Warp Eval

NVIDIAOfficialCreated by NVIDIA Corporation

Skill
Workflow
Software development
Apache-2.0
Medium risk
Has scripts

From the author

Evaluate whether an existing hot path is a credible NVIDIA Warp candidate. Use for irregular or spatial queries, particle or geometry simulation, branch-heavy loops, many small launches, host fallbacks, or large intermediates. CPU-only code and absent GPU dependencies are normal unless NVIDIA is prohibited. Exclude required cross-vendor or CPU-only deployment, vendor-lowered dense or NN layers, general Warp API questions, and already-selected Warp kernels. Contribution policy alone is not exclusion.

— the description by NVIDIA Corporation

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

The warp-eval skill provides a structured, evidence-based methodology to determine if specific hot paths in an existing codebase are viable candidates for NVIDIA Warp. It emphasizes objective reporting over recommendations, requiring users to define clear performance metrics and semantic contracts before any prototyping occurs. The process includes static screening, rigorous baseline comparisons, and authorized benchmarking, ensuring that Warp is only considered after in-project alternatives have been exhausted. The skill enforces strict gates to abort evaluations that do not meet technical or policy requirements, ensuring that only credible candidates proceed to measurement.

What it does

Screens code bottlenecks against specific technical gates to determine Warp suitability.
Requires the establishment of a semantic contract and performance baseline before prototyping.
Enforces a ladder of in-project optimizations to be exhausted before introducing Warp.
Produces a standardized evaluation report containing measurements, evidence, and caveats.
Ensures all performance claims are backed by synchronized, end-to-end measurements.
Provides scripts for synchronized timing, memory profiling, and report validation.

When to use it

When investigating performance bottlenecks involving irregular queries, simulations, or branch-heavy loops.
When you need to objectively assess if Warp provides a measurable benefit over existing in-project baselines.
When you require a reproducible, audit-ready report on potential GPU acceleration candidates.
When you need to determine if a performance issue is material enough to warrant a new dependency.

Advantages

Prevents premature optimization by requiring algorithmic improvements first.
Provides a clear, objective framework to stop evaluations that are not technically viable.
Ensures performance measurements include realistic overheads like JIT, transfers, and synchronization.
Maintains production code integrity by working in isolated, independent patches.
Standardizes the reporting of facts, hypotheses, and unknowns.

Usage examples

“Evaluate if the particle simulation loop in my project is a good candidate for Warp.”: The agent will identify the candidate seam, derive the contract, check initial gates, and provide a resource preview before asking for authorization to proceed with profiling.
“I have a performance bottleneck in my geometry processing code; can you run a warp-eval?”: The agent will initiate the screening process, request the target repository, and begin identifying candidate patterns to determine if a formal evaluation is warranted.

Install

Download the skill as a ZIP:
https://mp-staging.autogineer.com/agent-skills/nvidia/warp-eval/download
Unzip it into the cross-client skills folder of your project:
```
mkdir -p .agents/skills && unzip warp-eval.zip -d .agents/skills
```

Documentation by NVIDIA Corporation

Warp evaluation

Purpose

Collect reproducible evidence about how a narrow seam in an existing codebase would behave in NVIDIA Warp. Report facts; the user decides.

Name Warp as the option under evaluation in the first line, state that no adoption recommendation will follow, and do not treat the triggering performance request as authorization to experiment.

Measured evaluations produce warp-evaluation-report/: the report, one independently applicable diff per solution, the drivers, and raw results. Never modify production code. Exits before measured work create no directory.

Hard rules

These override any local reasoning.

Default objectives are latency, throughput, and peak or retained memory. Count maintainability, ergonomics, packaging, extensibility, autodiff or new functionality only when the user names it; otherwise report them as constraints or costs, not benefits.
Correctness is a gate. If the incumbent is buggy or its contract unclear, abort that comparison until an independent oracle or clarified contract exists. Report the defect without prescribing a response.
Never predict savings from source shape or another workload. Materialized memory comes from profiler or allocator evidence, not source a compiler may fuse.
"It can be written in Warp" is a hypothesis, never a proven opportunity.
Gates are exact, never adjacent or analogous. Fire a gate only when every condition in its definition is established. Gate D requires a maintained implementation confirmed to execute with CUDA on an NVIDIA GPU; a fast native CPU library is a baseline, not Gate D.
Label every claim as observed fact, measurement, hypothesis or unknown.
Time the user-visible stage, not the kernel. Include Warp's cold import/init/JIT in the real process regime, transfers, launches, Python launch loops, structure build/refit, allocation, conversion, validation, compaction and synchronization. Report each cost and the end-to-end difference.
No measured work without explicit authorization. After stage 1, stop and ask before profiling, environment changes, prototyping, benchmarking or GPU use.
Authorized measured work includes Warp. If profiling exposes no gate, continue through the strongest in-project baseline, minimum Warp prototype and end-to-end comparison. If Warp is out of scope, stop before profiling.
Never infer environment intent. NVIDIA deployment and ownership of an optional compiled dependency are product decisions. Abort only on a stated constraint; otherwise ask once and stop (AWAITING INTENT).
Warp measurements are CUDA-only and synchronized. Resolve an explicit NVIDIA CUDA device; discard CPU resolution and dispatch-only timing.
Never recommend or rank adoption options. Report facts, measurements, hypotheses and unknowns per seam and regime.
Stop at the report. No rollout or production edits.

Requirements

Static screening requires the target repository and its stated product constraints. Measured work additionally requires explicit authorization, an NVIDIA CUDA GPU, target-project dependencies and a representative workload.

Limitations

Warp CPU kernels are serial and outside this skill's evaluation scope. "Run it on the CPU instead" is not a performance fallback.
Mesh and geometry queries compute in float32/int32 even behind a float64 public API. Absolute error grows with coordinate magnitude.
Warp does not follow semver. Feature releases can break APIs, only the newest feature line is maintained, deprecations run roughly four monthly releases.
latest docs track development, not the shipped release. Pin to the target project's own Warp version; if it has none, use the current stable and say which.
Kernel-only builtins are not resolvable from Python — hasattr(wp, "mesh_query_point") is False on a release that has it. Probe the stub file or that version's docs before concluding a builtin is absent.
enable_backward=False (kernel, module or global) removes adjoint codegen. If nothing differentiates through the seam, set it before measuring compile cost.

Output Format

State	Reached when	Report directory
`ABORT`	Any gate establishes that the evaluated seam cannot satisfy the stated scope	No if nothing was measured; otherwise preserve the evidence already collected
`AWAITING INTENT`	The environment gates turn on a fact only the user has	No — one question, both branches concrete
`AWAITING AUTHORIZATION`	A candidate pattern survives stage 1	No — early findings plus scope/resource preview
`INCOMPLETE`	Authorized work cannot obtain representative evidence required for the scoped evaluation	Yes — preserve collected evidence and name the one missing artifact
Report delivered	Authorized work produced measurements, or stopped after the report directory existed	Yes — facts per seam and regime, with missing evidence explicit

Use this exact shape for an early exit: ABORT — Gate <letter>: <cited fact>; <why the scoped Warp evaluation cannot proceed>. Do not name a preferred alternative. Reporting rules: references/evidence-and-reporting.md. A delivered directory follows the template's fixed order: schema and provenance, authorization and evaluation state, stage census, one B<n> evidence section per seam/regime, caveats, then environment/reproduction. solutions/, benchmarks/ and results/ contain every linked artifact.

Examples

Gate exit: ABORT — Gate A: deployment.md requires one implementation with AMD, Apple and NVIDIA parity; a Warp-specific path cannot satisfy this scope.

Surviving candidate: Name the seam and pattern, label inferred facts as assumptions, state that no gate has fired, preview the profile/baseline/Warp prototype/benchmark scope and its cost, then ask the separate intent and authorization questions from references/authorization-checkpoint.md.

Inputs

Required: the target repository and a performance, memory or scale problem with a candidate seam. Optional: explicit deployment/packaging constraints, existing profiles or logs, representative datasets and acceptance criteria. Prompt constraints take precedence over repository policy/configuration, then existing logs. User corrections override inference. Never substitute an assumption for a stated fact or measurement.

Available scripts

Script	Purpose	Arguments
`scripts/driver-template.py`	Copy once per bottleneck; define workloads and variants	Edit placeholders, then run the copied driver
`scripts/measure.py`	Import from drivers for synchronized timing, memory and isolated cases	Python API; do not execute directly
`scripts/validate_report_schema.py`	Validate the delivered report and evidence links	`<report-directory>`

Use run_script("scripts/validate_report_schema.py", args=["warp-evaluation-report"]) when supported; otherwise invoke the script with Python and the report directory.

Troubleshooting

No representative workload: mark the scope INCOMPLETE and name the missing artifact; do not invent data or fire Gate F.
Warp resolves to CPU or no CUDA device: discard the run and stop before correctness or timing claims.
Report validation fails: fix the report or referenced artifact; never waive the schema error.

Instructions

Every stage before the last can end the evaluation. Stop as soon as a gate fires; do not gather evidence that cannot change the scoped facts.

1. Read the code, derive the contract, check the gates

Identify a candidate and its metric with references/target-patterns.md.
Derive devices/residency, dtypes/shapes, sizes, frequency, process lifetime, gradients and packaging from the repository. Infer before asking.
State the inferred contract in one line and invite correction. Unknown hardware, counts, sizes and tolerances are assumptions, never measurements. Every inference remains open to correction and cannot satisfy a gate that requires a stated fact or measurement.
Check Gates A–E before profiling. Check Gate F now only if representative evidence already exists; otherwise carry it into stage 2. Every gate uses only the exact boundaries in references/rejection-gates.md.

Gate	Fires when
A	Production is stated CPU-only or to need non-NVIDIA portability, with no acceptable optional CUDA path
B	Data must cross the host/device boundary per small or infrequent call and the boundary cannot be widened
C	The region is dense tensor algebra already mapped to a tuned framework or vendor library
D	A mature CUDA implementation already meets the contract, and no non-performance objective was requested
E	A stated policy blocks Warp's dependency, compilation, cache or fallback obligations
F	Representative evidence proves the region too small a share of its requested metric for any backend to move it

Gate F can fire in stage 1 only from representative evidence that already exists — a supplied profile, structural bound, or arithmetic on figures the user quoted. If that evidence does not exist, Gate F remains open until stage 2 profiling; inferred values never fire it.
Gates A and E need a stated constraint. A CPU implementation, another accelerator, no Warp dependency, or a small dependency list proves nothing.
When A/E are unresolved and a pattern survives, ask whether an optional NVIDIA path is acceptable: named extra, soft import, existing fallback, default install unchanged. Every affirmative answer must say explicitly that Warp will be prototyped and benchmarked; conditions constrain only that Warp scope. A negative or undecided answer means ABORT.
Do not ask when another gate fired, the repository answers, or no pattern matched. No pattern means no profiling.
If a candidate survives, combine any intent question with the authorization checkpoint — early findings, exact scope, stages, resource cost — then stop. Stage 2 requires both settled intent and explicit authorization.

2. Profile the real application

Requires explicit authorization and a settled intent question.

Profile with the project's own profiler and representative entry points.
Measure synchronized end-to-end stage time and peak memory before choosing a backend. Report which entry points were profiled and which a gate screened.
Prioritize further measurement by observed cost, not source appearance.
Name each measurement by the public method and variant actually invoked. A fallback is an execution regime of that public seam, not a different operation, and a cheaper sibling method cannot screen out the named method.
Confirm the timed branch ran on the intended device. Unchanged cost and near-zero device allocation between host and device inputs exposes a host fallback.
Measure the stage's free-stage ceiling and stubbed floor through the public boundary. Do not subtract per-op timings.
If the measured ceiling proves the candidate cannot move its own metric, ABORT the affected scope under Gate F, preserve the evidence already collected, and stop. The existing authorization already covered this materiality check; do not ask for authorization again.
If representative coverage is unavailable — no representative dataset, runnable entry point or production distribution — record the single missing artifact, mark the affected scope INCOMPLETE, and stop. This is missing evidence, not Gate F and not ABORT. An invented workload cannot prove materiality.

Protocol: references/benchmark-protocol.md.

3. Form falsifiable hypotheses

Record per candidate: source, bottleneck evidence, objective, narrow seam, mechanism Warp could change, strongest incumbent, risks, acceptance threshold, cheapest falsifying experiment. Screen against references/target-patterns.md; if none survives, write the report and stop.

4. Write the contract before the prototype

Define values, dtypes, shapes, devices, errors, mutation, ordering, ties, capacity/overflow, topology/degeneracy, tolerances, required gradients, streams, ownership, aliasing, invalidation, concurrency, capture, teardown and fallback.

Fix tolerances before seeing Warp output. Never weaken a contract after a mismatch.
ABORT before prototyping if the proposed seam cannot satisfy a required contract.
Pre-register, before timing: workload provenance, the state variable and production range controlling cost, tuning knobs, incumbent run-to-run spread, and the oracle applied to every implementation.

Hazards and adversarial checks: references/semantic-contract.md.

5. Improve the baseline first

Algorithm before backend: (1) a better or output-sensitive algorithm; (2) chunking, tiling, sparse output, layout, rematerialization; (3) the incumbent framework's compiler and native primitives; (4) what the project already depends on — its own accelerator backend, a parallel idiom it ships but leaves off, or a capability an existing dependency exposes and nobody wired up; (5) only then narrow Warp.

Compare only in-scope options: the improved incumbent, capabilities reachable through current dependencies, and Warp. Do not add unrelated libraries.
Search declared dependencies for dormant backends, flags and bindings before calling a route absent.
Keep the improved incumbent as the oracle. A result against an untuned, incorrect or asymptotically inferior baseline does not establish a backend comparison.

Close every in-project route before prototyping Warp:

State	What it takes to claim it
measured	timed through the same boundary as the baseline
absent	a cited declaration, symbol table or missing flag proves it is unavailable
waived	you asked the user and they chose to skip it; record their words

A capability present but unbound is reachable, not absent. If exposing it costs no more than the planned Warp seam, measure it first. Ladder details: references/baselines.md. Waived routes do not block stage 6; every route not explicitly waived must be measured or evidenced absent before the Warp prototype begins.

6. Prototype the minimum unit

Work in a separate copy, production unchanged, integration seam off by default. Prototype only enough to test the hypothesis.
Select an explicit CUDA device for every Warp prototype and verify that Warp resolves it as CUDA before correctness or performance work. Never exercise or report Warp's CPU backend.
Size the unit by shared data and structure lifetime, not function boundaries: build/refit/query costs that amortize together are one unit.
Name variants before timing. Preserve each solution as an independent patch against the pinned baseline, and prove it applies cleanly and reproduces the measured result.
Surface unsupported cases and overflow before timing.
Run the adversarial contract checks, audit correctness at production scale, and deliberately break a branch to prove the tests fail. Any failed required check returns ABORT for the affected scope.
Before treating a disappointing compute-bound or reuse-heavy result as representative, consider the stronger formulations in references/target-patterns.md. One naive kernel does not bound Warp's potential.

7. Benchmark the whole boundary

Copy scripts/driver-template.py once per bottleneck and measure through scripts/measure.py. Do not hand-roll timing or memory.
Warp launches are asynchronous. The measurement helper must synchronize the selected device immediately before starting and after enqueueing every timed region, before stopping its wall timer.
Include every rule-7 cost, the public call, the immediate downstream stage, realistic sizes, and cold/warm process regimes.
Transcribe every report cell from emitted JSON; absent records read not measured.
Report null_test, and one-time costs both separately and amortized. Serialize GPU measurements under an exclusive device lock. Below 1.5× is no measured difference.
ABORT for a seam and regime whose predeclared end-to-end performance or memory requirement fails, after preserving the measurements.

Full protocol: references/benchmark-protocol.md.

8. Summarize the evidence

Per seam and execution regime, report semantic results, strongest-baseline measurements, workload provenance, time, memory, lifecycle, portability and ownership facts.
Use pass, fail, not measured, not available, no representative data, unknown or n/a only where a stated criterion makes that status objective. Missing evidence remains missing.
Mark the report complete only when every evaluated seam and regime includes an end-to-end Warp measurement. Use aborted — <gate and scope> when a gate fired, or incomplete — <missing evidence and scope> when representative evidence was unavailable. Preserve everything collected. Never deliver an incumbent-only report as a completed warp-eval.
Give every stage worth ≥10 % of the measured total a census row with a status and a one-line reason, including the screened-out stages.
Check each gate against the metric that candidate's pattern exhausts, not the study's headline objective.

9. Hand back

Use assets/warp-evaluation-report-template.md unchanged in schema. Record authorization and scope.
One table per bottleneck: baseline, current-dependency solutions, then Warp; absolute time and peak memory, ratios, contract status, evidence gaps.
Include prose only where it explains the measurements or their bounds. Working artifacts belong in results/.

uv run python scripts/validate_report_schema.py <report-directory>

Run the validator once the first measurement establishes a census/table, and again before delivery. Fix every error.
Verify builds and imports from artifacts, not exit status. Ship the exact drivers run. Then stop.

Files

ATTRIBUTION.mdAdded by AutoGineer897 B
BENCHMARK.md4.3 KiB
LICENSELicence file added by AutoGineer11.2 KiB
SKILL.md19.2 KiB
assets/warp-evaluation-report-template.md5.8 KiB
config/skillspector-baseline.yaml475 B
evals/evals.json22.3 KiB
evals/files/gate-a-required-portability/deployment.md638 B
evals/files/gate-a-required-portability/neighbors.pyScript872 B
evals/files/gate-b-unwidenable-host-boundary/architecture.md794 B
evals/files/gate-b-unwidenable-host-boundary/scoring.pyScript880 B
evals/files/gate-c-dense-vendor-lowered/feature_stage.pyScript615 B
evals/files/gate-d-cpu-incumbent-not-cuda/baseline.md472 B
evals/files/gate-d-cpu-incumbent-not-cuda/operation_api.pyScript440 B
evals/files/gate-d-mature-incumbent/baseline.md960 B
evals/files/gate-d-mature-incumbent/contact_api.pyScript948 B
evals/files/gate-e-runtime-obligations-forbidden/candidates.pyScript735 B
evals/files/gate-e-runtime-obligations-forbidden/packaging-policy.md845 B
evals/files/gate-f-immaterial-stage/profile.json603 B
evals/files/gate-f-immaterial-stage/spatial_filter.pyScript646 B
evals/files/pattern-fused-edge-intermediate/context.md742 B
evals/files/pattern-fused-edge-intermediate/message_passing.pyScript521 B
evals/files/pattern-sequential-recurrence/adaptive_filter.pyScript462 B
evals/files/pattern-sequential-recurrence/context.md596 B
evals/files/proceed-persistent-mesh-query/context.md767 B
evals/files/proceed-persistent-mesh-query/queries.pyScript804 B
references/authorization-checkpoint.md3.8 KiB
references/baselines.md12.2 KiB
references/benchmark-protocol.md16.3 KiB
references/evidence-and-reporting.md7.2 KiB
references/rejection-gates.md8.2 KiB
references/semantic-contract.md13.3 KiB
references/target-patterns.md16.3 KiB
scripts/driver-template.pyScript3.4 KiB
scripts/measure.pyScript20.8 KiB
scripts/validate_report_schema.pyScript13.5 KiB
skill-card.md4.4 KiB
skill.oms.sig12.5 KiB

references/baselines.md

Baselines — algorithm first, backend second

The most common way a Warp evaluation goes wrong is beating a baseline nobody would defend.

The ladder

Work down it in order, then measure Warp. An in-project route meeting the objective is an observed fact, not a reason to omit the Warp comparison from an authorized warp-eval.

A better algorithm — asymptotically better, or output-sensitive so it stops doing work proportional to a quantity that does not matter.
Representation and memory strategy — chunking, tiling, slabs, sparse output, layout change, rematerialization.
The incumbent framework's compiler and native primitives.
What the project already depends on — its own accelerator backend, a parallel idiom it ships but leaves off, or a capability an existing dependency exposes and nobody wired up.
Only then a narrow Warp implementation.

Rungs 1–4 are limited to what the project can reach without a new dependency. Warp is the only addition this skill puts on the table. Rungs 1 and 2 routinely end the study on their own.

The dependency set is what the manifests declare, not what happens to be installed or imported: pyproject.toml (including [project.optional-dependencies]), setup.cfg/setup.py, requirements*.txt, environment.yml, and the lockfile if one is committed. An optional extra is in-project — a capability it ships is a rung-4 route, not a new dependency.

Rung 1 is mandatory when the diagnosis is algorithmic

If the cause you wrote down is that the incumbent uses the wrong algorithm — a linear scan where an index belongs, a rebuild where a refit belongs, a materialized candidate set where a traversal belongs — rung 1 means implementing that algorithm on the incumbent's own hardware, not tuning the existing one. Tuning the wrong algorithm's constants measures the tuning, and leaves the backend comparison unable to separate the algorithm was wrong from the hardware was slow.

Until the right algorithm has run on the incumbent's hardware, no speedup may be attributed to the accelerator: measure it, evidence it impossible, or get it waived on the record. An honest, unoptimized implementation is enough to separate the two effects, so no external project is needed. If writing one is impractical, mark the backend comparison not measured.

Rung 4 supplies the direct comparison

If the project already ships a GPU path — numba.cuda, CuPy, its own CUDA kernels — the baseline for a new-backend proposal is that backend running the same algorithm, not the CPU path it currently uses. Port the candidate kernel to the dependency the project already has, measure it, and only then ask what a new dependency adds. If the existing backend closes the gap, record the observed parity or difference and still run the Warp comparison.

Skipping this rung turns a CPU-versus-GPU result into an apparent Warp-versus-everything result, and hides the common real finding: that nobody had moved the loop to the GPU at all.

When the project has no accelerator backend

The ordinary case, so rung 4 needs a substitute rather than a "not applicable", in this order:

The project's own parallel idiom, turned on — an OpenMP pragma in a sibling kernel, a numba.prange, an n_jobs/n_threads left at its default, a chunked path used elsewhere. This costs no new dependency and is frequently the whole finding.
A capability an existing dependency exposes but the project never wired up. Bindings routinely cover only part of a native API. Read the dependency's surface before concluding it cannot help; the fix may be a binding addition rather than a new dependency.

Record the substitution explicitly: "the project ships no accelerator backend; rung 4 is its own OpenMP idiom applied to this seam". A rung answered "no" and left there is how a serial baseline becomes the thing a GPU is measured against.

Do not substitute a library outside the project's declared dependencies. The question is whether Warp beats what the project can already reach, not whether it beats the ecosystem. If no existing dependency can express the better algorithm, say so — "no in-project route to a BVH query; the only options are a binding addition or Warp" is the finding.

Measure the in-project route even when it is the harder one to measure. Do not benchmark whatever installs in one command while deferring the route that needs a binding written, a flag plumbed through or a kernel ported. If reaching the existing dependency's capability takes an afternoon of implementation, that afternoon is part of the evaluation. Defer it only with the user's agreement, and say plainly that the direct comparison is missing.

Parallelism is part of the baseline's definition

A baseline is not "the incumbent" but "the incumbent at the parallelism the deployment actually has". A CPU baseline measured below the deployment's thread count has not been compared. State the thread count of every timed configuration and the core counts of both the measuring host and the target.

Preserve the oracle

Keep the strongest correct implementation selectable and use it as the correctness oracle for every later comparison. If you improve the incumbent here, the improved version — not the original — is what the prototype must beat.

Strongest practical baseline by incumbent

Incumbent	Baseline you must beat
NumPy	Vectorized formulation with tuned linked libraries and realistic CPU threading
CuPy	Built-in/library op, default accelerator backend, fusion, and a reasonable custom kernel
Numba	Optimized launch configuration, device-resident arrays, cached specialization
JAX	Outermost practical `jit`, correct static specialization, device-resident inputs, blocked completion; plus `scan`/`fori_loop` chunking and Pallas where justified
PyTorch	Built-in/fused op and `torch.compile`; plus any existing custom op that is the real production baseline
C++	Optimized compiler settings, representative threading/vectorization, relevant native libraries
CUDA	Tuned kernel or library, realistic streams and graphs, production compiler settings

Domain libraries can supply correctness references, but are performance baselines only when the project already declares them: Embree or Open3D for CPU BVH point/ray queries; a maintained CUDA BVH library; an indexed CUDA neighbor-search library; cuCIM/CuPy for image and label operations; libigl for geometry references; OptiX where full traversal semantics matter.

Include any approximate cached structure the project already supports. A voxel SDF, a precomputed grid or a coarse index answers a cheaper question and routinely beats an exact GPU path by a wide margin. It is not a drop-in competitor and its error must be stated — but if the product tolerates that error, report its measured time, memory and error beside Warp.

Equivalent work

Hold these constant, or a comparison means nothing: accepted input domain and filtering; outputs and ordering where contractually required; dtype, precision and fast-math policy; convergence criterion or iteration count; initialization and seeds; structure quality, or an explicitly justified difference; forward/backward requirement; synchronization semantics; inclusion of pre- and post-processing.

The invoked public operation and configuration are part of the work. A benchmark of a helper, fallback or sibling operation cannot support a row or screening claim about another operation. Hold input domain, option values, output cardinality, ordering, duplicate handling and missing-value representation constant. If a prototype implements only a subset, scope every result to that subset and leave the rest unmeasured.

Where APIs genuinely differ, state exactly what extra work each performs. An indexed CUDA neighbor library typically returns nearest-K neighbors and squared distances and includes self, while a fixed-capacity Warp path returns arbitrary-order neighbors plus an overflow flag. Both facts belong in the report.

Compare optimized with optimized

Default-versus-default comparisons inflate ratios dramatically, and collapse once both sides are tuned. State which comparison you ran; if you only had time for default/default, say so rather than presenting it as a backend speedup.
The rule is symmetric: a candidate that loses on its own unset defaults has not been compared, it has been misconfigured. Name the knobs before benchmarking, sweep the ones with defensible physical meaning, and report the tuned comparison alongside the default. Watch for a default heuristic wrong for the data's intrinsic dimension — a spatial-index cell size derived from bounding-box volume is far too coarse for points lying on a surface, and the candidate then loses to brute force. A knob that cannot be set from data the caller already has is a finding about the seam's API, not a license to hand-pick it per measurement.
Transferable optimizations belong to both sides. Reformulations, precomputation, blocking, fusing a division into a reciprocal and hoisting invariants are almost always backend-neutral. A speedup found while tuning the Warp kernel and never tried on the baseline is a formulation result reported as a backend result, and it routinely reverses. Port the idea, re-measure both, report the pair; if the incumbent gains as much, the idea belongs in the "improve the incumbent" finding. Re-run the correctness audit — a reformulation can change numerics.
Do not change the backend and the workload in one measurement. A headline ratio from porting and from processing many systems concurrently is both effects multiplied. Report the single-item comparison separately from the batched one; the gap is what the port bought.

A fast wrong baseline is not a baseline

An incumbent that does not satisfy its own contract cannot be a performance competitor. A fixed-iteration solver that has not converged, a truncated neighbor list, or a tolerance-collapsed geometric test all look fast while producing wrong answers — exclude them and mark them failing. Conversely, an intentionally weak baseline is evidence only that the baseline needs improving.

When the incumbent is the defective one

It also stops being a usable oracle, and "% agreement with the incumbent" inverts: the more correct an implementation is, the worse it scores. The candidate disagreeing on more elements may be the only one that is right.

Construct a reference satisfying the intended contract and make it the oracle. Say how you built it and why you believe it.
Classify every disagreement as a correction or a regression, judged by the incumbent's own metric, and report the two counts separately. "Differs on N" is not a finding; "corrects N₁, regresses N₂" is.
Compare each candidate against the constructed reference as well as against the incumbent. Comparing only against a defective incumbent conflates the bug being fixed with the candidate's own tie behavior.
Report the defect as a finding in its own right and abort the performance comparison until a valid oracle exists (hard rule 2 in SKILL.md).

A green test is not evidence the fast path is correct — check what its fixture exercises. A parity test between a project's reference and accelerated path can pass because the fixture is degenerate in exactly the variable the two paths disagree on: a straight rod where the dropped term is proportional to curvature, a zero mask, an empty batch, a single element. The guard exists, passes, and is structurally incapable of failing. Before trusting an accelerated incumbent as an oracle, confirm its fixture is non-degenerate in the quantities that distinguish the implementations; if it is not, build one that is and re-run there.

When the alternative is the defective one

An in-project alternative that turns out to be wrong — a backend that does not run with its own driver, or whose results diverge from the reference on non-degenerate input — is not a baseline, however attractive its time. Name the defect in its row's note, keep its raw results so the exclusion is auditable, and do not let its number stand as something the candidate had to beat.

Requirements and compatibility

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Needs Python

The author says: Screening, static evaluation and reporting need no GPU. Measuring Warp requires an NVIDIA CUDA GPU, the target project's dependencies and a representative workload; without them, abort before profiling.

Security

What our scanner noticed· Medium risk

Large base64-encoded blobs
Scripts that your agent may run

This skill includes scripts that your agent may run. Read them before you use it.

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by

NVIDIA Corporation

Publisher

NVIDIA (Official)

Source

NVIDIA/skills · skills/warp-eval · commit fce3cca

Licence

Apache License 2.0 (redistribution allowed)

LICENSE (added by AutoGineer from LICENSE-APACHE)
ATTRIBUTION.md (written by AutoGineer)

Synced

2026-11-08

AutoGineer is not affiliated with or endorsed by NVIDIA Corporation or NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download warp-eval.zip