From 2f1e883c71ae35d569a663ff3711b969674769cc Mon Sep 17 00:00:00 2001
From: Pranav Sharma <emailpranav@gmail.com>
Date: Fri, 22 Feb 2019 19:29:06 -0800
Subject: [PATCH] Don't use mkldnn and tvm for release pkgs. (#511)

* support non-tensor types

* support non-tensor types.

* support non-tensor types.

* fix compilation issues

* fix compilation issues

* Build without mkldnn for release packages. We'll default to MLAS.

* Remove tvm as well

* Add openmp
---
 .../core/providers/cuda/cuda_execution_provider.cc    |  2 +-
 .../azure-pipelines/azure-pipelines-py-packaging.yml  | 11 ++++++-----
 2 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/onnxruntime/core/providers/cuda/cuda_execution_provider.cc b/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
index c3b7aae053..d1599486b3 100644
--- a/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
+++ b/onnxruntime/core/providers/cuda/cuda_execution_provider.cc
@@ -765,7 +765,7 @@ static void RegisterCudaKernels(KernelRegistry& kernel_registry) {
   kernel_registry.Register(BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 7, 9, double, Upsample)>());
   kernel_registry.Register(BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 7, 9, MLFloat16, Upsample)>());
   kernel_registry.Register(BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 7, 9, int32_t, Upsample)>());
-  kernel_registry.Register(BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 7, 9, uint8_t, Upsample)>());  
+  kernel_registry.Register(BuildKernelCreateInfo<ONNX_OPERATOR_VERSIONED_TYPED_KERNEL_CLASS_NAME(kCudaExecutionProvider, kOnnxDomain, 7, 9, uint8_t, Upsample)>());
 }
 
 std::shared_ptr<KernelRegistry> GetCudaKernelRegistry() {
diff --git a/tools/ci_build/github/azure-pipelines/azure-pipelines-py-packaging.yml b/tools/ci_build/github/azure-pipelines/azure-pipelines-py-packaging.yml
index c8d10b5b3f..fa9aaff75d 100644
--- a/tools/ci_build/github/azure-pipelines/azure-pipelines-py-packaging.yml
+++ b/tools/ci_build/github/azure-pipelines/azure-pipelines-py-packaging.yml
@@ -16,7 +16,7 @@ jobs:
       displayName: 'Run build script'
       inputs:
         scriptPath: 'tools/ci_build/github/linux/run_dockerbuild.sh'
-        args: '-c Release -o ubuntu16.04 -d cpu -r $(Build.BinariesDirectory) -p $(python.version) -x "--build_wheel --use_mkldnn --use_tvm --test_data_url $(TestDataUrl) --test_data_checksum $(TestDataChecksum)"'
+        args: '-c Release -o ubuntu16.04 -d cpu -r $(Build.BinariesDirectory) -p $(python.version) -x "--use_openmp --build_wheel --test_data_url $(TestDataUrl) --test_data_checksum $(TestDataChecksum)"'
 
     - task: CopyFiles@2
       displayName: 'Copy Python Wheel to:  $(Build.ArtifactStagingDirectory)'
@@ -49,7 +49,7 @@ jobs:
       displayName: 'Run build script'
       inputs:
         scriptPath: 'tools/ci_build/github/linux/run_dockerbuild.sh'
-        args: '-c Release -o ubuntu16.04 -d gpu -c cuda9.1-cudnn7.1 -r $(Build.BinariesDirectory) -p $(python.version) -x "--build_wheel --use_mkldnn --test_data_url $(TestDataUrl) --test_data_checksum $(TestDataChecksum)"'
+        args: '-c Release -o ubuntu16.04 -d gpu -c cuda9.1-cudnn7.1 -r $(Build.BinariesDirectory) -p $(python.version) -x "--use_openmp --build_wheel --test_data_url $(TestDataUrl) --test_data_checksum $(TestDataChecksum)"'
 
     - task: CopyFiles@2
       displayName: 'Copy Python Wheel to:  $(Build.ArtifactStagingDirectory)'
@@ -89,7 +89,7 @@ jobs:
       displayName: 'Run build script'
       inputs:
         filename: 'build.bat'
-        arguments: ' --build_dir $(buildDirectory) --config Release --build_wheel --use_mkldnn'
+        arguments: ' --build_dir $(buildDirectory) --config Release --use_openmp --build_wheel'
         workingFolder: "$(Build.SourcesDirectory)"
 
     - task: CopyFiles@2
@@ -151,7 +151,8 @@ jobs:
       displayName: 'Run build script'
       inputs:
         filename: 'build.bat'
-        arguments: ' --use_cuda --cuda_home="C:\local\cuda-9.1.85-windows10-x64-0" --cudnn_home="C:\local\cudnn-9.1-windows10-x64-v7.1\cuda" --build_dir $(buildDirectory) --config Release --build_wheel'
+        arguments: ' --use_cuda --cuda_home="C:\local\cuda-9.1.85-windows10-x64-0"
+      --cudnn_home="C:\local\cudnn-9.1-windows10-x64-v7.1\cuda" --build_dir $(buildDirectory) --config Release --use_openmp --build_wheel'
         workingFolder: "$(Build.SourcesDirectory)"
 
     - task: CopyFiles@2
@@ -197,7 +198,7 @@ jobs:
     - script: |
         sudo python -m pip install numpy
         sudo xcode-select --switch /Applications/Xcode_10.app/Contents/Developer
-        ./build.sh --config Release --skip_submodule_sync --parallel --use_mkldnn --build_wheel
+        ./build.sh --config Release --skip_submodule_sync --parallel --use_openmp --build_wheel
       displayName: 'Command Line Script'
 
     - task: CopyFiles@2