Page MenuHomec4science
Files F94101005

simd_math.h
No OneTemporary
Actions

File Metadata

Created
Tue, Dec 3, 22:11

simd_math.h
View Options

#ifndef SIMD_MATH
#define SIMD_MATH_
//
#include <immintrin.h>
//
//
//
#ifdef __INTEL_COMPILER
//
#ifdef __AVX512F__
//#warning "simd_math AVX512F detected"
inline __m512d operator + (__m512d a, __m512d b) {return _mm512_add_pd(a, b);}
inline __m512d operator - (__m512d a, __m512d b) {return _mm512_sub_pd(a, b);}
inline __m512d operator * (__m512d a, __m512d b) {return _mm512_mul_pd(a, b);}
inline __m512d operator / (__m512d a, __m512d b) {return _mm512_div_pd(a, b);}
#else
inline __m256d operator + (__m256d a, __m256d b) {return _mm256_add_pd(a, b);}
inline __m256d operator - (__m256d a, __m256d b) {return _mm256_sub_pd(a, b);}
inline __m256d operator * (__m256d a, __m256d b) {return _mm256_mul_pd(a, b);}
inline __m256d operator / (__m256d a, __m256d b) {return _mm256_div_pd(a, b);}
#endif
#endif
#ifdef __AVX512F__
#warning "AVX512F detected"
//#define __INV RCP_2NR_AVX512
#define __INV _mm512_rcp28_pd
//#define __INV _mm512_rcp14_pd
#define __SQRT _mm512_sqrt_pd
#else
//#define __INV RCP
#define __INV RCP_1NR
//#define __INV RCP_2NR
//#define __SQRT _mm256_sqrt_pd
//#define __SQRT SQRT
//#define __SQRT SQRT_1NR
#define __SQRT SQRT_2NR
#endif
inline __m256d RCP(const __m256d d)
{
const __m128 b = _mm256_cvtpd_ps(d);
const __m128 rcp = _mm_rcp_ps (b);
__m256d x0 = _mm256_cvtps_pd(rcp);
//
return x0;
}
//
//
//
inline __m256d RCP_1NR(const __m256d d)
{
const __m128 b = _mm256_cvtpd_ps(d);
const __m128 rcp = _mm_rcp_ps (b);
__m256d x0 = _mm256_cvtps_pd(rcp);
//
x0 = x0 + x0 - d*x0*x0;
// //
return x0;
}
//
//
//
inline __m256d RCP_2NR(const __m256d d)
{
const __m128 b = _mm256_cvtpd_ps(d);
const __m128 rcp = _mm_rcp_ps (b);
__m256d x0 = _mm256_cvtps_pd(rcp);
//
x0 = x0 + x0 - d*x0*x0;
x0 = x0 + x0 - d*x0*x0;
//
return x0;
}
inline __m512d RCP_2NR_AVX512(const __m512d d)
{
__m512d x0;
//x0 = _mm512_div_pd(_mm512_set1_pd(1.), d);
//return x0;
x0 = _mm512_rcp28_pd (d);
//
x0 = x0 + x0 - d*x0*x0;
x0 = x0 + x0 - d*x0*x0;
//
return x0;
}
inline __m256d SQRT(const __m256d d)
{
const __m128 b = _mm256_cvtpd_ps(d);
const __m128 rcp = _mm_sqrt_ps (b);
__m256d x0 = _mm256_cvtps_pd(rcp);
return x0;
}
//
//
//
inline __m256d SQRT_1NR(const __m256d d)
{
const __m128 b = _mm256_cvtpd_ps(d);
const __m128 rcp = _mm_sqrt_ps (b);
__m256d x0 = _mm256_cvtps_pd(rcp);
__m256d half = _mm256_set1_pd(0.5);
__m256d three = _mm256_set1_pd(3.);
__m256d a = RCP_1NR(d);
x0 = half*x0*(three - x0*x0*a);
return x0;
}
//
//
//
inline __m256d SQRT_2NR(const __m256d d)
{
const __m128 b = _mm256_cvtpd_ps(d);
const __m128 rcp = _mm_sqrt_ps (b);
__m256d x0 = _mm256_cvtps_pd(rcp);
__m256d half = _mm256_set1_pd(0.5);
__m256d three = _mm256_set1_pd(3.);
__m256d a = RCP_2NR(d);
x0 = half*x0*(three - x0*x0*a);
x0 = half*x0*(three - x0*x0*a);
return x0;
}
#endif

Event Timeline

c4science · Help