libAthena
/
athena
mirror of https://github.com/libAthena/athena.git
2
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Constexpr constructors for simd types

This commit is contained in:
Jack Andersen 2019-02-23 12:03:57 -10:00
parent c1a1c6e754
commit ef7a9633c6
3 changed files with 9 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
include/athena/simd/parallelism_v2_simd.hpp
View File

@ -1274,9 +1274,10 @@ public:
__s_.__broadcast(v); __s_.__broadcast(v);
} }
#endif #endif
simd(_Tp __rv) { __s_.__broadcast(__rv); }
simd(_Tp a, _Tp b, _Tp c = {}, _Tp d = {}) { __s_.__set4(a, b, c, d); } constexpr simd(_Tp __rv) : __s_(__rv) {}
constexpr simd(_Tp a, _Tp b, _Tp c = {}, _Tp d = {}) : __s_(a, b, c, d) {}
// generator constructor // generator constructor
template <class _Generator, template <class _Generator,

2
include/athena/simd/simd_avx.hpp
View File

@ -22,7 +22,9 @@ public:
sse_data[__index] = __val; sse_data[__index] = __val;
__storage_ = _mm256_load_pd(sse_data.data()); __storage_ = _mm256_load_pd(sse_data.data());
} }
constexpr __simd_storage(double a, double b, double c, double d) : __storage_{a, b, c, d} {}
void __set4(double a, double b, double c, double d) noexcept { __storage_ = _mm256_set_pd(d, c, b, a); } void __set4(double a, double b, double c, double d) noexcept { __storage_ = _mm256_set_pd(d, c, b, a); }
constexpr __simd_storage(double rv) : __storage_{rv, rv, rv, rv} {}
void __broadcast(double __val) noexcept { __storage_ = _mm256_set1_pd(__val); } void __broadcast(double __val) noexcept { __storage_ = _mm256_set1_pd(__val); }
double __dot2(const __simd_storage<double, m256d_abi>& other) const noexcept { double __dot2(const __simd_storage<double, m256d_abi>& other) const noexcept {
alignas(32) std::array<double, 4> sse_data; alignas(32) std::array<double, 4> sse_data;

4
include/athena/simd/simd_sse.hpp
View File

@ -41,7 +41,9 @@ public:
sse_data[__index] = __val; sse_data[__index] = __val;
__storage_ = _mm_load_ps(sse_data.data()); __storage_ = _mm_load_ps(sse_data.data());
} }
constexpr __simd_storage(float a, float b, float c, float d) : __storage_{a, b, c, d} {}
void __set4(float a, float b, float c, float d) noexcept { __storage_ = _mm_set_ps(d, c, b, a); } void __set4(float a, float b, float c, float d) noexcept { __storage_ = _mm_set_ps(d, c, b, a); }
constexpr __simd_storage(float rv) : __storage_{rv, rv, rv, rv} {}
void __broadcast(float __val) noexcept { __storage_ = _mm_set1_ps(__val); } void __broadcast(float __val) noexcept { __storage_ = _mm_set1_ps(__val); }
float __dot2(const __simd_storage<float, m128_abi>& other) const noexcept { float __dot2(const __simd_storage<float, m128_abi>& other) const noexcept {
#if __SSE4_1__ #if __SSE4_1__
@ -219,10 +221,12 @@ public:
sse_data[__index % 2] = __val; sse_data[__index % 2] = __val;
__storage_[__index / 2] = _mm_load_pd(sse_data.data()); __storage_[__index / 2] = _mm_load_pd(sse_data.data());
} }
constexpr __simd_storage(double a, double b, double c, double d) : __storage_{__m128d{a, b}, __m128d{c, d}} {}
void __set4(double a, double b, double c, double d) noexcept { void __set4(double a, double b, double c, double d) noexcept {
__storage_[0] = _mm_set_pd(b, a); __storage_[0] = _mm_set_pd(b, a);
__storage_[1] = _mm_set_pd(d, c); __storage_[1] = _mm_set_pd(d, c);
} }
constexpr __simd_storage(double rv) : __storage_{__m128d{rv, rv}, __m128d{rv, rv}} {}
void __broadcast(double __val) noexcept { void __broadcast(double __val) noexcept {
for (int i = 0; i < 2; ++i) for (int i = 0; i < 2; ++i)
__storage_[i] = _mm_set1_pd(__val); __storage_[i] = _mm_set1_pd(__val);