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

Allow to do gemv and ger buffer allocation on the stack

Browse Source 
ger and gemv call blas_memory_alloc/free which in their turn
call blas_lock. blas_lock create thread contention when matrices
are small and the number of thread is high enough. We avoid
call blas_memory_alloc by replacing it with stack allocation.
This can be enabled with:
make -DMAX_STACK_ALLOC=2048
The given size (in byte) must be high enough to avoid thread contention
and small enough to avoid stack overflow.

Fix #478
This commit is contained in:
Jerome Robert
2014-12-26 14:42:00 +01:00
parent cbb3ab80e7
commit e9d9a8eae3
4 changed files with 32 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
interface/ger.c
View File

@@ -171,7 +171,15 @@ void CNAME(enum CBLAS_ORDER order,
if (incy < 0) y -= (n - 1) * incy;
if (incx < 0) x -= (m - 1) * incx;
#ifdef MAX_STACK_ALLOC
int stack_alloc_size = m;
if(stack_alloc_size > MAX_STACK_ALLOC / sizeof(FLOAT))
stack_alloc_size = 0;
FLOAT stack_buffer[stack_alloc_size];
buffer = stack_alloc_size ? stack_buffer : (FLOAT *)blas_memory_alloc(1);
#else
buffer = (FLOAT *)blas_memory_alloc(1);
#endif
#ifdef SMPTEST
nthreads = num_cpu_avail(2);
@@ -190,7 +198,10 @@ void CNAME(enum CBLAS_ORDER order,
}
#endif
blas_memory_free(buffer);
#ifdef MAX_STACK_ALLOC
if(!stack_alloc_size)
#endif
blas_memory_free(buffer);
FUNCTION_PROFILE_END(1, m * n + m + n, 2 * m * n);