[microNPU][2a] Add CascaderGraph for cascading analysis #9469
Conversation
This file contains hidden or 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
This was referenced Nov 8, 2021
mbaret
force-pushed
the
ethosu-cascader-2
branch
2 times, most recently
from
3d4b396 to
22447b4
Compare
mbaret
requested review from
areusch,
comaniac,
icemelon,
jroesch,
junrushao,
merrymercy,
tqchen,
yzhliu and
zhiics
as code owners
Contributor
Author
mbaret
force-pushed
the
ethosu-cascader-2
branch
from
22447b4 to
cbec313
Compare
jacobbohlin
reviewed
Dec 20, 2021
jacobbohlin
reviewed
Dec 20, 2021
Contributor
jacobbohlin
left a comment
jacobbohlin
left a comment
There was a problem hiding this comment.
LGTM - Just a question and very minor suggestions.
| public: | ||
| void VisitAttrs(AttrVisitor* v); | ||
|
|
||
| /*! \return The shape of the tensor */ |
Contributor
There was a problem hiding this comment.
What's the purpose of all these functions, as opposed to having public class members?
Contributor
Author
There was a problem hiding this comment.
They reflect the members into Python.
Contributor
There was a problem hiding this comment.
The reason I'm asking is because PerformanceInfo reflects its members by making them public and wondered why it's done differently here. Is it required for immutability since Tensor has a MUTABLE_OBJECT_REF?
Contributor
Author
There was a problem hiding this comment.
Making the members public isn't enough on its own to reflect them into Python, PerformanceInfo still defines VisitAttrs. Tensor having MUTABLE_OBJECT_REF is so that AddProducer/Consumer work as these are non-const methods.
manupak
reviewed
Dec 21, 2021
cmake/modules/contrib/EthosU.cmake
Outdated
| file(GLOB COMPILER_ETHOSU_SRCS | ||
| CONFIGURE_DEPENDS src/relay/backend/contrib/ethosu/* | ||
| CONFIGURE_DEPENDS src/contrib/ethosu/cascader/*) | ||
| CONFIGURE_DEPENDS src/contrib/ethosu/cascader/* |
Contributor
There was a problem hiding this comment.
If we have source files that defines object definitions, we need to include the irrespective of cmake config.
See here : #9630
Contributor
Author
There was a problem hiding this comment.
Understood, good catch. I've added these files to the 'else' condition here.
mbaret
force-pushed
the
ethosu-cascader-2
branch
from
cbec313 to
82558c7
Compare
A CascaderGraph augments a TE graph with additional information needed by the cascading algorithms. This includes defining a strict ordering on the operators as well as including all the Propagators needed to do the affine analysis of cascades. The CascaderGraph consists of two object types, Parts and Tensors. A Part is an augmented operator which includes the Propagators and a Tensor is similar to a TE tensor but stores additional information like compression ratio. Change-Id: I0d147ef649e4187678ff13208d7921e52aa78238
Change-Id: Icb8f0fd19645f0c0b2730e8db0d2d2b41a34e2c1
Change-Id: I0a541c30813a97e570a5c49fa501daf2b563e66a
Change-Id: I7e0089d55c3bc4eb5d773177de14abf98b5f63a0
mbaret
force-pushed
the
ethosu-cascader-2
branch
from
2d6b666 to
324497c
Compare
Change-Id: Idacf97a9672698d18421f17f4f5b9f62563b1509
Contributor
Author
ylc
pushed a commit
to ylc/tvm
that referenced
this pull request
Jan 7, 2022
A CascaderGraph augments a TE graph with additional information needed by the cascading algorithms. This includes defining a strict ordering on the operators as well as including all the Propagators needed to do the affine analysis of cascades. The CascaderGraph consists of two object types, Parts and Tensors. A Part is an augmented operator which includes the Propagators and a Tensor is similar to a TE tensor but stores additional information like compression ratio.
ylc
pushed a commit
to ylc/tvm
that referenced
this pull request
Jan 13, 2022
A CascaderGraph augments a TE graph with additional information needed by the cascading algorithms. This includes defining a strict ordering on the operators as well as including all the Propagators needed to do the affine analysis of cascades. The CascaderGraph consists of two object types, Parts and Tensors. A Part is an augmented operator which includes the Propagators and a Tensor is similar to a TE tensor but stores additional information like compression ratio.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
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.
RFC: apache/tvm-rfcs#37
Issue: #9429
A CascaderGraph augments a TE graph with additional information needed by the cascading algorithms. This includes defining a strict ordering on the operators as well as including all the Propagators needed to do the affine analysis of cascades.
The CascaderGraph consists of two object types, Parts and Tensors. A Part is an augmented operator which includes the Propagators and a Tensor is similar to a TE tensor but stores additional information like compression ratio.