[performance] FeatGraph TVM kernels support #2136
Closed
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
* change edge_ids behavior and C++ impl * fix unittests; remove utils.Index in edge_id * pass mx and th tests * pass tf test * add aten::Scatter_ * Add nonzero; impl CSRGetDataAndIndices/CSRSliceMatrix * CSRGetData and CSRGetDataAndIndices passed tests * CSRSliceMatrix basic tests * fix bug in empty slice * CUDA CSRHasDuplicate * has_node; has_edge_between * predecessors, successors * deprecate send/recv; fix send_and_recv * deprecate send/recv; fix send_and_recv * in_edges; out_edges; all_edges; apply_edges * in deg/out deg * subgraph/edge_subgraph * adj * in_subgraph/out_subgraph * sample neighbors * set/get_n/e_repr * wip: working on refactoring all idtypes * pass ndata/edata tests on gpu * fix * stash * workaround nonzero issue * stash * nx conversion * test_hetero_basics except update routines * test_update_routines * test_hetero_basics for pytorch * more fixes * WIP: flatten graph * wip: flatten * test_flatten * test_to_device * fix bug in to_homo * fix bug in CSRSliceMatrix * pass subgraph test * fix send_and_recv * fix filter * test_heterograph * passed all pytorch tests * fix mx unittest * fix pytorch test_nn * fix all unittests for PyTorch * passed all mxnet tests * lint * fix tf nn test * pass all tf tests * lint * lint * change deprecation * try fix compile * lint * update METIDS * fix utest * fix * fix utests * try debug * revert * small fix * fix utests
* upd * upd * upd * fix * upd * upd * upd * upd * upd * trigger * +1s
* mutation add_nodes and add_edges * Add support for remove_edges, remove_nodes, add_selfloop, remove_selfloop * Fix Co-authored-by: Ubuntu <[email protected]>
* add nodesy * All three * Fix * lint * Add some test case * Fix * Fix * Fix * Fix * Fix * Fix * fix * triger * Fix * fix Co-authored-by: Ubuntu <[email protected]>
added 4 commits
August 27, 2020 07:14
update sparse, add cache for partitioned graphs, pass tests
yzh119
requested review from
yzh119 and
jermainewang
and removed request for
yzh119
yzh119
requested changes
Sep 1, 2020
There was a problem hiding this comment.
Congratulations, and thanks for your contribution to this PR.
I've left some comments, please address these concerns.
I suggest preserving old kernel code instead of deleting them. When users want to use tvm kernels, the gspmm call would be dispatched to gspmm_featgraph, likewise for gsddmm.
| # pass edge_mapping to tvm only when array packing will be used | ||
| use_idx = edge_shuffled and num_feat_partitions > 1 and not use_bcast and use_e | ||
| f_input = [indptr, indices] | ||
| key = (num_rows, num_cols, nnz, op, reduce_op, u_shp, e_shp, use_idx, \ |
There was a problem hiding this comment.
Used for indexing, should find a more elegant way.
python/dgl/sparse.py
Outdated
| # num_row_partitions, num_col_partitions = 2, 2 | ||
| # num_feat_partitions = 2 | ||
| if target == 'cuda' and (num_row_partitions > 1 or num_col_partitions > 1 or num_feat_partitions > 1): | ||
| print('Partitioning not supported on GPU') |
| else: | ||
| raise NotImplementedError | ||
| # check if used for sampling | ||
| generic_shape = nnz == 0 and num_rows == 0 and num_cols == 0 |
There was a problem hiding this comment.
If user would like to deal with generic shape, nnz, num_rows and num_cols need to be set to all zeros?
There was a problem hiding this comment.
Right. We can use another flag, if you prefer.
fix some type bugs. eliminate reshape operator
dimensions of feature tensor have same type minor fixes
yzh119
approved these changes
Sep 11, 2020
There was a problem hiding this comment.
LGTM, @jermainewang I think this PR is ready to be merged.
Let's finish the remaining stuff in future PRs:
- support sampling.
- support fp16.
- support GE-SpMM.
yzh119
changed the title
|
Things todo before merging this PR.
|
Closed
|
After discussion we decide to use AOT solution (see #2367). |
6 tasks
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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 branch merges FeatGraph TVM kernels, and implement necessary functions to be able to use them. When it is completed, it should be able to compile optimal kernels, perform partitioning if necessary, given a specific workload just-in-time though TVM.
Checklist