Torch integration

.azure-pipelines/integration-test-rocm.yml

@@ -94,7 +94,7 @@ jobs:
       script: |
         set -e
         export PATH=/usr/local/mpi/bin:$PATH
-        sudo mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN -x LD_PRELOAD="$(pwd)/build/apps/nccl/libmscclpp_nccl.so" \
+        sudo mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN -x LD_PRELOAD="$(pwd)/build/lib/libmscclpp_nccl.so" \
           -x ALLREDUCE_SMALL_MSG_BOUNDARY=32K -x ALLREDUCE_LARGE_MSG_BOUNDARY=1M ./rccl-tests/build/all_reduce_perf -b 1K -e 1G -f 2 -d half -G 20 -w 10 -n 100
       workingDirectory: '$(System.DefaultWorkingDirectory)'
 
@@ -106,7 +106,7 @@ jobs:
       script: |
         set -e
         export PATH=/usr/local/mpi/bin:$PATH
-        sudo mpirun -np 8 --bind-to numa --allow-run-as-root -x LD_PRELOAD=$(pwd)/build/apps/nccl/libmscclpp_nccl.so -x NCCL_DEBUG=WARN \
+        sudo mpirun -np 8 --bind-to numa --allow-run-as-root -x LD_PRELOAD=$(pwd)/build/lib/libmscclpp_nccl.so -x NCCL_DEBUG=WARN \
           -x ALLREDUCEPKT_IP_JSON_FILE=./msccl-users/execution-files/allreduce_mi300_packet.json \
           -x ALLREDUCE_IP_JSON_FILE=./msccl-users/execution-files/allreduce_mi300_sm_mscclpp.json \
           -x ALLREDUCE_SMALL_MSG_BOUNDARY=32K -x ALLREDUCE_LARGE_MSG_BOUNDARY=1M ./rccl-tests/build/all_reduce_perf \

.azure-pipelines/templates/integration-test.yaml

@@ -74,13 +74,13 @@ steps:
       parallel-ssh -o . -t 0 -h ${HOSTFILE} -x "-i ${KeyFilePath}" \
         -O $SSH_OPTION 'sudo docker exec -t mscclpp-test bash -c " \
         export PATH=/usr/local/mpi/bin:\$PATH;                     \
-        export LD_LIBRARY_PATH=/root/mscclpp/build:\$LD_LIBRARY_PATH;  \
+        export LD_LIBRARY_PATH=/root/mscclpp/build/lib:\$LD_LIBRARY_PATH;  \
         cd /root/mscclpp;                                         \
         set -e;                                                   \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -o output.jsonl;       \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl;  \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 2 -o output.jsonl;  \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 3 -o output.jsonl"'
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -o output.jsonl;       \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl;  \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 2 -o output.jsonl;  \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allgather_test_perf -b 1K -e 1G -f 2 -k 3 -o output.jsonl"'
       kill $CHILD_PID
     workingDirectory: '$(System.DefaultWorkingDirectory)'
 
@@ -101,9 +101,9 @@ steps:
         -O $SSH_OPTION 'sudo docker exec -t mscclpp-test bash -c "\
         set -e;                                                   \
         export PATH=/usr/local/mpi/bin:\$PATH;                    \
-        export LD_LIBRARY_PATH=/root/mscclpp/build:\$LD_LIBRARY_PATH;  \
+        export LD_LIBRARY_PATH=/root/mscclpp/build/lib:\$LD_LIBRARY_PATH;  \
         cd /root/mscclpp;                                         \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/sendrecv_test_perf -b 1K -e 1G -f 2 -o output.jsonl"'
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/sendrecv_test_perf -b 1K -e 1G -f 2 -o output.jsonl"'
       kill $CHILD_PID
     workingDirectory: '$(System.DefaultWorkingDirectory)'
 
@@ -124,15 +124,15 @@ steps:
         -O $SSH_OPTION 'sudo docker exec -t mscclpp-test bash -c "\
         set -e;                                                   \
         export PATH=/usr/local/mpi/bin:\$PATH;                     \
-        export LD_LIBRARY_PATH=/root/mscclpp/build:\$LD_LIBRARY_PATH;  \
+        export LD_LIBRARY_PATH=/root/mscclpp/build/lib:\$LD_LIBRARY_PATH;  \
         cd /root/mscclpp;                                         \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -o output.jsonl;                 \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl;            \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 2 -o output.jsonl;            \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 3 -o output.jsonl;            \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 4 -o output.jsonl;            \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 12M -e 48M -i 3145728 2 -k 5 -o output.jsonl;  \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/allreduce_test_perf -b 24K -e 768K -i 24576 -k 6 -w 100 -n 100 -o output.jsonl"'
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -o output.jsonl;                 \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl;            \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 2 -o output.jsonl;            \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 3 -o output.jsonl;            \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allreduce_test_perf -b 1K -e 1G -f 2 -k 4 -o output.jsonl;            \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allreduce_test_perf -b 12M -e 48M -i 3145728 2 -k 5 -o output.jsonl;  \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/allreduce_test_perf -b 24K -e 768K -i 24576 -k 6 -w 100 -n 100 -o output.jsonl"'
       kill $CHILD_PID
     workingDirectory: '$(System.DefaultWorkingDirectory)'
 
@@ -152,10 +152,10 @@ steps:
         -O $SSH_OPTION 'sudo docker exec -t mscclpp-test bash -c "\
         set -e;                                                   \
         export PATH=/usr/local/mpi/bin:\$PATH;                    \
-        export LD_LIBRARY_PATH=/root/mscclpp/build:\$LD_LIBRARY_PATH; \
+        export LD_LIBRARY_PATH=/root/mscclpp/build/lib:\$LD_LIBRARY_PATH; \
         cd /root/mscclpp;                                         \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/alltoall_test_perf -b 1K -e 1G -f 2 -o output.jsonl; \
-        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/test/mscclpp-test/alltoall_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl"'
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/alltoall_test_perf -b 1K -e 1G -f 2 -o output.jsonl; \
+        mpirun --allow-run-as-root -np 8 --bind-to numa -x MSCCLPP_DEBUG=WARN ./build/bin/mscclpp-test/alltoall_test_perf -b 1K -e 1G -f 2 -k 1 -o output.jsonl"'
       kill $CHILD_PID
     workingDirectory: '$(System.DefaultWorkingDirectory)'
 
@@ -177,7 +177,7 @@ steps:
         set -e;                                                   \
         cd /root/mscclpp;                                         \
         export PATH=/usr/local/mpi/bin:\$PATH;                    \
-        export LD_LIBRARY_PATH=/root/mscclpp/build:\$LD_LIBRARY_PATH; \
+        export LD_LIBRARY_PATH=/root/mscclpp/build/lib:\$LD_LIBRARY_PATH; \
         python3 test/mscclpp-test/check_perf_result.py --perf-file output.jsonl --baseline-file ${{ parameters.perfBaselineFile }}"'
       kill $CHILD_PID
     workingDirectory: '$(System.DefaultWorkingDirectory)'
@@ -200,7 +200,7 @@ steps:
         set -e;                                                    \
         cd /root/mscclpp;                                          \
         export PATH=/usr/local/mpi/bin:\$PATH;                     \
-        export LD_LIBRARY_PATH=/root/mscclpp/build:\$LD_LIBRARY_PATH; \
+        export LD_LIBRARY_PATH=/root/mscclpp/build/lib:\$LD_LIBRARY_PATH; \
         python3 -m pip install .;                                     \
         mpirun --allow-run-as-root -tag-output -x LD_LIBRARY_PATH=/usr/local/cuda/compat:$LD_LIBRARY_PATH -x MSCCLPP_HOME=/root/mscclpp -np 8 python3 ./python/mscclpp_benchmark/allreduce_bench.py"'
       kill $CHILD_PID
@@ -223,9 +223,9 @@ steps:
         -O $SSH_OPTION 'sudo docker exec -t mscclpp-test bash -c "\
         set -e;                                                   \
         export PATH=/usr/local/mpi/bin:\$PATH;                    \
-        export LD_LIBRARY_PATH=/root/mscclpp/build:\$LD_LIBRARY_PATH; \
+        export LD_LIBRARY_PATH=/root/mscclpp/build/lib:\$LD_LIBRARY_PATH; \
         cd /root/mscclpp;                                         \
-        ./build/test/perf/fifo_test"'
+        ./build/bin/perf/fifo_test"'
       kill $CHILD_PID
     workingDirectory: '$(System.DefaultWorkingDirectory)'