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

Merge pull request #4741 from shivammonaka/Pthread_Scalability_Improvement 

Enhancing Core Utilization in BLAS Calls: A Scalable Architecture
This commit is contained in:
Martin Kroeker 2024-06-20 13:36:23 +02:00 committed by GitHub
parent 0773695a5c 9e22d70957
commit 7e9a4ba427
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 12 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
driver/level3/level3_thread.c
View File

@ -570,6 +570,8 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
InitializeCriticalSection((PCRITICAL_SECTION)&level3_lock);
#else
static pthread_mutex_t level3_lock = PTHREAD_MUTEX_INITIALIZER;
static pthread_cond_t level3_wakeup = PTHREAD_COND_INITIALIZER;
volatile static BLASLONG CPU_AVAILABLE = MAX_CPU_NUMBER;
#endif
blas_arg_t newarg;
@ -639,6 +641,12 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
EnterCriticalSection((PCRITICAL_SECTION)&level3_lock);
#else
pthread_mutex_lock(&level3_lock);
while(CPU_AVAILABLE < nthreads) {
pthread_cond_wait(&level3_wakeup, &level3_lock);
}
CPU_AVAILABLE -= nthreads;
WMB;
pthread_mutex_unlock(&level3_lock);
#endif
#ifdef USE_ALLOC_HEAP
@ -783,6 +791,10 @@ static int gemm_driver(blas_arg_t *args, BLASLONG *range_m, BLASLONG
#elif defined(OS_WINDOWS)
LeaveCriticalSection((PCRITICAL_SECTION)&level3_lock);
#else
pthread_mutex_lock(&level3_lock);
CPU_AVAILABLE += nthreads;
WMB;
pthread_cond_signal(&level3_wakeup);
pthread_mutex_unlock(&level3_lock);
#endif