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

Lazyly reinit threads after a fork in OMP mode 

This initializes the per-thread memory buffers which get
cleared/released on a fork via pthread_at_fork. Not doing so leads to
each thread calling blas_memory_alloc on almost every execution which
slows down the code significantly as the threads race for the memory
allocation using locks to serialize that.
This commit is contained in:
Alexander Grund 2020-10-01 15:41:42 +02:00
parent d3c0d6811b
commit 3094fc6c83
No known key found for this signature in database
GPG Key ID: E92C451FC21EF13F
1 changed files with 18 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

18
driver/others/blas_server_omp.c
View File

@ -48,6 +48,21 @@
#else
#ifndef likely
#ifdef __GNUC__
#define likely(x) __builtin_expect(!!(x), 1)
#else
#define likely(x) (x)
#endif
#endif
#ifndef unlikely
#ifdef __GNUC__
#define unlikely(x) __builtin_expect(!!(x), 0)
#else
#define unlikely(x) (x)
#endif
#endif
#ifndef OMP_SCHED
#define OMP_SCHED static
#endif
@ -350,6 +365,9 @@ static void exec_threads(blas_queue_t *queue, int buf_index){
int exec_blas(BLASLONG num, blas_queue_t *queue){
// Handle lazy re-init of the thread-pool after a POSIX fork
if (unlikely(blas_server_avail == 0)) blas_thread_init();
BLASLONG i, buf_index;
if ((num <= 0) || (queue == NULL)) return 0;