From 9daf0bc3d1e3d22bd10856e5b3dfcc51f6a990db Mon Sep 17 00:00:00 2001 From: akolliasAMD <99202231+akolliasAMD@users.noreply.github.com> Date: Tue, 21 Mar 2023 08:48:39 -0600 Subject: [PATCH] Test Fixes (#710) * splitting CI tests in running SP first and MP second * set device before hipStreamSynchronize on tests [ROCm/rccl commit: 9a0d4a07a64645f03b44f8464b20b815f8251bd4] --- projects/rccl/.jenkins/common.groovy | 3 ++- projects/rccl/test/common/TestBedChild.cpp | 1 + 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/projects/rccl/.jenkins/common.groovy b/projects/rccl/.jenkins/common.groovy index c24ec7151b..98df3d346e 100644 --- a/projects/rccl/.jenkins/common.groovy +++ b/projects/rccl/.jenkins/common.groovy @@ -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) diff --git a/projects/rccl/test/common/TestBedChild.cpp b/projects/rccl/test/common/TestBedChild.cpp index feb9c0a346..a256de4192 100644 --- a/projects/rccl/test/common/TestBedChild.cpp +++ b/projects/rccl/test/common/TestBedChild.cpp @@ -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])); }