avoid invalidating textures by hashing it's value

also increase upload buffer size once again
RSDuck · Jul 15, 2021 · 8d56af7 · 8d56af7
1 parent 0666d4e
commit 8d56af7
Show file tree

Hide file tree

Showing 4 changed files with 32 additions and 6 deletions.
diff --git a/src/GPU3D_Deko.cpp b/src/GPU3D_Deko.cpp
@@ -8,6 +8,9 @@
 #include <assert.h>
 #include <switch.h>
 
+#define XXH_STATIC_LINKING_ONLY
+#include "xxhash/xxhash.h"
+
 #include <arm_neon.h>
 
 using Gfx::EmuCmdBuf;
@@ -635,9 +638,10 @@ DekoRenderer::TexCacheEntry& DekoRenderer::GetTexture(u32 texParam, u32 palBase)
     if (fmt != 7)
     {
         key |= (u64)palBase << 32;
-        if (fmt != 5)
+        if (fmt == 5)
             key &= ~((u64)1 << 29);
     }
+    //printf("%" PRIx64 " %" PRIx32 " %" PRIx32 "\n", key, texParam, palBase);
 
     assert(fmt != 0 && "no texture is not a texture format!");
 
@@ -724,6 +728,8 @@ DekoRenderer::TexCacheEntry& DekoRenderer::GetTexture(u32 texParam, u32 palBase)
         u8* texData = &GPU::VRAMFlat_Texture[addr];
         u16* palData = (u16*)(GPU::VRAMFlat_TexPal + palAddr);
 
+        //assert(entry.TexPalStart+entry.TexPalSize <= 128*1024*1024);
+
         bool color0Transparent = texParam & (1 << 29);
 
         switch (fmt)
@@ -736,6 +742,14 @@ DekoRenderer::TexCacheEntry& DekoRenderer::GetTexture(u32 texParam, u32 palBase)
         }
     }
 
+    for (int i = 0; i < 2; i++)
+    {
+        if (entry.TextureRAMSize[i])
+            entry.TextureHash[i] = XXH3_64bits(&GPU::VRAMFlat_Texture[entry.TextureRAMStart[i]], entry.TextureRAMSize[i]);
+    }
+    if (entry.TexPalSize)
+        entry.TexPalHash = XXH3_64bits(&GPU::VRAMFlat_TexPal[entry.TexPalStart], entry.TexPalSize);
+
     auto& texArrays = TexArrays[widthLog2][heightLog2];
     auto& freeTextures = FreeTextures[widthLog2][heightLog2];
 
@@ -846,7 +860,12 @@ void DekoRenderer::RenderFrame()
                     for (u32 j = startEntry; j < startEntry + entriesCount; j++)
                     {
                         if (GetRangedBitMask(j, startBit, bitsCount) & textureDirty.Data[j])
-                            goto invalidate;
+                        {
+                            u64 newTexHash = XXH3_64bits(&GPU::VRAMFlat_Texture[entry.TextureRAMStart[i]], entry.TextureRAMSize[i]);
+
+                            if (newTexHash != entry.TextureHash[i])
+                                goto invalidate;
+                        }
                     }
                 }
             }
@@ -861,7 +880,11 @@ void DekoRenderer::RenderFrame()
                 for (u32 j = startEntry; j < startEntry + entriesCount; j++)
                 {
                     if (GetRangedBitMask(j, startBit, bitsCount) & texPalDirty.Data[j])
-                        goto invalidate;
+                    {
+                        u64 newPalHash = XXH3_64bits(&GPU::VRAMFlat_TexPal[entry.TexPalStart], entry.TexPalSize);
+                        if (newPalHash != entry.TexPalHash)
+                            goto invalidate;
+                    }
                 }
             }
 

diff --git a/src/GPU3D_Deko.h b/src/GPU3D_Deko.h
@@ -238,6 +238,9 @@ class DekoRenderer : public Renderer3D
         u32 TexPalStart, TexPalSize;
         u8 WidthLog2, HeightLog2;
         TexArrayEntry Texture;
+
+        u64 TextureHash[2];
+        u64 TexPalHash;
     };
     std::unordered_map<u64, TexCacheEntry> TexCache;
 

diff --git a/src/frontend/switch/UploadBuffer.cpp b/src/frontend/switch/UploadBuffer.cpp
@@ -28,8 +28,8 @@ DkGpuAddr UploadBuffer::UploadData(dk::CmdBuf cmdbuf, u32 size, u8* data)
     DkGpuAddr srcGpu = Gfx::DataHeap->GpuAddr(Buffers[CurBuffer]) + CurOffset;
     memcpy(srcCpu, data, size);
 
-    CurOffset += size + 31;
-    CurOffset &= ~31;
+    CurOffset += size + 63;
+    CurOffset &= ~63;
     return srcGpu;
 }
 

diff --git a/src/frontend/switch/UploadBuffer.h b/src/frontend/switch/UploadBuffer.h
@@ -16,7 +16,7 @@ class UploadBuffer
 
     u32 LastFlushBuffer = 0;
 private:
-    static const u32 SegmentSize = 8*1024*1024;
+    static const u32 SegmentSize = 16*1024*1024;
     GpuMemHeap::Allocation Buffers[3];
     dk::Fence Fences[3] = {};
     u32 CurBuffer = 0, CurOffset = 0;