This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
Infra to use tvm write op kernels #15550
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
yzhliu
requested review from
anirudh2290,
eric-haibin-lin and
szha
as code owners
Closed
ZhennanQin
reviewed
Jul 16, 2019
| void TVMBroadcastCompute(const nnvm::NodeAttrs& attrs, | ||
| const mxnet::OpContext& ctx, | ||
| const std::vector<TBlob>& inputs, | ||
| const std::vector<OpReqType>& req, |
There was a problem hiding this comment.
Just an example to show how it works
There was a problem hiding this comment.
I mean does tvm have good support for all kinds of OpReqType, like inplace, addto?
There was a problem hiding this comment.
good question. I don't have a good answer to it as for now.
ZhennanQin
reviewed
Jul 16, 2019
ZhennanQin
reviewed
Jul 16, 2019
ZhennanQin
reviewed
Jul 16, 2019
ZhennanQin
reviewed
Jul 16, 2019
| @@ -734,3 +734,8 @@ def write_all_str(module_file, module_all_list): | |||
|
|
|||
| ctypes.pythonapi.PyCapsule_New.restype = ctypes.py_object | |||
| ctypes.pythonapi.PyCapsule_GetPointer.restype = ctypes.c_void_p | |||
|
|
|||
| from .runtime import Features | |||
| if Features().is_enabled("TVM_OP"): | |||
There was a problem hiding this comment.
Seems TVM only supports python because of this code. Shall we load libtvmop automatically along with loading libmxnet? Then we don't do extra change with front-end language.
There was a problem hiding this comment.
It should be doable. while might need non-trivial surgery to both. cc @tqchen if you have insight.
marcoabreu
reviewed
Jul 16, 2019
marcoabreu
previously requested changes
Jul 16, 2019
|
@marcoabreu cmake support added |
tqchen
approved these changes
Jul 18, 2019
tqchen
reviewed
Jul 18, 2019
larroy
reviewed
Jul 18, 2019
larroy
reviewed
Jul 18, 2019
src/operator/tvmop/op_module.cc
Outdated
| *module_ptr_ = module; | ||
| } | ||
|
|
||
| inline PackedFunc GetFunction(const std::shared_ptr<Module> &module, |
|
Please review again @tqchen @marcoabreu @ZhennanQin @larroy |
marcoabreu
approved these changes
Jul 18, 2019
ZhennanQin
reviewed
Jul 18, 2019
contrib/tvmop/opdef.py
Outdated
| ---------- | ||
| name : str | ||
| function name | ||
| target : bool |
ZhennanQin
reviewed
Jul 18, 2019
larroy
reviewed
Jul 19, 2019
| @@ -185,6 +189,9 @@ enum : unsigned { | |||
| SIGNAL_HANDLER, | |||
| DEBUG, | |||
|
|
|||
| // TVM operator | |||
| TVM_OP, | |||
larroy
reviewed
Jul 19, 2019
larroy
reviewed
Jul 19, 2019
ZhennanQin
approved these changes
Jul 19, 2019
apeforest
reviewed
Jul 19, 2019
apeforest
reviewed
Jul 19, 2019
|
@mxnet-label-bot add [Backend, Operator, pr-awaiting-response] |
marcoabreu
added
Backend
szha
reviewed
Jul 19, 2019
mikemwx
reviewed
Aug 5, 2019
| .set_attr<mxnet::FCompute>("FCompute<cpu>", mxnet::op::TVMBroadcastCompute<func_vadd_cpu>) | ||
| #if MXNET_USE_CUDA | ||
| .set_attr<mxnet::FCompute>("FCompute<gpu>", mxnet::op::TVMBroadcastCompute<func_vadd_gpu>); | ||
| #endif // MXNET_USE_CUDA |
There was a problem hiding this comment.
There will be a missing semicolon if MXNET_USE_CUDA is off.
There was a problem hiding this comment.
may change to this
#if MXNET_USE_CUDA
.set_attr<mxnet::FCompute>("FCompute<gpu>", mxnet::op::TVMBroadcastCompute<func_vadd_gpu>)
#endif // MXNET_USE_CUDA
.set_attr<mxnet::FCompute>("FCompute<cpu>", mxnet::op::TVMBroadcastCompute<func_vadd_cpu>);
anirudhacharya
pushed a commit
to anirudhacharya/mxnet
that referenced
this pull request
Aug 20, 2019
* intra to use tvm write op kernels * add cmake support for tvm op * fix header lint * cleanup USE_TVM_OP logic in Makefile * add doc, cmake def, etc. * fix doc * test rand shape * add with_seed to test * improve err msg. add #if
access2rohit
pushed a commit
to access2rohit/incubator-mxnet
that referenced
this pull request
Sep 25, 2019
* intra to use tvm write op kernels * add cmake support for tvm op * fix header lint * cleanup USE_TVM_OP logic in Makefile * add doc, cmake def, etc. * fix doc * test rand shape * add with_seed to test * improve err msg. add #if
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This PR implements an infra to let users write op kernels in pure Python. One broadcast operator example can be found in,
More background and discussion can be found here: #15465