optimizations for trsm

This commit is contained in:
wernsaar 2014-07-25 11:59:17 +02:00
parent 1b10ff129a
commit 3ea4dadd30
2 changed files with 18 additions and 6 deletions

View File

@ -128,6 +128,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
for(jjs = js; jjs < js + min_j; jjs += min_jj){
min_jj = min_j + js - jjs;
if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
else
if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
GEMM_ONCOPY(min_l, min_jj, b + (ls + jjs * ldb) * COMPSIZE, ldb, sb + min_l * (jjs - js) * COMPSIZE);
@ -194,6 +196,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
for(jjs = js; jjs < js + min_j; jjs += min_jj){
min_jj = min_j + js - jjs;
if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
else
if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
GEMM_ONCOPY(min_l, min_jj, b + (ls - min_l + jjs * ldb) * COMPSIZE, ldb, sb + min_l * (jjs - js) * COMPSIZE);

View File

@ -123,6 +123,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
for(jjs = js; jjs < js + min_j; jjs += min_jj){
min_jj = min_j + js - jjs;
if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
else
if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
#ifndef TRANSA
@ -177,6 +179,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
for(jjs = 0; jjs < min_j - min_l - ls + js; jjs += min_jj){
min_jj = min_j - min_l - ls + js - jjs;
if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
else
if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
#ifndef TRANSA
@ -238,6 +242,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
for(jjs = js; jjs < js + min_j; jjs += min_jj){
min_jj = min_j + js - jjs;
if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
else
if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
#ifndef TRANSA
@ -297,6 +303,8 @@ int CNAME(blas_arg_t *args, BLASLONG *range_m, BLASLONG *range_n, FLOAT *sa, FLO
for(jjs = 0; jjs < min_j - js + ls; jjs += min_jj){
min_jj = min_j - js + ls - jjs;
if (min_jj > GEMM_UNROLL_N*3) min_jj = GEMM_UNROLL_N*3;
else
if (min_jj > GEMM_UNROLL_N) min_jj = GEMM_UNROLL_N;
#ifndef TRANSA