[MXNET-425] MLK-DNN Integration

• Type: Improvement
• Status: To Do
• Priority: Minor
• Resolution: Unresolved
• Sprint: Apache Sprint 4, Apache Sprint 5, Apache Sprint 6, Apache Sprint 7, Apache Sprint 8
• Assignee: Alexander
• Reporter: Alexander
• Created: 16/May/18 17:04
• Updated: 01/Aug/18 07:07

Description

Integrate Intel MKL-DNN features into mxnet.

Sub-Tasks

• Add C++ unit tests for act operator: Done
• Add C++ unit tests for deconvolutions operator: In Review
• Add C++ unit tests for fully connected operator: In Progress
• Add C++ unit tests for convolutions operator: In Review
• Add C++ unit tests for lrn operator: In Review
• Add C++ unit tests for batch-norm operator: To Do
• Add C++ unit tests for concat operator: Done
• Add C++ unit tests for softmax operator: In Progress
• Add C++ unit tests for sum operator: Done
• Add C++ unit tests for pooling operator: Done
• Add kAddTo request type to C++ unit test framework: Done
• Test backward pass operations: Done
• Python test cases for MXNET MKLDNN integration: In Progress
• Integrate MKLRNN into RNN operator: To Do
• Fix the Python test failure for Gluon API/layers: To Do
• Fix MKLDNN performance regression: Done
• Unit test for MKLDNNSum helper util: Done
• Create unit test for CreateMKLDNNMem/CommitOutput: Done
• Fallback MKLDNN Activation for Sparse Input: Done
• Fallback MKLDNN LRN for Sparse Input: Done
• Update Benchmark AI Conv Benchmark with MKLDNN build: In Progress
• Fallback when using non-MKLDNN supported operators: In Progress
• Add RNN Benchmarks to Benchmark AI for MXNET MKLDNN: To Do
• Add RNN Benchmarks to Benchmark AI for MXNET GPU: To Do
• Add RNN Benchmarks to Benchmark AI for Tensorflow (Keras) MKLDNN: To Do
• Add RNN Benchmarks to Benchmark AI for Tensorflow (Keras) GPU: To Do
• Add RNN Benchmarks to Benchmark AI for PyTorch MKLDNN: To Do
• Add Conv Benchmarks to Benchmark AI for PyTorch GPU: To Do
• Add Conv Benchmarks to Benchmark AI for Tensorflow (Keras) MKLDNN: To Do
• Add Conv Benchmarks to Benchmark AI for Tensorflow (Keras) GPU: To Do
• Update Python Distribution with MKLDNN: To Do
• Update R Distribution with MKLDNN: To Do
• Update Scala Distribution with MKLDNN: To Do