New alpaka backend rules #110
Conversation
|
A new Pull Request was created by @smuzaffar for branch scramv3. @aandvalenzuela, @cmsbuild, @iarspider, @smuzaffar can you please review it and eventually sign? Thanks. |
|
cms-bot internal usage |
SCRAM/GMake/Makefile.rules
Outdated
| Enable/Disable alpaka gpus backends if not support on local host. It also calls\n\ | ||
| Also set native compute capabilities for the enabled alpaka backends.\n" |
|
test parameters:
|
|
please test |
Co-authored-by: Andrea Bocci <[email protected]>
|
Pull request #110 was updated. |
|
Pull request #110 was updated. |
|
When there is a change to the tool file, could you make |
Co-authored-by: Andrea Bocci <[email protected]>
|
Pull request #110 was updated. |
SCRAM/utils/backend-setup.sh
Outdated
| # remove existing capabilities flags | ||
| sed -r -i -e '/ROCM_FLAGS/s/--offload-arch=gfx[0-9a-f]+//g' ${TOOL}.tmp | ||
| # remove empty ROCM_FLAGS lines | ||
| sed -r -i -e '/ROCM_FLAGS=" *"/d' ${TOOL}.tmp | ||
|
|
||
| #add support for the capabilities found on this machine | ||
| for CAP in $CAPS; do | ||
| sed -i -e "s#</client>#</client>\n <flags ROCM_FLAGS=\"--offload-arch=${CAP}\"/>#" ${TOOL}.tmp | ||
| done |
There was a problem hiding this comment.
@fwyzard , may be something like this should be better for cuda tool too
There was a problem hiding this comment.
Do you mean splitting the capabilities to one entry per line ?
OK for me, but we have to change the way the CUDA flags are defined in the tool files.
There was a problem hiding this comment.
But if you mean adding it as a separate entry only when done by this script, sure, OK for me.
There was a problem hiding this comment.
yes, separate entry when done by the script. Basically it will remove all existing -gencode arch=compute_*,code=[sm_*,compute_*] entries from the tool file , delete any empty CUDA_FLAGS line and then add new entries on new line after </client>. This does not require changing existing cuda tool file and should also work if we decided to change the cuda file to have multiple CUDA_FLAGS lines
|
Pull request #110 was updated. |
|
please test |
|
+1 Summary: https://cmssdt.cern.ch/SDT/jenkins-artifacts/pull-request-integration/PR-9cf212/41488/summary.html Comparison SummarySummary:
|
|
+externals |
|
This pull request is fully signed and it will be integrated in one of the next scramv3 IBs (tests are also fine). This pull request will now be reviewed by the release team before it's merged. @sextonkennedy, @rappoccio, @mandrenguyen, @antoniovilela (and backports should be raised in the release meeting by the corresponding L2) |
cmsbuild
added
fully-signed
externals-approved
and removed
pending-signatures
externals-pending
labels
Added new build rules to enable/disable alpaka backends (
cuda/rocm). Also allow to change the cuda/rocm compute capabilities (see cms-sw/cmssw#45859 (comment) for details)