gl_rasterizer: Use GPU virtual addr when possible for framebuffers.

src/core/hle/service/nvdrv/devices/nvdisp_disp0.cpp

@@ -19,14 +19,20 @@ u32 nvdisp_disp0::ioctl(Ioctl command, const std::vector<u8>& input, std::vector
 
 void nvdisp_disp0::flip(u32 buffer_handle, u32 offset, u32 format, u32 width, u32 height,
                         u32 stride, NVFlinger::BufferQueue::BufferTransformFlags transform) {
-    VAddr addr = nvmap_dev->GetObjectAddress(buffer_handle);
+
-    NGLOG_WARNING(Service,
+    const auto& object{nvmap_dev->GetObject(buffer_handle)};
-                  "Drawing from address {:X} offset {:08X} Width {} Height {} Stride {} Format {}",
+    ASSERT(object);
-                  addr, offset, width, height, stride, format);
+    ASSERT(object->status == nvmap::Object::Status::Allocated);
 
     using PixelFormat = Tegra::FramebufferConfig::PixelFormat;
-    const Tegra::FramebufferConfig framebuffer{
+    const Tegra::FramebufferConfig framebuffer{object->cpu_addr,
-        addr, offset, width, height, stride, static_cast<PixelFormat>(format), transform};
+                                               object->gpu_addr,
+                                               offset,
+                                               width,
+                                               height,
+                                               stride,
+                                               static_cast<PixelFormat>(format),
+                                               transform};
 
     Core::System::GetInstance().perf_stats.EndGameFrame();
 

src/core/hle/service/nvdrv/devices/nvhost_as_gpu.cpp

@@ -85,8 +85,8 @@ u32 nvhost_as_gpu::Remap(const std::vector<u8>& input, std::vector<u8>& output)
         u64 size = static_cast<u64>(entry.pages) << 0x10;
         ASSERT(size <= object->size);
 
-        Tegra::GPUVAddr returned = gpu.memory_manager->MapBufferEx(object->addr, offset, size);
+        object->gpu_addr = gpu.memory_manager->MapBufferEx(object->cpu_addr, offset, size);
-        ASSERT(returned == offset);
+        ASSERT(object->gpu_addr == offset);
     }
     std::memcpy(output.data(), entries.data(), output.size());
     return 0;
@@ -122,10 +122,12 @@ u32 nvhost_as_gpu::MapBufferEx(const std::vector<u8>& input, std::vector<u8>& ou
     auto& gpu = Core::System::GetInstance().GPU();
 
     if (params.flags & 1) {
-        params.offset = gpu.memory_manager->MapBufferEx(object->addr, params.offset, object->size);
+        object->gpu_addr =
+            gpu.memory_manager->MapBufferEx(object->cpu_addr, params.offset, object->size);
     } else {
-        params.offset = gpu.memory_manager->MapBufferEx(object->addr, object->size);
+        object->gpu_addr = gpu.memory_manager->MapBufferEx(object->cpu_addr, object->size);
     }
+    params.offset = object->gpu_addr;
 
     // Create a new mapping entry for this operation.
     ASSERT_MSG(buffer_mappings.find(params.offset) == buffer_mappings.end(),

src/core/hle/service/nvdrv/devices/nvmap.cpp

@@ -11,13 +11,6 @@
 
 namespace Service::Nvidia::Devices {
 
-VAddr nvmap::GetObjectAddress(u32 handle) const {
-    auto object = GetObject(handle);
-    ASSERT(object);
-    ASSERT(object->status == Object::Status::Allocated);
-    return object->addr;
-}
-
 u32 nvmap::ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) {
     switch (static_cast<IoctlCommand>(command.raw)) {
     case IoctlCommand::Create:
@@ -70,7 +63,7 @@ u32 nvmap::IocAlloc(const std::vector<u8>& input, std::vector<u8>& output) {
     object->flags = params.flags;
     object->align = params.align;
     object->kind = params.kind;
-    object->addr = params.addr;
+    object->cpu_addr = params.addr;
     object->status = Object::Status::Allocated;
 
     NGLOG_DEBUG(Service_NVDRV, "called, addr={:X}", params.addr);

src/core/hle/service/nvdrv/devices/nvmap.h

@@ -11,6 +11,7 @@
 #include "common/common_types.h"
 #include "common/swap.h"
 #include "core/hle/service/nvdrv/devices/nvdevice.h"
+#include "video_core/memory_manager.h"
 
 namespace Service::Nvidia::Devices {
 
@@ -19,9 +20,6 @@ public:
     nvmap() = default;
     ~nvmap() override = default;
 
-    /// Returns the allocated address of an nvmap object given its handle.
-    VAddr GetObjectAddress(u32 handle) const;
-
     u32 ioctl(Ioctl command, const std::vector<u8>& input, std::vector<u8>& output) override;
 
     /// Represents an nvmap object.
@@ -32,7 +30,8 @@ public:
         u32 flags;
         u32 align;
         u8 kind;
-        VAddr addr;
+        VAddr cpu_addr;
+        Tegra::GPUVAddr gpu_addr;
         Status status;
         u32 refcount;
     };

src/video_core/gpu.h

@@ -48,7 +48,8 @@ struct FramebufferConfig {
         UNREACHABLE();
     }
 
-    VAddr address;
+    VAddr cpu_addr;
+    Tegra::GPUVAddr gpu_addr;
     u32 offset;
     u32 width;
     u32 height;

src/video_core/rasterizer_interface.h

@@ -51,8 +51,9 @@ public:
     }
 
     /// Attempt to use a faster method to display the framebuffer to screen
-    virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config, VAddr framebuffer_addr,
+    virtual bool AccelerateDisplay(const Tegra::FramebufferConfig& config,
-                                   u32 pixel_stride, ScreenInfo& screen_info) {
+                                   Tegra::GPUVAddr framebuffer_addr, u32 pixel_stride,
+                                   ScreenInfo& screen_info) {
         return false;
     }
 

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -482,7 +482,7 @@ bool RasterizerOpenGL::AccelerateFill(const void* config) {
 }
 
 bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& config,
-                                         VAddr framebuffer_addr, u32 pixel_stride,
+                                         Tegra::GPUVAddr framebuffer_addr, u32 pixel_stride,
                                          ScreenInfo& screen_info) {
     if (!framebuffer_addr) {
         return {};

src/video_core/renderer_opengl/gl_rasterizer.h

@@ -36,8 +36,9 @@ public:
     bool AccelerateDisplayTransfer(const void* config) override;
     bool AccelerateTextureCopy(const void* config) override;
     bool AccelerateFill(const void* config) override;
-    bool AccelerateDisplay(const Tegra::FramebufferConfig& framebuffer, VAddr framebuffer_addr,
+    bool AccelerateDisplay(const Tegra::FramebufferConfig& framebuffer,
-                           u32 pixel_stride, ScreenInfo& screen_info) override;
+                           Tegra::GPUVAddr framebuffer_addr, u32 pixel_stride,
+                           ScreenInfo& screen_info) override;
     bool AccelerateDrawBatch(bool is_indexed) override;
 
     /// OpenGL shader generated for a given Maxwell register state

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -33,8 +33,10 @@ struct FormatTuple {
 /*static*/ SurfaceParams SurfaceParams::CreateForTexture(
     const Tegra::Texture::FullTextureInfo& config) {
 
+    const auto& gpu{Core::System::GetInstance().GPU()};
     SurfaceParams params{};
     params.addr = config.tic.Address();
+    params.cpu_addr = gpu.memory_manager->GpuToCpuAddress(params.addr).get_value_or(0);
     params.is_tiled = config.tic.IsTiled();
     params.block_height = params.is_tiled ? config.tic.BlockHeight() : 0,
     params.pixel_format = PixelFormatFromTextureFormat(config.tic.format);
@@ -50,8 +52,10 @@ struct FormatTuple {
 /*static*/ SurfaceParams SurfaceParams::CreateForFramebuffer(
     const Tegra::Engines::Maxwell3D::Regs::RenderTargetConfig& config) {
 
+    const auto& gpu{Core::System::GetInstance().GPU()};
     SurfaceParams params{};
     params.addr = config.Address();
+    params.cpu_addr = *gpu.memory_manager->GpuToCpuAddress(params.addr);
     params.is_tiled = true;
     params.block_height = Tegra::Texture::TICEntry::DefaultBlockHeight;
     params.pixel_format = PixelFormatFromRenderTargetFormat(config.format);
@@ -97,11 +101,6 @@ static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType
     return {};
 }
 
-VAddr SurfaceParams::GetCpuAddr() const {
-    const auto& gpu = Core::System::GetInstance().GPU();
-    return *gpu.memory_manager->GpuToCpuAddress(addr);
-}
-
 static bool IsPixelFormatASTC(PixelFormat format) {
     switch (format) {
     case PixelFormat::ASTC_2D_4X4:
@@ -228,7 +227,7 @@ MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 64
 void CachedSurface::LoadGLBuffer() {
     ASSERT(params.type != SurfaceType::Fill);
 
-    u8* const texture_src_data = Memory::GetPointer(params.GetCpuAddr());
+    u8* const texture_src_data = Memory::GetPointer(params.cpu_addr);
 
     ASSERT(texture_src_data);
 
@@ -254,7 +253,7 @@ void CachedSurface::LoadGLBuffer() {
 
 MICROPROFILE_DEFINE(OpenGL_SurfaceFlush, "OpenGL", "Surface Flush", MP_RGB(128, 192, 64));
 void CachedSurface::FlushGLBuffer() {
-    u8* const dst_buffer = Memory::GetPointer(params.GetCpuAddr());
+    u8* const dst_buffer = Memory::GetPointer(params.cpu_addr);
 
     ASSERT(dst_buffer);
     ASSERT(gl_buffer.size() ==
@@ -450,7 +449,7 @@ void RasterizerCacheOpenGL::MarkSurfaceAsDirty(const Surface& surface) {
 }
 
 Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params) {
-    if (params.addr == 0 || params.height * params.width == 0) {
+    if (params.cpu_addr == 0 || params.height * params.width == 0) {
         return {};
     }
 
@@ -473,19 +472,16 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params) {
     return surface;
 }
 
-Surface RasterizerCacheOpenGL::TryFindFramebufferSurface(VAddr cpu_addr) const {
+Surface RasterizerCacheOpenGL::TryFindFramebufferSurface(Tegra::GPUVAddr gpu_addr) const {
-    // Tries to find the GPU address of a framebuffer based on the CPU address. This is because
+    // Tries to find a framebuffer based on a GPU address. We iterate through all cached
-    // final output framebuffers are specified by CPU address, but internally our GPU cache uses
+    // framebuffers, and compare their starting GPU address to the one provided. This is obviously
-    // GPU addresses. We iterate through all cached framebuffers, and compare their starting CPU
+    // not great, and won't work if the framebuffer overlaps surfaces.
-    // address to the one provided. This is obviously not great, and won't work if the
-    // framebuffer overlaps surfaces.
 
     std::vector<Surface> surfaces;
     for (const auto& surface : surface_cache) {
         const auto& params = surface.second->GetSurfaceParams();
-        const VAddr surface_cpu_addr = params.GetCpuAddr();
+        if (gpu_addr >= params.addr && gpu_addr < (params.addr + params.size_in_bytes)) {
-        if (cpu_addr >= surface_cpu_addr && cpu_addr < (surface_cpu_addr + params.size_in_bytes)) {
+            ASSERT_MSG(gpu_addr == params.addr, "overlapping surfaces are unsupported");
-            ASSERT_MSG(cpu_addr == surface_cpu_addr, "overlapping surfaces are unsupported");
             surfaces.push_back(surface.second);
         }
     }

src/video_core/renderer_opengl/gl_rasterizer_cache.h

@@ -256,9 +256,6 @@ struct SurfaceParams {
                GetFormatBpp(pixel_format) / CHAR_BIT;
     }
 
-    /// Returns the CPU virtual address for this surface
-    VAddr GetCpuAddr() const;
-
     /// Returns true if the specified region overlaps with this surface's region in Switch memory
     bool IsOverlappingRegion(Tegra::GPUVAddr region_addr, size_t region_size) const {
         return addr <= (region_addr + region_size) && region_addr <= (addr + size_in_bytes);
@@ -272,6 +269,7 @@ struct SurfaceParams {
         const Tegra::Engines::Maxwell3D::Regs::RenderTargetConfig& config);
 
     Tegra::GPUVAddr addr;
+    VAddr cpu_addr;
     bool is_tiled;
     u32 block_height;
     PixelFormat pixel_format;
@@ -350,7 +348,7 @@ public:
     void MarkSurfaceAsDirty(const Surface& surface);
 
     /// Tries to find a framebuffer GPU address based on the provided CPU address
-    Surface TryFindFramebufferSurface(VAddr cpu_addr) const;
+    Surface TryFindFramebufferSurface(Tegra::GPUVAddr gpu_addr) const;
 
     /// Write any cached resources overlapping the region back to memory (if dirty)
     void FlushRegion(Tegra::GPUVAddr addr, size_t size);

src/video_core/renderer_opengl/renderer_opengl.cpp

@@ -137,8 +137,6 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
                                         ScreenInfo& screen_info) {
     const u32 bytes_per_pixel{Tegra::FramebufferConfig::BytesPerPixel(framebuffer.pixel_format)};
     const u64 size_in_bytes{framebuffer.stride * framebuffer.height * bytes_per_pixel};
-    const VAddr framebuffer_addr{framebuffer.address + framebuffer.offset};
-
     // Framebuffer orientation handling
     framebuffer_transform_flags = framebuffer.transform_flags;
 
@@ -146,17 +144,18 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
     // only allows rows to have a memory alignement of 4.
     ASSERT(framebuffer.stride % 4 == 0);
 
-    if (!Rasterizer()->AccelerateDisplay(framebuffer, framebuffer_addr, framebuffer.stride,
+    if (!Rasterizer()->AccelerateDisplay(framebuffer, framebuffer.gpu_addr + framebuffer.offset,
-                                         screen_info)) {
+                                         framebuffer.stride, screen_info)) {
+
         // Reset the screen info's display texture to its own permanent texture
         screen_info.display_texture = screen_info.texture.resource.handle;
 
-        Memory::RasterizerFlushVirtualRegion(framebuffer_addr, size_in_bytes,
+        const VAddr cpu_addr{framebuffer.cpu_addr + framebuffer.offset};
-                                             Memory::FlushMode::Flush);
+        Memory::RasterizerFlushVirtualRegion(cpu_addr, size_in_bytes, Memory::FlushMode::Flush);
 
         VideoCore::MortonCopyPixels128(framebuffer.width, framebuffer.height, bytes_per_pixel, 4,
-                                       Memory::GetPointer(framebuffer_addr),
+                                       Memory::GetPointer(cpu_addr), gl_framebuffer_data.data(),
-                                       gl_framebuffer_data.data(), true);
+                                       true);
 
         state.texture_units[0].texture_2d = screen_info.texture.resource.handle;
         state.Apply();