[4a/10] [CMSIS-NN] Calculate CMSIS-NN buffer size with respect to architecture extensions #9338
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
Mousius
requested review from
MarisaKirisame,
ZihengJiang,
anijain2305,
areusch,
comaniac,
jroesch,
junrushao,
jwfromm,
leandron,
manupak,
mbaret,
mbrookhart,
merrymercy,
slyubomirsky,
tqchen,
trevor-m,
vinx13,
wweic,
yzhliu and
zhiics
as code owners
Member
Author
|
CC @ashutosh-arm @manupa-arm |
Mousius
force-pushed
the
cmsisnn-buffer-calculations
branch
from
b468ff7 to
c347020
Compare
Mousius
changed the title
asparkhi
reviewed
Oct 22, 2021
Contributor
asparkhi
left a comment
asparkhi
left a comment
There was a problem hiding this comment.
Thanks Chris! LGTM. Some minor comments in there.
This correctly calculates the buffer sizes for a variety of targets based on the `-mcpu` and `-mattr` flags passed to the `cmsis-nn` code generator. Added for Conv2d, Depthwise Conv2d and Average Pool.
Mousius
force-pushed
the
cmsisnn-buffer-calculations
branch
from
9795a5b to
11a2dad
Compare
Mousius
changed the title
asparkhi
reviewed
Jan 4, 2022
| ASSERT_EQ(depthwise_conv2d_with_channels(kHasDSP, 4, 6), 0); | ||
| ASSERT_EQ(depthwise_conv2d_with_channels(kHasDSP, 8, 7), 0); | ||
| ASSERT_EQ(depthwise_conv2d_with_channels(kHasMVE, 4, 6), 0); | ||
| ASSERT_EQ(depthwise_conv2d_with_channels(kHasMVE, 8, 7), 0); |
Contributor
There was a problem hiding this comment.
I like it that you have included multiples of 4 cases here for channel 😄
Contributor
|
Thanks @Mousius for this support and also taking care of adding the buffer support for all the other layers that needed buffer size calculations. LGTM! |
ylc
pushed a commit
to ylc/tvm
that referenced
this pull request
Jan 7, 2022
apache#9338) This correctly calculates the buffer sizes for a variety of targets based on the `-mcpu` and `-mattr` flags passed to the `cmsis-nn` code generator. Added for Conv2d, Depthwise Conv2d and Average Pool.
ylc
pushed a commit
to ylc/tvm
that referenced
this pull request
Jan 13, 2022
apache#9338) This correctly calculates the buffer sizes for a variety of targets based on the `-mcpu` and `-mattr` flags passed to the `cmsis-nn` code generator. Added for Conv2d, Depthwise Conv2d and Average Pool.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 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.
This correctly calculates the buffer sizes for a variety of targets based on the
-mcpuand-mattrflags passed to thecmsis-nncode generator.Added for Conv2d, Depthwise Conv2d and Average Pool.