-
Notifications
You must be signed in to change notification settings - Fork 263
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
NCCL Library Integration #156
Conversation
657995b
to
94b52af
Compare
@@ -118,6 +118,7 @@ function(caffe_print_configuration_summary) | |||
caffe_status(" USE_LEVELDB : ${USE_LEVELDB}") | |||
caffe_status(" USE_LMDB : ${USE_LMDB}") | |||
caffe_status(" ALLOW_LMDB_NOLOCK : ${ALLOW_LMDB_NOLOCK}") | |||
caffe_status(" USE_NCCL : ${USE_NCCL}") |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Can you add this down below with the other CUDA options?
https://github.com/NVIDIA/caffe/blob/v0.15.1/cmake/Summary.cmake#L144-L148
CMake didn't auto-detect and use the deb package installation of NCCL automatically. I really like that feature for cuDNN (one less flag to set). |
@lukeyeager NCCL does not come as a Deb package. This is old school |
@@ -424,7 +424,7 @@ RegisterBrewFunction(time); | |||
|
|||
int main(int argc, char** argv) { | |||
// Print output to stderr (while still logging). | |||
FLAGS_alsologtostderr = 1; | |||
FLAGS_alsologtostderr = true; |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Was this causing a problem?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
No, just a purist habit... :) This flag is boolean.
https://github.com/NVIDIA/nccl/releases
It can find it, but it doesn't use it until I specify |
Agreed, if found it should be the default. If not found it should probably warn at that point that multi gpu will be disabled.
|
94b52af
to
6b4d102
Compare
Notes from Luke and Simon have been addressed, thank you. |
# ---[ NCCL | ||
if(USE_NCCL) | ||
add_definitions(-DUSE_NCCL) | ||
find_package(NCCL REQUIRED) |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Now I can't build Caffe with CUDA unless I have NCCL. I'd like to see this behavior (same as with USE_CUDNN):
USE_NCCL | found NCCL? | use? |
---|---|---|
On |
Yes | Yes |
On |
No | ERROR |
unset | Yes | Yes |
unset | No | No |
Off |
* | No |
6b4d102
to
f43e15e
Compare
|
@lukeyeager yes, I've seen it and fixing it now. |
bb8c03f
to
ad43d82
Compare
ad43d82
to
97d59ad
Compare
All redundant changes removed, CMake is green. |
No description provided.