Sindhu/bfloat16 support #399
Conversation
- Enabled --var build to use parallel executor integrating weights-on-device and data pipelining - moved ngraph_var files outside the var build
sayantan-nervana
force-pushed
the
master
branch
from
367d3db to
d74ac48
Compare
|
Will your work help with this issue #447 ? |
| REGISTER_NGRAPH_STUB_KERNEL("NGraphApplyMomentum"); | ||
| REGISTER_NGRAPH_STUB_KERNEL( | ||
| "NGraphAssignAdd"); //*input[0] = *input[0] + input[1] | ||
| REGISTER_NGRAPH_STUB_KERNEL( |
There was a problem hiding this comment.
REGISTER_NGRAPH_STUB_KERNEL registers only for bfloat type? since its defn is:
#define REGISTER_NGRAPH_STUB_KERNEL(optype) \
REGISTER_KERNEL_BUILDER( \
Name(optype).Device(DEVICE_CPU).TypeConstraint<bfloat16>("T"), \
NGStubOp);
in that case is this replacement ok?
There was a problem hiding this comment.
changed the macros. the names were not matching the definition.
ngraph_bridge/ngraph_utils.cc
Outdated
| @@ -223,6 +223,9 @@ Status TensorToStream(std::ostream& ostream, const Tensor& tensor) { | |||
| case DT_BOOL: | |||
| TensorDataToStream<bool>(ostream, n_elements, data); | |||
| break; | |||
| case DT_BFLOAT16: | |||
| TensorDataToStream<bool>(ostream, n_elements, data); | |||
There was a problem hiding this comment.
It says <bool> in the template. copy-paste error perhaps.
There was a problem hiding this comment.
Good catch. Not sure what the corresponding data type for bfloat is.
There was a problem hiding this comment.
We can throw an error or return a bad status for now I guess
| // Since nGraph-bridge OPs work on TF DEVICE_CPU we are registering stub | ||
| // bfloat16 | ||
| // kernels here. The expectation is when we register the stub kernels for | ||
| // bfloat16 | ||
| // TF is going to assign DEVICE_CPU to the respective Ops and we will | ||
| // encapsulate them |
There was a problem hiding this comment.
// Since nGraph-bridge OPs work on TF DEVICE_CPU we are registering stub
// bfloat16 kernels here. The expectation is when we register the stub kernels
// for bfloat16 TF is going to assign DEVICE_CPU to the respective Ops and
// we will encapsulate them
test/python/test_bfloat16.py
Outdated
| @@ -0,0 +1,131 @@ | |||
| # ============================================================================== | |||
| # Copyright 2019 Intel Corporation | |||
There was a problem hiding this comment.
Copyright 2019-2020 Intel Corporation
Add support in bridge for TF Graph with Ops that take in bfloat data type inputs.