GPU: Partially implemented the Fermi2D surface copy operation.

The hardware allows for some rather complicated operations to be performed on the data during the copy, this is not implemented.
Only same-format same-size raw copies are implemented for now.

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;
+            }
         }
     };
 
@@ -625,6 +657,10 @@ void CopyBlock(const Kernel::Process& process, VAddr dest_addr, VAddr src_addr,
     }
 }
 
+void CopyBlock(VAddr dest_addr, VAddr src_addr, size_t size) {
+    CopyBlock(*Core::CurrentProcess(), dest_addr, src_addr, size);
+}
+
 boost::optional<PAddr> TryVirtualToPhysicalAddress(const VAddr addr) {
     if (addr == 0) {
         return 0;

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

src/video_core/command_processor.cpp

@@ -24,41 +24,18 @@ namespace Tegra {
 
 enum class BufferMethods {
     BindObject = 0,
-    SetGraphMacroCode = 0x45,
-    SetGraphMacroCodeArg = 0x46,
-    SetGraphMacroEntry = 0x47,
-    CountBufferMethods = 0x100,
+    CountBufferMethods = 0x40,
 };
 
 void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params) {
-    LOG_WARNING(HW_GPU, "Processing method %08X on subchannel %u value %08X remaining params %u",
-                method, subchannel, value, remaining_params);
-
-    if (method == static_cast<u32>(BufferMethods::SetGraphMacroEntry)) {
-        // Prepare to upload a new macro, reset the upload counter.
-        LOG_DEBUG(HW_GPU, "Uploading GPU macro %08X", value);
-        current_macro_entry = value;
-        current_macro_code.clear();
-        return;
-    }
-
-    if (method == static_cast<u32>(BufferMethods::SetGraphMacroCodeArg)) {
-        // Append a new code word to the current macro.
-        current_macro_code.push_back(value);
-
-        // There are no more params remaining, submit the code to the 3D engine.
-        if (remaining_params == 0) {
-            maxwell_3d->SubmitMacroCode(current_macro_entry, std::move(current_macro_code));
-            current_macro_entry = InvalidGraphMacroEntry;
-            current_macro_code.clear();
-        }
-
-        return;
-    }
+    NGLOG_WARNING(HW_GPU,
+                  "Processing method {:08X} on subchannel {} value "
+                  "{:08X} remaining params {}",
+                  method, subchannel, value, remaining_params);
 
     if (method == static_cast<u32>(BufferMethods::BindObject)) {
         // Bind the current subchannel to the desired engine id.
-        LOG_DEBUG(HW_GPU, "Binding subchannel %u to engine %u", subchannel, value);
+        NGLOG_DEBUG(HW_GPU, "Binding subchannel {} to engine {}", subchannel, value);
         ASSERT(bound_engines.find(subchannel) == bound_engines.end());
         bound_engines[subchannel] = static_cast<EngineID>(value);
         return;
@@ -66,7 +43,7 @@ void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params)
 
     if (method < static_cast<u32>(BufferMethods::CountBufferMethods)) {
         // TODO(Subv): Research and implement these methods.
-        LOG_ERROR(HW_GPU, "Special buffer methods other than Bind are not implemented");
+        NGLOG_ERROR(HW_GPU, "Special buffer methods other than Bind are not implemented");
         return;
     }
 
@@ -90,11 +67,9 @@ void GPU::WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params)
 }
 
 void GPU::ProcessCommandList(GPUVAddr address, u32 size) {
-    // TODO(Subv): PhysicalToVirtualAddress is a misnomer, it converts a GPU VAddr into an
-    // application VAddr.
-    const VAddr head_address = memory_manager->PhysicalToVirtualAddress(address);
-    VAddr current_addr = head_address;
-    while (current_addr < head_address + size * sizeof(CommandHeader)) {
+    const boost::optional<VAddr> head_address = memory_manager->GpuToCpuAddress(address);
+    VAddr current_addr = *head_address;
+    while (current_addr < *head_address + size * sizeof(CommandHeader)) {
         const CommandHeader header = {Memory::Read32(current_addr)};
         current_addr += sizeof(u32);
 

src/video_core/engines/fermi_2d.cpp

@@ -2,12 +2,71 @@
 // Licensed under GPLv2 or any later version
 // Refer to the license.txt file included.
 
+#include "core/memory.h"
 #include "video_core/engines/fermi_2d.h"
+#include "video_core/textures/decoders.h"
 
 namespace Tegra {
 namespace Engines {
 
-void Fermi2D::WriteReg(u32 method, u32 value) {}
+Fermi2D::Fermi2D(MemoryManager& memory_manager) : memory_manager(memory_manager) {}
+
+void Fermi2D::WriteReg(u32 method, u32 value) {
+    ASSERT_MSG(method < Regs::NUM_REGS,
+               "Invalid Fermi2D register, increase the size of the Regs structure");
+
+    regs.reg_array[method] = value;
+
+    switch (method) {
+    case FERMI2D_REG_INDEX(trigger): {
+        HandleSurfaceCopy();
+        break;
+    }
+    }
+}
+
+void Fermi2D::HandleSurfaceCopy() {
+    NGLOG_WARNING(HW_GPU, "Requested a surface copy with operation {}",
+                  static_cast<u32>(regs.operation));
+
+    const GPUVAddr source = regs.src.Address();
+    const GPUVAddr dest = regs.dst.Address();
+
+    // TODO(Subv): Only same-format and same-size copies are allowed for now.
+    ASSERT(regs.src.format == regs.dst.format);
+    ASSERT(regs.src.width * regs.src.height == regs.dst.width * regs.dst.height);
+
+    // TODO(Subv): Only raw copies are implemented.
+    ASSERT(regs.operation == Regs::Operation::SrcCopy);
+
+    const VAddr source_cpu = *memory_manager.GpuToCpuAddress(source);
+    const VAddr dest_cpu = *memory_manager.GpuToCpuAddress(dest);
+
+    u32 src_bytes_per_pixel = RenderTargetBytesPerPixel(regs.src.format);
+    u32 dst_bytes_per_pixel = RenderTargetBytesPerPixel(regs.dst.format);
+
+    if (regs.src.linear == regs.dst.linear) {
+        // If the input layout and the output layout are the same, just perform a raw copy.
+        Memory::CopyBlock(dest_cpu, source_cpu,
+                          src_bytes_per_pixel * regs.dst.width * regs.dst.height);
+        return;
+    }
+
+    u8* src_buffer = Memory::GetPointer(source_cpu);
+    u8* dst_buffer = Memory::GetPointer(dest_cpu);
+
+    if (!regs.src.linear && regs.dst.linear) {
+        // If the input is tiled and the output is linear, deswizzle the input and copy it over.
+        Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel,
+                                  dst_bytes_per_pixel, src_buffer, dst_buffer, true,
+                                  regs.src.block_height);
+    } else {
+        // If the input is linear and the output is tiled, swizzle the input and copy it over.
+        Texture::CopySwizzledData(regs.src.width, regs.src.height, src_bytes_per_pixel,
+                                  dst_bytes_per_pixel, dst_buffer, src_buffer, false,
+                                  regs.dst.block_height);
+    }
+}
 
 } // namespace Engines
 } // namespace Tegra

src/video_core/engines/fermi_2d.h

@@ -4,19 +4,106 @@
 
 #pragma once
 
+#include <array>
+#include "common/assert.h"
+#include "common/bit_field.h"
+#include "common/common_funcs.h"
 #include "common/common_types.h"
+#include "video_core/gpu.h"
+#include "video_core/memory_manager.h"
 
 namespace Tegra {
 namespace Engines {
 
+#define FERMI2D_REG_INDEX(field_name)                                                              \
+    (offsetof(Tegra::Engines::Fermi2D::Regs, field_name) / sizeof(u32))
+
 class Fermi2D final {
 public:
-    Fermi2D() = default;
+    explicit Fermi2D(MemoryManager& memory_manager);
     ~Fermi2D() = default;
 
     /// Write the value to the register identified by method.
     void WriteReg(u32 method, u32 value);
+
+    struct Regs {
+        static constexpr size_t NUM_REGS = 0x258;
+
+        struct Surface {
+            RenderTargetFormat format;
+            BitField<0, 1, u32> linear;
+            union {
+                BitField<0, 4, u32> block_depth;
+                BitField<4, 4, u32> block_height;
+                BitField<8, 4, u32> block_width;
+            };
+            u32 depth;
+            u32 layer;
+            u32 pitch;
+            u32 width;
+            u32 height;
+            u32 address_high;
+            u32 address_low;
+
+            GPUVAddr Address() const {
+                return static_cast<GPUVAddr>((static_cast<GPUVAddr>(address_high) << 32) |
+                                             address_low);
+            }
+        };
+        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);
+
+                Surface dst;
+
+                INSERT_PADDING_WORDS(2);
+
+                Surface src;
+
+                INSERT_PADDING_WORDS(0x15);
+
+                Operation operation;
+
+                INSERT_PADDING_WORDS(0x9);
+
+                // TODO(Subv): This is only a guess.
+                u32 trigger;
+
+                INSERT_PADDING_WORDS(0x1A3);
+            };
+            std::array<u32, NUM_REGS> reg_array;
+        };
+    } regs{};
+
+    MemoryManager& memory_manager;
+
+private:
+    /// Performs the copy from the source surface to the destination surface as configured in the
+    /// registers.
+    void HandleSurfaceCopy();
 };
 
+#define ASSERT_REG_POSITION(field_name, position)                                                  \
+    static_assert(offsetof(Fermi2D::Regs, field_name) == position * 4,                             \
+                  "Field " #field_name " has invalid position")
+
+ASSERT_REG_POSITION(dst, 0x80);
+ASSERT_REG_POSITION(src, 0x8C);
+ASSERT_REG_POSITION(operation, 0xAB);
+ASSERT_REG_POSITION(trigger, 0xB5);
+#undef ASSERT_REG_POSITION
+
 } // namespace Engines
 } // namespace Tegra

src/video_core/engines/maxwell_3d.cpp

@@ -22,10 +22,6 @@ constexpr u32 MacroRegistersStart = 0xE00;
 Maxwell3D::Maxwell3D(MemoryManager& memory_manager)
     : memory_manager(memory_manager), macro_interpreter(*this) {}
 
-void Maxwell3D::SubmitMacroCode(u32 entry, std::vector<u32> code) {
-    uploaded_macros[entry * 2 + MacroRegistersStart] = std::move(code);
-}
-
 void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
     auto macro_code = uploaded_macros.find(method);
     // The requested macro must have been uploaded already.
@@ -37,9 +33,6 @@ void Maxwell3D::CallMacroMethod(u32 method, std::vector<u32> parameters) {
 }
 
 void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
-    ASSERT_MSG(method < Regs::NUM_REGS,
-               "Invalid Maxwell3D register, increase the size of the Regs structure");
-
     auto debug_context = Core::System::GetInstance().GetGPUDebugContext();
 
     // It is an error to write to a register other than the current macro's ARG register before it
@@ -68,6 +61,9 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
         return;
     }
 
+    ASSERT_MSG(method < Regs::NUM_REGS,
+               "Invalid Maxwell3D register, increase the size of the Regs structure");
+
     if (debug_context) {
         debug_context->OnEvent(Tegra::DebugContext::Event::MaxwellCommandLoaded, nullptr);
     }
@@ -75,6 +71,10 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
     regs.reg_array[method] = value;
 
     switch (method) {
+    case MAXWELL3D_REG_INDEX(macros.data): {
+        ProcessMacroUpload(value);
+        break;
+    }
     case MAXWELL3D_REG_INDEX(code_address.code_address_high):
     case MAXWELL3D_REG_INDEX(code_address.code_address_low): {
         // Note: For some reason games (like Puyo Puyo Tetris) seem to write 0 to the CODE_ADDRESS
@@ -141,11 +141,17 @@ void Maxwell3D::WriteReg(u32 method, u32 value, u32 remaining_params) {
     }
 }
 
+void Maxwell3D::ProcessMacroUpload(u32 data) {
+    // Store the uploaded macro code to interpret them when they're called.
+    auto& macro = uploaded_macros[regs.macros.entry * 2 + MacroRegistersStart];
+    macro.push_back(data);
+}
+
 void Maxwell3D::ProcessQueryGet() {
     GPUVAddr sequence_address = regs.query.QueryAddress();
     // Since the sequence address is given as a GPU VAddr, we have to convert it to an application
     // VAddr before writing.
-    VAddr address = memory_manager.PhysicalToVirtualAddress(sequence_address);
+    boost::optional<VAddr> address = memory_manager.GpuToCpuAddress(sequence_address);
 
     // TODO(Subv): Support the other query units.
     ASSERT_MSG(regs.query.query_get.unit == Regs::QueryUnit::Crop,
@@ -153,7 +159,7 @@ void Maxwell3D::ProcessQueryGet() {
     ASSERT_MSG(regs.query.query_get.short_query,
                "Writing the entire query result structure is unimplemented");
 
-    u32 value = Memory::Read32(address);
+    u32 value = Memory::Read32(*address);
     u32 result = 0;
 
     // TODO(Subv): Support the other query variables
@@ -173,7 +179,7 @@ void Maxwell3D::ProcessQueryGet() {
     case Regs::QueryMode::Write2: {
         // Write the current query sequence to the sequence address.
         u32 sequence = regs.query.query_sequence;
-        Memory::Write32(address, sequence);
+        Memory::Write32(*address, sequence);
 
         // TODO(Subv): Write the proper query response structure to the address when not using short
         // mode.
@@ -186,8 +192,8 @@ void Maxwell3D::ProcessQueryGet() {
 }
 
 void Maxwell3D::DrawArrays() {
-    LOG_DEBUG(HW_GPU, "called, topology=%d, count=%d", regs.draw.topology.Value(),
-              regs.vertex_buffer.count);
+    NGLOG_DEBUG(HW_GPU, "called, topology={}, count={}",
+                static_cast<u32>(regs.draw.topology.Value()), regs.vertex_buffer.count);
     ASSERT_MSG(!(regs.index_array.count && regs.vertex_buffer.count), "Both indexed and direct?");
 
     auto debug_context = Core::System::GetInstance().GetGPUDebugContext();
@@ -225,10 +231,10 @@ void Maxwell3D::ProcessCBData(u32 value) {
     // Don't allow writing past the end of the buffer.
     ASSERT(regs.const_buffer.cb_pos + sizeof(u32) <= regs.const_buffer.cb_size);
 
-    VAddr address =
-        memory_manager.PhysicalToVirtualAddress(buffer_address + regs.const_buffer.cb_pos);
+    boost::optional<VAddr> address =
+        memory_manager.GpuToCpuAddress(buffer_address + regs.const_buffer.cb_pos);
 
-    Memory::Write32(address, value);
+    Memory::Write32(*address, value);
 
     // Increment the current buffer position.
     regs.const_buffer.cb_pos = regs.const_buffer.cb_pos + 4;
@@ -238,10 +244,10 @@ Texture::TICEntry Maxwell3D::GetTICEntry(u32 tic_index) const {
     GPUVAddr tic_base_address = regs.tic.TICAddress();
 
     GPUVAddr tic_address_gpu = tic_base_address + tic_index * sizeof(Texture::TICEntry);
-    VAddr tic_address_cpu = memory_manager.PhysicalToVirtualAddress(tic_address_gpu);
+    boost::optional<VAddr> tic_address_cpu = memory_manager.GpuToCpuAddress(tic_address_gpu);
 
     Texture::TICEntry tic_entry;
-    Memory::ReadBlock(tic_address_cpu, &tic_entry, sizeof(Texture::TICEntry));
+    Memory::ReadBlock(*tic_address_cpu, &tic_entry, sizeof(Texture::TICEntry));
 
     ASSERT_MSG(tic_entry.header_version == Texture::TICHeaderVersion::BlockLinear ||
                    tic_entry.header_version == Texture::TICHeaderVersion::Pitch,
@@ -268,10 +274,10 @@ Texture::TSCEntry Maxwell3D::GetTSCEntry(u32 tsc_index) const {
     GPUVAddr tsc_base_address = regs.tsc.TSCAddress();
 
     GPUVAddr tsc_address_gpu = tsc_base_address + tsc_index * sizeof(Texture::TSCEntry);
-    VAddr tsc_address_cpu = memory_manager.PhysicalToVirtualAddress(tsc_address_gpu);
+    boost::optional<VAddr> tsc_address_cpu = memory_manager.GpuToCpuAddress(tsc_address_gpu);
 
     Texture::TSCEntry tsc_entry;
-    Memory::ReadBlock(tsc_address_cpu, &tsc_entry, sizeof(Texture::TSCEntry));
+    Memory::ReadBlock(*tsc_address_cpu, &tsc_entry, sizeof(Texture::TSCEntry));
     return tsc_entry;
 }
 
@@ -293,7 +299,7 @@ std::vector<Texture::FullTextureInfo> Maxwell3D::GetStageTextures(Regs::ShaderSt
          current_texture < tex_info_buffer_end; current_texture += sizeof(Texture::TextureHandle)) {
 
         Texture::TextureHandle tex_handle{
-            Memory::Read32(memory_manager.PhysicalToVirtualAddress(current_texture))};
+            Memory::Read32(*memory_manager.GpuToCpuAddress(current_texture))};
 
         Texture::FullTextureInfo tex_info{};
         // TODO(Subv): Use the shader to determine which textures are actually accessed.

src/video_core/engines/maxwell_3d.h

@@ -31,7 +31,7 @@ public:
     /// Register structure of the Maxwell3D engine.
     /// TODO(Subv): This structure will need to be made bigger as more registers are discovered.
     struct Regs {
-        static constexpr size_t NUM_REGS = 0xE36;
+        static constexpr size_t NUM_REGS = 0xE00;
 
         static constexpr size_t NumRenderTargets = 8;
         static constexpr size_t NumViewports = 16;
@@ -322,7 +322,15 @@ public:
 
         union {
             struct {
-                INSERT_PADDING_WORDS(0x200);
+                INSERT_PADDING_WORDS(0x45);
+
+                struct {
+                    INSERT_PADDING_WORDS(1);
+                    u32 data;
+                    u32 entry;
+                } macros;
+
+                INSERT_PADDING_WORDS(0x1B8);
 
                 struct {
                     u32 address_high;
@@ -605,7 +613,7 @@ public:
                     u32 size[MaxShaderStage];
                 } tex_info_buffers;
 
-                INSERT_PADDING_WORDS(0x102);
+                INSERT_PADDING_WORDS(0xCC);
             };
             std::array<u32, NUM_REGS> reg_array;
         };
@@ -637,9 +645,6 @@ public:
     /// Write the value to the register identified by method.
     void WriteReg(u32 method, u32 value, u32 remaining_params);
 
-    /// Uploads the code for a GPU macro program associated with the specified entry.
-    void SubmitMacroCode(u32 entry, std::vector<u32> code);
-
     /// Returns a list of enabled textures for the specified shader stage.
     std::vector<Texture::FullTextureInfo> GetStageTextures(Regs::ShaderStage stage) const;
 
@@ -670,6 +675,9 @@ private:
      */
     void CallMacroMethod(u32 method, std::vector<u32> parameters);
 
+    /// Handles writes to the macro uploading registers.
+    void ProcessMacroUpload(u32 data);
+
     /// Handles a write to the QUERY_GET register.
     void ProcessQueryGet();
 
@@ -687,6 +695,7 @@ private:
     static_assert(offsetof(Maxwell3D::Regs, field_name) == position * 4,                           \
                   "Field " #field_name " has invalid position")
 
+ASSERT_REG_POSITION(macros, 0x45);
 ASSERT_REG_POSITION(rt, 0x200);
 ASSERT_REG_POSITION(viewport_transform[0], 0x280);
 ASSERT_REG_POSITION(viewport, 0x300);

src/video_core/gpu.cpp

@@ -12,7 +12,7 @@ namespace Tegra {
 GPU::GPU() {
     memory_manager = std::make_unique<MemoryManager>();
     maxwell_3d = std::make_unique<Engines::Maxwell3D>(*memory_manager);
-    fermi_2d = std::make_unique<Engines::Fermi2D>();
+    fermi_2d = std::make_unique<Engines::Fermi2D>(*memory_manager);
     maxwell_compute = std::make_unique<Engines::MaxwellCompute>();
 }
 
@@ -22,4 +22,16 @@ const Tegra::Engines::Maxwell3D& GPU::Get3DEngine() const {
     return *maxwell_3d;
 }
 
+u32 RenderTargetBytesPerPixel(RenderTargetFormat format) {
+    ASSERT(format != RenderTargetFormat::NONE);
+
+    switch (format) {
+    case RenderTargetFormat::RGBA8_UNORM:
+    case RenderTargetFormat::RGB10_A2_UNORM:
+        return 4;
+    default:
+        UNIMPLEMENTED_MSG("Unimplemented render target format %u", static_cast<u32>(format));
+    }
+}
+
 } // namespace Tegra

src/video_core/gpu.h

@@ -21,6 +21,9 @@ enum class RenderTargetFormat : u32 {
     RGBA8_SRGB = 0xD6,
 };
 
+/// Returns the number of bytes per pixel of each rendertarget format.
+u32 RenderTargetBytesPerPixel(RenderTargetFormat format);
+
 class DebugContext;
 
 /**
@@ -86,8 +89,6 @@ public:
     }
 
 private:
-    static constexpr u32 InvalidGraphMacroEntry = 0xFFFFFFFF;
-
     /// Writes a single register in the engine bound to the specified subchannel
     void WriteReg(u32 method, u32 subchannel, u32 value, u32 remaining_params);
 
@@ -100,11 +101,6 @@ private:
     std::unique_ptr<Engines::Fermi2D> fermi_2d;
     /// Compute engine
     std::unique_ptr<Engines::MaxwellCompute> maxwell_compute;
-
-    /// Entry of the macro that is currently being uploaded
-    u32 current_macro_entry = InvalidGraphMacroEntry;
-    /// Code being uploaded for the current macro
-    std::vector<u32> current_macro_code;
 };
 
 } // namespace Tegra

src/video_core/memory_manager.cpp

@@ -8,90 +8,112 @@
 
 namespace Tegra {
 
-PAddr MemoryManager::AllocateSpace(u64 size, u64 align) {
-    boost::optional<PAddr> paddr = FindFreeBlock(size, align);
-    ASSERT(paddr);
+GPUVAddr MemoryManager::AllocateSpace(u64 size, u64 align) {
+    boost::optional<GPUVAddr> gpu_addr = FindFreeBlock(size, align);
+    ASSERT(gpu_addr);
 
     for (u64 offset = 0; offset < size; offset += PAGE_SIZE) {
-        ASSERT(PageSlot(*paddr + offset) == static_cast<u64>(PageStatus::Unmapped));
-        PageSlot(*paddr + offset) = static_cast<u64>(PageStatus::Allocated);
+        ASSERT(PageSlot(*gpu_addr + offset) == static_cast<u64>(PageStatus::Unmapped));
+        PageSlot(*gpu_addr + offset) = static_cast<u64>(PageStatus::Allocated);
     }
 
-    return *paddr;
+    return *gpu_addr;
 }
 
-PAddr MemoryManager::AllocateSpace(PAddr paddr, u64 size, u64 align) {
+GPUVAddr MemoryManager::AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align) {
     for (u64 offset = 0; offset < size; offset += PAGE_SIZE) {
-        ASSERT(PageSlot(paddr + offset) == static_cast<u64>(PageStatus::Unmapped));
-        PageSlot(paddr + offset) = static_cast<u64>(PageStatus::Allocated);
+        ASSERT(PageSlot(gpu_addr + offset) == static_cast<u64>(PageStatus::Unmapped));
+        PageSlot(gpu_addr + offset) = static_cast<u64>(PageStatus::Allocated);
     }
 
-    return paddr;
+    return gpu_addr;
 }
 
-PAddr MemoryManager::MapBufferEx(VAddr vaddr, u64 size) {
-    boost::optional<PAddr> paddr = FindFreeBlock(size, PAGE_SIZE);
-    ASSERT(paddr);
+GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, u64 size) {
+    boost::optional<GPUVAddr> gpu_addr = FindFreeBlock(size, PAGE_SIZE);
+    ASSERT(gpu_addr);
 
     for (u64 offset = 0; offset < size; offset += PAGE_SIZE) {
-        ASSERT(PageSlot(*paddr + offset) == static_cast<u64>(PageStatus::Unmapped));
-        PageSlot(*paddr + offset) = vaddr + offset;
+        ASSERT(PageSlot(*gpu_addr + offset) == static_cast<u64>(PageStatus::Unmapped));
+        PageSlot(*gpu_addr + offset) = cpu_addr + offset;
     }
 
-    return *paddr;
+    MappedRegion region{cpu_addr, *gpu_addr, size};
+    mapped_regions.push_back(region);
+
+    return *gpu_addr;
 }
 
-PAddr MemoryManager::MapBufferEx(VAddr vaddr, PAddr paddr, u64 size) {
-    ASSERT((paddr & PAGE_MASK) == 0);
+GPUVAddr MemoryManager::MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size) {
+    ASSERT((gpu_addr & PAGE_MASK) == 0);
 
     for (u64 offset = 0; offset < size; offset += PAGE_SIZE) {
-        ASSERT(PageSlot(paddr + offset) == static_cast<u64>(PageStatus::Allocated));
-        PageSlot(paddr + offset) = vaddr + offset;
+        ASSERT(PageSlot(gpu_addr + offset) == static_cast<u64>(PageStatus::Allocated));
+        PageSlot(gpu_addr + offset) = cpu_addr + offset;
     }
 
-    return paddr;
+    MappedRegion region{cpu_addr, gpu_addr, size};
+    mapped_regions.push_back(region);
+
+    return gpu_addr;
 }
 
-boost::optional<PAddr> MemoryManager::FindFreeBlock(u64 size, u64 align) {
-    PAddr paddr = 0;
+boost::optional<GPUVAddr> MemoryManager::FindFreeBlock(u64 size, u64 align) {
+    GPUVAddr gpu_addr = 0;
     u64 free_space = 0;
     align = (align + PAGE_MASK) & ~PAGE_MASK;
 
-    while (paddr + free_space < MAX_ADDRESS) {
-        if (!IsPageMapped(paddr + free_space)) {
+    while (gpu_addr + free_space < MAX_ADDRESS) {
+        if (!IsPageMapped(gpu_addr + free_space)) {
             free_space += PAGE_SIZE;
             if (free_space >= size) {
-                return paddr;
+                return gpu_addr;
             }
         } else {
-            paddr += free_space + PAGE_SIZE;
+            gpu_addr += free_space + PAGE_SIZE;
             free_space = 0;
-            paddr = Common::AlignUp(paddr, align);
+            gpu_addr = Common::AlignUp(gpu_addr, align);
         }
     }
 
     return {};
 }
 
-VAddr MemoryManager::PhysicalToVirtualAddress(PAddr paddr) {
-    VAddr base_addr = PageSlot(paddr);
+boost::optional<VAddr> MemoryManager::GpuToCpuAddress(GPUVAddr gpu_addr) {
+    VAddr base_addr = PageSlot(gpu_addr);
     ASSERT(base_addr != static_cast<u64>(PageStatus::Unmapped));
-    return base_addr + (paddr & PAGE_MASK);
+
+    if (base_addr == static_cast<u64>(PageStatus::Allocated)) {
+        return {};
+    }
+
+    return base_addr + (gpu_addr & PAGE_MASK);
 }
 
-bool MemoryManager::IsPageMapped(PAddr paddr) {
-    return PageSlot(paddr) != static_cast<u64>(PageStatus::Unmapped);
+std::vector<GPUVAddr> MemoryManager::CpuToGpuAddress(VAddr cpu_addr) const {
+    std::vector<GPUVAddr> results;
+    for (const auto& region : mapped_regions) {
+        if (cpu_addr >= region.cpu_addr && cpu_addr < (region.cpu_addr + region.size)) {
+            u64 offset = cpu_addr - region.cpu_addr;
+            results.push_back(region.gpu_addr + offset);
+        }
+    }
+    return results;
 }
 
-VAddr& MemoryManager::PageSlot(PAddr paddr) {
-    auto& block = page_table[(paddr >> (PAGE_BITS + PAGE_TABLE_BITS)) & PAGE_TABLE_MASK];
+bool MemoryManager::IsPageMapped(GPUVAddr gpu_addr) {
+    return PageSlot(gpu_addr) != static_cast<u64>(PageStatus::Unmapped);
+}
+
+VAddr& MemoryManager::PageSlot(GPUVAddr gpu_addr) {
+    auto& block = page_table[(gpu_addr >> (PAGE_BITS + PAGE_TABLE_BITS)) & PAGE_TABLE_MASK];
     if (!block) {
         block = std::make_unique<PageBlock>();
         for (unsigned index = 0; index < PAGE_BLOCK_SIZE; index++) {
             (*block)[index] = static_cast<u64>(PageStatus::Unmapped);
         }
     }
-    return (*block)[(paddr >> PAGE_BITS) & PAGE_BLOCK_MASK];
+    return (*block)[(gpu_addr >> PAGE_BITS) & PAGE_BLOCK_MASK];
 }
 
 } // namespace Tegra

src/video_core/memory_manager.h

@@ -6,8 +6,11 @@
 
 #include <array>
 #include <memory>
+#include <vector>
+
+#include <boost/optional.hpp>
+
 #include "common/common_types.h"
-#include "core/memory.h"
 
 namespace Tegra {
 
@@ -18,20 +21,21 @@ class MemoryManager final {
 public:
     MemoryManager() = default;
 
-    PAddr AllocateSpace(u64 size, u64 align);
-    PAddr AllocateSpace(PAddr paddr, u64 size, u64 align);
-    PAddr MapBufferEx(VAddr vaddr, u64 size);
-    PAddr MapBufferEx(VAddr vaddr, PAddr paddr, u64 size);
-    VAddr PhysicalToVirtualAddress(PAddr paddr);
+    GPUVAddr AllocateSpace(u64 size, u64 align);
+    GPUVAddr AllocateSpace(GPUVAddr gpu_addr, u64 size, u64 align);
+    GPUVAddr MapBufferEx(VAddr cpu_addr, u64 size);
+    GPUVAddr MapBufferEx(VAddr cpu_addr, GPUVAddr gpu_addr, u64 size);
+    boost::optional<VAddr> GpuToCpuAddress(GPUVAddr gpu_addr);
+    std::vector<GPUVAddr> CpuToGpuAddress(VAddr cpu_addr) const;
 
     static constexpr u64 PAGE_BITS = 16;
     static constexpr u64 PAGE_SIZE = 1 << PAGE_BITS;
     static constexpr u64 PAGE_MASK = PAGE_SIZE - 1;
 
 private:
-    boost::optional<PAddr> FindFreeBlock(u64 size, u64 align = 1);
-    bool IsPageMapped(PAddr paddr);
-    VAddr& PageSlot(PAddr paddr);
+    boost::optional<GPUVAddr> FindFreeBlock(u64 size, u64 align = 1);
+    bool IsPageMapped(GPUVAddr gpu_addr);
+    VAddr& PageSlot(GPUVAddr gpu_addr);
 
     enum class PageStatus : u64 {
         Unmapped = 0xFFFFFFFFFFFFFFFFULL,
@@ -48,6 +52,14 @@ private:
 
     using PageBlock = std::array<VAddr, PAGE_BLOCK_SIZE>;
     std::array<std::unique_ptr<PageBlock>, PAGE_TABLE_SIZE> page_table{};
+
+    struct MappedRegion {
+        VAddr cpu_addr;
+        GPUVAddr gpu_addr;
+        u64 size;
+    };
+
+    std::vector<MappedRegion> mapped_regions;
 };
 
 } // namespace Tegra

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) {

src/video_core/renderer_opengl/gl_rasterizer.cpp

@@ -116,7 +116,7 @@ RasterizerOpenGL::RasterizerOpenGL() {
 
     glEnable(GL_BLEND);
 
-    LOG_CRITICAL(Render_OpenGL, "Sync fixed function OpenGL state here!");
+    NGLOG_CRITICAL(Render_OpenGL, "Sync fixed function OpenGL state here!");
 }
 
 RasterizerOpenGL::~RasterizerOpenGL() {
@@ -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->PhysicalToVirtualAddress(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);
@@ -233,8 +232,8 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
         // Fetch program code from memory
         GLShader::ProgramCode program_code;
         const u64 gpu_address{gpu.regs.code_address.CodeAddress() + shader_config.offset};
-        const VAddr cpu_address{gpu.memory_manager.PhysicalToVirtualAddress(gpu_address)};
-        Memory::ReadBlock(cpu_address, program_code.data(), program_code.size() * sizeof(u64));
+        const boost::optional<VAddr> cpu_address{gpu.memory_manager.GpuToCpuAddress(gpu_address)};
+        Memory::ReadBlock(*cpu_address, program_code.data(), program_code.size() * sizeof(u64));
         GLShader::ShaderSetup setup{std::move(program_code)};
 
         GLShader::ShaderEntries shader_resources;
@@ -253,8 +252,8 @@ void RasterizerOpenGL::SetupShaders(u8* buffer_ptr, GLintptr buffer_offset) {
             break;
         }
         default:
-            LOG_CRITICAL(HW_GPU, "Unimplemented shader index=%d, enable=%d, offset=0x%08X", index,
-                         shader_config.enable.Value(), shader_config.offset);
+            NGLOG_CRITICAL(HW_GPU, "Unimplemented shader index={}, enable={}, offset={:#010X}",
+                           index, shader_config.enable.Value(), shader_config.offset);
             UNREACHABLE();
         }
 
@@ -394,9 +393,9 @@ void RasterizerOpenGL::DrawArrays() {
     GLintptr index_buffer_offset = 0;
     if (is_indexed) {
         const auto& memory_manager = Core::System().GetInstance().GPU().memory_manager;
-        const VAddr index_data_addr{
-            memory_manager->PhysicalToVirtualAddress(regs.index_array.StartAddress())};
-        Memory::ReadBlock(index_data_addr, offseted_buffer, index_buffer_size);
+        const boost::optional<VAddr> index_data_addr{
+            memory_manager->GpuToCpuAddress(regs.index_array.StartAddress())};
+        Memory::ReadBlock(*index_data_addr, offseted_buffer, index_buffer_size);
 
         index_buffer_offset = buffer_offset;
         offseted_buffer += index_buffer_size;
@@ -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 +559,8 @@ 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.addr = res_cache.TryFindFramebufferGpuAddress(framebuffer_addr).get_value_or(0);
     src_params.width = std::min(framebuffer.width, pixel_stride);
     src_params.height = framebuffer.height;
     src_params.stride = pixel_stride;
@@ -659,9 +659,9 @@ u32 RasterizerOpenGL::SetupConstBuffers(Maxwell::ShaderStage stage, GLuint progr
         buffer_draw_state.enabled = true;
         buffer_draw_state.bindpoint = current_bindpoint + bindpoint;
 
-        VAddr addr = gpu.memory_manager->PhysicalToVirtualAddress(buffer.address);
+        boost::optional<VAddr> addr = gpu.memory_manager->GpuToCpuAddress(buffer.address);
         std::vector<u8> data(used_buffer.GetSize() * sizeof(float));
-        Memory::ReadBlock(addr, data.data(), data.size());
+        Memory::ReadBlock(*addr, data.data(), data.size());
 
         glBindBuffer(GL_SHADER_STORAGE_BUFFER, buffer_draw_state.ssbo);
         glBufferData(GL_SHADER_STORAGE_BUFFER, data.size(), data.data(), GL_DYNAMIC_DRAW);

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;

src/video_core/renderer_opengl/gl_rasterizer_cache.cpp

@@ -41,18 +41,15 @@ struct FormatTuple {
     GLenum format;
     GLenum type;
     bool compressed;
-    // How many pixels in the original texture are equivalent to one pixel in the compressed
-    // texture.
-    u32 compression_factor;
 };
 
 static constexpr std::array<FormatTuple, SurfaceParams::MaxPixelFormat> tex_format_tuples = {{
-    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, false, 1},                     // ABGR8
-    {GL_RGB, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, false, 1},                        // B5G6R5
-    {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, false, 1},               // A2B10G10R10
-    {GL_COMPRESSED_RGB_S3TC_DXT1_EXT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, true, 16},   // DXT1
-    {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true, 16}, // DXT23
-    {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true, 16}, // DXT45
+    {GL_RGBA8, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8_REV, false},                    // ABGR8
+    {GL_RGB, GL_RGB, GL_UNSIGNED_SHORT_5_6_5_REV, false},                       // B5G6R5
+    {GL_RGB10_A2, GL_RGBA, GL_UNSIGNED_INT_2_10_10_10_REV, false},              // A2B10G10R10
+    {GL_COMPRESSED_RGB_S3TC_DXT1_EXT, GL_RGB, GL_UNSIGNED_INT_8_8_8_8, true},   // DXT1
+    {GL_COMPRESSED_RGBA_S3TC_DXT3_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true}, // DXT23
+    {GL_COMPRESSED_RGBA_S3TC_DXT5_EXT, GL_RGBA, GL_UNSIGNED_INT_8_8_8_8, true}, // DXT45
 }};
 
 static const FormatTuple& GetFormatTuple(PixelFormat pixel_format, ComponentType component_type) {
@@ -83,26 +80,30 @@ 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) {
-    constexpr u32 bytes_per_pixel = SurfaceParams::GetFormatBpp(format) / 8;
+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) / CHAR_BIT;
     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);
+        NGLOG_WARNING(Render_OpenGL, "need to use correct swizzle/GOB parameters!");
+        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 +111,8 @@ 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 +221,9 @@ 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 +344,13 @@ 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) &&
@@ -349,9 +358,9 @@ bool CachedSurface::CanFill(const SurfaceParams& dest_surface,
         boost::icl::last_next(fill_interval) <= end && // dest_surface is within our fill range
         dest_surface.FromInterval(fill_interval).GetInterval() ==
             fill_interval) { // make sure interval is a rectangle in dest surface
-        if (fill_size * 8 != dest_surface.GetFormatBpp()) {
+        if (fill_size * CHAR_BIT != dest_surface.GetFormatBpp()) {
             // Check if bits repeat for our fill_size
-            const u32 dest_bytes_per_pixel = std::max(dest_surface.GetFormatBpp() / 8, 1u);
+            const u32 dest_bytes_per_pixel = std::max(dest_surface.GetFormatBpp() / CHAR_BIT, 1u);
             std::vector<u8> fill_test(fill_size * dest_bytes_per_pixel);
 
             for (u32 i = 0; i < dest_bytes_per_pixel; ++i)
@@ -456,15 +465,15 @@ 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;
 
     if (gl_buffer == nullptr) {
-        gl_buffer_size = width * height * GetGLBytesPerPixel(pixel_format);
+        gl_buffer_size = GetActualWidth() * GetActualHeight() * GetGLBytesPerPixel(pixel_format);
         gl_buffer.reset(new u8[gl_buffer_size]);
     }
 
@@ -479,14 +488,15 @@ void CachedSurface::LoadGLBuffer(VAddr load_start, VAddr load_end) {
         std::memcpy(&gl_buffer[start_offset], texture_src_data + start_offset,
                     bytes_per_pixel * width * height);
     } else {
-        morton_to_gl_fns[static_cast<size_t>(pixel_format)](
-            stride, block_height, height, &gl_buffer[0], addr, load_start, load_end);
+        morton_to_gl_fns[static_cast<size_t>(pixel_format)](GetActualWidth(), block_height,
+                                                            GetActualHeight(), &gl_buffer[0], addr,
+                                                            load_start, 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;
 
@@ -536,7 +546,8 @@ void CachedSurface::UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint
 
     MICROPROFILE_SCOPE(OpenGL_TextureUL);
 
-    ASSERT(gl_buffer_size == width * height * GetGLBytesPerPixel(pixel_format));
+    ASSERT(gl_buffer_size ==
+           GetActualWidth() * GetActualHeight() * GetGLBytesPerPixel(pixel_format));
 
     // Load data from memory to the surface
     GLint x0 = static_cast<GLint>(rect.left);
@@ -571,11 +582,9 @@ void CachedSurface::UploadGLTexture(const MathUtil::Rectangle<u32>& rect, GLuint
     glActiveTexture(GL_TEXTURE0);
     if (tuple.compressed) {
         glCompressedTexImage2D(GL_TEXTURE_2D, 0, tuple.internal_format,
-                               static_cast<GLsizei>(rect.GetWidth()),
-                               static_cast<GLsizei>(rect.GetHeight()), 0,
-                               rect.GetWidth() * rect.GetHeight() *
-                                   GetGLBytesPerPixel(pixel_format) / tuple.compression_factor,
-                               &gl_buffer[buffer_offset]);
+                               static_cast<GLsizei>(rect.GetWidth() * GetCompresssionFactor()),
+                               static_cast<GLsizei>(rect.GetHeight() * GetCompresssionFactor()), 0,
+                               size, &gl_buffer[buffer_offset]);
     } else {
         glTexSubImage2D(GL_TEXTURE_2D, 0, x0, y0, static_cast<GLsizei>(rect.GetWidth()),
                         static_cast<GLsizei>(rect.GetHeight()), tuple.format, tuple.type,
@@ -945,6 +954,33 @@ Surface RasterizerCacheOpenGL::GetSurface(const SurfaceParams& params, ScaleMatc
     return surface;
 }
 
+boost::optional<Tegra::GPUVAddr> RasterizerCacheOpenGL::TryFindFramebufferGpuAddress(
+    VAddr cpu_addr) const {
+    // Tries to find the GPU address of a framebuffer based on the CPU address. This is because
+    // final output framebuffers are specified by CPU address, but internally our GPU cache uses GPU
+    // addresses. We iterate through all cached framebuffers, and compare their starting CPU address
+    // to the one provided. This is obviously not great, and won't work if the framebuffer overlaps
+    // surfaces.
+
+    std::vector<Tegra::GPUVAddr> gpu_addresses;
+    for (const auto& pair : surface_cache) {
+        for (const auto& surface : pair.second) {
+            const VAddr surface_cpu_addr = surface->GetCpuAddr();
+            if (cpu_addr >= surface_cpu_addr && cpu_addr < (surface_cpu_addr + surface->size)) {
+                ASSERT_MSG(cpu_addr == surface_cpu_addr, "overlapping surfaces are unsupported");
+                gpu_addresses.push_back(surface->addr);
+            }
+        }
+    }
+
+    if (gpu_addresses.empty()) {
+        return {};
+    }
+
+    ASSERT_MSG(gpu_addresses.size() == 1, ">1 surface is unsupported");
+    return gpu_addresses[0];
+}
+
 SurfaceRect_Tuple RasterizerCacheOpenGL::GetSurfaceSubRect(const SurfaceParams& params,
                                                            ScaleMatch match_res_scale,
                                                            bool load_if_create) {
@@ -1028,11 +1064,11 @@ Surface RasterizerCacheOpenGL::GetTextureSurface(const Tegra::Texture::FullTextu
     auto& gpu = Core::System::GetInstance().GPU();
 
     SurfaceParams params;
-    params.addr = gpu.memory_manager->PhysicalToVirtualAddress(config.tic.Address());
-    params.width = config.tic.Width();
-    params.height = config.tic.Height();
+    params.addr = config.tic.Address();
     params.is_tiled = config.tic.IsTiled();
     params.pixel_format = SurfaceParams::PixelFormatFromTextureFormat(config.tic.format);
+    params.width = config.tic.Width() / params.GetCompresssionFactor();
+    params.height = config.tic.Height() / params.GetCompresssionFactor();
 
     // TODO(Subv): Different types per component are not supported.
     ASSERT(config.tic.r_type.Value() == config.tic.g_type.Value() &&
@@ -1045,7 +1081,7 @@ 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();
@@ -1073,11 +1109,10 @@ 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
-    LOG_WARNING(Render_OpenGL, "hard-coded for render target 0!");
+    NGLOG_WARNING(Render_OpenGL, "hard-coded for render target 0!");
 
     // update resolution_scale_factor and reset cache if changed
     // TODO (bunnei): This code was ported as-is from Citra, and is technically not thread-safe. We
@@ -1106,7 +1141,7 @@ SurfaceSurfaceRect_Tuple RasterizerCacheOpenGL::GetFramebufferSurfaces(
     color_params.block_height = Tegra::Texture::TICEntry::DefaultBlockHeight;
     SurfaceParams depth_params = color_params;
 
-    color_params.addr = memory_manager->PhysicalToVirtualAddress(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();
@@ -1122,8 +1157,8 @@ SurfaceSurfaceRect_Tuple RasterizerCacheOpenGL::GetFramebufferSurfaces(
     // Make sure that framebuffers don't overlap if both color and depth are being used
     if (using_color_fb && using_depth_fb &&
         boost::icl::length(color_vp_interval & depth_vp_interval)) {
-        LOG_CRITICAL(Render_OpenGL, "Color and depth framebuffer memory regions overlap; "
-                                    "overlapping framebuffers not supported!");
+        NGLOG_CRITICAL(Render_OpenGL, "Color and depth framebuffer memory regions overlap; "
+                                      "overlapping framebuffers not supported!");
         using_depth_fb = false;
     }
 
@@ -1222,7 +1257,8 @@ 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 +1297,7 @@ 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 +1333,8 @@ 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;
 
@@ -1390,10 +1427,10 @@ void RasterizerCacheOpenGL::UnregisterSurface(const Surface& surface) {
     surface_cache.subtract({surface->GetInterval(), SurfaceSet{surface}});
 }
 
-void RasterizerCacheOpenGL::UpdatePagesCachedCount(VAddr 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;
+void RasterizerCacheOpenGL::UpdatePagesCachedCount(Tegra::GPUVAddr addr, u64 size, int delta) {
+    const u64 num_pages = ((addr + size - 1) >> Tegra::MemoryManager::PAGE_BITS) -
+                          (addr >> Tegra::MemoryManager::PAGE_BITS) + 1;
+    const u64 page_start = addr >> Tegra::MemoryManager::PAGE_BITS;
     const u64 page_end = page_start + num_pages;
 
     // Interval maps will erase segments if count reaches 0, so if delta is negative we have to
@@ -1406,8 +1443,10 @@ 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)
+                                                    << Tegra::MemoryManager::PAGE_BITS;
+        const Tegra::GPUVAddr interval_end_addr = boost::icl::last_next(interval)
+                                                  << Tegra::MemoryManager::PAGE_BITS;
         const u64 interval_size = interval_end_addr - interval_start_addr;
 
         if (delta > 0 && count == delta)

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>() &&
@@ -82,23 +84,49 @@ struct SurfaceParams {
         Invalid = 4,
     };
 
-    static constexpr unsigned int GetFormatBpp(PixelFormat format) {
+    /**
+     * Gets the compression factor for the specified PixelFormat. This applies to just the
+     * "compressed width" and "compressed height", not the overall compression factor of a
+     * compressed image. This is used for maintaining proper surface sizes for compressed texture
+     * formats.
+     */
+    static constexpr u32 GetCompresssionFactor(PixelFormat format) {
         if (format == PixelFormat::Invalid)
             return 0;
 
-        constexpr std::array<unsigned int, MaxPixelFormat> bpp_table = {
+        constexpr std::array<u32, MaxPixelFormat> compression_factor_table = {{
+            1, // ABGR8
+            1, // B5G6R5
+            1, // A2B10G10R10
+            4, // DXT1
+            4, // DXT23
+            4, // DXT45
+        }};
+
+        ASSERT(static_cast<size_t>(format) < compression_factor_table.size());
+        return compression_factor_table[static_cast<size_t>(format)];
+    }
+    u32 GetCompresssionFactor() const {
+        return GetCompresssionFactor(pixel_format);
+    }
+
+    static constexpr u32 GetFormatBpp(PixelFormat format) {
+        if (format == PixelFormat::Invalid)
+            return 0;
+
+        constexpr std::array<u32, MaxPixelFormat> bpp_table = {{
             32,  // ABGR8
             16,  // B5G6R5
             32,  // A2B10G10R10
             64,  // DXT1
             128, // DXT23
             128, // DXT45
-        };
+        }};
 
         ASSERT(static_cast<size_t>(format) < bpp_table.size());
         return bpp_table[static_cast<size_t>(format)];
     }
-    unsigned int GetFormatBpp() const {
+    u32 GetFormatBpp() const {
         return GetFormatBpp(pixel_format);
     }
 
@@ -253,6 +281,24 @@ struct SurfaceParams {
     // Returns the region of the biggest valid rectange within interval
     SurfaceInterval GetCopyableInterval(const Surface& src_surface) const;
 
+    /**
+     * Gets the actual width (in pixels) of the surface. This is provided because `width` is used
+     * for tracking the surface region in memory, which may be compressed for certain formats. In
+     * this scenario, `width` is actually the compressed width.
+     */
+    u32 GetActualWidth() const {
+        return width * GetCompresssionFactor();
+    }
+
+    /**
+     * Gets the actual height (in pixels) of the surface. This is provided because `height` is used
+     * for tracking the surface region in memory, which may be compressed for certain formats. In
+     * this scenario, `height` is actually the compressed height.
+     */
+    u32 GetActualHeight() const {
+        return height * GetCompresssionFactor();
+    }
+
     u32 GetScaledWidth() const {
         return width * res_scale;
     }
@@ -277,6 +323,8 @@ 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 +333,9 @@ 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;
@@ -325,15 +374,15 @@ struct CachedSurface : SurfaceParams {
         if (format == PixelFormat::Invalid)
             return 0;
 
-        return SurfaceParams::GetFormatBpp(format) / 8;
+        return SurfaceParams::GetFormatBpp(format) / CHAR_BIT;
     }
 
     std::unique_ptr<u8[]> gl_buffer;
     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,
@@ -362,6 +411,9 @@ public:
     Surface GetSurface(const SurfaceParams& params, ScaleMatch match_res_scale,
                        bool load_if_create);
 
+    /// Tries to find a framebuffer GPU address based on the provided CPU address
+    boost::optional<Tegra::GPUVAddr> TryFindFramebufferGpuAddress(VAddr cpu_addr) const;
+
     /// Attempt to find a subrect (resolution scaled) of a surface, otherwise loads a texture from
     /// Switch memory to OpenGL and caches it (if not already cached)
     SurfaceRect_Tuple GetSurfaceSubRect(const SurfaceParams& params, ScaleMatch match_res_scale,
@@ -381,10 +433,10 @@ 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 +445,7 @@ 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 +457,7 @@ 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;

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),
@@ -269,10 +270,9 @@ void RendererOpenGL::ConfigureFramebufferTexture(TextureInfo& texture,
     GLint internal_format;
     switch (framebuffer.pixel_format) {
     case Tegra::FramebufferConfig::PixelFormat::ABGR8:
-        // Use RGBA8 and swap in the fragment shader
         internal_format = GL_RGBA;
         texture.gl_format = GL_RGBA;
-        texture.gl_type = GL_UNSIGNED_INT_8_8_8_8;
+        texture.gl_type = GL_UNSIGNED_INT_8_8_8_8_REV;
         gl_framebuffer_data.resize(texture.width * texture.height * 4);
         break;
     default:
@@ -302,8 +302,8 @@ void RendererOpenGL::DrawScreenTriangles(const ScreenInfo& screen_info, float x,
             right = texcoords.left;
         } else {
             // Other transformations are unsupported
-            LOG_CRITICAL(Render_OpenGL, "Unsupported framebuffer_transform_flags=%d",
-                         framebuffer_transform_flags);
+            NGLOG_CRITICAL(Render_OpenGL, "Unsupported framebuffer_transform_flags={}",
+                           static_cast<u32>(framebuffer_transform_flags));
             UNIMPLEMENTED();
         }
     }
@@ -428,9 +428,9 @@ bool RendererOpenGL::Init() {
     const char* gpu_vendor{reinterpret_cast<char const*>(glGetString(GL_VENDOR))};
     const char* gpu_model{reinterpret_cast<char const*>(glGetString(GL_RENDERER))};
 
-    LOG_INFO(Render_OpenGL, "GL_VERSION: %s", gl_version);
-    LOG_INFO(Render_OpenGL, "GL_VENDOR: %s", gpu_vendor);
-    LOG_INFO(Render_OpenGL, "GL_RENDERER: %s", gpu_model);
+    NGLOG_INFO(Render_OpenGL, "GL_VERSION: {}", gl_version);
+    NGLOG_INFO(Render_OpenGL, "GL_VENDOR: {}", gpu_vendor);
+    NGLOG_INFO(Render_OpenGL, "GL_RENDERER: {}", gpu_model);
 
     Core::Telemetry().AddField(Telemetry::FieldType::UserSystem, "GPU_Vendor", gpu_vendor);
     Core::Telemetry().AddField(Telemetry::FieldType::UserSystem, "GPU_Model", gpu_model);

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"
 
@@ -26,9 +27,8 @@ static u32 GetSwizzleOffset(u32 x, u32 y, u32 image_width, u32 bytes_per_pixel,
     return address;
 }
 
-static void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
-                             u8* swizzled_data, u8* unswizzled_data, bool unswizzle,
-                             u32 block_height) {
+void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
+                      u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height) {
     u8* data_ptrs[2];
     for (unsigned y = 0; y < height; ++y) {
         for (unsigned x = 0; x < width; ++x) {

src/video_core/textures/decoders.h

@@ -17,6 +17,10 @@ namespace Texture {
 std::vector<u8> UnswizzleTexture(VAddr address, TextureFormat format, u32 width, u32 height,
                                  u32 block_height = TICEntry::DefaultBlockHeight);
 
+/// Copies texture data from a buffer and performs swizzling/unswizzling as necessary.
+void CopySwizzledData(u32 width, u32 height, u32 bytes_per_pixel, u32 out_bytes_per_pixel,
+                      u8* swizzled_data, u8* unswizzled_data, bool unswizzle, u32 block_height);
+
 /**
  * Decodes an unswizzled texture into a A8R8G8B8 texture.
  */

src/video_core/video_core.cpp

@@ -24,9 +24,9 @@ bool Init(EmuWindow* emu_window) {
     g_renderer = std::make_unique<RendererOpenGL>();
     g_renderer->SetWindow(g_emu_window);
     if (g_renderer->Init()) {
-        LOG_DEBUG(Render, "initialized OK");
+        NGLOG_DEBUG(Render, "initialized OK");
     } else {
-        LOG_CRITICAL(Render, "initialization failed !");
+        NGLOG_CRITICAL(Render, "initialization failed !");
         return false;
     }
     return true;
@@ -36,7 +36,7 @@ bool Init(EmuWindow* emu_window) {
 void Shutdown() {
     g_renderer.reset();
 
-    LOG_DEBUG(Render, "shutdown OK");
+    NGLOG_DEBUG(Render, "shutdown OK");
 }
 
 } // namespace VideoCore

src/yuzu/debugger/graphics/graphics_surface.cpp

@@ -378,10 +378,10 @@ void GraphicsSurfaceWidget::OnUpdate() {
     // TODO: Implement a good way to visualize alpha components!
 
     QImage decoded_image(surface_width, surface_height, QImage::Format_ARGB32);
-    VAddr address = gpu.memory_manager->PhysicalToVirtualAddress(surface_address);
+    boost::optional<VAddr> address = gpu.memory_manager->GpuToCpuAddress(surface_address);
 
     auto unswizzled_data =
-        Tegra::Texture::UnswizzleTexture(address, surface_format, surface_width, surface_height);
+        Tegra::Texture::UnswizzleTexture(*address, surface_format, surface_width, surface_height);
 
     auto texture_data = Tegra::Texture::DecodeTexture(unswizzled_data, surface_format,
                                                       surface_width, surface_height);
@@ -437,9 +437,9 @@ void GraphicsSurfaceWidget::SaveSurface() {
             pixmap->save(&file, "PNG");
     } else if (selectedFilter == bin_filter) {
         auto& gpu = Core::System::GetInstance().GPU();
-        VAddr address = gpu.memory_manager->PhysicalToVirtualAddress(surface_address);
+        boost::optional<VAddr> address = gpu.memory_manager->GpuToCpuAddress(surface_address);
 
-        const u8* buffer = Memory::GetPointer(address);
+        const u8* buffer = Memory::GetPointer(*address);
         ASSERT_MSG(buffer != nullptr, "Memory not accessible");
 
         QFile file(filename);