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

potential workaround for problems on Windows and with some Intel compilers

Browse source
This commit is contained in:
Martin Reinecke 2021-10-07 09:18:40 +02:00
parent 6374a3a1ff
commit 0954bdb351
1 changed files with 26 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

26
libsharp2/sharp_core_inc.c
View file

@ -94,6 +94,7 @@ typedef union
sxdata_s s; sxdata_s s;
} sxdata_u; } sxdata_u;
#if 0
static inline void Tvnormalize (Tv * restrict val, Tv * restrict scale, static inline void Tvnormalize (Tv * restrict val, Tv * restrict scale,
double maxval) double maxval)
{ {
@ -114,6 +115,31 @@ static inline void Tvnormalize (Tv * restrict val, Tv * restrict scale,
mask = vand_mask(vlt(vabs(*val),vfmin),vne(*val,vzero)); mask = vand_mask(vlt(vabs(*val),vfmin),vne(*val,vzero));
} }
} }
#else
static inline void Tvnormalize (Tv * restrict val_, Tv * restrict scale_,
double maxval)
{
Tv val=*val_, scale=*scale_;
const Tv vfmin=vload(sharp_fsmall*maxval), vfmax=vload(maxval);
const Tv vfsmall=vload(sharp_fsmall), vfbig=vload(sharp_fbig);
Tm mask = vgt(vabs(val),vfmax);
while (vanyTrue(mask))
{
vmuleq_mask(mask,val,vfsmall);
vaddeq_mask(mask,scale,vone);
mask = vgt(vabs(val),vfmax);
}
mask = vand_mask(vlt(vabs(val),vfmin),vne(val,vzero));
while (vanyTrue(mask))
{
vmuleq_mask(mask,val,vfbig);
vsubeq_mask(mask,scale,vone);
mask = vand_mask(vlt(vabs(val),vfmin),vne(val,vzero));
}
*val_ = val;
*scale_ = scale;
}
#endif
static void mypow(Tv val, int npow, const double * restrict powlimit, static void mypow(Tv val, int npow, const double * restrict powlimit,
Tv * restrict resd, Tv * restrict ress) Tv * restrict resd, Tv * restrict ress)