diff options
| -rw-r--r-- | source/slang/glsl.meta.slang | 10 | ||||
| -rw-r--r-- | source/slang/hlsl.meta.slang | 22 | ||||
| -rw-r--r-- | tests/hlsl-intrinsic/countbits.slang | 27 |
3 files changed, 50 insertions, 9 deletions
diff --git a/source/slang/glsl.meta.slang b/source/slang/glsl.meta.slang index 523d9af90..f18e07fbd 100644 --- a/source/slang/glsl.meta.slang +++ b/source/slang/glsl.meta.slang @@ -1402,15 +1402,7 @@ __generic<let N:int> [require(cpp_cuda_glsl_hlsl_spirv, shader5_sm_5_0)] public vector<uint,N> bitCount(vector<uint,N> value) { - __target_switch - { - case glsl: __intrinsic_asm "bitCount"; - case spirv: return spirv_asm { - result:$$vector<uint,N> = OpBitCount $value - }; - default: - VECTOR_MAP_UNARY(uint, N, countbits, value); - } + return countbits(value); } [__readNone] diff --git a/source/slang/hlsl.meta.slang b/source/slang/hlsl.meta.slang index 909bebb0d..0a440f986 100644 --- a/source/slang/hlsl.meta.slang +++ b/source/slang/hlsl.meta.slang @@ -6917,6 +6917,7 @@ vector<T,N> cospi(vector<T,N> x) // Population count [__readNone] +[ForceInline] [require(cpp_cuda_glsl_hlsl_metal_spirv, shader5_sm_5_0)] uint countbits(uint value) { @@ -6936,6 +6937,27 @@ uint countbits(uint value) } } +__generic <let N : int> +[__readNone] +[ForceInline] +[require(cpp_cuda_glsl_hlsl_metal_spirv, shader5_sm_5_0)] +vector<uint, N> countbits(vector<uint, N> value) +{ + __target_switch + { + case hlsl: + __intrinsic_asm "countbits"; + case glsl: + __intrinsic_asm "bitCount"; + case metal: + __intrinsic_asm "popcount"; + case spirv: + return spirv_asm {OpBitCount $$vector<uint, N> result $value}; + default: + VECTOR_MAP_UNARY(uint, N, countbits, value); + } +} + // Cross product // TODO: SPIRV does not support integer vectors. __generic<T : __BuiltinFloatingPointType> diff --git a/tests/hlsl-intrinsic/countbits.slang b/tests/hlsl-intrinsic/countbits.slang new file mode 100644 index 000000000..da6828e87 --- /dev/null +++ b/tests/hlsl-intrinsic/countbits.slang @@ -0,0 +1,27 @@ +//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHK):-slang -compute -cpu +//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHK):-slang -compute -dx11 +//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHK):-slang -compute -dx12 +//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHK):-slang -vk -compute +//DISABLE_TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHK):-slang -cuda -compute +//TEST(compute):COMPARE_COMPUTE_EX(filecheck-buffer=CHK):-slang -mtl -compute + +//CHK:1 + +//TEST_INPUT:ubuffer(data=[0], stride=4):out,name=outputBuffer +RWStructuredBuffer<int> outputBuffer; + +[numthreads(1, 1, 1)] +void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) +{ + uint r1 = countbits(0b1); + uint2 r2 = countbits(uint2(0b0, 0b1)); + uint3 r3 = countbits(uint3(0b0, 0b1, 0b11)); + uint4 r4 = countbits(uint4(0b0, 0b1, 0b11, 0b111)); + + outputBuffer[0] = true + && (r1 == 1) + && (r2.x == 0 && r2.y == 1) + && (r3.x == 0 && r3.y == 1 && r3.z == 2) + && (r4.x == 0 && r4.y == 1 && r4.z == 2 && r4.w == 3) + ; +} |
