Refactor cuda packaging pipeline (microsoft#22542)

<!-- Describe your changes. -->

<!-- - Why is this change required? What problem does it solve?
- If it fixes an open issue, please link to the issue here. -->

tools/ci_build/github/azure-pipelines/py-cuda-alt-packaging-pipeline.yml

@@ -1,10 +1,5 @@
 trigger: none
-resources:
-  repositories:
-  - repository: 1esPipelines
-    type: git
-    name: 1ESPipelineTemplates/1ESPipelineTemplates
-    ref: refs/tags/release
+
 parameters:
   - name: enable_linux_cuda
     type: boolean
@@ -22,29 +17,11 @@ parameters:
       - Release
       - RelWithDebInfo
       - MinSizeRel
-extends:
-  # The pipeline extends the 1ES PT which will inject different SDL and compliance tasks.
-  # For non-production pipelines, use "Unofficial" as defined below.
-  # For productions pipelines, use "Official".
-  template: v1/1ES.Official.PipelineTemplate.yml@1esPipelines
-  parameters:
-    sdl:
-      tsa:
-        enabled: true
-      codeSignValidation:
-        enabled: true
-        break: true
-      policheck:
-        enabled: true
-        exclusionsFile: '$(Build.SourcesDirectory)\tools\ci_build\policheck_exclusions.xml'
-    pool:
-      name: 'onnxruntime-Win-CPU-2022'  # Name of your hosted pool
-      os: windows  # OS of the image. This value cannot be a variable. Allowed values: windows, linux, macOS
 
-    stages:
-      - template: stages/py-gpu-packaging-stage.yml
-        parameters:
-          enable_linux_cuda: ${{ parameters.enable_linux_cuda }}
-          enable_windows_cuda: ${{ parameters.enable_windows_cuda }}
-          cmake_build_type: ${{ parameters.cmake_build_type }}
-          cuda_version: '11.8'
+stages:
+  - template: stages/py-gpu-packaging-stage.yml
+    parameters:
+      enable_linux_cuda: ${{ parameters.enable_linux_cuda }}
+      enable_windows_cuda: ${{ parameters.enable_windows_cuda }}
+      cmake_build_type: ${{ parameters.cmake_build_type }}
+      cuda_version: '11.8'

tools/ci_build/github/azure-pipelines/py-cuda-packaging-pipeline.yml

@@ -24,22 +24,13 @@ parameters:
       - RelWithDebInfo
       - MinSizeRel
 
-
 extends:
   # The pipeline extends the 1ES PT which will inject different SDL and compliance tasks.
   # For non-production pipelines, use "Unofficial" as defined below.
   # For productions pipelines, use "Official".
   template: v1/1ES.Official.PipelineTemplate.yml@1esPipelines
   parameters:
-    sdl:
-      tsa:
-        enabled: true
-      codeSignValidation:
-        enabled: true
-        break: true
-      policheck:
-        enabled: true
-        exclusionsFile: '$(Build.SourcesDirectory)\tools\ci_build\policheck_exclusions.xml'
+    # Update the pool with your team's 1ES hosted pool.
     pool:
       name: 'onnxruntime-Win-CPU-2022'  # Name of your hosted pool
       os: windows  # OS of the image. This value cannot be a variable. Allowed values: windows, linux, macOS

tools/ci_build/github/azure-pipelines/py-dml-packaging-pipeline.yml

@@ -1,10 +1,5 @@
 trigger: none
-resources:
-  repositories:
-  - repository: 1esPipelines
-    type: git
-    name: 1ESPipelineTemplates/1ESPipelineTemplates
-    ref: refs/tags/release
+
 parameters:
   - name: cmake_build_type
     type: string
@@ -14,28 +9,10 @@ parameters:
       - Release
       - RelWithDebInfo
       - MinSizeRel
-extends:
-  # The pipeline extends the 1ES PT which will inject different SDL and compliance tasks.
-  # For non-production pipelines, use "Unofficial" as defined below.
-  # For productions pipelines, use "Official".
-  template: v1/1ES.Official.PipelineTemplate.yml@1esPipelines
-  parameters:
-    sdl:
-      tsa:
-        enabled: true
-      codeSignValidation:
-        enabled: true
-        break: true
-      policheck:
-        enabled: true
-        exclusionsFile: '$(Build.SourcesDirectory)\tools\ci_build\policheck_exclusions.xml'
-
-    pool:
-      name: 'onnxruntime-Win-CPU-2022'  # Name of your hosted pool
-      os: windows  # OS of the image. This value cannot be a variable. Allowed values: windows, linux, macOS
 
-    stages:
-      - template: stages/py-gpu-packaging-stage.yml
-        parameters:
-          enable_windows_dml: true
-          cmake_build_type: ${{ parameters.cmake_build_type }}
+stages:
+  - template: stages/py-gpu-packaging-stage.yml
+    parameters:
+      enable_windows_dml: true
+      cmake_build_type: ${{ parameters.cmake_build_type }}
+      publish_symbols: true

tools/ci_build/github/azure-pipelines/py-packaging-pipeline.yml

@@ -77,3 +77,5 @@ stages:
     build_py_parameters: ${{ parameters.build_py_parameters }}
     cmake_build_type: ${{ parameters.cmake_build_type }}
     qnn_sdk_version: ${{ parameters.qnn_sdk_version }}
+    publish_symbols: true
+

tools/ci_build/github/azure-pipelines/stages/py-gpu-packaging-stage.yml

@@ -48,6 +48,10 @@ parameters:
     - '3.12'
     - '3.13'
 
+- name: publish_symbols
+  type: boolean
+  default: false
+
 stages:
   - ${{ if eq(parameters.enable_windows_cuda, true) }}:
     - ${{ each python_version in parameters.PythonVersions }}:
@@ -56,8 +60,10 @@ stages:
           PYTHON_VERSION: ${{ python_version }}
           EP_NAME: gpu
           CudaVersion: ${{ parameters.cuda_version }}
-          EP_BUILD_FLAGS: --use_dml --enable_lto --cuda_home=$(Agent.TempDirectory)\v${{ parameters.cuda_version }} --cmake_extra_defines "CMAKE_CUDA_ARCHITECTURES=52;60;61;70;75;80"
-          use_tensorrt: True
+          ${{ if eq(parameters.cuda_version, '11.8') }}:
+            EP_BUILD_FLAGS: --enable_lto --use_tensorrt --tensorrt_home=$(Agent.TempDirectory)\TensorRT-10.4.0.26.Windows10.x86_64.cuda-11.8 --cuda_home=$(Agent.TempDirectory)\v11.8  --cmake_extra_defines "CMAKE_CUDA_ARCHITECTURES=52;60;61;70;75;80"
+          ${{ if eq(parameters.cuda_version, '12.2') }}:
+            EP_BUILD_FLAGS: --enable_lto --use_tensorrt --tensorrt_home=$(Agent.TempDirectory)\TensorRT-10.4.0.26.Windows10.x86_64.cuda-12.6 --cuda_home=$(Agent.TempDirectory)\v12.2  --cmake_extra_defines "CMAKE_CUDA_ARCHITECTURES=52;60;61;70;75;80"
 
   - ${{ if eq(parameters.enable_linux_cuda, true) }}:
       - template: py-linux-gpu-stage.yml
@@ -70,7 +76,8 @@ stages:
           ${{ if eq(parameters.cuda_version, '11.8') }}:
             docker_base_image: onnxruntimebuildcache.azurecr.io/internal/azureml/onnxruntime/build/cuda11_x64_almalinux8_gcc11:20241120.3
           ${{ if eq(parameters.cuda_version, '12.2') }}:
-            docker_base_image: onnxruntimebuildcache.azurecr.io/internal/azureml/onnxruntime/build/cuda12_x64_ubi8_gcc12:20241120.3
+            docker_base_image: onnxruntimebuildcache.azurecr.io/internal/azureml/onnxruntime/build/cuda12_x64_ubi8_gcc12:20241020.1
+            trt_version: 10.4.0.26-1.cuda12.6
 
   - ${{ if eq(parameters.enable_windows_dml, true) }}:
     - ${{ each python_version in parameters.PythonVersions }}:
@@ -79,5 +86,7 @@ stages:
           MACHINE_POOL: 'onnxruntime-Win2022-GPU-dml-A10'
           PYTHON_VERSION: ${{ python_version }}
           EP_BUILD_FLAGS: --use_dml --cmake_extra_defines CMAKE_SYSTEM_VERSION=10.0.18362.0 --enable_wcos
+          ENV_SETUP_SCRIPT: setup_env.bat
           EP_NAME: directml
+          publish_symbols: ${{ parameters.publish_symbols }}
           cmake_build_type: ${{ parameters.cmake_build_type }}

tools/ci_build/github/azure-pipelines/stages/py-linux-gpu-stage.yml

@@ -99,7 +99,6 @@ stages:
           mv $(Build.BinariesDirectory)/dist ./dist
           pushd dist
           find . -name \*.whl -exec unzip -qq -o {} \;
-          rm -r onnxruntime
           popd
           pushd ${{ parameters.cmake_build_type }} 
           find . -name \*.whl -exec unzip -qq -o {} \;