Performance: Avoid rewrite work in passes when no patterns can match by taalexander · Pull Request #5315 · NVIDIA/cuda-quantum

taalexander · 2026-08-28T15:51:56Z

Six passes currently invoke MLIR's greedy rewrite driver
even when the IR contains none of the operation types their patterns can
match. On large modules, that turns a pass with no relevant work into a full IR
traversal.

This change gives those passes a small typed check. It walks the IR
once, stops at the first relevant operation, and returns early when there is no
possible pattern root. If a matching op is present, the pass runs exactly as it did
before.

I've applied this to:

LiftArrayAlloc
GlobalizeArrayValues
PyFrontendCleanup
GetConcreteMatrix
ConstantPropagation
StatePreparation

Performance

Benchmarking results a fixed number of operations:

Gates	Baseline pass time	New pass time	Time saved	Reduction	Speedup	Baseline process wall	New process wall
500,000	5.7525 s	1.7238 s	4.0287 s	70.0%	3.34x	8.73 s	4.42 s
1,000,000	12.8370 s	2.9031 s	9.9339 s	77.4%	4.42x	18.83 s	8.03 s
1,500,000	20.4164 s	7.5225 s	12.8939 s	63.2%	2.71x	31.29 s	20.50 s

At 1.5 million gates, every individual pass spent substantially less time:

Pass	Baseline	New	Time saved	Reduction	Speedup
`LiftArrayAlloc`	2.1607 s	0.5827 s	1.5780 s	73.0%	3.71x
`GlobalizeArrayValues`	2.3258 s	0.5546 s	1.7712 s	76.2%	4.19x
`PyFrontendCleanup`	2.1621 s	0.4843 s	1.6778 s	77.6%	4.46x
`GetConcreteMatrix`	2.5007 s	0.7070 s	1.7937 s	71.7%	3.54x
`ConstantPropagation`	2.6656 s	0.7523 s	1.9133 s	71.8%	3.54x
`StatePreparation`	3.2388 s	1.1060 s	2.1328 s	65.9%	2.93x

Avoid invoking six narrow greedy rewrite passes when their typed pattern roots are absent. A private variadic helper performs one interrupting operation walk, while root-present execution retains the complete existing rewrite path. This reduces root-free MLIR pass time by 66% to 78% across 100k to 1.5M gate inputs in the initial benchmark matrix. Baseline and candidate outputs were byte-identical in every measured root-free case and existing root-present comparisons. Verified with the focused transform and AST lit tests, all 41 optimizer unit tests, full hardware and emulation pipeline comparisons, clang-format, and git diff --check. Signed-off-by: Thomas Alexander <talexander@nvidia.com>

github-actions · 2026-08-28T16:34:24Z

CI Summary (`push`) — ✅ passed

Run #33203169644 · ✅ 7 · ⏩ 7 · ❌ 0 · ⛔ 0

Top-level jobs (14)

Job	Result
`binaries`	⏩ skipped
`build_and_test`	✅ success
`changes`	✅ success
`config_devdeps`	✅ success
`config_source_build`	⏩ skipped
`config_wheeldeps`	✅ success
`devdeps`	✅ success
`docker_image`	⏩ skipped
`gen_code_coverage`	⏩ skipped
`metadata`	✅ success
`python_metapackages`	⏩ skipped
`python_wheels`	⏩ skipped
`source_build`	⏩ skipped
`wheeldeps`	✅ success

⏩ Skipped jobs (7) — intentionally skipped on PR builds; run on merge_group / workflow_dispatch

Job
`binaries`
`config_source_build`
`docker_image`
`gen_code_coverage`
`python_metapackages`
`python_wheels`
`source_build`

All sub-jobs (43) — every matrix leg, with links

Job	Status	Link
Build and test (amd64, gcc12, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, gcc12, openmpi) / Dev environment (Python)	✅ success	view
Build and test (amd64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, llvm, openmpi) / Dev environment (Python)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Python)	✅ success	view
CI Summary	❔ in_progress	view
Check for stable CUDA-Q changes	✅ success	view
Configure build (devdeps)	✅ success	view
Configure build (source_build)	⏩ skipped	view
Configure build (wheeldeps)	✅ success	view
Create CUDA Quantum installer	⏩ skipped	view
Create Docker images	⏩ skipped	view
Create Python metapackages	⏩ skipped	view
Create Python wheels	⏩ skipped	view
Gen code coverage	⏩ skipped	view
Load dependencies (amd64, gcc12) / Caching	✅ success	view
Load dependencies (amd64, gcc12) / Finalize	✅ success	view
Load dependencies (amd64, gcc12) / Metadata	✅ success	view
Load dependencies (amd64, llvm) / Caching	✅ success	view
Load dependencies (amd64, llvm) / Finalize	✅ success	view
Load dependencies (amd64, llvm) / Metadata	✅ success	view
Load dependencies (arm64, gcc12) / Caching	✅ success	view
Load dependencies (arm64, gcc12) / Finalize	✅ success	view
Load dependencies (arm64, gcc12) / Metadata	✅ success	view
Load dependencies (arm64, llvm) / Caching	✅ success	view
Load dependencies (arm64, llvm) / Finalize	✅ success	view
Load dependencies (arm64, llvm) / Metadata	✅ success	view
Load source build cache	⏩ skipped	view
Load wheel dependencies (amd64, 12.6) / Caching	✅ success	view
Load wheel dependencies (amd64, 12.6) / Finalize	✅ success	view
Load wheel dependencies (amd64, 12.6) / Metadata	✅ success	view
Load wheel dependencies (amd64, 13.0) / Caching	✅ success	view
Load wheel dependencies (amd64, 13.0) / Finalize	✅ success	view
Load wheel dependencies (amd64, 13.0) / Metadata	✅ success	view
Load wheel dependencies (arm64, 12.6) / Caching	✅ success	view
Load wheel dependencies (arm64, 12.6) / Finalize	✅ success	view
Load wheel dependencies (arm64, 12.6) / Metadata	✅ success	view
Load wheel dependencies (arm64, 13.0) / Caching	✅ success	view
Load wheel dependencies (arm64, 13.0) / Finalize	✅ success	view
Load wheel dependencies (arm64, 13.0) / Metadata	✅ success	view
Prepare cache clean-up	❔ in_progress	view
Retrieve PR info	✅ success	view

✅ Required checks (6/6) — declared in .github/required-checks.yml for push

Required check	Status	Link
Build and test (amd64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, llvm, openmpi) / Dev environment (Python)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (arm64, llvm, openmpi) / Dev environment (Python)	✅ success	view
Build and test (amd64, gcc12, openmpi) / Dev environment (Debug)	✅ success	view
Build and test (amd64, gcc12, openmpi) / Dev environment (Python)	✅ success	view

Signed-off-by: Thomas Alexander <talexander@nvidia.com>

atgeller

One nit, otherwise LGTM.

Describe scaling checks, rewrite-driver choices, avoiding repeated IR work, and timing and tracing. Rename the operation-presence helper for clarity. Signed-off-by: Thomas Alexander <talexander@nvidia.com>

schweitzpgi · 2026-08-31T16:15:30Z

    LLVM_DEBUG(llvm::dbgs() << "Before constant prop:\n" << func << '\n');

    if (failed(
            applyPatternsGreedily(func.getOperation(), std::move(patterns)))) {


From the description, I was surprised to see another pre-traversal being added instead of converting these passes to partial conversion passes in favor of greedy rewrites. Is there a reason the latter was not done?

Some of these passes do use the greedy rewrite behaviour so it was not drop in. Another approach would be something like I did in #5318. I'll take another look.

taalexander requested review from atgeller and schweitzpgi August 28, 2026 15:51

taalexander added the performance label Aug 28, 2026

github-actions Bot added the core compiler label Aug 28, 2026

Document cheap early exits for compiler passes

dca8357

Signed-off-by: Thomas Alexander <talexander@nvidia.com>

github-actions Bot added the documentation Improvements or additions to documentation label Aug 28, 2026

taalexander added the agentic-autoresearch Tracking automated development by agentic systems label Aug 28, 2026

atgeller approved these changes Aug 28, 2026

View reviewed changes

Comment thread cudaq/lib/Optimizer/Transforms/PassDetails.h Outdated

docs: add pass performance guidance

ab02c24

Describe scaling checks, rewrite-driver choices, avoiding repeated IR work, and timing and tracing. Rename the operation-presence helper for clarity. Signed-off-by: Thomas Alexander <talexander@nvidia.com>

schweitzpgi reviewed Aug 31, 2026

View reviewed changes

Navigation Menu

Uh oh!

Performance: Avoid rewrite work in passes when no patterns can match - #5315

Performance: Avoid rewrite work in passes when no patterns can match#5315
taalexander wants to merge 3 commits into
NVIDIA:mainfrom
taalexander:perf/skip-root-free-greedy-passes

taalexander commented Aug 28, 2026

Uh oh!

github-actions Bot commented Aug 28, 2026 •

edited

Loading

Uh oh!

atgeller left a comment

Uh oh!

Uh oh!

schweitzpgi Aug 31, 2026

Uh oh!

taalexander Aug 31, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

Uh oh!

Conversation

taalexander commented Aug 28, 2026

Performance

Uh oh!

github-actions Bot commented Aug 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CI Summary (push) — ✅ passed

Uh oh!

atgeller left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

schweitzpgi Aug 31, 2026

Choose a reason for hiding this comment

Uh oh!

taalexander Aug 31, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

github-actions Bot commented Aug 28, 2026 •

edited

Loading

CI Summary (`push`) — ✅ passed