Merge pull request #10155 from FernandoS27/reactive-flushing-new

Y.F.C. bring back Reactive Flushing

5 files changed, 59 insertions, 8 deletions

diff --git a/src/video_core/texture_cache/image_info.cpp b/src/video_core/texture_cache/image_info.cpp
index 11f3f78a1..e8ddde691 100644
--- a/src/video_core/texture_cache/image_info.cpp
+++ b/src/video_core/texture_cache/image_info.cpp
@@ -4,6 +4,7 @@
 #include <fmt/format.h>
 
 #include "common/assert.h"
+#include "common/settings.h"
 #include "video_core/surface.h"
 #include "video_core/texture_cache/format_lookup_table.h"
 #include "video_core/texture_cache/image_info.h"
@@ -22,6 +23,8 @@ using VideoCore::Surface::PixelFormat;
 using VideoCore::Surface::SurfaceType;
 
 ImageInfo::ImageInfo(const TICEntry& config) noexcept {
+    forced_flushed = config.IsPitchLinear() && !Settings::values.use_reactive_flushing.GetValue();
+    dma_downloaded = forced_flushed;
     format = PixelFormatFromTextureInfo(config.format, config.r_type, config.g_type, config.b_type,
                                         config.a_type, config.srgb_conversion);
     num_samples = NumSamples(config.msaa_mode);
@@ -117,6 +120,9 @@ ImageInfo::ImageInfo(const TICEntry& config) noexcept {
 
 ImageInfo::ImageInfo(const Maxwell3D::Regs::RenderTargetConfig& ct,
                      Tegra::Texture::MsaaMode msaa_mode) noexcept {
+    forced_flushed =
+        ct.tile_mode.is_pitch_linear && !Settings::values.use_reactive_flushing.GetValue();
+    dma_downloaded = forced_flushed;
     format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(ct.format);
     rescaleable = false;
     if (ct.tile_mode.is_pitch_linear) {
@@ -155,6 +161,9 @@ ImageInfo::ImageInfo(const Maxwell3D::Regs::RenderTargetConfig& ct,
 
 ImageInfo::ImageInfo(const Maxwell3D::Regs::Zeta& zt, const Maxwell3D::Regs::ZetaSize& zt_size,
                      Tegra::Texture::MsaaMode msaa_mode) noexcept {
+    forced_flushed =
+        zt.tile_mode.is_pitch_linear && !Settings::values.use_reactive_flushing.GetValue();
+    dma_downloaded = forced_flushed;
     format = VideoCore::Surface::PixelFormatFromDepthFormat(zt.format);
     size.width = zt_size.width;
     size.height = zt_size.height;
@@ -195,6 +204,9 @@ ImageInfo::ImageInfo(const Maxwell3D::Regs::Zeta& zt, const Maxwell3D::Regs::Zet
 
 ImageInfo::ImageInfo(const Fermi2D::Surface& config) noexcept {
     UNIMPLEMENTED_IF_MSG(config.layer != 0, "Surface layer is not zero");
+    forced_flushed = config.linear == Fermi2D::MemoryLayout::Pitch &&
+                     !Settings::values.use_reactive_flushing.GetValue();
+    dma_downloaded = forced_flushed;
     format = VideoCore::Surface::PixelFormatFromRenderTargetFormat(config.format);
     rescaleable = false;
     if (config.linear == Fermi2D::MemoryLayout::Pitch) {
diff --git a/src/video_core/texture_cache/image_info.h b/src/video_core/texture_cache/image_info.h
index 4b7dfa315..8a4cb0cbd 100644
--- a/src/video_core/texture_cache/image_info.h
+++ b/src/video_core/texture_cache/image_info.h
@@ -39,6 +39,8 @@ struct ImageInfo {
     u32 tile_width_spacing = 0;
     bool rescaleable = false;
     bool downscaleable = false;
+    bool forced_flushed = false;
+    bool dma_downloaded = false;
 };
 
 } // namespace VideoCommon
diff --git a/src/video_core/texture_cache/image_view_base.cpp b/src/video_core/texture_cache/image_view_base.cpp
index bcad40353..d134b6738 100644
--- a/src/video_core/texture_cache/image_view_base.cpp
+++ b/src/video_core/texture_cache/image_view_base.cpp
@@ -4,7 +4,6 @@
 #include <algorithm>
 
 #include "common/assert.h"
-#include "common/settings.h"
 #include "video_core/compatible_formats.h"
 #include "video_core/surface.h"
 #include "video_core/texture_cache/formatter.h"
@@ -26,8 +25,7 @@ ImageViewBase::ImageViewBase(const ImageViewInfo& info, const ImageInfo& image_i
     ASSERT_MSG(VideoCore::Surface::IsViewCompatible(image_info.format, info.format, false, true),
                "Image view format {} is incompatible with image format {}", info.format,
                image_info.format);
-    const bool is_async = Settings::values.use_asynchronous_gpu_emulation.GetValue();
-    if (image_info.type == ImageType::Linear && is_async) {
+    if (image_info.forced_flushed) {
         flags |= ImageViewFlagBits::PreemtiveDownload;
     }
     if (image_info.type == ImageType::e3D && info.type != ImageViewType::e3D) {
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index b5297e76b..e1198dcf8 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -491,6 +491,32 @@ void TextureCache<P>::DownloadMemory(VAddr cpu_addr, size_t size) {
 }
 
 template <class P>
+std::optional<VideoCore::RasterizerDownloadArea> TextureCache<P>::GetFlushArea(VAddr cpu_addr,
+                                                                               u64 size) {
+    std::optional<VideoCore::RasterizerDownloadArea> area{};
+    ForEachImageInRegion(cpu_addr, size, [&](ImageId, ImageBase& image) {
+        if (False(image.flags & ImageFlagBits::GpuModified)) {
+            return;
+        }
+        if (!area) {
+            area.emplace();
+            area->start_address = cpu_addr;
+            area->end_address = cpu_addr + size;
+            area->preemtive = true;
+        }
+        area->start_address = std::min(area->start_address, image.cpu_addr);
+        area->end_address = std::max(area->end_address, image.cpu_addr_end);
+        for (auto image_view_id : image.image_view_ids) {
+            auto& image_view = slot_image_views[image_view_id];
+            image_view.flags |= ImageViewFlagBits::PreemtiveDownload;
+        }
+        area->preemtive &= image.info.forced_flushed;
+        image.info.forced_flushed = true;
+    });
+    return area;
+}
+
+template <class P>
 void TextureCache<P>::UnmapMemory(VAddr cpu_addr, size_t size) {
     std::vector<ImageId> deleted_images;
     ForEachImageInRegion(cpu_addr, size, [&](ImageId id, Image&) { deleted_images.push_back(id); });
@@ -683,6 +709,7 @@ void TextureCache<P>::CommitAsyncFlushes() {
                 download_info.async_buffer_id = last_async_buffer_id;
             }
         }
+
         if (any_none_dma) {
             auto download_map = runtime.DownloadStagingBuffer(total_size_bytes, true);
             for (const PendingDownload& download_info : download_ids) {
@@ -695,6 +722,7 @@ void TextureCache<P>::CommitAsyncFlushes() {
             }
             uncommitted_async_buffers.emplace_back(download_map);
         }
+
         async_buffers.emplace_back(std::move(uncommitted_async_buffers));
         uncommitted_async_buffers.clear();
     }
@@ -783,17 +811,22 @@ void TextureCache<P>::PopAsyncFlushes() {
 }
 
 template <class P>
-ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand) {
+ImageId TextureCache<P>::DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload) {
     const ImageInfo dst_info(operand);
     const ImageId dst_id = FindDMAImage(dst_info, operand.address);
     if (!dst_id) {
         return NULL_IMAGE_ID;
     }
-    const auto& image = slot_images[dst_id];
+    auto& image = slot_images[dst_id];
     if (False(image.flags & ImageFlagBits::GpuModified)) {
         // No need to waste time on an image that's synced with guest
         return NULL_IMAGE_ID;
     }
+    if (!is_upload && !image.info.dma_downloaded) {
+        // Force a full sync.
+        image.info.dma_downloaded = true;
+        return NULL_IMAGE_ID;
+    }
     const auto base = image.TryFindBase(operand.address);
     if (!base) {
         return NULL_IMAGE_ID;
@@ -1290,7 +1323,6 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
             all_siblings.push_back(overlap_id);
         } else {
             bad_overlap_ids.push_back(overlap_id);
-            overlap.flags |= ImageFlagBits::BadOverlap;
         }
     };
     ForEachImageInRegion(cpu_addr, size_bytes, region_check);
@@ -1401,7 +1433,12 @@ ImageId TextureCache<P>::JoinImages(const ImageInfo& info, GPUVAddr gpu_addr, VA
         ImageBase& aliased = slot_images[aliased_id];
         aliased.overlapping_images.push_back(new_image_id);
         new_image.overlapping_images.push_back(aliased_id);
-        new_image.flags |= ImageFlagBits::BadOverlap;
+        if (aliased.info.resources.levels == 1 && aliased.overlapping_images.size() > 1) {
+            aliased.flags |= ImageFlagBits::BadOverlap;
+        }
+        if (new_image.info.resources.levels == 1 && new_image.overlapping_images.size() > 1) {
+            new_image.flags |= ImageFlagBits::BadOverlap;
+        }
     }
     RegisterImage(new_image_id);
     return new_image_id;
diff --git a/src/video_core/texture_cache/texture_cache_base.h b/src/video_core/texture_cache/texture_cache_base.h
index 758b7e212..0720494e5 100644
--- a/src/video_core/texture_cache/texture_cache_base.h
+++ b/src/video_core/texture_cache/texture_cache_base.h
@@ -179,6 +179,8 @@ public:
     /// Download contents of host images to guest memory in a region
     void DownloadMemory(VAddr cpu_addr, size_t size);
 
+    std::optional<VideoCore::RasterizerDownloadArea> GetFlushArea(VAddr cpu_addr, u64 size);
+
     /// Remove images in a region
     void UnmapMemory(VAddr cpu_addr, size_t size);
 
@@ -205,7 +207,7 @@ public:
     /// Pop asynchronous downloads
     void PopAsyncFlushes();
 
-    [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand);
+    [[nodiscard]] ImageId DmaImageId(const Tegra::DMA::ImageOperand& operand, bool is_upload);
 
     [[nodiscard]] std::pair<Image*, BufferImageCopy> DmaBufferImageCopy(
         const Tegra::DMA::ImageCopy& copy_info, const Tegra::DMA::BufferOperand& buffer_operand,