Additional tracing and resource estimation target improvements#4432
Open
taalexander wants to merge 8 commits intoNVIDIA:mainfrom
Open
Additional tracing and resource estimation target improvements#4432taalexander wants to merge 8 commits intoNVIDIA:mainfrom
taalexander wants to merge 8 commits intoNVIDIA:mainfrom
Conversation
Signed-off-by: Thomas Alexander <talexander@nvidia.com>
Signed-off-by: Thomas Alexander <talexander@nvidia.com>
Signed-off-by: Thomas Alexander <talexander@nvidia.com>
Signed-off-by: Thomas Alexander <talexander@nvidia.com>
Signed-off-by: Thomas Alexander <talexander@nvidia.com>
Signed-off-by: Thomas Alexander <talexander@nvidia.com>
Signed-off-by: Thomas Alexander <talexander@nvidia.com>
CI Summary — ✅ passedRun #25229541107 · trigger Top-level jobs (13)
⏩ Skipped jobs (7) — intentionally skipped on PR builds; run on merge_group / workflow_dispatch
All sub-jobs (50) — every matrix leg, with links
✅ Required checks (8/8) — declared in
|
| Required check | Status | Link |
|---|---|---|
| Build and test (amd64, clang16, openmpi) / Dev environment (Debug) | ✅ success | view |
| Build and test (amd64, clang16, openmpi) / Dev environment (Python) | ✅ success | view |
| Build and test (amd64, gcc11, openmpi) / Dev environment (Debug) | ✅ success | view |
| Build and test (amd64, gcc11, 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 |
| Build and test (arm64, clang16, openmpi) / Dev environment (Debug) | ✅ success | view |
| Build and test (arm64, clang16, openmpi) / Dev environment (Python) | ✅ success | view |
schweitzpgi
reviewed
May 4, 2026
| jit-low-level-pipeline: "func.func(apply-control-negations,canonicalize,cse),symbol-dce" | ||
| # Target high-level stage: materialize registered custom-op matrices and | ||
| # synthesize them to gates before target lowering. | ||
| jit-high-level-pipeline: "get-concrete-matrix,unitary-synthesis" |
Collaborator
There was a problem hiding this comment.
Before this pipeline is run, we should always run
pm.addNestedPass<func::FuncOp>(cudaq::opt::createQuakeAddDeallocs());
pm.addNestedPass<func::FuncOp>(cudaq::opt::createQuakeAddMetadata());
pm.addPass(cudaq::opt::createQuakePropagateMetadata());
pm.addNestedPass<func::FuncOp>(cudaq::opt::createUnwindLowering());
pm.addNestedPass<func::FuncOp>(createCanonicalizerPass());
pm.addNestedPass<func::FuncOp>(cudaq::opt::createClassicalMemToReg());
cudaq::opt::createClassicalOptimizationPipeline(pm, std::nullopt,
{options.allowEarlyExit});
pm.addPass(cudaq::opt::createGlobalizeArrayValues());
pm.addNestedPass<func::FuncOp>(createCanonicalizerPass());
pm.addPass(cudaq::opt::createUnitarySynthesis());
pm.addNestedPass<func::FuncOp>(createCanonicalizerPass());
pm.addPass(cudaq::opt::createApplySpecialization(
{.constantPropagation = options.applyConstProp}));
cudaq::opt::addAggressiveInlining(pm);as well as some other passes.
Collaborator
There was a problem hiding this comment.
So unitary-synthesis is already part of that pipeline and GetConcreteMatrix realy only matters if you're using user-defined custom ops.
| # Target mid-level stage: inline synthesized helper calls, decompose to a CX | ||
| # routing basis, prepare wire metadata, optionally route, then lower routed | ||
| # two-qubit work to CZ. With no device argument, routing is bypassed. | ||
| jit-mid-level-pipeline: "apply-op-specialization,aggressive-inlining,decomposition{basis=h,rx,ry,rz,x,x(1)},func.func(add-dealloc,combine-quantum-alloc,canonicalize,factor-quantum-alloc,memtoreg),add-wireset,func.func(assign-wire-indices),qubit-mapping{device=%DEVICE:bypass%},func.func(delay-measurements,regtomem),decomposition{basis=h,rx,ry,rz,x,z(1)}" |
Collaborator
There was a problem hiding this comment.
Before this pipeline, we'll run
cudaq::opt::createClassicalOptimizationPipeline(pm);
cudaq::opt::addDecomposition(pm, {std::string("U3ToRotations")});
pm.addNestedPass<func::FuncOp>(createCanonicalizerPass());
pm.addNestedPass<func::FuncOp>(cudaq::opt::createMultiControlDecomposition());The mid-level pipeline is where we normally would put custom (per target) gate set mappings, routings, etc. So this is good. Do you really want delay-measurements? That's sort of a weird hack pass for the IQM target.
Collaborator
There was a problem hiding this comment.
Here, apply op specialization and inlining should have already been done in the high-level pipeline. If we're doing them here again and they are actually transforming code in some way, it would be good to have an example and understand what's wrong.
| # two-qubit work to CZ. With no device argument, routing is bypassed. | ||
| jit-mid-level-pipeline: "apply-op-specialization,aggressive-inlining,decomposition{basis=h,rx,ry,rz,x,x(1)},func.func(add-dealloc,combine-quantum-alloc,canonicalize,factor-quantum-alloc,memtoreg),add-wireset,func.func(assign-wire-indices),qubit-mapping{device=%DEVICE:bypass%},func.func(delay-measurements,regtomem),decomposition{basis=h,rx,ry,rz,x,z(1)}" | ||
| # Target low-level cleanup stage: remove wire-set symbols left by routing prep. | ||
| jit-low-level-pipeline: "symbol-dce" |
Collaborator
There was a problem hiding this comment.
And before this pipeline string
pm.addNestedPass<func::FuncOp>(createCanonicalizerPass());
pm.addNestedPass<func::FuncOp>(createCSEPass());
pm.addPass(createSymbolDCEPass());So, here we're re-running symbol-dce.
Unfortunately, there is the caveat that instead of simply applying the pipeline as constructed, the pipelines that are strings can be mutated by the launch code in ad hoc ways, which is a bad idea.
Collaborator
There was a problem hiding this comment.
There is another user-defined pipeline that can be specified to run after this low-level one called post-codegen-passes. I think we can ignore that one and just use low-level for our purposes here.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Follow ups as noted by @schweitzpgi and @boschmitt.