diff options
Diffstat (limited to 'source/slang')
| -rw-r--r-- | source/slang/hlsl.meta.slang | 58 |
1 files changed, 57 insertions, 1 deletions
diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang index 329a73a33..46851269f 100644 --- a/source/slang/hlsl.meta.slang +++ b/source/slang/hlsl.meta.slang @@ -56,9 +56,12 @@ struct ByteAddressBuffer __target_intrinsic(glsl, "atomicAdd($0, $1)") __glsl_version(430) __glsl_extension(GL_EXT_shader_atomic_float) -//__glsl_extension(GL_EXT_gpu_shader5) float __atomicAdd(__ref float value, float amount); +// Helper for hlsl, using nvAPI +__target_intrinsic(hlsl, "NvInterlockedAddUint64($0, $1, $2)") +uint2 __atomicAdd(RWByteAddressBuffer buf, uint offset, uint2); + // Int versions require glsl 4.30 // https://www.khronos.org/registry/OpenGL-Refpages/gl4/html/atomicAdd.xhtml @@ -70,6 +73,10 @@ __target_intrinsic(glsl, "atomicAdd($0, $1)") __glsl_version(430) uint __atomicAdd(__ref uint value, uint amount); +__target_intrinsic(glsl, "atomicAdd($0, $1)") +__glsl_version(430) +__glsl_extension(GL_EXT_shader_atomic_int64) +int64_t __atomicAdd(__ref int64_t value, int64_t amount); __intrinsic_op($(kIROp_ByteAddressBufferLoad)) T __byteAddressBufferLoad<T>(ByteAddressBuffer buffer, int offset); @@ -192,6 +199,9 @@ ${{{{ // NvAPI support on DX // NOTE! To use this feature on HLSL, the shader needs to include 'nvHLSLExtns.h' from the NvAPI SDK // + + // Fp32 + __target_intrinsic(hlsl, "($3 = NvInterlockedAddFp32($0, $1, $2))") __target_intrinsic(cuda, "(*$3 = atomicAdd((float*)$0._getPtrAt($1), $2))") void InterlockedAddFp32(uint byteAddress, float valueToAdd, out float originalValue); @@ -203,6 +213,8 @@ ${{{{ originalValue = __atomicAdd(buf[byteAddress / 4], valueToAdd); } + // Without returning original value + __target_intrinsic(hlsl, "(NvInterlockedAddFp32($0, $1, $2))") __target_intrinsic(cuda, "atomicAdd((float*)$0._getPtrAt($1), $2)") void InterlockedAddFp32(uint byteAddress, float valueToAdd); @@ -214,6 +226,50 @@ ${{{{ __atomicAdd(buf[byteAddress / 4], valueToAdd); } + // Int64 + __cuda_sm_version(6.0) + __target_intrinsic(cuda, "(*$3 = atomicAdd((uint64_t*)$0._getPtrAt($1), $2))") + void InterlockedAddI64(uint byteAddress, int64_t valueToAdd, out int64_t originalValue); + + __specialized_for_target(hlsl) + void InterlockedAddI64(uint byteAddress, int64_t inValueToAdd, out int64_t outOriginalValue) + { + uint2 valueToAdd; + valueToAdd.x = uint(inValueToAdd); + valueToAdd.y = uint(uint64_t(inValueToAdd) >> 32); + + const uint2 originalValue = __atomicAdd(this, byteAddress, valueToAdd); + outOriginalValue = (int64_t(originalValue.y) << 32) | originalValue.x; + } + + __specialized_for_target(glsl) + void InterlockedAddI64(uint byteAddress, int64_t valueToAdd, out int64_t originalValue) + { + RWStructuredBuffer<int64_t> buf = __getEquivalentStructuredBuffer<int64_t>(this); + originalValue = __atomicAdd(buf[byteAddress / 8], valueToAdd); + } + + // Without returning original value + __cuda_sm_version(6.0) + __target_intrinsic(cuda, "atomicAdd((uint64_t*)$0._getPtrAt($1), $2)") + void InterlockedAddI64(uint byteAddress, int64_t valueToAdd); + + __specialized_for_target(hlsl) + void InterlockedAddI64(uint byteAddress, int64_t inValueToAdd) + { + uint2 valueToAdd; + valueToAdd.x = uint(inValueToAdd); + valueToAdd.y = uint(uint64_t(inValueToAdd) >> 32); + __atomicAdd(this, byteAddress, valueToAdd); + } + + __specialized_for_target(glsl) + void InterlockedAddI64(uint byteAddress, int64_t valueToAdd) + { + RWStructuredBuffer<int64_t> buf = __getEquivalentStructuredBuffer<int64_t>(this); + __atomicAdd(buf[byteAddress / 8], valueToAdd); + } + ${{{{ } }}}} |