7 files changed, 348 insertions, 181 deletions

diff --git a/src/core/hle/service/fs/archive.cpp b/src/core/hle/service/fs/archive.cpp
index e9588cb72..cc51ede0c 100644
--- a/src/core/hle/service/fs/archive.cpp
+++ b/src/core/hle/service/fs/archive.cpp
@@ -114,6 +114,7 @@ ResultVal<bool> File::SyncRequest() {
                 return read.Code();
             }
             cmd_buff[2] = static_cast<u32>(*read);
+            Memory::RasterizerFlushAndInvalidateRegion(Memory::VirtualToPhysicalAddress(address), length);
             break;
         }
 
diff --git a/src/core/hle/service/gsp_gpu.cpp b/src/core/hle/service/gsp_gpu.cpp
index 0c655395e..211fcf599 100644
--- a/src/core/hle/service/gsp_gpu.cpp
+++ b/src/core/hle/service/gsp_gpu.cpp
@@ -4,6 +4,7 @@
 
 #include "common/bit_field.h"
 #include "common/microprofile.h"
+#include "common/profiler.h"
 
 #include "core/memory.h"
 #include "core/hle/kernel/event.h"
@@ -15,8 +16,6 @@
 
 #include "video_core/gpu_debugger.h"
 #include "video_core/debug_utils/debug_utils.h"
-#include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
 
 #include "gsp_gpu.h"
 
@@ -291,8 +290,6 @@ static void FlushDataCache(Service::Interface* self) {
     u32 size    = cmd_buff[2];
     u32 process = cmd_buff[4];
 
-    VideoCore::g_renderer->Rasterizer()->InvalidateRegion(Memory::VirtualToPhysicalAddress(address), size);
-
     // TODO(purpasmart96): Verify return header on HW
 
     cmd_buff[1] = RESULT_SUCCESS.raw; // No error
@@ -408,6 +405,8 @@ void SignalInterrupt(InterruptId interrupt_id) {
     g_interrupt_event->Signal();
 }
 
+MICROPROFILE_DEFINE(GPU_GSP_DMA, "GPU", "GSP DMA", MP_RGB(100, 0, 255));
+
 /// Executes the next GSP command
 static void ExecuteCommand(const Command& command, u32 thread_id) {
     // Utility function to convert register ID to address
@@ -419,18 +418,21 @@ static void ExecuteCommand(const Command& command, u32 thread_id) {
 
     // GX request DMA - typically used for copying memory from GSP heap to VRAM
     case CommandId::REQUEST_DMA:
-        VideoCore::g_renderer->Rasterizer()->FlushRegion(Memory::VirtualToPhysicalAddress(command.dma_request.source_address),
-                                                            command.dma_request.size);
+    {
+        MICROPROFILE_SCOPE(GPU_GSP_DMA);
+
+        // TODO: Consider attempting rasterizer-accelerated surface blit if that usage is ever possible/likely
+        Memory::RasterizerFlushRegion(Memory::VirtualToPhysicalAddress(command.dma_request.source_address),
+                            command.dma_request.size);
+        Memory::RasterizerFlushAndInvalidateRegion(Memory::VirtualToPhysicalAddress(command.dma_request.dest_address),
+                            command.dma_request.size);
 
         memcpy(Memory::GetPointer(command.dma_request.dest_address),
                Memory::GetPointer(command.dma_request.source_address),
                command.dma_request.size);
         SignalInterrupt(InterruptId::DMA);
-
-        VideoCore::g_renderer->Rasterizer()->InvalidateRegion(Memory::VirtualToPhysicalAddress(command.dma_request.dest_address),
-                                                          command.dma_request.size);
         break;
-
+    }
     // TODO: This will need some rework in the future. (why?)
     case CommandId::SUBMIT_GPU_CMDLIST:
     {
@@ -517,13 +519,8 @@ static void ExecuteCommand(const Command& command, u32 thread_id) {
 
     case CommandId::CACHE_FLUSH:
     {
-        for (auto& region : command.cache_flush.regions) {
-            if (region.size == 0)
-                break;
-
-            VideoCore::g_renderer->Rasterizer()->InvalidateRegion(
-                Memory::VirtualToPhysicalAddress(region.address), region.size);
-        }
+        // NOTE: Rasterizer flushing handled elsewhere in CPU read/write and other GPU handlers
+        // Use command.cache_flush.regions to implement this handler
         break;
     }
 
diff --git a/src/core/hle/service/y2r_u.cpp b/src/core/hle/service/y2r_u.cpp
index 22f373adf..1672ad775 100644
--- a/src/core/hle/service/y2r_u.cpp
+++ b/src/core/hle/service/y2r_u.cpp
@@ -12,9 +12,6 @@
 #include "core/hle/service/y2r_u.h"
 #include "core/hw/y2r.h"
 
-#include "video_core/renderer_base.h"
-#include "video_core/video_core.h"
-
 ////////////////////////////////////////////////////////////////////////////////////////////////////
 // Namespace Y2R_U
 
@@ -262,13 +259,12 @@ static void SetAlpha(Service::Interface* self) {
 static void StartConversion(Service::Interface* self) {
     u32* cmd_buff = Kernel::GetCommandBuffer();
 
-    HW::Y2R::PerformConversion(conversion);
-
     // dst_image_size would seem to be perfect for this, but it doesn't include the gap :(
     u32 total_output_size = conversion.input_lines *
         (conversion.dst.transfer_unit + conversion.dst.gap);
-    VideoCore::g_renderer->Rasterizer()->InvalidateRegion(
-        Memory::VirtualToPhysicalAddress(conversion.dst.address), total_output_size);
+    Memory::RasterizerFlushAndInvalidateRegion(Memory::VirtualToPhysicalAddress(conversion.dst.address), total_output_size);
+
+    HW::Y2R::PerformConversion(conversion);
 
     LOG_DEBUG(Service_Y2R, "called");
     completion_event->Signal();
diff --git a/src/core/hw/gpu.cpp b/src/core/hw/gpu.cpp
index 7e2f9cdfa..2fe856293 100644
--- a/src/core/hw/gpu.cpp
+++ b/src/core/hw/gpu.cpp
@@ -115,21 +115,39 @@ inline void Write(u32 addr, const T data) {
                 u8* start = Memory::GetPhysicalPointer(config.GetStartAddress());
                 u8* end = Memory::GetPhysicalPointer(config.GetEndAddress());
 
-                if (config.fill_24bit) {
-                    // fill with 24-bit values
-                    for (u8* ptr = start; ptr < end; ptr += 3) {
-                        ptr[0] = config.value_24bit_r;
-                        ptr[1] = config.value_24bit_g;
-                        ptr[2] = config.value_24bit_b;
+                // TODO: Consider always accelerating and returning vector of
+                //       regions that the accelerated fill did not cover to
+                //       reduce/eliminate the fill that the cpu has to do.
+                //       This would also mean that the flush below is not needed.
+                //       Fill should first flush all surfaces that touch but are
+                //       not completely within the fill range.
+                //       Then fill all completely covered surfaces, and return the
+                //       regions that were between surfaces or within the touching
+                //       ones for cpu to manually fill here.
+                if (!VideoCore::g_renderer->Rasterizer()->AccelerateFill(config)) {
+                    Memory::RasterizerFlushAndInvalidateRegion(config.GetStartAddress(), config.GetEndAddress() - config.GetStartAddress());
+
+                    if (config.fill_24bit) {
+                        // fill with 24-bit values
+                        for (u8* ptr = start; ptr < end; ptr += 3) {
+                            ptr[0] = config.value_24bit_r;
+                            ptr[1] = config.value_24bit_g;
+                            ptr[2] = config.value_24bit_b;
+                        }
+                    } else if (config.fill_32bit) {
+                        // fill with 32-bit values
+                        if (end > start) {
+                            u32 value = config.value_32bit;
+                            size_t len = (end - start) / sizeof(u32);
+                            for (size_t i = 0; i < len; ++i)
+                                memcpy(&start[i * sizeof(u32)], &value, sizeof(u32));
+                        }
+                    } else {
+                        // fill with 16-bit values
+                        u16 value_16bit = config.value_16bit.Value();
+                        for (u8* ptr = start; ptr < end; ptr += sizeof(u16))
+                            memcpy(ptr, &value_16bit, sizeof(u16));
                     }
-                } else if (config.fill_32bit) {
-                    // fill with 32-bit values
-                    for (u32* ptr = (u32*)start; ptr < (u32*)end; ++ptr)
-                        *ptr = config.value_32bit;
-                } else {
-                    // fill with 16-bit values
-                    for (u16* ptr = (u16*)start; ptr < (u16*)end; ++ptr)
-                        *ptr = config.value_16bit;
                 }
 
                 LOG_TRACE(HW_GPU, "MemoryFill from 0x%08x to 0x%08x", config.GetStartAddress(), config.GetEndAddress());
@@ -139,8 +157,6 @@ inline void Write(u32 addr, const T data) {
                 } else {
                     GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PSC1);
                 }
-
-                VideoCore::g_renderer->Rasterizer()->InvalidateRegion(config.GetStartAddress(), config.GetEndAddress() - config.GetStartAddress());
             }
 
             // Reset "trigger" flag and set the "finish" flag
@@ -161,184 +177,185 @@ inline void Write(u32 addr, const T data) {
             if (Pica::g_debug_context)
                 Pica::g_debug_context->OnEvent(Pica::DebugContext::Event::IncomingDisplayTransfer, nullptr);
 
-            u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress());
-            u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress());
-
-            if (config.is_texture_copy) {
-                u32 input_width = config.texture_copy.input_width * 16;
-                u32 input_gap = config.texture_copy.input_gap * 16;
-                u32 output_width = config.texture_copy.output_width * 16;
-                u32 output_gap = config.texture_copy.output_gap * 16;
-
-                size_t contiguous_input_size = config.texture_copy.size / input_width * (input_width + input_gap);
-                VideoCore::g_renderer->Rasterizer()->FlushRegion(config.GetPhysicalInputAddress(), contiguous_input_size);
-
-                u32 remaining_size = config.texture_copy.size;
-                u32 remaining_input = input_width;
-                u32 remaining_output = output_width;
-                while (remaining_size > 0) {
-                    u32 copy_size = std::min({ remaining_input, remaining_output, remaining_size });
+            if (!VideoCore::g_renderer->Rasterizer()->AccelerateDisplayTransfer(config)) {
+                u8* src_pointer = Memory::GetPhysicalPointer(config.GetPhysicalInputAddress());
+                u8* dst_pointer = Memory::GetPhysicalPointer(config.GetPhysicalOutputAddress());
 
-                    std::memcpy(dst_pointer, src_pointer, copy_size);
-                    src_pointer += copy_size;
-                    dst_pointer += copy_size;
+                if (config.is_texture_copy) {
+                    u32 input_width = config.texture_copy.input_width * 16;
+                    u32 input_gap = config.texture_copy.input_gap * 16;
+                    u32 output_width = config.texture_copy.output_width * 16;
+                    u32 output_gap = config.texture_copy.output_gap * 16;
 
-                    remaining_input -= copy_size;
-                    remaining_output -= copy_size;
-                    remaining_size -= copy_size;
+                    size_t contiguous_input_size = config.texture_copy.size / input_width * (input_width + input_gap);
+                    Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), contiguous_input_size);
 
-                    if (remaining_input == 0) {
-                        remaining_input = input_width;
-                        src_pointer += input_gap;
-                    }
-                    if (remaining_output == 0) {
-                        remaining_output = output_width;
-                        dst_pointer += output_gap;
-                    }
-                }
+                    size_t contiguous_output_size = config.texture_copy.size / output_width * (output_width + output_gap);
+                    Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), contiguous_output_size);
 
-                LOG_TRACE(HW_GPU, "TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> 0x%08X(%u+%u), flags 0x%08X",
-                    config.texture_copy.size,
-                    config.GetPhysicalInputAddress(), input_width, input_gap,
-                    config.GetPhysicalOutputAddress(), output_width, output_gap,
-                    config.flags);
+                    u32 remaining_size = config.texture_copy.size;
+                    u32 remaining_input = input_width;
+                    u32 remaining_output = output_width;
+                    while (remaining_size > 0) {
+                        u32 copy_size = std::min({ remaining_input, remaining_output, remaining_size });
 
-                size_t contiguous_output_size = config.texture_copy.size / output_width * (output_width + output_gap);
-                VideoCore::g_renderer->Rasterizer()->InvalidateRegion(config.GetPhysicalOutputAddress(), contiguous_output_size);
+                        std::memcpy(dst_pointer, src_pointer, copy_size);
+                        src_pointer += copy_size;
+                        dst_pointer += copy_size;
 
-                GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
-                break;
-            }
+                        remaining_input -= copy_size;
+                        remaining_output -= copy_size;
+                        remaining_size -= copy_size;
 
-            if (config.scaling > config.ScaleXY) {
-                LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u", config.scaling.Value());
-                UNIMPLEMENTED();
-                break;
-            }
+                        if (remaining_input == 0) {
+                            remaining_input = input_width;
+                            src_pointer += input_gap;
+                        }
+                        if (remaining_output == 0) {
+                            remaining_output = output_width;
+                            dst_pointer += output_gap;
+                        }
+                    }
 
-            if (config.input_linear && config.scaling != config.NoScale) {
-                LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
-                UNIMPLEMENTED();
-                break;
-            }
+                    LOG_TRACE(HW_GPU, "TextureCopy: 0x%X bytes from 0x%08X(%u+%u)-> 0x%08X(%u+%u), flags 0x%08X",
+                        config.texture_copy.size,
+                        config.GetPhysicalInputAddress(), input_width, input_gap,
+                        config.GetPhysicalOutputAddress(), output_width, output_gap,
+                        config.flags);
 
-            bool horizontal_scale = config.scaling != config.NoScale;
-            bool vertical_scale = config.scaling == config.ScaleXY;
+                    GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
+                    break;
+                }
 
-            u32 output_width = config.output_width >> horizontal_scale;
-            u32 output_height = config.output_height >> vertical_scale;
+                if (config.scaling > config.ScaleXY) {
+                    LOG_CRITICAL(HW_GPU, "Unimplemented display transfer scaling mode %u", config.scaling.Value());
+                    UNIMPLEMENTED();
+                    break;
+                }
 
-            u32 input_size = config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
-            u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
+                if (config.input_linear && config.scaling != config.NoScale) {
+                    LOG_CRITICAL(HW_GPU, "Scaling is only implemented on tiled input");
+                    UNIMPLEMENTED();
+                    break;
+                }
 
-            VideoCore::g_renderer->Rasterizer()->FlushRegion(config.GetPhysicalInputAddress(), input_size);
+                int horizontal_scale = config.scaling != config.NoScale ? 1 : 0;
+                int vertical_scale = config.scaling == config.ScaleXY ? 1 : 0;
 
-            for (u32 y = 0; y < output_height; ++y) {
-                for (u32 x = 0; x < output_width; ++x) {
-                    Math::Vec4<u8> src_color;
+                u32 output_width = config.output_width >> horizontal_scale;
+                u32 output_height = config.output_height >> vertical_scale;
 
-                    // Calculate the [x,y] position of the input image
-                    // based on the current output position and the scale
-                    u32 input_x = x << horizontal_scale;
-                    u32 input_y = y << vertical_scale;
+                u32 input_size = config.input_width * config.input_height * GPU::Regs::BytesPerPixel(config.input_format);
+                u32 output_size = output_width * output_height * GPU::Regs::BytesPerPixel(config.output_format);
 
-                    if (config.flip_vertically) {
-                        // Flip the y value of the output data,
-                        // we do this after calculating the [x,y] position of the input image
-                        // to account for the scaling options.
-                        y = output_height - y - 1;
-                    }
+                Memory::RasterizerFlushRegion(config.GetPhysicalInputAddress(), input_size);
+                Memory::RasterizerFlushAndInvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
 
-                    u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
-                    u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
-                    u32 src_offset;
-                    u32 dst_offset;
+                for (u32 y = 0; y < output_height; ++y) {
+                    for (u32 x = 0; x < output_width; ++x) {
+                        Math::Vec4<u8> src_color;
 
-                    if (config.input_linear) {
-                        if (!config.dont_swizzle) {
-                            // Interpret the input as linear and the output as tiled
-                            u32 coarse_y = y & ~7;
-                            u32 stride = output_width * dst_bytes_per_pixel;
+                        // Calculate the [x,y] position of the input image
+                        // based on the current output position and the scale
+                        u32 input_x = x << horizontal_scale;
+                        u32 input_y = y << vertical_scale;
 
-                            src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
-                            dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + coarse_y * stride;
-                        } else {
-                           // Both input and output are linear
-                            src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
-                            dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
+                        if (config.flip_vertically) {
+                            // Flip the y value of the output data,
+                            // we do this after calculating the [x,y] position of the input image
+                            // to account for the scaling options.
+                            y = output_height - y - 1;
                         }
-                    } else {
-                        if (!config.dont_swizzle) {
-                            // Interpret the input as tiled and the output as linear
-                            u32 coarse_y = input_y & ~7;
-                            u32 stride = config.input_width * src_bytes_per_pixel;
 
-                            src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) + coarse_y * stride;
-                            dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
+                        u32 dst_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.output_format);
+                        u32 src_bytes_per_pixel = GPU::Regs::BytesPerPixel(config.input_format);
+                        u32 src_offset;
+                        u32 dst_offset;
+
+                        if (config.input_linear) {
+                            if (!config.dont_swizzle) {
+                                // Interpret the input as linear and the output as tiled
+                                u32 coarse_y = y & ~7;
+                                u32 stride = output_width * dst_bytes_per_pixel;
+
+                                src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
+                                dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + coarse_y * stride;
+                            } else {
+                                // Both input and output are linear
+                                src_offset = (input_x + input_y * config.input_width) * src_bytes_per_pixel;
+                                dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
+                            }
                         } else {
-                            // Both input and output are tiled
-                            u32 out_coarse_y = y & ~7;
-                            u32 out_stride = output_width * dst_bytes_per_pixel;
-
-                            u32 in_coarse_y = input_y & ~7;
-                            u32 in_stride = config.input_width * src_bytes_per_pixel;
-
-                            src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) + in_coarse_y * in_stride;
-                            dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + out_coarse_y * out_stride;
+                            if (!config.dont_swizzle) {
+                                // Interpret the input as tiled and the output as linear
+                                u32 coarse_y = input_y & ~7;
+                                u32 stride = config.input_width * src_bytes_per_pixel;
+
+                                src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) + coarse_y * stride;
+                                dst_offset = (x + y * output_width) * dst_bytes_per_pixel;
+                            } else {
+                                // Both input and output are tiled
+                                u32 out_coarse_y = y & ~7;
+                                u32 out_stride = output_width * dst_bytes_per_pixel;
+
+                                u32 in_coarse_y = input_y & ~7;
+                                u32 in_stride = config.input_width * src_bytes_per_pixel;
+
+                                src_offset = VideoCore::GetMortonOffset(input_x, input_y, src_bytes_per_pixel) + in_coarse_y * in_stride;
+                                dst_offset = VideoCore::GetMortonOffset(x, y, dst_bytes_per_pixel) + out_coarse_y * out_stride;
+                            }
                         }
-                    }
 
-                    const u8* src_pixel = src_pointer + src_offset;
-                    src_color = DecodePixel(config.input_format, src_pixel);
-                    if (config.scaling == config.ScaleX) {
-                        Math::Vec4<u8> pixel = DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
-                        src_color = ((src_color + pixel) / 2).Cast<u8>();
-                    } else if (config.scaling == config.ScaleXY) {
-                        Math::Vec4<u8> pixel1 = DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
-                        Math::Vec4<u8> pixel2 = DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
-                        Math::Vec4<u8> pixel3 = DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
-                        src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
-                    }
+                        const u8* src_pixel = src_pointer + src_offset;
+                        src_color = DecodePixel(config.input_format, src_pixel);
+                        if (config.scaling == config.ScaleX) {
+                            Math::Vec4<u8> pixel = DecodePixel(config.input_format, src_pixel + src_bytes_per_pixel);
+                            src_color = ((src_color + pixel) / 2).Cast<u8>();
+                        } else if (config.scaling == config.ScaleXY) {
+                            Math::Vec4<u8> pixel1 = DecodePixel(config.input_format, src_pixel + 1 * src_bytes_per_pixel);
+                            Math::Vec4<u8> pixel2 = DecodePixel(config.input_format, src_pixel + 2 * src_bytes_per_pixel);
+                            Math::Vec4<u8> pixel3 = DecodePixel(config.input_format, src_pixel + 3 * src_bytes_per_pixel);
+                            src_color = (((src_color + pixel1) + (pixel2 + pixel3)) / 4).Cast<u8>();
+                        }
 
-                    u8* dst_pixel = dst_pointer + dst_offset;
-                    switch (config.output_format) {
-                    case Regs::PixelFormat::RGBA8:
-                        Color::EncodeRGBA8(src_color, dst_pixel);
-                        break;
+                        u8* dst_pixel = dst_pointer + dst_offset;
+                        switch (config.output_format) {
+                        case Regs::PixelFormat::RGBA8:
+                            Color::EncodeRGBA8(src_color, dst_pixel);
+                            break;
 
-                    case Regs::PixelFormat::RGB8:
-                        Color::EncodeRGB8(src_color, dst_pixel);
-                        break;
+                        case Regs::PixelFormat::RGB8:
+                            Color::EncodeRGB8(src_color, dst_pixel);
+                            break;
 
-                    case Regs::PixelFormat::RGB565:
-                        Color::EncodeRGB565(src_color, dst_pixel);
-                        break;
+                        case Regs::PixelFormat::RGB565:
+                            Color::EncodeRGB565(src_color, dst_pixel);
+                            break;
 
-                    case Regs::PixelFormat::RGB5A1:
-                        Color::EncodeRGB5A1(src_color, dst_pixel);
-                        break;
+                        case Regs::PixelFormat::RGB5A1:
+                            Color::EncodeRGB5A1(src_color, dst_pixel);
+                            break;
 
-                    case Regs::PixelFormat::RGBA4:
-                        Color::EncodeRGBA4(src_color, dst_pixel);
-                        break;
+                        case Regs::PixelFormat::RGBA4:
+                            Color::EncodeRGBA4(src_color, dst_pixel);
+                            break;
 
-                    default:
-                        LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x", config.output_format.Value());
-                        break;
+                        default:
+                            LOG_ERROR(HW_GPU, "Unknown destination framebuffer format %x", config.output_format.Value());
+                            break;
+                        }
                     }
                 }
-            }
 
-            LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), dst format %x, flags 0x%08X",
+                LOG_TRACE(HW_GPU, "DisplayTriggerTransfer: 0x%08x bytes from 0x%08x(%ux%u)-> 0x%08x(%ux%u), dst format %x, flags 0x%08X",
                       config.output_height * output_width * GPU::Regs::BytesPerPixel(config.output_format),
                       config.GetPhysicalInputAddress(), config.input_width.Value(), config.input_height.Value(),
                       config.GetPhysicalOutputAddress(), output_width, output_height,
                       config.output_format.Value(), config.flags);
+            }
 
             g_regs.display_transfer_config.trigger = 0;
             GSP_GPU::SignalInterrupt(GSP_GPU::InterruptId::PPF);
-
-            VideoCore::g_renderer->Rasterizer()->InvalidateRegion(config.GetPhysicalOutputAddress(), output_size);
         }
         break;
     }
diff --git a/src/core/hw/gpu.h b/src/core/hw/gpu.h
index a00adbf53..da4c345b4 100644
--- a/src/core/hw/gpu.h
+++ b/src/core/hw/gpu.h
@@ -78,7 +78,7 @@ struct Regs {
 
     INSERT_PADDING_WORDS(0x4);
 
-    struct {
+    struct MemoryFillConfig {
         u32 address_start;
         u32 address_end;
 
@@ -165,7 +165,7 @@ struct Regs {
 
     INSERT_PADDING_WORDS(0x169);
 
-    struct {
+    struct DisplayTransferConfig {
         u32 input_address;
         u32 output_address;
 
diff --git a/src/core/memory.cpp b/src/core/memory.cpp
index 7de5bd15d..ee9b69f81 100644
--- a/src/core/memory.cpp
+++ b/src/core/memory.cpp
@@ -15,6 +15,9 @@
 #include "core/memory_setup.h"
 #include "core/mmio.h"
 
+#include "video_core/renderer_base.h"
+#include "video_core/video_core.h"
+
 namespace Memory {
 
 enum class PageType {
@@ -22,8 +25,12 @@ enum class PageType {
     Unmapped,
     /// Page is mapped to regular memory. This is the only type you can get pointers to.
     Memory,
+    /// Page is mapped to regular memory, but also needs to check for rasterizer cache flushing and invalidation
+    RasterizerCachedMemory,
     /// Page is mapped to a I/O region. Writing and reading to this page is handled by functions.
     Special,
+    /// Page is mapped to a I/O region, but also needs to check for rasterizer cache flushing and invalidation
+    RasterizerCachedSpecial,
 };
 
 struct SpecialRegion {
@@ -57,6 +64,12 @@ struct PageTable {
      * the corresponding entry in `pointers` MUST be set to null.
      */
     std::array<PageType, NUM_ENTRIES> attributes;
+
+    /**
+     * Indicates the number of externally cached resources touching a page that should be
+     * flushed before the memory is accessed
+     */
+    std::array<u8, NUM_ENTRIES> cached_res_count;
 };
 
 /// Singular page table used for the singleton process
@@ -72,8 +85,15 @@ static void MapPages(u32 base, u32 size, u8* memory, PageType type) {
     while (base != end) {
         ASSERT_MSG(base < PageTable::NUM_ENTRIES, "out of range mapping at %08X", base);
 
+        // Since pages are unmapped on shutdown after video core is shutdown, the renderer may be null here
+        if (current_page_table->attributes[base] == PageType::RasterizerCachedMemory ||
+            current_page_table->attributes[base] == PageType::RasterizerCachedSpecial) {
+            RasterizerFlushAndInvalidateRegion(VirtualToPhysicalAddress(base << PAGE_BITS), PAGE_SIZE);
+        }
+
         current_page_table->attributes[base] = type;
         current_page_table->pointers[base] = memory;
+        current_page_table->cached_res_count[base] = 0;
 
         base += 1;
         if (memory != nullptr)
@@ -84,6 +104,7 @@ static void MapPages(u32 base, u32 size, u8* memory, PageType type) {
 void InitMemoryMap() {
     main_page_table.pointers.fill(nullptr);
     main_page_table.attributes.fill(PageType::Unmapped);
+    main_page_table.cached_res_count.fill(0);
 }
 
 void MapMemoryRegion(VAddr base, u32 size, u8* target) {
@@ -107,6 +128,28 @@ void UnmapRegion(VAddr base, u32 size) {
 }
 
 /**
+ * Gets a pointer to the exact memory at the virtual address (i.e. not page aligned)
+ * using a VMA from the current process
+ */
+static u8* GetPointerFromVMA(VAddr vaddr) {
+    u8* direct_pointer = nullptr;
+
+    auto& vma = Kernel::g_current_process->vm_manager.FindVMA(vaddr)->second;
+    switch (vma.type) {
+    case Kernel::VMAType::AllocatedMemoryBlock:
+        direct_pointer = vma.backing_block->data() + vma.offset;
+        break;
+    case Kernel::VMAType::BackingMemory:
+        direct_pointer = vma.backing_memory;
+        break;
+    default:
+        UNREACHABLE();
+    }
+
+    return direct_pointer + (vaddr - vma.base);
+}
+
+/**
  * This function should only be called for virtual addreses with attribute `PageType::Special`.
  */
 static MMIORegionPointer GetMMIOHandler(VAddr vaddr) {
@@ -126,6 +169,7 @@ template <typename T>
 T Read(const VAddr vaddr) {
     const u8* page_pointer = current_page_table->pointers[vaddr >> PAGE_BITS];
     if (page_pointer) {
+        // NOTE: Avoid adding any extra logic to this fast-path block
         T value;
         std::memcpy(&value, &page_pointer[vaddr & PAGE_MASK], sizeof(T));
         return value;
@@ -139,8 +183,22 @@ T Read(const VAddr vaddr) {
     case PageType::Memory:
         ASSERT_MSG(false, "Mapped memory page without a pointer @ %08X", vaddr);
         break;
+    case PageType::RasterizerCachedMemory:
+    {
+        RasterizerFlushRegion(VirtualToPhysicalAddress(vaddr), sizeof(T));
+
+        T value;
+        std::memcpy(&value, GetPointerFromVMA(vaddr), sizeof(T));
+        return value;
+    }
     case PageType::Special:
         return ReadMMIO<T>(GetMMIOHandler(vaddr), vaddr);
+    case PageType::RasterizerCachedSpecial:
+    {
+        RasterizerFlushRegion(VirtualToPhysicalAddress(vaddr), sizeof(T));
+
+        return ReadMMIO<T>(GetMMIOHandler(vaddr), vaddr);
+    }
     default:
         UNREACHABLE();
     }
@@ -153,6 +211,7 @@ template <typename T>
 void Write(const VAddr vaddr, const T data) {
     u8* page_pointer = current_page_table->pointers[vaddr >> PAGE_BITS];
     if (page_pointer) {
+        // NOTE: Avoid adding any extra logic to this fast-path block
         std::memcpy(&page_pointer[vaddr & PAGE_MASK], &data, sizeof(T));
         return;
     }
@@ -165,9 +224,23 @@ void Write(const VAddr vaddr, const T data) {
     case PageType::Memory:
         ASSERT_MSG(false, "Mapped memory page without a pointer @ %08X", vaddr);
         break;
+    case PageType::RasterizerCachedMemory:
+    {
+        RasterizerFlushAndInvalidateRegion(VirtualToPhysicalAddress(vaddr), sizeof(T));
+
+        std::memcpy(GetPointerFromVMA(vaddr), &data, sizeof(T));
+        break;
+    }
     case PageType::Special:
         WriteMMIO<T>(GetMMIOHandler(vaddr), vaddr, data);
         break;
+    case PageType::RasterizerCachedSpecial:
+    {
+        RasterizerFlushAndInvalidateRegion(VirtualToPhysicalAddress(vaddr), sizeof(T));
+
+        WriteMMIO<T>(GetMMIOHandler(vaddr), vaddr, data);
+        break;
+    }
     default:
         UNREACHABLE();
     }
@@ -179,6 +252,10 @@ u8* GetPointer(const VAddr vaddr) {
         return page_pointer + (vaddr & PAGE_MASK);
     }
 
+    if (current_page_table->attributes[vaddr >> PAGE_BITS] == PageType::RasterizerCachedMemory) {
+        return GetPointerFromVMA(vaddr);
+    }
+
     LOG_ERROR(HW_Memory, "unknown GetPointer @ 0x%08x", vaddr);
     return nullptr;
 }
@@ -187,6 +264,69 @@ u8* GetPhysicalPointer(PAddr address) {
     return GetPointer(PhysicalToVirtualAddress(address));
 }
 
+void RasterizerMarkRegionCached(PAddr start, u32 size, int count_delta) {
+    if (start == 0) {
+        return;
+    }
+
+    u32 num_pages = ((start + size - 1) >> PAGE_BITS) - (start >> PAGE_BITS) + 1;
+    PAddr paddr = start;
+
+    for (unsigned i = 0; i < num_pages; ++i) {
+        VAddr vaddr = PhysicalToVirtualAddress(paddr);
+        u8& res_count = current_page_table->cached_res_count[vaddr >> PAGE_BITS];
+        ASSERT_MSG(count_delta <= UINT8_MAX - res_count, "Rasterizer resource cache counter overflow!");
+        ASSERT_MSG(count_delta >= -res_count, "Rasterizer resource cache counter underflow!");
+
+        // Switch page type to cached if now cached
+        if (res_count == 0) {
+            PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
+            switch (page_type) {
+            case PageType::Memory:
+                page_type = PageType::RasterizerCachedMemory;
+                current_page_table->pointers[vaddr >> PAGE_BITS] = nullptr;
+                break;
+            case PageType::Special:
+                page_type = PageType::RasterizerCachedSpecial;
+                break;
+            default:
+                UNREACHABLE();
+            }
+        }
+
+        res_count += count_delta;
+
+        // Switch page type to uncached if now uncached
+        if (res_count == 0) {
+            PageType& page_type = current_page_table->attributes[vaddr >> PAGE_BITS];
+            switch (page_type) {
+            case PageType::RasterizerCachedMemory:
+                page_type = PageType::Memory;
+                current_page_table->pointers[vaddr >> PAGE_BITS] = GetPointerFromVMA(vaddr & ~PAGE_MASK);
+                break;
+            case PageType::RasterizerCachedSpecial:
+                page_type = PageType::Special;
+                break;
+            default:
+                UNREACHABLE();
+            }
+        }
+        paddr += PAGE_SIZE;
+    }
+}
+
+void RasterizerFlushRegion(PAddr start, u32 size) {
+    if (VideoCore::g_renderer != nullptr) {
+        VideoCore::g_renderer->Rasterizer()->FlushRegion(start, size);
+    }
+}
+
+void RasterizerFlushAndInvalidateRegion(PAddr start, u32 size) {
+    if (VideoCore::g_renderer != nullptr) {
+        VideoCore::g_renderer->Rasterizer()->FlushAndInvalidateRegion(start, size);
+    }
+}
+
 u8 Read8(const VAddr addr) {
     return Read<u8>(addr);
 }
diff --git a/src/core/memory.h b/src/core/memory.h
index 5af72b7a7..9caa3c3f5 100644
--- a/src/core/memory.h
+++ b/src/core/memory.h
@@ -148,4 +148,20 @@ VAddr PhysicalToVirtualAddress(PAddr addr);
  */
 u8* GetPhysicalPointer(PAddr address);
 
+/**
+ * Adds the supplied value to the rasterizer resource cache counter of each
+ * page touching the region.
+ */
+void RasterizerMarkRegionCached(PAddr start, u32 size, int count_delta);
+
+/**
+ * Flushes any externally cached rasterizer resources touching the given region.
+ */
+void RasterizerFlushRegion(PAddr start, u32 size);
+
+/**
+ * Flushes and invalidates any externally cached rasterizer resources touching the given region.
+ */
+void RasterizerFlushAndInvalidateRegion(PAddr start, u32 size);
+
 }