-
Notifications
You must be signed in to change notification settings - Fork 82
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
how to improve the speed of network:forward #68
Comments
This must be the reason |
Yes, I can confirm this. When I compile OpenBlas for Android-Torch, it becomes much more faster than before. But it only works on the version of ARM64. Because OpenBlas only supports hard-float for ARMV7. |
Hi ExenVitor, care to write more detailed instructions on how to compile torch-android with OpenBlas? Thanks! |
Let me leaborate, I'm building for ARM64 (ARCH=${ARCH:-"v8"}) and linking OpenBLAS library /opt/OpenBLAS/lib/libopenblas.a I get this error: /opt/OpenBLAS/lib/libopenblas.a: error adding symbols: File in wrong format OpenBlas compiled using this command: make TARGET=ARMV8 BINARY=64 HOSTCC=gcc CC=aarch64-linux-android-gcc NOFORTRAN=1 USE_OPENMP=1 |
@antoniosimunovic Sorry for the late reply. It looks like you are linking System's OpenBLAS library. You need install the |
Thanks, that is the solution! I've put the install script before the make command in build.sh. |
check notfound
|
Thanks @ExenVitor for your excellent contribution |
I am using cpu only,and it takes about 4500ms to using the function network:forward to process an image(96x96).
how to use OpenMP or OpenCL to improve the speed?
And it is very strange that I use torch.setnumthreads(1) is much more faster than torch.setnumthreads(4)
The text was updated successfully, but these errors were encountered: