[hexagon][testing] add TVMScript elemwise-add #11490
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
|
Here's an example of the tabular output now produced: |
|
Note: This PR starts to factor some benchmarking logic out of the individual benchmark functions, but it's very much a WIP. I'd like to safe a more thorough refactoring for if/when it's truly necessary. |
|
FYI I just force-pushed a fix, since probably nobody had a chance to review the PR yet. |
csullivan
reviewed
May 27, 2022
| # ----------------------------------------------------------------------------------------------- | ||
|
|
||
| # Hexagon v69 allows more dtypes, but we're sticking with v68 for now. | ||
| for dtype in [ |
There was a problem hiding this comment.
nit:
These loops can be hoisted out into tvm.testing.parameters, e.g.
class TestEWBenchmarks:
dtype = tvm.testing.parameter("int8")
num_vectors_per_tensor = tvm.testing.parameters([1, 16, 64, 512])
...
@tvm.testing.requires_hexagon
def test_elemwise_add_tvmcript(hexagon_session : HexagonLauncherRPC, dtype, num_vectors_per_tensor, ...):
In doing so, each testing option becomes a separate pytest. The benchmarking utilities could be defined in the class scope to capture the state across multiple tests.
There was a problem hiding this comment.
It's a good idea. If possible I'd like to tackle that in a future PR, once we have more benchmark configurations written.
cconvey
commented
May 27, 2022
cconvey
commented
May 27, 2022
cconvey
commented
May 27, 2022
|
looks like there are multiple benchmark files. I suggest to move them under |
cconvey
changed the title
|
dude, it's called TVMScript :-) |
|
UPDATE: Doesn't seem so. |
cconvey
changed the title
|
@csullivan : Just a heads-up, I just pushed a major reworking of the PR. |
|
Sample output from the current revision of the PR: |
cconvey
commented
Jun 1, 2022
cconvey
commented
Jun 1, 2022
Comment on lines
+36
to
+38
| _HEXAGON_TARGET = tvm.target.hexagon("v69", link_params=True) | ||
|
|
||
| _SUPER_TARGET = tvm.target.Target(_HEXAGON_TARGET, host=_HEXAGON_TARGET) |
There was a problem hiding this comment.
I'm looking for terminology suggestions here.
cconvey
commented
Jun 2, 2022
Replace TE-based elementwise-add benchmark with a TVMScript-based one. Update Hexagon target architecture from v68 to v69. As a result, the benchmark now requires a version of Hexagon SDK newer than 4.4.0.1. Version 4.5.0.3 is known to work.
cconvey
changed the title
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.
Recreate the existing TE-based elementwise-add benchmark
in TVMScript.
The TE-based benchmark has several features not (yet)
present in the new TVMScript one:
mem_scope='global.vtcm'Small refactoring of how these benchmarks use the
BenchmarkTableclass.cc @mehrdadh