From 8ede2f139e72bcdd1092c06314e1eb2b9f37775e Mon Sep 17 00:00:00 2001
From: PeixuanZuo <94887879+PeixuanZuo@users.noreply.github.com>
Date: Mon, 24 Jul 2023 13:57:48 +0800
Subject: [PATCH] [ROCm] Optimize ROCm CI pipeline 2 (#16691)

- Set `KERNEL_EXPLORER_TEST_USE_CUPY=1` to replace numpy with cupy on
kernel explorer test.

KERNEL_EXPLORER_TEST_USE_CUPY=0 The CPU utilization is shown as below:

![image](https://github.com/microsoft/onnxruntime/assets/94887879/91724b78-0b4e-4cbd-ad88-83cad9976472)

KERNEL_EXPLORER_TEST_USE_CUPY=1 The CPU utilization is shown as below:

![image](https://github.com/microsoft/onnxruntime/assets/94887879/58239911-667c-4d5f-bb78-deca60d0266f)


- Use `Bash@3`.
- Update shell script.
---
 .../orttraining-pai-ci-pipeline.yml           | 21 ++++----
 .../migraphx-ci-pipeline-env.Dockerfile       |  3 +-
 tools/ci_build/github/pai/pai_clean_device.sh | 51 +++++++++++--------
 .../pai/pai_huggingface_bert_large_test.sh    | 21 +++++---
 .../pai/rocm-ci-pipeline-env.Dockerfile       |  4 +-
 5 files changed, 59 insertions(+), 41 deletions(-)

diff --git a/tools/ci_build/github/azure-pipelines/orttraining-pai-ci-pipeline.yml b/tools/ci_build/github/azure-pipelines/orttraining-pai-ci-pipeline.yml
index 4e4073ae84..1295f84142 100644
--- a/tools/ci_build/github/azure-pipelines/orttraining-pai-ci-pipeline.yml
+++ b/tools/ci_build/github/azure-pipelines/orttraining-pai-ci-pipeline.yml
@@ -133,12 +133,11 @@ jobs:
       DockerBuildArgs: "--build-arg BUILD_UID=$( id -u )"
       Repository: onnxruntimetrainingrocm-cibuild-rocm$(RocmVersion)-test
 
-  - task: CmdLine@2
+  - task: Bash@3
     inputs:
-      script: |-
-        echo "Select agent: $(Agent.Name), GPU: $HIP_VISIBLE_DEVICES, render: $DRIVER_RENDER"
-        bash $(Build.SourcesDirectory)/tools/ci_build/github/pai/pai_clean_device.sh $(Agent.Name) $HIP_VISIBLE_DEVICES
-      workingDirectory: $(Build.SourcesDirectory)
+      targetType: filePath
+      filePath: $(Build.SourcesDirectory)/tools/ci_build/github/pai/pai_clean_device.sh
+      arguments: -n $(Agent.Name) -d $HIP_VISIBLE_DEVICES -r $DRIVER_RENDER
     displayName: 'Check ROCm Environment'
 
   - task: CmdLine@2
@@ -182,6 +181,7 @@ jobs:
               set -ex; \
               export KERNEL_EXPLORER_BUILD_DIR=/build/$(BuildConfig); \
               export KERNEL_EXPLORER_BATCHED_GEMM_MAX_BATCH_SIZE=8; \
+              export KERNEL_EXPLORER_TEST_USE_CUPY=1; \
               pytest /onnxruntime_src/onnxruntime/python/tools/kernel_explorer/ -n 8 --reruns 1 --durations=100"
       workingDirectory: $(Build.SourcesDirectory)
     displayName: 'Run kernel explorer tests'
@@ -206,7 +206,7 @@ jobs:
               set -ex; \
               export PYTHONPATH=/build/$(BuildConfig); \
               python -m onnxruntime.training.ortmodule.torch_cpp_extensions.install; \
-              bash /onnxruntime_src/tools/ci_build/github/pai/pai_huggingface_bert_large_test.sh $(RocmVersion)"
+              bash /onnxruntime_src/tools/ci_build/github/pai/pai_huggingface_bert_large_test.sh -v $(RocmVersion)"
       workingDirectory: $(Build.SourcesDirectory)
     displayName: 'Run Python Hugging-Face BERT-L test'
     condition: succeededOrFailed()
@@ -250,11 +250,12 @@ jobs:
     displayName: 'Run orttraining_ortmodule_tests.py'
     condition: succeededOrFailed()
 
-  - task: CmdLine@2
+
+  - task: Bash@3
     inputs:
-      script: |-
-        bash $(Build.SourcesDirectory)/tools/ci_build/github/pai/pai_clean_device.sh $(Agent.Name) $HIP_VISIBLE_DEVICES
-      workingDirectory: $(Build.SourcesDirectory)
+      targetType: filePath
+      filePath: $(Build.SourcesDirectory)/tools/ci_build/github/pai/pai_clean_device.sh
+      arguments: -n $(Agent.Name) -d $HIP_VISIBLE_DEVICES -r $DRIVER_RENDER
     displayName: 'Clean ROCm Environment'
     condition: always()
 
diff --git a/tools/ci_build/github/linux/docker/migraphx-ci-pipeline-env.Dockerfile b/tools/ci_build/github/linux/docker/migraphx-ci-pipeline-env.Dockerfile
index 90e9731a35..d3bca26875 100644
--- a/tools/ci_build/github/linux/docker/migraphx-ci-pipeline-env.Dockerfile
+++ b/tools/ci_build/github/linux/docker/migraphx-ci-pipeline-env.Dockerfile
@@ -6,7 +6,8 @@ ARG MIGRAPHX_VERSION=rocm-5.5.0
 ENV DEBIAN_FRONTEND noninteractive
 ENV MIGRAPHX_DISABLE_FAST_GELU=1
 
-RUN apt-get clean && apt-get update -y && apt-get upgrade -y && apt-get install -y locales unzip
+RUN apt-get update -y && apt-get upgrade -y && apt-get autoremove -y && \
+    apt-get install -y locales unzip && apt-get clean -y
 RUN locale-gen en_US.UTF-8
 RUN update-locale LANG=en_US.UTF-8
 ENV LC_ALL C.UTF-8
diff --git a/tools/ci_build/github/pai/pai_clean_device.sh b/tools/ci_build/github/pai/pai_clean_device.sh
index 4ebbcf4920..98b680d4f4 100755
--- a/tools/ci_build/github/pai/pai_clean_device.sh
+++ b/tools/ci_build/github/pai/pai_clean_device.sh
@@ -1,38 +1,47 @@
 #!/bin/bash
 set -ex
 
-agentName=$1
-target_device=$2
-echo "agent name $agentName"
-echo "agent target device : $target_device"
+usage() { echo "Usage: $0 [-n <agent name>] [-d <target device>] [-r <driver render>]" 1>&2; exit 1; }
 
-echo -e "\n ---- rocm-smi"
+while getopts "n:d:r:" parameter_Option
+do case "${parameter_Option}"
+in
+n) AGENT_NAME=${OPTARG};;
+d) TARGET_DEVICE=${OPTARG};;
+r) DRIVER_RENDER=${OPTARG};;
+*) usage ;;
+esac
+done
+
+echo "Agent Name: $AGENT_NAME, Target Device: $TARGET_DEVICE, Driver Render: $DRIVER_RENDER"
+
+echo -e "\n ---- Execute rocm-smi"
 rocm-smi
 
-echo -e "\n ---- rocm-smi --showpids"
+echo -e "\n ---- Execute rocm-smi --showpids"
 rocm-smi --showpids
 
-echo -e "\n ---- rocm-smi --showpidgpus"
+echo -e "\n ---- Execute rocm-smi --showpidgpus"
 rocm-smi --showpidgpus
 
-echo -e "\n ---- rocm-smi --showpids detail"
+echo -e "\n ---- Execute rocm-smi --showpids detail"
 rocm-smi --showpids | awk '$1 ~/[0-9]+/{if((NR>6)) {print $1}}' | xargs -I {} ps {}
 
-echo -e "\n ---- rocm-smi --showmeminfo"
+echo -e "\n ---- Execute rocm-smi --showmeminfo"
 rocm-smi --showmeminfo vram vis_vram gtt
 
-echo -e "\n ---- Clean up the process that is using the target device"
-gpu_details=$(rocm-smi --showpidgpus)
-pid_lines=$(echo "$gpu_details" | grep -n "DRM device" | cut -d ":" -f 1)
-pid_lines_array=($pid_lines)
+echo -e "\n ---- Clean up processes that use the target device $TARGET_DEVICE"
+GPU_USED_BY_PIDS=$(rocm-smi --showpidgpus)
+PID_NUMBERS_LINES=$(echo "$GPU_USED_BY_PIDS" | grep -n "DRM device" | cut -d ":" -f 1)
+PID_NUMBERS_LINES_ARRAY=($PID_NUMBERS_LINES)
 
-for ((i = 0; i < ${#pid_lines_array[@]}; i++)); do
-    pid_line=${pid_lines_array[$i]}
-    pid=$(echo "$gpu_details" | awk '{print $2}' | sed -n "${pid_line}p")
-    gpu_line=$((pid_line + 1))
-    pid_gpu=$(echo "$gpu_details" | sed -n "${gpu_line}p" | sed -e 's/^[ ]*//g' | sed -e 's/[ ]*$//g')
-    if [ "$pid_gpu" == "$target_device" ]; then
-        echo "kill pid: $pid, gpu: $pid_gpu"
-        kill -9 $pid
+for ((i = 0; i < ${#PID_NUMBERS_LINES_ARRAY[@]}; i++)); do
+    PID_NUMBER_LINE=${PID_NUMBERS_LINES_ARRAY[$i]}
+    PID_NUMBER=$(echo "$GPU_USED_BY_PIDS" | awk '{print $2}' | sed -n "${PID_NUMBER_LINE}p")
+    GPU_USED_BY_PID_LINE=$((PID_NUMBER_LINE + 1))
+    GPU_USED_BY_PID=$(echo "$GPU_USED_BY_PIDS" | sed -n "${GPU_USED_BY_PID_LINE}p" | sed -e 's/^[ ]*//g' | sed -e 's/[ ]*$//g')
+    if [ "$GPU_USED_BY_PID" == "$TARGET_DEVICE" ]; then
+        echo "kill pid: $PID_NUMBER, using gpu: $GPU_USED_BY_PID"
+        kill -9 "$PID_NUMBER"
     fi
 done
diff --git a/tools/ci_build/github/pai/pai_huggingface_bert_large_test.sh b/tools/ci_build/github/pai/pai_huggingface_bert_large_test.sh
index 3b8e828fe9..fb4dbeb2e7 100644
--- a/tools/ci_build/github/pai/pai_huggingface_bert_large_test.sh
+++ b/tools/ci_build/github/pai/pai_huggingface_bert_large_test.sh
@@ -2,15 +2,22 @@
 
 set -ex
 
-rocm_version=$1
-mi200_gpus=$(rocm-smi --showproductname | grep -c "MI250" | xargs)
+usage() { echo "Usage: $0 [-v <ROCm version>]" 1>&2; exit 1; }
 
-echo "mi200_gpus: $mi200_gpus"
+while getopts "v:" parameter_Option
+do case "${parameter_Option}"
+in
+v) ROCM_VERSION=${OPTARG};;
+*) usage ;;
+esac
+done
 
-if [ "$mi200_gpus" -gt "0" ]; then
-  result_file=ci-mi200.huggingface.bert-large-rocm${rocm_version}.json
+MI200_DEVICE_NUMBERS=$(rocm-smi --showproductname | grep -c "MI250" | xargs)
+
+if [ "$MI200_DEVICE_NUMBERS" -gt "0" ]; then
+  RESULT_FILE=ci-mi200.huggingface.bert-large-rocm${ROCM_VERSION}.json
 else
-  result_file=ci-mi100.huggingface.bert-large-rocm${rocm_version}.json
+  RESULT_FILE=ci-mi100.huggingface.bert-large-rocm${ROCM_VERSION}.json
 fi
 
 python \
@@ -33,4 +40,4 @@ cat ci-pipeline-actual.json
 
 python /onnxruntime_src/orttraining/tools/ci_test/compare_huggingface.py \
   ci-pipeline-actual.json \
-  /onnxruntime_src/orttraining/tools/ci_test/results/${result_file}
+  /onnxruntime_src/orttraining/tools/ci_test/results/"$RESULT_FILE"
diff --git a/tools/ci_build/github/pai/rocm-ci-pipeline-env.Dockerfile b/tools/ci_build/github/pai/rocm-ci-pipeline-env.Dockerfile
index 540aeaf351..7540856913 100644
--- a/tools/ci_build/github/pai/rocm-ci-pipeline-env.Dockerfile
+++ b/tools/ci_build/github/pai/rocm-ci-pipeline-env.Dockerfile
@@ -1,6 +1,6 @@
-FROM rocm/pytorch:rocm5.5_ubuntu20.04_py3.8_pytorch_1.13.1
+FROM rocm/cupy:rocm5.5.0_ubuntu20.04_py3.8_pytorch2.0.0_cupy13.0.0
 
-RUN apt-get update -y && apt-get upgrade -y
+RUN apt-get update -y && apt-get upgrade -y && apt-get autoremove -y && apt-get clean -y
 
 WORKDIR /stage