diff options
15 files changed, 154 insertions, 4 deletions
diff --git a/prelude/slang-cpp-scalar-intrinsics.h b/prelude/slang-cpp-scalar-intrinsics.h index 63fe9c926..6c577733d 100644 --- a/prelude/slang-cpp-scalar-intrinsics.h +++ b/prelude/slang-cpp-scalar-intrinsics.h @@ -3,6 +3,11 @@ #include "../slang.h" +#if SLANG_PROCESSOR_X86_64 && SLANG_VC +// If we have visual studio and 64 bit processor, we can assume we have popcnt, and can include x86 intrinsics +# include <intrin.h> +#endif + #ifdef SLANG_PRELUDE_NAMESPACE namespace SLANG_PRELUDE_NAMESPACE { #endif @@ -192,7 +197,22 @@ SLANG_FORCE_INLINE double U32_asdouble(uint32_t low, uint32_t hi) return u.d; } - +SLANG_FORCE_INLINE uint32_t U32_countbits(uint32_t v) +{ +#if SLANG_GCC_FAMILY + return __builtin_popcount(v); +#elif SLANG_PROCESSOR_X86_64 && SLANG_VC + return __popcnt(v); +#else + uint32_t c = 0; + while (v) + { + c++; + v &= v - 1; + } + return c; +#endif +} #ifdef SLANG_PRELUDE_NAMESPACE diff --git a/prelude/slang-cuda-prelude.h b/prelude/slang-cuda-prelude.h index 8d100b0db..f78814486 100644 --- a/prelude/slang-cuda-prelude.h +++ b/prelude/slang-cuda-prelude.h @@ -161,6 +161,13 @@ SLANG_CUDA_CALL double U32_asdouble(uint32_t low, uint32_t hi) return u.d; } +SLANG_CUDA_CALL uint32_t U32_countbits(uint32_t v) +{ + // https://docs.nvidia.com/cuda/cuda-math-api/group__CUDA__MATH__INTRINSIC__INT.html#group__CUDA__MATH__INTRINSIC__INT_1g43c9c7d2b9ebf202ff1ef5769989be46 + return __popc(v); +} + + /* !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! */ diff --git a/source/slang/slang-emit-cpp.cpp b/source/slang/slang-emit-cpp.cpp index db442d131..5dd028182 100644 --- a/source/slang/slang-emit-cpp.cpp +++ b/source/slang/slang-emit-cpp.cpp @@ -969,6 +969,8 @@ void CPPSourceEmitter::_emitGetAtDefinition(const UnownedStringSlice& funcName, IRType* srcType = funcType->getParamType(0); + emitSpecializedOperationDefinitionPreamble(specOp); + IRType* retType = specOp->returnType; emitType(retType); m_writer->emit("& "); @@ -1046,6 +1048,8 @@ void CPPSourceEmitter::_emitConstructConvertDefinition(const UnownedStringSlice& IRType* srcType = funcType->getParamType(1); IRType* retType = specOp->returnType; + emitSpecializedOperationDefinitionPreamble(specOp); + emitType(retType); writer->emit(" "); writer->emit(funcName); @@ -1112,6 +1116,8 @@ void CPPSourceEmitter::_emitConstructFromScalarDefinition(const UnownedStringSli IRType* srcType = funcType->getParamType(1); IRType* retType = specOp->returnType; + emitSpecializedOperationDefinitionPreamble(specOp); + emitType(retType); writer->emit(" "); writer->emit(funcName); diff --git a/source/slang/slang-hlsl-intrinsic-set.h b/source/slang/slang-hlsl-intrinsic-set.h index d55c908ae..df1677b17 100644 --- a/source/slang/slang-hlsl-intrinsic-set.h +++ b/source/slang/slang-hlsl-intrinsic-set.h @@ -123,7 +123,9 @@ just constructXXXFromScalar. Would be good if there was a suitable name to encom x(ConstructFromScalar, "", 1) \ \ x(GetAt, "", 2) \ - x(SetAt, "", 3) + x(SetAt, "", 3) \ + \ + x(CountBits, "countbits", 1) struct HLSLIntrinsic { diff --git a/tests/compute/bit-cast-double.slang b/tests/hlsl-intrinsic/bit-cast-double.slang index 6b0925a58..b52c9de2b 100644 --- a/tests/compute/bit-cast-double.slang +++ b/tests/hlsl-intrinsic/bit-cast-double.slang @@ -1,7 +1,8 @@ //TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -cpu //TEST(compute):COMPARE_COMPUTE_EX:-slang -compute //TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 -//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute +//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-slang -vk -compute +//TEST(compute):COMPARE_COMPUTE_EX:-slang -cuda -compute //TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name=outputBuffer RWStructuredBuffer<int> outputBuffer; diff --git a/tests/compute/bit-cast-double.slang.expected.txt b/tests/hlsl-intrinsic/bit-cast-double.slang.expected.txt index bc856dafa..bc856dafa 100644 --- a/tests/compute/bit-cast-double.slang.expected.txt +++ b/tests/hlsl-intrinsic/bit-cast-double.slang.expected.txt diff --git a/tests/compute/bit-cast.slang b/tests/hlsl-intrinsic/bit-cast.slang index a8fec61e3..4ecd911d7 100644 --- a/tests/compute/bit-cast.slang +++ b/tests/hlsl-intrinsic/bit-cast.slang @@ -2,6 +2,7 @@ //TEST(compute):COMPARE_COMPUTE_EX:-slang -compute //TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 //TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute +//TEST(compute):COMPARE_COMPUTE_EX:-cuda -compute //TEST_INPUT:ubuffer(data=[0 0 0 0 0 0 0 0 0 0 0 0], stride=4):out,name outputBuffer diff --git a/tests/compute/bit-cast.slang.expected.txt b/tests/hlsl-intrinsic/bit-cast.slang.expected.txt index f5414793f..f5414793f 100644 --- a/tests/compute/bit-cast.slang.expected.txt +++ b/tests/hlsl-intrinsic/bit-cast.slang.expected.txt diff --git a/tests/hlsl-intrinsic/scalar-int.slang b/tests/hlsl-intrinsic/scalar-int.slang new file mode 100644 index 000000000..7fccc8645 --- /dev/null +++ b/tests/hlsl-intrinsic/scalar-int.slang @@ -0,0 +1,25 @@ +//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute +//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute +//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 +//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute +//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute + +//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer +RWStructuredBuffer<int> outputBuffer; + +[numthreads(4, 1, 1)] +void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) +{ + int idx = int(dispatchThreadID.x); + + int ti =0; + + ti += max(2, idx); + ti += min(idx, 1); + ti += abs(idx - 2); + ti += (idx * 3) % 5; + + ti += clamp(idx * 10, 11, 23); + + outputBuffer[idx] = ti; +}
\ No newline at end of file diff --git a/tests/hlsl-intrinsic/scalar-int.slang.expected.txt b/tests/hlsl-intrinsic/scalar-int.slang.expected.txt new file mode 100644 index 000000000..4efe33dc6 --- /dev/null +++ b/tests/hlsl-intrinsic/scalar-int.slang.expected.txt @@ -0,0 +1,4 @@ +F +12 +18 +20 diff --git a/tests/hlsl-intrinsic/scalar-uint.slang b/tests/hlsl-intrinsic/scalar-uint.slang new file mode 100644 index 000000000..eb16bb3b8 --- /dev/null +++ b/tests/hlsl-intrinsic/scalar-uint.slang @@ -0,0 +1,26 @@ +//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute +//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute +//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 +//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute +//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute + +//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer +RWStructuredBuffer<int> outputBuffer; + +[numthreads(4, 1, 1)] +void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) +{ + uint idx = dispatchThreadID.x; + + uint ti = 0; + + ti += max(2, idx); + ti += min(idx, 1); + ti += (idx * 3) %5; + + ti += clamp(idx * 10, 11, 23); + + ti += countbits(idx * 13); + + outputBuffer[idx] = ti; +}
\ No newline at end of file diff --git a/tests/hlsl-intrinsic/scalar-uint.slang.expected.txt b/tests/hlsl-intrinsic/scalar-uint.slang.expected.txt new file mode 100644 index 000000000..1a7befbe1 --- /dev/null +++ b/tests/hlsl-intrinsic/scalar-uint.slang.expected.txt @@ -0,0 +1,4 @@ +D +14 +1B +23 diff --git a/tests/hlsl-intrinsic/vector-int.slang b/tests/hlsl-intrinsic/vector-int.slang new file mode 100644 index 000000000..9a3f1faaa --- /dev/null +++ b/tests/hlsl-intrinsic/vector-int.slang @@ -0,0 +1,50 @@ +//TEST(compute):COMPARE_COMPUTE_EX:-cpu -compute +//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute +//TEST(compute):COMPARE_COMPUTE_EX:-slang -compute -dx12 +//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-vk -compute +//TEST(compute, vulkan):COMPARE_COMPUTE_EX:-cuda -compute + +//TEST_INPUT:ubuffer(data=[0 0 0 0], stride=4):out,name outputBuffer +RWStructuredBuffer<int> outputBuffer; + +[numthreads(4, 1, 1)] +void computeMain(uint3 dispatchThreadID : SV_DispatchThreadID) +{ + int idx = int(dispatchThreadID.x); + + int3 a = { idx + 10, idx - 9, idx + 3}; + int3 b = { idx * 2, idx * 3, idx * 10}; + + int3 t = { 0, 0, 0}; + + t += max(a, b); + t += min(a, b); + t += abs(a); + t += b % 5; + + t += clamp(a, int3(10), int3(23)); + + // Swizzle + t += a.zyx; + // Swizzle from scalar + t += idx.xxx; + +#if 0 + // TODO(JS): On C++ not all accesses are turned into getAt/setAt, so disable for now. + // Reads + { + t += int3(a[idx % 3], a[0], b[2]); + } + + // Writes + { + int3 v = int3(b[(idx + 1) % 3], b[(idx + 2) % 3], b[(idx + 3) % 3]); + v[1] += v[0]; + v[idx & 1] += v[2]; + + t += v; + } +#endif + + outputBuffer[idx] = t.x + t.y + t.z; +}
\ No newline at end of file diff --git a/tests/hlsl-intrinsic/vector-int.slang.expected.txt b/tests/hlsl-intrinsic/vector-int.slang.expected.txt new file mode 100644 index 000000000..e4c658883 --- /dev/null +++ b/tests/hlsl-intrinsic/vector-int.slang.expected.txt @@ -0,0 +1,4 @@ +3C +5B +75 +8F diff --git a/tools/render-test/bind-location.cpp b/tools/render-test/bind-location.cpp index 30b9de0f8..114ca7e8e 100644 --- a/tools/render-test/bind-location.cpp +++ b/tools/render-test/bind-location.cpp @@ -482,7 +482,7 @@ BindLocation BindSet::toIndex(const BindLocation& loc, Index index) const size_t baseOffset = loc.m_bindPointSet->m_points[category].m_offset; - if (category == SLANG_PARAMETER_CATEGORY_UNIFORM && uniformValue != loc.m_value) + if (category == slang::ParameterCategory::Uniform && uniformValue != loc.m_value) { baseOffset = 0; } |