diff options
| author | untodesu <kirill@untode.su> | 2025-03-15 16:22:09 +0500 |
|---|---|---|
| committer | untodesu <kirill@untode.su> | 2025-03-15 16:22:09 +0500 |
| commit | 3bf42c6ff3805a0d42bbc661794a95ff31bedc26 (patch) | |
| tree | 05049955847504808d6bed2bb7b155f8b03807bb /deps/include/glm/detail/func_integer_simd.inl | |
| parent | 02294547dcde0d4ad76e229106702261e9f10a51 (diff) | |
| download | voxelius-3bf42c6ff3805a0d42bbc661794a95ff31bedc26.tar.bz2 voxelius-3bf42c6ff3805a0d42bbc661794a95ff31bedc26.zip | |
Add whatever I was working on for the last month
Diffstat (limited to 'deps/include/glm/detail/func_integer_simd.inl')
| -rw-r--r-- | deps/include/glm/detail/func_integer_simd.inl | 65 |
1 files changed, 65 insertions, 0 deletions
diff --git a/deps/include/glm/detail/func_integer_simd.inl b/deps/include/glm/detail/func_integer_simd.inl new file mode 100644 index 0000000..5600c84 --- /dev/null +++ b/deps/include/glm/detail/func_integer_simd.inl @@ -0,0 +1,65 @@ +#include "../simd/integer.h" + +#if GLM_ARCH & GLM_ARCH_SSE2_BIT + +namespace glm{ +namespace detail +{ + template<qualifier Q> + struct compute_bitfieldReverseStep<4, uint, Q, true, true> + { + GLM_FUNC_QUALIFIER static vec<4, uint, Q> call(vec<4, uint, Q> const& v, uint Mask, uint Shift) + { + __m128i const set0 = v.data; + + __m128i const set1 = _mm_set1_epi32(static_cast<int>(Mask)); + __m128i const and1 = _mm_and_si128(set0, set1); + __m128i const sft1 = _mm_slli_epi32(and1, static_cast<int>(Shift)); + + __m128i const set2 = _mm_andnot_si128(set0, _mm_set1_epi32(-1)); + __m128i const and2 = _mm_and_si128(set0, set2); + __m128i const sft2 = _mm_srai_epi32(and2, static_cast<int>(Shift)); + + __m128i const or0 = _mm_or_si128(sft1, sft2); + + return or0; + } + }; + + template<qualifier Q> + struct compute_bitfieldBitCountStep<4, uint, Q, true, true> + { + GLM_FUNC_QUALIFIER static vec<4, uint, Q> call(vec<4, uint, Q> const& v, uint Mask, uint Shift) + { + __m128i const set0 = v.data; + + __m128i const set1 = _mm_set1_epi32(static_cast<int>(Mask)); + __m128i const and0 = _mm_and_si128(set0, set1); + __m128i const sft0 = _mm_slli_epi32(set0, static_cast<int>(Shift)); + __m128i const and1 = _mm_and_si128(sft0, set1); + __m128i const add0 = _mm_add_epi32(and0, and1); + + return add0; + } + }; +}//namespace detail + +# if GLM_ARCH & GLM_ARCH_AVX_BIT + template<> + GLM_FUNC_QUALIFIER int bitCount(uint x) + { + return _mm_popcnt_u32(x); + } + +# if(GLM_MODEL == GLM_MODEL_64) + template<> + GLM_FUNC_QUALIFIER int bitCount(detail::uint64 x) + { + return static_cast<int>(_mm_popcnt_u64(x)); + } +# endif//GLM_MODEL +# endif//GLM_ARCH + +}//namespace glm + +#endif//GLM_ARCH & GLM_ARCH_SSE2_BIT |
