[Adreno] Add documentation for Adreno deployment #13393

dsbarinov1 · 2022-11-15T13:23:05Z

Purpose:
assist TVM users compile and deploy on Adreno by expanding our documentation and providing sample scripts in TVM.

Information about PR:
The present PR consists globally of 3 parts.

The first part is an introductory article on compilation and deployment of neural networks on Adreno, covering such topics as: «Build TVM for Android/Adreno», «Advantages of textures» and «Differences in compilation and deployment of models for Adreno devices».

The second part is a straightforward example script for compiling and inferring models at different precisions for Adreno devices.

The third part is auxiliary files, images, etc.

tvm-bot · 2022-11-15T13:23:08Z

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.

cc @echuraev, @elvin-n _{See #10317 for details}
Built docs for commit 98ee8b5 can be found here.

_{Generated by tvm-bot}

echuraev · 2022-11-15T13:51:25Z

@masahi, @driazati, @areusch Could you please help? This PR contains two image files. We saw that other images are stored here. How this PR should be merged? First the images should be added to the tlc-pack/web-data repository and only after that this PR might be merged to the tvm/main?

driazati

For the images we prefer to keep binary files outside of the repo for readability / size reasons, can you instead make a PR to add them to https://github.com/tlc-pack/web-data and tag me in it? Once that's done you can link to them in rst e.g.

just saw your comment @echuraev, we can merge the PR to tlc-pack/web-data first then use that on this PR to make sure it's all good, then merge this one

docs/how_to/deploy/adreno.rst

echuraev · 2022-11-16T06:15:48Z

docs/how_to/deploy/adreno.rst

+
+For the complete step-py-step process of compiling and deploying models on
+Adreno, including selection of precision, running the inference of the
+model, getting the predictions, and measuring the performance please refer to this tutorial: `How To Deploy model on Adreno <https://tvm.apache.org/docs/how_to/deploy_models/deploy_model_on_adreno.html>`_


Probably the link can be relative

I could not find references for creating relative links in other apache documents.
@driazati what do you think about that? Should we somehow make relative links to the other docs?

echuraev · 2022-11-16T06:18:55Z

docs/how_to/deploy/adreno.rst

+* Operations which can be converted if they follow by converted cluster
+* Operations never be converted to the float16 data type  
+This list is defined in the ToMixedPrecision implementation here 
+`relay/transform/mixed_precision.py <https://github.com/apache/tvm/blob/main/python/tvm/relay/transform/mixed_precision.py#L34>`_ 


Probably it would be better to have relative links. Not sure if it will work with specifying line number. But anyway, line number might be not actual after adding something to this file.

Couldn't make a relative link to other site.

docs/how_to/deploy/adreno.rst

driazati · 2022-11-21T19:56:53Z

hey @dsbarinov1 just a note on the CI failures, you'll probably need to rebase your changes on the latest main in order to get CI running properly

Purpose: assist TVM users compile and deploy on Adreno by expanding our documentation and providing sample scripts in TVM. Information about PR: The present PR consists globally of 3 parts. The first part is an introductory article on compilation and deployment of neural networks on Adreno, covering such topics as: «Build TVM for Android/Adreno», «Advantages of textures» and «Differences in compilation and deployment of models for Adreno devices». The second part is a straightforward example script for compiling and inferring models at different precisions for Adreno devices. The third part is auxiliary files, images, etc.

…in docs to work without this extension

echuraev · 2022-11-25T08:14:39Z

docs/how_to/deploy/adreno.rst

+Alternatively, to build a TVM via docker using OpenCL-Headers and set-up
+with Android TVM RPC, refer to this guide: `Deploy the Pretrained Model on Android <https://tvm.apache.org/docs/how_to/deploy_models/deploy_model_on_android.html>`_.


Phrase about OpenCL-Headers is not actual now. OpenCL Headers are not necessary for building TVM now and this documentation also was updated in this commit: 2bb3382

Suggested change

Alternatively, to build a TVM via docker using OpenCL-Headers and set-up

with Android TVM RPC, refer to this guide: `Deploy the Pretrained Model on Android <https://tvm.apache.org/docs/how_to/deploy_models/deploy_model_on_android.html>`_.

Since there are many required packages for Android, you can use the official Docker Image to build TVM. For more information refer to this guide: `Deploy the Pretrained Model on Android <https://tvm.apache.org/docs/how_to/deploy_models/deploy_model_on_android.html>`_.

supersat · 2022-11-25T09:40:19Z

Seems like PR #13073 broke these instructions, and now additionally requires specifying -DUSE_MICRO=OFF to cmake

echuraev

LGTM. Thanks

masahi · 2022-11-28T07:49:38Z

@echuraev You should be able to merge now.

dsbarinov1 force-pushed the dbarinov/adreno_docs_release branch from 9bf1929 to bb8696d Compare November 15, 2022 13:37

driazati reviewed Nov 15, 2022

View reviewed changes

echuraev requested changes Nov 16, 2022

View reviewed changes

dsbarinov1 added a commit to dsbarinov1/web-data that referenced this pull request Nov 18, 2022

Add images for apache/tvm#13393

0d1e89b

driazati pushed a commit to tlc-pack/web-data that referenced this pull request Nov 18, 2022

Add images for apache/tvm#13393 (#17)

02ed358

dsbarinov1 force-pushed the dbarinov/adreno_docs_release branch from 23a78ab to 28b1562 Compare November 22, 2022 11:41

dsbarinov1 added 18 commits November 25, 2022 02:06

Add correct links to images + small fixes

f99a22e

Remove images (.png)

7700637

Add request_hook in deploy_model_on_adreno.py

f9178ab

Fix trailing newline + add license

0809fb6

No newline at the EOF + blanks

53484be

Fix request hook placing

c427d82

Fix style

0e0a8ef

Fix trailing

1ca0d49

Fix whitespaces

a8783f0

Fix whitespaces v2

29f84d2

Add newline at adreno.rst EOF

edef640

Add license to adreno.rst

632fed6

Remove sphinx 'autosectionlabel' extension + modify cross-references …

4a95c81

…in docs to work without this extension

Set default values to tracker_host and tracker_port

039df07

Add local_demo to be able to autogenerate docs

12e5362

Fix quotes

8e45ad4

Fix benchmark

81a60ce

dsbarinov1 force-pushed the dbarinov/adreno_docs_release branch from 962b89a to 81a60ce Compare November 24, 2022 23:06

echuraev reviewed Nov 25, 2022

View reviewed changes

supersat mentioned this pull request Nov 25, 2022

[CI Problem] cross-compilations are broken #13482

Closed

echuraev approved these changes Nov 25, 2022

View reviewed changes

.

98ee8b5

dsbarinov1 force-pushed the dbarinov/adreno_docs_release branch from 6836a6f to 98ee8b5 Compare November 25, 2022 17:46

echuraev merged commit c38a0c5 into apache:main Nov 28, 2022

leandron mentioned this pull request Feb 1, 2023

TVM v0.11.0 Release Candidate Notes #13899

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Adreno] Add documentation for Adreno deployment #13393

[Adreno] Add documentation for Adreno deployment #13393

dsbarinov1 commented Nov 15, 2022 •

edited

Loading

tvm-bot commented Nov 15, 2022 •

edited

Loading

echuraev commented Nov 15, 2022

driazati left a comment •

edited

Loading

echuraev Nov 16, 2022

dsbarinov1 Nov 21, 2022

echuraev Nov 16, 2022

dsbarinov1 Nov 21, 2022

driazati commented Nov 21, 2022

echuraev Nov 25, 2022

supersat commented Nov 25, 2022

echuraev left a comment

masahi commented Nov 28, 2022

		Alternatively, to build a TVM via docker using OpenCL-Headers and set-up
		with Android TVM RPC, refer to this guide: `Deploy the Pretrained Model on Android <https://tvm.apache.org/docs/how_to/deploy_models/deploy_model_on_android.html>`_.

	Alternatively, to build a TVM via docker using OpenCL-Headers and set-up
	with Android TVM RPC, refer to this guide: `Deploy the Pretrained Model on Android <https://tvm.apache.org/docs/how_to/deploy_models/deploy_model_on_android.html>`_.
	Since there are many required packages for Android, you can use the official Docker Image to build TVM. For more information refer to this guide: `Deploy the Pretrained Model on Android <https://tvm.apache.org/docs/how_to/deploy_models/deploy_model_on_android.html>`_.

[Adreno] Add documentation for Adreno deployment #13393

[Adreno] Add documentation for Adreno deployment #13393

Conversation

dsbarinov1 commented Nov 15, 2022 • edited Loading

tvm-bot commented Nov 15, 2022 • edited Loading

echuraev commented Nov 15, 2022

driazati left a comment • edited Loading

Choose a reason for hiding this comment

echuraev Nov 16, 2022

Choose a reason for hiding this comment

dsbarinov1 Nov 21, 2022

Choose a reason for hiding this comment

echuraev Nov 16, 2022

Choose a reason for hiding this comment

dsbarinov1 Nov 21, 2022

Choose a reason for hiding this comment

driazati commented Nov 21, 2022

echuraev Nov 25, 2022

Choose a reason for hiding this comment

supersat commented Nov 25, 2022

echuraev left a comment

Choose a reason for hiding this comment

masahi commented Nov 28, 2022

dsbarinov1 commented Nov 15, 2022 •

edited

Loading

tvm-bot commented Nov 15, 2022 •

edited

Loading

driazati left a comment •

edited

Loading