Create combined GPU tarball and zip file package (#8827)

* Add onnxruntime_providers_shared.dll into gpu nuget package * Modify for test * Temporarily remove for test * Modify for test * Modify for test * Test packging Windows combined GPU * Test packging Windows combined GPU * Test packging Windows combined GPU * Test packging Windows combined GPU * modify for test * modify for test * fix bug * Modify for test * Modify for test * Modify for test * Modify for test * Modify for test * Modify for test * Modify for test * Modify for test * Prepare for PR * Prepare for PR * Code refactor * Rename proper Artifact name * Rename intermediate Artifact names * Revert Artifact Names * Rename Artifact Names * Modify Artifact name * Modify Artifact name * Modify Artifact name * Update Java package * Update Java package * fix bug to change artifact name * Fix bug for the wrong file path * Fix no fetching correct artifact and test * temporarily modify for test * undo the change for test
2026-05-31 23:27:43 +00:00 · 2021-08-25 13:51:18 -07:00 · 2021-08-25 13:51:18 -07:00 · 32ecbf4691
commit 32ecbf4691
parent cee79526fd
10 changed files with 253 additions and 53 deletions
--- a/tools/ci_build/github/azure-pipelines/c-api-noopenmp-packaging-pipelines.yml
+++ b/tools/ci_build/github/azure-pipelines/c-api-noopenmp-packaging-pipelines.yml
@ -60,7 +60,7 @@ jobs:
      parameters:
          arch: 'linux-x64'
          buildConfig: 'Release'
-          artifactName: 'onnxruntime-java-linux-gpu-x64'
+          artifactName: 'onnxruntime-java-linux-x64-cuda'
          version: '$(OnnxRuntimeVersion)'
          libraryName: 'libonnxruntime.so'
          nativeLibraryName: 'libonnxruntime4j_jni.so'
@ -68,8 +68,8 @@ jobs:
    - template: templates/c-api-artifacts-package-and-publish-steps-posix.yml
      parameters:
          buildConfig: 'Release'
-          artifactName: 'onnxruntime-linux-x64-gpu-$(OnnxRuntimeVersion)'
-          artifactNameNoVersionString:  'onnxruntime-linux-x64-gpu'
+          artifactName: 'onnxruntime-linux-x64-cuda-$(OnnxRuntimeVersion)'
+          artifactNameNoVersionString:  'onnxruntime-linux-x64-cuda'
          libraryName: 'libonnxruntime.so.$(OnnxRuntimeVersion)'
          commitId: $(OnnxRuntimeGitCommitHash)

@ -79,6 +79,11 @@ jobs:
    - template: templates/clean-agent-build-directory-step.yml

 - template: linux-gpu-tensorrt-packaging-pipeline.yml
+  parameters:
+      artifactName: 'onnxruntime-linux-x64-tensorrt-$(OnnxRuntimeVersion)'
+      artifactNameNoVersionString: 'onnxruntime-linux-x64-tensorrt'
+      buildJava: true
+      buildJavaOption: '--build_java'

 - template: templates/win-ci.yml
  parameters:
@ -89,7 +94,7 @@ jobs:
    EnvSetupScript: setup_env_gpu.bat
    buildArch: x64
    msbuildPlatform: x64
-    packageName: gpu-x64
+    packageName: x64-cuda
    buildparameter: --use_cuda --cuda_version=11.1 --cuda_home="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1" --cudnn_home="C:\local\cudnn-11.4-windows-x64-v8.2.2.26\cuda" --enable_onnx_tests --enable_wcos --build_java --cmake_extra_defines "CMAKE_CUDA_ARCHITECTURES=37;52;60;61;70;75;80"
    runTests: ${{ parameters.RunOnnxRuntimeTests }}
    buildJava: true
@ -104,7 +109,7 @@ jobs:
    EnvSetupScript: setup_env_gpu.bat
    buildArch: x64
    msbuildPlatform: x64
-    packageName: tensorrt-x64
+    packageName: x64-tensorrt
    buildparameter: --use_tensorrt --tensorrt_home="C:\local\TensorRT-8.0.1.6.Windows10.x86_64.cuda-11.3.cudnn8.2" --cuda_version=11.1 --cuda_home="C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1" --cudnn_home="C:\local\cudnn-11.4-windows-x64-v8.2.2.26\cuda" --enable_onnx_tests --enable_wcos --cmake_extra_defines "CMAKE_CUDA_ARCHITECTURES=37;52;60;61;70;75;80"
    runTests: ${{ parameters.RunOnnxRuntimeTests }}
    buildJava: false 
@ -130,16 +135,23 @@ jobs:
    displayName: 'Download Pipeline Artifact - Win x64'
    inputs:
      buildType: 'current'
-      artifactName: 'drop-onnxruntime-java-win-gpu-x64'
+      artifactName: 'drop-onnxruntime-java-win-x64-cuda'
      targetPath: '$(Build.BinariesDirectory)\java-artifact\onnxruntime-java-win-x64'

  - task: DownloadPipelineArtifact@2
    displayName: 'Download Pipeline Artifact - Linux x64'
    inputs:
      buildType: 'current'
-      artifactName: 'drop-onnxruntime-java-linux-gpu-x64'
+      artifactName: 'drop-onnxruntime-java-linux-x64-cuda'
      targetPath: '$(Build.BinariesDirectory)\java-artifact\onnxruntime-java-linux-x64'

+  - task: DownloadPipelineArtifact@2
+    displayName: 'Download Pipeline Artifact - Linux x64'
+    inputs:
+      buildType: 'current'
+      artifactName: 'drop-onnxruntime-java-linux-x64-tensorrt'
+      targetPath: '$(Build.BinariesDirectory)\java-artifact\onnxruntime-java-linux-x64-tensorrt'
+
  - task: PowerShell@2
    displayName: 'PowerShell Script'
    inputs:
@ -149,8 +161,6 @@ jobs:
      showWarnings: true
      workingDirectory: '$(Build.BinariesDirectory)\java-artifact'

-
-
  - task: CopyFiles@2
    displayName: 'Copy Java Files to Artifact Staging Directory'
    inputs:
@ -252,7 +262,103 @@ jobs:
    displayName: 'Clean Agent Directories'
    condition: always()

+- job: Linux_Packaging_combined_GPU
+  workspace:
+    clean: all
+  pool: 'Onnxruntime-Linux-GPU'
+  dependsOn:
+  - Linux_C_API_Packaging_GPU_x64
+  - Linux_C_API_Packaging_GPU_TensorRT_x64
+  condition: succeeded()
+  steps:
+  - checkout: self
+    submodules: false
+  - template: templates/set-version-number-variables-step.yml
+  - task: DownloadPipelineArtifact@2
+    displayName: 'Download Pipeline Artifact - Combined GPU'
+    inputs:
+      artifactName: 'onnxruntime-linux-x64-cuda'
+      targetPath: '$(Build.BinariesDirectory)/tgz-artifacts'

+  - task: DownloadPipelineArtifact@2
+    displayName: 'Download Pipeline Artifact - Combined GPU'
+    inputs:
+      artifactName: 'onnxruntime-linux-x64-tensorrt'
+      targetPath: '$(Build.BinariesDirectory)/tgz-artifacts'
+
+  - task: ShellScript@2
+    displayName: 'Shell Script'
+    inputs:
+      scriptPath: 'tools/ci_build/github/linux/extract_and_bundle_gpu_package.sh'
+      args: '-a $(Build.BinariesDirectory)/tgz-artifacts'
+      workingDirectory: '$(Build.BinariesDirectory)/tgz-artifacts'
+
+  - task: ArchiveFiles@2
+    inputs:
+      rootFolderOrFile: '$(Build.BinariesDirectory)/tgz-artifacts/onnxruntime-linux-x64-gpu'
+      includeRootFolder: true
+      archiveType: 'tar' # Options: zip, 7z, tar, wim
+      tarCompression: 'gz'
+      archiveFile: '$(Build.ArtifactStagingDirectory)/onnxruntime-linux-x64-gpu-$(OnnxRuntimeVersion).tgz'
+      replaceExistingArchive: true
+
+  - task: PublishPipelineArtifact@1
+    inputs:
+      targetPath: '$(Build.ArtifactStagingDirectory)/onnxruntime-linux-x64-gpu-$(OnnxRuntimeVersion).tgz'
+      artifactName: 'onnxruntime-linux-x64-gpu'
+
+- job: Windows_Packaging_combined_GPU
+  workspace:
+    clean: all
+  pool: 'Win-CPU-2021'
+  dependsOn:
+  - Windows_Packaging_gpu
+  - Windows_Packaging_tensorrt
+  condition: succeeded()
+  steps:
+  - checkout: self
+    submodules: false
+  - task: DownloadPipelineArtifact@2
+    displayName: 'Download Pipeline Artifact - Combined GPU'
+    inputs:
+      artifactName: 'onnxruntime-win-x64-cuda'
+      targetPath: '$(Build.BinariesDirectory)/zip-artifacts'
+
+  - task: DownloadPipelineArtifact@2
+    displayName: 'Download Pipeline Artifact - Combined GPU'
+    inputs:
+      artifactName: 'onnxruntime-win-x64-tensorrt'
+      targetPath: '$(Build.BinariesDirectory)/zip-artifacts'
+
+  - task: PowerShell@2
+    displayName: 'PowerShell Script'
+    inputs:
+      targetType: filePath
+      filePath: $(Build.SourcesDirectory)\tools\ci_build\github\windows\extract_zip_files_gpu.ps1
+
+  - script: |
+     dir
+    workingDirectory: '$(Build.BinariesDirectory)/zip-artifacts'
+    displayName: 'List artifacts'
+
+  - task: BatchScript@1
+    displayName: 'Bundle CUDA/TRT EP binaries'
+    inputs:
+      filename: $(Build.SourcesDirectory)\tools\ci_build\github\windows\bundle_dlls_gpu.bat
+      workingFolder: $(Build.BinariesDirectory)\zip-artifacts
+
+  - task: CopyFiles@2
+    displayName: 'Copy zip file to: $(Build.ArtifactStagingDirectory)'
+    inputs:
+      SourceFolder: '$(Build.BinariesDirectory)\zip-artifacts'
+      Contents: 'onnxruntime-win-x64-gpu-*.zip'
+      TargetFolder: '$(Build.ArtifactStagingDirectory)'
+
+  - task: PublishPipelineArtifact@0
+    displayName: 'Publish Pipeline Combined GPU Package Artifact'
+    inputs:
+      artifactName: 'onnxruntime-win-x64-gpu'
+      targetPath: '$(Build.ArtifactStagingDirectory)'

 - job: NuGet_Packaging_GPU
  workspace:
@ -272,25 +378,25 @@ jobs:
  - task: DownloadPipelineArtifact@2
    displayName: 'Download Pipeline Artifact - NuGet'
    inputs:
-      artifactName: 'onnxruntime-win-gpu-x64'
+      artifactName: 'onnxruntime-win-x64-cuda'
      targetPath: '$(Build.BinariesDirectory)/nuget-artifact'

  - task: DownloadPipelineArtifact@2
    displayName: 'Download Pipeline Artifact - NuGet'
    inputs:
-      artifactName: 'onnxruntime-win-tensorrt-x64'
+      artifactName: 'onnxruntime-win-x64-tensorrt'
      targetPath: '$(Build.BinariesDirectory)/nuget-artifact'

  - task: DownloadPipelineArtifact@2
    displayName: 'Download Pipeline Artifact - NuGet'
    inputs:
-      artifactName: 'onnxruntime-linux-x64-gpu'
+      artifactName: 'onnxruntime-linux-x64-cuda'
      targetPath: '$(Build.BinariesDirectory)/nuget-artifact'

  - task: DownloadPipelineArtifact@2
    displayName: 'Download Pipeline Artifact - NuGet'
    inputs:
-      artifactName: 'onnxruntime-linux-x64-gpu-tensorrt'
+      artifactName: 'onnxruntime-linux-x64-tensorrt'
      targetPath: '$(Build.BinariesDirectory)/nuget-artifact'

  - task: DownloadPipelineArtifact@2
@ -357,7 +463,7 @@ jobs:
  - task: BatchScript@1
    displayName: 'Bundle Native NuGet and other binaries'
    inputs:
-      filename: $(Build.SourcesDirectory)\tools\ci_build\github\windows\bundle_dlls_gpu.bat
+      filename: $(Build.SourcesDirectory)\tools\ci_build\github\windows\bundle_dlls_gpu_nuget.bat
      workingFolder: $(Build.BinariesDirectory)\RelWithDebInfo\RelWithDebInfo\nuget-artifacts

  - task: CopyFiles@2
--- a/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-packaging-pipeline.yml
+++ b/tools/ci_build/github/azure-pipelines/linux-gpu-tensorrt-packaging-pipeline.yml
@ -1,3 +1,20 @@
+parameters:
+- name: artifactName  
+  type: string
+  default: 'onnxruntime-linux-x64-gpu-tensorrt-$(OnnxRuntimeVersion)'
+
+- name: artifactNameNoVersionString
+  type: string
+  default: 'onnxruntime-linux-x64-gpu-tensorrt'
+
+- name: buildJava
+  type: boolean
+  default: false
+
+- name: buildJavaOption
+  type: string
+  default: ''
+
 jobs:

 - job: Linux_C_API_Packaging_GPU_TensorRT_x64
@ -22,14 +39,24 @@ jobs:
          docker run --gpus all -e CC=/opt/rh/devtoolset-10/root/usr/bin/cc -e CXX=/opt/rh/devtoolset-10/root/usr/bin/c++ -e CFLAGS="-Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -fstack-protector-strong -fstack-clash-protection -fcf-protection -O3 -Wl,--strip-all" -e CXXFLAGS="-Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS -fstack-protector-strong -fstack-clash-protection -fcf-protection -O3 -Wl,--strip-all" -e NVIDIA_VISIBLE_DEVICES=all --rm --volume /data/onnx:/data/onnx:ro --volume $(Build.SourcesDirectory):/onnxruntime_src --volume $(Build.BinariesDirectory):/build \
          --volume /data/models:/build/models:ro --volume $HOME/.onnx:/home/onnxruntimedev/.onnx -e NIGHTLY_BUILD onnxruntimecuda114xtrt80build \
          /opt/python/cp37-cp37m/bin/python3 /onnxruntime_src/tools/ci_build/build.py --build_dir /build --config Release \
-          --skip_submodule_sync --parallel --build_shared_lib --use_tensorrt --cuda_version=$(CUDA_VERSION) --cuda_home=/usr/local/cuda-$(CUDA_VERSION) --cudnn_home=/usr --tensorrt_home=/usr --cmake_extra_defines CMAKE_CUDA_HOST_COMPILER=/opt/rh/devtoolset-10/root/usr/bin/cc 'CMAKE_CUDA_ARCHITECTURES=37;50;52;60;61;70;75;80'
+          --skip_submodule_sync --parallel --build_shared_lib ${{ parameters.buildJavaOption }} --use_tensorrt --cuda_version=$(CUDA_VERSION) --cuda_home=/usr/local/cuda-$(CUDA_VERSION) --cudnn_home=/usr --tensorrt_home=/usr --cmake_extra_defines CMAKE_CUDA_HOST_COMPILER=/opt/rh/devtoolset-10/root/usr/bin/cc 'CMAKE_CUDA_ARCHITECTURES=37;50;52;60;61;70;75;80'
        workingDirectory: $(Build.SourcesDirectory)

+    - ${{ if eq(parameters.buildJava, true) }}:
+        - template: templates/java-api-artifacts-package-and-publish-steps-posix.yml
+          parameters:
+              arch: 'linux-x64'
+              buildConfig: 'Release'
+              artifactName: 'onnxruntime-java-linux-x64-tensorrt'
+              version: '$(OnnxRuntimeVersion)'
+              libraryName: 'libonnxruntime.so'
+              nativeLibraryName: 'libonnxruntime4j_jni.so'
+
    - template: templates/c-api-artifacts-package-and-publish-steps-posix.yml
      parameters:
          buildConfig: 'Release'
-          artifactName: 'onnxruntime-linux-x64-gpu-tensorrt-$(OnnxRuntimeVersion)'
-          artifactNameNoVersionString:  'onnxruntime-linux-x64-gpu-tensorrt'
+          artifactName: ${{ parameters.artifactName }} 
+          artifactNameNoVersionString: ${{ parameters.artifactNameNoVersionString }}  
          libraryName: 'libonnxruntime.so.$(OnnxRuntimeVersion)'
          commitId: $(OnnxRuntimeGitCommitHash)

--- a/tools/ci_build/github/azure-pipelines/templates/c-api-artifacts-package-and-publish-steps-windows.yml
+++ b/tools/ci_build/github/azure-pipelines/templates/c-api-artifacts-package-and-publish-steps-windows.yml
@ -48,6 +48,11 @@ steps:
          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime_providers_cuda.pdb $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime_providers_cuda.lib $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib

+          # copy trt ep libraries only when trt ep is enabled
+          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime_providers_tensorrt.dll $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
+          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime_providers_tensorrt.pdb $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
+          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime_providers_tensorrt.lib $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
+
          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime.pdb $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime.lib $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
          copy $(Build.SourcesDirectory)\include\onnxruntime\core\session\onnxruntime_*.h  $(Build.BinariesDirectory)\${{parameters.artifactName}}\include
@ -64,20 +69,6 @@ steps:

        workingDirectory: '$(Build.BinariesDirectory)\${{parameters.buildConfig}}'

-    - task: CmdLine@2
-      displayName: 'Copy TensorRT EP build artifacts for zipping'
-      condition: and(succeeded(), eq('${{ parameters.trtEnabled}}', true))
-      inputs:
-        script: |
-          mkdir $(Build.BinariesDirectory)\${{parameters.artifactName}}
-          mkdir $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
-          mkdir $(Build.BinariesDirectory)\${{parameters.artifactName}}\include
-
-          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime_providers_tensorrt.dll $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
-          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime.dll $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
-          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime.pdb $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
-          copy $(Build.BinariesDirectory)\${{parameters.buildConfig}}\${{parameters.buildConfig}}\onnxruntime.lib $(Build.BinariesDirectory)\${{parameters.artifactName}}\lib
-
    - template: win-esrp-dll.yml
      parameters:
        FolderPath: '$(Build.BinariesDirectory)\${{parameters.artifactName}}'
--- a/tools/ci_build/github/linux/extract_and_bundle_gpu_package.sh
+++ b/tools/ci_build/github/linux/extract_and_bundle_gpu_package.sh
@ -0,0 +1,28 @@
+#!/bin/bash
+set -e -o -x
+
+while getopts a: parameter_Option
+do case "${parameter_Option}"
+in
+a) ARTIFACT_DIR=${OPTARG};;
+esac
+done
+
+EXIT_CODE=1
+
+uname -a
+
+cd $ARTIFACT_DIR 
+
+mkdir -p $ARTIFACT_DIR/onnxruntime-linux-x64-tensorrt
+tar zxvf $ARTIFACT_DIR/onnxruntime-linux-x64-tensorrt-*.tgz -C onnxruntime-linux-x64-tensorrt
+rm $ARTIFACT_DIR/onnxruntime-linux-x64-tensorrt-*.tgz
+
+mkdir -p $ARTIFACT_DIR/onnxruntime-linux-x64-gpu
+TAR_NAME=`ls $ARTIFACT_DIR/onnxruntime-linux-x64-cuda-*.tgz`
+tar zxvf $ARTIFACT_DIR/onnxruntime-linux-x64-cuda-*.tgz -C onnxruntime-linux-x64-gpu
+rm $ARTIFACT_DIR/onnxruntime-linux-x64-cuda-*.tgz
+
+cp onnxruntime-linux-x64-tensorrt/*/lib/libonnxruntime.so* onnxruntime-linux-x64-gpu/*/lib
+cp onnxruntime-linux-x64-tensorrt/*/lib/libonnxruntime_providers_tensorrt.so onnxruntime-linux-x64-gpu/*/lib
+cp onnxruntime-linux-x64-tensorrt/*/lib/libonnxruntime_providers_shared.so onnxruntime-linux-x64-gpu/*/lib
--- a/tools/ci_build/github/linux/java_copy_strip_binary.sh
+++ b/tools/ci_build/github/linux/java_copy_strip_binary.sh
@ -50,6 +50,11 @@ then
        cp $BINARY_DIR/$BUILD_CONFIG/libonnxruntime_providers_shared.so $BINARY_DIR/$ARTIFACT_NAME/$NATIVE_FOLDER/libonnxruntime_providers_shared.so
        cp $BINARY_DIR/$BUILD_CONFIG/libonnxruntime_providers_cuda.so $BINARY_DIR/$ARTIFACT_NAME/$NATIVE_FOLDER/libonnxruntime_providers_cuda.so
    fi
+    # Add tensorrt provider if it exists
+    if [[ -f "$BINARY_DIR/$BUILD_CONFIG/libonnxruntime_providers_tensorrt.so" ]]; then
+        cp $BINARY_DIR/$BUILD_CONFIG/libonnxruntime_providers_shared.so $BINARY_DIR/$ARTIFACT_NAME/$NATIVE_FOLDER/libonnxruntime_providers_shared.so
+        cp $BINARY_DIR/$BUILD_CONFIG/libonnxruntime_providers_tensorrt.so $BINARY_DIR/$ARTIFACT_NAME/$NATIVE_FOLDER/libonnxruntime_providers_tensorrt.so
+    fi
 fi

 find $BINARY_DIR/$ARTIFACT_NAME -ls
--- a/tools/ci_build/github/windows/bundle_dlls_gpu.bat
+++ b/tools/ci_build/github/windows/bundle_dlls_gpu.bat
@ -1,22 +1,21 @@
 REM Copyright (c) Microsoft Corporation. All rights reserved.
 REM Licensed under the MIT License.

-REM for available runtime identifiers, see https://github.com/dotnet/corefx/blob/release/3.1/pkg/Microsoft.NETCore.Platforms/runtime.json
 set PATH=%CD%;%PATH%
-SETLOCAL EnableDelayedExpansion 
-FOR /R %%i IN (*.nupkg) do (
+SETLOCAL EnableDelayedExpansion
+FOR /R %%i IN (*.zip) do (
   set filename=%%~ni
-   IF NOT "!filename:~25,7!"=="Managed" (
-       mkdir runtimes\linux-x64\native
-       move onnxruntime-linux-x64-gpu\lib\libonnxruntime_providers_* runtimes\linux-x64\native
-       move onnxruntime-linux-x64-gpu-tensorrt\lib\libonnxruntime.so.1* runtimes\linux-x64\native\libonnxruntime.so
-       move onnxruntime-linux-x64-gpu-tensorrt\lib\libonnxruntime_providers_shared.so runtimes\linux-x64\native\libonnxruntime_providers_shared.so
-       move onnxruntime-linux-x64-gpu-tensorrt\lib\libonnxruntime_providers_tensorrt.so runtimes\linux-x64\native\libonnxruntime_providers_tensorrt.so
-       mkdir runtimes\win-x64\native
-       move onnxruntime-win-tensorrt-x64\lib\onnxruntime_providers_tensorrt.dll runtimes\win-x64\native\onnxruntime_providers_tensorrt.dll
-       move onnxruntime-win-tensorrt-x64\lib\onnxruntime.dll runtimes\win-x64\native\onnxruntime.dll
-       move onnxruntime-win-tensorrt-x64\lib\onnxruntime.lib runtimes\win-x64\native\onnxruntime.lib
-       move onnxruntime-win-tensorrt-x64\lib\onnxruntime.pdb runtimes\win-x64\native\onnxruntime.pdb
-       7z a  %%~ni.nupkg runtimes
+   IF "!filename:~20,3!"=="gpu" (
+       mkdir !filename!\lib
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_tensorrt.dll !filename!\lib\onnxruntime_providers_tensorrt.dll
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_tensorrt.lib !filename!\lib\onnxruntime_providers_tensorrt.lib
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_tensorrt.pdb !filename!\lib\onnxruntime_providers_tensorrt.pdb
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_shared.dll !filename!\lib\onnxruntime_providers_shared.dll
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_shared.lib !filename!\lib\onnxruntime_providers_shared.lib
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_shared.pdb !filename!\lib\onnxruntime_providers_shared.pdb
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime.dll !filename!\lib\onnxruntime.dll
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime.lib !filename!\lib\onnxruntime.lib
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime.pdb !filename!\lib\onnxruntime.pdb
+       7z a  %%~ni.zip !filename!\lib
   )
-) 
+)
--- a/tools/ci_build/github/windows/bundle_dlls_gpu_nuget.bat
+++ b/tools/ci_build/github/windows/bundle_dlls_gpu_nuget.bat
@ -0,0 +1,23 @@
+REM Copyright (c) Microsoft Corporation. All rights reserved.
+REM Licensed under the MIT License.
+
+REM for available runtime identifiers, see https://github.com/dotnet/corefx/blob/release/3.1/pkg/Microsoft.NETCore.Platforms/runtime.json
+set PATH=%CD%;%PATH%
+SETLOCAL EnableDelayedExpansion 
+FOR /R %%i IN (*.nupkg) do (
+   set filename=%%~ni
+   IF NOT "!filename:~25,7!"=="Managed" (
+       mkdir runtimes\linux-x64\native
+       move onnxruntime-linux-x64-gpu\lib\libonnxruntime_providers_* runtimes\linux-x64\native
+       move onnxruntime-linux-x64-tensorrt\lib\libonnxruntime.so.1* runtimes\linux-x64\native\libonnxruntime.so
+       move onnxruntime-linux-x64-tensorrt\lib\libonnxruntime_providers_shared.so runtimes\linux-x64\native\libonnxruntime_providers_shared.so
+       move onnxruntime-linux-x64-tensorrt\lib\libonnxruntime_providers_tensorrt.so runtimes\linux-x64\native\libonnxruntime_providers_tensorrt.so
+       mkdir runtimes\win-x64\native
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_tensorrt.dll runtimes\win-x64\native\onnxruntime_providers_tensorrt.dll
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime_providers_shared.dll runtimes\win-x64\native\onnxruntime_providers_shared.dll
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime.dll runtimes\win-x64\native\onnxruntime.dll
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime.lib runtimes\win-x64\native\onnxruntime.lib
+       move onnxruntime-win-x64-tensorrt\lib\onnxruntime.pdb runtimes\win-x64\native\onnxruntime.pdb
+       7z a  %%~ni.nupkg runtimes
+   )
+) 
--- a/tools/ci_build/github/windows/extract_nuget_files_gpu.ps1
+++ b/tools/ci_build/github/windows/extract_nuget_files_gpu.ps1
@ -27,10 +27,10 @@ Foreach-Object {

 New-Item -Path $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\external\protobuf\cmake\RelWithDebInfo -ItemType directory

-Copy-Item -Path $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-win-gpu-x64-*\lib\* -Destination $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo
+Copy-Item -Path $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-win-x64-cuda-*\lib\* -Destination $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo
 Copy-Item -Path $Env:BUILD_BINARIESDIRECTORY\extra-artifact\protoc.exe $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\external\protobuf\cmake\RelWithDebInfo


-Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts | Where-Object { $_.Name -match 'onnxruntime-linux-x64-gpu-\d{1,}\.\d{1,}\.\d{1,}$' } | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-linux-x64-gpu
-Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-linux-x64-gpu-tensorrt-* | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-linux-x64-gpu-tensorrt
-Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-win-tensorrt-x64-* | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-win-tensorrt-x64
+Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts | Where-Object { $_.Name -match 'onnxruntime-linux-x64-cuda-\d{1,}\.\d{1,}\.\d{1,}$' } | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-linux-x64-gpu
+Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-linux-x64-tensorrt-* | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-linux-x64-tensorrt
+Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-win-x64-tensorrt-* | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\RelWithDebInfo\RelWithDebInfo\nuget-artifacts\onnxruntime-win-x64-tensorrt
--- a/tools/ci_build/github/windows/extract_zip_files_gpu.ps1
+++ b/tools/ci_build/github/windows/extract_zip_files_gpu.ps1
@ -0,0 +1,20 @@
+# Copyright (c) Microsoft Corporation. All rights reserved.
+# Licensed under the MIT License.
+
+Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\zip-artifacts -Filter *.zip | 
+Foreach-Object {
+ $cmd = "7z.exe x $($_.FullName) -y -o$Env:BUILD_BINARIESDIRECTORY\zip-artifacts"
+ Write-Output $cmd
+ Invoke-Expression -Command $cmd
+}
+
+Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\zip-artifacts -Filter *onnxruntime-win-x64-cuda*.zip |
+Foreach-Object {
+ $($_.FullName) -match '.*onnxruntime-win-x64-cuda-(.*).zip'
+ $version=$matches[1]
+ Rename-Item -Path $($_.FullName) -NewName onnxruntime-win-x64-gpu-$version.zip
+}
+
+Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\zip-artifacts | Where-Object { $_.Name -match 'onnxruntime-win-x64-cuda-\d{1,}\.\d{1,}\.\d{1,}$' } | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\zip-artifacts\onnxruntime-win-x64-gpu
+
+Get-ChildItem $Env:BUILD_BINARIESDIRECTORY\zip-artifacts | Where-Object { $_.Name -match 'onnxruntime-win-x64-tensorrt-\d{1,}\.\d{1,}\.\d{1,}$' } | Rename-Item -NewName $Env:BUILD_BINARIESDIRECTORY\zip-artifacts\onnxruntime-win-x64-tensorrt
--- a/tools/ci_build/github/windows/jar_gpu_packaging.ps1
+++ b/tools/ci_build/github/windows/jar_gpu_packaging.ps1
@ -4,7 +4,8 @@
 $ErrorActionPreference = "Stop"
 Write-Output "Start"
 dir
-pushd onnxruntime-java-linux-x64
+Copy-Item -Path $Env:BUILD_BINARIESDIRECTORY\java-artifact\onnxruntime-java-linux-x64\ai\onnxruntime\native\linux-x64\libonnxruntime_providers_cuda.so -Destination $Env:BUILD_BINARIESDIRECTORY\java-artifact\onnxruntime-java-linux-x64-tensorrt\ai\onnxruntime\native\linux-x64 
+pushd onnxruntime-java-linux-x64-tensorrt
 Write-Output "Run 7z"
 7z a $Env:BUILD_BINARIESDIRECTORY\java-artifact\onnxruntime-java-win-x64\testing.jar libcustom_op_library.so
 Remove-Item -Path libcustom_op_library.so
@ -15,4 +16,4 @@ ren onnxruntime-$Env:ONNXRUNTIMEVERSION.jar onnxruntime_gpu-$Env:ONNXRUNTIMEVERS
 ren onnxruntime-$Env:ONNXRUNTIMEVERSION-javadoc.jar onnxruntime_gpu-$Env:ONNXRUNTIMEVERSION-javadoc.jar
 ren onnxruntime-$Env:ONNXRUNTIMEVERSION-sources.jar onnxruntime_gpu-$Env:ONNXRUNTIMEVERSION-sources.jar
 ren onnxruntime-$Env:ONNXRUNTIMEVERSION.pom onnxruntime_gpu-$Env:ONNXRUNTIMEVERSION.pom
-popd
+popd