From b18c1fb1bbbcc235e9a1cbbb2704fed2e5895e61 Mon Sep 17 00:00:00 2001
From: Ameer J <52414509+ameerj@users.noreply.github.com>
Date: Sun, 6 Aug 2023 14:49:34 -0400
Subject: [PATCH] HACK: Avoid swizzling and reuploading ASTC image every frame

---
 .../renderer_opengl/gl_texture_cache.cpp           |  9 ++++++++-
 .../renderer_vulkan/vk_staging_buffer_pool.cpp     |  1 +
 .../renderer_vulkan/vk_texture_cache.cpp           | 10 +++++++++-
 src/video_core/texture_cache/texture_cache.h       | 14 ++++++++++++--
 src/video_core/textures/decoders.cpp               |  9 +++++++++
 5 files changed, 39 insertions(+), 4 deletions(-)

diff --git a/src/video_core/renderer_opengl/gl_texture_cache.cpp b/src/video_core/renderer_opengl/gl_texture_cache.cpp
index 9cafd2983..9ca544ec9 100644
--- a/src/video_core/renderer_opengl/gl_texture_cache.cpp
+++ b/src/video_core/renderer_opengl/gl_texture_cache.cpp
@@ -554,7 +554,14 @@ void TextureCacheRuntime::Finish() {
 }
 
 StagingBufferMap TextureCacheRuntime::UploadStagingBuffer(size_t size) {
-    return staging_buffer_pool.RequestUploadBuffer(size);
+    static StagingBufferMap result;
+    static size_t last_size = 0;
+    if (size == last_size) {
+        return result;
+    }
+    last_size = size;
+    result = staging_buffer_pool.RequestUploadBuffer(size);
+    return result;
 }
 
 StagingBufferMap TextureCacheRuntime::DownloadStagingBuffer(size_t size) {
diff --git a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
index ce92f66ab..27328dab2 100644
--- a/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
+++ b/src/video_core/renderer_vulkan/vk_staging_buffer_pool.cpp
@@ -214,6 +214,7 @@ StagingBufferPool::StagingBuffersCache& StagingBufferPool::GetCache(MemoryUsage
 }
 
 void StagingBufferPool::ReleaseCache(MemoryUsage usage) {
+    return;
     ReleaseLevel(GetCache(usage), current_delete_level);
 }
 
diff --git a/src/video_core/renderer_vulkan/vk_texture_cache.cpp b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
index b3e17c332..51be57b19 100644
--- a/src/video_core/renderer_vulkan/vk_texture_cache.cpp
+++ b/src/video_core/renderer_vulkan/vk_texture_cache.cpp
@@ -846,7 +846,15 @@ void TextureCacheRuntime::Finish() {
 }
 
 StagingBufferRef TextureCacheRuntime::UploadStagingBuffer(size_t size) {
-    return staging_buffer_pool.Request(size, MemoryUsage::Upload);
+    static StagingBufferRef result;
+    static size_t last_size = 0;
+    if (size == last_size) {
+        return result;
+    }
+    LOG_ERROR(Debug, "Called");
+    last_size = size;
+    result = staging_buffer_pool.Request(size, MemoryUsage::Upload);
+    return result;
 }
 
 StagingBufferRef TextureCacheRuntime::DownloadStagingBuffer(size_t size, bool deferred) {
diff --git a/src/video_core/texture_cache/texture_cache.h b/src/video_core/texture_cache/texture_cache.h
index 4457b366f..6b87987de 100644
--- a/src/video_core/texture_cache/texture_cache.h
+++ b/src/video_core/texture_cache/texture_cache.h
@@ -1021,8 +1021,18 @@ void TextureCache<P>::UploadImageContents(Image& image, StagingBuffer& staging)
     const GPUVAddr gpu_addr = image.gpu_addr;
 
     if (True(image.flags & ImageFlagBits::AcceleratedUpload)) {
-        gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
-                              VideoCommon::CacheType::NoTextureCache);
+        static u64 last_size = 0;
+        bool has_run = false;
+        if (last_size == image.unswizzled_size_bytes) {
+            has_run = true;
+        }
+        last_size = image.unswizzled_size_bytes;
+
+        if (!has_run) {
+            LOG_ERROR(Debug, "Called");
+            gpu_memory->ReadBlock(gpu_addr, mapped_span.data(), mapped_span.size_bytes(),
+                                  VideoCommon::CacheType::NoTextureCache);
+        }
         const auto uploads = FullUploadSwizzles(image.info);
         runtime.AccelerateImageUpload(image, staging, uploads);
         return;
diff --git a/src/video_core/textures/decoders.cpp b/src/video_core/textures/decoders.cpp
index 95bcdd37b..dde4b0904 100644
--- a/src/video_core/textures/decoders.cpp
+++ b/src/video_core/textures/decoders.cpp
@@ -88,6 +88,7 @@ void SwizzleImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32
     }
 }
 
+
 template <bool TO_LINEAR, u32 BYTES_PER_PIXEL>
 void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 width, u32 height,
                         u32 depth, u32 origin_x, u32 origin_y, u32 extent_x, u32 num_lines,
@@ -95,6 +96,14 @@ void SwizzleSubrectImpl(std::span<u8> output, std::span<const u8> input, u32 wid
     // The origin of the transformation can be configured here, leave it as zero as the current API
     // doesn't expose it.
     static constexpr u32 origin_z = 0;
+    static u32 last_width = 0;
+    static u32 last_height = 0;
+    if (last_width == width && last_height == height) {
+        return;
+    }
+    LOG_ERROR(Debug, "Called");
+    last_width = width;
+    last_height = height;
 
     // We can configure here a custom pitch
     // As it's not exposed 'width * BYTES_PER_PIXEL' will be the expected pitch.