           const Tegra::Engines::Fermi2D::Config& copy_config) = 0;
 
     void Register(TSurface surface) {
         std::lock_guard lock{mutex};
@@ -282,10 +277,11 @@ protected:
         return new_surface;
     }
 
-    TSurface GetFermiSurface(const Tegra::Engines::Fermi2D::Regs::Surface& config) {
+    std::pair<TSurface, TView> GetFermiSurface(
+        const Tegra::Engines::Fermi2D::Regs::Surface& config) {
         SurfaceParams params = SurfaceParams::CreateForFermiCopySurface(config);
         const GPUVAddr gpu_addr = config.Address();
-        return GetSurface(gpu_addr, params, true).first;
+        return GetSurface(gpu_addr, params, true);
     }
 
     Core::System& system;
@@ -551,7 +547,21 @@ private:
             if (view.has_value()) {
                 const bool is_mirage = !current_surface->MatchFormat(params.pixel_format);
                 if (is_mirage) {
-                    LOG_CRITICAL(HW_GPU, "Mirage View Unsupported");
+                    // On a mirage view, we need to recreate the surface under this new view
+                    // and then obtain a view again.
+                    SurfaceParams new_params = current_surface->GetSurfaceParams();
+                    const u32 wh = SurfaceParams::ConvertWidth(
+                        new_params.width, new_params.pixel_format, params.pixel_format);
+                    const u32 hh = SurfaceParams::ConvertHeight(
+                        new_params.height, new_params.pixel_format, params.pixel_format);
+                    new_params.width = wh;
+                    new_params.height = hh;
+                    new_params.pixel_format = params.pixel_format;
+                    std::pair<TSurface, TView> pair = RebuildSurface(current_surface, new_params);
+                    std::optional<TView> mirage_view =
+                        pair.first->EmplaceView(params, gpu_addr, candidate_size);
+                    if (mirage_view)
+                        return {pair.first, *mirage_view};
                     return RecycleSurface(overlaps, params, gpu_addr, preserve_contents, false);
                 }
                 return {current_surface, *view};