Aquila-Consortium/libsharp2
0
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

cleanup

Browse source
This commit is contained in:
Martin Reinecke 2018-12-15 13:02:56 +01:00
parent c750162e2b
commit 0b8222393f
12 changed files with 281 additions and 428 deletions
Download patch file Download diff file Expand all files Collapse all files

71
libsharp/sharp_complex_hacks.h
View file

@ -25,131 +25,126 @@
/* \file sharp_complex_hacks.h
* support for converting vector types and complex numbers
*
* Copyright (C) 2012-2016 Max-Planck-Society
* Copyright (C) 2012-2018 Max-Planck-Society
* Author: Martin Reinecke
*/
#ifndef SHARP_COMPLEX_HACKS_H
#define SHARP_COMPLEX_HACKS_H
#ifdef __cplusplus
#error This header file cannot be included from C++, only from C
#endif
#include <math.h>
#include <complex.h>
#include "sharp_vecsupport.h"
#define UNSAFE_CODE
#if (VLEN==1)
static inline complex double vhsum_cmplx(Tv a, Tv b)
static inline _Complex double vhsum_cmplx(Tv a, Tv b)
{ return a+_Complex_I*b; }
static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c, Tv d,
complex double * restrict c1, complex double * restrict c2)
_Complex double * restrict c1, _Complex double * restrict c2)
{ *c1 += a+_Complex_I*b; *c2 += c+_Complex_I*d; }
static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d,
complex double * restrict cc)
_Complex double * restrict cc)
{ cc[0] += a+_Complex_I*b; cc[1] += c+_Complex_I*d; }
#endif
#if (VLEN==2)
static inline complex double vhsum_cmplx (Tv a, Tv b)
static inline _Complex double vhsum_cmplx (Tv a, Tv b)
{
#if defined(__SSE3__)
Tv tmp = _mm_hadd_pd(a,b);
#else
Tv tmp = vadd(_mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)),
_mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0)));
Tv tmp = _mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)) +
_mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0));
#endif
union {Tv v; complex double c; } u;
union {Tv v; _Complex double c; } u;
u.v=tmp; return u.c;
}
static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c,
Tv d, complex double * restrict c1, complex double * restrict c2)
Tv d, _Complex double * restrict c1, _Complex double * restrict c2)
{
#ifdef UNSAFE_CODE
#if defined(__SSE3__)
vaddeq(*((__m128d *)c1),_mm_hadd_pd(a,b));
vaddeq(*((__m128d *)c2),_mm_hadd_pd(c,d));
*((__m128d *)c1) += _mm_hadd_pd(a,b);
*((__m128d *)c2) += _mm_hadd_pd(c,d);
#else
vaddeq(*((__m128d *)c1),vadd(_mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)),
_mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0))));
vaddeq(*((__m128d *)c2),vadd(_mm_shuffle_pd(c,d,_MM_SHUFFLE2(0,1)),
_mm_shuffle_pd(c,d,_MM_SHUFFLE2(1,0))));
*((__m128d *)c1) += _mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)) +
_mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0));
*((__m128d *)c2) += _mm_shuffle_pd(c,d,_MM_SHUFFLE2(0,1)) +
_mm_shuffle_pd(c,d,_MM_SHUFFLE2(1,0));
#endif
#else
union {Tv v; complex double c; } u1, u2;
union {Tv v; _Complex double c; } u1, u2;
#if defined(__SSE3__)
u1.v = _mm_hadd_pd(a,b); u2.v=_mm_hadd_pd(c,d);
#else
u1.v = vadd(_mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)),
_mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0)));
u2.v = vadd(_mm_shuffle_pd(c,d,_MM_SHUFFLE2(0,1)),
_mm_shuffle_pd(c,d,_MM_SHUFFLE2(1,0)));
u1.v = _mm_shuffle_pd(a,b,_MM_SHUFFLE2(0,1)) +
_mm_shuffle_pd(a,b,_MM_SHUFFLE2(1,0));
u2.v = _mm_shuffle_pd(c,d,_MM_SHUFFLE2(0,1)) +
_mm_shuffle_pd(c,d,_MM_SHUFFLE2(1,0));
#endif
*c1+=u1.c; *c2+=u2.c;
#endif
}
static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d,
complex double * restrict cc)
_Complex double * restrict cc)
{ vhsum_cmplx2(a,b,c,d,cc,cc+1); }
#endif
#if (VLEN==4)
static inline complex double vhsum_cmplx (Tv a, Tv b)
static inline _Complex double vhsum_cmplx (Tv a, Tv b)
{
Tv tmp=_mm256_hadd_pd(a,b);
Tv tmp2=_mm256_permute2f128_pd(tmp,tmp,1);
tmp=_mm256_add_pd(tmp,tmp2);
#ifdef UNSAFE_CODE
complex double ret;
_Complex double ret;
*((__m128d *)&ret)=_mm256_extractf128_pd(tmp, 0);
return ret;
#else
union {Tv v; complex double c[2]; } u;
union {Tv v; _Complex double c[2]; } u;
u.v=tmp; return u.c[0];
#endif
}
static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c, Tv d,
complex double * restrict c1, complex double * restrict c2)
_Complex double * restrict c1, _Complex double * restrict c2)
{
Tv tmp1=_mm256_hadd_pd(a,b), tmp2=_mm256_hadd_pd(c,d);
Tv tmp3=_mm256_permute2f128_pd(tmp1,tmp2,49),
tmp4=_mm256_permute2f128_pd(tmp1,tmp2,32);
tmp1=vadd(tmp3,tmp4);
tmp1=tmp3+tmp4;
#ifdef UNSAFE_CODE
*((__m128d *)c1)=_mm_add_pd(*((__m128d *)c1),_mm256_extractf128_pd(tmp1, 0));
*((__m128d *)c2)=_mm_add_pd(*((__m128d *)c2),_mm256_extractf128_pd(tmp1, 1));
#else
union {Tv v; complex double c[2]; } u;
union {Tv v; _Complex double c[2]; } u;
u.v=tmp1;
*c1+=u.c[0]; *c2+=u.c[1];
#endif
}
static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d,
complex double * restrict cc)
_Complex double * restrict cc)
{
Tv tmp1=_mm256_hadd_pd(a,b), tmp2=_mm256_hadd_pd(c,d);
Tv tmp3=_mm256_permute2f128_pd(tmp1,tmp2,49),
tmp4=_mm256_permute2f128_pd(tmp1,tmp2,32);
tmp1=vadd(tmp3,tmp4);
tmp1=tmp3+tmp4;
#ifdef UNSAFE_CODE
_mm256_storeu_pd((double *)cc,
_mm256_add_pd(_mm256_loadu_pd((double *)cc),tmp1));
#else
union {Tv v; complex double c[2]; } u;
union {Tv v; _Complex double c[2]; } u;
u.v=tmp1;
cc[0]+=u.c[0]; cc[1]+=u.c[1];
#endif
@ -159,18 +154,18 @@ static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d,
#if (VLEN==8)
static inline complex double vhsum_cmplx(Tv a, Tv b)
static inline _Complex double vhsum_cmplx(Tv a, Tv b)
{ return _mm512_reduce_add_pd(a)+_Complex_I*_mm512_reduce_add_pd(b); }
static inline void vhsum_cmplx2 (Tv a, Tv b, Tv c, Tv d,
complex double * restrict c1, complex double * restrict c2)
_Complex double * restrict c1, _Complex double * restrict c2)
{
*c1 += _mm512_reduce_add_pd(a)+_Complex_I*_mm512_reduce_add_pd(b);
*c2 += _mm512_reduce_add_pd(c)+_Complex_I*_mm512_reduce_add_pd(d);
}
static inline void vhsum_cmplx_special (Tv a, Tv b, Tv c, Tv d,
complex double * restrict cc)
_Complex double * restrict cc)
{ vhsum_cmplx2(a,b,c,d,cc,cc+1); }
#endif