fix omp flag #871
fix omp flag #871
Conversation
|
@wojtuss Please help review, Thanks |
|
@luotao1 Can you help review this? |
|
Solve issue #429 |
deploy/cpp_infer/src/main.cpp
Outdated
| #ifdef USE_MKL | ||
| #pragma omp parallel | ||
| for (auto i = 0; i < 10; i++) { | ||
| assert(config.cpu_math_library_num_threads == omp_get_num_threads()); | ||
| } | ||
| #endif | ||
|
|
There was a problem hiding this comment.
I think it is better to keep this asset. Because before, MKL was not linked properly, but it does not report any error. Developers and users don't know if it is linked properly. In the code where MKL is used, it is only running on 1 thread, omp_get_num_threads()=1, even config.cpu_math_library_num_threads is set to 8.
There was a problem hiding this comment.
I have doubts about whether this check should be an assert here. In my opinion a user should be informed with a warning rather than prevented from running Paddle when the number of threads is not appropriate. It would be good to have this assert in a test, not in this file.
@luotao1 , what do you think about it?
Other than that LGTM.
|
@OliverLPH Please help test it！ |
lidanqing-intel
force-pushed
the
develop-fix-omp-flag
branch
2 times, most recently
from
1b02139
to
2a7b74d
Compare
|
@lidanqing-intel mkldnn 0 mkldnn 0
|
machine:Linux i9
Compile option: WITH_MKL ON
mkldnn 0
cpu_math_library_num_threads 1
花费了17.8732秒
mkldnn 0
cpu_math_library_num_threads 8
花费了14.1399秒
Windiws machine Please help verify on i7 performance @OliverLPH