Test Fixes (#710)
* splitting CI tests in running SP first and MP second
* set device before hipStreamSynchronize on tests
[ROCm/rccl commit: 9a0d4a07a6]
Этот коммит содержится в:
@@ -23,7 +23,8 @@ def runTestCommand (platform, project, gfilter)
|
||||
cd ${project.paths.project_build_prefix}/build/release/test
|
||||
${sudo} ulimit -l unlimited
|
||||
ulimit -a
|
||||
${sudo} RCCL_ENABLE_SIGNALHANDLER=0 NCCL_DEBUG=INFO HSA_FORCE_FINE_GRAIN_PCIE=1 ./rccl-UnitTests --gtest_filter=${gfilter} --gtest_output=xml --gtest_color=yes
|
||||
${sudo} UT_PROCESS_MASK=1 RCCL_ENABLE_SIGNALHANDLER=0 NCCL_DEBUG=INFO HSA_FORCE_FINE_GRAIN_PCIE=1 ./rccl-UnitTests --gtest_filter=${gfilter} --gtest_output=xml --gtest_color=yes
|
||||
${sudo} UT_PROCESS_MASK=2 RCCL_ENABLE_SIGNALHANDLER=0 NCCL_DEBUG=INFO HSA_FORCE_FINE_GRAIN_PCIE=1 ./rccl-UnitTests --gtest_filter=${gfilter} --gtest_output=xml --gtest_color=yes
|
||||
"""
|
||||
|
||||
platform.runCommand(this, command)
|
||||
|
||||
@@ -657,6 +657,7 @@ namespace RcclUnitTesting
|
||||
for (int localRank : localRanksToExecute)
|
||||
{
|
||||
if (this->verbose) INFO("Starting synchronization for rank %d\n", localRank);
|
||||
CHECK_HIP(hipSetDevice(this->deviceIds[localRank]));
|
||||
for (int i = 0; i < this->numStreamsPerGroup; i++)
|
||||
CHECK_HIP(hipStreamSynchronize(this->streams[localRank][i]));
|
||||
}
|
||||
|
||||
Ссылка в новой задаче
Block a user