Only dirty the uniform UVSCALEOFFSET when really needed

Broken out from #17479 With OpenGL, greatly reduces the amount of glUniform4fv calls in many games (and similar in the other backends).
hrydgard · May 25, 2023 · 80e47b7 · 80e47b7
1 parent 82934b9
commit 80e47b7
Show file tree

Hide file tree

Showing 8 changed files with 28 additions and 23 deletions.
diff --git a/GPU/Common/DrawEngineCommon.cpp b/GPU/Common/DrawEngineCommon.cpp
@@ -236,7 +236,7 @@ void DrawEngineCommon::DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex
  // Code checks this reg directly, not just the vtype ID.
  if (!prevThrough) {
  gstate.vertType |= GE_VTYPE_THROUGH;
- gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE | DIRTY_FRAGMENTSHADER_STATE | DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_UVSCALEOFFSET | DIRTY_CULLRANGE);
+ gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE | DIRTY_FRAGMENTSHADER_STATE | DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE);
  }
 
  int bytesRead;
@@ -246,7 +246,7 @@ void DrawEngineCommon::DispatchSubmitImm(GEPrimitiveType prim, TransformedVertex
 
  if (!prevThrough) {
  gstate.vertType &= ~GE_VTYPE_THROUGH;
- gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE | DIRTY_FRAGMENTSHADER_STATE | DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_UVSCALEOFFSET | DIRTY_CULLRANGE);
+ gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE | DIRTY_FRAGMENTSHADER_STATE | DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_CULLRANGE);
  }
 }
 

diff --git a/GPU/Common/ShaderId.cpp b/GPU/Common/ShaderId.cpp
@@ -377,7 +377,7 @@ void ComputeFragmentShaderID(FShaderID *id_out, const ComputedPipelineState &pip
  id.SetBit(FS_BIT_COLOR_WRITEMASK, colorWriteMask);
 
  // All framebuffers are array textures in Vulkan now.
- if (gstate_c.arrayTexture && g_Config.iGPUBackend == (int)GPUBackend::VULKAN) {
+ if (gstate_c.textureIsArray && g_Config.iGPUBackend == (int)GPUBackend::VULKAN) {
  id.SetBit(FS_BIT_SAMPLE_ARRAY_TEXTURE);
  }
 

diff --git a/GPU/Common/TextureCacheCommon.cpp b/GPU/Common/TextureCacheCommon.cpp
@@ -406,6 +406,7 @@ TexCacheEntry *TextureCacheCommon::SetTexture() {
  Unbind();
  gstate_c.SetTextureIs3D(false);
  gstate_c.SetTextureIsArray(false);
+ gstate_c.SetTextureIsFramebuffer(false);
  return nullptr;
  }
 
@@ -573,6 +574,7 @@ TexCacheEntry *TextureCacheCommon::SetTexture() {
  gstate_c.SetTextureIs3D((entry->status & TexCacheEntry::STATUS_3D) != 0);
  gstate_c.SetTextureIsArray(false);
  gstate_c.SetTextureIsBGRA((entry->status & TexCacheEntry::STATUS_BGRA) != 0);
+ gstate_c.SetTextureIsFramebuffer(false);
 
  if (rehash) {
  // Update in case any of these changed.
@@ -681,6 +683,7 @@ TexCacheEntry *TextureCacheCommon::SetTexture() {
  gstate_c.curTextureHeight = h;
  gstate_c.SetTextureIs3D((entry->status & TexCacheEntry::STATUS_3D) != 0);
  gstate_c.SetTextureIsArray(false); // Ordinary 2D textures still aren't used by array view in VK. We probably might as well, though, at this point..
+ gstate_c.SetTextureIsFramebuffer(false);
 
  failedTexture_ = false;
  nextTexture_ = entry;
@@ -1154,6 +1157,7 @@ void TextureCacheCommon::SetTextureFramebuffer(const AttachCandidate &candidate)
  gstate_c.Dirty(DIRTY_FRAGMENTSHADER_STATE);
  }
  gstate_c.SetTextureIsBGRA(false);
+ gstate_c.SetTextureIsFramebuffer(true);
  gstate_c.curTextureXOffset = fbInfo.xOffset;
  gstate_c.curTextureYOffset = fbInfo.yOffset;
  u32 texW = (u32)gstate.getTextureWidth(0);

diff --git a/GPU/Common/VertexDecoderCommon.cpp b/GPU/Common/VertexDecoderCommon.cpp
@@ -1176,7 +1176,7 @@ void VertexDecoder::SetVertexType(u32 fmt, const VertexDecoderOptions &options,
 
  steps_[numSteps_++] = morphcount == 1 ? colstep[col] : colstep_morph[col];
 
- // All color formats decode to DEC_U8_4 currently.
+ // All color formats decode to DEC_U8_4.
  // They can become floats later during transform though.
  decFmt.c0fmt = DEC_U8_4;
  decFmt.c0off = decOff;

diff --git a/GPU/Common/VertexDecoderCommon.h b/GPU/Common/VertexDecoderCommon.h
@@ -336,7 +336,7 @@ class VertexDecoder {
 
  u32 VertexType() const { return fmt_; }
 
- const DecVtxFormat &GetDecVtxFmt() { return decFmt; }
+ const DecVtxFormat &GetDecVtxFmt() const { return decFmt; }
 
  void DecodeVerts(u8 *decoded, const void *verts, int indexLowerBound, int indexUpperBound) const;
 

diff --git a/GPU/GPUCommonHW.cpp b/GPU/GPUCommonHW.cpp
@@ -149,7 +149,7 @@ const CommonCommandTableEntry commonCommandTable[] = {
  { GE_CMD_TEXFORMAT, FLAG_FLUSHBEFOREONCHANGE, DIRTY_TEXTURE_IMAGE },
  { GE_CMD_TEXLEVEL, FLAG_EXECUTEONCHANGE, DIRTY_TEXTURE_PARAMS, &GPUCommonHW::Execute_TexLevel },
  { GE_CMD_TEXLODSLOPE, FLAG_FLUSHBEFOREONCHANGE, DIRTY_TEXTURE_PARAMS },
- { GE_CMD_TEXADDR0, FLAG_FLUSHBEFOREONCHANGE, DIRTY_TEXTURE_IMAGE | DIRTY_UVSCALEOFFSET },
+ { GE_CMD_TEXADDR0, FLAG_FLUSHBEFOREONCHANGE, DIRTY_TEXTURE_IMAGE },
  { GE_CMD_TEXADDR1, FLAG_FLUSHBEFOREONCHANGE, DIRTY_TEXTURE_PARAMS },
  { GE_CMD_TEXADDR2, FLAG_FLUSHBEFOREONCHANGE, DIRTY_TEXTURE_PARAMS },
  { GE_CMD_TEXADDR3, FLAG_FLUSHBEFOREONCHANGE, DIRTY_TEXTURE_PARAMS },
@@ -827,13 +827,14 @@ void GPUCommonHW::FastRunLoop(DisplayList &list) {
 }
 
 void GPUCommonHW::Execute_VertexType(u32 op, u32 diff) {
- if (diff)
+ if (diff) {
+ // TODO: We only need to dirty vshader-state here if the output format will be different.
  gstate_c.Dirty(DIRTY_VERTEXSHADER_STATE);
- if (diff & (GE_VTYPE_TC_MASK | GE_VTYPE_THROUGH_MASK)) {
- gstate_c.Dirty(DIRTY_UVSCALEOFFSET);
- // Switching between through and non-through, we need to invalidate a bunch of stuff.
- if (diff & GE_VTYPE_THROUGH_MASK)
+
+ if (diff & GE_VTYPE_THROUGH_MASK) {
+ // Switching between through and non-through, we need to invalidate a bunch of stuff.
  gstate_c.Dirty(DIRTY_RASTER_STATE | DIRTY_VIEWPORTSCISSOR_STATE | DIRTY_FRAGMENTSHADER_STATE | DIRTY_GEOMETRYSHADER_STATE | DIRTY_CULLRANGE);
+ }
  }
 }
 
@@ -844,8 +845,6 @@ void GPUCommonHW::Execute_VertexTypeSkinning(u32 op, u32 diff) {
  gstate.vertType ^= diff;
  Flush();
  gstate.vertType ^= diff;
- if (diff & (GE_VTYPE_TC_MASK | GE_VTYPE_THROUGH_MASK))
- gstate_c.Dirty(DIRTY_UVSCALEOFFSET);
  // In this case, we may be doing weights and morphs.
  // Update any bone matrix uniforms so it uses them correctly.
  if ((op & GE_VTYPE_MORPHCOUNT_MASK) != 0) {
@@ -1125,8 +1124,6 @@ void GPUCommonHW::Execute_Prim(u32 op, u32 diff) {
 
 void GPUCommonHW::Execute_Bezier(u32 op, u32 diff) {
  // We don't dirty on normal changes anymore as we prescale, but it's needed for splines/bezier.
- gstate_c.Dirty(DIRTY_UVSCALEOFFSET);
-
  gstate_c.framebufFormat = gstate.FrameBufFormat();
 
  // This also make skipping drawing very effective.
@@ -1198,8 +1195,6 @@ void GPUCommonHW::Execute_Bezier(u32 op, u32 diff) {
 
 void GPUCommonHW::Execute_Spline(u32 op, u32 diff) {
  // We don't dirty on normal changes anymore as we prescale, but it's needed for splines/bezier.
- gstate_c.Dirty(DIRTY_UVSCALEOFFSET);
-
  gstate_c.framebufFormat = gstate.FrameBufFormat();
 
  // This also make skipping drawing very effective.
@@ -1289,7 +1284,6 @@ void GPUCommonHW::Execute_TexSize0(u32 op, u32 diff) {
  if (diff || gstate_c.IsDirty(DIRTY_TEXTURE_IMAGE | DIRTY_TEXTURE_PARAMS)) {
  gstate_c.curTextureWidth = gstate.getTextureWidth(0);
  gstate_c.curTextureHeight = gstate.getTextureHeight(0);
- gstate_c.Dirty(DIRTY_UVSCALEOFFSET);
  // We will need to reset the texture now.
  gstate_c.Dirty(DIRTY_TEXTURE_PARAMS);
  }

diff --git a/GPU/GPUState.h b/GPU/GPUState.h
@@ -569,8 +569,8 @@ struct GPUStateCache {
  }
  }
  void SetTextureIsArray(bool isArrayTexture) { // VK only
- if (arrayTexture != isArrayTexture) {
- arrayTexture = isArrayTexture;
+ if (textureIsArray != isArrayTexture) {
+ textureIsArray = isArrayTexture;
  Dirty(DIRTY_FRAGMENTSHADER_STATE);
  }
  }
@@ -580,6 +580,12 @@ struct GPUStateCache {
  Dirty(DIRTY_FRAGMENTSHADER_STATE);
  }
  }
+ void SetTextureIsFramebuffer(bool isFramebuffer) {
+ if (textureIsFramebuffer != isFramebuffer) {
+ textureIsFramebuffer = isFramebuffer;
+ Dirty(DIRTY_UVSCALEOFFSET);
+ }
+ }
  void SetUseFlags(u32 newFlags) {
  if (newFlags != useFlags_) {
  if (useFlags_ != 0)
@@ -614,7 +620,8 @@ struct GPUStateCache {
 
  bool bgraTexture;
  bool needShaderTexClamp;
- bool arrayTexture;
+ bool textureIsArray;
+ bool textureIsFramebuffer;
  bool useFlagsChanged;
 
  float morphWeights[8];

diff --git a/GPU/Vulkan/DrawEngineVulkan.cpp b/GPU/Vulkan/DrawEngineVulkan.cpp
@@ -775,7 +775,7 @@ void DrawEngineVulkan::DoFlush() {
  textureCache_->ApplyTexture();
  textureCache_->GetVulkanHandles(imageView, sampler);
  if (imageView == VK_NULL_HANDLE)
- imageView = (VkImageView)draw_->GetNativeObject(gstate_c.arrayTexture ? Draw::NativeObject::NULL_IMAGEVIEW_ARRAY : Draw::NativeObject::NULL_IMAGEVIEW);
+ imageView = (VkImageView)draw_->GetNativeObject(gstate_c.textureIsArray ? Draw::NativeObject::NULL_IMAGEVIEW_ARRAY : Draw::NativeObject::NULL_IMAGEVIEW);
  if (sampler == VK_NULL_HANDLE)
  sampler = nullSampler_;
  }
@@ -910,7 +910,7 @@ void DrawEngineVulkan::DoFlush() {
  textureCache_->ApplyTexture();
  textureCache_->GetVulkanHandles(imageView, sampler);
  if (imageView == VK_NULL_HANDLE)
- imageView = (VkImageView)draw_->GetNativeObject(gstate_c.arrayTexture ? Draw::NativeObject::NULL_IMAGEVIEW_ARRAY : Draw::NativeObject::NULL_IMAGEVIEW);
+ imageView = (VkImageView)draw_->GetNativeObject(gstate_c.textureIsArray ? Draw::NativeObject::NULL_IMAGEVIEW_ARRAY : Draw::NativeObject::NULL_IMAGEVIEW);
  if (sampler == VK_NULL_HANDLE)
  sampler = nullSampler_;
  }