From d45f9e4d26aa3205a8496893cbaf9b60efb330fe Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Mon, 13 Feb 2023 18:02:13 +0000
Subject: [PATCH] Loosen some texture WaR sync when possible

By keeping track of the stages reading the image we can do more fine-grained WaR prevention, as opposed to waiting for all commands to complete.
---
 .../cpp/skyline/gpu/interconnect/fermi_2d.cpp |  2 +-
 .../main/cpp/skyline/gpu/texture/texture.cpp  | 23 +++++++++++++------
 .../main/cpp/skyline/gpu/texture/texture.h    |  6 +++++
 3 files changed, 23 insertions(+), 8 deletions(-)

diff --git a/app/src/main/cpp/skyline/gpu/interconnect/fermi_2d.cpp b/app/src/main/cpp/skyline/gpu/interconnect/fermi_2d.cpp
index b543758e..4a4e7bee 100644
--- a/app/src/main/cpp/skyline/gpu/interconnect/fermi_2d.cpp
+++ b/app/src/main/cpp/skyline/gpu/interconnect/fermi_2d.cpp
@@ -151,7 +151,7 @@ namespace skyline::gpu::interconnect {
                 std::array<TextureView *, 1> sampledImages{srcTextureView.get()};
                 executor.AddSubpass(std::move(executionCallback), {{static_cast<i32>(dstRectX), static_cast<i32>(dstRectY)}, {dstRectWidth, dstRectHeight} },
                                     sampledImages, {}, {dst}, {}, false,
-                                    vk::PipelineStageFlagBits::eAllCommands, vk::PipelineStageFlagBits::eAllCommands);
+                                    vk::PipelineStageFlagBits::eAllGraphics, vk::PipelineStageFlagBits::eAllGraphics);
             }
         );
 
diff --git a/app/src/main/cpp/skyline/gpu/texture/texture.cpp b/app/src/main/cpp/skyline/gpu/texture/texture.cpp
index 6595d1bf..ba23cace 100644
--- a/app/src/main/cpp/skyline/gpu/texture/texture.cpp
+++ b/app/src/main/cpp/skyline/gpu/texture/texture.cpp
@@ -1007,22 +1007,31 @@ namespace skyline::gpu {
         lastRenderPassUsage = renderPassUsage;
         lastRenderPassIndex = renderPassIndex;
 
-        if (renderPassUsage == texture::RenderPassUsage::RenderTarget)
+        if (renderPassUsage == texture::RenderPassUsage::RenderTarget) {
             pendingStageMask = vk::PipelineStageFlagBits::eVertexShader |
-                               vk::PipelineStageFlagBits::eTessellationControlShader |
-                               vk::PipelineStageFlagBits::eTessellationEvaluationShader |
-                               vk::PipelineStageFlagBits::eGeometryShader |
-                               vk::PipelineStageFlagBits::eFragmentShader |
-                               vk::PipelineStageFlagBits::eComputeShader;
-        else if (renderPassUsage == texture::RenderPassUsage::None)
+                vk::PipelineStageFlagBits::eTessellationControlShader |
+                vk::PipelineStageFlagBits::eTessellationEvaluationShader |
+                vk::PipelineStageFlagBits::eGeometryShader |
+                vk::PipelineStageFlagBits::eFragmentShader |
+                vk::PipelineStageFlagBits::eComputeShader;
+            readStageMask = {};
+        } else if (renderPassUsage == texture::RenderPassUsage::None) {
             pendingStageMask = {};
+            readStageMask = {};
+        }
     }
 
     texture::RenderPassUsage Texture::GetLastRenderPassUsage() {
         return lastRenderPassUsage;
     }
 
+    vk::PipelineStageFlags Texture::GetReadStageMask() {
+        return readStageMask;
+    }
+
     void Texture::PopulateReadBarrier(vk::PipelineStageFlagBits dstStage, vk::PipelineStageFlags &srcStageMask, vk::PipelineStageFlags &dstStageMask) {
+        readStageMask |= dstStage;
+
         if (!(pendingStageMask & dstStage))
             return;
 
diff --git a/app/src/main/cpp/skyline/gpu/texture/texture.h b/app/src/main/cpp/skyline/gpu/texture/texture.h
index 63c36092..f04b70d2 100644
--- a/app/src/main/cpp/skyline/gpu/texture/texture.h
+++ b/app/src/main/cpp/skyline/gpu/texture/texture.h
@@ -411,6 +411,7 @@ namespace skyline::gpu {
         u32 lastRenderPassIndex{}; //!< The index of the last render pass that used this texture
         texture::RenderPassUsage lastRenderPassUsage{texture::RenderPassUsage::None}; //!< The type of usage in the last render pass
         vk::PipelineStageFlags pendingStageMask{}; //!< List of pipeline stages that are yet to be flushed for reads since the last time this texture was used an an RT
+        vk::PipelineStageFlags readStageMask{}; //!< Set of pipeline stages that this texture has been read in since it was last used as an RT
 
         friend TextureManager;
         friend TextureView;
@@ -613,6 +614,11 @@ namespace skyline::gpu {
          */
         texture::RenderPassUsage GetLastRenderPassUsage();
 
+        /**
+         * @return The set of stages this texture has been read in since it was last used as an RT
+         */
+        vk::PipelineStageFlags GetReadStageMask();
+
         /**
          * @brief Populates the input src and dst stage masks with appropriate read barrier parameters for the current texture state
          */