[EP Perf] Add CI option to enable TRT-OSS parser (#19448)

### Description
<!-- Describe your changes. -->
* Introducing CI option to enable TRT-OSS parser, during ep perf
testing:

![image](https://github.com/microsoft/onnxruntime/assets/109183385/a9ba6393-6b94-4b8f-8ca4-ba7bc7954504)

By default, open-sourced onnx-tensorrt parser listed under
[cmake/deps.txt](https://github.com/microsoft/onnxruntime/blob/main/cmake/deps.txt#L39-L40)
will be used if enabling this option.


### To verify this option and check the difference during ORT image
build:
If this option is enabled:
<img width="649" alt="image"
src="https://github.com/microsoft/onnxruntime/assets/109183385/3b778583-451e-4617-ba8c-c064442e60fd">

If this option is not enabled (by default):
<img width="683" alt="image"
src="https://github.com/microsoft/onnxruntime/assets/109183385/cd8383ba-eff4-4536-94ab-a1424bb858ab">

* update default usage of cmake/trt version to the latest

### Motivation and Context
Make it easier to test oss parser and find potential gap between
tensorrt builtin/oss parser.

Schedule runs with oss parser will be set after this PR gets merged

onnxruntime/python/tools/tensorrt/perf/build/build_image.py

@@ -45,7 +45,7 @@ def get_common_docker_build_args(args: argparse.Namespace) -> List[str]:
     :return: A list of common 'docker build' arguments.
     """
 
-    return [
+    command = [
         "--no-cache",
         "-t",
         f"{args.image_name}",
@@ -54,6 +54,14 @@ def get_common_docker_build_args(args: argparse.Namespace) -> List[str]:
         "--build-arg",
         f"ONNXRUNTIME_BRANCH={args.branch}",
     ]
+    if args.use_tensorrt_oss_parser:
+        command.extend(
+            [
+                "--build-arg",
+                "PARSER_CONFIG=--use_tensorrt_oss_parser",
+            ]
+        )
+    return command
 
 
 def is_valid_ver_str(version: str, min_comps: int = 0, max_comps: int = 0) -> bool:
@@ -187,7 +195,7 @@ def parse_arguments() -> argparse.Namespace:
     parser.add_argument("-r", "--repo_path", required=True, help="Path to the onnxruntime repository")
     parser.add_argument("-i", "--image_name", required=True, help="The resulting Docker image name")
     parser.add_argument("-b", "--branch", default="main", help="Name of the onnxruntime git branch to checkout")
-    parser.add_argument("-t", "--trt_version", default="8.4.1.5", help="TensorRT version (e.g., 8.4.1.5)")
+    parser.add_argument("-t", "--trt_version", default="8.6.1.6", help="TensorRT version (e.g., 8.6.1.6)")
     parser.add_argument("-a", "--cuda_arch", default="75", help="CUDA architecture (e.g., 75)")
 
     # Command-line options for installing TensorRT from binaries.
@@ -208,6 +216,12 @@ def parse_arguments() -> argparse.Namespace:
         help="CUDA version (e.g., 8.6) used to find TensorRT EA binary tar.gz package",
     )
     parser.add_argument("--trt_bins_dir", default="", help="Directory containing TensorRT tar.gz package")
+    parser.add_argument(
+        "--use_tensorrt_oss_parser",
+        action="store_true",
+        default=False,
+        help="Use TensorRT OSS Parser",
+    )
 
     return parser.parse_args()
 

onnxruntime/python/tools/tensorrt/perf/build/ort_build_latest.py

@@ -13,6 +13,12 @@ def parse_arguments():
     parser.add_argument("-b", "--branch", required=False, default="master", help="Github branch to test perf off of")
     parser.add_argument("-s", "--save", required=False, help="Directory to archive wheel file")
     parser.add_argument("-a", "--use_archived", required=False, help="Archived wheel file")
+    parser.add_argument(
+        "--use_tensorrt_oss_parser",
+        action="store_true",
+        default=False,
+        help="Use TensorRT OSS Parser",
+    )
     args = parser.parse_args()
     return args
 
@@ -35,14 +41,14 @@ def install_new_ort_wheel(ort_master_path):
 def main():
     args = parse_arguments()
 
-    cmake_tar = "cmake-3.18.4-Linux-x86_64.tar.gz"
+    cmake_tar = "cmake-3.28.3-linux-x86_64.tar.gz"
     if not os.path.exists(cmake_tar):
-        subprocess.run(["wget", "-c", "https://cmake.org/files/v3.18/" + cmake_tar], check=True)
+        subprocess.run(["wget", "-c", "https://cmake.org/files/v3.28/" + cmake_tar], check=True)
     tar = tarfile.open(cmake_tar)
     tar.extractall()
     tar.close()
 
-    os.environ["PATH"] = os.path.join(os.path.abspath("cmake-3.18.4-Linux-x86_64"), "bin") + ":" + os.environ["PATH"]
+    os.environ["PATH"] = os.path.join(os.path.abspath("cmake-3.28.3-linux-x86_64"), "bin") + ":" + os.environ["PATH"]
     os.environ["CUDACXX"] = os.path.join(args.cuda_home, "bin", "nvcc")
 
     ort_master_path = args.ort_master_path
@@ -57,24 +63,24 @@ def main():
         subprocess.run(["git", "fetch"], check=True)
         subprocess.run(["git", "checkout", args.branch], check=True)
         subprocess.run(["git", "pull", "origin", args.branch], check=True)
-        subprocess.run(
-            [
-                "./build.sh",
-                "--config",
-                "Release",
-                "--use_tensorrt",
-                "--tensorrt_home",
-                args.tensorrt_home,
-                "--cuda_home",
-                args.cuda_home,
-                "--cudnn",
-                "/usr/lib/x86_64-linux-gnu",
-                "--build_wheel",
-                "--skip_tests",
-                "--parallel",
-            ],
-            check=True,
-        )
+        command = [
+            "./build.sh",
+            "--config",
+            "Release",
+            "--use_tensorrt",
+            "--tensorrt_home",
+            args.tensorrt_home,
+            "--cuda_home",
+            args.cuda_home,
+            "--cudnn",
+            "/usr/lib/x86_64-linux-gnu",
+            "--build_wheel",
+            "--skip_tests",
+            "--parallel",
+        ]
+        if args.use_tensorrt_oss_parser:
+            command.append("--use_tensorrt_oss_parser")
+        subprocess.run(command, check=True)
 
         ort_wheel_file = install_new_ort_wheel(ort_master_path)
 

tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-daily-perf-pipeline.yml

@@ -15,6 +15,11 @@ parameters:
   - 8.6.1.6
   - BIN
 
+- name: UseTensorrtOssParser
+  displayName: Use TensorRT-OSS Parser
+  type: boolean
+  default: false
+
 - name: ModelGroups
   type: object
   default: 
@@ -73,7 +78,7 @@ jobs:
       value: ort-image-$(Build.BuildId)
 
   steps:
-    - ${{ if eq(parameters.TrtVersion, 'BIN') }}:
+    - ${{ if and(eq(parameters.TrtVersion, 'BIN'), eq(parameters.UseTensorrtOssParser, false)) }}:
       - script: 'ls -al $(trtBinsDir)'
         displayName: 'Show available TensorRT .tar.gz packages'
 
@@ -83,11 +88,19 @@ jobs:
       - script: 'python3 $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build/build_image.py -r $(Build.SourcesDirectory) -i $(image) -b $(branchName) -t $(trtVersion) -a 75 --install_bin --tar_cuda_version=$(tarCudaVersion) --tar_cudnn_version=$(tarCudnnVersion) --trt_bins_dir=.'
         displayName: 'Install TensorRT from binaries and build latest ORT Image'
         workingDirectory: '$(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build'
-    - ${{ else }}:
+    
+    # Build ORT with TensorRT built-in parser 
+    - ${{ if and(ne(parameters.TrtVersion, 'BIN'), eq(parameters.UseTensorrtOssParser, false)) }}:
       - script: 'python3 $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build/build_image.py -r $(Build.SourcesDirectory) -i $(image) -b $(branchName) -t $(trtVersion) -a 75'
-        displayName: 'Build latest ORT Image'
+        displayName: 'Build latest ORT Image with TensorRT built-in parser'
         workingDirectory: '$(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build'
-        
+    
+    # Build ORT with TensorRT OSS parser 
+    - ${{ if and(ne(parameters.TrtVersion, 'BIN'), eq(parameters.UseTensorrtOssParser, true)) }}:
+      - script: 'python3 $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build/build_image.py -r $(Build.SourcesDirectory) -i $(image) -b $(branchName) -t $(trtVersion) -a 75 --use_tensorrt_oss_parser'
+        displayName: 'Build latest ORT Image with TensorRT OSS parser'
+        workingDirectory: '$(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/build'
+    
     - ${{ if eq(parameters.MemTest, true) }}:
       - script: '$(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/mem_test/run_mem_test_docker.sh -d $(image) -p $(Build.SourcesDirectory)/onnxruntime/python/tools/tensorrt/perf/mem_test/ -w /code/ -l false'
         displayName: 'Run Memory Test'

tools/ci_build/github/linux/docker/Dockerfile.ubuntu_cuda11_8_tensorrt8_6

@@ -82,8 +82,9 @@ RUN if [ -z "$ONNXRUNTIME_COMMIT_ID" ] ; then echo "Building branch ${ONNXRUNTIM
     git reset --hard ${ONNXRUNTIME_COMMIT_ID} && git submodule update --recursive ; fi
 
 # Build ORT
-ENV CUDA_MODULE_LOADING "LAZY"
-RUN /bin/sh build.sh --parallel --build_shared_lib --cuda_home /usr/local/cuda --cudnn_home /usr/lib/x86_64-linux-gnu/ --use_tensorrt --tensorrt_home /usr/lib/x86_64-linux-gnu/ --config Release --build_wheel --skip_tests --skip_submodule_sync --cmake_extra_defines '"CMAKE_CUDA_ARCHITECTURES='${CMAKE_CUDA_ARCHITECTURES}'"'
+ENV CUDA_MODULE_LOADING "LAZY" 
+ARG PARSER_CONFIG=""
+RUN /bin/sh build.sh ${PARSER_CONFIG} --parallel --build_shared_lib --cuda_home /usr/local/cuda --cudnn_home /usr/lib/x86_64-linux-gnu/ --use_tensorrt --tensorrt_home /usr/lib/x86_64-linux-gnu/ --config Release --build_wheel --skip_tests --skip_submodule_sync --cmake_extra_defines '"CMAKE_CUDA_ARCHITECTURES='${CMAKE_CUDA_ARCHITECTURES}'"'
 
 # Switch to root to continue following steps of CI
 USER root