[EISW-165337] Port and Adapt Custom VPUx Changes for LLVM 20.x Integration #126

sramasit · 2025-05-15T01:18:46Z

Based on the following discussion:
https://llvm.discourse.group/t/declarative-assembly-format-requirement-for-type-presence/4399

Relax checks in OperationParser - it allows to skip value type specification, if the value was already defined in the same block.

Differential Revision: https://reviews.llvm.org/D111650

Summary

This PR ports several custom VPUX plugin changes from npu/release/19.x to npu/release/20.x to maintain consistency across both branches and enable smoother compiler integration with LLVM 20.x.
Some cherry-picked changes did not apply cleanly or introduced runtime/test failures due to differences in LLVM 20.x APIs and dialect behaviors. Fixes for these issues have been squashed into the corresponding commits, so certain commits in this PR may differ slightly from their original versions in npu/release/19.x.

JIRA ticket

EISW-165337

Related PR in NPU Compiler and/or OpenVINO repository with sub-module update

PR-19736

Other related tickets

List tickets for additional work, eg, something was found during review but you agreed to address it in another Jira

E-xxxxx

andrey-golubev · 2025-06-30T09:05:57Z

I think you need to cherry-pick a couple more patches (see the current state of 19.x):

7f2cd91 - Maxim's update
544ee1a - my recent addition

mlir/CMakeLists.txt

llvm/include/llvm/ADT/FunctionExtras.h

llvm/cmake/modules/TableGen.cmake

ZoranZomborat

Ported changes for Quant dialect seem fine

andrey-golubev · 2025-07-15T09:30:49Z

On the patches,

Why do we need b3d3082 once again? I'd imagine that "standard" LLVM clang-format / pr-code-format should be enough. After all, this is a "fork" repository.
What is this needed for: 1ed9846? (Maybe I've missed the discussion) Also, as far as I see, this is a "revert" of 561395e -- which was also added by us?! -> Can we just delete these two commits completely then?
Could these three: a1793a8, 8a923ae, 13bf57c be moved to the tip of the HEAD? I don't see them in 19.x at all also.

Other than that, I don't see c7caf07 being ported but I think we've discussed this before with @ArtemySkrebkov and the consensus is that we don't need it.

* Fix <anonymous> declared here * Fix Ubuntu24 (intel#33)

* Expanding Quant dialect with Quantile Quantized type * Adding quantile mlir tests * Adding check on quantiles array size and updated mlir tests

Bumps [certifi](https://github.com/certifi/python-certifi) from 2024.6.2 to 2024.7.4. - [Commits](certifi/python-certifi@2024.06.02...2024.07.04) --- updated-dependencies: - dependency-name: certifi dependency-type: direct:production ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

…ntel#60)

Bumps [cryptography](https://github.com/pyca/cryptography) from 42.0.4 to 43.0.1. - [Changelog](https://github.com/pyca/cryptography/blob/main/CHANGELOG.rst) - [Commits](pyca/cryptography@42.0.4...43.0.1) --- updated-dependencies: - dependency-name: cryptography dependency-type: direct:production ... Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* relax negative scale conditions * adapt tests now handling negative scales

* Adding missing check in QuantileQuantizedPerAxisType::verify * Adding minor test for illegal quantile array size for per axis type

This reverses partially changed in commit 2c1ae80

…bolTable (#80590)" This reverts commit 7d055af.

Update pr-code-format.yml from llvm-project

…om crash diagnostics

* Bump cryptography from 43.0.1 to 44.0.1 in /llvm/utils/git Bumps [cryptography](https://github.com/pyca/cryptography) from 43.0.1 to 44.0.1. - [Changelog](https://github.com/pyca/cryptography/blob/main/CHANGELOG.rst) - [Commits](pyca/cryptography@43.0.1...44.0.1) --- updated-dependencies: - dependency-name: cryptography dependency-version: 44.0.1 dependency-type: direct:production ... Signed-off-by: dependabot[bot] <[email protected]> * Bump tar-fs from 2.1.1 to 2.1.2 in /mlir/utils/vscode Bumps [tar-fs](https://github.com/mafintosh/tar-fs) from 2.1.1 to 2.1.2. - [Commits](mafintosh/tar-fs@v2.1.1...v2.1.2) --- updated-dependencies: - dependency-name: tar-fs dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> * Bump tar-fs from 2.1.1 to 2.1.2 in /lldb/tools/lldb-dap Bumps [tar-fs](https://github.com/mafintosh/tar-fs) from 2.1.1 to 2.1.2. - [Commits](mafintosh/tar-fs@v2.1.1...v2.1.2) --- updated-dependencies: - dependency-name: tar-fs dependency-version: 2.1.2 dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> * Bump jinja2 from 3.1.4 to 3.1.6 in /llvm/docs Bumps [jinja2](https://github.com/pallets/jinja) from 3.1.4 to 3.1.6. - [Release notes](https://github.com/pallets/jinja/releases) - [Changelog](https://github.com/pallets/jinja/blob/main/CHANGES.rst) - [Commits](pallets/jinja@3.1.4...3.1.6) --- updated-dependencies: - dependency-name: jinja2 dependency-version: 3.1.6 dependency-type: direct:production ... Signed-off-by: dependabot[bot] <[email protected]> * Bump llvm/utils/git/requirements.txt cryptography --------- Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

…ARNINGS is ON (intel#99) Signed-off-by: Zhu, Shaojie <[email protected]>

…and-line parser reset When LLVM TableGen flags are globally appended, mlir-src-sharder fails with an unknown argument error because its command-line parser is reset during execution and does not retain the inherited flags.

…ck) (intel#142) Add a new API to access all blobs that are stored in the blob manager. The main purpose (as of now) is to allow users of dialect resources to iterate over all blobs, especially when the blobs are no longer used in IR (e.g. the operation that uses the blob is deleted) and thus cannot be easily accessed without manual tracking of keys.

* Bump requests from 2.32.3 to 2.32.4 in /llvm/docs Bumps [requests](https://github.com/psf/requests) from 2.32.3 to 2.32.4. - [Release notes](https://github.com/psf/requests/releases) - [Changelog](https://github.com/psf/requests/blob/main/HISTORY.md) - [Commits](psf/requests@v2.32.3...v2.32.4) --- updated-dependencies: - dependency-name: requests dependency-version: 2.32.4 dependency-type: direct:production ... Signed-off-by: dependabot[bot] <[email protected]> * Bump requests from 2.32.2 to 2.32.4 in /llvm/utils/git Bumps [requests](https://github.com/psf/requests) from 2.32.2 to 2.32.4. - [Release notes](https://github.com/psf/requests/releases) - [Changelog](https://github.com/psf/requests/blob/main/HISTORY.md) - [Commits](psf/requests@v2.32.2...v2.32.4) --- updated-dependencies: - dependency-name: requests dependency-version: 2.32.4 dependency-type: direct:production ... Signed-off-by: dependabot[bot] <[email protected]> * Bump brace-expansion from 2.0.1 to 2.0.2 in /mlir/utils/tree-sitter-mlir Bumps [brace-expansion](https://github.com/juliangruber/brace-expansion) from 2.0.1 to 2.0.2. - [Release notes](https://github.com/juliangruber/brace-expansion/releases) - [Commits](juliangruber/brace-expansion@v2.0.1...v2.0.2) --- updated-dependencies: - dependency-name: brace-expansion dependency-version: 2.0.2 dependency-type: indirect ... Signed-off-by: dependabot[bot] <[email protected]> * Update README badges --------- Signed-off-by: dependabot[bot] <[email protected]> Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

This is an implementation for [RFC: Supporting Sub-Channel Quantization in MLIR](https://discourse.llvm.org/t/rfc-supporting-sub-channel-quantization-in-mlir/82694). In order to make the review process easier, the PR has been divided into the following commit labels: 1. **Add implementation for sub-channel type:** Includes the class design for `UniformQuantizedSubChannelType`, printer/parser and bytecode read/write support. The existing types (per-tensor and per-axis) are unaltered. 2. **Add implementation for sub-channel type:** Lowering of `quant.qcast` and `quant.dcast` operations to Linalg operations. 3. **Adding C/Python Apis:** We first define he C-APIs and build the Python-APIs on top of those. 4. **Add pass to normalize generic ....:** This pass normalizes sub-channel quantized types to per-tensor per-axis types, if possible. A design note: - **Explicitly storing the `quantized_dimensions`, even when they can be derived for ranked tensor.** While it's possible to infer quantized dimensions from the static shape of the scales (or zero-points) tensor for ranked data tensors ([ref](https://discourse.llvm.org/t/rfc-supporting-sub-channel-quantization-in-mlir/82694/3) for background), there are cases where this can lead to ambiguity and issues with round-tripping. ``` Consider the example: tensor<2x4x!quant.uniform<i8:f32:{0:2, 0:2}, {{s00:z00, s01:z01}}>> ``` The shape of the scales tensor is [1, 2], which might suggest that only axis 1 is quantized. While this inference is technically correct, as the block size for axis 0 is a degenerate case (equal to the dimension size), it can cause problems with round-tripping. Therefore, even for ranked tensors, we are explicitly storing the quantized dimensions. Suggestions welcome! PS: I understand that the upcoming holidays may impact your schedule, so please take your time with the review. There's no rush.

This commit introduces several CMake options to conditionally enable or disable specific dialects and conversions. This allows for a more modular build and helps resolve build failures in configurations where certain dialects are not needed or are causing issues. Part-of: EISW-163563, EISW-170058, EISW-163564

mlir-tblgen fails to generate doc for IRDL because while parsing IRDL.td it also finds the definition of the builtin dialect and cannot determine what dialect to generate documentation for. This is because IRDL.td includes IRDLOps.td which includes BuiltinAttributes.td which in turns includes BuiltinDialect.td.

nikita-kud · 2025-10-07T09:09:57Z

The PR should not have been merged because we had not finished the discussion. I did not revert the PR, but dropped last commit: Fix APInt call

sramasit force-pushed the sramasit/20.x branch 5 times, most recently from c6d1bc0 to 1b1f99c Compare May 29, 2025 04:52

sramasit force-pushed the sramasit/20.x branch 2 times, most recently from ee8aadd to ecbf209 Compare June 12, 2025 00:44

sramasit force-pushed the sramasit/20.x branch 6 times, most recently from 783b49c to 9ef830b Compare June 27, 2025 22:52

sramasit requested review from Maxim-Doronin, andrey-golubev and nikita-kud June 28, 2025 05:15

nikita-kud reviewed Jun 30, 2025

View reviewed changes

mlir/CMakeLists.txt Outdated Show resolved Hide resolved

nikita-kud reviewed Jun 30, 2025

View reviewed changes

llvm/include/llvm/ADT/FunctionExtras.h Outdated Show resolved Hide resolved

nikita-kud reviewed Jun 30, 2025

View reviewed changes

llvm/cmake/modules/TableGen.cmake Outdated Show resolved Hide resolved

ZoranZomborat approved these changes Jul 4, 2025

View reviewed changes

sramasit force-pushed the sramasit/20.x branch 5 times, most recently from 2307dd8 to b3d3082 Compare July 15, 2025 06:24

sramasit changed the title ~~Port custom changes from 19.x to 20.x~~ [EISW-165337] Port custom changes from 19.x to 20.x Jul 15, 2025

sramasit changed the title ~~[EISW-165337] Port custom changes from 19.x to 20.x~~ [EISW-165337] Port and Adapt Custom VPUx Changes for LLVM 20.x Integration Jul 15, 2025

sramasit force-pushed the sramasit/20.x branch from 2698d5e to 2fde0db Compare July 23, 2025 02:55

DariaMityagina and others added 22 commits October 6, 2025 12:24

Port Ubunu-24 warning fixes to npu/release/18.x (intel#45)

ee6bba2

* Fix <anonymous> declared here * Fix Ubuntu24 (intel#33)

Extend Quant dialect with Quantile Quantization type (intel#53)

30e56d6

* Expanding Quant dialect with Quantile Quantized type * Adding quantile mlir tests * Adding check on quantiles array size and updated mlir tests

Extending QuantileQuantizedType with quantileType mlir::Type member (i…

f7c71d8

…ntel#60)

Bump cryptography and certifi versions (intel#65)

164988d

Relax negative scales conditions in Quant Dialect (intel#61)

8e1745e

* relax negative scale conditions * adapt tests now handling negative scales

Adding missing check in QuantileQuantizedPerAxisType::verify (intel#66)

695a2f1

* Adding missing check in QuantileQuantizedPerAxisType::verify * Adding minor test for illegal quantile array size for per axis type

Adding new quantileType types to mlir test (intel#85)

f1f8d16

Add Value back to SideEffectInterfaces

fd8af31

This reverses partially changed in commit 2c1ae80

Revert "[mlir][Symbol] Add verification that symbol's parent is a Sym…

282f06a

…bolTable (#80590)" This reverts commit 7d055af.

Linux precommit (intel#88)

7dca4bd

GitHub Workflow changes

d694cab

Update pr-code-format.yml from llvm-project

Update CMake configuration in Linux precommit workflow to enable cust…

126e8c3

…om crash diagnostics

Fix thread handle leak on Windows (intel#133)

6c523ef

EISW-130780: remove all wd flag to enable warnings when LLVM_ENABLE_W…

98e7c78

…ARNINGS is ON (intel#99) Signed-off-by: Zhu, Shaojie <[email protected]>

Security update 2025ww27 (intel#145)

d00ef59

sramasit force-pushed the sramasit/20.x branch from df5c947 to cfa0fa5 Compare October 6, 2025 19:28

sramasit and others added 3 commits October 6, 2025 12:49

Fix APInt call

ddec344

sramasit force-pushed the sramasit/20.x branch from cfa0fa5 to ddec344 Compare October 6, 2025 19:50

nikita-kud approved these changes Oct 6, 2025

View reviewed changes

sramasit merged commit 0ad3ee1 into intel:npu/release/20.x Oct 7, 2025
9 checks passed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[EISW-165337] Port and Adapt Custom VPUx Changes for LLVM 20.x Integration #126

[EISW-165337] Port and Adapt Custom VPUx Changes for LLVM 20.x Integration #126

Uh oh!

sramasit commented May 15, 2025 •

edited

Loading

Uh oh!

andrey-golubev commented Jun 30, 2025 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ZoranZomborat left a comment

Uh oh!

andrey-golubev commented Jul 15, 2025

Uh oh!

Uh oh!

nikita-kud commented Oct 7, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

17 participants

[EISW-165337] Port and Adapt Custom VPUx Changes for LLVM 20.x Integration #126

[EISW-165337] Port and Adapt Custom VPUx Changes for LLVM 20.x Integration #126

Uh oh!

Conversation

sramasit commented May 15, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

JIRA ticket

Related PR in NPU Compiler and/or OpenVINO repository with sub-module update

Other related tickets

Uh oh!

andrey-golubev commented Jun 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

ZoranZomborat left a comment

Choose a reason for hiding this comment

Uh oh!

andrey-golubev commented Jul 15, 2025

Uh oh!

Uh oh!

nikita-kud commented Oct 7, 2025

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

17 participants

sramasit commented May 15, 2025 •

edited

Loading

andrey-golubev commented Jun 30, 2025 •

edited

Loading