simd_math_avx512f.h
No OneTemporary
Actions

Subscribers

None

File Metadata

	#ifndef SIMD_MATH
	#define SIMD_MATH_
	//
	#include <immintrin.h>
	//
	//
	//

	#ifdef __INTEL_COMPILER
	#warning "simd_math AVX512F detected"
	#define ADD(x,y) _mm512_add_pd(x, y)
	#define SUB(x,y) _mm512_sub_pd(x, y)
	#define MUL(x,y) _mm512_mul_pd(x, y)
	#define SET(x) _mm512_set1_pd(x)
	//
	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);}
	#endif

	#define __INV RCP_AVX512
	//#define __INV RCP_1NR_AVX512
	//#define __INV RCP_2NR_AVX512
	//#define __INV _mm512_rcp28_pd
	//#define __INV _mm512_rcp14_pd
	#define __SQRT _mm512_sqrt_pd
	//
	//
	inline __m512d RCP_AVX512(const __m512d d)
	{
	#warning "RCP_AVX512"
	__m512d x0;
	//x0 = _mm512_div_pd(_mm512_set1_pd(1.), d);
	//return x0;
	x0 = _mm512_rcp28_pd (d);
	//
	return x0;
	}

	//
	inline __m512d RCP_1NR_AVX512(const __m512d d)
	{
	#warning "RCP_1NR_AVX512"

	__m512d x0;
	//x0 = _mm512_div_pd(_mm512_set1_pd(1.), d);
	//return x0;
	x0 = _mm512_rcp28_pd (d);
	//
	x0 = SUB(ADD(x0, x0), MUL(d, MUL(x0, x0)));
	//
	return x0;
	}
	//
	//
	inline __m512d RCP_2NR_AVX512(const __m512d d)
	{
	#warning "RCP_2NR_AVX512"

	__m512d x0;
	//x0 = _mm512_div_pd(_mm512_set1_pd(1.), d);
	//return x0;
	x0 = _mm512_rcp28_pd (d);
	//
	x0 = SUB(ADD(x0, x0), MUL(d, MUL(x0, x0)));
	x0 = SUB(ADD(x0, x0), MUL(d, MUL(x0, x0)));
	//
	return x0;
	}
	//
	//
	//
	#endif