Add `OPENBLAS_DEFAULT_NUM_THREADS`

This allows Julia to set a default number of threads (usually `1`) to be
used when no other thread counts are specified [0], to short-circuit the
default OpenBLAS thread initialization routine that spins up a different
number of threads than Julia would otherwise choose.

The reason to add a new environment variable is that we want to be able
to configure OpenBLAS to avoid performing its initial memory
allocation/thread startup, as that can consume significant amounts of
memory, but we still want to be sensitive to legacy codebases that set
things like `OMP_NUM_THREADS` or `GOTOBLAS_NUM_THREADS`.  Creating a new
environment variable that is openblas-specific and is not already
publicly used to control the overall number of threads of programs like
Julia seems to be the best way forward.

[0] https://github.com/JuliaLang/julia/pull/46844
This commit is contained in:
Elliot Saba 2022-09-22 10:38:36 -07:00
parent 667d0e0b48
commit d2ce93179f
2 changed files with 9 additions and 1 deletions

View File

@ -823,6 +823,8 @@ void gotoblas_affinity_init(void) {
if (numprocs == 0) numprocs = readenv_atoi("OMP_NUM_THREADS"); if (numprocs == 0) numprocs = readenv_atoi("OMP_NUM_THREADS");
if (numprocs == 0) numprocs = readenv_atoi("OPENBLAS_DEFAULT_NUM_THREADS");
numnodes = 1; numnodes = 1;
if (numprocs == 1) { if (numprocs == 1) {

View File

@ -67,10 +67,16 @@ void openblas_read_env() {
openblas_env_thread_timeout=(unsigned int)ret; openblas_env_thread_timeout=(unsigned int)ret;
ret=0; ret=0;
if (readenv(p,"OPENBLAS_NUM_THREADS")) ret = atoi(p); if (readenv(p,"OPENBLAS_DEFAULT_NUM_THREADS")) ret = atoi(p);
if(ret<0) ret=0; if(ret<0) ret=0;
openblas_env_openblas_num_threads=ret; openblas_env_openblas_num_threads=ret;
ret=0;
if (readenv(p,"OPENBLAS_NUM_THREADS")) ret = atoi(p);
if(ret<0) ret=0;
if(ret != 0 || openblas_env_openblas_num_threads == 0)
openblas_env_openblas_num_threads=ret;
ret=0; ret=0;
if (readenv(p,"GOTO_NUM_THREADS")) ret = atoi(p); if (readenv(p,"GOTO_NUM_THREADS")) ret = atoi(p);
if(ret<0) ret=0; if(ret<0) ret=0;