Skip to content

[QNN] Implement quantized avg_pool2d#15057

Merged
masahi merged 4 commits intoapache:mainfrom
kparzysz-quic:qnn.avgpool
Jun 15, 2023
Merged

[QNN] Implement quantized avg_pool2d#15057
masahi merged 4 commits intoapache:mainfrom
kparzysz-quic:qnn.avgpool

Conversation

@kparzysz-quic
Copy link
Copy Markdown
Contributor

  • qnn.avg_pool2d is integrated into Relay and integration tests are added
  • FQ2I is modified to pick up qnn.avg_pool2d
  • Canonicalization and layout conversion for qnn.avg_pool2d are implemented
  • Dynamic PoolArea computation is implemented for both qnn and slice_op avg_pool2d to exclude padding area in averaging calculation when count_include_pad=False

@tvm-bot
Copy link
Copy Markdown
Collaborator

tvm-bot commented Jun 7, 2023

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

Generated by tvm-bot

3 similar comments
@tvm-bot
Copy link
Copy Markdown
Collaborator

tvm-bot commented Jun 7, 2023

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

Generated by tvm-bot

@tvm-bot
Copy link
Copy Markdown
Collaborator

tvm-bot commented Jun 7, 2023

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

Generated by tvm-bot

@tvm-bot
Copy link
Copy Markdown
Collaborator

tvm-bot commented Jun 7, 2023

Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.

Generated by tvm-bot

@kparzysz-quic kparzysz-quic force-pushed the qnn.avgpool branch 4 times, most recently from 9495547 to 894d3ab Compare June 7, 2023 19:48
@masahi masahi self-assigned this Jun 8, 2023
Comment thread tests/python/contrib/test_hexagon/test_qnn_op_integration.py Outdated
Comment thread src/relay/qnn/op/avg_pool2d.cc Outdated
Comment thread src/relay/qnn/op/avg_pool2d.cc Outdated
Comment thread src/relay/qnn/op/avg_pool2d.cc Outdated
Comment thread src/relay/qnn/op/avg_pool2d.cc Outdated
Comment thread src/relay/qnn/op/avg_pool2d.cc Outdated
Comment thread src/relay/op/nn/pooling_common.h Outdated
Comment thread src/relay/qnn/op/avg_pool2d.cc
Comment thread python/tvm/topi/hexagon/slice_ops/avg_pool2d.py Outdated
Comment thread python/tvm/topi/hexagon/utils.py
@kparzysz-quic
Copy link
Copy Markdown
Contributor Author

Thanks Masa! I'm planning to address your comments this weekend.

fhossein-quic and others added 3 commits June 14, 2023 08:39
* qnn.avg_pool2d is integrated into Relay and integration tests are added
* FQ2I is modified to pick up qnn.avg_pool2d
* Canonicalization and layout conversion for qnn.avg_pool2d are implemented
* Dynamic PoolArea computation is implemented for both qnn and slice_op avg_pool2d
  to exclude padding area in averaging calculation when count_include_pad=False

Co-authored-by: Fateme Hosseini <quic_fhossein@quicinc.com>
Co-authored-by: Jyotsna Verma <jverma@quicinc.com>
Co-authored-by: Anirudh Sundar <quic_sanirudh@quicinc.com>
Co-authored-by: Venkat Rasagna Reddy Komatireddy <quic_rasagna@quicinc.com>
Merge test_wo_qnn_canonicalization.py into test_qnn_op_integration.py.
@kparzysz-quic
Copy link
Copy Markdown
Contributor Author

kparzysz-quic commented Jun 14, 2023

Rebased to fix the GPU build error. Edit: also folded the linter fixes into the previous commit.

@masahi masahi merged commit e6c9d69 into apache:main Jun 15, 2023
@kparzysz-quic kparzysz-quic deleted the qnn.avgpool branch July 5, 2023 12:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants