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

simplify Ylm data structures

Browse source
This commit is contained in:
Martin Reinecke 2019-01-11 09:27:04 +01:00
parent b0b0875def
commit 253b253467
3 changed files with 52 additions and 55 deletions
Download patch file Download diff file Expand all files Collapse all files

80
libsharp/sharp_core.c
View file

@ -190,8 +190,8 @@ NOINLINE static void iter_to_ieee(const sharp_Ylmgen_C * restrict gen,
{
if (l+4>gen->lmax) {*l_=gen->lmax+1;return;}
below_limit=1;
Tv a1=vload(gen->ab[il ].f[0]), b1=vload(gen->ab[il ].f[1]);
Tv a2=vload(gen->ab[il+1].f[0]), b2=vload(gen->ab[il+1].f[1]);
Tv a1=vload(gen->coef[il ][0]), b1=vload(gen->coef[il ][1]);
Tv a2=vload(gen->coef[il+1][0]), b2=vload(gen->coef[il+1][1]);
for (int i=0; i<nv2; ++i)
{
d->lam1[i] = (a1*d->csq[i] + b1)*d->lam2[i] + d->lam1[i];
@ -205,7 +205,7 @@ NOINLINE static void iter_to_ieee(const sharp_Ylmgen_C * restrict gen,
}
NOINLINE static void alm2map_kernel(s0data_v * restrict d,
const sharp_ylmgen_dbl2 * restrict ab, const dcmplx * restrict alm,
const sharp_ylmgen_dbl2 * restrict coef, const dcmplx * restrict alm,
int l, int il, int lmax, int nv2)
{
if (nv2==nv0)
@ -216,8 +216,8 @@ NOINLINE static void alm2map_kernel(s0data_v * restrict d,
Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1]));
Tv ar3=vload(creal(alm[l+2])), ai3=vload(cimag(alm[l+2]));
Tv ar4=vload(creal(alm[l+3])), ai4=vload(cimag(alm[l+3]));
Tv a1=vload(ab[il ].f[0]), b1=vload(ab[il ].f[1]);
Tv a2=vload(ab[il+1].f[0]), b2=vload(ab[il+1].f[1]);
Tv a1=vload(coef[il ][0]), b1=vload(coef[il ][1]);
Tv a2=vload(coef[il+1][0]), b2=vload(coef[il+1][1]);
for (int i=0; i<nv0; ++i)
{
d->p1r[i] += d->lam2[i]*ar1;
@ -241,8 +241,8 @@ NOINLINE static void alm2map_kernel(s0data_v * restrict d,
Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1]));
Tv ar3=vload(creal(alm[l+2])), ai3=vload(cimag(alm[l+2]));
Tv ar4=vload(creal(alm[l+3])), ai4=vload(cimag(alm[l+3]));
Tv a1=vload(ab[il ].f[0]), b1=vload(ab[il ].f[1]);
Tv a2=vload(ab[il+1].f[0]), b2=vload(ab[il+1].f[1]);
Tv a1=vload(coef[il ][0]), b1=vload(coef[il ][1]);
Tv a2=vload(coef[il+1][0]), b2=vload(coef[il+1][1]);
for (int i=0; i<nv2; ++i)
{
d->p1r[i] += d->lam2[i]*ar1;
@ -262,7 +262,7 @@ NOINLINE static void alm2map_kernel(s0data_v * restrict d,
{
Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ]));
Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1]));
Tv a=vload(ab[il].f[0]), b=vload(ab[il].f[1]);
Tv a=vload(coef[il][0]), b=vload(coef[il][1]);
for (int i=0; i<nv2; ++i)
{
d->p1r[i] += d->lam2[i]*ar1;
@ -286,7 +286,7 @@ NOINLINE static void calc_alm2map (sharp_job * restrict job,
if (l>lmax) return;
job->opcnt += (lmax+1-l) * 6*nth;
const sharp_ylmgen_dbl2 * restrict ab = gen->ab;
const sharp_ylmgen_dbl2 * restrict coef = gen->coef;
const dcmplx * restrict alm=job->almtmp;
int full_ieee=1;
for (int i=0; i<nv2; ++i)
@ -299,7 +299,7 @@ NOINLINE static void calc_alm2map (sharp_job * restrict job,
{
Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ]));
Tv ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1]));
Tv a=vload(ab[il].f[0]), b=vload(ab[il].f[1]);
Tv a=vload(coef[il][0]), b=vload(coef[il][1]);
full_ieee=1;
for (int i=0; i<nv2; ++i)
{
@ -323,17 +323,17 @@ NOINLINE static void calc_alm2map (sharp_job * restrict job,
d->lam1[i] *= d->corfac[i];
d->lam2[i] *= d->corfac[i];
}
alm2map_kernel(d, ab, alm, l, il, lmax, nv2);
alm2map_kernel(d, coef, alm, l, il, lmax, nv2);
}
NOINLINE static void map2alm_kernel(s0data_v * restrict d,
const sharp_ylmgen_dbl2 * restrict ab, dcmplx * restrict alm, int l,
const sharp_ylmgen_dbl2 * restrict coef, dcmplx * restrict alm, int l,
int il, int lmax, int nv2)
{
for (; l<=lmax-2; il+=2, l+=4)
{
Tv a1=vload(ab[il ].f[0]), b1=vload(ab[il ].f[1]);
Tv a2=vload(ab[il+1].f[0]), b2=vload(ab[il+1].f[1]);
Tv a1=vload(coef[il ][0]), b1=vload(coef[il ][1]);
Tv a2=vload(coef[il+1][0]), b2=vload(coef[il+1][1]);
Tv atmp1[4] = {vzero, vzero, vzero, vzero};
Tv atmp2[4] = {vzero, vzero, vzero, vzero};
for (int i=0; i<nv2; ++i)
@ -354,7 +354,7 @@ NOINLINE static void map2alm_kernel(s0data_v * restrict d,
}
for (; l<=lmax; ++il, l+=2)
{
Tv a=vload(ab[il].f[0]), b=vload(ab[il].f[1]);
Tv a=vload(coef[il][0]), b=vload(coef[il][1]);
Tv atmp[4] = {vzero, vzero, vzero, vzero};
for (int i=0; i<nv2; ++i)
{
@ -380,7 +380,7 @@ NOINLINE static void calc_map2alm (sharp_job * restrict job,
if (l>lmax) return;
job->opcnt += (lmax+1-l) * 6*nth;
const sharp_ylmgen_dbl2 * restrict ab = gen->ab;
const sharp_ylmgen_dbl2 * restrict coef = gen->coef;
dcmplx * restrict alm=job->almtmp;
int full_ieee=1;
for (int i=0; i<nv2; ++i)
@ -391,7 +391,7 @@ NOINLINE static void calc_map2alm (sharp_job * restrict job,
while((!full_ieee) && (l<=lmax))
{
Tv a=vload(ab[il].f[0]), b=vload(ab[il].f[1]);
Tv a=vload(coef[il][0]), b=vload(coef[il][1]);
Tv atmp[4] = {vzero, vzero, vzero, vzero};
full_ieee=1;
for (int i=0; i<nv2; ++i)
@ -417,13 +417,13 @@ NOINLINE static void calc_map2alm (sharp_job * restrict job,
d->lam1[i] *= d->corfac[i];
d->lam2[i] *= d->corfac[i];
}
map2alm_kernel(d, ab, alm, l, il, lmax, nv2);
map2alm_kernel(d, coef, alm, l, il, lmax, nv2);
}
NOINLINE static void iter_to_ieee_spin (const sharp_Ylmgen_C * restrict gen,
sxdata_v * restrict d, int * restrict l_, int nv2)
{
const sharp_ylmgen_dbl2 * restrict fx = gen->fx;
const sharp_ylmgen_dbl2 * restrict fx = gen->coef;
Tv prefac=vload(gen->prefac[gen->m]),
prescale=vload(gen->fscale[gen->m]);
Tv limscale=vload(sharp_limscale);
@ -474,8 +474,8 @@ NOINLINE static void iter_to_ieee_spin (const sharp_Ylmgen_C * restrict gen,
{
if (l+2>gen->lmax) {*l_=gen->lmax+1;return;}
below_limit=1;
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
for (int i=0; i<nv2; ++i)
{
d->l1p[i] = (d->cth[i]*fx10 - fx11)*d->l2p[i] - d->l1p[i];
@ -500,8 +500,8 @@ NOINLINE static void alm2map_spin_kernel(sxdata_v * restrict d,
int lsave = l;
while (l<=lmax)
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv agr1=vload(creal(alm[2*l ])), agi1=vload(cimag(alm[2*l ])),
acr1=vload(creal(alm[2*l+1])), aci1=vload(cimag(alm[2*l+1]));
Tv agr2=vload(creal(alm[2*l+2])), agi2=vload(cimag(alm[2*l+2])),
@ -525,8 +525,8 @@ NOINLINE static void alm2map_spin_kernel(sxdata_v * restrict d,
l=lsave;
while (l<=lmax)
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv agr1=vload(creal(alm[2*l ])), agi1=vload(cimag(alm[2*l ])),
acr1=vload(creal(alm[2*l+1])), aci1=vload(cimag(alm[2*l+1]));
Tv agr2=vload(creal(alm[2*l+2])), agi2=vload(cimag(alm[2*l+2])),
@ -559,7 +559,7 @@ NOINLINE static void calc_alm2map_spin (sharp_job * restrict job,
if (l>lmax) return;
job->opcnt += (lmax+1-l) * 23*nth;
const sharp_ylmgen_dbl2 * restrict fx = gen->fx;
const sharp_ylmgen_dbl2 * restrict fx = gen->coef;
const dcmplx * restrict alm=job->almtmp;
int full_ieee=1;
for (int i=0; i<nv2; ++i)
@ -572,8 +572,8 @@ NOINLINE static void calc_alm2map_spin (sharp_job * restrict job,
while((!full_ieee) && (l<=lmax))
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv agr1=vload(creal(alm[2*l ])), agi1=vload(cimag(alm[2*l ])),
acr1=vload(creal(alm[2*l+1])), aci1=vload(cimag(alm[2*l+1]));
Tv agr2=vload(creal(alm[2*l+2])), agi2=vload(cimag(alm[2*l+2])),
@ -636,8 +636,8 @@ NOINLINE static void map2alm_spin_kernel(sxdata_v * restrict d,
int lsave=l;
while (l<=lmax)
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv agr1=vzero, agi1=vzero, acr1=vzero, aci1=vzero;
Tv agr2=vzero, agi2=vzero, acr2=vzero, aci2=vzero;
for (int i=0; i<nv2; ++i)
@ -660,8 +660,8 @@ NOINLINE static void map2alm_spin_kernel(sxdata_v * restrict d,
l=lsave;
while (l<=lmax)
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv agr1=vzero, agi1=vzero, acr1=vzero, aci1=vzero;
Tv agr2=vzero, agi2=vzero, acr2=vzero, aci2=vzero;
for (int i=0; i<nv2; ++i)
@ -693,7 +693,7 @@ NOINLINE static void calc_map2alm_spin (sharp_job * restrict job,
if (l>lmax) return;
job->opcnt += (lmax+1-l) * 23*nth;
const sharp_ylmgen_dbl2 * restrict fx = gen->fx;
const sharp_ylmgen_dbl2 * restrict fx = gen->coef;
dcmplx * restrict alm=job->almtmp;
int full_ieee=1;
for (int i=0; i<nv2; ++i)
@ -714,8 +714,8 @@ NOINLINE static void calc_map2alm_spin (sharp_job * restrict job,
while((!full_ieee) && (l<=lmax))
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv agr1=vzero, agi1=vzero, acr1=vzero, aci1=vzero;
Tv agr2=vzero, agi2=vzero, acr2=vzero, aci2=vzero;
full_ieee=1;
@ -766,8 +766,8 @@ NOINLINE static void alm2map_deriv1_kernel(sxdata_v * restrict d,
{
while (l<=lmax)
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])),
ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1]));
for (int i=0; i<nv2; ++i)
@ -803,7 +803,7 @@ NOINLINE static void calc_alm2map_deriv1(sharp_job * restrict job,
if (l>lmax) return;
job->opcnt += (lmax+1-l) * 17*nth;
const sharp_ylmgen_dbl2 * restrict fx = gen->fx;
const sharp_ylmgen_dbl2 * restrict fx = gen->coef;
const dcmplx * restrict alm=job->almtmp;
int full_ieee=1;
for (int i=0; i<nv2; ++i)
@ -816,8 +816,8 @@ NOINLINE static void calc_alm2map_deriv1(sharp_job * restrict job,
while((!full_ieee) && (l<=lmax))
{
Tv fx10=vload(fx[l+1].f[0]),fx11=vload(fx[l+1].f[1]);
Tv fx20=vload(fx[l+2].f[0]),fx21=vload(fx[l+2].f[1]);
Tv fx10=vload(fx[l+1][0]),fx11=vload(fx[l+1][1]);
Tv fx20=vload(fx[l+2][0]),fx21=vload(fx[l+2][1]);
Tv ar1=vload(creal(alm[l ])), ai1=vload(cimag(alm[l ])),
ar2=vload(creal(alm[l+1])), ai2=vload(cimag(alm[l+1]));
full_ieee=1;