minor update

microsoft · Jun 28, 2023 · 8ee39ef · 8ee39ef
1 parent 504b24e
commit 8ee39ef
Show file tree

Hide file tree

Showing 3 changed files with 5 additions and 2 deletions.
diff --git a/test/deploy/run_tests.sh b/test/deploy/run_tests.sh
@@ -25,6 +25,10 @@ function run_mscclpp_test()
     -x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
     -npernode 8 /root/mscclpp/build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 1 -o /root/mscclpp/output.jsonl
 
+  /usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile /root/mscclpp/hostfile_mpi \
+    -x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \
+    -npernode 8 /root/mscclpp/build/test/mscclpp-test/allreduce_test_perf -b 3K -e 3G -f 2 -k 3 -o /root/mscclpp/output.jsonl
+
   echo "==================Run alltoall_test_perf on 2 nodes========================="
   /usr/local/mpi/bin/mpirun --allow-run-as-root -np 16 --bind-to numa -hostfile /root/mscclpp/hostfile_mpi \
     -x MSCCLPP_DEBUG=WARN -x LD_LIBRARY_PATH=/root/mscclpp/build:$LD_LIBRARY_PATH \

diff --git a/test/mscclpp-test/allreduce_test.cu b/test/mscclpp-test/allreduce_test.cu
@@ -70,6 +70,7 @@ __device__ void localReduceScatter(int* buff, int* scratch, int rank, int nRanks
   }
   int isComm = (threadIdx.x == 0) && (blockIdx.x == 0);
   int startRankInNode = (rank / nRanksPerNode) * nRanksPerNode;
+  int rankIdexInNode = rank % nRanksPerNode;
 
   for (int i = 1; i < nRanksPerNode; ++i) {
     int remoteSendToRank = (rank + i) % nRanksPerNode + startRankInNode;
@@ -79,7 +80,6 @@ __device__ void localReduceScatter(int* buff, int* scratch, int rank, int nRanks
 
     mscclpp::channel::SimpleDeviceChannel& devFstSendChan = constDevFstRoundChans[peerSendId];
     mscclpp::channel::SimpleDeviceChannel& devFstRecvChan = constDevFstRoundChans[peerRecvId];
-    int rankIdexInNode = rank % nRanksPerNode;
     size_t srcOffset =
         (((rankIdexInNode + i) % nRanksPerNode + startChunkIndex) * chunkSize + offsetInChunk) * sizeof(int);
     size_t dstOffset = rank * chunkSize * sizeof(int);

diff --git a/test/mscclpp-test/common.cc b/test/mscclpp-test/common.cc
@@ -507,7 +507,6 @@ int main(int argc, char* argv[]) {
             "[-c,--check <0/1>] \n\t"
             "[-T,--timeout <time in seconds>] \n\t"
             "[-G,--cudagraph <num graph launches>] \n\t"
-            "[-C,--report_cputime <0/1>] \n\t"
             "[-a,--average <0/1/2/3> report average iteration time <0=RANK0/1=AVG/2=MIN/3=MAX>] \n\t"
             "[-k,--kernel_num <kernel number of commnication primitive>] \n\t"
             "[-o, --output_file <output file name>] \n\t"