Support for HW format conversions for RWTexture on CUDA (#1840)

* #include an absolute path didn't work - because paths were taken to always be relative.

* Fix for writing to RWTexture with half types on CUDA.

* CUDA half functionality doc updates.

* First pass support for sust.p RWTexture format conversion on write.

* Tidy up implementation of $C.
Made clamping mode #define able.

* A simple test for RWTexture CUDA format conversion.

1 files changed, 32 insertions, 1 deletions

diff --git a/prelude/slang-cuda-prelude.h b/prelude/slang-cuda-prelude.h
index a00e8f744..a439d274a 100644
--- a/prelude/slang-cuda-prelude.h
+++ b/prelude/slang-cuda-prelude.h
@@ -63,6 +63,15 @@
  
 #ifndef SLANG_CUDA_BOUNDARY_MODE
 #   define SLANG_CUDA_BOUNDARY_MODE cudaBoundaryModeZero
+
+// Can be one of SLANG_CUDA_PTX_BOUNDARY_MODE. Only applies *PTX* emitted CUDA operations
+// which currently is just RWTextureRW format writes
+// 
+// .trap         causes an execution trap on out-of-bounds addresses
+// .clamp        stores data at the nearest surface location (sized appropriately)
+// .zero         drops stores to out-of-bounds addresses 
+
+#   define SLANG_PTX_BOUNDARY_MODE "zero"
 #endif
 
 struct TypeInfo
@@ -371,9 +380,31 @@ SLANG_SURFACE_WRITE(surf1DLayeredwrite, (int x, int layer), (x, layer))
 SLANG_SURFACE_WRITE(surf2DLayeredwrite, (int x, int y, int layer), (x, y, layer))
 SLANG_SURFACE_WRITE(surfCubemapwrite, (int x, int y, int face), (x, y, face))
 SLANG_SURFACE_WRITE(surfCubemapLayeredwrite, (int x, int y, int layerFace), (x, y, layerFace))
-     
+
 #endif
 
+// Support for doing format conversion when writing to a surface/RWTexture
+
+template <typename T>
+SLANG_FORCE_INLINE SLANG_CUDA_CALL void surf1Dwrite_convert(T, cudaSurfaceObject_t surfObj, int x, cudaSurfaceBoundaryMode boundaryMode);
+template <typename T>
+SLANG_FORCE_INLINE SLANG_CUDA_CALL void surf2Dwrite_convert(T, cudaSurfaceObject_t surfObj, int x, int y, cudaSurfaceBoundaryMode boundaryMode);
+
+// https://docs.nvidia.com/cuda/inline-ptx-assembly/index.html
+// https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#surface-instructions-sust
+
+template <>
+SLANG_FORCE_INLINE SLANG_CUDA_CALL void surf1Dwrite_convert<float>(float v, cudaSurfaceObject_t surfObj, int x, cudaSurfaceBoundaryMode boundaryMode)
+{
+    asm volatile ( "{sust.p.1d.b32." SLANG_PTX_BOUNDARY_MODE " [%0, {%1}], {%2};}\n\t" :: "l"(surfObj),"r"(x),"f"(v));     
+}
+ 
+template <>
+SLANG_FORCE_INLINE SLANG_CUDA_CALL void surf2Dwrite_convert<float>(float v, cudaSurfaceObject_t surfObj, int x, int y, cudaSurfaceBoundaryMode boundaryMode)
+{
+    asm volatile ( "{sust.p.2d.b32." SLANG_PTX_BOUNDARY_MODE " [%0, {%1,%2}], {%3};}\n\t" :: "l"(surfObj),"r"(x),"r"(y),"f"(v));
+}
+
 // ----------------------------- F32 -----------------------------------------
 
 // Unary