[microNPU] Add various options to the cascader #10509
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
jacobbohlin
requested review from
zhiics,
trevor-m,
comaniac,
mbaret,
manupak,
jroesch,
slyubomirsky,
icemelon,
MarisaKirisame,
ZihengJiang,
yzhliu,
vinx13,
mbrookhart,
jwfromm,
anijain2305,
wweic,
junrushao,
tqchen,
areusch and
merrymercy
as code owners
ekalda
reviewed
Apr 21, 2022
There was a problem hiding this comment.
Thanks @jacobbohlin, looks good to me in general, mostly nits and questions :)
| if len(self.output_shape) != len(other.output_shape): | ||
| return False | ||
|
|
||
| return other >= self |
There was a problem hiding this comment.
Is it supposed to be similar to the comparison in __ge__?
There was a problem hiding this comment.
Yes, I thought the easiest way to define strict less than in this case was as the inverse of greater or equal, i.e. __ge__.
| output_cycles = self._get_output_cycles( | ||
| op_type, op_str, ifm_dtype, ofm_dtype, activation | ||
| ) | ||
| output_cycles *= reduce(lambda a, b: a * b, block_shape, 1) |
There was a problem hiding this comment.
Maybe np.prod(block_shape) for simplicity
There was a problem hiding this comment.
That's probably simpler, I will change it.
| break | ||
| if select_proposal_idx != -1: | ||
| # Manually select proposal based on index | ||
| proposal_choice = proposals[select_proposal_idx % len(proposals)] |
There was a problem hiding this comment.
What is the % len(proposals) for?
There was a problem hiding this comment.
It's just to ensure that we always select a proposal. I can add a comment to clarify.
| /*! \brief Flag to disable pareto culling for proposals to allow non pareto-optimal proposals */ | ||
| bool disable_pareto_proposals; | ||
| /*! \brief Whether to consider multi-dimensional striping */ | ||
| bool multi_dimensional_striping; |
There was a problem hiding this comment.
nit: maybe enable_multi_dimensional_striping to match with other similar variables
Comment on lines
58
to
70
| TVM_ATTR_FIELD(dev_force_block_config) | ||
| .describe( | ||
| "Force the block config to a given value; format = " | ||
| "\"[BLK_HEIGHT]x[BLK_WIDTH]x[BLK_DEPTH]\"") | ||
| .set_default(""); |
There was a problem hiding this comment.
Should that dev option also also come with a dev_warning?
Comment on lines
49
to
100
| expected_stripe_configs = { | ||
| cs.StripeConfig([1, 1], [400, 400], [1, 1], [1, 2], [400, 400], [0, 0]), | ||
| cs.StripeConfig([1, 1], [400, 400], [1, 1], [2, 1], [400, 400], [0, 0]), | ||
| cs.StripeConfig([200, 1], [400, 400], [200, 1], [1, 2], [2, 400], [0, 0]), | ||
| cs.StripeConfig([200, 1], [400, 400], [200, 1], [2, 1], [2, 400], [0, 0]), | ||
| cs.StripeConfig([400, 1], [400, 400], [400, 1], [2, 1], [1, 400], [0, 0]), | ||
| cs.StripeConfig([1, 200], [400, 400], [1, 200], [1, 2], [400, 2], [0, 0]), | ||
| cs.StripeConfig([1, 200], [400, 400], [1, 200], [2, 1], [400, 2], [0, 0]), | ||
| cs.StripeConfig([200, 200], [400, 400], [200, 200], [2, 1], [2, 2], [0, 0]), | ||
| cs.StripeConfig([200, 200], [400, 400], [200, 200], [1, 2], [2, 2], [0, 0]), | ||
| cs.StripeConfig([400, 200], [400, 400], [400, 200], [2, 1], [1, 2], [0, 0]), | ||
| cs.StripeConfig([1, 400], [400, 400], [1, 400], [1, 2], [400, 1], [0, 0]), | ||
| cs.StripeConfig([200, 400], [400, 400], [200, 400], [1, 2], [2, 1], [0, 0]), | ||
| cs.StripeConfig([400, 400], [400, 400], [400, 400], [1, 2], [1, 1], [0, 0]), | ||
| } | ||
|
|
||
| output_stripe_configs = _generate_output_stripe_configs( | ||
| part=part_1, stripe_factors=stripe_factors, multi_dimensional=True | ||
| ) | ||
|
|
||
| assert len(output_stripe_configs) == len(expected_stripe_configs) | ||
| assert set(output_stripe_configs) == expected_stripe_configs |
There was a problem hiding this comment.
I think it would be good to have some tests for the various culling disabling options. Is it possible to compare sets of plans, proposals or block configs similarly to how it is done here for the stripe configs? In that case we could have similar regressions tests for non optimal proposals, plans and block configs.
jacobbohlin
force-pushed
the
cascader-add-options
branch
from
4dbc61d to
61a78c7
Compare
jacobbohlin
force-pushed
the
cascader-add-options
branch
2 times, most recently
from
a3e03cd to
eb788dc
Compare
|
Might need to retrigger the CI... |
jacobbohlin
force-pushed
the
cascader-add-options
branch
from
eb788dc to
f518c57
Compare
* Added option to toggle multi-dimensional striping, it is disabled by default because it has a very high computational cost. Single dimension striping shares most of the benefit with greatly reduced cost. * Added multiple developer/debugging options prefixed with 'dev_' Also added these options to tvmc. * Added cascader logging, if enabled it will dump information about the cascader proposals to a 'cascader_log.json' file. Co-authored-by: Matthew Barrett <[email protected]> Change-Id: I2ec59ae0bd84b73b2cc4bc56d39e3831b0aeec27
jacobbohlin
force-pushed
the
cascader-add-options
branch
from
f518c57 to
0e304b8
Compare
Also added enable_striping to plan_generator.h Change-Id: I496b30ed6af6f0730087329cd81a69c5040a5e4d
jacobbohlin
force-pushed
the
cascader-add-options
branch
from
0e304b8 to
4dddb0e
Compare
|
Thanks all! This is merged now |
|
Thank you! |
mehrdadh
pushed a commit
to mehrdadh/tvm
that referenced
this pull request
May 16, 2022
* [microNPU] Added options to Cascader * Added option to toggle multi-dimensional striping, it is disabled by default because it has a very high computational cost. Single dimension striping shares most of the benefit with greatly reduced cost. * Added multiple developer/debugging options prefixed with 'dev_' Also added these options to tvmc. * Added cascader logging, if enabled it will dump information about the cascader proposals to a 'cascader_log.json' file. Co-authored-by: Matthew Barrett <[email protected]> Change-Id: I2ec59ae0bd84b73b2cc4bc56d39e3831b0aeec27 * Updated memory_reduction testcases Also added enable_striping to plan_generator.h Change-Id: I496b30ed6af6f0730087329cd81a69c5040a5e4d Co-authored-by: Matthew Barrett <[email protected]>
shtinsa
pushed a commit
to Deelvin/tvm
that referenced
this pull request
May 17, 2022
* [microNPU] Added options to Cascader * Added option to toggle multi-dimensional striping, it is disabled by default because it has a very high computational cost. Single dimension striping shares most of the benefit with greatly reduced cost. * Added multiple developer/debugging options prefixed with 'dev_' Also added these options to tvmc. * Added cascader logging, if enabled it will dump information about the cascader proposals to a 'cascader_log.json' file. Co-authored-by: Matthew Barrett <[email protected]> Change-Id: I2ec59ae0bd84b73b2cc4bc56d39e3831b0aeec27 * Updated memory_reduction testcases Also added enable_striping to plan_generator.h Change-Id: I496b30ed6af6f0730087329cd81a69c5040a5e4d Co-authored-by: Matthew Barrett <[email protected]>
shingjan
pushed a commit
to shingjan/tvm
that referenced
this pull request
May 17, 2022
* [microNPU] Added options to Cascader * Added option to toggle multi-dimensional striping, it is disabled by default because it has a very high computational cost. Single dimension striping shares most of the benefit with greatly reduced cost. * Added multiple developer/debugging options prefixed with 'dev_' Also added these options to tvmc. * Added cascader logging, if enabled it will dump information about the cascader proposals to a 'cascader_log.json' file. Co-authored-by: Matthew Barrett <[email protected]> Change-Id: I2ec59ae0bd84b73b2cc4bc56d39e3831b0aeec27 * Updated memory_reduction testcases Also added enable_striping to plan_generator.h Change-Id: I496b30ed6af6f0730087329cd81a69c5040a5e4d Co-authored-by: Matthew Barrett <[email protected]>
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.
This PR adds multiple options to the cascader. One option is to toggle multi-dimensional striping and it is disabled by default. The reasoning is that exploring striping in more than 1 dimension has a very high computational cost.
The remainder of the options are aimed towards developers/debugging and are prefixed with dev_. The options are exposed to tvmc. It also introduces the ability to dump information about the cascader proposals to a .json file.
Co-authored-by: Matthew Barrett [email protected]