Intel® Optimizations for TensorFlow* 1.15.0

This release of Intel® Optimized TensorFlow is based on the TensorFlow v1.15.0 tag (https://github.com/tensorflow/tensorflow/tree/v1.15.0) as built with support for Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN). For features and fixes that were introduced in TensorFlow 1.15.0, please see the TensorFlow 1.15.0 release notes. This build contains the following features and fixes:

New functionality and usability improvements:

• The last 1.x release for TensorFlow

• Support context-based graph rewrite to improve performance of tf.data operations

• Support for new Tensorflow operators FusedBatchNormV3 and Addv2

• Support for Intel® MKL-DNN version 0.20.6

Bug fixes:

• Remove redundant transpose for Conv3D

• Modification to graph rewrite pass

• Change 4 unit tests caused by node label changes

• Clean up MKL-ML code

• Fix 3D-Dilated Conv build crash issue

• Add a build flag for MKL-DNN v1.0

• Fix some variables initialization and Clang format

• Fix quantized Matmul unit test

• Fix UT failure of 'NodeRewrite_BiasAddGrad_Positive1'

• Fix clang compilation errors in TF-MKL variant

• Fix 4 autograph unit test failures

• Fix the unit test "print_selective_registration_header_test"

• Fix static scan issue in mkl_layout_pass.cc

• Fix build failure for MKL-DNN v1.x

• Fix unit test //tensorflow/core:graph_mkl_layout_pass

• Updated the test analyzer_cli_test.py to be compatible with recent changes in the graph rewrite logic for MKL-DNN support

• Fix Conv3D output tensor shape when the tensor is empty

• Replace redundant attribute function with a generic function in MKL layout pass

• Fix "Missing 1-th output from" crash caused by MKL MaxPooling

• Enabled Conv2D fprop for MKL-DNN v1.0

• To set default OMP_NUM_THREADS to a smaller value for unit tests

• Enable context-based graph rewrite

• Reverting changes made to Add + Conv fusion during MKL-DNN v1.x integration

• Fix for MKL Slice on block data format

• Fix the issue which fails to create memory descriptor in MKL Concat

• Enable 3d shapes in the MklSlice op

• Fix Batchmatmul regression

• Upgrade giflib to fix CVE-2019-15133

Additional security and performance patches:

• tensorflow#33064: Upgrade curl to fix CVE-2019-5481 and CVE-2019-5482

• tensorflow#31916: Upgrade the checkerframework component to the version 2.10.0

• tensorflow#33523: Upgrade sqlite to 3.30.1 to fix CVE-2019-16168

• a7230a7: Upgrade kafka to 1.2.2 to fix the CVE-2019-17543

• f55fe27: Fixes variable initialization issues discovered by static code scan

Intel® Optimizations for TensorFlow* 2.0.0

This release of Intel® Optimized TensorFlow is based on the TensorFlow v2.0.0 tag (https://github.com/tensorflow/tensorflow/tree/v2.0.0) as built with support for Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN). For features and fixes that were introduced in TensorFlow 2.0.0, please see the TensorFlow 2.0.0 release notes. This build contains the following features and fixes:

New functionality and usability improvements:

• Improved performance of eager (pure) execution

• Support context-based graph rewrite to improve performance of tf.data operations

• Support for new Tensorflow operators FusedBatchNormV3 and Addv2

• Support for Intel® MKL-DNN version 0.20.3

Bug fixes:

• Remove redundant transpose for Conv3D

• Modification to graph rewrite pass

• Change 4 unit tests caused by node label changes

• Clean up MKL-ML code

• Fix 3D-Dilated Conv build crash issue

• Add a build flag for MKL-DNN v1.0

• Fix some variables initialization and Clang format

• Fix quantized Matmul unit test

• Fix UT failure of 'NodeRewrite_BiasAddGrad_Positive1'

• Fix clang compilation errors in TF-MKL variant

• Fix 4 autograph unit test failures

• Fix the unit test "print_selective_registration_header_test"

• Fix static scan issue in mkl_layout_pass.cc

• Fix build failure for MKL-DNN v1.x

• Fix unit test //tensorflow/core:graph_mkl_layout_pass

• Updated the test analyzer_cli_test.py to be compatible with recent changes in the graph rewrite logic for MKL-DNN support

• Fix Conv3D output tensor shape when the tensor is empty

• Replace redundant attribute function with a generic function in MKL layout pass

• Fix "Missing 1-th output from" crash caused by MKL MaxPooling

• To set default OMP_NUM_THREADS to a smaller value for unit tests

• Reverting changes made to Add + Conv fusion during MKL-DNN v1.x integration

• Fix Batchmatmul regression

• Upgrade giflib to fix CVE-2019-15133

• Improve Layer docstrings in regards to autocasting

Additional security and performance patches:

• tensorflow#33064: Upgrade curl to fix CVE-2019-5481 and CVE-2019-5482

• tensorflow#31916: Upgrading the checkerframework component to the version 2.10.0.

• tensorflow#33523: Upgrade sqlite to 3.30.1 to fix CVE-2019-16168

• tensorflow#31469: Fixes variable initialization issues discovered by static code scan and minor formatting issues.

• 07f7b19: Update Google Cloud C++ Client to the v0.14.0 release.

• b7ca5fa: Fixes variable initialization issues discovered by static code scan.

Intel® Optimizations for TensorFlow* 1.14.0 (take 2)

This release of Intel® Optimized TensorFlow is based on the TensorFlow v1.14.0 tag (https://github.com/tensorflow/tensorflow/tree/v1.14.0) as built with support for Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN). For features and fixes that were introduced in TensorFlow 1.14.0, please see the TensorFlow 1.14.0 release notes. This build contains the following features fix:

• Upgrade curl to 7.66.0 for CVE-2019-5481 and CVE-2019-5482

Intel® Optimizations for TensorFlow* 1.13.2

This release of Intel® Optimized TensorFlow is based on the TensorFlow v1.13.2 tag (https://github.com/tensorflow/tensorflow/tree/v1.13.2) as built with support for Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN). For features and fixes that were introduced in TensorFlow 1.13.2, please see the TensorFlow 1.13.2 release notes. This build contains the following fixes in addition to those from upstream:

• Upgrade curl to 7.66.0 for CVE-2019-5481 and CVE-2019-5482
• Upgrade sqlite to 3.29 for CVE-2019-16168
• Fix several issues discovered by static code scans
• Upgrade the checkerframework component to version 2.10.0
• Upgrading giflib to fix CVE-2019-15133

Intel® Optimizations for TensorFlow* 1.14.0

This release of Intel® Optimized TensorFlow is based on the TensorFlow v1.14.0 tag (https://github.com/tensorflow/tensorflow/tree/v1.14.0) as built with support for Intel® Math Kernel Library for Deep Neural Networks (Intel® MKL-DNN). For features and fixes that were introduced in TensorFlow 1.14.0, please see the TensorFlow 1.14.0 release notes. This build contains the following features and fixes:

New functionality and usability improvements:

• Support for 8-bit (quantized) inference on many popular models (see the Model Zoo for Intel® Architecture)
• Initial support for BFloat16 precision operations.
• Optimized FusedBatchNormV2
• Support for Intel® MKL-DNN version 0.18

Bug fixes:

• Fix for a dynamic linker error when using MKL on MacOS
• Fix for windows build with MKL
• Minor Features and Bug Fixes:
• Enable concat with MKLDNN.
• Make MKL 'tf.reshape(tensor, shape)' work when 'tensor' has 0 dimension and 'shape' has unknown dimension.
• Parallelize UnsortedSegmentSum on CPU deivce.
• Disable BFloat16 transpose op
• Adding support for quantized type gather_nd op registration
• Disabling MatMul fusions in grappler for MKL backend
• Update mkl addn op for multi-input
• Enable FusedBatchNormV2 and its grad operator via MKL-DNN API
• Enabled 2D MKLDNN supported concat when all inputs are mkl tensors.
• Add support for Dequantize Op in MKL-DNN
• Fix incorrect way to dump optimized graph
• MKL conditional compilation
• Adding support for MKL Quantized Concat

Additionally, this release contains the following security and performance patches:

• tensorflow#31469: Fixes variable initialization issues discovered by static code scan and minor formatting issues.
• tensorflow#31746: Fixes a member variable initialization issue detected by static code scans.
• tensorflow#30288: Fixes variable initialization issues discovered by static code scan and minor formatting issues.
• e229d3c: Added support for MKL version of BatchMatMul to fix a performance issue.
• tensorflow#31675: Upgrading curl to 7.65.3 to fix CVE-2019-5443.
• 6b6771c): Fixes a build issue caused by the curl upgrade to 7.65.3.
• 62cd702: Upgrades checker framework version to 2.10.0 in order to meet security requirements.