Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Good First Issue] [Snippets] [ARM]: Enable FakeQuantize tokenization #28508

Open
a-sidorova opened this issue Jan 17, 2025 · 2 comments
Open
Assignees
Labels
category: CPU OpenVINO CPU plugin good first issue Good for newcomers platform: arm OpenVINO on ARM / ARM64

Comments

@a-sidorova
Copy link
Contributor

Context

Snippets is a highly specialized JIT (just-in-time) compiler for computational graphs. It provides a scalable approach to operations' fusions and enablement. As a typical compiler, Snippets have frontend (tokenizer), optimizer and backend (generator).

The first of the Snippets pipeline, Tokenization, identifies parts of the initial ov::Model that can be lowered by Snippets efficiently, and tokenizes them into one whole node - Subgraph.
The second step of the pipeline (Optimizer) is applying common and device-specific optimizations to Subgraph and getting lowered representation of tokenized Subgraph.
Finally, the last stage is code emission. The target generator maps every operation in the IR to a binary code emitter JIT Emitter, which is then used to produce a piece of executable code. As a result, we produce an executable that performs calculations described by the initial input ov::Model.

The purpose of this issue is enabling FakeQuantize operation tokenization in Snippets for ARM devices.
Snippets decomposes FakeQuantize into several simple elementwise operations using FakeQuantizeDecomposition pass. This pass is called after the op tokenization into Subgraph.

Prerequisites

Recommended to use ARM CPU based platform for development (e.g. Mac, Raspberry Pi etc). The cross-compilation with an emulator (e.g. QEMU) using is still option: cmake -DCMAKE_TOOLCHAIN_FILE=../cmake/arm64.toolchain.cmake ...

What needs to be done?

  • Firstly, enable tests which are currently disabled on aarch64 platforms. For that need to update this line to enable smoke*Snippets*_FQDecomposition_* tests. Launch tests (how to launch them - please see the section Tests below). There should be failed tests.
  • Support all the missing operations that occurred during the decomposition pass in the ARM64 machine for the code generation.
  • Enable FakeQuantize op tokenization in tokenizer callback in CPU Plugin - update the is_supported_op.
  • Launch tests again - the tests should be green.

Tests

Tests are disabled in default build, so ensure to add -DENABLE_TESTS=ON into cmake command.

GoogleTest is used for testing. CPU functional test target is ov_cpu_func_tests. You can use GoogleTest filter:

./bin/[platform]/[build_type]/ov_cpu_func_tests --gtest_filter="*smoke*Snippets*FQDecomposition*"

Examples

Resources

Contact points

@a-sidorova, @dmitry-gorokhov

@a-sidorova a-sidorova added category: CPU OpenVINO CPU plugin good first issue Good for newcomers platform: arm OpenVINO on ARM / ARM64 labels Jan 17, 2025
@github-project-automation github-project-automation bot moved this to Contributors Needed in Good first issues Jan 17, 2025
@srinjoydutta03
Copy link
Contributor

.take

Copy link
Contributor

Thank you for looking into this issue! Please let us know if you have any questions or require any help.

@a-sidorova a-sidorova moved this from Contributors Needed to Assigned in Good first issues Jan 20, 2025
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
category: CPU OpenVINO CPU plugin good first issue Good for newcomers platform: arm OpenVINO on ARM / ARM64
Projects
Status: Assigned
Development

No branches or pull requests

2 participants