floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge pull request #744 from jeromerobert/bug731 

Bug731
This commit is contained in:
Zhang Xianyi 2016-01-20 11:18:21 -06:00
parent 692d9c881c 14db1ca508
commit 6caa40302e
4 changed files with 19 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
CONTRIBUTORS.md
View File

@ -121,6 +121,11 @@ In chronological order:
* [2014-10-10] trmm and sgemm kernels (optimized for APM's X-Gene 1). * [2014-10-10] trmm and sgemm kernels (optimized for APM's X-Gene 1).
ARMv8 support. ARMv8 support.
* Jerome Robert <jeromerobert@gmx.com>
* [2015-01-01] Speed-up small `ger` and `gemv` using stack allocation (bug #478)
* [2015-12-23] `stack_check` in `gemv.c` (bug #722)
* [2016-01-19] Disable multi-threading in `ger` and `swap` for small matrices (bug #731)
* Dan Kortschak * Dan Kortschak
* [2015-01-07] Added test for drotmg bug #484. * [2015-01-07] Added test for drotmg bug #484.

7
interface/ger.c
View File

@ -174,8 +174,11 @@ void CNAME(enum CBLAS_ORDER order,
STACK_ALLOC(m, FLOAT, buffer); STACK_ALLOC(m, FLOAT, buffer);
#ifdef SMPTEST #ifdef SMPTEST
nthreads = num_cpu_avail(2); // Threshold chosen so that speed-up is > 1 on a Xeon E5-2630
if(1L * m * n > 24L * GEMM_MULTITHREAD_THRESHOLD)
nthreads = num_cpu_avail(2);
else
nthreads = 1;
if (nthreads == 1) { if (nthreads == 1) {
#endif #endif

7
interface/swap.c
View File

@ -77,12 +77,13 @@ void CNAME(blasint n, FLOAT *x, blasint incx, FLOAT *y, blasint incy){
if (incy < 0) y -= (n - 1) * incy; if (incy < 0) y -= (n - 1) * incy;
#ifdef SMP #ifdef SMP
nthreads = num_cpu_avail(1);
//disable multi-thread when incx==0 or incy==0 //disable multi-thread when incx==0 or incy==0
//In that case, the threads would be dependent. //In that case, the threads would be dependent.
if (incx == 0 || incy == 0) if (incx == 0 || incy == 0 || n < 2097152 * GEMM_MULTITHREAD_THRESHOLD / sizeof(FLOAT))
nthreads = 1; nthreads = 1;
else
nthreads = num_cpu_avail(1);
if (nthreads == 1) { if (nthreads == 1) {
#endif #endif

6
interface/zger.c
View File

@ -213,7 +213,11 @@ void CNAME(enum CBLAS_ORDER order,
buffer = (FLOAT *)blas_memory_alloc(1); buffer = (FLOAT *)blas_memory_alloc(1);
#ifdef SMPTEST #ifdef SMPTEST
nthreads = num_cpu_avail(2); // Threshold chosen so that speed-up is > 1 on a Xeon E5-2630
if(1L * m * n > 3L * sizeof(FLOAT) * GEMM_MULTITHREAD_THRESHOLD)
nthreads = num_cpu_avail(2);
else
nthreads = 1;
if (nthreads == 1) { if (nthreads == 1) {
#endif #endif