Integrate rocBLAS fp16 api into Caffe2 (#14882)
authorrohithkrn <rohith.nallamaddi@gmail.com>
Tue, 11 Dec 2018 01:25:46 +0000 (17:25 -0800)
committerFacebook Github Bot <facebook-github-bot@users.noreply.github.com>
Tue, 11 Dec 2018 01:54:06 +0000 (17:54 -0800)
commit7e2b074219fad6d2b09b379423e83b2295b29df2
tree0f47bb2e53291946e2ce96397d21efe7498faa69
parent92f3616f3695ca0ec79e4d583b086cefbcef8aed
Integrate rocBLAS fp16 api into Caffe2 (#14882)

Summary:
This PR integrates rocBLAS half and mixed precision APIs in to Caffe2.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/14882

Differential Revision: D13407840

Pulled By: bddppq

fbshipit-source-id: 75cb0d74da066776fa66575f1d255e879d36121e
caffe2/core/common_gpu.h
caffe2/operators/fully_connected_op_gpu.cc
caffe2/python/operator_test/fc_operator_test.py
caffe2/python/operator_test/matmul_op_test.py
caffe2/python/operator_test/momentum_sgd_test.py
caffe2/sgd/fp16_momentum_sgd_op.cu
caffe2/utils/math_gpu.cu
cmake/Dependencies.cmake
tools/amd_build/pyHIPIFY/cuda_to_hip_mappings.py