move UI rect fetch to vertex shader

src/gpu/gpu_shared.cgh

@@ -83,31 +83,33 @@ Enum(G_BasicSamplerKind)
 //~ Resource dereference
 
 #if IsGpu
-  //- Scalar/Uniform dereference
+  // NOTE: Uniform dereferencing is faster than Non-Uniform on AMD hardware
-  SamplerState                               G_SDeref(G_SamplerStateRef r)         { u32 idx = r.v; return SamplerDescriptorHeap[idx]; }
-  template<typename T> StructuredBuffer<T>   G_SDeref(G_StructuredBufferRef r)     { u32 idx = r.v; return ResourceDescriptorHeap[idx]; }
-  ByteAddressBuffer                          G_SDeref(G_ByteAddressBufferRef r)    { u32 idx = r.v; return ResourceDescriptorHeap[idx]; }
-  template<typename T> Texture1D<T>          G_SDeref(G_Texture1DRef r)            { u32 idx = r.v; return ResourceDescriptorHeap[idx]; }
-  template<typename T> Texture2D<T>          G_SDeref(G_Texture2DRef r)            { u32 idx = r.v; return ResourceDescriptorHeap[idx]; }
-  template<typename T> Texture3D<T>          G_SDeref(G_Texture3DRef r)            { u32 idx = r.v; return ResourceDescriptorHeap[idx]; }
-  template<typename T> RWStructuredBuffer<T> G_SDerefRW(G_StructuredBufferRef r)   { u32 idx = r.v + 1; return ResourceDescriptorHeap[idx]; }
-  RWByteAddressBuffer                        G_SDerefRW(G_ByteAddressBufferRef r)  { u32 idx = r.v + 1; return ResourceDescriptorHeap[idx]; }
-  template<typename T> RWTexture1D<T>        G_SDerefRW(G_Texture1DRef r)          { u32 idx = r.v + 1; return ResourceDescriptorHeap[idx]; }
-  template<typename T> RWTexture2D<T>        G_SDerefRW(G_Texture2DRef r)          { u32 idx = r.v + 1; return ResourceDescriptorHeap[idx]; }
-  template<typename T> RWTexture3D<T>        G_SDerefRW(G_Texture3DRef r)          { u32 idx = r.v + 1; return ResourceDescriptorHeap[idx]; }
 
-  //- Vector/Non-Uniform dereference (slower on AMD)
+  //- Scalar/Uniform dereference
-  SamplerState                               G_VDeref(G_SamplerStateRef r)         { u32 idx = r.v; return SamplerDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  SamplerState                               G_SDeref(G_SamplerStateRef r)         { return SamplerDescriptorHeap[r.v]; }
-  template<typename T> StructuredBuffer<T>   G_VDeref(G_StructuredBufferRef r)     { u32 idx = r.v; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> StructuredBuffer<T>   G_SDeref(G_StructuredBufferRef r)     { return ResourceDescriptorHeap[r.v]; }
-  ByteAddressBuffer                          G_VDeref(G_ByteAddressBufferRef r)    { u32 idx = r.v; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  ByteAddressBuffer                          G_SDeref(G_ByteAddressBufferRef r)    { return ResourceDescriptorHeap[r.v]; }
-  template<typename T> Texture1D<T>          G_VDeref(G_Texture1DRef r)            { u32 idx = r.v; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> Texture1D<T>          G_SDeref(G_Texture1DRef r)            { return ResourceDescriptorHeap[r.v]; }
-  template<typename T> Texture2D<T>          G_VDeref(G_Texture2DRef r)            { u32 idx = r.v; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> Texture2D<T>          G_SDeref(G_Texture2DRef r)            { return ResourceDescriptorHeap[r.v]; }
-  template<typename T> Texture3D<T>          G_VDeref(G_Texture3DRef r)            { u32 idx = r.v; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> Texture3D<T>          G_SDeref(G_Texture3DRef r)            { return ResourceDescriptorHeap[r.v]; }
-  template<typename T> RWStructuredBuffer<T> G_VDerefRW(G_StructuredBufferRef r)   { u32 idx = r.v + 1; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> RWStructuredBuffer<T> G_SDerefRW(G_StructuredBufferRef r)   { return ResourceDescriptorHeap[r.v + 1]; }
-  RWByteAddressBuffer                        G_VDerefRW(G_ByteAddressBufferRef r)  { u32 idx = r.v + 1; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  RWByteAddressBuffer                        G_SDerefRW(G_ByteAddressBufferRef r)  { return ResourceDescriptorHeap[r.v + 1]; }
-  template<typename T> RWTexture1D<T>        G_VDerefRW(G_Texture1DRef r)          { u32 idx = r.v + 1; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> RWTexture1D<T>        G_SDerefRW(G_Texture1DRef r)          { return ResourceDescriptorHeap[r.v + 1]; }
-  template<typename T> RWTexture2D<T>        G_VDerefRW(G_Texture2DRef r)          { u32 idx = r.v + 1; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> RWTexture2D<T>        G_SDerefRW(G_Texture2DRef r)          { return ResourceDescriptorHeap[r.v + 1]; }
-  template<typename T> RWTexture3D<T>        G_VDerefRW(G_Texture3DRef r)          { u32 idx = r.v + 1; return ResourceDescriptorHeap[NonUniformResourceIndex(idx)]; }
+  template<typename T> RWTexture3D<T>        G_SDerefRW(G_Texture3DRef r)          { return ResourceDescriptorHeap[r.v + 1]; }
+
+  //- Vector/Non-Uniform dereference
+  SamplerState                               G_VDeref(G_SamplerStateRef r)         { return SamplerDescriptorHeap[NonUniformResourceIndex(r.v)]; }
+  template<typename T> StructuredBuffer<T>   G_VDeref(G_StructuredBufferRef r)     { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v)]; }
+  ByteAddressBuffer                          G_VDeref(G_ByteAddressBufferRef r)    { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v)]; }
+  template<typename T> Texture1D<T>          G_VDeref(G_Texture1DRef r)            { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v)]; }
+  template<typename T> Texture2D<T>          G_VDeref(G_Texture2DRef r)            { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v)]; }
+  template<typename T> Texture3D<T>          G_VDeref(G_Texture3DRef r)            { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v)]; }
+  template<typename T> RWStructuredBuffer<T> G_VDerefRW(G_StructuredBufferRef r)   { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v + 1)]; }
+  RWByteAddressBuffer                        G_VDerefRW(G_ByteAddressBufferRef r)  { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v + 1)]; }
+  template<typename T> RWTexture1D<T>        G_VDerefRW(G_Texture1DRef r)          { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v + 1)]; }
+  template<typename T> RWTexture2D<T>        G_VDerefRW(G_Texture2DRef r)          { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v + 1)]; }
+  template<typename T> RWTexture3D<T>        G_VDerefRW(G_Texture3DRef r)          { return ResourceDescriptorHeap[NonUniformResourceIndex(r.v + 1)]; }
 #endif
 
 ////////////////////////////////////////////////////////////

src/meta/meta.c

@@ -563,6 +563,7 @@ void M_BuildEntryPoint(WaveLaneCtx *lane)
       PushStringToList(perm, &cp.warnings_dxc, Lit("-Wno-unused-local-typedef"));
       PushStringToList(perm, &cp.warnings_dxc, Lit("-Wno-conversion"));
       PushStringToList(perm, &cp.warnings_dxc, Lit("-Wno-switch"));
+      // PushStringToList(perm, &cp.warnings_dxc, Lit("-Wno-inline-asm"));  // Disables false-positive "Gradient operations are not affected by wave-sensitive data or control flow."
     }
   }
 

src/pp/pp_vis/pp_vis_gpu.g

@@ -192,7 +192,7 @@ ImplComputeShader2D(V_BackdropDownCS)
   {
     bd_up = G_SDeref<Vec4>(frame.backdrop_mips[mip_idx - 1]);
   }
-  RWTexture2D<Vec4> bd_down = G_VDerefRW<Vec4>(frame.backdrop_mips[mip_idx]);
+  RWTexture2D<Vec4> bd_down = G_SDerefRW<Vec4>(frame.backdrop_mips[mip_idx]);
 
   Vec2 down_dims = countof(bd_down);
 
@@ -316,9 +316,9 @@ ImplVertexShader(V_QuadVS, V_QuadPSInput)
 
   V_QuadPSInput result;
   result.sv_position = Vec4(NdcFromPos(screen_pos, frame.screen_dims).xy, 0, 1);
-  result.quad_idx = SV_InstanceID;
   result.world_pos = world_pos;
   result.samp_uv = samp_uv;
+  result.quad = quad;
   return result;
 }
 
@@ -328,11 +328,10 @@ ImplVertexShader(V_QuadVS, V_QuadPSInput)
 ImplPixelShader(V_QuadPS, V_QuadPSOutput, V_QuadPSInput input)
 {
   V_SharedFrame frame = G_SDeref<V_SharedFrame>(V_GpuConst_Frame)[0];
-  StructuredBuffer<V_Quad> quads = G_SDeref<V_Quad>(frame.quads);
   SamplerState sampler = G_SDeref(frame.basic_samplers[G_BasicSamplerKind_PointClamp]);
   RWTexture2D<u32> occluders = G_SDerefRW<u32>(frame.occluders);
 
-  V_Quad quad = quads[input.quad_idx];
+  V_Quad quad = input.quad;
   Texture2D<Vec4> tex = G_VDeref<Vec4>(quad.tex);
 
   Vec2 world_pos = input.world_pos;
@@ -1173,7 +1172,7 @@ ImplComputeShader2D(V_BloomDownCS)
     }
   }
 
-  if (IsInside(bloom_pos, down_dims))
+  if (all(bloom_pos < down_dims))
   {
     bloom_down[bloom_pos] = result;
   }

src/pp/pp_vis/pp_vis_gpu.gh

@@ -4,9 +4,9 @@
 Struct(V_QuadPSInput)
 {
   Semantic(Vec4, sv_position);
-  Semantic(nointerpolation u32, quad_idx);
   Semantic(Vec2, world_pos);
   Semantic(Vec2, samp_uv);
+  Semantic(nointerpolation V_Quad, quad);
 };
 
 Struct(V_QuadPSOutput)

src/ui/ui_gpu.g

@@ -8,8 +8,8 @@ ImplVertexShader(UI_DRectVS, UI_DRectPSInput)
 {
   UI_GpuParams params = G_SDeref<UI_GpuParams>(UI_GpuConst_Params)[0];
   StructuredBuffer<UI_GpuRect> rects = G_SDeref<UI_GpuRect>(params.rects);
-  UI_GpuRect rect = rects[SV_InstanceID];
 
+  UI_GpuRect rect = rects[SV_InstanceID];
   Vec2 rect_uv = RectUvFromIdx(SV_VertexID);
   Vec2 tex_uv = lerp(rect.tex_slice_uv.p0, rect.tex_slice_uv.p1, rect_uv);
   Vec2 target_pos = lerp(rect.bounds.p0, rect.bounds.p1, rect_uv);
@@ -17,15 +17,13 @@ ImplVertexShader(UI_DRectVS, UI_DRectPSInput)
   UI_DRectPSInput result;
   {
     result.sv_position = Vec4(NdcFromPos(target_pos, Vec2(params.target_size).xy), 0, 1);
-    result.rect_idx = SV_InstanceID;
-
     result.base_background_premul = Premul(rect.background_lin);
     result.base_border_premul = Premul(rect.border_lin);
     result.tint_premul = Premul(rect.tint_lin);
     result.debug_premul = Premul(rect.debug_lin);
-
     result.rect_uv = rect_uv;
     result.tex_uv = tex_uv;
+    result.rect = rect;
   }
   return result;
 }
@@ -36,13 +34,11 @@ ImplVertexShader(UI_DRectVS, UI_DRectPSInput)
 ImplPixelShader(UI_DRectPS, UI_DRectPSOutput, UI_DRectPSInput input)
 {
   UI_GpuParams params = G_SDeref<UI_GpuParams>(UI_GpuConst_Params)[0];
-  StructuredBuffer<UI_GpuRect> rects = G_SDeref<UI_GpuRect>(params.rects);
   SamplerState sampler = G_SDeref(params.sampler);
 
-  UI_GpuRect rect = rects[input.rect_idx];
+  UI_GpuRect rect = input.rect;
-
-  Vec2 p = input.sv_position.xy;
   Vec2 rect_uv = input.rect_uv;
+  Vec2 p = input.sv_position.xy;
   Vec2 p0 = rect.bounds.p0;
   Vec2 p1 = rect.bounds.p1;
 

src/ui/ui_gpu.gh

@@ -4,13 +4,13 @@
 Struct(UI_DRectPSInput)
 {
   Semantic(Vec4, sv_position);
-  Semantic(nointerpolation u32, rect_idx);
   Semantic(Vec4, base_background_premul);
   Semantic(Vec4, base_border_premul);
   Semantic(Vec4, tint_premul);
   Semantic(Vec4, debug_premul);
   Semantic(Vec2, rect_uv);
   Semantic(Vec2, tex_uv);
+  nointerpolation Semantic(UI_GpuRect, rect);
 };
 
 Struct(UI_DRectPSOutput)