gl_rasterizer_cache: Update to be based on GPU addresses, not CPU addresses.

10 files changed, 122 insertions, 67 deletions

diff --git a/src/core/memory.cpp b/src/core/memory.cpp
index 291bf066f..ff0420c56 100644
--- a/src/core/memory.cpp
+++ b/src/core/memory.cpp
@@ -325,15 +325,29 @@ u8* GetPhysicalPointer(PAddr address) {
     return target_pointer;
 }
 
-void RasterizerMarkRegionCached(VAddr start, u64 size, bool cached) {
-    if (start == 0) {
+void RasterizerMarkRegionCached(Tegra::GPUVAddr gpu_addr, u64 size, bool cached) {
+    if (gpu_addr == 0) {
         return;
     }
 
-    u64 num_pages = ((start + size - 1) >> PAGE_BITS) - (start >> PAGE_BITS) + 1;
-    VAddr vaddr = start;
+    // Iterate over a contiguous CPU address space, which corresponds to the specified GPU address
+    // space, marking the region as un/cached. The region is marked un/cached at a granularity of
+    // CPU pages, hence why we iterate on a CPU page basis (note: GPU page size is different). This
+    // assumes the specified GPU address region is contiguous as well.
+
+    u64 num_pages = ((gpu_addr + size - 1) >> PAGE_BITS) - (gpu_addr >> PAGE_BITS) + 1;
+    for (unsigned i = 0; i < num_pages; ++i, gpu_addr += PAGE_SIZE) {
+        boost::optional<VAddr> maybe_vaddr =
+            Core::System::GetInstance().GPU().memory_manager->GpuToCpuAddress(gpu_addr);
+        // The GPU <-> CPU virtual memory mapping is not 1:1
+        if (!maybe_vaddr) {
+            LOG_ERROR(HW_Memory,
+                      "Trying to flush a cached region to an invalid physical address %08X",
+                      gpu_addr);
+            continue;
+        }
+        VAddr vaddr = *maybe_vaddr;
 
-    for (unsigned i = 0; i < num_pages; ++i, vaddr += PAGE_SIZE) {
         PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
 
         if (cached) {
@@ -347,6 +361,10 @@ void RasterizerMarkRegionCached(VAddr start, u64 size, bool cached) {
                 page_type = PageType::RasterizerCachedMemory;
                 current_page_table->pointers[vaddr >> PAGE_BITS] = nullptr;
                 break;
+            case PageType::RasterizerCachedMemory:
+                // There can be more than one GPU region mapped per CPU region, so it's common that
+                // this area is already marked as cached.
+                break;
             default:
                 UNREACHABLE();
             }
@@ -357,6 +375,10 @@ void RasterizerMarkRegionCached(VAddr start, u64 size, bool cached) {
                 // It is not necessary for a process to have this region mapped into its address
                 // space, for example, a system module need not have a VRAM mapping.
                 break;
+            case PageType::Memory:
+                // There can be more than one GPU region mapped per CPU region, so it's common that
+                // this area is already unmarked as cached.
+                break;
             case PageType::RasterizerCachedMemory: {
                 u8* pointer = GetPointerFromVMA(vaddr & ~PAGE_MASK);
                 if (pointer == nullptr) {
@@ -394,19 +416,29 @@ void RasterizerFlushVirtualRegion(VAddr start, u64 size, FlushMode mode) {
 
         VAddr overlap_start = std::max(start, region_start);
         VAddr overlap_end = std::min(end, region_end);
+
+        std::vector<Tegra::GPUVAddr> gpu_addresses =
+            Core::System::GetInstance().GPU().memory_manager->CpuToGpuAddress(overlap_start);
+
+        if (gpu_addresses.empty()) {
+            return;
+        }
+
         u64 overlap_size = overlap_end - overlap_start;
 
-        auto* rasterizer = VideoCore::g_renderer->Rasterizer();
-        switch (mode) {
-        case FlushMode::Flush:
-            rasterizer->FlushRegion(overlap_start, overlap_size);
-            break;
-        case FlushMode::Invalidate:
-            rasterizer->InvalidateRegion(overlap_start, overlap_size);
-            break;
-        case FlushMode::FlushAndInvalidate:
-            rasterizer->FlushAndInvalidateRegion(overlap_start, overlap_size);
-            break;
+        for (const auto& gpu_address : gpu_addresses) {
+            auto* rasterizer = VideoCore::g_renderer->Rasterizer();
+            switch (mode) {
+            case FlushMode::Flush:
+                rasterizer->FlushRegion(gpu_address, overlap_size);
+                break;
+            case FlushMode::Invalidate:
+                rasterizer->InvalidateRegion(gpu_address, overlap_size);
+                break;
+            case FlushMode::FlushAndInvalidate:
+                rasterizer->FlushAndInvalidateRegion(gpu_address, overlap_size);
+                break;
+            }
         }
     };
 
diff --git a/src/core/memory.h b/src/core/memory.h
index e9b8ca873..3f56a2c6a 100644
--- a/src/core/memory.h
+++ b/src/core/memory.h
@@ -14,6 +14,7 @@
 #include <boost/optional.hpp>
 #include "common/common_types.h"
 #include "core/memory_hook.h"
+#include "video_core/memory_manager.h"
 
 namespace Kernel {
 class Process;
@@ -258,7 +259,7 @@ enum class FlushMode {
 /**
  * Mark each page touching the region as cached.
  */
-void RasterizerMarkRegionCached(VAddr start, u64 size, bool cached);
+void RasterizerMarkRegionCached(Tegra::GPUVAddr start, u64 size, bool cached);
 
 /**
  * Flushes and invalidates any externally cached rasterizer resources touching the given virtual
diff --git a/src/video_core/memory_manager.h b/src/video_core/memory_manager.h
index 7d745101f..08140c83a 100644
--- a/src/video_core/memory_manager.h
+++ b/src/video_core/memory_manager.h
@@ -11,7 +11,6 @@
 #include <boost/optional.hpp>
 
 #include "common/common_types.h"
-#include "core/memory.h"
 
 namespace Tegra {
 
diff --git a/src/video_core/rasterizer_interface.h b/src/video_core/rasterizer_interface.h
index 36629dd11..f0e48a802 100644
--- a/src/video_core/rasterizer_interface.h
+++ b/src/video_core/rasterizer_interface.h
@@ -6,6 +6,7 @@
 
 #include "common/common_types.h"
 #include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
 
 struct ScreenInfo;
 
@@ -25,14 +26,14 @@ public:
     virtual void FlushAll() = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
-    virtual void FlushRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushRegion(Tegra::GPUVAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be invalidated
-    virtual void InvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void InvalidateRegion(Tegra::GPUVAddr addr, u64 size) = 0;
 
     /// Notify rasterizer that any caches of the specified region should be flushed to Switch memory
     /// and invalidated
-    virtual void FlushAndInvalidateRegion(VAddr addr, u64 size) = 0;
+    virtual void FlushAndInvalidateRegion(Tegra::GPUVAddr addr, u64 size) = 0;
 
     /// Attempt to use a faster method to perform a display transfer with is_texture_copy = 0
     virtual bool AccelerateDisplayTransfer(const void* config) {
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.cpp b/src/video_core/renderer_opengl/gl_rasterizer.cpp
index 71612790b..bc9368877 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer.cpp
@@ -150,9 +150,8 @@ std::pair<u8*, GLintptr> RasterizerOpenGL::SetupVertexArrays(u8* array_ptr,
         u64 size = end - start + 1;
 
         // Copy vertex array data
-        const VAddr data_addr{*memory_manager->GpuToCpuAddress(start)};
-        res_cache.FlushRegion(data_addr, size, nullptr);
-        Memory::ReadBlock(data_addr, array_ptr, size);
+        res_cache.FlushRegion(start, size, nullptr);
+        Memory::ReadBlock(*memory_manager->GpuToCpuAddress(start), array_ptr, size);
 
         // Bind the vertex array to the buffer at the current offset.
         glBindVertexBuffer(index, stream_buffer->GetHandle(), buffer_offset, vertex_array.stride);
@@ -519,17 +518,17 @@ void RasterizerOpenGL::FlushAll() {
     res_cache.FlushAll();
 }
 
-void RasterizerOpenGL::FlushRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::FlushRegion(Tegra::GPUVAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     res_cache.FlushRegion(addr, size);
 }
 
-void RasterizerOpenGL::InvalidateRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::InvalidateRegion(Tegra::GPUVAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     res_cache.InvalidateRegion(addr, size, nullptr);
 }
 
-void RasterizerOpenGL::FlushAndInvalidateRegion(VAddr addr, u64 size) {
+void RasterizerOpenGL::FlushAndInvalidateRegion(Tegra::GPUVAddr addr, u64 size) {
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
     res_cache.FlushRegion(addr, size);
     res_cache.InvalidateRegion(addr, size, nullptr);
@@ -560,7 +559,7 @@ bool RasterizerOpenGL::AccelerateDisplay(const Tegra::FramebufferConfig& framebu
     MICROPROFILE_SCOPE(OpenGL_CacheManagement);
 
     SurfaceParams src_params;
-    src_params.addr = framebuffer_addr;
+    src_params.cpu_addr = framebuffer_addr;
     src_params.width = std::min(framebuffer.width, pixel_stride);
     src_params.height = framebuffer.height;
     src_params.stride = pixel_stride;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer.h b/src/video_core/renderer_opengl/gl_rasterizer.h
index 544714b95..9709e595e 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer.h
@@ -11,6 +11,7 @@
 #include <glad/glad.h>
 #include "common/common_types.h"
 #include "video_core/engines/maxwell_3d.h"
+#include "video_core/memory_manager.h"
 #include "video_core/rasterizer_interface.h"
 #include "video_core/renderer_opengl/gl_rasterizer_cache.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
@@ -29,9 +30,9 @@ public:
     void DrawArrays() override;
     void NotifyMaxwellRegisterChanged(u32 method) override;
     void FlushAll() override;
-    void FlushRegion(VAddr addr, u64 size) override;
-    void InvalidateRegion(VAddr addr, u64 size) override;
-    void FlushAndInvalidateRegion(VAddr addr, u64 size) override;
+    void FlushRegion(Tegra::GPUVAddr addr, u64 size) override;
+    void InvalidateRegion(Tegra::GPUVAddr addr, u64 size) override;
+    void FlushAndInvalidateRegion(Tegra::GPUVAddr addr, u64 size) override;
     bool AccelerateDisplayTransfer(const void* config) override;
     bool AccelerateTextureCopy(const void* config) override;
     bool AccelerateFill(const void* config) override;
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
index ced648c12..d139d51e9 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.cpp
@@ -83,26 +83,30 @@ static u16 GetResolutionScaleFactor() {
 }
 
 template <bool morton_to_gl, PixelFormat format>
-void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, VAddr base, VAddr start,
-                VAddr end) {
+void MortonCopy(u32 stride, u32 block_height, u32 height, u8* gl_buffer, Tegra::GPUVAddr base,
+                Tegra::GPUVAddr start, Tegra::GPUVAddr end) {
     constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / 8;
     constexpr u32 gl_bytes_per_pixel = CachedSurface::GetGLBytesPerPixel(format);
+    const auto& gpu = Core::System::GetInstance().GPU();
 
     if (morton_to_gl) {
         auto data = Tegra::Texture::UnswizzleTexture(
-            base, SurfaceParams::TextureFormatFromPixelFormat(format), stride, height,
-            block_height);
+            *gpu.memory_manager->GpuToCpuAddress(base),
+            SurfaceParams::TextureFormatFromPixelFormat(format), stride, height, block_height);
         std::memcpy(gl_buffer, data.data(), data.size());
     } else {
         // TODO(bunnei): Assumes the default rendering GOB size of 16 (128 lines). We should check
         // the configuration for this and perform more generic un/swizzle
         LOG_WARNING(Render_OpenGL, "need to use correct swizzle/GOB parameters!");
-        VideoCore::MortonCopyPixels128(stride, height, bytes_per_pixel, gl_bytes_per_pixel,
-                                       Memory::GetPointer(base), gl_buffer, morton_to_gl);
+        VideoCore::MortonCopyPixels128(
+            stride, height, bytes_per_pixel, gl_bytes_per_pixel,
+            Memory::GetPointer(*gpu.memory_manager->GpuToCpuAddress(base)), gl_buffer,
+            morton_to_gl);
     }
 }
 
-static constexpr std::array<void (*)(u32, u32, u32, u8*, VAddr, VAddr, VAddr),
+static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr, Tegra::GPUVAddr,
+                                     Tegra::GPUVAddr),
                             SurfaceParams::MaxPixelFormat>
     morton_to_gl_fns = {
         MortonCopy<true, PixelFormat::ABGR8>,       MortonCopy<true, PixelFormat::B5G6R5>,
@@ -110,7 +114,8 @@ static constexpr std::array<void (*)(u32, u32, u32, u8*, VAddr, VAddr, VAddr),
         MortonCopy<true, PixelFormat::DXT23>,       MortonCopy<true, PixelFormat::DXT45>,
 };
 
-static constexpr std::array<void (*)(u32, u32, u32, u8*, VAddr, VAddr, VAddr),
+static constexpr std::array<void (*)(u32, u32, u32, u8*, Tegra::GPUVAddr, Tegra::GPUVAddr,
+                                     Tegra::GPUVAddr),
                             SurfaceParams::MaxPixelFormat>
     gl_to_morton_fns = {
         MortonCopy<false, PixelFormat::ABGR8>,
@@ -219,9 +224,9 @@ SurfaceParams SurfaceParams::FromInterval(SurfaceInterval interval) const {
     SurfaceParams params = *this;
     const u32 tiled_size = is_tiled ? 8 : 1;
     const u64 stride_tiled_bytes = BytesInPixels(stride * tiled_size);
-    VAddr aligned_start =
+    Tegra::GPUVAddr aligned_start =
         addr + Common::AlignDown(boost::icl::first(interval) - addr, stride_tiled_bytes);
-    VAddr aligned_end =
+    Tegra::GPUVAddr aligned_end =
         addr + Common::AlignUp(boost::icl::last_next(interval) - addr, stride_tiled_bytes);
 
     if (aligned_end - aligned_start > stride_tiled_bytes) {
@@ -342,6 +347,13 @@ bool SurfaceParams::CanTexCopy(const SurfaceParams& texcopy_params) const {
     return FromInterval(texcopy_params.GetInterval()).GetInterval() == texcopy_params.GetInterval();
 }
 
+VAddr SurfaceParams::GetCpuAddr() const {
+    // When this function is used, only cpu_addr or (GPU) addr should be set, not both
+    ASSERT(!(cpu_addr && addr));
+    const auto& gpu = Core::System::GetInstance().GPU();
+    return cpu_addr.get_value_or(*gpu.memory_manager->GpuToCpuAddress(addr));
+}
+
 bool CachedSurface::CanFill(const SurfaceParams& dest_surface,
                             SurfaceInterval fill_interval) const {
     if (type == SurfaceType::Fill && IsRegionValid(fill_interval) &&
@@ -456,10 +468,10 @@ void RasterizerCacheOpenGL::CopySurface(const Surface& src_surface, const Surfac
 }
 
 MICROPROFILE_DEFINE(OpenGL_SurfaceLoad, "OpenGL", "Surface Load", MP_RGB(128, 64, 192));
-void CachedSurface::LoadGLBuffer(VAddr load_start, VAddr load_end) {
+void CachedSurface::LoadGLBuffer(Tegra::GPUVAddr load_start, Tegra::GPUVAddr load_end) {
     ASSERT(type != SurfaceType::Fill);
 
-    u8* const texture_src_data = Memory::GetPointer(addr);
+    u8* const texture_src_data = Memory::GetPointer(GetCpuAddr());
     if (texture_src_data == nullptr)
         return;
 
@@ -485,8 +497,8 @@ void CachedSurface::LoadGLBuffer(VAddr load_start, VAddr load_end) {
 }
 
 MICROPROFILE_DEFINE(OpenGL_SurfaceFlush, "OpenGL", "Surface Flush", MP_RGB(128, 192, 64));
-void CachedSurface::FlushGLBuffer(VAddr flush_start, VAddr flush_end) {
-    u8* const dst_buffer = Memory::GetPointer(addr);
+void CachedSurface::FlushGLBuffer(Tegra::GPUVAddr flush_start, Tegra::GPUVAddr flush_end) {
+    u8* const dst_buffer = Memory::GetPointer(GetCpuAddr());
     if (dst_buffer == nullptr)
         return;
 
@@ -1028,7 +1040,7 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu
     auto& gpu = Core::System::GetInstance().GPU();
 
     SurfaceParams params;
-    params.addr = *gpu.memory_manager->GpuToCpuAddress(config.tic.Address());
+    params.addr = config.tic.Address();
     params.width = config.tic.Width();
     params.height = config.tic.Height();
     params.is_tiled = config.tic.IsTiled();
@@ -1045,7 +1057,7 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu
         params.block_height = config.tic.BlockHeight();
     } else {
         // Use the texture-provided stride value if the texture isn't tiled.
-        params.stride = params.PixelsInBytes(config.tic.Pitch());
+        params.stride = static_cast<u32>(params.PixelsInBytes(config.tic.Pitch()));
     }
 
     params.UpdateParams();
@@ -1073,7 +1085,6 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu
 SurfaceSurfaceRect_Tuple RasterizerCacheOpenGL::GetFramebufferSurfaces(
     bool using_color_fb, bool using_depth_fb, const MathUtil::Rectangle<s32>& viewport) {
     const auto& regs = Core::System().GetInstance().GPU().Maxwell3D().regs;
-    const auto& memory_manager = Core::System().GetInstance().GPU().memory_manager;
     const auto& config = regs.rt[0];
 
     // TODO(bunnei): This is hard corded to use just the first render buffer
@@ -1106,7 +1117,7 @@ SurfaceSurfaceRect_Tuple RasterizerCacheOpenGL::GetFramebufferSurfaces(
     color_params.block_height = Tegra::Texture::TICEntry::DefaultBlockHeight;
     SurfaceParams depth_params = color_params;
 
-    color_params.addr = *memory_manager->GpuToCpuAddress(config.Address());
+    color_params.addr = config.Address();
     color_params.pixel_format = SurfaceParams::PixelFormatFromRenderTargetFormat(config.format);
     color_params.component_type = SurfaceParams::ComponentTypeFromRenderTarget(config.format);
     color_params.UpdateParams();
@@ -1222,7 +1233,8 @@ void RasterizerCacheOpenGL::DuplicateSurface(const Surface& src_surface,
     }
 }
 
-void RasterizerCacheOpenGL::ValidateSurface(const Surface& surface, VAddr addr, u64 size) {
+void RasterizerCacheOpenGL::ValidateSurface(const Surface& surface, Tegra::GPUVAddr addr,
+                                            u64 size) {
     if (size == 0)
         return;
 
@@ -1261,7 +1273,7 @@ void RasterizerCacheOpenGL::ValidateSurface(const Surface& surface, VAddr addr,
     }
 }
 
-void RasterizerCacheOpenGL::FlushRegion(VAddr addr, u64 size, Surface flush_surface) {
+void RasterizerCacheOpenGL::FlushRegion(Tegra::GPUVAddr addr, u64 size, Surface flush_surface) {
     if (size == 0)
         return;
 
@@ -1297,7 +1309,8 @@ void RasterizerCacheOpenGL::FlushAll() {
     FlushRegion(0, Kernel::VMManager::MAX_ADDRESS);
 }
 
-void RasterizerCacheOpenGL::InvalidateRegion(VAddr addr, u64 size, const Surface& region_owner) {
+void RasterizerCacheOpenGL::InvalidateRegion(Tegra::GPUVAddr addr, u64 size,
+                                             const Surface& region_owner) {
     if (size == 0)
         return;
 
@@ -1390,7 +1403,7 @@ void RasterizerCacheOpenGL::UnregisterSurface(const Surface& surface) {
     surface_cache.subtract({surface->GetInterval(), SurfaceSet{surface}});
 }
 
-void RasterizerCacheOpenGL::UpdatePagesCachedCount(VAddr addr, u64 size, int delta) {
+void RasterizerCacheOpenGL::UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta) {
     const u64 num_pages =
         ((addr + size - 1) >> Memory::PAGE_BITS) - (addr >> Memory::PAGE_BITS) + 1;
     const u64 page_start = addr >> Memory::PAGE_BITS;
@@ -1406,8 +1419,10 @@ void RasterizerCacheOpenGL::UpdatePagesCachedCount(VAddr addr, u64 size, int del
         const auto interval = pair.first & pages_interval;
         const int count = pair.second;
 
-        const VAddr interval_start_addr = boost::icl::first(interval) << Memory::PAGE_BITS;
-        const VAddr interval_end_addr = boost::icl::last_next(interval) << Memory::PAGE_BITS;
+        const Tegra::GPUVAddr interval_start_addr = boost::icl::first(interval)
+                                                    << Memory::PAGE_BITS;
+        const Tegra::GPUVAddr interval_end_addr = boost::icl::last_next(interval)
+                                                  << Memory::PAGE_BITS;
         const u64 interval_size = interval_end_addr - interval_start_addr;
 
         if (delta > 0 && count == delta)
diff --git a/src/video_core/renderer_opengl/gl_rasterizer_cache.h b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
index bf0fabb29..5f77f4e61 100644
--- a/src/video_core/renderer_opengl/gl_rasterizer_cache.h
+++ b/src/video_core/renderer_opengl/gl_rasterizer_cache.h
@@ -17,12 +17,14 @@
 #ifdef __GNUC__
 #pragma GCC diagnostic pop
 #endif
+#include <boost/optional.hpp>
 #include <glad/glad.h>
 #include "common/assert.h"
 #include "common/common_funcs.h"
 #include "common/common_types.h"
 #include "common/math_util.h"
 #include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
 #include "video_core/renderer_opengl/gl_resource_manager.h"
 #include "video_core/textures/texture.h"
 
@@ -30,9 +32,9 @@ struct CachedSurface;
 using Surface = std::shared_ptr<CachedSurface>;
 using SurfaceSet = std::set<Surface>;
 
-using SurfaceRegions = boost::icl::interval_set<VAddr>;
-using SurfaceMap = boost::icl::interval_map<VAddr, Surface>;
-using SurfaceCache = boost::icl::interval_map<VAddr, SurfaceSet>;
+using SurfaceRegions = boost::icl::interval_set<Tegra::GPUVAddr>;
+using SurfaceMap = boost::icl::interval_map<Tegra::GPUVAddr, Surface>;
+using SurfaceCache = boost::icl::interval_map<Tegra::GPUVAddr, SurfaceSet>;
 
 using SurfaceInterval = SurfaceCache::interval_type;
 static_assert(std::is_same<SurfaceRegions::interval_type, SurfaceCache::interval_type>() &&
@@ -277,6 +279,8 @@ struct SurfaceParams {
         return pixels * GetFormatBpp(pixel_format) / CHAR_BIT;
     }
 
+    VAddr GetCpuAddr() const;
+
     bool ExactMatch(const SurfaceParams& other_surface) const;
     bool CanSubRect(const SurfaceParams& sub_surface) const;
     bool CanExpand(const SurfaceParams& expanded_surface) const;
@@ -285,8 +289,9 @@ struct SurfaceParams {
     MathUtil::Rectangle<u32> GetSubRect(const SurfaceParams& sub_surface) const;
     MathUtil::Rectangle<u32> GetScaledSubRect(const SurfaceParams& sub_surface) const;
 
-    VAddr addr = 0;
-    VAddr end = 0;
+    Tegra::GPUVAddr addr = 0;
+    Tegra::GPUVAddr end = 0;
+    boost::optional<VAddr> cpu_addr;
     u64 size = 0;
 
     u32 width = 0;
@@ -332,8 +337,8 @@ struct CachedSurface : SurfaceParams {
     size_t gl_buffer_size = 0;
 
     // Read/Write data in Switch memory to/from gl_buffer
-    void LoadGLBuffer(VAddr load_start, VAddr load_end);
-    void FlushGLBuffer(VAddr flush_start, VAddr flush_end);
+    void LoadGLBuffer(Tegra::GPUVAddr load_start, Tegra::GPUVAddr load_end);
+    void FlushGLBuffer(Tegra::GPUVAddr flush_start, Tegra::GPUVAddr flush_end);
 
     // Upload/Download data in gl_buffer in/to this surface's texture
     void UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint read_fb_handle,
@@ -381,10 +386,10 @@ public:
     SurfaceRect_Tuple GetTexCopySurface(const SurfaceParams& params);
 
     /// Write any cached resources overlapping the region back to memory (if dirty)
-    void FlushRegion(VAddr addr, u64 size, Surface flush_surface = nullptr);
+    void FlushRegion(Tegra::GPUVAddr addr, u64 size, Surface flush_surface = nullptr);
 
     /// Mark region as being invalidated by region_owner (nullptr if Switch memory)
-    void InvalidateRegion(VAddr addr, u64 size, const Surface& region_owner);
+    void InvalidateRegion(Tegra::GPUVAddr addr, u64 size, const Surface& region_owner);
 
     /// Flush all cached resources tracked by this cache manager
     void FlushAll();
@@ -393,7 +398,7 @@ private:
     void DuplicateSurface(const Surface& src_surface, const Surface& dest_surface);
 
     /// Update surface's texture for given region when necessary
-    void ValidateSurface(const Surface& surface, VAddr addr, u64 size);
+    void ValidateSurface(const Surface& surface, Tegra::GPUVAddr addr, u64 size);
 
     /// Create a new surface
     Surface CreateSurface(const SurfaceParams& params);
@@ -405,7 +410,7 @@ private:
     void UnregisterSurface(const Surface& surface);
 
     /// Increase/decrease the number of surface in pages touching the specified region
-    void UpdatePagesCachedCount(VAddr addr, u64 size, int delta);
+    void UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta);
 
     SurfaceCache surface_cache;
     PageMap cached_pages;
diff --git a/src/video_core/renderer_opengl/renderer_opengl.cpp b/src/video_core/renderer_opengl/renderer_opengl.cpp
index baff2c7af..a266e21cf 100644
--- a/src/video_core/renderer_opengl/renderer_opengl.cpp
+++ b/src/video_core/renderer_opengl/renderer_opengl.cpp
@@ -152,7 +152,8 @@ void RendererOpenGL::LoadFBToScreenInfo(const Tegra::FramebufferConfig& framebuf
         screen_info.display_texture = screen_info.texture.resource.handle;
         screen_info.display_texcoords = MathUtil::Rectangle<float>(0.f, 0.f, 1.f, 1.f);
 
-        Rasterizer()->FlushRegion(framebuffer_addr, size_in_bytes);
+        Memory::RasterizerFlushVirtualRegion(framebuffer_addr, size_in_bytes,
+                                             Memory::FlushMode::Flush);
 
         VideoCore::MortonCopyPixels128(framebuffer.width, framebuffer.height, bytes_per_pixel, 4,
                                        Memory::GetPointer(framebuffer_addr),
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index e0509f0ce..9c3ae875c 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -4,6 +4,7 @@
 
 #include <cstring>
 #include "common/assert.h"
+#include "core/memory.h"
 #include "video_core/textures/decoders.h"
 #include "video_core/textures/texture.h"