Skip to content

Commit

Permalink
some bugs fixed, included in Makefile
Browse files Browse the repository at this point in the history
  • Loading branch information
@urbach
urbach committed Sep 6, 2012
1 parent 0b61ce2 commit 141d577
Show file tree
Hide file tree
Showing 3 changed files with 42 additions and 58 deletions.
3 changes: 2 additions & 1 deletion linalg/Makefile.in
View file
Original file line number Diff line number Diff line change
Expand Up @@ -45,7 +45,8 @@ liblinalg_TARGETS = assign_add_mul_r_add_mul \
scalar_prod_r_bi assign_add_mul_r_bi assign_mul_add_r_bi \
convert_eo_to_lexic assign_mul_add_mul_r mul_add_mul_r \
assign_mul_add_mul_add_mul_r mattimesvec \
scalar_prod_su3spinor
scalar_prod_su3spinor \
assign_mul_add_r_and_square

liblinalg_STARGETS = diff assign_add_mul_r assign_mul_add_r square_norm

Expand Down
96 changes: 39 additions & 57 deletions linalg/assign_mul_add_r_and_square.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -37,63 +37,39 @@ double assign_mul_add_r_and_square(spinor * const R, const double c, spinor * co
#ifdef MPI
double ALIGN mres;
#endif
vector4double r0, r1, r2, r3, r4, r5;

#ifdef OMP
#pragma omp parallel private(res, r0, r1, r2, r3, r4, r5) reduction(+: res)
#pragma omp parallel reduction(+: res)
{
#endif
vector4double x0, x1, x2, x3, x4, x5, y0, y1, y2, y3, y4, y5;
vector4double z0, z1, z2, z3, z4, z5, k;
vector4double r0, r1, r2, r3, r4, r5;
double *s, *r;
double ALIGN _c = c;
#ifndef OMP
__prefetch_by_load(S);
__prefetch_by_load(R);
#endif

k = vec_splats(_c);
__alignx(32, s);
__alignx(32, r);
__alignx(32, S);
__alignx(32, R);
r0 = vec_splats(0.);
r1 = vec_splats(0.);
r2 = vec_splats(0.);
r3 = vec_splats(0.);
r4 = vec_splats(0.);
r5 = vec_splats(0.);

s=(double*)((spinor *) S);
r=(double*)((spinor *) R);
x0 = vec_ld(0, r);
x1 = vec_ld(0, r+4);
x2 = vec_ld(0, r+8);
x3 = vec_ld(0, r+12);
x4 = vec_ld(0, r+16);
x5 = vec_ld(0, r+20);
y0 = vec_ld(0, s);
y1 = vec_ld(0, s+4);
y2 = vec_ld(0, s+8);
y3 = vec_ld(0, s+12);
y4 = vec_ld(0, s+16);
y5 = vec_ld(0, s+20);
z0 = vec_madd(k, x0, y0);
z1 = vec_madd(k, x1, y1);
z2 = vec_madd(k, x2, y2);
z3 = vec_madd(k, x3, y3);
z4 = vec_madd(k, x4, y4);
z5 = vec_madd(k, x5, y5);
vec_st(z0, 0, r);
vec_st(z1, 0, r+4);
vec_st(z2, 0, r+8);
vec_st(z3, 0, r+12);
vec_st(z4, 0, r+16);
vec_st(z5, 0, r+20);
r0 = vec_mul(z0, z0);
r1 = vec_mul(z1, z1);
r2 = vec_mul(z2, z2);
r3 = vec_mul(z3, z3);
r4 = vec_mul(z4, z4);
r5 = vec_mul(z5, z5);

#ifdef OMP
#pragma omp for
#endif
#pragma unroll(4)
for(int i = 1; i < N; i++) {
//#pragma unroll(4)
for(int i = 0; i < N; i++) {
s=(double*)((spinor *) S + i);
r=(double*)((spinor *) R + i);
__prefetch_by_load(S + i + 1);
Expand Down Expand Up @@ -128,7 +104,6 @@ double assign_mul_add_r_and_square(spinor * const R, const double c, spinor * co
r3 = vec_madd(z3, z3, r3);
r4 = vec_madd(z4, z4, r4);
r5 = vec_madd(z5, z5, r5);

}
x0 = vec_add(r0, r1);
x1 = vec_add(r2, r3);
Expand All @@ -140,7 +115,13 @@ double assign_mul_add_r_and_square(spinor * const R, const double c, spinor * co
#ifdef OMP
} /* OpenMP closing brace */
#endif
return;
# ifdef MPI
if(parallel) {
MPI_Allreduce(&res, &mres, 1, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
return(mres);
}
#endif
return(res);
}

#else
Expand All @@ -150,51 +131,52 @@ double assign_mul_add_r_and_square(spinor * const R, const double c, spinor * co

double assign_mul_add_r_and_square(spinor * const R, const double c, spinor * const S,
const int N, const int parallel) {
double res;
#ifdef MPI
double mres;
#endif

#ifdef OMP
#pragma omp parallel
#pragma omp parallel reduction(+ : res)
{
#endif
spinor *r,*s;
double res = 0.;
#ifdef MPI
double mres;
#endif

res = 0.;
/* Change due to even-odd preconditioning : VOLUME to VOLUME/2 */
#ifdef OMP
#pragma omp for private(res) reduction(+ : res)
#pragma omp for
#endif
for (int ix = 0; ix < N; ++ix) {
r = R + ix;
s = S + ix;

r->s0.c0 = c * r->s0.c0 + s->s0.c0;
sum += creal(r->s0.c0)*creal(r->s0.c0) + cimag(r->s0.c0)*cimag(r->s0.c0);
res += creal(r->s0.c0)*creal(r->s0.c0) + cimag(r->s0.c0)*cimag(r->s0.c0);
r->s0.c1 = c * r->s0.c1 + s->s0.c1;
sum += creal(r->s0.c1)*creal(r->s0.c1) + cimag(r->s0.c1)*cimag(r->s0.c1);
res += creal(r->s0.c1)*creal(r->s0.c1) + cimag(r->s0.c1)*cimag(r->s0.c1);
r->s0.c2 = c * r->s0.c2 + s->s0.c2;
sum += creal(r->s0.c2)*creal(r->s0.c2) + cimag(r->s0.c2)*cimag(r->s0.c2);
res += creal(r->s0.c2)*creal(r->s0.c2) + cimag(r->s0.c2)*cimag(r->s0.c2);

r->s1.c0 = c * r->s1.c0 + s->s1.c0;
sum += creal(r->s1.c0)*creal(r->s1.c0) + cimag(r->s1.c0)*cimag(r->s1.c0);
res += creal(r->s1.c0)*creal(r->s1.c0) + cimag(r->s1.c0)*cimag(r->s1.c0);
r->s1.c1 = c * r->s1.c1 + s->s1.c1;
sum += creal(r->s1.c1)*creal(r->s1.c1) + cimag(r->s1.c1)*cimag(r->s1.c1);
res += creal(r->s1.c1)*creal(r->s1.c1) + cimag(r->s1.c1)*cimag(r->s1.c1);
r->s1.c2 = c * r->s1.c2 + s->s1.c2;
sum += creal(r->s1.c2)*creal(r->s1.c2) + cimag(r->s1.c2)*cimag(r->s1.c2);
res += creal(r->s1.c2)*creal(r->s1.c2) + cimag(r->s1.c2)*cimag(r->s1.c2);

r->s2.c0 = c * r->s2.c0 + s->s2.c0;
sum += creal(r->s2.c0)*creal(r->s2.c0) + cimag(r->s2.c0)*cimag(r->s2.c0);
res += creal(r->s2.c0)*creal(r->s2.c0) + cimag(r->s2.c0)*cimag(r->s2.c0);
r->s2.c1 = c * r->s2.c1 + s->s2.c1;
sum += creal(r->s2.c1)*creal(r->s2.c1) + cimag(r->s2.c1)*cimag(r->s2.c1);
res += creal(r->s2.c1)*creal(r->s2.c1) + cimag(r->s2.c1)*cimag(r->s2.c1);
r->s2.c2 = c * r->s2.c2 + s->s2.c2;
sum += creal(r->s2.c2)*creal(r->s2.c2) + cimag(r->s2.c2)*cimag(r->s2.c2);
res += creal(r->s2.c2)*creal(r->s2.c2) + cimag(r->s2.c2)*cimag(r->s2.c2);

r->s3.c0 = c * r->s3.c0 + s->s3.c0;
sum += creal(r->s3.c0)*creal(r->s3.c0) + cimag(r->s3.c0)*cimag(r->s3.c0);
res += creal(r->s3.c0)*creal(r->s3.c0) + cimag(r->s3.c0)*cimag(r->s3.c0);
r->s3.c1 = c * r->s3.c1 + s->s3.c1;
sum += creal(r->s3.c1)*creal(r->s3.c1) + cimag(r->s3.c1)*cimag(r->s3.c1);
res += creal(r->s3.c1)*creal(r->s3.c1) + cimag(r->s3.c1)*cimag(r->s3.c1);
r->s3.c2 = c * r->s3.c2 + s->s3.c2;
sum += creal(r->s3.c2)*creal(r->s3.c2) + cimag(r->s3.c2)*cimag(r->s3.c2);
res += creal(r->s3.c2)*creal(r->s3.c2) + cimag(r->s3.c2)*cimag(r->s3.c2);
}
#ifdef OMP
} /* OpenMP closing brace */
Expand Down
1 change: 1 addition & 0 deletions linalg_eo.h
View file
Original file line number Diff line number Diff line change
Expand Up @@ -35,6 +35,7 @@
#include "linalg/assign.h"
/* #include "linalg/deri_linalg.h" */
#include "linalg/assign_mul_add_r.h"
#include "linalg/assign_mul_add_r_and_square.h"
#include "linalg/scalar_prod.h"
#include "linalg/mul_diff_mul.h"
#include "linalg/assign_add_mul.h"
Expand Down

0 comments on commit 141d577

Please sign in to comment.