Skip to content

Add generate_identity_sequences helper and replace lambdas with named functors#4283

Closed
assistant-librarian[bot] wants to merge 5 commits intodevelopfrom
import/develop/ROCm_composable_kernel/pr-3628
Closed

Add generate_identity_sequences helper and replace lambdas with named functors#4283
assistant-librarian[bot] wants to merge 5 commits intodevelopfrom
import/develop/ROCm_composable_kernel/pr-3628

Conversation

@assistant-librarian
Copy link
Copy Markdown
Contributor

@assistant-librarian assistant-librarian bot commented Feb 3, 2026 •
edited by DDEle
Loading

Summary

  • Add generate_identity_sequences<N>() helper that returns Tuple<Sequence<0>, Sequence<1>, ..., Sequence<N-1>>
  • Replace lambdas with named functors in transform_tensor_descriptor
  • Add unpack_and_merge_sequences helper functor
  • Reduces transform_tensor_descriptor instantiations from 388 to 32 (92% reduction)

Motivation

Multiple call sites use generate_tuple([](auto i) { return Sequence<i>{}; }, Number<N>{}) pattern. A named helper reduces lambda instantiations.

Additionally, each lambda in transform_tensor_descriptor creates a unique closure type, causing the function to be instantiated separately for every call site. Named functors share a single type, so the compiler reuses the same instantiation.

Changes

Part 1: generate_identity_sequences helper

  • Replaces common lambda pattern for generating identity sequences
  • Each lambda expression creates a unique closure type, causing separate template instantiations at every call site
  • Named helper shares a single type across all uses

Part 2: Named functors in transform_tensor_descriptor

  • Add unpack_and_merge_sequences helper to replace lambda in GetNumOfHiddenDimension
  • Use generate_identity_sequences in matrix_padder.hpp

Test Plan

  • Added 7 unit tests:
    • 4 tests for generate_identity_sequences
    • 3 tests for unpack_and_merge_sequences
  • Waiting for full CI

Related PRs

This PR merges the functionality from:

Part of PR stack for issue #4229 (Reduce CK/CKTile Build Times)

Note: This PR supersedes ROCm/composable_kernel#3588 and ROCm/composable_kernel#3589, which can be closed once this is merged.

🔁 Imported from ROCm/composable_kernel#3628
🧑‍💻 Originally authored by @tenpercent

tenpercent and others added 5 commits January 29, 2026 14:26
@tenpercent @claude
Optimize tensor descriptor functor template instantiation
bce6ec1 
Replace inline lambdas with named functor structs in transform_tensor_descriptor
to reduce template instantiation overhead and improve compile times.

Changes:
- Add three named functors in tensor_descriptor.hpp:
  - convert_visible_to_hidden_id: maps visible dimension ID to hidden ID
  - convert_visible_ids_to_hidden_ids: maps sequence of visible IDs to hidden IDs
  - generate_arithmetic_sequence_from_scan: generates consecutive hidden dim ID ranges

- Add utility functions in sequence_helper.hpp and tuple_helper.hpp:
  - unpack_and_merge_sequences(): unpacks tuple of sequences and merges them
  - generate_identity_sequences(): creates Tuple<Sequence<0>, Sequence<1>, ...>

- Update 14 call sites across threadwise transfer, wrapper, and device files
  to use generate_identity_sequences() instead of generate_tuple with lambdas

- Add comprehensive unit tests:
  - unit_sequence_helper.cpp: tests for new utility functions
  - unit_tensor_descriptor_functors.cpp: tests for new functors

Co-Authored-By: Claude <noreply@anthropic.com>
@tenpercent
Merge branch 'develop' into tenpercent/tensor-descriptor-functor-opti…
fa98184 
…mization
@tenpercent
Merge branch 'develop' into tenpercent/tensor-descriptor-functor-opti…
2ec116f 
…mization
@tenpercent
Merge branch 'develop' into tenpercent/tensor-descriptor-functor-opti…
c6a8d06 
…mization
@systems-assistant
Merge commit 'c6a8d0613893760f87232c97c3af549de4cb6dc0' into import/d…
772c8dd 
…evelop/ROCm_composable_kernel/pr-3628
@assistant-librarian assistant-librarian bot added the external contribution Code contribution from users community.. label Feb 3, 2026
This was referenced Feb 10, 2026
Draft
Draft
@illsilin
Copy link
Copy Markdown
Contributor

illsilin commented Feb 17, 2026

Hi @tenpercent, please resolve the merge conflict and run CI.

@tenpercent tenpercent mentioned this pull request Feb 23, 2026
Merged
2 tasks
@tenpercent
Copy link
Copy Markdown
Contributor

tenpercent commented Feb 23, 2026

Closing in favor of PR #4828 which has a cleaner commit history.

@tenpercent tenpercent closed this Feb 23, 2026
shumway pushed a commit that referenced this pull request Feb 28, 2026
@tenpercent @claude
Add generate_identity_sequences helper and replace lambdas with named…
7de19bb 
… functors (#4828)

## Summary

- Add `generate_identity_sequences<N>()` helper that returns
`Tuple<Sequence<0>, Sequence<1>, ..., Sequence<N-1>>`
- Replace lambdas with named functors in `transform_tensor_descriptor`
- Add `unpack_and_merge_sequences` helper functor
- Reduces `transform_tensor_descriptor` instantiations from 388 to 32
(92% reduction)

## Motivation

Multiple call sites use `generate_tuple([](auto i) { return
Sequence<i>{}; }, Number<N>{})` pattern. A named helper reduces lambda
instantiations.

Additionally, each lambda in `transform_tensor_descriptor` creates a
unique closure type, causing the function to be instantiated separately
for every call site. Named functors share a single type, so the compiler
reuses the same instantiation.

## Changes

### Part 1: generate_identity_sequences helper
- Replaces common lambda pattern for generating identity sequences
- Each lambda expression creates a unique closure type, causing separate
template instantiations at every call site
- Named helper shares a single type across all uses

### Part 2: Named functors in transform_tensor_descriptor
- Add `unpack_and_merge_sequences` helper to replace lambda in
`GetNumOfHiddenDimension`
- Use `generate_identity_sequences` in `matrix_padder.hpp`

## Test Plan

- [x] Added 7 unit tests:
  - 4 tests for `generate_identity_sequences`
  - 3 tests for `unpack_and_merge_sequences`
- [ ] Waiting for full CI

## Related PRs

This PR merges the functionality from:
- ROCm/composable_kernel#3588 (generate_identity_sequences helper)
- ROCm/composable_kernel#3589 (Named functors in
transform_tensor_descriptor)

Part of PR stack for issue #4229 (Reduce CK/CKTile Build Times)

**Note:** This PR supersedes #4283, ROCm/composable_kernel#3588 and
ROCm/composable_kernel#3589, which can be closed once this is merged.

---
🔁 Imported from
[ROCm/composable_kernel#3628](ROCm/composable_kernel#3628)
🧑‍💻 Originally authored by @tenpercent

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
kokolchin pushed a commit to kokolchin/rocm-libraries that referenced this pull request Mar 4, 2026
@tenpercent @claude @kokolchin
Add generate_identity_sequences helper and replace lambdas with named…
f818d6c 
… functors (ROCm#4828)

## Summary

- Add `generate_identity_sequences<N>()` helper that returns
`Tuple<Sequence<0>, Sequence<1>, ..., Sequence<N-1>>`
- Replace lambdas with named functors in `transform_tensor_descriptor`
- Add `unpack_and_merge_sequences` helper functor
- Reduces `transform_tensor_descriptor` instantiations from 388 to 32
(92% reduction)

## Motivation

Multiple call sites use `generate_tuple([](auto i) { return
Sequence<i>{}; }, Number<N>{})` pattern. A named helper reduces lambda
instantiations.

Additionally, each lambda in `transform_tensor_descriptor` creates a
unique closure type, causing the function to be instantiated separately
for every call site. Named functors share a single type, so the compiler
reuses the same instantiation.

## Changes

### Part 1: generate_identity_sequences helper
- Replaces common lambda pattern for generating identity sequences
- Each lambda expression creates a unique closure type, causing separate
template instantiations at every call site
- Named helper shares a single type across all uses

### Part 2: Named functors in transform_tensor_descriptor
- Add `unpack_and_merge_sequences` helper to replace lambda in
`GetNumOfHiddenDimension`
- Use `generate_identity_sequences` in `matrix_padder.hpp`

## Test Plan

- [x] Added 7 unit tests:
  - 4 tests for `generate_identity_sequences`
  - 3 tests for `unpack_and_merge_sequences`
- [ ] Waiting for full CI

## Related PRs

This PR merges the functionality from:
- ROCm/composable_kernel#3588 (generate_identity_sequences helper)
- ROCm/composable_kernel#3589 (Named functors in
transform_tensor_descriptor)

Part of PR stack for issue ROCm#4229 (Reduce CK/CKTile Build Times)

**Note:** This PR supersedes ROCm#4283, ROCm/composable_kernel#3588 and
ROCm/composable_kernel#3589, which can be closed once this is merged.

---
🔁 Imported from
[ROCm/composable_kernel#3628](ROCm/composable_kernel#3628)
🧑‍💻 Originally authored by @tenpercent

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
NaveenElumalaiAMD pushed a commit that referenced this pull request Mar 6, 2026
@tenpercent @claude
Add generate_identity_sequences helper and replace lambdas with named…
183cbd5 
… functors (#4828)

## Summary

- Add `generate_identity_sequences<N>()` helper that returns
`Tuple<Sequence<0>, Sequence<1>, ..., Sequence<N-1>>`
- Replace lambdas with named functors in `transform_tensor_descriptor`
- Add `unpack_and_merge_sequences` helper functor
- Reduces `transform_tensor_descriptor` instantiations from 388 to 32
(92% reduction)

## Motivation

Multiple call sites use `generate_tuple([](auto i) { return
Sequence<i>{}; }, Number<N>{})` pattern. A named helper reduces lambda
instantiations.

Additionally, each lambda in `transform_tensor_descriptor` creates a
unique closure type, causing the function to be instantiated separately
for every call site. Named functors share a single type, so the compiler
reuses the same instantiation.

## Changes

### Part 1: generate_identity_sequences helper
- Replaces common lambda pattern for generating identity sequences
- Each lambda expression creates a unique closure type, causing separate
template instantiations at every call site
- Named helper shares a single type across all uses

### Part 2: Named functors in transform_tensor_descriptor
- Add `unpack_and_merge_sequences` helper to replace lambda in
`GetNumOfHiddenDimension`
- Use `generate_identity_sequences` in `matrix_padder.hpp`

## Test Plan

- [x] Added 7 unit tests:
  - 4 tests for `generate_identity_sequences`
  - 3 tests for `unpack_and_merge_sequences`
- [ ] Waiting for full CI

## Related PRs

This PR merges the functionality from:
- ROCm/composable_kernel#3588 (generate_identity_sequences helper)
- ROCm/composable_kernel#3589 (Named functors in
transform_tensor_descriptor)

Part of PR stack for issue #4229 (Reduce CK/CKTile Build Times)

**Note:** This PR supersedes #4283, ROCm/composable_kernel#3588 and
ROCm/composable_kernel#3589, which can be closed once this is merged.

---
🔁 Imported from
[ROCm/composable_kernel#3628](ROCm/composable_kernel#3628)
🧑‍💻 Originally authored by @tenpercent

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
jovanau pushed a commit to jovanau/rocm-libraries that referenced this pull request Mar 19, 2026
@tenpercent @claude @jovanau
Add generate_identity_sequences helper and replace lambdas with named…
e183717 
… functors (ROCm#4828)

## Summary

- Add `generate_identity_sequences<N>()` helper that returns
`Tuple<Sequence<0>, Sequence<1>, ..., Sequence<N-1>>`
- Replace lambdas with named functors in `transform_tensor_descriptor`
- Add `unpack_and_merge_sequences` helper functor
- Reduces `transform_tensor_descriptor` instantiations from 388 to 32
(92% reduction)

## Motivation

Multiple call sites use `generate_tuple([](auto i) { return
Sequence<i>{}; }, Number<N>{})` pattern. A named helper reduces lambda
instantiations.

Additionally, each lambda in `transform_tensor_descriptor` creates a
unique closure type, causing the function to be instantiated separately
for every call site. Named functors share a single type, so the compiler
reuses the same instantiation.

## Changes

### Part 1: generate_identity_sequences helper
- Replaces common lambda pattern for generating identity sequences
- Each lambda expression creates a unique closure type, causing separate
template instantiations at every call site
- Named helper shares a single type across all uses

### Part 2: Named functors in transform_tensor_descriptor
- Add `unpack_and_merge_sequences` helper to replace lambda in
`GetNumOfHiddenDimension`
- Use `generate_identity_sequences` in `matrix_padder.hpp`

## Test Plan

- [x] Added 7 unit tests:
  - 4 tests for `generate_identity_sequences`
  - 3 tests for `unpack_and_merge_sequences`
- [ ] Waiting for full CI

## Related PRs

This PR merges the functionality from:
- ROCm/composable_kernel#3588 (generate_identity_sequences helper)
- ROCm/composable_kernel#3589 (Named functors in
transform_tensor_descriptor)

Part of PR stack for issue ROCm#4229 (Reduce CK/CKTile Build Times)

**Note:** This PR supersedes ROCm#4283, ROCm/composable_kernel#3588 and
ROCm/composable_kernel#3589, which can be closed once this is merged.

---
🔁 Imported from
[ROCm/composable_kernel#3628](ROCm/composable_kernel#3628)
🧑‍💻 Originally authored by @tenpercent

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
johannes-graner pushed a commit that referenced this pull request Mar 20, 2026
@tenpercent @claude @johannes-graner
Add generate_identity_sequences helper and replace lambdas with named…
cbf6d95 
… functors (#4828)

## Summary

- Add `generate_identity_sequences<N>()` helper that returns
`Tuple<Sequence<0>, Sequence<1>, ..., Sequence<N-1>>`
- Replace lambdas with named functors in `transform_tensor_descriptor`
- Add `unpack_and_merge_sequences` helper functor
- Reduces `transform_tensor_descriptor` instantiations from 388 to 32
(92% reduction)

## Motivation

Multiple call sites use `generate_tuple([](auto i) { return
Sequence<i>{}; }, Number<N>{})` pattern. A named helper reduces lambda
instantiations.

Additionally, each lambda in `transform_tensor_descriptor` creates a
unique closure type, causing the function to be instantiated separately
for every call site. Named functors share a single type, so the compiler
reuses the same instantiation.

## Changes

### Part 1: generate_identity_sequences helper
- Replaces common lambda pattern for generating identity sequences
- Each lambda expression creates a unique closure type, causing separate
template instantiations at every call site
- Named helper shares a single type across all uses

### Part 2: Named functors in transform_tensor_descriptor
- Add `unpack_and_merge_sequences` helper to replace lambda in
`GetNumOfHiddenDimension`
- Use `generate_identity_sequences` in `matrix_padder.hpp`

## Test Plan

- [x] Added 7 unit tests:
  - 4 tests for `generate_identity_sequences`
  - 3 tests for `unpack_and_merge_sequences`
- [ ] Waiting for full CI

## Related PRs

This PR merges the functionality from:
- ROCm/composable_kernel#3588 (generate_identity_sequences helper)
- ROCm/composable_kernel#3589 (Named functors in
transform_tensor_descriptor)

Part of PR stack for issue #4229 (Reduce CK/CKTile Build Times)

**Note:** This PR supersedes #4283, ROCm/composable_kernel#3588 and
ROCm/composable_kernel#3589, which can be closed once this is merged.

---
🔁 Imported from
[ROCm/composable_kernel#3628](ROCm/composable_kernel#3628)
🧑‍💻 Originally authored by @tenpercent

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Reviewers

@illsilin illsilin Awaiting requested review from illsilin

@carlushuang carlushuang Awaiting requested review from carlushuang

@qianfengz qianfengz Awaiting requested review from qianfengz

@aosewski aosewski Awaiting requested review from aosewski

@poyenc poyenc Awaiting requested review from poyenc

@geyyer geyyer Awaiting requested review from geyyer

@bartekxk bartekxk Awaiting requested review from bartekxk

@andriy-ca andriy-ca Awaiting requested review from andriy-ca

@afagaj afagaj Awaiting requested review from afagaj

@asleepzzz asleepzzz Awaiting requested review from asleepzzz

@tenpercent tenpercent Awaiting requested review from tenpercent

@ThomasNing ThomasNing Awaiting requested review from ThomasNing

@coderfeli coderfeli Awaiting requested review from coderfeli

Assignees

@cgmillette cgmillette

@illsilin illsilin

Labels

external contribution Code contribution from users community.. imported pr project: composablekernel

Projects

None yet

Milestone

No milestone

Development

Successfully merging this pull request may close these issues.

3 participants

@illsilin @tenpercent @cgmillette