Restart-aware optimization decisions, 2026-10-17

The CUDA ownership consolidation and stable-norm/control launch change are retained; the CPU four-row GEMV proposal is rejected, leaving the original two-row kernel. The initial CUDA resource variant is also rejected and remains separately reported. No complete wall-time study establishes a universal speedup or a pass for every row’s 5% regression budget.

The final supplemental CUDA study has no confirmed regression, but its whole verdict is inconclusive. Its 64 native/cuBLAS-to-Torch comparisons retain 56 lower and eight higher measured medians. All eight higher medians are cuBLAS Jacobi fresh API: all four sizes in host and device execution, with ratios 1.108486–1.463973. Native CUDA’s 32 comparisons and all measured Session/prepared comparisons are lower descriptively. These observations cover the declared fixtures and scopes, not statistical or universal wins.

All five A/B campaigns retain 176 fresh-process artifacts, 2,144 eligible outcomes and 536 strict candidate/case rows, including both rejected variants and every framework control. The 57 instrumented CUDA captures and 84 CPU component observations are separate diagnostics. The preceding restart-aware baseline retains its own 108 artifacts/1,336 results and 668 characterization rows; none is silently pooled into this evaluation. Its unremeasured modes remain historical context.

Completed ownership decisions

The baseline was compiled from 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7. Its stable public source tag is benchmarks/restart-baseline-20261017. The rejected resource-layout variant was compiled from e5c587889906511b0f9e74f42d88d8e857dea3d0. The separate borrowed-span follow-up was compiled from ce45ea9cc6d38cea05ef04cff8a321507e9684d5. Final merge heads do not replace those actual compiled identities. The complete tables retain actual source, stage, module and manifest identities; they are not inferred from a later checkout.

Each ownership campaign contains eight configurations, 32 fresh-process artifacts, 384 eligible results and 96 strict candidate/case rows. Both retain all same-stage I/M/R gates and all native/cuBLAS/Torch observations.

Completed ownership campaigns

Variant	Whole verdict	Within 5% budget	Inconclusive	Regression
Initial resource layout, rejected	regression	12	83	1
Borrowed-span follow-up, adopted for ownership	inconclusive	34	62	0

Every implementation’s strict row inventory

Variant	Implementation	Within 5% budget	Inconclusive	Regression
Initial	cuda-native	6	25	1
Initial	cuda-cublas	2	30	0
Initial	torch-cuda-eager	4	28	0
Follow-up	cuda-native	13	19	0
Follow-up	cuda-cublas	14	18	0
Follow-up	torch-cuda-eager	7	25	0

The original regression is native dense-256, Jacobi, prepared scope and host execution: current/baseline ratio 1.16509, 95% interval [1.05821, 1.25940], with unchanged 4/6/0 I/M/R. It remains in the final report even though the later variant has no confirmed regression. All other initial rows, including higher medians and inconclusive controls, remain too.

The follow-up was selected after observing the initial failure. Its 95% intervals are unadjusted for this selection and for evaluating another variant. It has its own fixed complete ABBA campaign, retains the same budget, and does not replace or pool with the rejected campaign. The observation does not establish that host-object layout caused the original regression. No control drift is subtracted, and no timing result is normalized by iterations.

The ownership change combines reduction/scalar storage and the optional full 4 MiB cuBLAS scratch prefix in one RAII allocation. An explicit borrowed byte span supplies the provider’s scratch; its handle is destroyed before that allocation. Fresh solve still owns fresh output/workspace/handle and optional Jacobi setup. Session and prepared contracts are unchanged.

Recorded diagnostics support one fewer cudaMalloc/cudaFree per fresh cuBLAS solve: ten fewer of each in the matching ten-call capture. Complete API inventories retain the extra single driver cuCtxSynchronize observed in a native-API diagnostic; not every synchronization count is unchanged. Allocation counts establish an allocation change, not an additive latency contribution or a general completed-wall improvement.

CPU and CUDA launch decisions

CPU four-row GEMV

The complete cpu-quad-abba/manifest.json retains 16 configurations, 64 artifacts and 832 eligible candidate/case results. Release includes both fusion settings; BLAS and two-thread OpenMP retain their declared controls. All 208 strict rows remain: 64 native-kernel rows, 16 private-BLAS rows and 128 NumPy/Torch controls. Every known I/M/R gate matches. The whole verdict is regression, and the four-row variant is not adopted.

The measured four-row source is 2d81b3d1409b0b428572cbf26fc7418e03bd270c. Tables labelled B/current describe that rejected measured source, not the production checkout. Only expanded correctness tests are retained from this CPU proposal; production keeps the original two-row dense kernel. No crossover threshold or automatic selection was added. The complete source/stage/binary identities and remaining-loss inventory below belong to the measured variant and must not be relabelled as latest-main performance.

Complete CPU strict inventory

Implementation	Rows	Within 5% budget	Inconclusive	Regression
native	64	15	45	4
native-blas	16	5	9	2
numpy-out	64	16	41	7
torch-eager-out	64	12	47	5

The unchanged BLAS provider and NumPy/Torch controls also retain regressions. These observations limit causal attribution; they neither justify subtracting control drift nor excuse the proposal’s failed adoption gate. All cases, including improvements and inconclusive results, remain in the tables.

The separate 84 fixed-vector component probes are retained below. They are one descriptive baseline pass and one candidate pass, not a confidence study. They use the four official seeded dense fixtures with vector=b and three padded integer tails with a separate fixed vector. Neither represents a changing CG direction. Matvec, fused reduction and whole-solve observations answer different questions; component medians cannot be summed into CG latency attribution or used to overrule the complete rejection. Odd-shape OpenMP rounding still requires the retained correctness tests; the component probes themselves use one thread.

CUDA stable-norm/control launch

The complete prepared/device study retains four configurations, 16 artifacts, 160 eligible results and 40 strict rows. Actual A is adopted-span source ce45ea9cc6d38cea05ef04cff8a321507e9684d5; actual B is 14c37277de16182fdc7f4cfcf9193e0b352791d1. Its whole verdict is regression, with the sole strict regression in the Torch control. No native/cuBLAS row is a confirmed regression; ten meet the 5% budget and fourteen remain inconclusive. This is not a complete-gate pass.

Complete prepared/device launch inventory

Implementation	Rows	Within 5% budget	Inconclusive	Regression
cuda-native	16	7	9	0
cuda-cublas	8	3	5	0
torch-cuda-eager	16	7	8	1

The Torch dense-64 Jacobi prepared/device row has current/baseline ratio 1.25757, 95% interval [1.12458, 1.42341]. It remains in the complete batch; no control drift is subtracted and no unrelated native ratio overrides it. All 24 native/cuBLAS medians are lower than their same-artifact Torch controls descriptively, without a confidence claim for those framework comparisons. The full candidate/control I/M/R and true-residual gates repeat.

The kernel runs the existing stable norm collective before residual/beta policy within the same 256-thread block. Initial and true-residual checks retain separate norm calls, while the ordinary post-update norm launch is merged. Arithmetic, residual priority and completed packet checks are retained. Explicit invalid-alpha exits save no launch. The six completed diagnostic captures are retained below as instrumentation, not completed-wall scores.

A separate supplemental scope study was fixed before any launch ABBA results were inspected. launch-scope-preregistration.json records UTC 2026-10-16T18:33:53.770121+00:00 and plan SHA256 3e9f579275d06bb4936f96e4759aeb120e2d7a96344b17750c1728b264887de8. All eight configurations and 32 processes completed, retaining 384 eligible outcomes and 96 strict rows. None/API is included in host and device modes; Jacobi is included in API, Session and prepared scopes in both modes. Every artifact includes native, cuBLAS and eager Torch on all four dense sizes.

Complete supplemental scope inventory

Implementation	Rows	Within 5% budget	Inconclusive	Regression
cuda-native	32	22	10	0
cuda-cublas	32	22	10	0
torch-cuda-eager	32	13	19	0

The whole verdict is inconclusive, not a complete-gate pass. Its observed work repeats across all four blocks for every candidate. All eight remaining higher framework medians are cuBLAS Jacobi fresh API; the complete per-size, host/device timing and I/M/R inventory below retains them. Session and prepared scope savings must not replace that fresh-call contract.

The study compares adopted-span source ce45ea9cc6d38cea05ef04cff8a321507e9684d5 against launch source 14c37277de16182fdc7f4cfcf9193e0b352791d1. The frozen launch-stage identity is daf34907780c5b6fd366347ff81161904289228e9cf5960e85c0c669887e6466; its CPU binary is unchanged from the adopted span stage. The initial prepared study and supplemental study remain separate, including their repeated prepared configurations. Neither replaces or pools with the other. The prepared study’s Torch regression is retained. Supplemental intervals are unadjusted for the exploratory sequence and multiple studies; preregistration before inspecting launch results does not make the whole sequence independent confirmatory evidence.

Current-source coverage

The initial resource and CPU four-row sources are rejected experiments, not production performance. Production retains the two-row CPU kernel and adopts the narrower borrowed-span ownership consolidation and stable-norm/control launch change. Launch observations describe actual frozen B source 14c37277de16182fdc7f4cfcf9193e0b352791d1. CUDA CSR is measured in device/prepared scope with standard CG and Jacobi PCG. Dense explicit-residual prepared execution has diagnostic captures only, not a framework timing campaign. Unmeasured CSR API/Session scopes and current BLAS/OpenMP scope characterizations remain historical; they do not inherit a win claim from a different source or from the preceding baseline.

All framework medians belong to the recorded frozen Python harness hashes. The later reference-boundary correction makes unusable recurrence squared norms fail before refresh/iteration-limit recovery, matching native priority. It has focused regression coverage but is absent from these timed harnesses. That edge-case fix and later merges do not turn old bytecode into a new measurement, even when normal fixtures retain their numerical work. The final checkout is not silently substituted for an actual compiled source. Source identity and same-spec I/M/R parity establish the boundaries of the evidence, not universal performance equivalence for unmeasured cases.

Method, estimators and environment

All fixed A/B runs use seed 20261017, three warmups, 36 complete Williams rounds and one solve per timed round. Each configuration launches A/B/B/A once, yielding two fresh-process artifact blocks per side. All declared candidates execute every round. Strict estimates pool the 72 recorded round averages per side; the hierarchical percentile bootstrap resamples artifact blocks and then rounds, with 10,000 resamples, seed 20261017 and 95% intervals. The budget is fixed at current/baseline ratio 1.05. Meeting this budget is evidence against a material regression under this policy, not proof of a speedup. Inconclusive is not a passing result. Framework controls receive the same strict analysis and remain visible without drift subtraction.

The characterization baseline instead reports the median of two process medians and their two-value range. It does not use the pooled-72 estimator or bootstrap. Do not mix its point ratios with strict A/B ratios, reuse historical processes as new A/B blocks, or attach a new interval to an old estimate. Cross-framework and cross-scope point comparisons are descriptive; strict comparison requires matching complete specs, input bytes and known I/M/R.

Restarts count actual successfully installed directions after true-residual replacement. Initial setup and failed/converged/last-iteration exits before replacement do not count. The explicit algorithm reports zero. Counts reset on every solve. Official SciPy CG/PCG and SuperLU restart counts remain unknown, never a fabricated zero. Unknown external work prevents a claim of full native numerical-work parity.

The hardware context is Intel Core i7-13700K, WSL2/Linux 6.6.87.2 and RTX 2060 sm75. The measured driver is 610.88, driver API 13030, CUDA runtime 12080, and NVCC 12.8.93. CPU compilation uses GNU 16.0.1; CUDA host compilation uses GNU 13.3.0, with Release -O3 -DNDEBUG and no fast-math. CPU OpenMP uses two threads and logical affinity [0,2]; other modes use one thread and affinity [0]. These logical constraints do not isolate the physical Windows host. GPU clocks, thermal state, Windows display activity and host background load are uncontrolled. Historical driver-591.86 evidence is not pooled with this driver-610.88 campaign.

Actual compiler flags, public runtime libraries, Python/packages, full threadpool inventories and loaded module hashes are audited. Pools are exact within the applicable preset/suite/configuration; CPU/GPU providers and dense/CSR import inventories need not be identical. Native mode selection does not change Torch’s project reference engine. These frozen sources keep the Python harness and comparator identical for each C++-only A/B study.

Lineage and diagnostics

The original characterization stopped after numerical worker 29 succeeded, because its parent audit mishandled deliberately absent Torch metadata in the official solver-strategy cohort. A separate repair reaudited those 29 without rerunning them and executed only positions 30–108 in the original order. The failed original manifest, claims, tools and all raw/import/log evidence remain unchanged. The continuation has a wall-time gap; it is not an uninterrupted campaign. No numerical worker was retried or discarded.

Every A/B campaign has its own frozen plan, claim, source stages, manifest, worker/import/log records and complete saved comparator results. The first ownership trial remains rejected; the post-selection follow-up is a separate exploratory evaluation. CPU and both launch campaigns retain that same lineage discipline. Raw JSON, binaries and Nsight traces remain ignored local evidence; public commands regenerate new observations, not historical times.

All 57 audited baseline/initial/follow-up/launch profile captures are retained below. Each retains its actual compiled revision/tree, driver path/SHA and separate orchestration-runner path/SHA. Resource probes time construction and destruction; solve probes retain their declared scope. Before/after independent true-residual gates retain identical known I/M/R.

CUDA API counts cover the full ten-call capture, including pre-NVTX/inter-call completion fences. They are not solely solver-internal counts. Outer NVTX phase wall times have a separate boundary. API, GPU and NVTX durations can overlap; do not add them into percentages or subtract independently sampled scope medians. Preserve every driver call and distinguish launch API counts from GPU kernel instances. A missing kernel CSV remains unreported; known zero requires the retained no-event export/log evidence.

Validation

Central validation of the merged implementation passes 92 CTest entries in each CPU release/BLAS/OpenMP preset and 127 in the CUDA preset, covering C++ and Python with declared optional-backend skips. CUDA pytest reports 737 passed and nine skipped; those skips concern unavailable BLAS/OpenMP or non-CUDA requirements, rather than skipping required GPU tests. Compute Sanitizer memcheck covers all 35 GPU tests with zero errors and zero leaked bytes; synccheck also passes. Relocated native CPU, BLAS and CUDA consumers pass, with the CUDA consumer requiring actual device execution. These checks establish tested correctness and ownership behavior, not a performance gate. Python, typing, documentation and distribution validation use the workflows described in Build, test and document.

Public reproduction

Follow Build, test and document, Installation and CUDA for CPython 3.12, the local environments, pinned requirements and project-local CUDA 12.8. Use the exact public measured sources in separate checkouts and build directories. Do not silently substitute the final merge head. The source identities above and complete tables name the actual measured versions. Ignored orchestration helpers are not a required benchmark dependency.

The following is the public CUDA eight-configuration ownership matrix. It keeps all three scopes and both native execution modes explicit; it does not compare different scopes with a strict gate. Set an absolute PYTHON and local CUDA_ROOT consistently when running from different source checkouts.

PROJECT="${PROJECT:-$PWD}"
for preconditioner in none jacobi; do
    if [ "$preconditioner" = none ]; then
        scopes=api
    else
        scopes='api session prepared'
    fi
    for scope in $scopes; do
        for mode in host device; do
            taskset -c 0 env CUBLAS_WORKSPACE_CONFIG=:4096:8 \
                make pyprof PRESET=cuda-release \
                PYTHON="$PROJECT/.venv-cuda/bin/python" \
                CUDA_ROOT="$PROJECT/.deps/cuda-12.8" \
                PYPROF_SUITE=cuda-dense-f64-v1 PYPROF_THREADS=1 \
                PYPROF_SCOPE="$scope" PYPROF_ROUNDS=36 \
                PYPROF_REPETITIONS=1 \
                PYPROF_OUTPUT="build/reproduce/cuda-${preconditioner}-${scope}-${mode}.json" \
                PYPROF_ARGS="--device cuda --cuda-device 0 --cuda-execution-mode $mode --implementations cuda-native,cuda-cublas,torch-cuda-eager --preconditioner $preconditioner --algorithm standard --fused --seed 20261017 --warmups 3 --rtol 1e-8 --atol 1e-12 --maxiter 2000 --residual-refresh 50"
        done
    done
done

This loop makes one fresh observation per configuration. To reproduce ABBA, run the same complete configuration from sources A/B/B/A with four distinct output paths, retaining both A files and both B files. Never overwrite a previous artifact. Compare corresponding configurations with the public scripts/compare_performance.py from the matching frozen current source, --max-regression 0.05 --bootstrap-samples 10000 --seed 20261017. Do not pool the rejected and selected ownership variants. Keep both selected intervals and the original failure when reporting the exploration.

For the completed CPU matrix, run the following separately at the exact A and rejected B sources. It creates one process observation per configuration; use distinct paths in the same A/B/B/A order for a strict reproduction.

PROJECT="${PROJECT:-$PWD}"
for preset in cpu-release cpu-blas cpu-openmp; do
    case "$preset" in
        cpu-release)
            fusions='--fused --no-fused'
            candidates=native,numpy-out,torch-eager-out
            threads=1
            affinity=0
            ;;
        cpu-blas)
            fusions=--fused
            candidates=native,native-blas,numpy-out,torch-eager-out
            threads=1
            affinity=0
            ;;
        cpu-openmp)
            fusions=--fused
            candidates=native,numpy-out,torch-eager-out
            threads=2
            affinity=0,2
            ;;
    esac
    for preconditioner in none jacobi; do
        for scope in api prepared; do
            for fusion in $fusions; do
                taskset -c "$affinity" make pyprof PRESET="$preset" \
                    PYTHON="$PROJECT/.venv/bin/python" \
                    PYPROF_SUITE=cpu-dense-f64-v1 \
                    PYPROF_THREADS="$threads" PYPROF_SCOPE="$scope" \
                    PYPROF_ROUNDS=36 PYPROF_REPETITIONS=1 \
                    PYPROF_OUTPUT="build/reproduce/${preset}-${preconditioner}-${scope}-${fusion#--}.json" \
                    PYPROF_ARGS="--implementations $candidates --preconditioner $preconditioner --algorithm standard $fusion --seed 20261017 --warmups 3 --rtol 1e-8 --atol 1e-12 --maxiter 2000 --residual-refresh 50"
            done
        done
    done
done

Set PROJECT once to the absolute checkout holding the pinned environments and toolkit. Keep it fixed when changing between measured source checkouts; the fallback assignment above retains an existing value. Build each exact source in its own directory, and inspect the recorded compiler/runtime and loaded extension identity before comparing new artifacts.

The prepared/device launch matrix is dense/CSR times none/Jacobi:

for suite in cuda-dense-f64-v1 cuda-csr-f64-v1; do
    if [ "$suite" = cuda-dense-f64-v1 ]; then
        candidates=cuda-native,cuda-cublas,torch-cuda-eager
    else
        candidates=cuda-native,torch-cuda-eager
    fi
    for preconditioner in none jacobi; do
        taskset -c 0 env CUBLAS_WORKSPACE_CONFIG=:4096:8 \
            make pyprof PRESET=cuda-release \
            PYTHON="$PROJECT/.venv-cuda/bin/python" \
            CUDA_ROOT="$PROJECT/.deps/cuda-12.8" \
            PYPROF_SUITE="$suite" PYPROF_THREADS=1 \
            PYPROF_SCOPE=prepared PYPROF_ROUNDS=36 \
            PYPROF_REPETITIONS=1 \
            PYPROF_OUTPUT="build/reproduce/launch-${suite}-${preconditioner}-prepared-device.json" \
            PYPROF_ARGS="--device cuda --cuda-device 0 --cuda-execution-mode device --implementations $candidates --preconditioner $preconditioner --algorithm standard --fused --seed 20261017 --warmups 3 --rtol 1e-8 --atol 1e-12 --maxiter 2000 --residual-refresh 50"
    done
done

Run this matrix independently at the recorded adopted-span and launch sources in A/B/B/A order, retaining four distinct files per configuration. The dense eight-configuration loop above also describes the preregistered supplemental study when run at these sources. It is a separate campaign; do not combine its prepared observations with this four-configuration study. Keep independent processes separate rather than treating 36 rounds as 36 independent runs.

CUDA API allocates fresh output/workspace and optional Jacobi per call. Session reuses iterative execution resources but creates fresh output and optional Jacobi per call. Prepared reuses output/workspace/operator/Jacobi; every call resets its initial guess. Torch uses matching scope policies with its documented caching allocator. All CUDA inputs are resident, and input uploads and independent correctness downloads are outside solve timing. Dispatch, required scalar/status synchronization and final completion stay inside. None of these scopes is host-to-host end-to-end latency.

For the explicit dense prepared diagnostic, use an existing Nsight Systems installation and the selected source’s staged package:

mkdir -p build/profiles
nsys profile --capture-range=cudaProfilerApi --capture-range-end=stop \
    --trace=cuda,cublas,nvtx --sample=none --cpuctxsw=none \
    --output build/profiles/native-explicit-device-dense1024 \
    taskset -c 0 env PYTHONPATH="$PWD/build/cuda-release/python" \
    CUBLAS_WORKSPACE_CONFIG=:4096:8 \
    "$PROJECT/.venv-cuda/bin/python" scripts/profile_solve.py \
    --suite cuda-dense-f64-v1 --case dense-1024 \
    --implementation cuda-native --device cuda --cuda-device 0 \
    --scope prepared --probe solve --cuda-execution-mode device \
    --algorithm explicit --preconditioner none --fused --threads 1 \
    --seed 20261017 --rtol 1e-8 --atol 1e-12 --maxiter 2000 \
    --residual-refresh 50 --warmups 3 --repetitions 10 --cuda-profiler \
    --output build/profiles/native-explicit-device-dense1024.json

This capture is diagnostic only. Nsight availability and WSL permissions are external prerequisites. It includes full-capture completion fences and cannot be combined with unprofiled scores or used as an additive phase breakdown.

CPU component diagnostic appendix — rejected four-row variant

All 84 observations / 42 matched pairs are retained, including higher candidate medians and every long-double oracle gate. The four-row candidate was rejected by the separate complete CG/PCG A/B campaign; production restored the original two-row kernel. These component observations do not replace that decision.

Each side has one fresh process per fixed operation, three warmups, 36 samples and one operation per sample, native CPU mode, one thread and affinity [0]. The entire baseline pass preceded the candidate pass. No ABBA interleaving or independent process repetitions were collected for these components. Ranges below are sample minima/maxima, not confidence intervals. No strict regression or winner is inferred.

Official sizes64/256/1024/2048 use the actual seed20261017 dense fixtures, whose full A/b/x0/truth hashes match the saved baseline. The fixed diagnostic vector is b; it does not represent every CG direction. Integer sizes129/130/131 have padded row stride n+5: A[i,j]=((i+3j)%7)-3, vector[i]=(i%5)-2, b[i]=(i%7)-3, with NaN padding excluded from arithmetic. Both sides reuse identical bytes.

Probe file reads, allocations, finite scans, independent oracle validation, warmups and output are outside operation timing. Public backend metadata and alias checks remain inside the timed call. The source-pinned C++ probe checks every vector entry and applicable scalar against scalar long-double arithmetic and propagated FP64 bounds before and after measurement. Successful exit proves both checks; the stored error/bound fields below are the post-measurement checks. No oracle is recomputed by this stdlib renderer.

These fixed-vector wall times are diagnostics, not framework scores, hardware bandwidth measurements or additive whole-CG phase attribution. Do not add operation medians or subtract separate scope timings. WSL affinity does not isolate physical host load, clocks or thermals.

All matched diagnostic timings

Case	Stride	Fused	Operation	A median ms	A min–max ms	B median ms	B min–max ms	B/A
dense-64	64	False	apply	0.000361000	0.000359000–0.000452000	0.000389000	0.000381000–0.048573000	1.077562
dense-64	64	False	apply_and_dot	0.000402000	0.000401000–0.000542000	0.000379000	0.000375000–0.000547000	0.942786
dense-64	64	False	residual_squared_norm	0.000425000	0.000422000–0.011278000	0.000400000	0.000394000–0.000524000	0.941176
dense-64	64	True	apply	0.000360500	0.000358000–0.011217000	0.000339000	0.000333000–0.010987000	0.940361
dense-64	64	True	apply_and_dot	0.000368000	0.000366000–0.000470000	0.000413000	0.000408000–0.000517000	1.122283
dense-64	64	True	residual_squared_norm	0.000407000	0.000399000–0.001641000	0.000360000	0.000357000–0.000451000	0.884521
dense-256	256	False	apply	0.004889500	0.004800000–0.022465000	0.004773000	0.004605000–0.029226000	0.976173
dense-256	256	False	apply_and_dot	0.004941500	0.004891000–0.035683000	0.004737000	0.004700000–0.063968000	0.958616
dense-256	256	False	residual_squared_norm	0.005623000	0.004953000–0.034259000	0.004957500	0.004745000–0.007224000	0.881647
dense-256	256	True	apply	0.004820000	0.004792000–0.018282000	0.004624500	0.004602000–0.025039000	0.959440
dense-256	256	True	apply_and_dot	0.004906000	0.004879000–0.027261000	0.004663500	0.004628000–0.026027000	0.950571
dense-256	256	True	residual_squared_norm	0.004941000	0.004899000–0.015916000	0.004665000	0.004632000–0.004826000	0.944141
dense-1024	1024	False	apply	0.161656500	0.137956000–0.182225000	0.152456000	0.129621000–0.196575000	0.943086
dense-1024	1024	False	apply_and_dot	0.173882000	0.139432000–0.233091000	0.170992000	0.137561000–0.270517000	0.983380
dense-1024	1024	False	residual_squared_norm	0.190878500	0.144089000–0.263680000	0.179298500	0.149665000–1.051264000	0.939333
dense-1024	1024	True	apply	0.156799000	0.139120000–0.190669000	0.136624500	0.134449000–0.217335000	0.871335
dense-1024	1024	True	apply_and_dot	0.181145500	0.140580000–0.230790000	0.199004000	0.140301000–0.449935000	1.098586
dense-1024	1024	True	residual_squared_norm	0.183166500	0.160023000–0.225488000	0.152435000	0.135637000–0.212076000	0.832221
dense-2048	2048	False	apply	1.916531000	0.955828000–4.313184000	1.671557500	0.896646000–3.751114000	0.872179
dense-2048	2048	False	apply_and_dot	1.619994500	1.115293000–3.410844000	1.513830000	0.797859000–4.204348000	0.934466
dense-2048	2048	False	residual_squared_norm	1.428512000	1.096583000–3.487011000	1.733059000	0.827660000–3.478580000	1.213192
dense-2048	2048	True	apply	1.380356500	0.859954000–3.118661000	1.386002500	0.732642000–3.352031000	1.004090
dense-2048	2048	True	apply_and_dot	1.826700500	0.915059000–4.993389000	1.220367000	0.861965000–2.941810000	0.668072
dense-2048	2048	True	residual_squared_norm	1.506064500	1.098146000–3.521191000	1.667759000	1.157505000–3.443238000	1.107362
integer-tail-129	134	False	apply	0.001307000	0.001297000–0.012591000	0.001496500	0.001295000–0.023596000	1.144989
integer-tail-129	134	False	apply_and_dot	0.001370000	0.001360000–0.007264000	0.001558500	0.001551000–0.001696000	1.137591
integer-tail-129	134	False	residual_squared_norm	0.001394000	0.001385000–0.001478000	0.001433000	0.001383000–0.012618000	1.027977
integer-tail-129	134	True	apply	0.001514000	0.001498000–0.017745000	0.001265500	0.001252000–0.038629000	0.835865
integer-tail-129	134	True	apply_and_dot	0.001586500	0.001569000–0.044206000	0.001363000	0.001280000–0.013667000	0.859124
integer-tail-129	134	True	residual_squared_norm	0.001348000	0.001341000–0.011256000	0.001334000	0.001276000–0.012161000	0.989614
integer-tail-130	135	False	apply	0.001418000	0.001406000–0.016137000	0.001356000	0.001346000–0.001452000	0.956276
integer-tail-130	135	False	apply_and_dot	0.001738500	0.001729000–0.002060000	0.001431000	0.001413000–0.013826000	0.823123
integer-tail-130	135	False	residual_squared_norm	0.001483000	0.001431000–0.012640000	0.001699000	0.001688000–0.001815000	1.145651
integer-tail-130	135	True	apply	0.001636500	0.001383000–0.097503000	0.001356500	0.001340000–0.036614000	0.828903
integer-tail-130	135	True	apply_and_dot	0.001420500	0.001407000–0.012487000	0.001373500	0.001358000–0.001466000	0.966913
integer-tail-130	135	True	residual_squared_norm	0.001424500	0.001408000–0.040305000	0.001389000	0.001372000–0.014208000	0.975079
integer-tail-131	136	False	apply	0.001580500	0.001563000–0.001677000	0.001545500	0.001524000–0.013927000	0.977855
integer-tail-131	136	False	apply_and_dot	0.001413000	0.001388000–0.011180000	0.001394000	0.001366000–0.102640000	0.986553
integer-tail-131	136	False	residual_squared_norm	0.001428500	0.001407000–0.011633000	0.001646500	0.001633000–0.013531000	1.152608
integer-tail-131	136	True	apply	0.001333500	0.001306000–0.012329000	0.001312000	0.001301000–0.011341000	0.983877
integer-tail-131	136	True	apply_and_dot	0.001420000	0.001410000–0.027486000	0.001565500	0.001555000–0.015704000	1.102465
integer-tail-131	136	True	residual_squared_norm	0.001480500	0.001462000–0.011378000	0.001353000	0.001343000–0.001451000	0.913880

All 84 finite post-measurement oracle records

Side	Run	Long-double bits	Vector max error	Vector max bound	Scalar error	Scalar bound	Finite
baseline	dense-64-fused0-apply	64	1.0196288258157438e-12	2.0618619623662891e-10	0	0	True
baseline	dense-64-fused0-apply_and_dot	64	1.0196288258157438e-12	2.0618619623662891e-10	1.6279955161735415e-10	2.0302254959166928e-07	True
baseline	dense-64-fused0-residual_squared_norm	64	7.5539574595495651e-13	2.0908316183909403e-10	8.7311491370201111e-08	2.2454420314339871e-05	True
baseline	dense-64-fused1-apply	64	1.0196288258157438e-12	2.0618619623662891e-10	0	0	True
baseline	dense-64-fused1-apply_and_dot	64	1.0196288258157438e-12	2.0618619623662891e-10	1.6279955161735415e-10	2.0302254959166928e-07	True
baseline	dense-64-fused1-residual_squared_norm	64	7.5539574595495651e-13	2.0908316183909403e-10	8.7311491370201111e-08	2.2454420314339871e-05	True
baseline	dense-256-fused0-apply	64	2.1898038937706588e-12	8.7769074042863557e-10	0	0	True
baseline	dense-256-fused0-apply_and_dot	64	2.1898038937706588e-12	8.7769074042863557e-10	2.2537278709933162e-09	1.9603692385893142e-06	True
baseline	dense-256-fused0-residual_squared_norm	64	2.2466473126314668e-12	8.8104926014635646e-10	2.046581357717514e-07	0.00021547992368137688	True
baseline	dense-256-fused1-apply	64	2.1898038937706588e-12	8.7769074042863557e-10	0	0	True
baseline	dense-256-fused1-apply_and_dot	64	2.1898038937706588e-12	8.7769074042863557e-10	2.2537278709933162e-09	1.9603692385893142e-06	True
baseline	dense-256-fused1-residual_squared_norm	64	2.2466473126314668e-12	8.8104926014635646e-10	2.046581357717514e-07	0.00021547992368137688	True
baseline	dense-1024-fused0-apply	64	4.0416558988454199e-12	4.1708565652771602e-09	0	0	True
baseline	dense-1024-fused0-apply_and_dot	64	4.0416558988454199e-12	4.1708565652771602e-09	3.328023012727499e-08	3.2850246876764954e-05	True
baseline	dense-1024-fused0-residual_squared_norm	64	4.1269210271366319e-12	4.1749363516710137e-09	1.2945383787155151e-07	0.0036328106130226736	True
baseline	dense-1024-fused1-apply	64	4.0416558988454199e-12	4.1708565652771602e-09	0	0	True
baseline	dense-1024-fused1-apply_and_dot	64	4.0416558988454199e-12	4.1708565652771602e-09	3.328023012727499e-08	3.2850246876764954e-05	True
baseline	dense-1024-fused1-residual_squared_norm	64	4.1269210271366319e-12	4.1749363516710137e-09	1.2945383787155151e-07	0.0036328106130226736	True
baseline	dense-2048-fused0-apply	64	6.0809135504769074e-12	1.1096243113407423e-08	0	0	True
baseline	dense-2048-fused0-apply_and_dot	64	6.0809135504769074e-12	1.1096243113407423e-08	2.0736479200422764e-08	0.00013626509015883027	True
baseline	dense-2048-fused0-residual_squared_norm	64	5.9530158580400894e-12	1.1101691666334374e-08	5.0524249672889709e-06	0.015375788579093783	True
baseline	dense-2048-fused1-apply	64	6.0809135504769074e-12	1.1096243113407423e-08	0	0	True
baseline	dense-2048-fused1-apply_and_dot	64	6.0809135504769074e-12	1.1096243113407423e-08	2.0736479200422764e-08	0.00013626509015883027	True
baseline	dense-2048-fused1-residual_squared_norm	64	5.9530158580400894e-12	1.1101691666334374e-08	5.0524249672889709e-06	0.015375788579093783	True
baseline	integer-tail-129-fused0-apply	64	0	8.0917494926782365e-12	0	0	True
baseline	integer-tail-129-fused0-apply_and_dot	64	0	8.0917494926782365e-12	0	1.2798349047216872e-09	True
baseline	integer-tail-129-fused0-residual_squared_norm	64	0	8.0950801617521136e-12	0	1.7782194827740646e-08	True
baseline	integer-tail-129-fused1-apply	64	0	8.0917494926782365e-12	0	0	True
baseline	integer-tail-129-fused1-apply_and_dot	64	0	8.0917494926782365e-12	0	1.2798349047216872e-09	True
baseline	integer-tail-129-fused1-residual_squared_norm	64	0	8.0950801617521136e-12	0	1.7782194827740646e-08	True
baseline	integer-tail-130-fused0-apply	64	0	8.2427398240272643e-12	0	0	True
baseline	integer-tail-130-fused0-apply_and_dot	64	0	8.2427398240272643e-12	0	1.3138130583457381e-09	True
baseline	integer-tail-130-fused0-residual_squared_norm	64	0	8.2440720916568161e-12	0	1.4466784215026626e-08	True
baseline	integer-tail-130-fused1-apply	64	0	8.2427398240272643e-12	0	0	True
baseline	integer-tail-130-fused1-apply_and_dot	64	0	8.2427398240272643e-12	0	1.3138130583457381e-09	True
baseline	integer-tail-130-fused1-residual_squared_norm	64	0	8.2440720916568161e-12	0	1.4466784215026626e-08	True
baseline	integer-tail-131-fused0-apply	64	0	8.3950624230058422e-12	0	0	True
baseline	integer-tail-131-fused0-apply_and_dot	64	0	8.3950624230058422e-12	0	1.3516050501039815e-09	True
baseline	integer-tail-131-fused0-residual_squared_norm	64	0	8.3979490028698708e-12	0	1.3550524036186618e-08	True
baseline	integer-tail-131-fused1-apply	64	0	8.3950624230058422e-12	0	0	True
baseline	integer-tail-131-fused1-apply_and_dot	64	0	8.3950624230058422e-12	0	1.3516050501039815e-09	True
baseline	integer-tail-131-fused1-residual_squared_norm	64	0	8.3979490028698708e-12	0	1.3550524036186618e-08	True
candidate	dense-64-fused0-apply	64	1.0196288258157438e-12	2.0618619623662891e-10	0	0	True
candidate	dense-64-fused0-apply_and_dot	64	1.0196288258157438e-12	2.0618619623662891e-10	1.6279955161735415e-10	2.0302254959166928e-07	True
candidate	dense-64-fused0-residual_squared_norm	64	7.5539574595495651e-13	2.0908316183909403e-10	8.7311491370201111e-08	2.2454420314339871e-05	True
candidate	dense-64-fused1-apply	64	1.0196288258157438e-12	2.0618619623662891e-10	0	0	True
candidate	dense-64-fused1-apply_and_dot	64	1.0196288258157438e-12	2.0618619623662891e-10	1.6279955161735415e-10	2.0302254959166928e-07	True
candidate	dense-64-fused1-residual_squared_norm	64	7.5539574595495651e-13	2.0908316183909403e-10	8.7311491370201111e-08	2.2454420314339871e-05	True
candidate	dense-256-fused0-apply	64	2.1898038937706588e-12	8.7769074042863557e-10	0	0	True
candidate	dense-256-fused0-apply_and_dot	64	2.1898038937706588e-12	8.7769074042863557e-10	2.2537278709933162e-09	1.9603692385893142e-06	True
candidate	dense-256-fused0-residual_squared_norm	64	2.2466473126314668e-12	8.8104926014635646e-10	2.046581357717514e-07	0.00021547992368137688	True
candidate	dense-256-fused1-apply	64	2.1898038937706588e-12	8.7769074042863557e-10	0	0	True
candidate	dense-256-fused1-apply_and_dot	64	2.1898038937706588e-12	8.7769074042863557e-10	2.2537278709933162e-09	1.9603692385893142e-06	True
candidate	dense-256-fused1-residual_squared_norm	64	2.2466473126314668e-12	8.8104926014635646e-10	2.046581357717514e-07	0.00021547992368137688	True
candidate	dense-1024-fused0-apply	64	4.0416558988454199e-12	4.1708565652771602e-09	0	0	True
candidate	dense-1024-fused0-apply_and_dot	64	4.0416558988454199e-12	4.1708565652771602e-09	3.328023012727499e-08	3.2850246876764954e-05	True
candidate	dense-1024-fused0-residual_squared_norm	64	4.1269210271366319e-12	4.1749363516710137e-09	1.2945383787155151e-07	0.0036328106130226736	True
candidate	dense-1024-fused1-apply	64	4.0416558988454199e-12	4.1708565652771602e-09	0	0	True
candidate	dense-1024-fused1-apply_and_dot	64	4.0416558988454199e-12	4.1708565652771602e-09	3.328023012727499e-08	3.2850246876764954e-05	True
candidate	dense-1024-fused1-residual_squared_norm	64	4.1269210271366319e-12	4.1749363516710137e-09	1.2945383787155151e-07	0.0036328106130226736	True
candidate	dense-2048-fused0-apply	64	6.0809135504769074e-12	1.1096243113407423e-08	0	0	True
candidate	dense-2048-fused0-apply_and_dot	64	6.0809135504769074e-12	1.1096243113407423e-08	2.0736479200422764e-08	0.00013626509015883027	True
candidate	dense-2048-fused0-residual_squared_norm	64	5.9530158580400894e-12	1.1101691666334374e-08	5.0524249672889709e-06	0.015375788579093783	True
candidate	dense-2048-fused1-apply	64	6.0809135504769074e-12	1.1096243113407423e-08	0	0	True
candidate	dense-2048-fused1-apply_and_dot	64	6.0809135504769074e-12	1.1096243113407423e-08	2.0736479200422764e-08	0.00013626509015883027	True
candidate	dense-2048-fused1-residual_squared_norm	64	5.9530158580400894e-12	1.1101691666334374e-08	5.0524249672889709e-06	0.015375788579093783	True
candidate	integer-tail-129-fused0-apply	64	0	8.0917494926782365e-12	0	0	True
candidate	integer-tail-129-fused0-apply_and_dot	64	0	8.0917494926782365e-12	0	1.2798349047216872e-09	True
candidate	integer-tail-129-fused0-residual_squared_norm	64	0	8.0950801617521136e-12	0	1.7782194827740646e-08	True
candidate	integer-tail-129-fused1-apply	64	0	8.0917494926782365e-12	0	0	True
candidate	integer-tail-129-fused1-apply_and_dot	64	0	8.0917494926782365e-12	0	1.2798349047216872e-09	True
candidate	integer-tail-129-fused1-residual_squared_norm	64	0	8.0950801617521136e-12	0	1.7782194827740646e-08	True
candidate	integer-tail-130-fused0-apply	64	0	8.2427398240272643e-12	0	0	True
candidate	integer-tail-130-fused0-apply_and_dot	64	0	8.2427398240272643e-12	0	1.3138130583457381e-09	True
candidate	integer-tail-130-fused0-residual_squared_norm	64	0	8.2440720916568161e-12	0	1.4466784215026626e-08	True
candidate	integer-tail-130-fused1-apply	64	0	8.2427398240272643e-12	0	0	True
candidate	integer-tail-130-fused1-apply_and_dot	64	0	8.2427398240272643e-12	0	1.3138130583457381e-09	True
candidate	integer-tail-130-fused1-residual_squared_norm	64	0	8.2440720916568161e-12	0	1.4466784215026626e-08	True
candidate	integer-tail-131-fused0-apply	64	0	8.3950624230058422e-12	0	0	True
candidate	integer-tail-131-fused0-apply_and_dot	64	0	8.3950624230058422e-12	0	1.3516050501039815e-09	True
candidate	integer-tail-131-fused0-residual_squared_norm	64	0	8.3979490028698708e-12	0	1.3550524036186618e-08	True
candidate	integer-tail-131-fused1-apply	64	0	8.3950624230058422e-12	0	0	True
candidate	integer-tail-131-fused1-apply_and_dot	64	0	8.3950624230058422e-12	0	1.3516050501039815e-09	True
candidate	integer-tail-131-fused1-residual_squared_norm	64	0	8.3979490028698708e-12	0	1.3550524036186618e-08	True

Baseline compiled revision: 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7.

Actual portable compiler argv: /usr/bin/c++ -O3 -DNDEBUG -std=c++20 -fPIC -Wall -Wextra -Wpedantic -I/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/cpu-four-row/headers-baseline /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/cpu-four-row/profile_gemv.cpp /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/baseline-stages/cpu-release/libcg_core.a -o /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/cpu-four-row/baseline-components/probe.

Candidate compiled revision: 2d81b3d1409b0b428572cbf26fc7418e03bd270c.

Actual portable compiler argv: /usr/bin/c++ -O3 -DNDEBUG -std=c++20 -fPIC -Wall -Wextra -Wpedantic -I/home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/cpu-four-row/headers-baseline /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/cpu-four-row/profile_gemv.cpp /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/cpu-quad-stages/cpu-release/libcg_core.a -o /home/threemonth03/Downloads/Conjugate-Gradient-Method-library/build/restart-optimization-20261017/cpu-four-row/candidate-components/probe.

Source, archive, executable and public-header identities

Role	Path	SHA256
renderer	build/restart-optimization-20261017/render_components_report.py	31a10e28d5d7422142542a356c646863a0dc13934797375e7e8cf6d13d6e356c
baseline plan	build/restart-optimization-20261017/cpu-four-row/baseline-components/plan.json	09f15a68b80705b942c57ed4633ebd44b2d8e28e4972c4df90ca09be7f11974b
baseline manifest	build/restart-optimization-20261017/cpu-four-row/baseline-components/manifest.json	5b10527756a3cc7c6f204c5f787a1d325d29877b92835bbb0e196e972541f069
baseline archive	build/restart-optimization-20261017/baseline-stages/cpu-release/libcg_core.a	e47471ca51084098a11eee8f0b45bb1b06b8a78b27f0b868408d8e709de6f3da
baseline claim	build/restart-optimization-20261017/cpu-four-row/baseline-components/execution-claim.json	fdb8d82238597bffaef0539d280c0713cbe3c2e0bb2de4e82c1d2c0eab0cdde9
baseline export	build/restart-optimization-20261017/cpu-four-row/baseline-components/exports.json	6552201a8072542783850a48af86aebe5f13a869c38fded7c0db299ec2f4aea2
baseline binary	build/restart-optimization-20261017/cpu-four-row/baseline-components/probe	2553ed19dfe90515e2684d8fec9dedeffa34e45c307f23ca4d7db7eaf608ee3a
baseline log	build/restart-optimization-20261017/cpu-four-row/baseline-components/compile.log	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline header 0	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/backend/cpu/backend.hpp	265479a84eb2313d23b821ecb04048a93a725370a6278b102a4caf0414a05409
baseline header 1	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/backend/residual_sign.hpp	6d10cde48e611fb016f0a083e2ba6a59758a4b5ca1875e7948f2c72db4c2c3d0
baseline header 2	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/matrix/dense_matrix_view.hpp	0869ab54e368c0dd4dcf91e2a877d48a0490147df5fe27af5052e425e29c282d
baseline header 3	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/storage/vector_view.hpp	dc49a8ef869421d50c8184a4b82dd750fe4d954b67252900edceb0bdac93d889
baseline compiler	/usr/bin/c++	546ddb2e5c2aadbdbcf2aaab30b8ac4f72abb8bb8f5fd51dc4bf6864a29491d7
baseline interpreter	.venv/bin/python	e50d468e8b0adfb05733f5b87b3cff34829c4a8c1aea50c865aa8bdfe4bb150f
baseline stage identity	build/restart-optimization-20261017/baseline-stages/cpu-release/identity.json	82f66c8a54d2c5a029ab6b62941d37650c1d2f07441ad6dd5e852b052e567b38
baseline source snapshot	build/restart-optimization-20261017/baseline-stages/cpu-release/source-files.json	ec6553bb49b270edfabdad0b64f6090b3bc614def58cdce5cb02836d837f48dd
candidate plan	build/restart-optimization-20261017/cpu-four-row/candidate-components/plan.json	31ca10eb08575f365c46119c3a567edec1bf372dc3b05d7cc039147ef31777fb
candidate manifest	build/restart-optimization-20261017/cpu-four-row/candidate-components/manifest.json	6aea12827f7e3be167b27fdc528ae14248a9b96593a188d0b792ba5e25083e2a
candidate archive	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/libcg_core.a	3ec4a15d19cba01a53955be4eb16e4ba8b600346257b15bb7b207d31f6be35dd
candidate claim	build/restart-optimization-20261017/cpu-four-row/candidate-components/execution-claim.json	c61d4f2f67db39e80448a2d3536862de96352101b6fcf988cbb8e880a132a0a7
candidate export	build/restart-optimization-20261017/cpu-four-row/baseline-components/exports.json	6552201a8072542783850a48af86aebe5f13a869c38fded7c0db299ec2f4aea2
candidate binary	build/restart-optimization-20261017/cpu-four-row/candidate-components/probe	bcf1910d426b94738f4ee566b40a423d1fc1fb049c535736dd64cfa1095697ff
candidate log	build/restart-optimization-20261017/cpu-four-row/candidate-components/compile.log	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate header 0	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/backend/cpu/backend.hpp	265479a84eb2313d23b821ecb04048a93a725370a6278b102a4caf0414a05409
candidate header 1	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/backend/residual_sign.hpp	6d10cde48e611fb016f0a083e2ba6a59758a4b5ca1875e7948f2c72db4c2c3d0
candidate header 2	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/matrix/dense_matrix_view.hpp	0869ab54e368c0dd4dcf91e2a877d48a0490147df5fe27af5052e425e29c282d
candidate header 3	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/storage/vector_view.hpp	dc49a8ef869421d50c8184a4b82dd750fe4d954b67252900edceb0bdac93d889
candidate compiler	/usr/bin/c++	546ddb2e5c2aadbdbcf2aaab30b8ac4f72abb8bb8f5fd51dc4bf6864a29491d7
candidate interpreter	.venv/bin/python	e50d468e8b0adfb05733f5b87b3cff34829c4a8c1aea50c865aa8bdfe4bb150f
candidate stage identity	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/identity.json	bd3a934ae3d0b8e43055634a6ba22fbc8870a405bce2a2f4082ef7aecdf2c5da
candidate source snapshot	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/source-files.json	8b5b51b4ddebd5428c2d95d2516e0f6c94cb1c9dd42e61a31c281f650254517c
frozen helper/oracle	build/restart-optimization-20261017/run_baseline.py	b368f781a3aa6516e366770b2d1f8c57df04da0f21f1c0274331c913ecc1ba51
frozen helper/oracle	build/restart-optimization-20261017/protocol.py	8814adcd9e90fa800e3654c248b1554fbc19e74a4a773dfcbe61e60187afdc0c
frozen helper/oracle	build/restart-optimization-20261017/worker.py	038fa8530e6ed876f52dcea09497e804840f74ebabb008d4adaa99db24f9456c
frozen helper/oracle	build/restart-optimization-20261017/freeze_stage.py	ea90c787125e9bb2f7977f40a25e5d4fe4e94afbe2a237b2dc3893ca678b68ab
frozen helper/oracle	build/restart-optimization-20261017/README.md	3f308246cf5540f79b24b43bd492cd57b35b6fe7e1f8c1e7a2977f6165cbea14
frozen helper/oracle	build/performance-optimization/ab_runner.py	54668d5459fec4041f125d21d38a6fc73074c5b7dedc539a765a8bde254c7ace
frozen helper/oracle	build/performance-optimization/run_baseline.py	213afaa21053d4f65dea506158dc65d60cc4587033bc388d0efccb362e76233f
frozen helper/oracle	build/restart-optimization-20261017/baseline-audit-repair/run_baseline.py	7173f06e038d6de74596ec06c506136705d3be7a8a346f6d3cb9d235c77d7cff
frozen helper/oracle	build/restart-optimization-20261017/baseline-audit-repair/protocol.py	5276f6ddface2b5011df30c4feea845e4ccd91e90b00d9d8c0bf092315f9dced
frozen helper/oracle	build/restart-optimization-20261017/baseline-audit-repair/worker.py	038fa8530e6ed876f52dcea09497e804840f74ebabb008d4adaa99db24f9456c
frozen helper/oracle	build/restart-optimization-20261017/baseline-audit-repair/freeze_stage.py	ea90c787125e9bb2f7977f40a25e5d4fe4e94afbe2a237b2dc3893ca678b68ab
frozen helper/oracle	build/restart-optimization-20261017/baseline-audit-repair/README.md	7ba0ff2cc21ce90529945aa3c84b13b9a40a4a8d90ab00ad5de9a18e73e2cc55
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/run_components.py	d8f711b1910205a8da2ba6e170bd3996c797fa1b2e5d6d3648031e20ea6ca2bd
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/component_common.py	e4acc393194643a5f2538e8354b544dcdb7f6270442d381d34d1da939e4be9ee
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/component_worker.py	6a52b987fa8521c8ca9067703440a248ecc6390c2ad491a829c32db055e993e1
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/profile_gemv.cpp	670fa8067d49ad5eb325bcd550768afa8e733934b9292fc009ddd3510255dedc
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/headers-baseline.json	cd96fb1ff29d0dda3077904daf4c508656be3751abd2e9dfc4ab4c48c02a7e4b
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/driver.md	d0eb6675f50952902333f60b1788c1bac2a9108de06ca620ed1023d52e58fdcc
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/backend/cpu/backend.hpp	265479a84eb2313d23b821ecb04048a93a725370a6278b102a4caf0414a05409
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/backend/residual_sign.hpp	6d10cde48e611fb016f0a083e2ba6a59758a4b5ca1875e7948f2c72db4c2c3d0
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/matrix/dense_matrix_view.hpp	0869ab54e368c0dd4dcf91e2a877d48a0490147df5fe27af5052e425e29c282d
frozen helper/oracle	build/restart-optimization-20261017/cpu-four-row/headers-baseline/cg/storage/vector_view.hpp	dc49a8ef869421d50c8184a4b82dd750fe4d954b67252900edceb0bdac93d889

Shared exact input identities

Case	Role	Path	SHA256
dense-64	matrix	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-64.A.f64	2ae2227d37228e9b4f5c5d9060961ed9c7a27c53135f51bcbfcf0c621f04b9f2
dense-64	rhs	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-64.b.f64	06946998cb419a52b0b367fc7b27c4941d1552abaa46d25552b8693f34d81fe4
dense-64	vector	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-64.vector.f64	06946998cb419a52b0b367fc7b27c4941d1552abaa46d25552b8693f34d81fe4
dense-256	matrix	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-256.A.f64	4453bfe9013b7aa2538e1d4e5f1aca18336ec1ce54ddf3d15c6cfa4643abac14
dense-256	rhs	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-256.b.f64	73d3bf719a3bc3266180bfee2a4de8840e9eb2ef8e425277453464134e76fc53
dense-256	vector	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-256.vector.f64	73d3bf719a3bc3266180bfee2a4de8840e9eb2ef8e425277453464134e76fc53
dense-1024	matrix	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-1024.A.f64	ce8175bcf567ab4f16594849c5c48510b7947f249be6c4216f4a5c68998d5bf5
dense-1024	rhs	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-1024.b.f64	1891b597da63bf9d4823dac02fa7633c06cedba3e4468e18fef568e56574a9f4
dense-1024	vector	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-1024.vector.f64	1891b597da63bf9d4823dac02fa7633c06cedba3e4468e18fef568e56574a9f4
dense-2048	matrix	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-2048.A.f64	d97407fb8fab562a10fb8cc40d8dca819aefca248681c12e2f10f5cd304db56d
dense-2048	rhs	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-2048.b.f64	1ac142af51538f14b47212b9ef4235467330dab06c45f80592fbdb7c09018371
dense-2048	vector	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/dense-2048.vector.f64	1ac142af51538f14b47212b9ef4235467330dab06c45f80592fbdb7c09018371
integer-tail-129	matrix	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-129.A.f64	538a2b692d7136266f3d8d289723076a175cf4355995ff85942aa76065dbd39d
integer-tail-129	rhs	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-129.b.f64	db811c05c5cc79a296ec9f7210975e9bb6891ec95836a5bc0cd09675ea68ee84
integer-tail-129	vector	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-129.vector.f64	32b1129fca2a5e1db81e2029494845bf1192c51adce81b47d8b5512f7c82bf39
integer-tail-130	matrix	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-130.A.f64	9766b0037e754753aee00e0437bd4197a7455b2cfad6f52ea011e409a6a3caac
integer-tail-130	rhs	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-130.b.f64	d6f72dd9601107715902f4aa1aa28956d148e0b5bb37e79d49874095cfc50223
integer-tail-130	vector	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-130.vector.f64	759dcf000800c7efc749756c153884b62f5ccdbb8cc2eabbc4bde05b8693ee61
integer-tail-131	matrix	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-131.A.f64	205582384beb5635dffd8e0310ff0087f899466f2c7c6659edcaccbb30a86ee4
integer-tail-131	rhs	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-131.b.f64	2a3e8c826093ad0769ebd0783ba20bc069803b23b2eea206071749cb4c36fca1
integer-tail-131	vector	build/restart-optimization-20261017/cpu-four-row/baseline-components/inputs/integer-tail-131.vector.f64	c8e913873a76341110dc19bf10e00fb3d931ed7961cb51a54a5c064bb2e3c272

All original raw, execution-sidecar and log identities

Side	Run	Raw path	Raw SHA256	Sidecar SHA256	Log SHA256
baseline	dense-64-fused0-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-64-fused0-apply.json	04b3164197bb2f8863117de1e02e32bfcd2c21b25f782b20ce8aa3025a4901c5	20ada4fc625491f6f97e248c8eab4fc7ec7761c48239cfa6ff1d55df3b55f75a	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-64-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-64-fused0-apply_and_dot.json	11098867feb66d67859a5777a7d1a9219e9a6d2e29177f8cbad65fc22306f513	0fde73fbee19f1f83f9512dc019f7026cd1b4ffe5776b1936e9ce2845bc472b0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-64-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-64-fused0-residual_squared_norm.json	d1e6ca038c074de23302ab08cc3aa7c77fdd8ae2b13632835276141ad5021e06	af7a650bcb44d5809ce36eb1db59cc37bdd13b7b5dacd243804dc4268cd2f804	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-64-fused1-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-64-fused1-apply.json	98f5e55c639c86287d6ca92dbf345818a12bd597a8fd2f7ddd609ef839b2c450	65003f8bef8d613a9fbc0093e1231d7402a08fc92f2092342e31c6766d06e666	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-64-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-64-fused1-apply_and_dot.json	1f9767c305b0b8dc8b99672b2ba02f0f25ac20901e7c47271172952fe50b02c6	6cd0bb0bed0967f7500ddbdd4dcf90399d7dee4da4aea5b8d9ed6c6e18715abc	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-64-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-64-fused1-residual_squared_norm.json	b98f628a1da3878b7ab8215fb52a419db0348e9a18a9deaa74a5e7cbc4406178	91195d400c6bfa9b69d586299a22f29a0ec111b6248a3aa88456f8c4abedb7ac	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-256-fused0-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-256-fused0-apply.json	2585e321daa89b5613e9e1ca5f2e3af5cbe9bedacece24c2aa0caa2a0ae7d814	a14c7f3c418b56860bbdf7f35c745c3c627d4e58e9f2f4977025922b68ec823c	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-256-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-256-fused0-apply_and_dot.json	2462f55b47a371f7986b4c6c35d9388614bb9ee9d68f3ad3f8b0af5773bc54ca	a257b4ddca678cc1e4eec804861fa045cfa0871252b2b06dc4e91f7da2188ed7	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-256-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-256-fused0-residual_squared_norm.json	05190c313db8d501c9452532a39e61ff1d0677d77a1a0591fe5ce2d69b34f554	207663390de87826dd8b0af5514974c837d3383a9f179897d0c52f3b8f26fa42	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-256-fused1-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-256-fused1-apply.json	860b0fbac5541378e74d8706ce47bf4d672d4603b4b7834954f373bb638918e5	8698fb1ce933e4f3014b001854cf16d027c7956096d8e56cae08ec5a3c6b78ec	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-256-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-256-fused1-apply_and_dot.json	6cc7e78b9e9881e6d8c4142682933fb016247e1a83a3152466b27a4feb95fcc0	574cea21d7e13854cfed5405245c161a2174530fcd70446411ace33bdfab4c21	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-256-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-256-fused1-residual_squared_norm.json	9466d336e966da0233f450c1c52d862e3bb1c829ac2564b9fc8b3b1e89ae1dab	3e0ddeb22642377656a29ed69ba105e7dd1fa1b3d33c2760f1f6e28c6f398bb9	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-1024-fused0-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-1024-fused0-apply.json	ee5947c0a101a90e5a1b5d3cce7da136841774c3953f27230c9ad08aaa759618	2c56f16e6303ccb5a82c0e3ec3f125c5109d4334fec86485ab63cbd6c8462e67	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-1024-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-1024-fused0-apply_and_dot.json	b0fa8f5e261df4bd08fa5a7823c16c2c3cf6981cdd3f8c06cd205b5744d294fa	74004498a9b27a9936c17174add4fb008045ecc00596bfb879ea720d8e049971	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-1024-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-1024-fused0-residual_squared_norm.json	5737d4501cefbcabef919f45f03516fedd706934650a02bae89d908a279d09b2	494ae78be7924eb660c074c4949fd7fadd42b4c241db289c34ff707a2f554dd8	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-1024-fused1-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-1024-fused1-apply.json	84a41ef638f192df389d8c545307df7de8d36a92c2c7b24d2723ea65d3891531	6e4042cdf6fb56e39e0f7941f95d3a19fdf365d81ca511551d117223f17f1c06	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-1024-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-1024-fused1-apply_and_dot.json	a82639605949938214557fd3bccd92bca295a75fbe65c7f17040510610d07def	27f4e044118837347e5825bc4c1f77db58ffb1c73b229de11237829dcdf50e31	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-1024-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-1024-fused1-residual_squared_norm.json	cff5a2bb1c5969fae7e845beecc3c67c5326e72dd7339e25449e8543aae21012	eff2c432cfd18af11b20c2f82f93faf62dc67354ce2f2097f9846b671e17950a	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-2048-fused0-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-2048-fused0-apply.json	47b6d40c3ae8f2cc30223f6c8fe643e860dc1826b8ef45e32c5700a8959dae16	32eb27cb86966f691e8d76362077b38f9fe34adef4595952bd827bd65178581d	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-2048-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-2048-fused0-apply_and_dot.json	26e7a080508a20c4dfbfb4d88c9c6622685cb8a604964731b1b0d81e6b05aa78	9d39aa9a84454ec956f10cf7ea9854bc7083c4a16bf98213d1ed5bfd58f60e91	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-2048-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-2048-fused0-residual_squared_norm.json	770b8312049cf78e394795b827098a9e7c4b295998241f701d811f508670e52f	7c2aeb74e548d70b06b01c27d0e9689b47570392306ad53cfab77fa277068c1f	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-2048-fused1-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-2048-fused1-apply.json	7f9d40e7939f94a5807a5a859b45474e099c1e0f6b89a6c2973e373d5b8f4b91	4d2a2a4c2f6c3a285879cd123944b9e67c29a6e1e0054a5267a4a69f92d6190a	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-2048-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-2048-fused1-apply_and_dot.json	726cbaf5dd9186fff23b4077392853011943cf9ad79ab7beb5686423a2c03c10	93e700ffbb32e6b69257f3788ec1b3f0a9d5df1bd60224698c5a607899172d94	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	dense-2048-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/dense-2048-fused1-residual_squared_norm.json	81a2ca14b92533b714e24f8871573425be195ae0c1d9157d425aa8f44ea435a8	1e4345f077fcdbc3d3b70010395e84e4387350a4c926cc63cb93e500632e93ae	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-129-fused0-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-129-fused0-apply.json	9ad86c354af6fb3efc647707c641505617cfaa2ae0b03d570a3af246b87f4ef8	2136ee3ba51e4515bd039a16a08d8f5a2602c1c549f1f48fd722276909f5c549	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-129-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-129-fused0-apply_and_dot.json	6379164fab7f7643cbc3ba60138e15adcde47438cb8c68a661d2078df1698d4b	cf9f36dbf6c6c3312039b303e77c79b29807e98633dd7b58fc99e641b6d572cc	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-129-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-129-fused0-residual_squared_norm.json	8be92776da8dad11bcc012b2d41c44785c763b64d79302a1d2058257d76a1ff4	d5525b28a6db52e2114c014538576ee480dc8f68326221cb18468b2c5924829b	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-129-fused1-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-129-fused1-apply.json	b762508cda20f73786cbf846696d9e9a451ccb1211c2be51bc9aa557996667f4	dbeb18c4f2a54c244bce59ed3eceb65be84a4c69d1521b6d8494b6516c772c0b	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-129-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-129-fused1-apply_and_dot.json	e05cc07afa20c8870eaec5ea05cfb0f23a996583c12c71467f298f7b92e61362	646fd0df4e8c632fe2370b75882e14a4bef4a63aa64db9e762cdc1521b1db4ea	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-129-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-129-fused1-residual_squared_norm.json	d427e38bd73c8d5c6f9922494dfb3254bc11406f354765de041d99be36693763	64b0c610540e83a9ecb1eccfd47d69bf60e0427b25dff2d068490a0779571a66	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-130-fused0-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-130-fused0-apply.json	f103f599d40950ce040bc5780c12c26ca9174e881405e3c0485ee7d65c7ba331	93ec28b46e4dd482aefb4af06a286dc3b4cae6cba4a009cf91c8731724556b48	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-130-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-130-fused0-apply_and_dot.json	35c5886d0fb453bf903858d63dbb2b95ca34abb0fa931b84be462298f808890c	785de9d4e6b00011abcf3766897dc2a4b28f6e7193c76ef3627628b5286a3f7c	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-130-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-130-fused0-residual_squared_norm.json	e59afe7a04beb4c59ed3947fa3db477f61c50d48d77f99a59f1b1f4abe70cd42	98f5ecd447fc98663bcb8631be5a8b49f39f7ef7535b62e9952554f8d7db4bb9	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-130-fused1-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-130-fused1-apply.json	3c29e45e3c8715f84b6ce7cc2b3c4003c5f883afa1f47c24d61d858d0fec4d80	b043a00dd3e75023073ac1ee7f178d4eab0837353ab8d4dd2f82a95db1e932dd	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-130-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-130-fused1-apply_and_dot.json	d9818516313a1d4e101ffce2b73e5b5ff2853d1669b597a5df0f7b939ae538d5	d7a6245dff80e31ec2f87ddfde981b975a2fe6e5bcfa7e5b9a1e8b3b1c7ae104	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-130-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-130-fused1-residual_squared_norm.json	7b973d7aab4bc577563b5fb56fbaf21c879045c8887cce5beb4c083a3657125a	7588918cd02600f180dc130977ac4899d118f74318e335b60d4f927dcff85d24	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-131-fused0-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-131-fused0-apply.json	de8ad637d93ada010cb0a4df737621348070066429bd2b4bb770430ea6b4ec28	0665d3e555b1faf0e81e36ed4f9d76a97101f1a87b51357321ef7ea58f3c7d93	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-131-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-131-fused0-apply_and_dot.json	98d7abc92bd21125ce52c37ef20f763d9de569a0b0dda7dc42574e99983751bd	2fd40dc5880e2b96939d26f46b8b500d1116a8cd38cb419d5fdfe5e96bb7daf9	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-131-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-131-fused0-residual_squared_norm.json	ed5724dc3d412d44e1861e92b18b842f60b1a158865af337d37987e3939689c5	2484c7ed00b0950f350f3173d961867e6422e74a42d1ce75ac9615ab217aa55e	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-131-fused1-apply	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-131-fused1-apply.json	cfa12d73421d3340413cfff52c41c98890e060a93d4110170f61d6127674434f	345d9d6646c3c93dd5004769164d407fb2c08aa4cbf88259c79ecc1627576170	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-131-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-131-fused1-apply_and_dot.json	d4c83a42da53c34c186157aa180826af58acc3566bcd5db7c6f61632a5f76ac3	ff7b70792c0ab071b346bca3247776eeace330b6e50893ebc8f50aedc56f5794	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
baseline	integer-tail-131-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/baseline-components/raw/integer-tail-131-fused1-residual_squared_norm.json	3c489126c52afc100c454b2ecc4f39f86901a656412f5a1ec4d875c5ef2bcf30	8edeea1cee2cd93b95b03312c1d46a381689e90c3d33a08d85bc54f68dd51607	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-64-fused0-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-64-fused0-apply.json	db831ce2e53d13a1cbf8e4ebf78b5dea58dc45c9c004b4f9ccb7686331093f4a	025ccd7d9fbe536f8666b2fd093bcd9928c303ea5301ddc292257c78a9868c79	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-64-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-64-fused0-apply_and_dot.json	53792a37d04beebd3ddbecbec1bc511d2a7d67e1bd364554332440b742abc915	cc78103362a5e1af8b56643b246ebbf881af7f796c0e0700006149d534554200	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-64-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-64-fused0-residual_squared_norm.json	339417b9538e004e34f4db25a6604fb78126c14a384e6dad9eef2a67a98ad0ad	848400e51815ffa4a753ac7ee3b0c6b5400bef5f686b6464e094719bd9badf5e	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-64-fused1-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-64-fused1-apply.json	1fd3d5526d0ebcf9c2d0972f63895bb406a154dd498559e7c1f7e6b1113215d0	363604d5ed10b6c01ace8c6c0444f0474d9057a8333b7b0836f6238e7ecca31e	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-64-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-64-fused1-apply_and_dot.json	2dc2dd559e90a290586af8419695a26f8228c6274ff7a88c4383f5260303ed36	cea74713f7ba0822cd02b7f904934d7f09e773bd4058d2e2776f53b6d9e7d9f4	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-64-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-64-fused1-residual_squared_norm.json	7c775b8c4310c810b398a6270f9220a9f311933465406b6961cf3f30f73e4d8e	e7c3e18c3b14dc21e96077158918b7ea2d9a992393d94363f7b52931ce36f798	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-256-fused0-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-256-fused0-apply.json	664945fda5670ecdd58670ffd4846431a37059dc1aba7d186b5be7f76e6ea115	e6056efd5aa9af81cd28050731ab2d5d5519ad03cbc5b33f9220cda7e441ee10	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-256-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-256-fused0-apply_and_dot.json	6ec3879c7227fadc122650ce7c9acbfff9467c966757300c3478cbdc2eb31d3e	58ed21d043e4580f7e27144f0ae580cdf63168057a7f2ef99fa7a10dc7b3234f	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-256-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-256-fused0-residual_squared_norm.json	a732b5ef44f4a2ac5c0c42867decc1413b9ef5d108862bff6e4c718ff52dec2c	3fcb43eb09edbfa3061b6a08dd391e47bd5b31aa9c29bf564a4cef77fb727dc1	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-256-fused1-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-256-fused1-apply.json	e2498b7b494c01db34a8b5fb19f2d06f404c9e0b97c93e58e8bc2c1d75eda0da	bffa9c45602d2502963ad4b17b98f8ddeb5626b970a9e36de3c762c73954eb91	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-256-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-256-fused1-apply_and_dot.json	bd0d9bbd043e2923625a7781dd87cd4732dae8c43aa1b23b48761ab09b06cf29	eb621ae733f2e3c5819b1a7bc9d78a4f065d74eb1f21b50981b0f25bddc40fe6	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-256-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-256-fused1-residual_squared_norm.json	2ae5a24d9a6720f12d89510133a838800a1be22534a63f87dd68ad6667061fa8	d204c88828028677cc78a0a064a8b2675793f1f34b0492940f9f5d7a38050c1d	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-1024-fused0-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-1024-fused0-apply.json	c9641aa172af7aafccf171f3da4fdb8d94b22bc5687c1779da80203070d62d89	b1c07b700723a5779fbe2892068df17467e2d68702eda50f4e3fbca61628d405	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-1024-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-1024-fused0-apply_and_dot.json	5fd380b8ce90eb2f7028cfa949aebc2a55e1248966df908a1a211c8a0c4c4ae5	519e7b6ad9c7d89b2c877e9ed0588ef36456d04ec36bbbb0ae5d379908a4bf2d	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-1024-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-1024-fused0-residual_squared_norm.json	fe5d1d80828b4b9bcb3027307a7031539af183c8c677d2e02e03d550c1a53fed	7c3824027fb56080a21a33bce01aa586054c818f208de86c1dbddfbde4c10a5c	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-1024-fused1-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-1024-fused1-apply.json	68241de7f2d6897989abdf3186dda11904fecca3e5ea2137aa40760cc3dd47c6	fc9eccb77e6360bdd6a6d1b8d0f088e1fb5efddedd96c00d5abce49f816865ed	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-1024-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-1024-fused1-apply_and_dot.json	6ed07e1e70094b2d7edba8d8086f7c9027bc223e947f2728845134f1bd08c5bd	8da1152e5e3b189203c3ab56ca00bc69c9d24fda2cf8da8900fcfefbb1bd67e2	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-1024-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-1024-fused1-residual_squared_norm.json	b29349474f679128d7f955e5c2804fa0b93b22adef212729e81b25a0c71e4318	aab8a9fdabd07dbd13c491d9751b187f86f070205287f9f553e8561368acb763	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-2048-fused0-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-2048-fused0-apply.json	495988025cde6b146c5a3c22e32eba70d9159dae7e64346431e8fa0113f30766	0a839dd47862f5e8cc7cdcdf422ebcbb912c8a6f89fa047b6df04cbce37e6084	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-2048-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-2048-fused0-apply_and_dot.json	7bb373bba5b0e3036a6cbde4b2b8e981b72eac304fe9e6caa659caa35d273038	90d83455c1437ef7362a6f9a4838bd3169584f46c2d233c6cfd5f4fa44e7313f	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-2048-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-2048-fused0-residual_squared_norm.json	470e7ce6bafdf459b11a9b36014f5827ee9dd5f70bf9495d96639b8d215734d6	47d1ca3138a0ed864e82ba43ee9ace0ad4f660ac39171fd375df7bed6fc8dec9	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-2048-fused1-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-2048-fused1-apply.json	3f837ffa397269e90e4f3572080123c19af3cce5c9a7d050fbaa3119eca536d0	faed75e3705144ace31b5d785170fb0e6638cb290f95ace2f8d636c8f5096df0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-2048-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-2048-fused1-apply_and_dot.json	e10544a50c352c411cb48b69f2dc3c1e8cfed0f8ede1963b79ca0a6e9bb01a55	47a7ddaf0644f6b7b7ddc729d8fed064bb3b63054ee9ee5735a8399ab23b2ce0	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	dense-2048-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/dense-2048-fused1-residual_squared_norm.json	5eae0121bdd51c61b118ab3712b1addc54ec979f72c104dacde2190fd9872253	e4f4bf6e712c357ecbc4771de16f83c116b2bf1a15dca11f172fa9e3e1195165	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-129-fused0-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-129-fused0-apply.json	931f4fe666b486a9934593db20246455ee1ecd8a7e2c2e1c2068d5aa824230fc	77c34d20754290ddd7fb7aa651f796688bb12ac1223b195fd15ea044cf5e6447	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-129-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-129-fused0-apply_and_dot.json	7d7542450496afe89a2154a16713bcabe0599d9879e037692380a462f048a59c	444eaf7e6bb2b91af7ec5ca48e05d2c82b21791afb4fa547873d7582e5f614a7	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-129-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-129-fused0-residual_squared_norm.json	01175084cf836cca1730ef322d2c28b99aca48dc4db92aed6f0f545c21cdc608	d0401c63d4f6cc8abc7c1a41b251d5377c0622fca0e869d97cbd09300a7bef45	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-129-fused1-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-129-fused1-apply.json	6eed34208e857485da200f08a44769f65039bc462cf6c42ec4a76b4d1b8eff70	d5468461bf2cd20be4487b324b0aef897cd7df0b0030018d8f3ecd7021da623c	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-129-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-129-fused1-apply_and_dot.json	aa443d02e206838bfc9856999bd3a9b377969b9990e073c2e49d823cf3c05ef7	03de9c24c801e33b30918a9dad73ec35bf7d584697219958d5baccde0f61bc7c	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-129-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-129-fused1-residual_squared_norm.json	627dbe9608750a7bd14f9b0944d19cfb3838c6c156a4456ca7ec08613e7dfec4	04e6571098e88a5166f206f8d410bb6ad5b7975d157dd3ca252efd01fa7650c9	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-130-fused0-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-130-fused0-apply.json	19ce044e8b8fdb780d844a396b2f6baf778a9fd7e09cce0c17f69dee483bed32	160c69499e8185544084edc25cbd77dd0bd013efe904facd6c19e182a2a492c8	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-130-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-130-fused0-apply_and_dot.json	6707db2e279124565a7ba498aee2e2413c4ec219030c32a581f742d8d2859658	6bcd05f637ec78001e07ea0688b21881c15c7f05ac39eab57ac287ea36d84e6b	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-130-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-130-fused0-residual_squared_norm.json	d065a75ad75defa014e4994d4c7bdc0075c215e26877b01339d2f7dde19d4b94	470d8c7f57b4cc5cc9cbe7e940c244212abbbce60a87c471cf8775b7b030f101	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-130-fused1-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-130-fused1-apply.json	652961529d96df843e573bec8678723e588463886a986ff5be577c3498c9d333	68c007f67f3d6c2359b8db803b2034d2640f3d674873b8527e68090e79734398	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-130-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-130-fused1-apply_and_dot.json	cdd1befc51ce2ef91f91f50e710a439f3b25b936f40790044e0229bd74f8fcff	6ae25b38c12ffedaaf2eb30c3173d7d943d8a97c2e57095b5296da0856a1c18d	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-130-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-130-fused1-residual_squared_norm.json	3529227a9e2f601b75df75f7774b66af3c1f1f33a9d5e00ceaa93d4efb7bfd7b	575d874178869972374beb41a2e34c0104ce54934b5098832c13545f851f4db7	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-131-fused0-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-131-fused0-apply.json	815c5c596ec9064502df86cbee0bb421bda3ec3b20dac7e98bad5805f5ae6129	ea13c76ff548cf8f8e7c85b489040ebd0870e40a96ba8c3e07207ce7bf85eddb	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-131-fused0-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-131-fused0-apply_and_dot.json	35908abe5c7efd11b71b881b39e0d2bf73df7c192ca2edb5f09a59f85b3e9e71	26d194d2038f1a2eeaa7d4e7e3fa93c59f305e2d5d5f84fbe730c4edbffae6fd	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-131-fused0-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-131-fused0-residual_squared_norm.json	f6daf9850d88f45c5521dd4acb7f26c78c3e917214161b98fc500cb60f1fcef3	0972c197c55c9ab541e9a30429d035d4c18ba04708fb526849ebad522fe12e89	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-131-fused1-apply	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-131-fused1-apply.json	cd47797b1d9672dd5c37c82786be4f0b693af3003de60a7b9eb57564f2437dc9	7231db445af80919a1f4e219d62d554f5bfe44ffa22e4dd20847a6057d15d382	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-131-fused1-apply_and_dot	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-131-fused1-apply_and_dot.json	2a068b776916b95fa9f55434f76c77d7b19ca978bfcd55d66962bd856021b005	08853273d04ed91a461e4e36539266b1c18a746bad9239a2f6b93467ff314b85	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
candidate	integer-tail-131-fused1-residual_squared_norm	build/restart-optimization-20261017/cpu-four-row/candidate-components/raw/integer-tail-131-fused1-residual_squared_norm.json	e161373d46650ae5a23e2451ef3ec0b55363c7c9077d770bdf409cc10e1f7047	17f09547dc0c2b0febb939f7212ff35efd3ff68375840d1f516159c8e9655a67	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855

All36 raw samples, exact input/compiler/worker argv, public-header closure and stage/archive identities are retained in component-analysis.json and the original immutable manifests. Paths under build/ are local ignored evidence, not public downloads. To reproduce, build the two recorded source revisions with the exact portable flags and public-header closure, export the named fixtures with seed20261017, and use the probe/worker argv recorded per row. Run all42 operations per side, preserving padding/vector/control policies and every result; new execution creates new timings.

Complete retained A/B tables and provenance

All536 strict rows below retain the five complete campaigns. B/current means the specific measured candidate, not a later production checkout. Rejected experiments remain visible alongside adopted changes. Cross-scope framework ratios are descriptive and cannot override any strict verdict.

Every campaign retains its complete preplanned A,B,B,A order, all native candidates and framework controls, two fresh-process blocks per side and every declared fixture. No failed comparison or regression is discarded.

Each block contains 36 complete Williams rounds, one solve per round, three warmups and seed 20261017. Strict A and B summaries pool their 72 normalized round samples. This estimator differs from the median of two process medians used in the earlier descriptive baseline. The A samples here are new runs in each ABBA campaign; historical baseline samples are not pooled into them.

The frozen production comparator uses the original 5% budget, 95% hierarchical percentile intervals, 10,000 resamples and seed 20261017. These saved intervals are audited without recalculation. Per-case intervals are not simultaneous coverage, serial round dependence can narrow them, and two artifact blocks give limited evidence of process variance. Control drift is not subtracted.

Known iterations/matvecs/restarts must match all four observations for each same candidate/configuration. Different frameworks/scopes can perform different accepted-tolerance work. No pure-kernel normalization or cross-scope strict comparison bypasses a failed gate. Resident CUDA scopes keep transfers outside timing; Session still creates fresh output and optional Jacobi while reusing only iterative workspace. Device execution remains explicit and does not change the Torch reference implementation.

CPU OpenMP uses two threads and logical affinity [0,2]; other presets use one thread/[0]. Compiler, runtime, package and full threadpool controls are audited per preset/suite/configuration. WSL affinity is not physical-host isolation. Thermals, clocks, Windows display and host load remain uncontrolled. Campaigns with different actual sources stay separate; no universal speedup is claimed.

fresh-abba (fresh) fixed A/B campaign

Whole-campaign verdict: regression. All 32 artifacts/384 candidate/case outcomes are retained.

Every native and framework strict result

Candidate	Role	Rows	Within 5%	Inconclusive	Regression	Comparison failed
cuda-native	native	32	6	25	1	0
cuda-cublas	native	32	2	30	0	0
torch-cuda-eager	framework control	32	4	28	0	0

All candidate rows

A/B and current/reference ratios below one indicate a lower measured numerator. Ranges bracket the two process medians and are not confidence intervals. Saved strict intervals compare current against A/B baseline; they do not give confidence for the separate current/native-to-framework comparison.

cuda-release-cuda-dense-f64-v1-none-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	5.990799	5.566319–6.075101	6.878946	6.361309–8.367543	1.14825	1.0158–1.5252	inconclusive	49/51/0	0.648887
dense-64	cuda-cublas	8.027313	7.900297–8.490600	8.586097	8.076489–10.819370	1.06961	0.9172–1.4040	inconclusive	49/51/0	0.611009
dense-64	torch-cuda-eager	15.572653	15.572653–15.582584	16.431311	15.620182–17.442043	1.05514	0.9666–1.2370	inconclusive	49/51/0	0.611009
dense-256	cuda-native	8.727754	8.082854–9.732482	8.868001	8.330844–9.609753	1.01607	0.8340–1.2008	inconclusive	67/70/1	0.979365
dense-256	cuda-cublas	11.162451	9.925811–12.249381	12.028764	10.432799–13.320502	1.07761	0.8459–1.3350	inconclusive	67/70/1	0.994421
dense-256	torch-cuda-eager	19.945405	18.923392–20.439822	21.938080	20.491255–23.595915	1.09991	0.9901–1.2841	inconclusive	67/70/1	0.977359
dense-1024	cuda-native	12.565022	12.020011–13.108141	13.930637	12.606834–15.409537	1.10868	0.9274–1.2981	inconclusive	70/73/1	0.928014
dense-1024	cuda-cublas	12.969179	12.619217–13.061041	14.666550	14.006196–14.776200	1.13088	1.0004–1.2527	inconclusive	70/73/1	0.928014
dense-1024	torch-cuda-eager	21.420195	20.310178–22.154037	23.441527	23.217088–23.735183	1.09437	1.0293–1.1914	inconclusive	70/73/1	0.928014
dense-2048	cuda-native	20.723479	20.651594–21.392236	21.500293	21.134262–21.735993	1.03748	0.9645–1.0732	inconclusive	70/73/1	0.871762
dense-2048	cuda-cublas	21.298671	20.981434–21.933403	21.685540	21.274891–22.638927	1.01816	0.9596–1.0950	inconclusive	70/73/1	0.871762
dense-2048	torch-cuda-eager	28.267417	27.856112–28.820904	29.355132	28.935983–30.218388	1.03848	0.9942–1.0880	inconclusive	70/73/1	0.871762

cuda-release-cuda-dense-f64-v1-none-api-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	4.958106	4.511310–5.414434	4.480462	4.367547–4.551558	0.90366	0.7701–1.0458	within 5%	49/51/0	0.648887
dense-64	cuda-cublas	7.238997	6.952524–7.439982	6.980843	6.976151–6.990018	0.96434	0.8320–1.1065	inconclusive	49/51/0	0.611009
dense-64	torch-cuda-eager	16.615526	14.945384–18.101284	15.322932	15.241425–15.335229	0.92221	0.8082–1.0833	inconclusive	49/51/0	0.611009
dense-256	cuda-native	7.318467	6.535211–9.310261	6.303477	6.072243–6.489874	0.86131	0.6276–1.0154	within 5%	67/70/1	0.979365
dense-256	cuda-cublas	9.420893	8.703219–9.973629	8.559380	7.517732–9.830570	0.90855	0.7192–1.1554	inconclusive	67/70/1	0.994421
dense-256	torch-cuda-eager	22.043427	21.068174–26.013520	20.136776	19.305942–20.988536	0.91350	0.7380–1.0044	within 5%	67/70/1	0.977359
dense-1024	cuda-native	9.356555	9.213931–9.416039	8.329847	8.205682–8.381606	0.89027	0.7577–0.9628	within 5%	70/73/1	0.928014
dense-1024	cuda-cublas	10.120622	9.675468–10.695447	9.713313	9.243750–10.045236	0.95975	0.8277–1.0643	inconclusive	70/73/1	0.928014
dense-1024	torch-cuda-eager	21.322642	21.201767–22.084372	19.749612	19.421816–20.349317	0.92623	0.8419–0.9730	within 5%	70/73/1	0.928014
dense-2048	cuda-native	18.239783	17.909893–19.024649	17.593325	16.732059–18.638730	0.96456	0.8556–1.0741	inconclusive	70/73/1	0.871762
dense-2048	cuda-cublas	19.247432	18.169121–20.327390	18.834974	17.273602–19.482138	0.97857	0.8437–1.0913	inconclusive	70/73/1	0.871762
dense-2048	torch-cuda-eager	29.938409	29.788967–30.445128	27.898975	26.573736–30.453212	0.93188	0.8628–1.0408	within 5%	70/73/1	0.871762

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.148835	1.067803–1.279684	1.293035	1.252372–1.329819	1.12552	0.9022–1.3085	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	3.175622	2.973723–3.575760	3.383411	3.341808–3.398331	1.06543	0.9041–1.1703	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.672486	2.404158–3.215788	3.078105	3.049989–3.113431	1.15178	0.8974–1.3509	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	1.025435	0.927322–1.047999	1.157195	1.072326–1.336868	1.12849	0.9767–1.4642	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	2.913485	2.726187–3.158002	3.376632	3.346917–3.442177	1.15897	1.0385–1.3284	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.312146	2.149276–2.520730	2.843663	2.745783–3.013510	1.22988	1.0232–1.4664	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.327343	1.292107–1.363278	1.486171	1.464773–1.516747	1.11966	0.9925–1.2398	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	3.088715	2.952999–3.317144	3.378528	3.370707–3.387824	1.09383	0.9927–1.1771	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.172534	2.023207–2.407300	2.456437	2.342446–2.532166	1.13068	0.9246–1.3485	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.668646	1.608481–1.691979	1.606190	1.489635–1.809052	0.96257	0.8471–1.1429	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	3.433789	3.304135–3.466320	3.214390	3.150298–3.428475	0.93611	0.8755–1.0668	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.703196	2.646180–2.744915	2.462680	2.347752–2.593837	0.91103	0.8058–1.0562	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.822068	0.811028–0.829899	0.785447	0.782057–0.803102	0.95545	0.8942–1.0508	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	2.848774	2.847453–2.879440	2.631548	2.586017–2.674647	0.92375	0.8335–0.9858	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.659801	2.530358–2.768668	2.575579	2.493769–2.792286	0.96834	0.8390–1.1219	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.744645	0.744501–0.744645	0.723645	0.719113–0.724900	0.97180	0.9136–1.0141	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	2.683412	2.579143–2.748382	2.570172	2.530750–2.664811	0.95780	0.9060–1.0646	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.198082	2.129034–2.250818	2.158510	2.053362–2.325639	0.98200	0.8523–1.1202	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.026048	1.001931–1.056528	1.059681	1.014325–1.089045	1.03278	0.9488–1.0969	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	2.712432	2.688716–2.744487	2.711186	2.633013–2.814614	0.99954	0.9087–1.1058	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.176731	2.109250–2.211806	2.282760	2.101776–2.385665	1.04871	0.8889–1.1532	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.344855	1.333973–1.348355	1.350888	1.329449–1.448594	1.00449	0.9696–1.1458	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	2.952529	2.918608–3.137034	3.028773	2.944456–3.232934	1.02582	0.9283–1.1207	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.303138	2.136286–2.471726	2.504496	2.458140–2.557809	1.08743	0.9435–1.2188	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.211960	1.169873–1.230563	1.246935	1.228885–1.254141	1.02886	0.9250–1.1015	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	1.242849	1.148566–1.438000	1.279424	1.244671–1.297194	1.02943	0.8305–1.1520	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	3.020078	2.699664–3.464084	3.123741	2.997441–3.304105	1.03432	0.8218–1.2449	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	1.065230	1.032461–1.126694	1.148883	1.090025–1.155936	1.07853	0.9579–1.1520	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	1.062603	1.031986–1.084192	1.105408	1.095670–1.113050	1.04028	0.9763–1.1587	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.523412	2.440196–2.568271	2.665855	2.646980–2.829159	1.05645	0.9691–1.2112	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.371506	1.336181–1.479844	1.574534	1.569454–1.574534	1.14803	1.0121–1.2606	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	1.224973	1.189229–1.250009	1.379453	1.339786–1.398062	1.12611	1.0463–1.2202	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.569501	2.299796–2.718920	2.942484	2.791991–3.258278	1.14516	1.0006–1.4770	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.784979	1.717823–1.926554	1.857889	1.799575–1.885850	1.04085	0.9096–1.1380	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	1.640162	1.523467–1.867833	1.640367	1.601130–1.685409	1.00012	0.7927–1.1252	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.792294	2.502795–3.175930	2.757448	2.720937–2.816527	0.98752	0.8291–1.1550	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.877521	0.838837–1.069317	0.836047	0.816065–0.862156	0.95274	0.7466–1.0432	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	0.972186	0.917690–1.013689	0.922022	0.908374–0.943342	0.94840	0.7857–1.0499	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.844071	2.603395–3.128595	2.745232	2.588211–2.842109	0.96525	0.8006–1.1376	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.807076	0.796100–0.812168	0.753803	0.750744–0.771586	0.93399	0.8914–1.0136	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	0.869463	0.868022–0.869463	0.839700	0.757115–0.908761	0.96577	0.8347–1.1050	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.772206	2.592509–2.994291	2.370183	2.370183–2.423693	0.85498	0.7455–1.0209	within 5%	4/6/0	0.209854
dense-1024	cuda-native	1.110392	1.099149–1.161992	1.078235	1.051075–1.094998	0.97104	0.8908–1.0711	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	0.958198	0.956453–0.962375	0.955894	0.904855–1.091875	0.99760	0.9039–1.2691	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.536409	2.477205–2.609453	2.437807	2.251127–2.786450	0.96113	0.8432–1.2604	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.486492	1.464488–1.560732	1.566050	1.535071–1.648108	1.05352	0.9515–1.1502	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	1.271726	1.250188–1.284863	1.345154	1.332597–1.361391	1.05774	0.9843–1.1244	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.605779	2.576443–2.644598	2.555153	2.522720–2.697633	0.98057	0.8924–1.1080	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.115659	1.077836–1.173229	1.290857	1.078509–1.720284	1.15704	0.9144–1.7486	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	1.203971	1.108972–1.363744	1.412484	1.132160–2.064938	1.17319	0.7996–1.9184	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.450992	2.265464–2.775006	3.124087	2.321696–4.076408	1.27462	0.8324–1.8110	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.906042	0.884392–0.926696	1.055621	1.001681–1.083035	1.16509	1.0582–1.2594	regression	4/6/0	0.209854
dense-256	cuda-cublas	0.929852	0.873166–0.988967	0.993488	0.956736–1.036269	1.06844	0.9585–1.2023	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	1.855552	1.782391–2.146524	2.149329	2.109222–2.151924	1.15832	0.8631–1.2412	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.225636	1.205553–1.234992	1.318239	1.274443–1.373106	1.07556	1.0112–1.1645	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	1.098774	1.073492–1.115592	1.136211	1.112354–1.140335	1.03407	0.9863–1.1030	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	1.922899	1.892256–1.955464	2.122151	1.987136–2.367614	1.10362	0.9851–1.3291	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.591410	1.590284–1.594009	1.677698	1.660426–1.690355	1.05422	0.9965–1.1038	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	1.429221	1.418883–1.435116	1.477210	1.459999–1.555427	1.03358	0.9718–1.1338	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.134995	2.105147–2.134995	2.189638	2.073052–2.303970	1.02559	0.9286–1.1570	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.744911	0.727204–0.766778	0.792945	0.769998–0.796427	1.06448	0.9741–1.1862	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	0.785900	0.773809–0.812098	0.811352	0.755027–0.848996	1.03239	0.9107–1.1452	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.159779	2.098845–2.220444	2.375952	2.199975–2.511372	1.10009	0.9363–1.2424	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.712705	0.702860–0.733469	0.763787	0.759822–0.767930	1.07167	0.9877–1.1832	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	0.726788	0.708832–0.731567	0.725878	0.700228–0.730787	0.99875	0.9222–1.0802	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.044892	2.000097–2.099070	2.056516	1.980679–2.111378	1.00568	0.8884–1.1201	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	0.998027	0.948211–1.044831	1.044334	1.010714–1.063390	1.04640	0.9592–1.1320	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	0.863951	0.825384–0.924487	0.890622	0.832925–0.919002	1.03087	0.9001–1.1313	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.107824	1.704477–2.252203	2.091747	2.015678–2.314008	0.99237	0.8538–1.3935	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.352891	1.294159–1.418786	1.439126	1.318542–1.818169	1.06374	0.9214–1.4335	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	1.230195	1.133638–1.299477	1.418946	1.168682–1.772522	1.15343	0.8916–1.5763	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.078502	1.938821–2.221036	2.274089	1.948582–3.035519	1.09410	0.8745–1.5811	inconclusive	3/5/0	0.836412

Current-source framework comparisons

These are descriptive ratios of current pooled-round medians from the same matched configuration, not strict comparisons across frameworks or scopes. Every higher current-native median remains listed below.

All current matched scopes

Configuration	Native	Reference	Lower/higher/equal	Current ratio range	Same known I/M/R
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4042–0.7324	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.5225–0.7387	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.2924–0.6306	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.4251–0.6751	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4069–0.6522	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	cuda-cublas	torch-cuda-eager	0/4/0	1.0992–1.3754	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3050–0.5394	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	cuda-cublas	torch-cuda-eager	0/4/0	1.0217–1.2093	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.3992–0.6738	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.4096–0.5949	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3045–0.6129	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3359–0.5264	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4132–0.7662	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.4521–0.6746	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3337–0.6328	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3415–0.6240	True

Complete current higher-median inventory

Configuration	Case	Native/reference	Native pooled ms	Reference pooled ms	Ratio	Native I/M/R	Reference I/M/R
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-64	cuda-cublas/torch-cuda-eager	3.383411	3.078105	1.09919	5/7/0; 5/7/0	5/7/0; 5/7/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-256	cuda-cublas/torch-cuda-eager	3.376632	2.843663	1.18742	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-1024	cuda-cublas/torch-cuda-eager	3.378528	2.456437	1.37538	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-2048	cuda-cublas/torch-cuda-eager	3.214390	2.462680	1.30524	3/5/0; 3/5/0	3/5/0; 3/5/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-64	cuda-cublas/torch-cuda-eager	2.631548	2.575579	1.02173	5/7/0; 5/7/0	5/7/0; 5/7/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-256	cuda-cublas/torch-cuda-eager	2.570172	2.158510	1.19072	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-1024	cuda-cublas/torch-cuda-eager	2.711186	2.282760	1.18768	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-2048	cuda-cublas/torch-cuda-eager	3.028773	2.504496	1.20933	3/5/0; 3/5/0	3/5/0; 3/5/0

Source, controls and raw evidence

• baseline/cuda-release: compiled 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7, tree 06f324529806d3edf005e4ecbb6a3c27f048f946, identity SHA256 7a956531880eaee793f22d75c4cc748f53c20ee8d5569fc0f2fe4cf67411d416. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: 760af83cc19b2e0c6b06f9373166d7bd7022d4a2a779677f5a3d141dec774240.

• current/cuda-release: compiled e5c587889906511b0f9e74f42d88d8e857dea3d0, tree 241772b9eb7a0ed25bfbf8e82bbe4118386991ea, identity SHA256 aba137408ee38d8d66ed6e90c092ad5cbe0afe31193999fd19cd936fa0d23d77. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: f66428197ece464a22112fd0f6aee8e922f92fb462a7e38d9b204b194a4d2213.

Manifest SHA256: f7d874200fee86d850deb96e962257fbe057dfe5f4e945740cd54b34bb24498f. Plan SHA256: a3b43a52d9cb01a262df97f064627f64729c25d7efb10a1e367854d682fbaf97.

Every full strict report and actual comparator

Configuration	Verdict	Report SHA256	Comparator SHA256	Comparator path
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	inconclusive	45aa248dd5adb06814f515099c0fc21a2964bd026370194db8d2e81cf1d43cc3	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	inconclusive	237a9c526d6895f9dbb2e3b987aafc932c99853fcf20d6023977f0ae0e7d6387	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	inconclusive	6d3b96e65c37f14d33db55f7c75e966d73687e7a6ad37968dccd3fab6e08537e	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	inconclusive	00d4ba3a0c05aad38d26233711ac9cdefece1d067178f0b7d820e9ffde73b31a	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	inconclusive	b1d543830fdd97e6fd51dd783000d078e7d14aeb375525096b3a5b477979d252	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	inconclusive	0035674076b58c384418c3440342656b350b5f75d29d7c3bd0281fc93877d8a5	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	regression	da560e993cef119a690924a5c8a040b93ddd9cbdfdd86ca9539dcaede37b5a3e	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	inconclusive	85ae1ff9ba60043eeff33bd255a9ffc58491e66d4e730cb2b1cd64f97b725611	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/compare_performance.py

All original raw/import/log identities

Run	Raw path	Raw SHA256	Import SHA256	Log SHA256
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b01-baseline.json	19dc286b5d6217ac233ef41653123f8093e7d0cfc181b7291397c90e288b34d7	b67c6ef8ee1427535a4e45cd1a53a6887918a2d82e2faa2ba67a600b7ea22559	c5ab8b88ea093a402532ebf7cce8fb9d1145bbaea60623d9886c590834009376
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b02-current.json	3319ad1c3e63f1518b2e69c10d01ecfb36b59702f17cfd3b6350b08a2c560fa1	e2d854eebdf85d12cdd607620aa1f4531661e1153e6a2590bacfcfde7f92194a	67a3188d10c26c700a5789ef541ed5d9d654e40b15dc48dc05f62eedbeb40a39
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b03-current.json	3da761940a0cac147fb0730521d73915de4055681d16d04da85f55690c500941	fb0be82f8f5a549e62b756afa0d87bcccf93d4500900f83b6eb47b25c99420fc	9e17f98ee11f95cc58e2025054b99a673e4ec17b7057e90d259a9abb7ac8f3e6
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b04-baseline.json	81a1e1b7b7fab5bc4f43d186549dc529f3d14a535e8b879752f34fc65bab1fa9	f1676f49404eea2f55379215c3adf2a00210876ad66d91e0de42e36477622160	d8c2f78202ba17c74c458affe1cf8bc9b5092555ab95aa750bce2168a65bdb91
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b01-baseline.json	b7153f1bfd694668cd43922bdbfa6ce208b5ff4e6ad34c2a9e8ec7be132b00b8	ac7c258386833a5191be0dbb0d6109fe76e20c116856d5abf08e2530cdb9268f	a76bcba96710ce2371e3d9e176d4250fa2eeab6fe2054cac17fb74154e91ac18
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b02-current.json	dadddebad67d99d3106b01db77d2c2a3e5d13aa8abc4715d2dfc06952c0902c9	c03920ea2a29c19f7e1a71872ed45177d5d967f38770ea0ed75dad15125ef6e6	6f2feaf712fc7c2909348abd559e997282e46c39a14b1aad4500f2296dd3092a
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b03-current.json	40304371589048188515f1a38b76b63f6fc0e71a8972cacf8e875a46bc169273	1cf215018dafce90e34cb716ca02eaedf3145272a406125e5f922dff6313599e	ed855b719b516cc44cd127967b8a869c4697b3fb2d5dc30de86e1d0b377cb215
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b04-baseline.json	b3f21f8a30ddf726836b0325a7eb7854e061de469007d41f3a355887a0d04f5f	96a9bbb8cb605b867c550d562ae93cad1b9079c6bd04c3aa794c4bd4e2790db7	1c768b631315f760e478232c7a1ba487c3b0713fd6730ca5bb663608d1297de1
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b01-baseline.json	5a3b4bb509b5e43ce68a12ad683b3c28cd4384f565dd77421631ed14e963d296	220713e7da8c00b1a71321d93e89035612544bcfcfc53dc221c70906d4f8b820	1cf8041c11d64c5f926994dfc79b7d465a2a99be2f31df401d2d6cdab4044968
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b02-current.json	960fbdabbaf07d05e120680bcdedc737e8ec4596c8d0e230c16c74112b35609a	d1a9d062282dfb17475d768db159e65a2d362a8f3535e815052b6588063a6dd4	8eb422dc89cb69b1e9de4e479b9bf99f841300ebe6be983c293bf0a9af918884
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b03-current.json	71737878bff6f9feae27ffbb6a683448f9c18604994c0af3f237ffe64cd2b7a1	38b9504d076b6ec45e8502787d15c113e193040841cc9620d360d14b5fc84e8a	6257e0f2035d7faf0c92312ebd4cbfd9d948b2acfcaacb70d80bc024484a654d
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b04-baseline.json	f972ca9d0a771daf35523d8523ecd7626765fde7d8bacd6a8c5b473ee2daf126	80a8c64fe3a2bb70fd5fb0e705c71df23816a34f6e762777d0b104d782ff7523	ee239d883675f9fc2e59200cda7ef769e4639eca8208b132263d5f793c60fd9b
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b01-baseline.json	3b5e7a52ef2ddfcedd4aa173f3577b46d3f3f0e65a8d36ef69533e4995a390c1	1498af8fac317680d3f4763cd0c25853ea5f8e605ce97cb9205a7ffe5819a22b	c240abe20841efa9b882a9bc2cc7682ac29198884d399e4ae8b322c5fdfd86bb
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b02-current.json	c58db327968b9d165ee14e196501f0e5ea92b78ed03ea04a93978ab702dcac70	6c02d14db8fcff17831f3d6d0a1809000413508bb45639134c4272777fca4213	74d8113df84a2cfa762c6538b3141bc27208ac94bd04072904b7603b09a03dde
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b03-current.json	9424dbe675ae821f025c91f0220d3de92bea66be8556120efe5df8a5e3cdd5ad	f93075ad51ca15d4d51d16f000075cff56c62d0d0f88630d3daf6fb10eebed8d	c1b524d0f72c95f706020f2a5621faddae701a79b80afd154c8b02cc5f155d7c
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b04-baseline.json	2e1366c87e77d9d8ece4f8b7bf202ae7da222a9d3babc6b54a9b2b6b8651d443	93e868405e27457dae68fbf5573d04eb401a2136254a08ad971ad45c4963d09e	0f00250a30be3b1705059565d99cde9d79f71a9a095b7d9150d274f8813c4ac3
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b01-baseline.json	116ef578940cb4bbbd145e89896bfc5ecbaeaeac05e0970a123638a9a9606ef0	d270e83b27ffdc6901d6290993c0743424409e3b12bd6c511e81bb8717dbaa4c	d314242a548164c8bf03ead5ee5462aa3824b7e3d9e65b5772011e8049d9d3bf
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b02-current.json	9fe83ca97e24e80564e97a269600651a2a2ae6598ff6897c7b0ae7f6ad657cab	c6e44f7be5f55233da79c01643ef4b27729e46a848df900d5215bc66d60c8fc2	78348ae51bad33805db7e3953f5319a997367eb3881583c80828bf6e6a658960
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b03-current.json	1b933c32294f59ae906c72d67cd88efe37bd61e096971a13082d06ebad59bf41	14294477812994237e85088eace88dc6673d3dd8031464909305e8442db93698	7df6be65ffeee2073ffa6eb19d665a8d668bc143a8fc5db5c38f59296aa8bab0
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b04-baseline.json	97d02180723d4a6397d941102d839ac543f0b32362a446812f6a9763ae2421ba	b95c3ea0a2009f9d52d09e71571bb824ce8ff0e6e16c17161b9d839f273807d3	f0c29ffa876f3fcebfaf6ee6ac2cb0a7c14236ccd5edc3ed9eafd07795d96cda
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b01-baseline.json	58b008b8fc40ba611c6c16aca6409d274e22a0cfad7838c78e2ebddff93a4bc8	9d208d4b6c4a65eec07c54c662d17cfcb7703de89ecaa610c27e037e84be87f6	75f6727813d4dfe40b4a24061833cf9b256549e75571dcc62250a7f9b971e3e3
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b02-current.json	33192be8317b07ce61440295e6bdaac48faac0382c48532b63dface06fe6df79	517f3e5c3e69732a73ca8924ddfcc2db524ff097a636f0b8e15162a125f061ca	a7c9165d95e684c4dc43b7f3ffc39c97183c2dc93319c72383e6bc5afd352294
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b03-current.json	02c8094acd1c0db1f8f681b2822955c656a43ba99bd6f11b174ae8bbd48d38c2	9d15ebb775aff094c7eace30d2bae242b7d2c52425d11f1b8b2ea5188ac26473	0533cb38b40bae5127002036b4f3fa4d46baf64088346e940b666fecd5552f77
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b04-baseline.json	d5154ca50752e560cf9df30cbb7be34eb774795f72c7a19fd7a2aafe383760f6	419b0c2845a35cb74e7ffc22692b00c5800dc4a9904a5267ee1c3854a05db229	6d8f67339cb664b4d96724ead6d8c5e106f70991d59beb02ebfd582636489693
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline.json	808a1fa271aa73543c381ac9bce55d3487dfc9060c8575d635fbe2afb8b51fa4	125683e26f3e6b0d9465b8722340bbcd52dbb0e7e8e574904b293b6b987d9218	d9429750188b64acf38c5d366f5329661dc0568e00b6540caaf8256044f47d06
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b02-current.json	842d90db8836356548927b9ed3a0738dbf59179b0cced8d36e86babd969d7b79	87721dd422d20fcb16f6f717b4f14d67510afb7045d81a2a224550206ef2314c	47fb6a3bc345c45799495cef94a506e31b4b49649ce28879b26bcd613b2eff4d
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b03-current.json	b9f7ba548ffcb83ecc0f39b5889f64f856a055a1742ec08bb8c8b383d54d07c5	5e7035b5fb77574621e3b0f820043b09c0fb29fb8049bd75d18c9f4d861046d1	ba2475e1a2dd05084b7f9999b5b9ef77b7347e1aa9342d4714183532fa74a8e2
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline.json	90854a50fbcb3063244c7303ed0038606d13adac78cbe91344a1ce1ccc7190bb	c396a0326e39d33c6612b7cf0b172ba06cc6993ee05bdffae10734d406cf8db0	0bb6092ef35c0b92f0bd1a09e4c893566a6dac5e7da389a532d85dce34a7d95b
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline.json	e8933534f1c6f38fba8fc653cfd19ec9f71f8fd2e8b3bcda008783e199c24b84	832ae606ca2d19af98140b5e6736def20e15704355976004d4e09bd241b13ff0	ab8057eaf6ca7412486d5e80262f712d2973336fe2d3de58106d0081ce6c6082
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current.json	79a870966d582ac6aabf93432860f2597677ebfb65d7f0cceca4987afeac19f7	38607aca1978e15d16c2618aecc72e917fda7b6c6b986d797bda12bbc2e697d0	daace775613c54f18c5a4e82d35d7b3b655fcdf7b9bbca84142a0194e900ca2f
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current.json	a7cd755e7a8eaa91518d36bb42f52a566f90940d0cd591324465c3755213c3e5	4988cea683b7492c08787eef557a1688eb06614bb8e9bb45e1c65f8a94ec1ffd	d017a1813afb3322539af5920b622775b9b638ff938116479bff2d614d649e2f
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline.json	31c5f7cdea34dd98a6a3d6f6971ca5985730d3485cddf0dd6723a1b9c3d90342	1aeddf7ece3c07d8cab1809cd96d5333e8260629dfb88418966a5ad93a0e3a31	b3692bffbc3c6cf081555f87245cc092cebdefc6762c1146412f8f94f808d48e

fresh-span-abba (fresh) fixed A/B campaign

Whole-campaign verdict: inconclusive. All 32 artifacts/384 candidate/case outcomes are retained.

This follow-up variant was selected after observing the rejected initial fresh-variant trial. Its intervals are unadjusted for that selection and multiple variants; they are exploratory rather than independent confirmatory evidence. The original rejected trial remains separately reported, without replacement, pooling or a claim that resource layout caused its regression.

Every native and framework strict result

Candidate	Role	Rows	Within 5%	Inconclusive	Regression	Comparison failed
cuda-native	native	32	13	19	0	0
cuda-cublas	native	32	14	18	0	0
torch-cuda-eager	framework control	32	7	25	0	0

All candidate rows

A/B and current/reference ratios below one indicate a lower measured numerator. Ranges bracket the two process medians and are not confidence intervals. Saved strict intervals compare current against A/B baseline; they do not give confidence for the separate current/native-to-framework comparison.

cuda-release-cuda-dense-f64-v1-none-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	6.492710	6.475779–6.879581	6.262711	6.242748–6.298239	0.96458	0.8169–0.9973	within 5%	49/51/0	0.648887
dense-64	cuda-cublas	8.596882	8.520694–8.639883	8.368508	8.240198–8.483339	0.97344	0.9156–1.0275	within 5%	49/51/0	0.611009
dense-64	torch-cuda-eager	16.055709	15.897201–16.472330	14.753621	14.633412–14.846294	0.91890	0.8348–1.0165	within 5%	49/51/0	0.611009
dense-256	cuda-native	9.564447	9.007661–10.216734	9.341464	9.315929–9.341464	0.97669	0.8505–1.0778	inconclusive	67/70/1	0.979365
dense-256	cuda-cublas	12.509035	12.184772–12.627019	10.911715	10.894336–11.191428	0.87231	0.8108–0.9683	within 5%	67/70/1	0.994421
dense-256	torch-cuda-eager	22.280898	21.522447–23.254687	20.682583	20.562271–20.735802	0.92827	0.8578–0.9761	within 5%	67/70/1	0.977359
dense-1024	cuda-native	12.915876	12.310626–13.792145	13.977778	12.603044–15.037432	1.08222	0.9129–1.2496	inconclusive	70/73/1	0.928014
dense-1024	cuda-cublas	13.594882	13.260993–14.199153	15.020853	13.835985–16.652614	1.10489	0.9131–1.2848	inconclusive	70/73/1	0.928014
dense-1024	torch-cuda-eager	21.166487	20.042276–22.461407	22.082067	21.658887–24.694976	1.04326	0.9472–1.2710	inconclusive	70/73/1	0.928014
dense-2048	cuda-native	21.044254	20.518545–21.894064	21.927488	21.678189–22.469369	1.04197	0.9836–1.1211	inconclusive	70/73/1	0.871762
dense-2048	cuda-cublas	22.227856	22.177400–22.259943	21.424021	21.275544–21.524456	0.96384	0.9331–1.0006	within 5%	70/73/1	0.871762
dense-2048	torch-cuda-eager	28.526084	28.012313–29.235084	29.681604	28.045909–30.358908	1.04051	0.9575–1.0866	inconclusive	70/73/1	0.871762

cuda-release-cuda-dense-f64-v1-none-api-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	4.707842	4.658784–4.727042	4.568682	4.527483–4.618936	0.97044	0.9186–1.0344	within 5%	49/51/0	0.648887
dense-64	cuda-cublas	6.975551	6.952021–7.443373	6.706884	6.391678–6.808772	0.96148	0.8265–1.0318	within 5%	49/51/0	0.611009
dense-64	torch-cuda-eager	15.121043	13.807603–16.062929	16.048172	15.898145–16.378561	1.06131	0.9517–1.2083	inconclusive	49/51/0	0.611009
dense-256	cuda-native	6.901218	6.894364–6.924187	6.776741	6.616523–6.792236	0.98196	0.9075–1.0186	within 5%	67/70/1	0.979365
dense-256	cuda-cublas	9.829019	9.178750–10.566930	8.874627	8.845365–9.104415	0.90290	0.7991–1.0106	within 5%	67/70/1	0.994421
dense-256	torch-cuda-eager	22.021948	20.911348–22.174436	21.938269	20.460935–23.026325	0.99620	0.8862–1.1095	inconclusive	67/70/1	0.977359
dense-1024	cuda-native	9.439955	9.343283–9.710426	9.420049	9.395634–9.420049	0.99789	0.8964–1.0743	inconclusive	70/73/1	0.928014
dense-1024	cuda-cublas	11.111570	11.111570–11.261017	10.973718	10.482556–11.124723	0.98759	0.8330–1.1361	inconclusive	70/73/1	0.928014
dense-1024	torch-cuda-eager	22.104254	22.104254–22.237153	21.454560	20.919110–22.013459	0.97061	0.9174–1.0228	within 5%	70/73/1	0.928014
dense-2048	cuda-native	18.151483	18.116229–18.252569	18.125956	17.965605–18.248862	0.99859	0.8950–1.0858	inconclusive	70/73/1	0.871762
dense-2048	cuda-cublas	19.073935	19.057133–19.111118	18.927781	18.822575–18.958759	0.99234	0.9337–1.0444	within 5%	70/73/1	0.871762
dense-2048	torch-cuda-eager	29.569926	29.234782–30.014788	27.701878	27.397935–27.991996	0.93683	0.9077–0.9637	within 5%	70/73/1	0.871762

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.266555	1.229469–1.394025	1.247036	1.233672–1.296671	0.98459	0.8000–1.2006	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	3.397706	3.389849–3.399818	3.378935	3.230551–3.688777	0.99448	0.9001–1.1367	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.982594	2.855068–3.156010	2.653832	2.598919–2.845048	0.88977	0.8074–1.0559	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.981962	0.981962–0.986020	1.077880	0.992350–1.130281	1.09768	0.9465–1.2368	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	2.928189	2.844587–3.117306	3.007370	2.944055–3.212563	1.02704	0.9398–1.1728	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.328559	2.303974–2.335638	2.410013	2.314339–2.580392	1.03498	0.9317–1.2260	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.372162	1.362731–1.372162	1.407542	1.307038–1.497481	1.02578	0.9335–1.1632	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	3.185598	3.114729–3.200362	3.396927	3.117334–3.509448	1.06634	0.9454–1.1476	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.337518	2.329017–2.367221	2.468707	2.467648–2.468707	1.05612	0.9767–1.2512	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.843573	1.829899–1.861125	1.618324	1.536388–1.846562	0.87782	0.7821–1.0562	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	3.872453	3.628766–4.234067	3.370900	3.198328–3.497922	0.87048	0.7416–0.9904	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	3.039229	3.039229–3.060577	2.440940	2.257434–2.512784	0.80314	0.6975–0.8870	within 5%	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.798799	0.795222–0.805222	0.941199	0.886746–1.003705	1.17827	1.0426–1.3068	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	2.798882	2.797344–2.811113	3.063679	2.856313–3.338705	1.09461	0.9679–1.3223	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.589501	2.517878–2.595079	2.919957	2.701872–3.143229	1.12761	0.9979–1.3239	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.788459	0.779026–0.792480	0.903934	0.733626–1.259531	1.14646	0.9002–1.7935	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	2.690084	2.688299–2.705843	3.097362	2.585812–4.159072	1.15140	0.9368–1.6266	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.356784	2.356416–2.356784	2.682766	2.060376–4.148799	1.13832	0.8326–1.9135	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.130004	1.121617–1.192038	1.208567	0.994286–1.425980	1.06952	0.8184–1.3366	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	3.087725	3.081171–3.090010	3.320677	2.902990–3.996428	1.07544	0.9021–1.4187	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.606009	2.344104–2.735156	2.706859	2.341058–3.457313	1.03870	0.8359–1.5683	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.489747	1.419120–1.573346	1.382803	1.377790–1.407878	0.92821	0.8274–1.0195	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	3.144896	3.069691–3.231789	3.201679	3.024430–3.383576	1.01806	0.9043–1.1430	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.488376	2.473180–2.579085	2.457333	2.271940–2.766853	0.98752	0.8585–1.1903	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.265761	1.264669–1.265761	1.222193	1.147981–1.247621	0.96558	0.8870–1.0235	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	1.256360	1.252603–1.274698	1.266875	1.227461–1.333627	1.00837	0.8976–1.1571	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	3.115334	3.069008–3.147870	3.161504	2.937587–3.543308	1.01482	0.8942–1.1930	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	1.136534	1.066345–1.230172	1.141242	1.113314–1.160324	1.00414	0.8275–1.2458	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	1.111992	1.087347–1.188699	1.161641	1.132564–1.165538	1.04465	0.9154–1.1326	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.795138	2.582121–3.229781	2.852864	2.726445–3.150254	1.02065	0.8011–1.2506	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.544483	1.322973–1.906103	1.641867	1.619741–1.647288	1.06305	0.8125–1.3614	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	1.289686	1.200869–1.685232	1.351320	1.330807–1.378114	1.04779	0.7590–1.2389	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.914446	2.354602–4.245139	3.254523	3.145407–3.254523	1.11669	0.6762–1.5223	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.825524	1.710330–2.102473	1.745843	1.628840–1.817692	0.95635	0.7619–1.1199	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	1.718138	1.482538–2.218884	1.605112	1.490426–1.737405	0.93422	0.6715–1.1768	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	3.003433	2.458655–4.278227	2.972180	2.485482–3.723283	0.98959	0.5720–1.5427	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.845392	0.809280–0.868799	0.880169	0.840543–0.989590	1.04114	0.9491–1.3235	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	0.910698	0.886803–0.962322	0.931658	0.918962–0.962815	1.02302	0.9172–1.1400	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.748053	2.741615–2.884665	2.791490	2.707608–2.984863	1.01581	0.9142–1.2219	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.786393	0.757095–0.886089	0.853531	0.807536–0.879088	1.08538	0.8853–1.2080	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	0.864039	0.821983–0.901191	0.864681	0.792915–1.096519	1.00074	0.8750–1.3637	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.626563	2.376130–2.908967	2.876340	2.620608–2.985366	1.09510	0.8733–1.2952	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.067304	1.063821–1.076678	1.079902	1.069503–1.086544	1.01180	0.9622–1.0354	within 5%	4/6/0	0.013684
dense-1024	cuda-cublas	0.964318	0.963151–0.965573	0.955630	0.950603–0.957287	0.99099	0.9444–1.0333	within 5%	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.308142	2.300397–2.342414	2.381022	2.381022–2.393305	1.03158	0.9463–1.1206	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.507021	1.460899–1.527416	1.519600	1.489032–1.534117	1.00835	0.9504–1.0609	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	1.309915	1.299411–1.340023	1.326377	1.320107–1.331659	1.01257	0.9462–1.0412	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.442630	2.387434–2.536990	2.550007	2.494976–2.560583	1.04396	0.9432–1.1190	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.086500	1.073859–1.103111	1.045322	1.039050–1.059414	0.96210	0.9202–1.0223	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	1.114066	1.111733–1.114643	1.082314	1.082314–1.093741	0.97150	0.9181–1.0603	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.411312	2.366190–2.445047	2.266996	2.259046–2.266996	0.94015	0.8789–1.0091	within 5%	5/7/0	0.0796191
dense-256	cuda-native	0.952408	0.949527–0.957789	0.916674	0.911605–0.923790	0.96248	0.8712–1.0279	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	0.959397	0.957405–0.961314	0.940688	0.904094–0.979206	0.98050	0.9144–1.0492	within 5%	4/6/0	0.209854
dense-256	torch-cuda-eager	1.931015	1.897104–2.005447	1.853854	1.846626–1.956182	0.96004	0.8915–1.0847	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.260374	1.195303–1.312348	1.271768	1.243469–1.324194	1.00904	0.9366–1.1187	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	1.177188	1.135690–1.186303	1.144365	1.107617–1.148024	0.97212	0.9047–1.0606	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.013682	1.777284–2.118470	1.922793	1.917353–1.924268	0.95486	0.8488–1.1198	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.614268	1.592365–1.742449	1.579406	1.527241–1.590879	0.97840	0.8574–1.0160	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	1.457574	1.454629–1.479019	1.375128	1.362593–1.378348	0.94344	0.8779–1.0072	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.191901	2.118638–2.202979	2.061099	2.030724–2.142581	0.94033	0.8639–1.0552	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.784088	0.732989–0.832766	0.729788	0.726796–0.731922	0.93075	0.7949–1.0236	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	0.890722	0.834290–0.965932	0.798679	0.788022–0.808387	0.89666	0.7797–1.0090	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.475205	2.214082–2.704721	2.253088	2.130445–2.330832	0.91026	0.7617–1.0816	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.714356	0.699291–0.744426	0.698982	0.698913–0.702105	0.97848	0.8944–1.0422	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	0.798995	0.744861–0.826347	0.740572	0.719957–0.753350	0.92688	0.8240–1.0414	within 5%	4/6/0	0.209854
dense-256	torch-cuda-eager	2.102325	2.087372–2.131005	1.864931	1.859509–1.871943	0.88708	0.8034–0.9827	within 5%	4/6/0	0.209854
dense-1024	cuda-native	1.047702	1.024793–1.093608	1.016387	0.994820–1.042295	0.97011	0.8507–1.0283	within 5%	4/6/0	0.013684
dense-1024	cuda-cublas	0.912474	0.853364–1.013491	0.885826	0.883447–0.890259	0.97080	0.8124–1.0615	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.273217	1.931328–2.503678	1.904384	1.779768–2.280957	0.83775	0.7005–1.1865	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.420087	1.404957–1.422676	1.420836	1.400667–1.432044	1.00053	0.9473–1.0393	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	1.250657	1.244795–1.285658	1.224201	1.215957–1.243715	0.97885	0.9299–1.0229	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.085856	1.943342–2.194827	2.062934	2.054965–2.088296	0.98901	0.8998–1.0926	inconclusive	3/5/0	0.836412

Current-source framework comparisons

These are descriptive ratios of current pooled-round medians from the same matched configuration, not strict comparisons across frameworks or scopes. Every higher current-native median remains listed below.

All current matched scopes

Configuration	Native	Reference	Lower/higher/equal	Current ratio range	Same known I/M/R
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4245–0.7388	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.5276–0.7218	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.2847–0.6543	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.4045–0.6833	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4473–0.6630	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	cuda-cublas	torch-cuda-eager	0/4/0	1.2479–1.3810	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3223–0.5627	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	cuda-cublas	torch-cuda-eager	0/4/0	1.0492–1.3029	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.3866–0.5874	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.4007–0.5400	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.2967–0.5959	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3006–0.5201	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4611–0.7663	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.4774–0.6672	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3239–0.6887	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3545–0.5934	True

Complete current higher-median inventory

Configuration	Case	Native/reference	Native pooled ms	Reference pooled ms	Ratio	Native I/M/R	Reference I/M/R
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-64	cuda-cublas/torch-cuda-eager	3.378935	2.653832	1.27323	5/7/0; 5/7/0	5/7/0; 5/7/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-256	cuda-cublas/torch-cuda-eager	3.007370	2.410013	1.24787	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-1024	cuda-cublas/torch-cuda-eager	3.396927	2.468707	1.37599	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-2048	cuda-cublas/torch-cuda-eager	3.370900	2.440940	1.38098	3/5/0; 3/5/0	3/5/0; 3/5/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-64	cuda-cublas/torch-cuda-eager	3.063679	2.919957	1.04922	5/7/0; 5/7/0	5/7/0; 5/7/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-256	cuda-cublas/torch-cuda-eager	3.097362	2.682766	1.15454	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-1024	cuda-cublas/torch-cuda-eager	3.320677	2.706859	1.22676	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-2048	cuda-cublas/torch-cuda-eager	3.201679	2.457333	1.30291	3/5/0; 3/5/0	3/5/0; 3/5/0

Source, controls and raw evidence

• baseline/cuda-release: compiled 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7, tree 06f324529806d3edf005e4ecbb6a3c27f048f946, identity SHA256 7a956531880eaee793f22d75c4cc748f53c20ee8d5569fc0f2fe4cf67411d416. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: 760af83cc19b2e0c6b06f9373166d7bd7022d4a2a779677f5a3d141dec774240.

• current/cuda-release: compiled ce45ea9cc6d38cea05ef04cff8a321507e9684d5, tree 6cf00b82bac6f75e89093d1eb627f36d4fcc7d9b, identity SHA256 e7232ead6d14f51f9f6081b982c63c3fc3c9c553876d2f4daa052ba5c8fc3d47. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: f5ae10070dd259e06e629113ac3c101d8c0e2248a2b5810239d3779cba11de25.

Manifest SHA256: 81cef076a347463044875f94f61748628e415182c63862426a32667553731b9a. Plan SHA256: 64813c89d17e146815643d1e907a7ca489b13c4592eb31f031a474c801d72d21.

Every full strict report and actual comparator

Configuration	Verdict	Report SHA256	Comparator SHA256	Comparator path
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	inconclusive	a66bf49d4492f69b67417b860266e4027163cba18e470191e85232ecf5e5796a	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	inconclusive	41505e326b15bb1e07a8c6da3397954cc1cd0c05d859abee258fcd10f18f76b0	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	inconclusive	7d0d0e0ff2325964d4567c345094db98aad1ce63506cd241450246080a90ac3a	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	inconclusive	139f2cf9a95260deb1863e67d68263437fcc6901eb3d679d97af378c5d827a03	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	inconclusive	9eca7380972c803754fd05f29c5e1d19e8fc2faef0d74f2375a12bdef40d3792	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	inconclusive	86a06f3c93323000fe099c801ef136586c8cc11d7b219a45ea1a2a54a4af8a69	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	inconclusive	79a2bb2b1da2f4f1d1edaf079f7acab75ec745a843c3b915fffeec25d335221b	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	inconclusive	fac1daea85ccd4602dd3481e1f7f843c8bd9f8c4235f96f3a51edff1b16f4545	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/compare_performance.py

All original raw/import/log identities

Run	Raw path	Raw SHA256	Import SHA256	Log SHA256
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b01-baseline.json	c6b400c87a95f6d8dfafca285dcb2f0a93d15fdd4837d48465b521587ed0d066	67723d89710a992f438388095c5b0c3e877c18ba0da082430283e9a0c532d629	2f639791e6904266c89a514c1fc05b437bb756858b25a4388be8239c3883b8e9
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b02-current.json	38bbdefd8db8ce26bf54e0788a10e945d98ae9e5ad289374784b7afe009209ac	aa77909c2320b54828d2be46ebad906f04b3d7cd8b4647eb28537560209b075e	1a0197963b49f08222843859dd86f41076784aa5408823b616ac5bd290849b26
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b03-current.json	3cebff6a21c00e05011b7f9ee036d3aa9d408cf428f270a4136e95cab9d79131	0aacaeed4b11a7c08f5304659da5506319d3fee4eee35981b07cdd48d6a05dde	d8869c69a6c6734632a82ee82d5ff6cdef5378a36511d11b64ec060afa2956f1
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b04-baseline.json	6f1e3463127defbc40ca75e6f42c89ac326bbaf1145ef4f2f69929879d4e2a41	5d17b384e3c21933d1ab608db89b9052e134dc0f07460f105774edb97127d70f	fea5a2bca3be12a8f46f01504c04c21fbb1c65e9d064f38936046e69dd3fa25d
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b01-baseline.json	c95cf4c0cf56952e39a51735093db77c7ec6ea2a5b513d3dd760c25e326c0cad	857b8bb5771859da054154f014bbce8c8e275e6e4f9a21cb8c76f1ee3d911240	3c9d39285e37859c5ec9ebeb0c93da33f28cedee2b18932ac43086ae0301c981
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b02-current.json	884e085f724adc8ca03baed6542e445a08e4034ee2c14edbf781c82d6a0834ec	b6342cf6ef4fec2a8e195c1932f1865ed3ed6b4c7cb2b46977436248b2dc9a8a	952118457d7656b35e83c177103c2bc148f3f3253235b83d250304456be4a084
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b03-current.json	91be7f0f6e80375c3aed1c5f852fbba7d3c076d88f0e1b2452b46f75e22aa6f6	b7e7c15b275cbe20d2a24b2dbc355e4e8c61e89c2eff0c275fc5163963ccf93d	185c1e0496d20a92dffa6dc529b3f52f09af5d292a5b41263338b76e5b2fd798
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b04-baseline.json	904b8b3bcab9a496b6488d6fd5eeb77c5a94b7dd7ee1769d7a10d05612c7e35c	f0ce5bd5e1804d3bd0506916eee8d588c732222dfef3156b03069cadb63e68ef	a7615e6f7ad8ccb3228b609c6c69e6d5305793e53d4c729d6969d8a9055b9ef6
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b01-baseline.json	272c9d7064b839a8907f75b4e8a5af082a50dce4d69da5ba1501c8a63919f2f5	52b40da52dc445e24bb60e3af10d4fe08c17c562170156e0f94b90383e6b79b7	d0688f54e3d2c30ad189df584d42a02079465a1c435df99ed83c5a8bc7c804e2
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b02-current.json	34f2959d5d982da366bca2b7e3d61c2f26d3395d0c4a3ce1b9295519160ff7b5	06fb238a0256878611bffacde8fb6e600dc977c64c70e76a72ca13f2652878b2	8644006411534ba538a6bd9d76f09a73cfec7d4536c783df21e821111e207ece
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b03-current.json	e9bf22c5b7ce570bba0af26b643fdd77cefeff3e189882854de6ecf7c1af26f8	815674b0cad91fd6073a6b833c5264751cbf88ae82733b378cdaeefdc6d2bc61	7157dc3717dd04916bc164c9313cd106fcea9236bd4746eee33a3174cf1b2787
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b04-baseline.json	5ca87533ada3c689c603ef17ebb2025a1297e45e609456658b6d8226eec20a44	43c5a16487a55e9c5b8a383af953107162097e1340648032c2729f8d0ef718ed	f54a22ba8d6ff18ce88bef58fb264983a858ef1a4ef67da0b1dede793955ffa7
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b01-baseline.json	c87eb3a21b22787b2bea3e604899ac523d91d7f8b77fe830c63262fe6ede30eb	b46bd35ce92e888044d4d5e6018d1def57578bc171e1659f5226a0504ec4c87d	d0e14ca072b311b5f3a09c15585cd8c5fe2513868f71f8a44a3325d0367dbe89
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b02-current.json	6e6f905353dcd7f59faf56f570c7aadefafb2d360d63a3c1241e74b7610da9d4	9b96f3dcfecdad52fec7946486c6ca481d15cc718f55b950424bcff6bce6d680	434d2920c9ca1312ae45441c75f7c570bcceea49dd5bf02ccac8f6c3027dd21c
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b03-current.json	5aa194539fc162368dca41b44f6e80a33eda5af45ffbc883863d45aa42ba0c53	e923f43ecd1948706c669fa0d76d37412ba47d608f1a082a0c462187be04037c	0c3346cc9f09f3c12b72e783a810eab528b8f433b95bbd259f69937b72115dcc
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b04-baseline.json	a240d4490c960a787ced80121bfa41880989b6033a691d8790158ec112fcf60e	6adf96da1656b347251b97ba764c05b469c7be5f284e81d779e85483e15c5ca7	a85c34c2fbf85f3385f15e93af9baf3dba08266c6c9bc2dc813b046cc207c6fc
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b01-baseline.json	68ec064da128e2957bdbb8ade4ac4f5eed81f133241607408f315bf0db8dc39a	7e1053b06c1f67df915262021ac45cddf118957bfafcb9815e6266864280a843	687f6504f9c19904c05242f4cd9790d42cfed004b7cb4213306f4248bc294824
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b02-current.json	4695162abeaa7fd1493217d6c949e71fb2339b4091b5ec32e47b106639d3e315	4fd4ae8199d727b6a8f7db990550c250396336c37577dafbd83e2610bcc8a7c3	5a194fa5ddff982513fbe91b51f8bee1a8af1cf23615e8a89dc86567288fdc4a
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b03-current.json	2303d8777a448c8a86a0f8dc30ef13f129a0de5f0db046b4f31c08e265c9edef	47108681ffde1b287bac517f1c9c34a89137c09ae70e472756688afaf6c7cc5d	bf0fa244ece34b59ddd75bbaa7c45e5ff4bff3a286b9d84ffb8559a52119a9f9
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b04-baseline.json	e5b6568e446700a389c971dc4941c85110325c1c3e9177d08c6e9c862f7c9dd7	486959e54c70b524eaedb2ea211de8e627c93a524599f1b2645460edc4059c59	23a0abfcbe4fa0d2ad68294c17556241f19428ac40365db4a7e7dd200b222aab
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b01-baseline.json	1662a130bb70262d64a9380f9c7b27c541b90b6198a885bf27d21bfb2548c68d	ed4cc9b237cf2a1521be0e665f50a9a5c8a6b19dab5ddd604384acc4de417367	0b0bfdf6faf5f9e71177a22bacc30c36fb39eebf260dc9540a4515bab16f2414
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b02-current.json	a5db30505215400c5f9ac3bb7618d038147c4ef4b35ec3e1d8f31255ebf9cdc4	241600c901471558a970530eda6c3f97a7a6dc51f0a19c92f306fb6ae8f89718	fdddab3cc1f7d4f4091a9deeade048fff7e4b2c46544144af6b937624c5e5c11
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b03-current.json	c12d2774da83657630798f6cdc96c45b161ae78e7fd0fec69a4157cb60db75a6	cd6ac5f08339ce9ea715f3a3ffd8cf040500a52b34cd3cba688626bc9b16aa0a	b81ae318170331b53f9084675e593ac0bb32dbd16c0fa2ca2612856ac11981da
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b04-baseline.json	b5b7d49bc31f892adacb35380111fd28da4259a3717c956b553f4ac09a69e366	53f6458c93c8831db163e485ea85cd89f4a8c95bc33797fa0cca3ac556cd7b0d	698a856714307da3b62a553d874ff2f32d6cc2d5d14c2df9ee7c788ed1d4789b
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline.json	57a6093752f488d2eab0b2555878c7b75f0b9301026279b92bbe2e94684925a2	084f2129c97e53bf00455eb2a09437c59ac37f6c4de2223f69f82b433be8d8d6	3455a0d8ba5a7bfbc194ec930746d1f49d1fc75f71b79a62947726bf7eeb3b3f
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b02-current.json	bacfd6d390a9e8ea92a6e80b56971f3c34060a52881764775cd917910e05e46f	2163bd9076e9628c96545d835b20f841fbccb4a0e7c2a03b71a62e82e0fd2215	cccc5e71270c617a972943f00cecc502dd38a38b0d36be509dd4c4b4b4fde8b0
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b03-current.json	f5f3d84a98d157f71888ae99df100c212e6e7129b71eb715305aaa0344e3d209	a88f0688a16f111195926b801d860b53eb7aad481bc9ca4aa868a49367f6ded1	28dbbeb4c0a71d06c1a56830e79f159bfb1e1f65435e18209b058a2ecda0d744
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline.json	10bf2f05eafe33ac064d7f553d9fd84962be53d8465fce2010e3c48c9d46b6da	cb6f4f11f815a027df4bbd7eaf7f4a0d0a199d695e7981aa860cb0ac30b3977d	6b1aceee01497393aed19f230595069875893c4b6ea49d01215bb77def53c390
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline.json	f450bb046f0f065d03b48b5518caa478a5db9405c56c8e2ea36486d40b9c63bd	32791041f4e29d2d84c58bf7508c85da6775964142af8dfd18cc4846ac4bad88	4dfc872d25aa3a99f6e89db4ebfb45816c0729cc01e052d449955995055348be
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current.json	6f666bb30a5040d591c21f296d40df3f41f456fdf56df00b1302526cdd8fdd5c	f99dd6df6ee812180a3867ca75f5386b03a417b63b36394bbbf17e94adbf6e6f	949bccddddd2fe89463178333d99ab61e52d7a558a824cb6e7ae0386110195e0
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current.json	33f1142823afb9d44a600400909e163e59e85413b8a42749d338ec035c8c07f7	2bccb25c3719c26a73e749d4604f9bca6a32eaf615379f95eeea614059f20963	833cc846b04061ac1a0835ee17bef12eba141c0517f4ac67e7a315e35039dc76
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline	build/restart-optimization-20261017/fresh-span-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline.json	f651010cdb1c358a4f7dbc1715c8b75cdac815a5196f00be45ab74cdfa121b82	c56d445a28c9dc972c8804da8905a812356cd8f36eb83695e318ed9f8d06163e	5f8fb89f46b677e9e921d559611e63241194cf8871baf4426492549f9ab24398

cpu-quad-abba (cpu) fixed A/B campaign

Whole-campaign verdict: regression. All 64 artifacts/832 candidate/case outcomes are retained.

Every native and framework strict result

Candidate	Role	Rows	Within 5%	Inconclusive	Regression	Comparison failed
native	native	64	15	45	4	0
numpy-out	framework control	64	16	41	7	0
torch-eager-out	framework control	64	12	47	5	0
native-blas	native	16	5	9	2	0

All candidate rows

A/B and current/reference ratios below one indicate a lower measured numerator. Ranges bracket the two process medians and are not confidence intervals. Saved strict intervals compare current against A/B baseline; they do not give confidence for the separate current/native-to-framework comparison.

cpu-release-cpu-dense-f64-v1-none-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.042038	0.038106–0.050751	0.049491	0.048831–0.051365	1.17728	0.8556–1.3839	inconclusive	49/51/0	0.607749
dense-64	numpy-out	0.216004	0.202672–0.281401	0.285015	0.273992–0.295851	1.31949	0.9056–1.4646	inconclusive	49/51/0	0.61272
dense-64	torch-eager-out	0.349208	0.330628–0.447311	0.447082	0.427642–0.456652	1.28028	0.8212–1.4385	inconclusive	49/51/0	0.635707
dense-256	native	0.484480	0.477195–0.504274	0.498094	0.478643–0.508410	1.02810	0.9102–1.1178	inconclusive	67/70/1	0.971216
dense-256	numpy-out	0.770166	0.758203–0.802014	0.815955	0.713938–0.909397	1.05945	0.8368–1.2363	inconclusive	67/70/1	0.969964
dense-256	torch-eager-out	0.953616	0.897260–1.041708	1.156014	0.995309–1.283745	1.21224	0.9337–1.5040	inconclusive	67/70/1	0.972357
dense-1024	native	10.457527	10.387518–10.631519	9.524603	9.346824–9.629594	0.91079	0.8624–0.9462	within 5%	70/73/1	0.928014
dense-1024	numpy-out	11.039568	10.871386–11.102237	10.950899	10.752448–11.021052	0.99197	0.9367–1.0348	within 5%	70/73/1	0.928014
dense-1024	torch-eager-out	12.570985	12.399534–12.684242	12.355033	12.275438–12.510957	0.98282	0.9408–1.0310	within 5%	70/73/1	0.928014
dense-2048	native	119.740431	119.740431–119.972992	107.810716	106.164976–110.135357	0.90037	0.8584–0.9434	within 5%	70/73/1	0.871762
dense-2048	numpy-out	100.697592	100.697592–101.255465	100.093237	97.741563–109.928444	0.99400	0.9529–1.1086	inconclusive	70/73/1	0.871762
dense-2048	torch-eager-out	104.664960	102.070537–105.384689	104.753747	101.140635–112.854564	1.00085	0.9356–1.1143	inconclusive	70/73/1	0.871762

cpu-release-cpu-dense-f64-v1-none-api-fused0-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.043701	0.039690–0.050471	0.056100	0.055732–0.056546	1.28371	1.0628–1.4743	regression	49/51/0	0.607749
dense-64	numpy-out	0.227460	0.210323–0.245940	0.282634	0.276720–0.289480	1.24257	1.1026–1.4001	regression	49/51/0	0.61272
dense-64	torch-eager-out	0.356376	0.309134–0.368166	0.474930	0.467484–0.487223	1.33267	1.2425–1.5929	regression	49/51/0	0.635707
dense-256	native	0.468238	0.439736–0.499845	0.543973	0.514590–0.577717	1.16175	1.0068–1.3218	inconclusive	67/70/1	0.971216
dense-256	numpy-out	0.653466	0.566266–0.744281	0.795365	0.794628–0.802897	1.21715	1.0358–1.4971	inconclusive	67/70/1	0.969964
dense-256	torch-eager-out	0.823513	0.733635–1.035921	1.139810	1.124154–1.155962	1.38408	1.0387–1.6379	inconclusive	67/70/1	0.972357
dense-1024	native	10.071204	10.044321–10.112115	10.085772	9.335278–12.462783	1.00145	0.9015–1.2896	inconclusive	70/73/1	0.928014
dense-1024	numpy-out	10.558459	10.410975–10.758313	11.141518	10.589441–13.842702	1.05522	0.9630–1.4850	inconclusive	70/73/1	0.928014
dense-1024	torch-eager-out	11.759586	11.635023–11.910004	12.897099	11.769146–14.003522	1.09673	0.9551–1.2706	inconclusive	70/73/1	0.928014
dense-2048	native	121.653656	120.672128–125.237603	126.676158	118.025280–132.960107	1.04129	0.9309–1.1342	inconclusive	70/73/1	0.871762
dense-2048	numpy-out	100.415560	99.108128–101.349485	124.629762	115.835740–140.956800	1.24114	1.0831–1.4541	regression	70/73/1	0.871762
dense-2048	torch-eager-out	104.527313	104.310374–104.938545	124.295817	115.321748–133.386662	1.18912	1.0496–1.3138	inconclusive	70/73/1	0.871762

cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.045463	0.045366–0.045784	0.037671	0.031343–0.044030	0.82861	0.6312–1.1517	inconclusive	49/51/0	0.607749
dense-64	numpy-out	0.327890	0.326711–0.340554	0.247629	0.240754–0.255865	0.75522	0.6312–0.9312	within 5%	49/51/0	0.61272
dense-64	torch-eager-out	0.498279	0.467086–0.597500	0.394910	0.390265–0.403298	0.79255	0.5890–0.9342	within 5%	49/51/0	0.635707
dense-256	native	0.528154	0.523107–0.567871	0.472382	0.454319–0.490812	0.89440	0.7881–0.9677	within 5%	67/70/1	0.971216
dense-256	numpy-out	0.922941	0.907277–1.092906	0.748051	0.697568–0.840042	0.81051	0.6234–0.9491	within 5%	67/70/1	0.969964
dense-256	torch-eager-out	1.294853	1.251946–1.346327	0.992642	0.938895–1.006969	0.76661	0.6644–0.9430	within 5%	67/70/1	0.972357
dense-1024	native	12.838892	12.178964–13.983548	12.077255	12.030092–12.260506	0.94068	0.8419–1.0322	within 5%	70/73/1	0.928014
dense-1024	numpy-out	12.638963	12.230132–13.341102	13.211063	12.988356–13.958096	1.04526	0.9599–1.1665	inconclusive	70/73/1	0.928014
dense-1024	torch-eager-out	14.139568	13.556372–14.994195	14.601103	14.035477–15.166621	1.03264	0.9107–1.1328	inconclusive	70/73/1	0.928014
dense-2048	native	127.292558	125.456946–131.899644	136.774067	136.708186–137.852006	1.07449	1.0063–1.1304	inconclusive	70/73/1	0.871762
dense-2048	numpy-out	93.683631	93.655204–93.760593	112.315505	104.391305–123.373615	1.19888	1.0800–1.3547	regression	70/73/1	0.871762
dense-2048	torch-eager-out	106.899927	98.194232–117.437431	125.381565	121.470311–129.998575	1.17289	1.0313–1.3421	inconclusive	70/73/1	0.871762

cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.044189	0.044015–0.044189	0.043073	0.042543–0.044735	0.97474	0.8533–1.2089	inconclusive	49/51/0	0.607749
dense-64	numpy-out	0.272817	0.250077–0.290573	0.288411	0.269136–0.333082	1.05716	0.9070–1.3412	inconclusive	49/51/0	0.61272
dense-64	torch-eager-out	0.452307	0.417258–0.477467	0.491169	0.472341–0.548358	1.08592	0.9773–1.3500	inconclusive	49/51/0	0.635707
dense-256	native	0.489382	0.464647–0.538067	0.488546	0.466158–0.507574	0.99829	0.8481–1.0967	inconclusive	67/70/1	0.971216
dense-256	numpy-out	0.779444	0.764717–0.790898	0.881588	0.665701–0.936291	1.13105	0.8010–1.2492	inconclusive	67/70/1	0.969964
dense-256	torch-eager-out	0.962159	0.916911–1.047567	1.051166	0.911379–1.144756	1.09251	0.7830–1.2693	inconclusive	67/70/1	0.972357
dense-1024	native	12.382461	11.866444–13.120945	11.921061	11.712153–12.117420	0.96274	0.8793–1.0455	within 5%	70/73/1	0.928014
dense-1024	numpy-out	12.204106	12.077063–12.351791	12.829164	12.662799–13.057472	1.05122	0.9583–1.1238	inconclusive	70/73/1	0.928014
dense-1024	torch-eager-out	13.587476	13.060068–13.878958	14.239138	14.059055–14.829439	1.04796	0.9832–1.1491	inconclusive	70/73/1	0.928014
dense-2048	native	145.815435	136.375575–156.510558	127.130635	125.981415–130.231523	0.87186	0.7525–0.9649	within 5%	70/73/1	0.871762
dense-2048	numpy-out	101.841864	96.127741–114.278188	94.764996	94.632415–96.747348	0.93051	0.7903–1.0170	within 5%	70/73/1	0.871762
dense-2048	torch-eager-out	125.922307	123.735653–132.946276	113.048113	108.448492–118.731271	0.89776	0.8060–0.9825	within 5%	70/73/1	0.871762

cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.015482	0.014863–0.017413	0.018815	0.017754–0.019886	1.21528	0.9796–1.5646	inconclusive	5/7/0	0.0796191
dense-64	numpy-out	0.042320	0.041096–0.044341	0.060068	0.050837–0.063053	1.41936	1.0602–1.5659	regression	5/7/0	0.0796191
dense-64	torch-eager-out	0.073302	0.070135–0.088708	0.113624	0.099184–0.118639	1.55008	1.0727–1.7043	regression	5/7/0	0.0796191
dense-256	native	0.056544	0.047074–0.073399	0.061164	0.050009–0.064848	1.08170	0.6681–1.3943	inconclusive	4/6/0	0.209854
dense-256	numpy-out	0.076975	0.062198–0.096728	0.097557	0.084087–0.107437	1.26739	0.8499–1.7395	inconclusive	4/6/0	0.209854
dense-256	torch-eager-out	0.136835	0.096295–0.186451	0.169849	0.120029–0.180671	1.24127	0.6211–1.8953	inconclusive	4/6/0	0.209854
dense-1024	native	0.945860	0.934688–0.971317	0.930875	0.923060–0.952358	0.98416	0.9270–1.0346	within 5%	4/6/0	0.013684
dense-1024	numpy-out	1.073007	1.067175–1.073007	1.072718	1.072718–1.080943	0.99973	0.9443–1.0792	inconclusive	4/6/0	0.013684
dense-1024	torch-eager-out	1.218091	1.191566–1.252444	1.293532	1.282047–1.298506	1.06193	0.9908–1.1173	inconclusive	4/6/0	0.013684
dense-2048	native	7.578267	7.475696–7.671149	6.310298	5.954938–7.206009	0.83268	0.7238–1.1381	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	6.197669	6.147465–6.220285	6.030121	5.595532–6.529207	0.97297	0.7832–1.1295	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	6.268376	6.021684–6.792315	6.551185	6.018681–7.112641	1.04512	0.7884–1.2038	inconclusive	3/5/0	0.836412

cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.018055	0.016443–0.020211	0.026266	0.023141–0.031273	1.45478	1.0569–1.9263	regression	5/7/0	0.0796191
dense-64	numpy-out	0.044401	0.043075–0.049570	0.069025	0.067081–0.069351	1.55458	1.0411–1.7027	inconclusive	5/7/0	0.0796191
dense-64	torch-eager-out	0.077837	0.073405–0.089402	0.133079	0.131712–0.133870	1.70971	1.3192–1.8631	regression	5/7/0	0.0796191
dense-256	native	0.071806	0.070410–0.072629	0.068281	0.063453–0.070042	0.95092	0.8520–1.0385	within 5%	4/6/0	0.209854
dense-256	numpy-out	0.105774	0.103247–0.114004	0.102357	0.102357–0.103022	0.96770	0.8428–1.0588	inconclusive	4/6/0	0.209854
dense-256	torch-eager-out	0.192489	0.174966–0.218127	0.178041	0.178041–0.179604	0.92494	0.7600–1.0596	inconclusive	4/6/0	0.209854
dense-1024	native	0.988382	0.893336–1.130740	0.970795	0.908275–1.023026	0.98221	0.7956–1.1509	inconclusive	4/6/0	0.013684
dense-1024	numpy-out	1.054271	0.967160–1.215243	1.108482	1.095954–1.132640	1.05142	0.8548–1.1794	inconclusive	4/6/0	0.013684
dense-1024	torch-eager-out	1.241844	1.178023–1.516934	1.316274	1.234677–1.401963	1.05993	0.8121–1.1986	inconclusive	4/6/0	0.013684
dense-2048	native	7.820111	6.628355–8.300432	6.418099	6.185906–7.941310	0.82072	0.7322–1.2210	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	6.890287	6.111979–7.544366	5.811723	5.578374–6.531288	0.84347	0.7145–1.1303	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	7.229095	6.420790–7.605319	6.444130	6.038529–6.731510	0.89142	0.7491–1.0884	inconclusive	3/5/0	0.836412

cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.008655	0.007138–0.009232	0.009581	0.009089–0.009911	1.10711	0.9807–1.4163	inconclusive	5/7/0	0.0796191
dense-64	numpy-out	0.044408	0.029879–0.061859	0.049668	0.048962–0.051452	1.11845	0.7714–1.7424	inconclusive	5/7/0	0.0796191
dense-64	torch-eager-out	0.080779	0.046984–0.112021	0.093152	0.091557–0.094792	1.15318	0.7770–2.0244	inconclusive	5/7/0	0.0796191
dense-256	native	0.053611	0.047677–0.069941	0.050781	0.044278–0.057374	0.94720	0.6241–1.2077	inconclusive	4/6/0	0.209854
dense-256	numpy-out	0.086014	0.076028–0.118541	0.087477	0.072407–0.092425	1.01700	0.6140–1.2599	inconclusive	4/6/0	0.209854
dense-256	torch-eager-out	0.128025	0.115958–0.153790	0.128892	0.091393–0.142948	1.00678	0.5765–1.2676	inconclusive	4/6/0	0.209854
dense-1024	native	1.540671	1.540671–1.543071	1.506552	1.481142–1.566455	0.97785	0.8061–1.1950	inconclusive	4/6/0	0.013684
dense-1024	numpy-out	1.266905	1.265508–1.295896	1.369044	1.338928–1.391082	1.08062	0.9194–1.1694	inconclusive	4/6/0	0.013684
dense-1024	torch-eager-out	1.625737	1.355144–1.672821	1.602343	1.563997–1.630384	0.98561	0.8224–1.2700	inconclusive	4/6/0	0.013684
dense-2048	native	9.262180	9.130049–9.298321	9.171795	8.272897–9.371507	0.99024	0.8407–1.1187	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	7.071290	6.493213–7.470228	7.947388	7.598876–8.372616	1.12390	0.8922–1.3583	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	7.970667	7.752719–8.335994	7.689222	7.677315–7.693801	0.96469	0.8748–1.0397	within 5%	3/5/0	0.836412

cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.008649	0.008121–0.009994	0.009292	0.009088–0.009628	1.07434	0.8695–1.2092	inconclusive	5/7/0	0.0796191
dense-64	numpy-out	0.031178	0.030160–0.035552	0.047380	0.046147–0.048661	1.51970	1.1342–1.6384	regression	5/7/0	0.0796191
dense-64	torch-eager-out	0.052431	0.049261–0.059672	0.083480	0.073039–0.090913	1.59220	1.0445–1.8486	inconclusive	5/7/0	0.0796191
dense-256	native	0.045408	0.043624–0.048582	0.059101	0.057912–0.059875	1.30156	1.1131–1.4361	regression	4/6/0	0.209854
dense-256	numpy-out	0.061679	0.060687–0.061847	0.093931	0.088812–0.098358	1.52289	1.2777–1.6840	regression	4/6/0	0.209854
dense-256	torch-eager-out	0.089441	0.072542–0.118157	0.135785	0.118618–0.140973	1.51814	0.9764–1.9804	inconclusive	4/6/0	0.209854
dense-1024	native	1.340797	1.242754–1.457267	1.475731	1.411318–1.494826	1.10064	0.8743–1.2591	inconclusive	4/6/0	0.013684
dense-1024	numpy-out	1.303882	1.199370–1.381997	1.447552	1.363893–1.509780	1.11019	0.8695–1.3175	inconclusive	4/6/0	0.013684
dense-1024	torch-eager-out	1.429414	1.357732–1.525151	1.608480	1.608480–1.628892	1.12527	0.8984–1.3639	inconclusive	4/6/0	0.013684
dense-2048	native	9.887118	9.368645–11.046350	8.704638	8.451689–8.994224	0.88040	0.7250–1.0989	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	6.467257	6.123685–7.784951	6.423397	6.288486–6.519978	0.99322	0.7681–1.0990	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	7.605853	7.340197–8.581200	8.045276	7.441446–8.959401	1.05777	0.8298–1.3102	inconclusive	3/5/0	0.836412

cpu-blas-cpu-dense-f64-v1-none-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.052203	0.050818–0.058204	0.048942	0.045938–0.050340	0.93753	0.7330–1.0512	inconclusive	49/51/0	0.607749
dense-64	native-blas	0.045922	0.045256–0.055877	0.042667	0.040752–0.046194	0.92911	0.6641–1.0448	within 5%	49/51/0	0.612054
dense-64	numpy-out	0.301914	0.299313–0.331434	0.278184	0.259766–0.290822	0.92140	0.7427–1.0029	within 5%	49/51/0	0.61272
dense-64	torch-eager-out	0.513254	0.478587–0.584865	0.449313	0.410478–0.484636	0.87542	0.6716–1.0658	inconclusive	49/51/0	0.635707
dense-256	native	0.485963	0.474413–0.493465	0.481976	0.468123–0.488113	0.99180	0.9390–1.0736	inconclusive	67/70/1	0.971216
dense-256	native-blas	0.474730	0.470595–0.476241	0.335533	0.328861–0.346406	0.70679	0.6620–0.7755	within 5%	67/70/1	0.979189
dense-256	numpy-out	0.847515	0.847515–0.849541	0.707725	0.698615–0.721323	0.83506	0.7764–0.8846	within 5%	67/70/1	0.969964
dense-256	torch-eager-out	1.238738	1.177280–1.268535	0.991865	0.945593–1.002665	0.80071	0.7274–0.8891	within 5%	67/70/1	0.972357
dense-1024	native	9.789756	9.729269–9.966482	9.172807	9.167500–9.216453	0.93698	0.8885–0.9825	within 5%	70/73/1	0.928014
dense-1024	native-blas	9.323401	8.800693–9.377906	8.974327	8.730684–9.119744	0.96256	0.9130–1.0547	inconclusive	70/73/1	0.928014
dense-1024	numpy-out	10.328694	10.263386–10.421463	10.348399	10.335130–10.373466	1.00191	0.9504–1.0614	inconclusive	70/73/1	0.928014
dense-1024	torch-eager-out	12.118351	11.943413–12.273738	11.932993	11.814045–12.045082	0.98470	0.9409–1.0299	within 5%	70/73/1	0.928014
dense-2048	native	110.671547	109.987273–111.026158	123.736721	116.248153–125.338402	1.11805	1.0145–1.1688	inconclusive	70/73/1	0.871762
dense-2048	native-blas	99.501520	97.522853–101.044513	103.599949	98.018648–111.902693	1.04119	0.9477–1.1643	inconclusive	70/73/1	0.871762
dense-2048	numpy-out	106.337981	105.077995–107.749245	112.036158	103.604195–117.647297	1.05359	0.9377–1.1743	inconclusive	70/73/1	0.871762
dense-2048	torch-eager-out	107.797856	105.469099–111.969113	117.203422	116.471084–119.352217	1.08725	0.9841–1.1644	inconclusive	70/73/1	0.871762

cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.042750	0.042500–0.044368	0.039211	0.036607–0.040938	0.91722	0.7414–1.0045	within 5%	49/51/0	0.607749
dense-64	native-blas	0.035370	0.031116–0.041829	0.035631	0.030812–0.037082	1.00735	0.7080–1.4083	inconclusive	49/51/0	0.612054
dense-64	numpy-out	0.315423	0.309986–0.358909	0.289305	0.287069–0.295054	0.91720	0.7653–0.9933	within 5%	49/51/0	0.61272
dense-64	torch-eager-out	0.535139	0.514803–0.588260	0.479405	0.475867–0.485261	0.89585	0.7372–0.9856	within 5%	49/51/0	0.635707
dense-256	native	0.515799	0.515799–0.516112	0.469185	0.469185–0.476060	0.90963	0.8662–0.9806	within 5%	67/70/1	0.971216
dense-256	native-blas	0.493503	0.487303–0.499342	0.472686	0.467697–0.482542	0.95782	0.9162–1.0132	within 5%	67/70/1	0.979189
dense-256	numpy-out	0.804721	0.797051–0.806351	0.689586	0.683435–0.706800	0.85693	0.8199–0.9606	within 5%	67/70/1	0.969964
dense-256	torch-eager-out	1.160993	1.155217–1.173793	0.968673	0.966001–0.982948	0.83435	0.7804–0.9033	within 5%	67/70/1	0.972357
dense-1024	native	13.499565	12.709122–14.318356	12.587576	12.572033–12.631638	0.93244	0.8608–1.0176	within 5%	70/73/1	0.928014
dense-1024	native-blas	12.561340	12.239774–14.190210	11.810638	11.726218–12.150629	0.94024	0.8006–1.0021	within 5%	70/73/1	0.928014
dense-1024	numpy-out	14.059589	13.411640–15.053088	13.707803	13.653808–13.852920	0.97498	0.8633–1.0539	inconclusive	70/73/1	0.928014
dense-1024	torch-eager-out	15.232900	14.280247–16.860312	14.936497	14.800920–15.031305	0.98054	0.8434–1.0609	inconclusive	70/73/1	0.928014
dense-2048	native	143.625237	130.627575–155.832783	133.696443	133.594631–133.755485	0.93087	0.8364–1.0575	inconclusive	70/73/1	0.871762
dense-2048	native-blas	108.001609	104.311105–109.439793	102.202950	101.193609–102.789967	0.94631	0.8705–1.0045	within 5%	70/73/1	0.871762
dense-2048	numpy-out	109.609177	103.320093–112.790036	104.550442	104.293179–106.001856	0.95385	0.9019–1.0343	within 5%	70/73/1	0.871762
dense-2048	torch-eager-out	107.946269	100.618544–116.453821	112.504743	112.278622–113.357761	1.04223	0.9334–1.1629	inconclusive	70/73/1	0.871762

cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.014847	0.013881–0.015772	0.014541	0.014461–0.016442	0.97939	0.7870–1.2103	inconclusive	5/7/0	0.0796191
dense-64	native-blas	0.017086	0.016679–0.017636	0.017209	0.016875–0.017470	1.00717	0.7632–1.2407	inconclusive	5/7/0	0.0796191
dense-64	numpy-out	0.060934	0.056723–0.067666	0.056890	0.053209–0.061480	0.93364	0.6998–1.1937	inconclusive	5/7/0	0.0796191
dense-64	torch-eager-out	0.112548	0.108313–0.118320	0.110706	0.108995–0.110908	0.98363	0.7180–1.1367	inconclusive	5/7/0	0.0796191
dense-256	native	0.057264	0.052848–0.059606	0.059688	0.052565–0.062391	1.04234	0.8627–1.2402	inconclusive	4/6/0	0.209854
dense-256	native-blas	0.055284	0.047932–0.062789	0.057804	0.047782–0.061405	1.04557	0.7466–1.3129	inconclusive	4/6/0	0.209854
dense-256	numpy-out	0.097103	0.096566–0.098763	0.100959	0.089750–0.108750	1.03971	0.8777–1.1834	inconclusive	4/6/0	0.209854
dense-256	torch-eager-out	0.161055	0.153261–0.167513	0.183249	0.154486–0.201778	1.13780	0.8702–1.3616	inconclusive	4/6/0	0.209854
dense-1024	native	1.067340	1.026217–1.081611	0.936248	0.917486–0.955413	0.87718	0.7897–0.9557	within 5%	4/6/0	0.013684
dense-1024	native-blas	0.978814	0.945925–1.008271	0.942434	0.922166–0.958318	0.96283	0.8960–1.1212	inconclusive	4/6/0	0.013684
dense-1024	numpy-out	1.161325	1.137908–1.229516	1.147377	1.127992–1.215865	0.98799	0.8781–1.0963	inconclusive	4/6/0	0.013684
dense-1024	torch-eager-out	1.428410	1.337514–1.668497	1.439147	1.327090–1.599923	1.00752	0.7814–1.2015	inconclusive	4/6/0	0.013684
dense-2048	native	7.299926	6.774863–7.713626	7.031703	6.988801–7.241459	0.96326	0.7658–1.2149	inconclusive	3/5/0	0.836412
dense-2048	native-blas	6.587444	6.587444–6.788151	5.788287	5.470110–6.665693	0.87868	0.7032–1.2306	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	6.113086	5.953421–6.484996	6.337937	6.109414–6.454701	1.03678	0.8394–1.1293	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	6.767993	6.690172–7.052211	7.209759	6.531131–7.397428	1.06527	0.8234–1.1824	inconclusive	3/5/0	0.836412

cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.007047	0.006861–0.007057	0.008824	0.008393–0.009202	1.25208	1.0984–1.4183	regression	5/7/0	0.0796191
dense-64	native-blas	0.006944	0.006944–0.007032	0.008604	0.008192–0.009921	1.23907	1.0791–1.5032	regression	5/7/0	0.0796191
dense-64	numpy-out	0.032341	0.032283–0.034788	0.050469	0.044636–0.056384	1.56055	1.1280–1.7997	regression	5/7/0	0.0796191
dense-64	torch-eager-out	0.051829	0.050334–0.054172	0.090479	0.088161–0.101347	1.74572	1.4553–2.0363	regression	5/7/0	0.0796191
dense-256	native	0.057734	0.055585–0.062919	0.058722	0.057654–0.059852	1.01712	0.8787–1.2278	inconclusive	4/6/0	0.209854
dense-256	native-blas	0.043659	0.039216–0.044876	0.056848	0.055963–0.062053	1.30209	1.0880–1.6689	regression	4/6/0	0.209854
dense-256	numpy-out	0.095132	0.085542–0.105620	0.092943	0.082989–0.095230	0.97699	0.7734–1.1202	inconclusive	4/6/0	0.209854
dense-256	torch-eager-out	0.139974	0.120566–0.160095	0.145616	0.142889–0.154712	1.04031	0.8298–1.3136	inconclusive	4/6/0	0.209854
dense-1024	native	1.889512	1.697399–2.029024	1.728854	1.692831–1.743683	0.91497	0.7930–1.0839	inconclusive	4/6/0	0.013684
dense-1024	native-blas	1.581378	1.448908–1.795345	1.516158	1.488968–1.567935	0.95876	0.7948–1.1165	inconclusive	4/6/0	0.013684
dense-1024	numpy-out	1.683308	1.555769–1.842799	1.693000	1.531552–1.859689	1.00576	0.8017–1.2105	inconclusive	4/6/0	0.013684
dense-1024	torch-eager-out	2.010719	1.941064–2.286630	2.210337	2.010807–2.497941	1.09928	0.8737–1.3293	inconclusive	4/6/0	0.013684
dense-2048	native	10.465901	9.743443–11.323823	11.146144	9.757210–12.855193	1.06500	0.7962–1.4271	inconclusive	3/5/0	0.836412
dense-2048	native-blas	7.867612	7.731559–7.924334	9.403459	8.300538–10.828259	1.19521	0.9958–1.4636	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	7.502494	7.222422–7.526651	9.116241	8.475718–9.500371	1.21509	0.9793–1.4016	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	10.291389	9.514822–11.283906	9.008320	8.155505–9.867854	0.87533	0.7053–1.1179	inconclusive	3/5/0	0.836412

cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.069964	0.068458–0.070649	0.069852	0.069852–0.069875	0.99841	0.9319–1.1218	inconclusive	49/51/0	0.607749
dense-64	numpy-out	0.208315	0.204798–0.212701	0.215673	0.215558–0.216716	1.03532	0.9766–1.0821	inconclusive	49/51/0	0.61272
dense-64	torch-eager-out	0.333528	0.333102–0.333589	0.332829	0.326782–0.343045	0.99790	0.9624–1.0917	inconclusive	49/51/0	0.635707
dense-256	native	0.358784	0.338190–0.364019	0.373569	0.367424–0.381783	1.04121	0.9802–1.1643	inconclusive	67/70/1	0.993806
dense-256	numpy-out	0.847326	0.794802–0.869617	0.579588	0.537435–0.632835	0.68402	0.6108–0.8021	within 5%	67/70/1	0.969964
dense-256	torch-eager-out	1.005043	0.885763–1.060450	0.791250	0.742814–0.841740	0.78728	0.6957–0.9616	within 5%	67/70/1	0.972357
dense-1024	native	12.250766	12.162724–12.328628	10.817203	10.478015–11.305290	0.88298	0.7617–1.0325	within 5%	70/73/1	0.928014
dense-1024	numpy-out	13.088453	12.613152–14.200572	11.908826	11.781162–11.908826	0.90987	0.7850–1.0171	within 5%	70/73/1	0.928014
dense-1024	torch-eager-out	16.836122	16.436981–16.963508	13.578562	13.508268–14.270492	0.80651	0.6936–0.9190	within 5%	70/73/1	0.928014
dense-2048	native	73.455928	71.248919–73.841260	68.010165	66.082171–68.345557	0.92586	0.8208–1.1301	inconclusive	70/73/1	0.871762
dense-2048	numpy-out	69.592870	69.375901–70.299243	66.694926	65.495268–68.728290	0.95836	0.9098–1.0138	within 5%	70/73/1	0.871762
dense-2048	torch-eager-out	68.273240	67.451418–73.251066	68.532064	62.516466–71.412288	1.00379	0.7707–1.1581	inconclusive	70/73/1	0.871762

cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.052873	0.052781–0.053293	0.052191	0.051741–0.053314	0.98710	0.9367–1.0622	inconclusive	49/51/0	0.607749
dense-64	numpy-out	0.203425	0.197418–0.204746	0.198362	0.196043–0.200947	0.97511	0.9454–1.0437	within 5%	49/51/0	0.61272
dense-64	torch-eager-out	0.296554	0.295398–0.298700	0.295524	0.291297–0.297707	0.99653	0.9312–1.0604	inconclusive	49/51/0	0.635707
dense-256	native	0.352040	0.324651–0.378273	0.387538	0.370636–0.418844	1.10084	0.9738–1.2902	inconclusive	67/70/1	0.993806
dense-256	numpy-out	0.670878	0.645710–0.679493	0.752991	0.666762–0.858802	1.12240	0.9669–1.3389	inconclusive	67/70/1	0.969964
dense-256	torch-eager-out	0.805469	0.783343–0.812639	0.932033	0.874379–1.010854	1.15713	1.0520–1.3257	regression	67/70/1	0.972357
dense-1024	native	14.235617	13.997974–15.551402	14.656363	13.866990–15.530318	1.02956	0.8691–1.1396	inconclusive	70/73/1	0.928014
dense-1024	numpy-out	13.915693	13.149484–14.206482	12.655776	12.444251–13.444104	0.90946	0.8305–1.2358	inconclusive	70/73/1	0.928014
dense-1024	torch-eager-out	16.948152	15.352688–17.226251	19.039321	17.513781–20.755053	1.12339	0.9750–1.3707	inconclusive	70/73/1	0.928014
dense-2048	native	88.003933	85.039827–88.974012	104.817928	93.166456–121.083851	1.19106	0.9472–1.5189	inconclusive	70/73/1	0.871762
dense-2048	numpy-out	69.109776	67.298872–71.240920	71.412644	66.189376–74.469460	1.03332	0.9079–1.1486	inconclusive	70/73/1	0.871762
dense-2048	torch-eager-out	89.510291	87.756258–93.114704	82.778267	73.905640–88.191157	0.92479	0.7699–1.2826	inconclusive	70/73/1	0.871762

cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.020064	0.019838–0.020608	0.019353	0.018535–0.023335	0.96459	0.8891–1.2299	inconclusive	5/7/0	0.0796191
dense-64	numpy-out	0.042443	0.042125–0.042527	0.043236	0.041378–0.051187	1.01871	0.9457–1.2786	inconclusive	5/7/0	0.0796191
dense-64	torch-eager-out	0.072801	0.071981–0.073389	0.072694	0.069410–0.093528	0.99854	0.9314–1.3434	inconclusive	5/7/0	0.0796191
dense-256	native	0.050425	0.045469–0.055061	0.059419	0.044521–0.084257	1.17835	0.7904–1.9746	inconclusive	4/6/0	0.209854
dense-256	numpy-out	0.084556	0.071368–0.098680	0.067947	0.065181–0.071235	0.80357	0.6418–1.0113	within 5%	4/6/0	0.209854
dense-256	torch-eager-out	0.143461	0.110938–0.157150	0.146321	0.109553–0.201478	1.01994	0.6818–1.8696	inconclusive	4/6/0	0.209854
dense-1024	native	1.028299	0.599132–1.085761	0.979938	0.970577–1.021329	0.95297	0.8683–1.7514	inconclusive	4/6/0	0.013684
dense-1024	numpy-out	1.237694	1.132184–2.147800	1.041165	0.996602–1.112097	0.84121	0.4559–1.0221	within 5%	4/6/0	0.013684
dense-1024	torch-eager-out	1.242117	1.042675–1.277740	1.239275	1.194782–1.257710	0.99771	0.8720–1.3148	inconclusive	4/6/0	0.013684
dense-2048	native	9.441247	8.869248–10.619672	8.575426	7.893645–9.268961	0.90829	0.7041–1.1720	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	7.223762	6.884082–8.432111	8.142355	7.895445–8.336089	1.12716	0.8448–1.4780	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	9.626836	9.100214–10.111403	9.197698	8.896773–9.510724	0.95542	0.7923–1.2398	inconclusive	3/5/0	0.836412

cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	native	0.011095	0.011041–0.011154	0.010927	0.010731–0.011123	0.98490	0.9572–1.0315	within 5%	5/7/0	0.0796191
dense-64	numpy-out	0.030765	0.030210–0.031339	0.029743	0.029639–0.029792	0.96678	0.9048–1.0010	within 5%	5/7/0	0.0796191
dense-64	torch-eager-out	0.048301	0.048056–0.048601	0.046539	0.046490–0.046539	0.96353	0.9362–0.9873	within 5%	5/7/0	0.0796191
dense-256	native	0.033208	0.030194–0.044512	0.034356	0.032236–0.037651	1.03457	0.7090–1.2725	inconclusive	4/6/0	0.209854
dense-256	numpy-out	0.065110	0.049401–0.093093	0.054981	0.050080–0.060639	0.84442	0.5343–1.2423	inconclusive	4/6/0	0.209854
dense-256	torch-eager-out	0.082896	0.061449–0.118122	0.068386	0.067404–0.068710	0.82495	0.5532–1.1589	inconclusive	4/6/0	0.209854
dense-1024	native	0.959715	0.942902–0.993173	1.441454	1.049789–1.690936	1.50196	0.4872–2.0268	inconclusive	4/6/0	0.013684
dense-1024	numpy-out	1.839719	1.825372–1.839719	1.922732	1.590231–2.030394	1.04512	0.7251–1.6261	inconclusive	4/6/0	0.013684
dense-1024	torch-eager-out	1.399378	1.300302–1.592701	1.725651	1.336782–2.351413	1.23316	0.4319–2.0907	inconclusive	4/6/0	0.013684
dense-2048	native	12.106832	11.156087–12.522657	12.848515	10.737441–13.488679	1.06126	0.7930–1.3025	inconclusive	3/5/0	0.836412
dense-2048	numpy-out	9.912330	9.806442–9.990468	9.816973	9.095509–9.816973	0.99038	0.7649–1.1819	inconclusive	3/5/0	0.836412
dense-2048	torch-eager-out	12.684347	10.877647–13.940584	13.091261	12.016633–14.441135	1.03208	0.8367–1.3580	inconclusive	3/5/0	0.836412

Current-source framework comparisons

These are descriptive ratios of current pooled-round medians from the same matched configuration, not strict comparisons across frameworks or scopes. Every higher current-native median remains listed below.

All current matched scopes

Configuration	Native	Reference	Lower/higher/equal	Current ratio range	Same known I/M/R
cpu-release-cpu-dense-f64-v1-none-api-fused1-host	native	numpy-out	3/1/0	0.1736–1.0771	True
cpu-release-cpu-dense-f64-v1-none-api-fused1-host	native	torch-eager-out	3/1/0	0.1107–1.0292	True
cpu-release-cpu-dense-f64-v1-none-api-fused0-host	native	numpy-out	3/1/0	0.1985–1.0164	True
cpu-release-cpu-dense-f64-v1-none-api-fused0-host	native	torch-eager-out	3/1/0	0.1181–1.0192	True
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host	native	numpy-out	3/1/0	0.1521–1.2178	True
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host	native	torch-eager-out	3/1/0	0.0954–1.0909	True
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host	native	numpy-out	3/1/0	0.1493–1.3415	True
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host	native	torch-eager-out	3/1/0	0.0877–1.1246	True
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host	native	numpy-out	3/1/0	0.3132–1.0465	True
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host	native	torch-eager-out	4/0/0	0.1656–0.9632	True
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host	native	numpy-out	3/1/0	0.3805–1.1043	True
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host	native	torch-eager-out	4/0/0	0.1974–0.9960	True
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native	numpy-out	2/2/0	0.1929–1.1541	True
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native	torch-eager-out	3/1/0	0.1029–1.1928	True
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host	native	numpy-out	2/2/0	0.1961–1.3551	True
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host	native	torch-eager-out	3/1/0	0.1113–1.0820	True
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host	native	numpy-out	3/1/0	0.1759–1.1044	True
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host	native	torch-eager-out	3/1/0	0.1089–1.0557	True
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host	native-blas	numpy-out	4/0/0	0.1534–0.9247	True
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host	native-blas	torch-eager-out	4/0/0	0.0950–0.8839	True
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host	native	numpy-out	3/1/0	0.1355–1.2788	True
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host	native	torch-eager-out	3/1/0	0.0818–1.1884	True
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host	native-blas	numpy-out	4/0/0	0.1232–0.9775	True
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host	native-blas	torch-eager-out	4/0/0	0.0743–0.9084	True
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host	native	numpy-out	3/1/0	0.2556–1.1095	True
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host	native	torch-eager-out	4/0/0	0.1313–0.9753	True
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host	native-blas	numpy-out	4/0/0	0.3025–0.9133	True
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host	native-blas	torch-eager-out	4/0/0	0.1554–0.8028	True
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native	numpy-out	2/2/0	0.1748–1.2227	True
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native	torch-eager-out	3/1/0	0.0975–1.2373	True
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native-blas	numpy-out	3/1/0	0.1705–1.0315	True
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native-blas	torch-eager-out	3/1/0	0.0951–1.0439	True
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host	native	numpy-out	3/1/0	0.3239–1.0197	True
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host	native	torch-eager-out	4/0/0	0.2099–0.9924	True
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host	native	numpy-out	2/2/0	0.2631–1.4678	True
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host	native	torch-eager-out	3/1/0	0.1766–1.2662	True
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host	native	numpy-out	3/1/0	0.4476–1.0532	True
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host	native	torch-eager-out	4/0/0	0.2662–0.9323	True
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native	numpy-out	3/1/0	0.3674–1.3088	True
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host	native	torch-eager-out	4/0/0	0.2348–0.9815	True

Complete current higher-median inventory

Configuration	Case	Native/reference	Native pooled ms	Reference pooled ms	Ratio	Native I/M/R	Reference I/M/R
cpu-release-cpu-dense-f64-v1-none-api-fused1-host	dense-2048	native/numpy-out	107.810716	100.093237	1.07710	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-none-api-fused1-host	dense-2048	native/torch-eager-out	107.810716	104.753747	1.02918	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-none-api-fused0-host	dense-2048	native/numpy-out	126.676158	124.629762	1.01642	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-none-api-fused0-host	dense-2048	native/torch-eager-out	126.676158	124.295817	1.01915	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host	dense-2048	native/numpy-out	136.774067	112.315505	1.21777	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host	dense-2048	native/torch-eager-out	136.774067	125.381565	1.09086	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host	dense-2048	native/numpy-out	127.130635	94.764996	1.34154	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host	dense-2048	native/torch-eager-out	127.130635	113.048113	1.12457	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host	dense-2048	native/numpy-out	6.310298	6.030121	1.04646	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host	dense-2048	native/numpy-out	6.418099	5.811723	1.10434	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-1024	native/numpy-out	1.506552	1.369044	1.10044	4/6/0; 4/6/0	4/6/0; 4/6/0
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-2048	native/numpy-out	9.171795	7.947388	1.15406	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-2048	native/torch-eager-out	9.171795	7.689222	1.19281	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host	dense-1024	native/numpy-out	1.475731	1.447552	1.01947	4/6/0; 4/6/0	4/6/0; 4/6/0
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host	dense-2048	native/numpy-out	8.704638	6.423397	1.35515	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host	dense-2048	native/torch-eager-out	8.704638	8.045276	1.08196	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host	dense-2048	native/numpy-out	123.736721	112.036158	1.10444	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host	dense-2048	native/torch-eager-out	123.736721	117.203422	1.05574	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host	dense-2048	native/numpy-out	133.696443	104.550442	1.27877	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host	dense-2048	native/torch-eager-out	133.696443	112.504743	1.18836	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host	dense-2048	native/numpy-out	7.031703	6.337937	1.10946	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-1024	native/numpy-out	1.728854	1.693000	1.02118	4/6/0; 4/6/0	4/6/0; 4/6/0
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-2048	native/numpy-out	11.146144	9.116241	1.22267	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-2048	native/torch-eager-out	11.146144	9.008320	1.23732	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-2048	native-blas/numpy-out	9.403459	9.116241	1.03151	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-2048	native-blas/torch-eager-out	9.403459	9.008320	1.04386	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host	dense-2048	native/numpy-out	68.010165	66.694926	1.01972	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host	dense-1024	native/numpy-out	14.656363	12.655776	1.15808	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host	dense-2048	native/numpy-out	104.817928	71.412644	1.46778	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host	dense-2048	native/torch-eager-out	104.817928	82.778267	1.26625	70/73/1; 70/73/1	70/73/1; 70/73/1
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host	dense-2048	native/numpy-out	8.575426	8.142355	1.05319	3/5/0; 3/5/0	3/5/0; 3/5/0
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host	dense-2048	native/numpy-out	12.848515	9.816973	1.30881	3/5/0; 3/5/0	3/5/0; 3/5/0

Source, controls and raw evidence

• baseline/cpu-release: compiled 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7, tree 06f324529806d3edf005e4ecbb6a3c27f048f946, identity SHA256 82f66c8a54d2c5a029ab6b62941d37650c1d2f07441ad6dd5e852b052e567b38. Loaded _core SHA256: 76168203038c854eec161fb28eac73b832ec0d0341d76e0266dfdc3514318611.

• baseline/cpu-blas: compiled 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7, tree 06f324529806d3edf005e4ecbb6a3c27f048f946, identity SHA256 c2ee4d3b54cb2f013799122cfb6bb254f102582931ebe16f341bbb3698ace8a9. Loaded _core SHA256: 81f5e99ede9ebf0d62e2ffcef1b4529183ce335a9d2e713d6a2d5373c7f63702.

• baseline/cpu-openmp: compiled 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7, tree 06f324529806d3edf005e4ecbb6a3c27f048f946, identity SHA256 425d959e2b85fc58d2bdfcd3bc1f41f95845478f6989adb1ff723d237920375d. Loaded _core SHA256: 17d44c0b297d33607aef6c51084304e1955b3e5961fec29409e4a13a57cb1ea1.

• current/cpu-release: compiled 2d81b3d1409b0b428572cbf26fc7418e03bd270c, tree 90f513866f9f2aa349b92c8958b1aaaebe24f1e0, identity SHA256 bd3a934ae3d0b8e43055634a6ba22fbc8870a405bce2a2f4082ef7aecdf2c5da. Loaded _core SHA256: 26a23a8dcbfa77ef6be4cfa34aa5c001f5a684edf94d77d403742afcd15b3fca.

• current/cpu-blas: compiled 2d81b3d1409b0b428572cbf26fc7418e03bd270c, tree 90f513866f9f2aa349b92c8958b1aaaebe24f1e0, identity SHA256 5f6ad045349cf3e70689876c9011497479502e32f9472a37e763be9ff8092467. Loaded _core SHA256: fba22b9264cbb2f4ef0109143883ab7bd475b85b92ce42b0b93b30d7483bf261.

• current/cpu-openmp: compiled 2d81b3d1409b0b428572cbf26fc7418e03bd270c, tree 90f513866f9f2aa349b92c8958b1aaaebe24f1e0, identity SHA256 3ab4d7e1b7d6f6ad63e6cde68708e5a075565a94d2a0cec0501832af4f82692e. Loaded _core SHA256: 2d58e607d8562879f6fbaaa8c0a6dc5643b3c3927b5e7a5e1bbf1be4c4aa3d31.

Manifest SHA256: a5ec759c657e79f7cd22547fb36794c328855f7a270f31720c13b079e4dce591. Plan SHA256: 7e0661254f1159b132252a445b93c8bd867da77bd952362f8d9b1a4940ea6418.

Every full strict report and actual comparator

Configuration	Verdict	Report SHA256	Comparator SHA256	Comparator path
cpu-release-cpu-dense-f64-v1-none-api-fused1-host	inconclusive	3e67a6348345165654564a23be227abff1118ed45ed429e7baa216624e9aef28	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-release-cpu-dense-f64-v1-none-api-fused0-host	regression	3d96f293b271cbb959025bf99cbb3f2d81f2943656ed3a422959e21e820e9de9	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host	regression	b8d566ef0f4ef96baf0847830a9a8fc08feae4afb663e4a0d263f6472f305ba1	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host	inconclusive	14b0f0ae955461e0aefcf2ab44d10fcf53a97cb45d26ede9d709a5fb6646c56a	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host	regression	e6020d59604ce6c331c1f18caea531223b7494b5a1acf6c59443c471318e4c17	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host	regression	bf44631d3ba9cc979a2bc9291f9127ad669c6aa853c2c5afee150c81d1d45b02	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host	inconclusive	ab3e9f053242e98810f880409ed235ca6bd07bee485ab4cd521ea7afa16d2740	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host	regression	57d76f3fad69758da3790309170da00c69eed55e012140598bf53933f9ce5510	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-release/tools/compare_performance.py
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host	inconclusive	c02763e66be880d3258388a45e602b2e4edc9408236136b3e476f79d57b828c0	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-blas/tools/compare_performance.py
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host	inconclusive	00bfc0956ec4e4a28217710bbc9e1fcc35b447cf7cc04684cca6f7c90c601f5c	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-blas/tools/compare_performance.py
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host	inconclusive	a3fbcc75f93dc6c2cd533fb3d6744b0ee065e456808a0555d2445c1d7f5612ee	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-blas/tools/compare_performance.py
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host	regression	68528bed32601a75b33465ecf2953706b6c4bde1e1034e5aa2aa8aee41d5f5fa	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-blas/tools/compare_performance.py
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host	inconclusive	e01dd4b27285aaf91e18befbcf9868ffd8e757321480ad9582c97b6b3624c4c8	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-openmp/tools/compare_performance.py
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host	regression	3c4638a66be3c7dadc44ba997ed2c27bf6bdffcfff48e1bd3fee9c6664376e74	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-openmp/tools/compare_performance.py
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host	inconclusive	85a4c0d54740cf21e6711e12ac251b802321ebccba6665ba1300a64629824cd8	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-openmp/tools/compare_performance.py
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host	inconclusive	e7736ccc6415925ad97d5517ebf6f7583d4ce28c857e1e4e340586e2b5e7a81f	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/cpu-quad-stages/cpu-openmp/tools/compare_performance.py

All original raw/import/log identities

Run	Raw path	Raw SHA256	Import SHA256	Log SHA256
cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b01-baseline.json	87338e369b6a4839c414a5d57657c53dec80bd1a66587429edf319ea7c368c6b	57d4335ea815368e616a27d53b9bc603abb9adfc574c8d5d11501229fceb9724	a8ba3a85b0d9386235bbd2f1d34ba29f7c761a9e83adfe53bcc8c2ff6d35c255
cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b02-current.json	0ac19d9a812ae435b66d0ea7d0331b587b08324b0f2fa50bfa6b05d76ce27f85	774093c654a8c3a06eb52694543110f3bec5f72a727e4f0f62dfda9cdf658ad2	bdbc907e6b8045d237197b4569b4303b392338576b0c7ce3b462a921ae1b6852
cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b03-current.json	157f9477a0b99eda693bae3ff72740dbd162e2e98f46ac825477e62dedb7c1ce	486ae209d90937ff863f494074bb11efb9624e71d38d6552282c57574485a223	b8a97fa61bd7baa4b1219b53c81868a41db92238ccaa321e4e2056b599112eb6
cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused1-host-b04-baseline.json	9841601b1dc2511c6c4c3e8de09bb6c21ab37bd0adb53bafe60551919b195cfc	6625af419557c0dd017466e59097a899a5ecd36c2a4133c9ebc5bbb261a0933d	8c10829e0f34650870a853c2c124ba1e4a8e82be3ffcf9f82412076d3aa953e6
cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b01-baseline.json	beb2478b2186e60128c8027a834d7ca6903121e9031db70124fb4c4cef2550bf	b7b4b5c7a19e2495b129050dbf09e4a05109a81277325c18e6ff72e38cbeb99f	36175de7f4029aa8038488be0caccd216c4d661723f562c0ab7186b633910657
cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b02-current.json	d452f9139e5a8cf62a94dedbe02074c837efb43bd0d6c3593910860579f270fa	57af0119ac585576d04176f4eca2c7734d8f25f75577d44c86be0ded6c33cd71	63a3131897610391c6c78f44d46354583cb76b43e3dca5eab5d7fd30fec42980
cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b03-current.json	0fbea281bb6ec73cf842b0e9edbbf67126b106fab09ef7fc171d731f71a0d522	ef46b3fa2950d8754e182d2c89b741758663df9e12336593f994be091bb20007	b024fee7e37bf880d7acabf0926a530333cbdc3e3934b89f1290de8678480ba7
cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-api-fused0-host-b04-baseline.json	a942b2afb8128da93964cb2149a870f2330cc4ae27f6f39adf60d9afaf334efe	fb3eb7ad9b8535c74d7ecd9cd5f46eea998451002fdfd07acc832d4123620d75	0d583a5ec8509c544820eaad316dbd66ca473ff50701bb229365d2d36c2b0ae8
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b01-baseline.json	322a4fef90ce9492640b27356844c442c0a1efb5f9cc971a7f652839fa9775e7	ed45901afa37c265a360dba66ea3e9bf11f7903f70bc2315388223112f66965f	532c15b54d654beb5b99cbc6400cc5201d7370fd34b0d7f992ea509cb59ed980
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b02-current.json	deb7aa8919400a6e3a4af4af681a76aa79d61e465d56abf26d4a283e0d7a6a74	ad7ad589b9b7455928b792eba5fd74faefd99a7dd63455bf69df89fba12aff03	9f9928adef823340d8d612a8e5c5deb152322979cc77b01c6d23f8a5f315c72b
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b03-current.json	113766534c8d308a98091f5a74593b4c509e87f2582803ad4769fca048e8d590	20112e902b50f710b32638c913804637a9267eac39985a11b4af10d2f891ab6d	354174aebc75fd34eab3a44beb5a192d0ddf0957497e45510b9ecc65ca39e229
cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused1-host-b04-baseline.json	43d6482ac311b9c6e2ed95460fa123e69ce453c5d71b37bae8834a472e7d79da	016c14c963d6ca1e922893012fefb8c9812a24c128c52103721c6f569cd80440	abfa7e8ba5a6d744e5b592a679331081ccd50d883acb86165b78eee7b32f4bbe
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b01-baseline.json	dc4e9fda08611e63dd26ed6e58a6f2914d4cacad0990382ed39996653d129860	3cf9b6e8bccfb4cfdbfb60103e75f5eae802c1c76f28fed84a13003cd2fbac5c	6679883aae79cc23f7b434f0a26a97c4b6d2e7a0b3f52de719f6e271945f0401
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b02-current.json	e6fe5f1b16bf23bdaa6a18ea9a584f05a9c5bd37b8a35dbd80c9138f927c3c84	23a94ea04ce9fd19525540ab8d9a7fac062515f2bc4496193058ac951b930028	0a20f03cc5a449aa109cb1b180eb8f4963bd42085d6b2e15e0696ee0a3dea2f0
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b03-current.json	0fd74d3a9249fbed9d15f663c21f8f47c07fc636105aab19a1687ed1825d1bc5	66e28331e14182e7d681f70f40dd6ffe61ce158469ea817a1881a4d737c30704	18541d45747db5693af9720c22eff50052a7ccf9c34c480dad3844cb29a8c660
cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-none-prepared-fused0-host-b04-baseline.json	38eb9591e38f2eed7478a68f2e8be8b7d223b198fd0738e22bd738d1eab86211	2aa88675193d95db46def7bfbbf6c0942f75e48f02a1ce261ebdabee15ea8a9b	8faf8790f29586842b8035115cde3b7ec9e8d0a8ab1866ea7ce6ce2dd9a43314
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b01-baseline.json	05f6de441e576f7d62476b2fa465a33e9b8c1bdab4cc2ac617b1452e9c29547a	89fc5f284d9c4dd73208decea1330ffd01eac7f47747749ab335e7d02c045b41	3ad38009fe0bea027945006b065bac51f45d3353a025eff1f330a6ad1d999125
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b02-current.json	d18d0c68c316b6648c2dc5a34a392908987430e0dcd1d3ba3b8f2d3cd612d159	a057f538ff1effb3b268905675d49b81c02080b5e36f4d3c42b6d1b10aebfcd8	252187d69908be0529979002ea3a4439004d12ff175cac8ac9f79fde943105fb
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b03-current.json	40b1852bcda59ccd05f5a3432ed78c6937a6da278ce5baaf85a156334c202ead	f4c860dac51bff6bf9dd59cf684edafc5f4b1cd52a854dfab1dbfa34f26fd7fe	3e8875d27e92b0afd562a1a3025b3c448def01317b558a948b51ee63a19c4c8e
cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused1-host-b04-baseline.json	54708e6a02e61aa7fe4c317459d8b69f4ef6296edfefe3fa5ff98a7c6281ffde	a1db7be8d26d86f4ab1cad386c88065d208d5e4ec0c178ef55db3ad66edd2b87	bb70a2c74fadb4d0ff1b3c9a05a47aec90f779050b1ccf909bc7c22fc908e266
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b01-baseline.json	4782cba580fd844c703bfb429fadd6ff40fb8a3ff7da59ba29704fa3a311dee7	13563f8fda9d17e8200113640e51cbd2b472ff1340a52fa6218826ad66206b43	65d8b09571385fa3730fd8b03e81ccef49ee62df2dce6a150291efa9be120a7b
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b02-current.json	1b1e600257c498a70076ccc17ff79b224577244ed04b13cdf4e422f60cb0879c	9dafcc13c0309c5d80d30e93e13cca80cb7802f9c4fee73918c440e58c3f7bd5	3f0d873aac4390360373971bf3e3db415403c215ac42b6fff27d5fba3723e0da
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b03-current.json	9cf803081601a4dd49ed693b0f1085fecc7c1ca30d3df47aa74e7a6cffacf523	42bde8b33ca6fb2b3331297d2960165998c8cce14302fa6bace5a4a900bf00a6	24d656ef7ea5fa592fff27a3f55374ffa87f6d3e5abc83d14e4a1c100c357cdb
cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-api-fused0-host-b04-baseline.json	e4ddde3e4cedcbda765a54bb1a01aa2fbcb625a3ced1a095811322e6d84c909e	6b5290e9cecf53e976b767c09ac6950fb13ffffabe6b75b2ac1f515b6bc52524	9ebbdae703797527cda719cf8a89f6f34036d458ff6605bbb6364f6d3d4472f9
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline.json	e3b3cb481d2b591264098e908eb99a16cbc4c90cd5387a43604a9b7cec1dfbd8	e6e321c9e1f5fb88eff27a34f6fd389fcb710324ac03d97efddeee5cbf495b66	5dbfbb0f950fef01b215367ae5e4869858fecf24dccacd430037c883109e1351
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b02-current.json	d6fae0c4ecd0cd705c39be343e4a9c0e493c0f8f13ab8eed0e341771c2672de4	8862fb89be47ce36a4317abe9f09c7d30ad5c0f2dbfd615632854b9614b5d822	a930a07d30b6cdf93c32e31afa9d03df7cbec65b01415e148918ca9f613ad120
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b03-current.json	b13dc2bccf1858366445e5416e59e5236b653911ae2532bf7a75781694c7960b	9ad835904a9afd1e009e271add5b4f5e740909e76500fb60cd55218c5aa6dcd7	1733302ebfa8287559e040d7e094cbddaa81f75e11923b66459f8b2b5efceff5
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline.json	a981799aa437e51ecabaee10c8a93626d63bde7fc61824645e37ec8ea9596685	2e571c1050d14c21d7c913a55f73df67b857c50c5c76eef89ee9ee34a431b9e9	5c0c4f0b0740ffb1349df6eabbf8df2dacd6f3fcaa917446dfb27e0c33e6a171
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b01-baseline.json	e94541d3bb94dfc6a77d0b1e4cab2befba488a703ec7b081f08402aa61a539d7	338c033f4614bd7d7f4dc6bd2f9c4b0f947502447cd2394c16ea912c1ecc1cdc	e36c98d02779b77735d9b27cf47428e4f0a00032bb5c239c217cad0b5453699d
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b02-current.json	419de2dc289ff85f64358515e46b299c959714e42000c71aeabb8aa4aaf2d738	7c34ff8b4ed5b84e1673a2aa6c12fe03cb4d4e1f3d68233410d0fa2e8cbe949b	7c717c7d3f5eb9c978065a4c7a2221e4e1060b2fcec1bf81aaa7f57cb9037a30
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b03-current.json	6351d9ebaf0e228c127ecd5fd97fdc7237728f9603e71deff69730b8a769ae93	eaaf54828086b6d22d4a4335c043e5e81c27e88992f0ed8810e75c249a0b6785	da3abb9d8a1086f4e567d40f13a22ccf7858b6134e085173b57f0acc5de8b27a
cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-release-cpu-dense-f64-v1-jacobi-prepared-fused0-host-b04-baseline.json	25578d56b0e35ffe903226f1d5b74da2f7872e7146d59646e6c098a2b19811ed	26f562699876835efa8d8fb758b0503374c0bc32d49695975558ab674a5dbb72	768bdab273f7fa21489a4bb6e4acf5bd9ebd2ef1a2a1dbea6cae60fa0d051cca
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b01-baseline.json	231f8a7c00990528b539f8dfedd28c09b7a5029905234f3582e17903a03882bf	cb6bcfebadb597558ed7695f2f94bf4f22a05c4ef9afb22c25556bd4b548a12d	536e78bcd056117280571f922cb4295eee28402dfcda21ba91fb9a02f9bddc3e
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b02-current.json	6fc5019782e5f678c9e5350c2cbb3bb7bed36f0521aab83872f836f1bc2095ab	f2c2f4c17cd1b69b65da99ff4ba6eefeee6dc12162af3e1e94ec19850f3a3ed7	50dba0663c0b3bc8d195329322a7263e382a63bfd007487690d57d81d4c4578f
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b03-current.json	ef235b127d5b73556f627896e2b535a3665cca20d435d5c3e8d8f571f649f0d6	086cfa88938b8f7ae0b8ff1d70c438981629b3b1476d398324542056e482f7b8	4e9692615bceafd63fee30ae65a6e9b974f2db38be8ca7883f42afe52919f59f
cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-api-fused1-host-b04-baseline.json	e36ecc9787a3f9757b38656f22f2344b0d0bdfa1d23c96c67688decf6237f2a6	fef2fa44930b71e845afdddd57feda28d8d98b4596509ad7926bcacf7ee7d81f	3d5e2377d1c2f85444084d9c2eca31b91ca8d2ddb36f4d3cee976a3877798fd5
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b01-baseline.json	fbf4e2b3cdd46d0ea13dac1d0fc0901931e02db2b0cf62b72c30fb0eb531ca7c	067b0eacd3dafcc5f2ab46e18fc46756d18b6ff9c5003a5fdd2bf7148c1d08ed	a53b236113e87c127268703544cda3249197a04929c97f4ff2e00e6253a6392a
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b02-current.json	fa1e3fcfd01cad90bfb2ec2d3bd9c332c4d7483b3e68b503a175f53c517280ea	08ac9afc66162a180a6ae9c1ad2d87b86dc2c410aa31be66a0e5880fe4bd0b48	a31aac1cfc83b0177d674e6608a501609bbc16ba618e7d973e533d54835be9a1
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b03-current.json	d3f1801ae87cdd2e10dcd077d1f41db8c7f71f9dd1eb500416c93f2e7a958154	4ec3544167836adc3e6cc7b10065d5a347f62d10f4c81c4fc707f9038fe21b4c	6fb656c9927a310ca2434a490dde767a72a7bf92ab5e7172380c67c4851f1e66
cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-none-prepared-fused1-host-b04-baseline.json	5b3daa55b91f6b33f8a72c5be0043508245a3d9e40d9916cde4864dcabb3db1a	89ae0a6b55f7a9feed266565a1799b164cd6083546eb3f6bdce1057cd76c8ec1	2f709b7b5042062956d0feeef4c9414b625daa9f963bcc268e52eacbddeb3a5e
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b01-baseline.json	d9f118d337623fe2fda735264a454aa0d83064c07464a355dcca030f38920cd0	eb3a60ad3149babc7c3f0406c4553ff5b919b34815809f5ab79061b79e21d9d3	416eaf2a3a507006068c2b1bb8eb31f4ee217ce3cb482a8e1eb8a6839b104ce3
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b02-current.json	aaa2817f2714b6fb872e007b9637d3225fc5351467c08312ebf9e5a5b3680ee7	7847b579f531edcf5b7636699b7ef24feae2383e30e797da69f625c64154b0dd	3fb6e2a8ac7d0517b4f96758d797f63329fd9d1dfa42e34f2c9ab0dcbc63299c
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b03-current.json	79b36f82b48c02477896f8d9d81ce30605b84a9d2791e9137c183f4d60ed444f	10e90bcceef699a7e1c34254e664ea61a0a8eac7219ed8f773b9f33779c188b1	43d568fa00427b30f77b712a67250c669e5e10511f393eb257cf8bd7f8c1564f
cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-api-fused1-host-b04-baseline.json	2039fd2b9ecc6d165bcb262a3c29da908924c8caec9a0e467b663b8c7cade3c2	8aeb4cdf0be47fe11ff2dacff9cec6d0e3ef02064bf53a20f1a86c87c903ff29	b13b8759eff38999b49b6478bcebd92160119ea0955cd0f891ea4c6a5f4b2654
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline.json	193461c9569d7bd77d54416126f047adddf63f54f3678d27584b3e0a0de4d7d4	a096a7482da30a735fbf808918a2d4ddcfa9f8d3d2e0671253dba2b3477f6610	e2e403f46aeb51c80e71f856729c461854114571e42de911fdc488d00b4c6085
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b02-current.json	a6c2db7ccab86ce0e796eed141cfa6773d35389345bb14db22a88ab46b87858e	bb7f6ca14f4e0b0f8a42249c14574d82374fcab2ff97291313ae373d77f005fc	00b15a232939c55da0bfdcd515e80a835ab83f59bab18eb0c4f108e392606e6d
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b03-current.json	713f303ef4aaacd36f59aee74c0be87b9d9718ec275ba847d7bca88918a3828e	4867f72317e3531cbd5898ce043a1a355d6f40e4eeb13f5d811071fd551c116b	44c583fb97cf2c6765b6b315c00eb626e098271d49f9318e0c36d499e86da026
cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-blas-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline.json	bc9c31fe613b106cf5e569882862539201607ff6c3a484467d09986dc38c3b83	387d456bc1a7db5d338a63bd6a0df71792c627fdc1d2d6a98846eaff2ec4aa73	6568cddb4157d8cf719eac0138bbeeb321cb08a2f7d588d7828cb8f7c2fa6fc5
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b01-baseline.json	e49ec9d453bb315b07b8050271198428f93b9062fadd785e2a0273bc0443c92d	79e3cfc4f7a743fedc1be951eb1f724651a5083d57f22cb759cb9c01cc444e58	782356b3bb1a3a7f07e52d86228a0ada663d0ce67b9a5d5456c791a7e4025baf
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b02-current.json	42c151026883abb2793f0dc77f01ceff0f9c966d2a5b61854e6f4ed0b3128d9e	d364810b1bdd81699d0f7d84d6556a753873efe2dea722e53c32239b8bb66ebc	afcc7029798bf828e99bc19167379f01e2a22553c4d245a2800e3be64a519a07
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b03-current.json	e09a657d8166c80e32bf5d022e4f3fc348173f016dc3392b85683c61df471c09	aeaceb33d31063a1e58a290036448b7d80e4a6ffee046f3f3e6a3f92b1ec6886	307141be5cd4cf8607576f53902da517a97e8e086b2344ad5b9856e1a7e7d46c
cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-api-fused1-host-b04-baseline.json	87b3584d7b8161ce7e95e066e017bdcf9d60d34dd58fd366b165786a0d147758	43c682a76c265764fbcfe905af75ed8c8230310730fff1365b9fe5525b14e7de	5e3407e1bcec7076c5e4fb97cc3a704b815db90e385f8543f7ec52792664c277
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b01-baseline.json	fde3dc6c3a87af2d849d3b477df60cefe7ff3f9a922db6650a9c62277698f946	8fc70404ba771f8c826b75247fae0f3340ceee4c829f45a6a48cb72bfd4e8001	1fee172758d54a724135e4bbc32e43edfafd6f099dc6733fe50bb71bd30ea228
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b02-current.json	27fb3fc27d7102319c600f85f4e961c0e46b4a578f5250c5284f30c614630440	6221722326c14faa15f8d162bca7d31e65f0b92b5c370fb2115341bf0c738d26	d35d782ac392baf3eb29bee295024747cd36c601fc0033a06a0995d9e9f25ac2
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b03-current.json	10855a1dd8097548d43fe8b990e853aab0fa12032ed0fd80ef52dda5383f533c	12d9162844942c40469d3334b0f096b0b07dcffab383f76bb7b546ad6d599654	6986a79ceb48f9e76576f3771b0ef6b54f03a8b9abf3eff6191d0c5304eeb7c0
cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-none-prepared-fused1-host-b04-baseline.json	912700f596ebe3442b044b70bac5ee22bc768124db8ad520382331f2b4831fdc	07bfc8505b5a6313a0243bd719ac6333f262b66c1725ae3042069aeb2792308e	8a2871dc43421da4ee973b626bc01551c6c2d95aff603c07832f71f4677b9814
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b01-baseline.json	3be161702bfd043750af70a9d62dd6c56a3ce6aaf8f1d6d426734edc01d606fc	cc1c8263a09c66f8524cba614dfb0fc8fd67bba2ba8a0421d5dbd4c2dda353f4	fda09f24414892d80082947d968e3b1afb22f91d5d3ad94749c8fa86f58eb05d
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b02-current.json	eb2f92cdbffa93c30b0dad400c4c1842627eacc7cfd89b027f83021299259af4	728c6d98ffeab4dd63d85af0b21f20a061208a12089343eb77b37abe4696ae02	b0e1f4b283b0867e29f10f3f7fe5745ee4552adbbf3a7d66913321764df8d32c
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b03-current.json	60acb7967326b2181c98e9b8872b75af8f91090bb38f80194b7938aa5d4e8ab0	e6dfae4e995bb36dd6b1a243f5093ecf8a9700d06b1a10f4abf9641d5d238d9d	f095633a3b6112fd85ab5cbfc080b4b9a1d1708115d6507d133772f731853cb2
cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-api-fused1-host-b04-baseline.json	cdb156e48c8ae81d568f7c055b73d81149c1e8c515caf6a2c02e5376bf87739d	5145e3aaf6ed238a03694d166b18deda525b640025570ccb831b500d463ea00c	abc9249c2c6c4c12f3d17cc44e6f30d1ee3d3242413947736c067a1f1df71a1d
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline.json	58d3cf8c6899924b2d8066db7d659ed975b290914672ac45a8f5a5f7681cdbb4	8ebbf8eced9c45eb0e1bc647caafdd5ad9474f345bfdabf78fe8170b48e256c2	9f373e470db3f3f67945b1fc50d22560489e7efb40ab8b187e681a410a9d06fa
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b02-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b02-current.json	5d7e1a15e3667ef45acb05ee27be76c9f8b6365552d2a478799698ba0d76ceb5	a77c2846942c988b6cbc7136c8350847b658715bcb8d854d6b23d568faee87e6	40c39d93eccfef811a9de74add3728efcae1b909acd9fe900ec2c697d0e0b4bc
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b03-current	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b03-current.json	bfd07e1cfa3806456c022342c325522b36924963ca2554a1e668bc84c2a9b900	a309a6df3bb0608ce9a47ac0029c5ffa95ceae540df3da9eaf3b29d35d942972	f9669c4256c71d489c0f6597a1d36d9e0d3ae1eba2c1faf23983ddc64fb88332
cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/cpu-quad-abba/raw/cpu-openmp-cpu-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline.json	314f1be3aa5e3086c8f44529356b3d0618d9fe47ce2ea6b1418c0e551b00fec8	43a2ac5df703cef12136ca782d097c0115166b7ef35b7748f45d8d46dda35bae	e6a617d62f625a13b985a0046de69c0e9a683a3e6d7c513eca97e1913cbc164d

launch-abba (launch) fixed A/B campaign

Whole-campaign verdict: regression. All 16 artifacts/160 candidate/case outcomes are retained.

Every native and framework strict result

Candidate	Role	Rows	Within 5%	Inconclusive	Regression	Comparison failed
cuda-native	native	16	7	9	0	0
cuda-cublas	native	8	3	5	0	0
torch-cuda-eager	framework control	16	7	8	1	0

All candidate rows

A/B and current/reference ratios below one indicate a lower measured numerator. Ranges bracket the two process medians and are not confidence intervals. Saved strict intervals compare current against A/B baseline; they do not give confidence for the separate current/native-to-framework comparison.

cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	4.813830	4.556268–5.550495	4.675959	4.664255–4.675959	0.97136	0.8148–1.0666	inconclusive	49/51/0	0.648887
dense-64	cuda-cublas	5.244673	5.126512–5.356832	5.042561	4.869444–5.580466	0.96146	0.8839–1.1368	inconclusive	49/51/0	0.611009
dense-64	torch-cuda-eager	16.846000	16.743127–16.893090	16.146093	15.254762–17.171689	0.95845	0.8559–1.0860	inconclusive	49/51/0	0.611009
dense-256	cuda-native	6.553387	6.523076–6.589938	7.002052	6.592493–7.248671	1.06846	0.9546–1.2605	inconclusive	67/70/1	0.979365
dense-256	cuda-cublas	7.117926	6.572613–7.163162	6.782412	6.387729–7.027467	0.95286	0.8460–1.0843	inconclusive	67/70/1	0.994421
dense-256	torch-cuda-eager	20.814728	20.379113–21.335915	21.279331	20.428297–23.146691	1.02232	0.9468–1.1632	inconclusive	67/70/1	0.977359
dense-1024	cuda-native	9.972689	9.891568–10.082052	9.296761	8.969531–9.324780	0.93222	0.8313–1.0360	within 5%	70/73/1	0.928014
dense-1024	cuda-cublas	9.121081	9.003308–9.400038	8.356147	8.272370–8.810037	0.91614	0.8272–1.0241	within 5%	70/73/1	0.928014
dense-1024	torch-cuda-eager	22.006749	21.843463–22.198815	21.489071	20.914811–22.545226	0.97648	0.9214–1.0432	within 5%	70/73/1	0.928014
dense-2048	cuda-native	17.643095	17.643095–17.645789	17.394528	16.077773–19.172180	0.98591	0.8752–1.1296	inconclusive	70/73/1	0.871762
dense-2048	cuda-cublas	16.857435	15.434188–16.938054	16.039076	15.595882–16.795307	0.95145	0.8600–1.1303	inconclusive	70/73/1	0.871762
dense-2048	torch-cuda-eager	28.044739	28.039097–28.074096	28.948305	27.654878–30.322626	1.03222	0.9677–1.0992	inconclusive	70/73/1	0.871762

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.728329	0.718538–0.731125	0.793611	0.793382–0.794507	1.08963	1.0057–1.3034	inconclusive	5/7/0	0.0796191
dense-64	cuda-cublas	0.815262	0.811589–0.819505	0.924808	0.886712–0.936325	1.13437	1.0261–1.3675	inconclusive	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.231371	2.210525–2.240413	2.806099	2.601447–3.035518	1.25757	1.1246–1.4234	regression	5/7/0	0.0796191
dense-256	cuda-native	0.682779	0.672356–0.692455	0.717371	0.665252–0.746328	1.05066	0.9396–1.1419	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	0.710753	0.659617–0.734714	0.734561	0.706212–0.740878	1.03350	0.9404–1.1624	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	1.776912	1.577338–1.978453	1.925737	1.868912–2.232284	1.08375	0.9258–1.5043	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.024139	1.024139–1.033605	0.983904	0.980726–0.986058	0.96071	0.9201–1.0124	within 5%	4/6/0	0.013684
dense-1024	cuda-cublas	0.890312	0.890312–0.900717	0.865443	0.861279–0.868336	0.97207	0.9068–1.0204	within 5%	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.063054	2.003967–2.079198	1.861300	1.799389–1.909934	0.90221	0.8432–1.0289	within 5%	4/6/0	0.013684
dense-2048	cuda-native	1.493251	1.452932–1.517471	1.354861	1.319800–1.377295	0.90732	0.8475–0.9595	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	1.340388	1.274400–1.389114	1.226325	1.187407–1.238725	0.91490	0.8511–0.9851	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.326382	2.130661–2.472691	2.049732	2.005732–2.136675	0.88108	0.8010–1.0166	within 5%	3/5/0	0.836412

cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
poisson-8	cuda-native	2.726065	2.608788–2.744769	2.271615	2.203330–2.314783	0.83329	0.7679–0.9020	within 5%	26/28/0	0.579621
poisson-8	torch-cuda-eager	8.167160	8.123282–8.167160	7.141832	6.724729–8.084623	0.87446	0.7745–1.0480	within 5%	26/28/0	0.579621
poisson-32	cuda-native	12.351859	12.351859–12.501883	10.142781	10.141766–10.204629	0.82115	0.7616–0.9193	within 5%	118/122/2	0.840189
poisson-32	torch-cuda-eager	35.556039	35.338629–35.750465	32.857698	32.658737–32.958426	0.92411	0.8894–0.9582	within 5%	118/122/2	0.840189
poisson-64	cuda-native	21.039181	20.199127–21.737904	19.249882	18.542243–20.657129	0.91495	0.8337–1.0268	within 5%	201/207/4	0.907427
poisson-64	torch-cuda-eager	62.768426	59.110305–65.638706	56.278432	53.656345–60.147633	0.89660	0.8158–1.0201	within 5%	201/207/4	0.907427
poisson-128	cuda-native	61.233210	60.160044–63.369062	60.074118	58.036546–61.842773	0.98107	0.9133–1.0719	inconclusive	627/641/12	0.989568
poisson-128	torch-cuda-eager	179.504109	169.993008–189.327686	175.607445	172.820148–182.200422	0.97829	0.9048–1.0778	inconclusive	627/641/12	0.989568

cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
poisson-8	cuda-native	3.209334	2.885346–3.536908	3.244278	3.180382–3.284078	1.01089	0.8233–1.1758	inconclusive	26/28/0	0.579621
poisson-8	torch-cuda-eager	12.119196	9.662614–13.498572	12.371728	12.250894–12.673979	1.02084	0.8101–1.3510	inconclusive	26/28/0	0.579621
poisson-32	cuda-native	13.737134	13.084474–14.073719	12.920089	12.648853–13.527864	0.94052	0.8255–1.1026	inconclusive	118/122/2	0.840189
poisson-32	torch-cuda-eager	47.675364	46.249704–49.681671	49.659800	47.601290–51.178409	1.04162	0.9559–1.1082	inconclusive	118/122/2	0.840189
poisson-64	cuda-native	23.085894	22.834246–23.554558	22.461419	20.958669–24.395005	0.97295	0.8839–1.0869	inconclusive	201/207/4	0.907427
poisson-64	torch-cuda-eager	78.673308	78.450570–78.916816	84.202037	74.901499–92.845909	1.07027	0.9377–1.1933	inconclusive	201/207/4	0.907427
poisson-128	cuda-native	67.216064	66.933736–67.486862	62.867447	62.646747–63.872352	0.93530	0.8927–0.9740	within 5%	627/641/12	0.989568
poisson-128	torch-cuda-eager	257.307359	256.046632–260.969516	240.192646	240.192646–240.737775	0.93349	0.8959–0.9854	within 5%	627/641/12	0.989568

Current-source framework comparisons

These are descriptive ratios of current pooled-round medians from the same matched configuration, not strict comparisons across frameworks or scopes. Every higher current-native median remains listed below.

All current matched scopes

Configuration	Native	Reference	Lower/higher/equal	Current ratio range	Same known I/M/R
cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.2896–0.6009	True
cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3123–0.5541	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.2828–0.6610	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3296–0.5983	True
cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3087–0.3421	True
cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.2602–0.2668	True

No current higher native median in this fixed matched inventory.

Source, controls and raw evidence

• baseline/cuda-release: compiled ce45ea9cc6d38cea05ef04cff8a321507e9684d5, tree 6cf00b82bac6f75e89093d1eb627f36d4fcc7d9b, identity SHA256 e7232ead6d14f51f9f6081b982c63c3fc3c9c553876d2f4daa052ba5c8fc3d47. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: f5ae10070dd259e06e629113ac3c101d8c0e2248a2b5810239d3779cba11de25.

• current/cuda-release: compiled 14c37277de16182fdc7f4cfcf9193e0b352791d1, tree 26fcec1817f59f8dd680980fffcdf8fe0907e628, identity SHA256 daf34907780c5b6fd366347ff81161904289228e9cf5960e85c0c669887e6466. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: 2b95332f5e7cfa622e5de9eeb109ed8d6b2798f7e9eb9f8ee7d686a9c1f71fbe.

Manifest SHA256: e01983ecfbb268ada98843af502a3b650a53e2ba7653c51a1f8195ab95848d2d. Plan SHA256: d3e77ce3b1aa79ad4a10df4a595702a227b180318dd3603cc60d22dfd98f3ac2.

Every full strict report and actual comparator

Configuration	Verdict	Report SHA256	Comparator SHA256	Comparator path
cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device	inconclusive	5508004489fade1785405ab6d12fea7f50d7acc09942e0d755c9fb7841051ac4	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	regression	e83a95ad83860bd5a7f48887b0cc82db8ac81e02f64fadf1294f8b66842ca661	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device	inconclusive	1b8ee57cbc7fd19b5570acfa24cf23951199ac95c9336d896302fc0d6dc429e8	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device	inconclusive	12282675d9c6d8959e724e03c9a56be03d6af0d71b6f506280b1312e1d1c4928	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py

All original raw/import/log identities

Run	Raw path	Raw SHA256	Import SHA256	Log SHA256
cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b01-baseline.json	6b8e219f2cdb1d3e6b5641fd241f75882c1e154252457ea1299f32b804a48fb5	316bd75d37ec2bca60fd3534582918db407ab4e2c28e10066f5d1dddee8cc77a	86742513986bbd3d3d84d7eedc8122585103f665c81f0ed3966cf24f28109309
cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b02-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b02-current.json	0fb1e8787f0ebb5fc7bb6dc0fef0c689e8659f6084f9b1f758e6cb4c38f5ab9c	5adb7be9b90a2bab5026815a695f523a4a96564857bd5f35025252ec90d4f11d	5c7e0044c08c551ea7a85714ebbc5272ed386e30cee9619573d9d598fd36dc7c
cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b03-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b03-current.json	917a535abcb0436d1f58df11bdd66c3c563f6dfcb7eff5602656e351a4e3b737	97668c610740b82b57f5396a7a474f76ccf08e713bd468ac145ae06ff58edd0a	7c3c3d5be3e7dd40602c09f2f6d35f87333c5aa2d819a5ce915738fb797b1332
cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-none-prepared-fused1-device-b04-baseline.json	dbf158f4d671a01aed9fd4c08f85ffa5b08b3604cf7c5b69de31d270dbbedac3	e7459a963961f43b55aca146326b49e065e4ec08f64d9c4b0d91ebe8ee1dea30	3d1393e53420dabaf5cbe35653ec0229bd3bbecf042d59cad94f3d7bcaa3d890
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline.json	8973810f23384690cae2d8c30669de93a28f594477c4a2a650acd61c4aba4c26	0fb273f04ee4c1c357642c87a49bf589257f6047805b1fbf27bc6dac37c85888	f2aecf87e7cb6d1fea3f6461df6856ca2ce84bbfc67cb2aa55224a94deaf2069
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current.json	5e555be2c6ae9375cd276a7a6145b2ccfe6377a1004a70bc720ca6e1c3fb7600	a7c5c5136d1c6255c23d695228e0615b2d36e23a1a94a8566a0e80cc724fbeee	c62f3eea585ded27ffa93a723e00c6995d6c38a938a9d47729052eaf62c27776
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current.json	89d56029f5cad57bfd31c773799fd63da12595aaf2d165142f9a19200f9a1228	f8610f5b873978a180fd973581d989cc325bbc7c6e1bbf9c7d9a953a3e840dd9	eeccc04d10c19553f669879de376441fe51f8b2a51d7914ed138f72ec51be273
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline.json	b9c29957b5c7c708c4252713c5e0e55608aafaff1d014b39e20d9ad4dce340e0	67f0fbee39a83b5fcc03942c055504575d9e820c8995f4c9bb3f262936369c15	357f939d3642b9c46b93ab94d767a59882d56d3a5101458ccc8d1121a2fd8b2b
cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b01-baseline.json	bd06d6676b8eedfb87dfbcd7ff7e52c126bca238ddcc3f1c60b53335fd31ecea	d04faa50713239c2663a87adf495e43aff459f70f90ab79b6d04272d885dff31	1b29354e4c90874e7939a1ae722dfd613df8c314b8377a821de1110983e0425a
cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b02-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b02-current.json	75f6931fe6a18553d0d6455a14fd88c40b4712ef50555e006f7b4bc4357df0ed	5b328bb5462c6a8d0c652f024b7822ce9b8d18f593ea9cee48eb8724e8e32d19	cdb52911f60174325f3a5a55fed20997e03111bc28ba217578fdcf816baa191a
cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b03-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b03-current.json	9afc6f2a05096285e30e4aff5aca898ccf32e9b9c5f8e69a21b8fa3fe4a47f96	e7606c28a11ddd6c805bb14eb038f7db2f9bae7876268e9e23aeca340ef65ccb	a173333301db54338357df1592295c2fd9c8d37228f60cb8d8e264451bea2a34
cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-none-prepared-fused1-device-b04-baseline.json	d7647f84b1d089dcef6ce13b1f51d74f22768654be417d90a3145f43c5c558d7	12245323d8ad5782ebfb2d51b6cb60dc21a7c9b4bbca234dc657051e952f4384	649b2058ea8c6353251c085c2cbaaf58610c04d8253b7384033a4c094b5c1b69
cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b01-baseline.json	49663fe9f18fd0ee506b11f7537feaefabe477462dd2f5c2685080b7f1185a6a	e0bf7422c9e2c28858e72955419b96207d66915786acec77cde9acb1de8d8742	5c9f151f308f086bb752e0d7d9c513756bccfd9d0d02df11331c588ef027ba1c
cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b02-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b02-current.json	0dd9c5ffa622af3ac668165bfe9a0b6a1ad280bb89b243599b08aa072f3347dd	d208395ffd58ca62ab21bbc15bbf7dd6e20dca11fb6947b8627e09c581aa3e60	c15a6e5a4524d181b80c6ea783457e7215ed45f237516664b93a070c6cb30084
cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b03-current	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b03-current.json	a5af15b1fec8a3d262edfb66456f630f8b2146537f7efaac639664745213719a	d90097e23b4bf47ffae87dabcee7e653f90bae74a86ad0b94954a32e52db2884	5118039dc393e32c61f9b7c8fa7090839c326a56e154af6d3f405342690ff253
cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-abba/raw/cuda-release-cuda-csr-f64-v1-jacobi-prepared-fused1-device-b04-baseline.json	4c5c703a7ee9924a33cd655ee56443414ff7b8b5f6aa0d54d6ff260c1f5f0c91	4e9c87d757fc0823bf139a270a62a3efc3d2a1753208e0ea90048bee989bc9ae	95db05f771691988468398dae0f338555ea1ce7581a528b828f962fa4cfe68fa

launch-scope-abba (fresh) fixed A/B campaign

Whole-campaign verdict: inconclusive. All 32 artifacts/384 candidate/case outcomes are retained.

This supplemental scope study was preregistered before any launch A/B results were inspected, following the earlier resource-variant exploration. Intervals remain unadjusted for that exploratory sequence and multiple studies. The separate prepared study and rejected resource/CPU trials remain reported without replacement, pooling or causal layout attribution.

Every native and framework strict result

Candidate	Role	Rows	Within 5%	Inconclusive	Regression	Comparison failed
cuda-native	native	32	22	10	0	0
cuda-cublas	native	32	22	10	0	0
torch-cuda-eager	framework control	32	13	19	0	0

All candidate rows

A/B and current/reference ratios below one indicate a lower measured numerator. Ranges bracket the two process medians and are not confidence intervals. Saved strict intervals compare current against A/B baseline; they do not give confidence for the separate current/native-to-framework comparison.

cuda-release-cuda-dense-f64-v1-none-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	5.558584	5.509592–5.668073	5.318354	5.317839–5.325962	0.95678	0.9255–0.9759	within 5%	49/51/0	0.648887
dense-64	cuda-cublas	7.624638	7.498715–7.903087	7.080199	7.077820–7.086291	0.92859	0.8815–0.9574	within 5%	49/51/0	0.611009
dense-64	torch-cuda-eager	12.122936	11.717379–13.200377	11.954493	11.684459–12.036627	0.98611	0.8839–1.0403	within 5%	49/51/0	0.611009
dense-256	cuda-native	7.904368	7.780297–8.094331	7.804919	7.748337–7.874125	0.98742	0.9495–1.0227	within 5%	67/70/1	0.979365
dense-256	cuda-cublas	9.700395	9.627548–9.766035	9.457528	9.367525–9.568084	0.97496	0.9365–1.0136	within 5%	67/70/1	0.994421
dense-256	torch-cuda-eager	17.199876	17.199876–17.242945	16.625322	16.547660–16.940137	0.96660	0.9242–1.0109	within 5%	67/70/1	0.977359
dense-1024	cuda-native	11.756305	11.693552–11.780366	12.222746	11.193752–12.662413	1.03968	0.9434–1.0931	inconclusive	70/73/1	0.928014
dense-1024	cuda-cublas	11.668278	11.583510–11.687221	12.423585	11.587998–12.561685	1.06473	0.9848–1.0896	inconclusive	70/73/1	0.928014
dense-1024	torch-cuda-eager	17.357174	17.326778–17.488304	18.983285	17.066225–20.092351	1.09369	0.9661–1.1665	inconclusive	70/73/1	0.928014
dense-2048	cuda-native	17.156406	17.025800–17.392353	17.457335	17.284956–17.558044	1.01754	0.9921–1.0369	within 5%	70/73/1	0.871762
dense-2048	cuda-cublas	17.443869	17.330283–17.531420	17.684451	17.539849–18.099760	1.01379	0.9970–1.0473	within 5%	70/73/1	0.871762
dense-2048	torch-cuda-eager	23.270290	22.834941–23.671836	23.488968	23.060504–24.277636	1.00940	0.9736–1.0646	inconclusive	70/73/1	0.871762

cuda-release-cuda-dense-f64-v1-none-api-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	3.920649	3.911244–3.965355	4.166663	3.904380–4.351502	1.06275	0.9719–1.1635	inconclusive	49/51/0	0.648887
dense-64	cuda-cublas	5.674605	5.660838–5.701799	6.245970	5.852079–6.523479	1.10069	1.0085–1.1734	inconclusive	49/51/0	0.611009
dense-64	torch-cuda-eager	11.829503	11.791700–11.978089	13.186150	11.773838–14.043598	1.11468	0.9614–1.2330	inconclusive	49/51/0	0.611009
dense-256	cuda-native	6.525169	5.803021–7.099715	6.279880	6.253046–6.333868	0.96241	0.8706–1.1059	inconclusive	67/70/1	0.979365
dense-256	cuda-cublas	8.098004	7.583664–9.648200	7.841983	7.740248–7.925288	0.96838	0.7812–1.0734	inconclusive	67/70/1	0.994421
dense-256	torch-cuda-eager	17.773696	16.815669–20.249087	18.344075	17.836344–19.156613	1.03209	0.8700–1.1431	inconclusive	67/70/1	0.977359
dense-1024	cuda-native	7.979628	7.909260–8.131789	8.191064	8.094971–8.249306	1.02650	0.9689–1.0766	inconclusive	70/73/1	0.928014
dense-1024	cuda-cublas	9.142264	9.078347–9.228347	9.199725	9.109329–9.317029	1.00629	0.9763–1.0308	within 5%	70/73/1	0.928014
dense-1024	torch-cuda-eager	17.415067	17.282888–17.498371	17.534715	17.528339–17.592651	1.00687	0.9868–1.0355	within 5%	70/73/1	0.928014
dense-2048	cuda-native	14.939487	14.795201–15.196542	15.076313	14.841594–15.274166	1.00916	0.9735–1.0466	within 5%	70/73/1	0.871762
dense-2048	cuda-cublas	15.813889	15.589552–16.143103	15.763234	15.628604–15.905750	0.99680	0.9599–1.0264	within 5%	70/73/1	0.871762
dense-2048	torch-cuda-eager	23.698869	23.505097–24.276404	24.046171	23.547532–25.060578	1.01465	0.9568–1.0740	inconclusive	70/73/1	0.871762

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.044177	0.984684–1.269582	1.001512	0.998546–1.016874	0.95914	0.7683–1.0403	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	2.810224	2.679733–3.329852	2.731109	2.729404–2.735117	0.97185	0.7716–1.0456	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.243271	2.129258–2.894052	2.149955	2.124598–2.152917	0.95840	0.6754–1.0491	within 5%	5/7/0	0.0796191
dense-256	cuda-native	0.947592	0.915763–1.011131	0.928508	0.915922–0.967511	0.97986	0.8537–1.0636	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	2.819311	2.646847–3.077898	2.880930	2.775056–3.044509	1.02186	0.8921–1.1588	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	2.056191	1.917993–2.409822	2.135820	1.984446–2.240715	1.03873	0.7911–1.2074	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.238280	1.224085–1.306517	1.356641	1.280091–1.497544	1.09558	0.8695–1.3717	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	3.030536	2.945783–3.167385	3.217680	2.965362–3.637247	1.06175	0.9229–1.2691	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.003415	1.928111–2.094355	2.242156	2.027974–2.600961	1.11917	0.9323–1.3541	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.535839	1.469008–1.567593	1.670551	1.554080–1.908949	1.08771	0.9872–1.3138	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	3.024857	2.944654–3.203375	3.448856	3.116245–4.023427	1.14017	0.9681–1.4108	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.142551	2.034492–2.332956	2.355818	2.082577–3.171981	1.09954	0.8813–1.5928	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.821586	0.732637–1.011873	0.708570	0.686554–0.726082	0.86244	0.6499–0.9917	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	2.712245	2.452931–3.611819	2.370928	2.287463–2.519506	0.87416	0.6199–1.0316	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.377959	2.095622–3.777839	2.138888	2.051811–2.201712	0.89946	0.5200–1.0589	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.757327	0.690966–0.951461	0.692977	0.691514–0.694163	0.91503	0.6389–1.0242	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	2.566451	2.364276–3.358207	2.400638	2.398173–2.404040	0.93539	0.7006–1.0218	within 5%	4/6/0	0.209854
dense-256	torch-cuda-eager	2.044604	1.859624–2.615505	1.874515	1.840013–1.885301	0.91681	0.6790–1.0388	within 5%	4/6/0	0.209854
dense-1024	cuda-native	1.013063	0.996272–1.033354	0.999636	0.978330–1.018757	0.98675	0.9177–1.0270	within 5%	4/6/0	0.013684
dense-1024	cuda-cublas	2.675207	2.565730–2.996505	2.604455	2.583083–2.677782	0.97355	0.8512–1.0491	within 5%	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.119394	1.931060–2.582912	2.018583	2.008831–2.023041	0.95243	0.7439–1.0653	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.407590	1.336974–1.409442	1.357466	1.301961–1.396719	0.96439	0.9188–1.0468	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	2.855716	2.853080–2.911271	2.838779	2.749462–2.907357	0.99407	0.9382–1.0242	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.084807	2.053697–2.132242	2.068932	2.020733–2.106725	0.99239	0.9305–1.0378	within 5%	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.128999	1.093133–1.149412	1.009567	0.979797–1.032594	0.89421	0.8461–0.9498	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	1.163079	1.138310–1.185890	1.073600	1.040179–1.074733	0.92307	0.8654–0.9689	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.636712	2.584957–2.684674	2.492222	2.446653–2.498412	0.94520	0.8784–1.0020	within 5%	5/7/0	0.0796191
dense-256	cuda-native	0.974251	0.959617–0.983511	0.942415	0.917953–0.943273	0.96732	0.9273–0.9974	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	0.980642	0.952458–0.998721	0.935273	0.918305–0.944341	0.95374	0.9130–0.9975	within 5%	4/6/0	0.209854
dense-256	torch-cuda-eager	2.105469	2.051251–2.180646	2.027782	1.976114–2.054682	0.96310	0.8979–1.0114	within 5%	4/6/0	0.209854
dense-1024	cuda-native	1.283670	1.271408–1.295298	1.266212	1.254118–1.279381	0.98640	0.9612–1.0306	within 5%	4/6/0	0.013684
dense-1024	cuda-cublas	1.149104	1.130325–1.158005	1.124510	1.114162–1.152782	0.97860	0.9575–1.0401	within 5%	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.191905	2.163638–2.208797	2.154080	2.091395–2.265408	0.98274	0.9345–1.0723	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.595825	1.586174–1.611390	1.587031	1.582491–1.591034	0.99449	0.9782–1.0147	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	1.407275	1.380462–1.417172	1.398157	1.397832–1.398318	0.99352	0.9749–1.0222	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.225759	2.173619–2.275061	2.221999	2.211921–2.236676	0.99831	0.9462–1.0404	within 5%	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.822370	0.810353–0.832868	0.780085	0.774211–0.786675	0.94858	0.9129–0.9857	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	0.877126	0.867170–0.900608	0.855031	0.855031–0.867625	0.97481	0.9277–1.0266	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.597586	2.470286–2.885368	2.530194	2.453485–2.595603	0.97406	0.8345–1.0563	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.788716	0.756352–0.809067	0.718750	0.683180–0.754148	0.91129	0.8345–1.0031	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	0.817607	0.816785–0.820140	0.739571	0.715701–0.784063	0.90456	0.8437–0.9841	within 5%	4/6/0	0.209854
dense-256	torch-cuda-eager	2.285092	2.222405–2.321416	2.074269	1.988693–2.170833	0.90774	0.8457–0.9888	within 5%	4/6/0	0.209854
dense-1024	cuda-native	1.035756	1.024296–1.041694	1.019724	1.008364–1.043616	0.98452	0.9572–1.0301	within 5%	4/6/0	0.013684
dense-1024	cuda-cublas	0.891263	0.888287–0.897706	0.910766	0.883686–0.927351	1.02188	0.9728–1.0563	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	2.166253	2.094848–2.205652	2.246345	2.235781–2.301702	1.03697	1.0050–1.1122	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.455598	1.452973–1.460220	1.429555	1.417057–1.472713	0.98211	0.9573–1.0241	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	1.288269	1.288269–1.288385	1.265831	1.249911–1.284274	0.98258	0.9378–1.0143	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	2.344484	2.280553–2.389727	2.304294	2.215015–2.397918	0.98286	0.9106–1.0572	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	1.080323	0.998317–1.123514	0.964716	0.947835–0.997580	0.89299	0.8288–0.9993	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	1.072694	1.036861–1.096950	1.023499	1.000215–1.053875	0.95414	0.9011–1.0280	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.198707	2.189704–2.198707	2.059556	2.018206–2.084269	0.93671	0.8774–0.9853	within 5%	5/7/0	0.0796191
dense-256	cuda-native	0.868017	0.868017–0.868276	0.899779	0.876784–0.987614	1.03659	0.9903–1.1528	inconclusive	4/6/0	0.209854
dense-256	cuda-cublas	0.878098	0.869495–0.879842	0.926142	0.887748–0.952435	1.05471	0.9903–1.1234	inconclusive	4/6/0	0.209854
dense-256	torch-cuda-eager	1.735846	1.732201–1.741756	1.900549	1.801062–2.051969	1.09488	0.9985–1.2187	inconclusive	4/6/0	0.209854
dense-1024	cuda-native	1.201768	1.158787–1.244525	1.263948	1.209654–1.350435	1.05174	0.9716–1.1715	inconclusive	4/6/0	0.013684
dense-1024	cuda-cublas	1.054764	1.014149–1.105970	1.117646	1.053180–1.187466	1.05962	0.9497–1.2009	inconclusive	4/6/0	0.013684
dense-1024	torch-cuda-eager	1.765022	1.615681–1.824547	1.947777	1.763957–2.064167	1.10354	0.9633–1.2766	inconclusive	4/6/0	0.013684
dense-2048	cuda-native	1.568860	1.545627–1.592413	1.574775	1.532530–1.612615	1.00377	0.9535–1.0581	inconclusive	3/5/0	0.836412
dense-2048	cuda-cublas	1.401909	1.372283–1.421055	1.372230	1.330302–1.412624	0.97883	0.9209–1.0374	within 5%	3/5/0	0.836412
dense-2048	torch-cuda-eager	1.961140	1.959844–1.967150	1.932678	1.815764–2.043515	0.98549	0.8978–1.0761	inconclusive	3/5/0	0.836412

cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device

Case	Candidate	A pooled ms	A process range ms	B pooled ms	B process range ms	B/A	Saved CI95	Strict verdict	Observed I/M/R	Max TrueR/tol
dense-64	cuda-native	0.731423	0.729778–0.732930	0.708711	0.706912–0.709467	0.96895	0.9378–1.0112	within 5%	5/7/0	0.0796191
dense-64	cuda-cublas	0.793694	0.765811–0.812902	0.761303	0.758501–0.766990	0.95919	0.9152–1.0202	within 5%	5/7/0	0.0796191
dense-64	torch-cuda-eager	2.161590	2.098235–2.249855	2.132350	2.090624–2.222311	0.98647	0.9238–1.0760	inconclusive	5/7/0	0.0796191
dense-256	cuda-native	0.689377	0.682396–0.696560	0.640475	0.636805–0.642178	0.92906	0.8929–0.9734	within 5%	4/6/0	0.209854
dense-256	cuda-cublas	0.722757	0.717894–0.732221	0.653315	0.641282–0.669409	0.90392	0.8577–0.9760	within 5%	4/6/0	0.209854
dense-256	torch-cuda-eager	1.875619	1.833271–1.905965	1.684773	1.674403–1.698031	0.89825	0.8586–0.9488	within 5%	4/6/0	0.209854
dense-1024	cuda-native	1.016820	1.015664–1.017135	0.966043	0.958612–0.976279	0.95006	0.9256–0.9783	within 5%	4/6/0	0.013684
dense-1024	cuda-cublas	0.880308	0.858733–0.892875	0.832805	0.819931–0.847012	0.94604	0.9103–0.9945	within 5%	4/6/0	0.013684
dense-1024	torch-cuda-eager	1.967815	1.830682–2.008050	1.809462	1.799338–1.812717	0.91953	0.8619–1.0124	within 5%	4/6/0	0.013684
dense-2048	cuda-native	1.389899	1.370019–1.402781	1.397696	1.390467–1.422143	1.00561	0.9796–1.0468	within 5%	3/5/0	0.836412
dense-2048	cuda-cublas	1.198602	1.180671–1.230419	1.220894	1.188011–1.258983	1.01860	0.9620–1.0822	inconclusive	3/5/0	0.836412
dense-2048	torch-cuda-eager	1.902791	1.851777–1.947435	1.971213	1.971213–1.972118	1.03596	0.9634–1.1471	inconclusive	3/5/0	0.836412

Current-source framework comparisons

These are descriptive ratios of current pooled-round medians from the same matched configuration, not strict comparisons across frameworks or scopes. Every higher current-native median remains listed below.

All current matched scopes

Configuration	Native	Reference	Lower/higher/equal	Current ratio range	Same known I/M/R
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4449–0.7432	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.5689–0.7529	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3160–0.6270	True
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.4275–0.6555	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4347–0.7091	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	cuda-cublas	torch-cuda-eager	0/4/0	1.2703–1.4640	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3313–0.6561	True
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	cuda-cublas	torch-cuda-eager	0/4/0	1.1085–1.3721	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4051–0.7142	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.4308–0.6292	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3083–0.6204	True
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3379–0.5493	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	cuda-native	torch-cuda-eager	4/0/0	0.4684–0.8148	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	cuda-cublas	torch-cuda-eager	4/0/0	0.4873–0.7100	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-native	torch-cuda-eager	4/0/0	0.3324–0.7091	True
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	cuda-cublas	torch-cuda-eager	4/0/0	0.3570–0.6194	True

Complete current higher-median inventory

Configuration	Case	Native/reference	Native pooled ms	Reference pooled ms	Ratio	Native I/M/R	Reference I/M/R
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-64	cuda-cublas/torch-cuda-eager	2.731109	2.149955	1.27031	5/7/0; 5/7/0	5/7/0; 5/7/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-256	cuda-cublas/torch-cuda-eager	2.880930	2.135820	1.34886	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-1024	cuda-cublas/torch-cuda-eager	3.217680	2.242156	1.43508	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	dense-2048	cuda-cublas/torch-cuda-eager	3.448856	2.355818	1.46397	3/5/0; 3/5/0	3/5/0; 3/5/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-64	cuda-cublas/torch-cuda-eager	2.370928	2.138888	1.10849	5/7/0; 5/7/0	5/7/0; 5/7/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-256	cuda-cublas/torch-cuda-eager	2.400638	1.874515	1.28067	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-1024	cuda-cublas/torch-cuda-eager	2.604455	2.018583	1.29024	4/6/0; 4/6/0	4/6/0; 4/6/0
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	dense-2048	cuda-cublas/torch-cuda-eager	2.838779	2.068932	1.37210	3/5/0; 3/5/0	3/5/0; 3/5/0

Source, controls and raw evidence

• baseline/cuda-release: compiled ce45ea9cc6d38cea05ef04cff8a321507e9684d5, tree 6cf00b82bac6f75e89093d1eb627f36d4fcc7d9b, identity SHA256 e7232ead6d14f51f9f6081b982c63c3fc3c9c553876d2f4daa052ba5c8fc3d47. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: f5ae10070dd259e06e629113ac3c101d8c0e2248a2b5810239d3779cba11de25.

• current/cuda-release: compiled 14c37277de16182fdc7f4cfcf9193e0b352791d1, tree 26fcec1817f59f8dd680980fffcdf8fe0907e628, identity SHA256 daf34907780c5b6fd366347ff81161904289228e9cf5960e85c0c669887e6466. Loaded _core SHA256: 84d7166a496b4190f373fbdf2ffd057af8c7ada67f53670aed9d30ba78ba764a. Loaded _cuda SHA256: 2b95332f5e7cfa622e5de9eeb109ed8d6b2798f7e9eb9f8ee7d686a9c1f71fbe.

Manifest SHA256: 7c5b216d188e49a1588d857478019fc3ab8a00f5b84ccb04772bff4a087554c5. Plan SHA256: 3e9f579275d06bb4936f96e4759aeb120e2d7a96344b17750c1728b264887de8.

Every full strict report and actual comparator

Configuration	Verdict	Report SHA256	Comparator SHA256	Comparator path
cuda-release-cuda-dense-f64-v1-none-api-fused1-host	inconclusive	65bc999b80e9e7ebe31af9c467d13f7ee53c70da1ecd4d126088617325df2f55	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-none-api-fused1-device	inconclusive	5d3b20c295a8955639aec2a8326ac62e54efc0468192e3f3a3306eb6e048d3e1	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host	inconclusive	e5cd53aeb787ff91bbeea8425d9e86680cae2c2a95e52e05b45f65616ca0e8c2	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device	inconclusive	608799dd2f62eebfd162cb7b41949bacbe706c5e6393ac8e15e13e34eb9d8c1d	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host	inconclusive	bced81b819e8c64080f5845665813cb35551ecc640a4cf78288bdc501c048405	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device	inconclusive	77765492506166f1647f6fd1663c472712f5e6e7d055ca6f6fa76c2a9af98f5c	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host	inconclusive	c76edd07ec43f8ebf4ad7c76230d1e4b9cc5fa992c2025a42bb23a776fee816c	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device	inconclusive	39e60ef69ee649d994450c802cad2022d1799b8c4c9df244c46d07308bf8de5f	b057d071ebe47e8356a38cc2c43624ea15ff13a34abbcb02b249c83c52f23063	build/restart-optimization-20261017/launch-stages/cuda-release/tools/compare_performance.py

All original raw/import/log identities

Run	Raw path	Raw SHA256	Import SHA256	Log SHA256
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b01-baseline.json	3c3dbf178d7285ff9f2e5574beef9edab84a6509dd78b2998a4e13592daa3031	8ba912d0cd539dd09ad2b8aef0f0fa536bcea8b0f8b2c3396360060df0b24f9d	ff223851666e8c58c6ec634eb34525eb73a58e854dcfd688843342d19467e0bc
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b02-current.json	792c2bfe701064c217f9663d36f660b8b170b03fc59ada03f4e96c26bd389bca	95f34167b0e55426bdf18044816022cc07661186b6179e757bf712b9cd6899e2	f0123982bd310b9f251246bcf1ea1a0e22b782cf4437755f395a326767ac4382
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b03-current.json	e00ab6e4b9fb269712925e41a2ab079b2e5def5285b851fa1d3a17ef0237cbfd	b1920fc4d3aada21dcd7bd35059126365dfbd62c8bd4bd2ec943a8fa41e2bc20	fd4e51ff1593f697cda47f24454d71a516ed8114d299471d516ca70c17915e51
cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-host-b04-baseline.json	eb177e9c7d62cde7194cdd32fd48d56467aa170be055bdae1623863642caf7c7	e8575a1ad5752af9fb65fb5a051d21dcd0d8c4e5128fa3177dcdc5eb7dfe7cf8	15a6dace9f5acbf482907686bb50cd04b75919c4913383b4e9fa2e57e45ed546
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b01-baseline.json	2985e4b99a3e87cac9c5ef97d16b24573708e41f286d0ca0c321c15c0a8b919e	420a7da2642b09b1a9d649f5ab95e73ba3a8fadbc4eda10677d4fba948c7db78	a869f9ff1d59afea852d7a050006635ea2506091b23ea243c6becb66523734a8
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b02-current.json	74709516dd6393235dd49570a3371d4e0ad17b06d5daa28be4997d1616a4b29f	e75ce35490941bd0e700ff0bb5bf4bfbc205a504c9caf0166bcf6798fe4b4002	54ba99b90217bd731714339b36eda801067ef0a99dae413be8a2862ddfc8e168
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b03-current.json	66263063c7f3a07f9c8d0be7b2ebfc9f1eecc13368a8bb28544fb6ff43bb27d0	f2cf08ddfde1bab0ce118bf50407a96d1f8486f8e0ec830a81d0ef077f0e6ec9	6b1bf3faf8e0c807918683dcc20d7719a03850a6c8aca060c10fd8c63b4bc174
cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-none-api-fused1-device-b04-baseline.json	d2aaa200c0a193ec68f57251c734a76ed211c93dd9298e7e4ebc258a29d9a93d	4854e8d9fc93ed0f52f8caea9855efe2a0ea981e73030aaf62f95ab0346a1cd0	72a3d5968a3c55cf6f0b4756625bc2c9aeadc93fade7ad5b1eea86f9c97481d2
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b01-baseline.json	d63f0191a753ece8923c19275e20f6c0fa95ed1d20a2fe3947f67c792ccf9cb6	34391af7a81086416324e92e3e36fe7fcaaeecb49f9eaf0b70592780f4c94535	c6978225f706dcb8abb09a2a1be90fef91cd3012c88d2f872dadd8c70a119804
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b02-current.json	d41301a9ce308e24a1833e25e4a382720defa44686f01732229ed99c366df647	b577acdc9560afe5e1903564a12f7105cd66d47cf27b73f39be0c44a6f6b81a6	34b6c1ec29700b5778df54914b6eff2bfe238da5b35a85ed9dfa099088663570
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b03-current.json	fbb76498ea87c802d08fb8a57707328b97f83f734233a5a083ac86965b18d250	b439d75ca788a15b82ee1646c8d2162f6e34264399908bf6f883b4a99bd0f42c	82790a527c39956f508765062498739485dbaa19e32eccdd35523fa0f851039d
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-host-b04-baseline.json	c6e1a3111e29df24a10eeaa784cfe6c6d4ebdc26fee8b4a1ba3f16bc2996341d	bf5a23ac8d2697a20c4fa849ec4556b8d9cfca3a13d416c0314cec2585afea4a	61aecc2d6bd8ed97d73d5bd4a833e60e80f3ea8ff16435c2e13cc05a326813d4
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b01-baseline.json	92677ef560dd4f03512ea88400ea77822d1889d1bdee8818e473b23db7a2f47a	38ab51455441107d308f17b1c6669d86eb7d7c8f920267966a1b9f6a7457215c	9bf61133ebb679abdf71f4bf32cad68506ca48ddc2a4108c55e652e23bb8821a
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b02-current.json	c0630003267fbd12eeac313251fc9a81c7b96cda680612f8e8a78c9fd6a3bd70	9352bc1746a13997acc65a3b38e0c7b283f283d543fa1cd5fe8d8e049ac34cc2	b3bc2c3fa06e5d0f425bf379b3cae5369f3aba8c67ec099aecfcc4157ac3f856
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b03-current.json	33ca5eaaae4ca31bda9c632c7bea195b1a4a1998614300d0b5c441ce6f0cf861	2f0300722e0e6b42846e99e84226cb326d9dae36641ca4092bd5fc52d8f65b5a	c9b7eceda180b9d42b884f3d878e341e7de80f198ecf501dfa89b17531092a76
cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-api-fused1-device-b04-baseline.json	1d402a18227fe597df79a28da04acd4e572944e92645d9f38d420bcf53d19d4b	4d76335599bb2839ae7bdbb2f2d9ac49aac978d8df7235bfa91da9e5f9462b70	eaed79d4ff4bcd4456bdbacd03cd66f6692e576edaaa69f17fa19a4d134e069e
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b01-baseline.json	fe20eb782832f6575e64e66ede05c7f697955cbb19cd78138d7ffe5b99fac5d5	5e750d8499a98c5810e302dfde1cde939b78df7fd86ca0205810243ec9da1dce	3e5cb9707da816e05fdf8d6034649556df81122f519b75bab9a5d2339355efed
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b02-current.json	1be1daf9f79c0006c46825ee21fc99c50cd5a54377bac0990fc4f94ce9b46f01	6f085c7357c29ca6ed1331f7b5b28add41ba163725ff6997dc01dc2f381b8118	914568b8a5bacd877242d237d6f4cd4c7995bc97f22b56592ee3f1bd1de6fbe1
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b03-current.json	576d8e855fd29ce5bea3c5e9bc603ef0693e43e2ff258ae33453d2be17bfcb0e	a4e06ebd371575d6f6f15febb1af3f589d58936e10d161d996193fed70799114	a43f4cb1beec6a3efaf3802d289944aa965f2715d317265e10675551e798ddd7
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-host-b04-baseline.json	d384ec4617bbae8fe0feb3bdc03692b121f5ce314d56225d9f4fcb8f7f6e8468	34609631ea27c535606cdcbb031874a753afe8dc9f2fde329fc1f377bef23776	9637c075f0dc923a71c0edb83773d50c86201dc5f1f3b2dcd29fb3ddeed033f8
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b01-baseline.json	bc2c2b29069d5ab7cf874d3a12b86d5a7cf0debf2d135ac461aa20979a953ff1	272be598bb7d5338a1c935e3ec2e6ff16e5fc44f5ea02d630c165896bb5c23b0	a16ba49b7753c5e29371d950d7682e631448ceb33319cbf98d178a1a26fdc457
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b02-current.json	8fac71825f6dc1b21a02672ed7c328548050d05eb511f4504de3e024a015df88	f2bd4dae06bb4de76a60ce2a3f9f2b27bcb6ad7e90d446922607e67ee20ab00a	e404edc96d4d7cc17224fb4f088e81374fdaac056f664f57ef21cb0a746ea1d5
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b03-current.json	67c3620878519c93958c997d9565f6682b68a960250296bd3dcc5f20cf2e3f54	619b6cf06ed3f39edb8c2628b8e2e111552f7f15bd6baa0bf04a4b60844b527f	e7f93d6563d06fa8382b3c2b2b4983dd637a206b4628a39527e92c4e8d22692f
cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-session-fused1-device-b04-baseline.json	524b6ffe7f470b650d7d6514ae5bfbd3c6acca12d589e2f07f93a4bc6c35f073	0f192a74011f86604869be9c33c29fb2acd8cc62df891e9ec2f8ba4f46baa654	3592a01b52b1426ca39b8d7d5c16bbd915fb2dfd3aa9a0016e5965716451ca82
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b01-baseline.json	ed1094deb58a41b123acffce95f285dfea9180e957b0046a06d44778a51934a4	b3f4bab83f9d217c040c93f24091a6891e2690b957203765db5fcda40eea63d8	fdf43a87504f105c9d0af034034bbb176156b2bc7809ecc68c16257ede97c21a
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b02-current.json	f2189bb701d809739f75fdf70443d6f9aef8a82501950bd03fe77ff6efb2557b	8c4071d4c52acf78c6cc1ac56b3c04cce742bdc697c6f0d6e28a0634329ab91e	9f5fe106cd020842c9104cf7fdcabd8b02f5aaa4a469f78ff4304941b7963396
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b03-current.json	af1b23c346464203929a8ae275aad11c9a16785bb32248b271cfbaad28d33b1e	8313b124b3662139938030bbc9c05943a54bb39e56e05b53f03ecab8eb73d106	8398c1bb7a48ad048ed5845dfd2273527928af2ac578a56ec6c735356a281875
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-host-b04-baseline.json	1af287f0733a577b65bb3dcbae004479fc9df9e078ed335c64b97bd7b11e6cbf	5d773e7f35a79e76b4fb25a4c22685e873faa2b5731b9a224dfab5bcb679cf82	99d172ccf1ccae4c214d035a83741d1ff01d8348dd623f1b18b61b144f84a703
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b01-baseline.json	882630699b3a09739faa378e04195fa2b62b3e830f5fc2d19b4974b8129c36bb	d5fad1030e81b841b65bcad74a45919e6431190c59806a33b79d215678015969	f8a8623e144e42b79c67c05b6ca0c76e0950b6bf4fa83db65d0d0067b42666a8
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b02-current.json	14228cd467968932bacf91a97fd265f8a27a182cd589056918977c1c1471c104	628271910d851bc46a732a171c35ad0a8d978fc2a160cd03e79447edfdda1be5	d7d1d19af697150316e5d035495b2e8bfc5f4dfb25217ca3beb7e25f5bac4fba
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b03-current.json	0931d55e9d86e4d81ec06dc27be13f998cd24e641153f6fdae9cafbbfd7a7474	8ae39a79728fb5a00eb7de239ca0b6201cf71064d8c40ebd32dd0a3588aae275	a54f2ec25dbd7035b2af667adc8907fefa9d598c7be0cd877856ade8e9d870b4
cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline	build/restart-optimization-20261017/launch-scope-abba/raw/cuda-release-cuda-dense-f64-v1-jacobi-prepared-fused1-device-b04-baseline.json	8dcdadaa3463521661d20584b2c8f66e753ecfe6ca8a3b12e09f6bdeacca7fd7	a99cb00cd91d0c27766b405021c5c36af4585f2e82c0d86c3a8006b7b40c7959	3b0fd904b82d2db3f2e4030e06d05fa42b6a35b78aa32abf7ef3abdf83873a97

Public reproduction

Follow Installation, CUDA and Methodology and reproduction with project-local dependencies. Fetch each actual compiled source recorded above, build both sides before quiet measurement with matching recorded compiler flags, and use the public make pyprof configurations from their immutable plans. Preserve all candidates, scope, M, fused flag, mode, threads, affinity and input suite. Run A,B,B,A in fresh processes with seed20261017/warmups3/rounds36/repetitions1 and distinct outputs. Compare all blocks with the matching current-stage scripts/compare_performance.py --max-regression .05 --bootstrap-samples 10000 --seed 20261017. The analysis JSON retains exact argv, source/build controls and all original paths/hashes. Ignored helpers/raw/stages are local evidence, not downloadable artifact links. Reproduction creates new timings.

Renderer SHA256: 3c2d22577ae5e382eee57e1aec1e4938e12efa1eed9bf0a2d4cd57561eafa1d7.

All 57 retained baseline/resource/launch diagnostic captures

Each of the four actual sources retains every diagnostic capture separately. No instrumented timing replaces an uninstrumented score or another capture.

Every phase retains its actual compiled source and all original captures; different source/scope timings are not pooled or subtracted.

profiles-baseline CUDA follow-up diagnostics

These instrumented captures are separate from the uninstrumented baseline and are not framework scores. All ten-repetition captures retain passing before/after independent true-residual gates with identical known iterations, matvecs and direction restarts. Resource probes gate a solve outside the resource-only probe; their gated work does not describe operations performed inside Workspace/prepare construction and destruction.

All CUDA API counts below are totals over each complete ten-call cudaProfiler capture. The driver synchronizes before every NVTX range as well as within the measured phase. Full-capture stream-sync counts therefore include those inter-call fences and must not be interpreted as solver-only counts. Launch API counts are distinct from the recorded GPU kernel instances.

NVTX total sums only the ten exact outer iteration ranges, which do not overlap each other. NVTX phase time, Python completed-wall medians, CUDA API durations and GPU kernel durations have different boundaries and can overlap. They must not be added into latency fractions or inferred by subtracting different scope medians. API calls can wait for prior GPU work. Captures have three warmups, seed 20261017, one host thread, affinity [0] and ten repetitions. Clocks, thermal state, Windows display activity and physical host isolation are uncontrolled. Single-process instrumentation does not establish a speedup.

Workspace probes include construction and destruction. Prepare probes include the operator snapshot, optional Jacobi setup and destruction. Session workspace construction is outside its solve capture; output and requested Jacobi setup remain per-call. No historical pinned-host-scalar allocation is claimed as a current cost. Empty workspace kernel exports record zero instances only when the unchanged Nsight stats log explicitly confirms no CUDA kernel data. An absent export remains unreported; its absence is not converted into a fabricated zero.

All 17 full-capture diagnostics

Capture	Instrumented wall median ms	10 outer NVTX total ms	Malloc/free	Launch APIs	GPU kernels	Kernel CSV	Memcpy/async	Stream/device sync	Gated I/M/R	Before TrueR	Before threshold
cuda-cublas-dense-64-api-device-solve-jacobi-standard	3.710663	41.866971	80/90	670	670	recorded	0/120	140/30	5/7/0	2.43445e-07	3.05762e-06
cuda-cublas-dense-256-api-device-solve-jacobi-standard	2.916256	34.481943	80/90	560	560	recorded	0/110	130/30	4/6/0	1.05477e-06	5.02623e-06
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	3.711502	40.730392	80/90	560	560	recorded	0/110	130/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	3.371965	35.248615	80/90	450	450	recorded	0/100	120/30	3/5/0	1.23878e-05	1.48107e-05
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	3.626654	41.482650	70/80	400	400	recorded	0/180	200/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	1.636027	17.867452	50/60	0	0	empty; no kernel events	0/0	20/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	1.868264	19.568045	60/70	0	0	empty; no kernel events	0/0	20/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	0.979402	12.676295	20/20	10	10	recorded	0/10	40/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-api-device-solve-jacobi-standard	1.261705	16.898935	50/50	500	500	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	4.074745	48.270787	0/0	720	720	recorded	0/210	190/20	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-session-device-solve-jacobi-standard	1.208967	15.409862	20/20	500	500	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	1.124743	14.017973	20/20	560	560	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	3.201754	36.395886	0/0	720	720	recorded	0/210	190/20	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	1.432609	17.084414	0/0	490	490	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	1.390553	18.336522	0/0	550	550	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	17.861432	175.607085	0/0	11990	11990	recorded	0/1320	1300/0	118/122/2	1.21108e-06	1.44144e-06
cuda-native-dense-1024-prepared-device-solve-none-explicit	21.695802	215.801420	0/0	9030	9030	recorded	0/1410	1440/0	69/139/0	8.92006e-06	1.04058e-05

All CUDA API calls remain in the table below, including profiler and completion fences outside the NVTX ranges. Raw original CSV rows and API durations remain in the analysis JSON; durations are not combined into additive attribution.

Complete observed API count inventory

Capture	API	Calls per ten-call capture
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaFree	90
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaLaunchKernel	670
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMalloc	80
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMemcpyAsync	120
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaStreamSynchronize	140
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaFree	90
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMalloc	80
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaFree	90
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMalloc	80
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaFree	90
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaLaunchKernel	450
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMalloc	80
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaLaunchKernel	400
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMemcpyAsync	180
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaStreamSynchronize	200
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaFree	60
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaMalloc	50
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaStreamSynchronize	20
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaFree	70
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaMalloc	60
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaStreamSynchronize	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaFree	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaLaunchKernel	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMalloc	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemcpy2DAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemcpyAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaStreamSynchronize	40
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaFree	50
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	500
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMalloc	50
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaDeviceSynchronize	20
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	720
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	210
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	190
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaFree	20
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	500
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMalloc	20
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaFree	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMalloc	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	130
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaDeviceSynchronize	20
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	720
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	210
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	190
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	490
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	550
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaLaunchKernel	11990
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	1320
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	1300
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaLaunchKernel	9030
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaMemcpyAsync	1410
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamSynchronize	1440

Compiled source and retained evidence

Actual compiled source: 1aae5d2d8c07243ce5e15b2fe04784ed24b12ad7; tree: 06f324529806d3edf005e4ecbb6a3c27f048f946.

Profile driver and campaign runner are different files and are independently hashed below. Raw JSON, CSVs, logs, Nsight reports and frozen source/binaries remain local ignored evidence; these paths are not public artifact downloads.

Source identities

Role	Path	SHA256
profile_driver	build/restart-optimization-20261017/baseline-stages/cuda-release/tools/profile_solve.py	20b805f3bd95a9f0c1a5d23105c17888e4bd285b754575c1540661df87b7e0c4
runner	build/restart-optimization-20261017/profile_current.py	eda0e6d1779698c6cc85d419c269cfb58150ef2c4cd8aad317ce066ea7f11d5e
stage_identity	build/restart-optimization-20261017/baseline-stages/cuda-release/identity.json	7a956531880eaee793f22d75c4cc748f53c20ee8d5569fc0f2fe4cf67411d416
manifest	build/restart-optimization-20261017/profiles-baseline/manifest.json	26d67c9d638a096bec24f03eab4274052706d7ca98aedc3cbd4c69a37bf334ac
plan	build/restart-optimization-20261017/profiles-baseline/plan.json	1dcd1c68e3f57f69031eec7893568f859476a5c4c06409a83d3cf66136133e8e
renderer	build/restart-optimization-20261017/render_followup_profiles.py	f74a687c7499ab92d6919368405045c780eb93974c217c57978e82e3f58bc488
render_helpers	build/restart-optimization-20261017/render_baseline.py	af180cd5ecb8e916450024b0d5fabd0f6b7088f9f41e0e720b95ea5552154ab0
baseline_inputs	build/restart-optimization-20261017/baseline-analysis.json	8d27f15818e58182a7c113e38024049849fc2ebc44048de4362cbd08e8385a9d

Every retained diagnostic artifact

Capture	Filename	SHA256
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard-stats.log	cd2ffcd93c9b812be9577b99e697850ef22c07ebb3995596d80e0decffb277a5
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.json	cb3aec02fc1bd7070c389306d6876f52dfb6fe89fbfcde0b2fdd4f368af62aac
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.log	273ecf6e8afca954c543c38fd151c6bb2264184ea1274ce110a55b7f26e4b985
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.nsys-rep	e9aadf3b59157de2a25ec8087268bd08a2c54c1eefd95c3037798dcec742d63f
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.sqlite	b1f77618e02ff6df8351f1472c51e7ef7f7a081de59afe9708c203d711228c3a
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_cuda_api_sum.csv	faf38535d5b3d09cb839adafb7e59af9785adc8c4e3ee4dbf83686283142c281
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	d8251fdabd8a62610a43c30aeb3ff01b3cc3400f655a166df69f60c64226b585
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	07ef20b449472563abc5208fd507c61d738c7e0838ddc0ab22ee80869e432e4e
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard-stats.log	65c51fbb846d17b8b033414d28fef0a0720800d35ac0ca8195131fe197b65a26
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.json	9dd566c3614d8be80975b5226b460ca12bc22ba973bcbc8cb324d238a85638c6
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.log	af2a6a8bb73335f93aac4036cb89be40bb1b6073fd9a32c50f19813eb4b00133
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.nsys-rep	d00cac17707cc4a6d4ddc13c1be171d9afb2b139d5e9059a4d4e20cde5497184
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.sqlite	1460b49c5cab621ccc869e8ec1edcdfc660ab1f1c03e6974818ce9e5045d0989
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_cuda_api_sum.csv	1e9c02e55bbc3705be87f01c23a6f4e6fce878846a48c36f6cf38b13f5326652
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	b820fc1b8b18f24b4614548d2cbf3de17a70bacccbeef6d721a246571860cad2
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	74f5169f8791be483e652deedc010317bdd8bd607f4b483c9b006df0da5f8de2
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard-stats.log	f5ebc1bed6146eebeef0a91743aaa5c3202d61c868dee8fb2e5c8ea81d263608
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.json	669799b8da6919871ac61b47cb262236d7d1669b9653406abb746f408c603ad9
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.log	5e6ce3814438141bd5bc4c7e51e9f57d6d2445d5731999e6b2d9d919d22cff74
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.nsys-rep	29d1194e1e574bf6aced89d59dce37d84356596a13ff15902d62a6f8a123cbf6
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.sqlite	b6c41e20b2b363eed3759a775b3edcd58e32c2829ba94fb28e2df0d6a9dc55cd
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	ef815a39d41c49cd682b8b57be3d3f759b80f2b19df6e450f54cdfb033832978
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	1eee9ec0f1e3ef479a5e7ce13f1acb6771839c40ba4caee2edaa91c2802ed6d4
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	10f3e35d0c9abba8adb8b1d6f767463e5c6ed8b36985f3892bf072a7dc71ebef
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard-stats.log	b3b6313d0c9f0108aa13536572c008bd89eb69316ad3d0b80abd00391aacf632
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.json	0c6430702709503821afaabf2e9334c769db9d9102efc0d166fdd3a7a80961c7
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.log	3d5defdee1246c4cf030783c3c8c78f63026035ec0f265c08421ffcd55e4e195
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.nsys-rep	545775261b6e68a8211bca55d408e7fb61fe19f97d132fb461a0f22e070f63ee
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.sqlite	faa90eb5254a84661a0e4c2eabb460eaf9bde0ca2d68f301dfb6e1da86405c62
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_cuda_api_sum.csv	015be3268a338f7f813e5820a06092bfe1f8cff9faa3310fe3aa5bc514fb91f1
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	913f41d65b37c77a71d51a4e34b967d535b0d84ac7c16dc83517ad667b6aafe2
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	6940bf061170bb0ec6b7434b4d7ee28de37980202c476f323a380386620302e8
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard-stats.log	9fb42ed49468fdd3d3b7523bd2a65fe3843bd2753bde6fe57c51b03ffba25621
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.json	1dabd8d66bdff72fb60bbc35a60953032052a6364f2faa6836f9625e5a8bcf7a
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.log	61658e570ba006da258fa060f90fbce378fd06494ce96006327954814ddd16f1
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.nsys-rep	d71a78c8dbcc8eefacfa14c7784db68b872c7bdc248c83d28572ba24997ac0f1
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.sqlite	96da2f7b7f2bacbb480e0363344e560e701cec0524de16f437521ce8a7a4fdd3
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_cuda_api_sum.csv	263a7fbb3206aa97acda067174d4909a1792ea1195561b333f5aa327dd461d66
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_cuda_gpu_kern_sum.csv	81d89882b728e7e2ac4352b689b7586af68c300682bd0e2908b49a125008d4d9
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_nvtx_pushpop_sum.csv	6181ba8c9f683252f6450af4a13bd0032f1acb6bd0b48dc71b3436165385ae2a
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard-stats.log	cdea0d87341ea5a12352c0d3918ad39b4e54bcaa0baba3748af46af312295642
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.json	3dc9ecc50283b010b4e357d57bbd414530d0320c7b5d037f47e7509b2a6db75d
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.log	265a1b64cef2b24237000a0290cce41084e1d29a8dda32cdaf4433d85c334d45
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.nsys-rep	f529e1995464bb02163a533708ea582ae9cd74f342c4e96fa1a4c031a1908300
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.sqlite	e86e545f91b8d6f982335823e250ea78a36eb88174f6ce91d8ee263ad98e361e
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_cuda_api_sum.csv	40a0f93f1449752dc45dc55f6b91c1a41eb3558c5213df67b80033cd071b4a11
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_cuda_gpu_kern_sum.csv	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_nvtx_pushpop_sum.csv	c777143aa2d07da6db36cec8a8ca734fe60eb8082efc7808ea52607b876c7cc9
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard-stats.log	bdda8bacc62f587f32ff2f557083dcd7451db60903d8fde94108edf156c7683a
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.json	14937b25b31d5d43da6b90fe15ef2d3201011382ed79dd3e7fd48a99ee1e3fc6
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.log	92cbc6981ba420efd07800cb33a334706ba4c485bc5ba76cc3997e1d9f64d147
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.nsys-rep	9a25f71cdecb3a1435777ad3ec53d5ea4340b65c70a447e4d67a7389d676671c
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.sqlite	8026e5b36f103d98959bbfb6780841ed5ef22906bcaf4d91c5dd5736611f6d70
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_cuda_api_sum.csv	967ab2a426752b4204f6a757429dc60938353a6dd4533a379538ffe9c61359d5
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_cuda_gpu_kern_sum.csv	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_nvtx_pushpop_sum.csv	2c5da3dd703b13db6fab229f034b4be00cbfe12b5e49835ea5c0fa15fc32b09d
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard-stats.log	336232dbca9cc3090e608912d36c40b03763850661dbfb9447a4fedbee117a25
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.json	a05c063370de2a98dd3b8fc25a6bb85b8d40b728efe6ce4fddebee0f92b1d932
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.log	00a6f78128745b75b2a84079e79a6142b659aae2e629cd384425a29cda1fb53b
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.nsys-rep	c27b23e9dea50c5075d5fcb6dcbf91084acff15031aac99b7b7dbdb086485590
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.sqlite	1659ed25f9d39f0dc2b02657a1d765bbe6c1343bc6f0c0397508e14fa5cd41c4
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_cuda_api_sum.csv	be116d7afa73183246c2229a12abde526ee2674ddb91d5f0c22c4685a655f914
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_cuda_gpu_kern_sum.csv	cb84ff5e05ff400d35a206bec109f127d6507695b86347c8b776bb5ab7aeda81
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_nvtx_pushpop_sum.csv	a8ed9d1ac4dace0d945c311f9c10a579769b3ef70ee51e7d531445b2eda14cca
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard-stats.log	3ead9c9f01a28f4d689ad9306a7b41cc775b174cc5205be1cd4940b4d843721f
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.json	f3ab382a71ee6a35c74e2743764cd36ce1d67cc79539d27011f62df391ca10db
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.log	d3e1bb62ba9f9235c51de0f1ab4dedfca6ce48a007ffa14a18097f9b11ede9a8
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.nsys-rep	df76f135a9952f97b232c82a5adc4bc96c3952061fef2de76e7a9b0e7fbb2d55
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.sqlite	8c5eca2aedf61a913dba393ef6141b4beb33a44f8a0dad080553d57a9e50cc1c
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	b139b7806cb872a3d145f11ce166dcef36703b9156e383434fdfa23c9a7efc36
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	89610801b887b5e7b652d3a142ffa726855856f4fa29f9a766ee1e8a54d08efa
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	77aa045aaa636f56bdcfe1ae2c463bbcd56ee7d23ba9ad2ad9813e732b2f33f8
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard-stats.log	d281a865123bd4964a4176381e01a42c659cc00d934ddd4cc16570372a1b2a81
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.json	a100d24fc754740269dc05a27785da24978f7f056c0b0e5ee952f6d7ec816e38
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.log	10c3bfa736f2550938bb7c60667da6b00c9b8e1e4653a511366f0816be56d66f
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.nsys-rep	c15e9047ef7b1a506e016cb9989d695d548f66ac304167641a01af4c6b10fdd1
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.sqlite	be538ba5b5d948a7c18998565b055db46a2813bfd753fb2fb06f3fb12470de43
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	7d7a392569501d744c8ca3c69c6d870d7f1c25e3ae0c32c294dbb5baedf4e337
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	3a8562ec354be02e04ed8ab012e11ebe5359cb3c2bc43909ff6083c2e39fbf62
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	e2b493af6cb7e8f26af09a66f61245b5ff056ce47f9b3f1c2da021d9d92e6c14
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard-stats.log	de1a3676de048878873cf4b1ec63c0c8c289208ba0d28e2af4f88c82fdab947e
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.json	8e348a57d416065c79e93f6cfee34236965181904d53579643b42511cd3e1e81
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.log	55af0e36584b8cda4a41ce8d6287c91d1b0fa137e8c91703ba94fc85019db6d5
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.nsys-rep	a94b0cd599f65fe44285e283ca6a278812f6d30611fae4cb15e9b16ac8f9c891
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.sqlite	3bf0ea2d690d28630d79a993c46e876e9e3734a6a6f0d06236222aa82f6e3985
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	e13210778ed77adbe20eccf04583043e8003c58cf3c8d40a2e93293c0026028a
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	f89dbf52f7ced12a0994b8251d13afb40202684721a0a66b2516aea17c09efe0
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	3271df9f9188cbbb97d68482d183f6847467a445740e6cb8b660e4339a481526
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard-stats.log	dbd418d496db5bf865c2adfc1070da9ff7716adce894814133757aafb66c5574
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.json	bb47d9ceafc1ecb2d287e3b8a7974a83f1187ebb92092bf0b3d474aa3834c86d
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.log	8898011a23583c50507e4daa3c9885f31d254ddde66f79f5a2f2e98089f0aeb3
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.nsys-rep	e58fffc4f7320ebc110b4051704487301ec852bdce35acd73003d8a407275696
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.sqlite	ee7ffd229ca3590befd160c679f06a1b7c5e2333b7ecb3f22433fb21c51c2a7d
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	dc05a20661aa0f5fa457dd3d0a4b76e53daa58e950d02cb760a78ddcae769d79
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	a254f5b8bc74e35a4195a1e5a54a1faf1bb47b0d393b3c31c32cda88d9680e8a
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	745059b005383dba9018309c5029477a3e064a636ef78521887fb24433d5772d
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard-stats.log	6ac47568d5ebd5eddf6211a2202c4e2cebc55fcad1d18d42c506e2cc2667b6a7
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.json	7f5dcf1e11942f98e73caa1295137683eeaac3271f04e162e1ed31cf28cf4785
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.log	f65cccdeff677d70f8c90ef89e2118c0d401f13670016cbf5db8a0bb441b6e85
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.nsys-rep	f0ce2636b259876695bb0c8a9a914c78ad0407f620b016fb609aeb3c359fc2e4
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.sqlite	c79154e58a3fa6344671d2702076b7a09bc81574f8a2a8fec57b07ea25bff09a
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	4dd964abc449f88d369a45302add66ac9eeb5bb6627c8282fe58a1e9ecb7ad03
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	e6d0de931ecfb9e37d71d40d60181f96a326dfb8e1d60b0050c51433656d4dda
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	34d533f5b03a5777acde853ba0022041d98415f909e0a7b9c74fc7ad9f5b7aa0
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard-stats.log	fd6c122ea7f8957105401e982ba78136b1fa8738307612b54ee2099b9c0f6ea6
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.json	50ae41c8f9d1172ce788097bd2bb1877e814113c43a0556366009996835df728
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.log	a6db9df1a45519f37af18dc8f6a2be80f44dbd53034e562e38d2908a47f87520
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	25ef0e234cd202d99d7a15e19f867a982319534382fb439b4839898effe6c52e
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.sqlite	47a2d72a1bf34910dfe1ab4c11fa2a6a7e0cd214dee7bef7e298fdae1704ff1c
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	c0fdc0947fc8569c8285f0b944a3abda680cbf66fc0b4826c966e7eae2e08d4f
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	47ed248c5100d4352aa7ac8e1b3dfb1744527dc58d0fe0e77a52512b52ec8562
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	727aada084f0264b1f79960a1e468fc12417e3f45a44de6e3165891e68b49b3c
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard-stats.log	14ef39d87f6a58b34c319037fd3708a4cb5cce8f8f4e5b6193310565964f5322
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.json	d8e3ababf04ce4ba64c3abe1436e1eebfe311c5449e96c94dd762feadf246772
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.log	5beca91753f1377e259a6f7bba5cf43dc9896f8f8ed37abe8ef4a46e38ed05a4
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	c7b1f37b97e4309f3c70d386fa46269353f233ba0650296787ea4d3416db15b1
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.sqlite	586fe85730a38e4c854ff47ed363fd9d23f22e0c9ee71d8db5235f47e641d722
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	eeffd849db180912ac9519acda4ed0ed0932256b42f8d2e4e839873f2ccce87c
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	be9eabea3ecf8dce2434baea7f5cdf1ca58edad7fd559932292f78cafb9aa0dd
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	021a745ac1a5f2c59f27f037b304e2646ad977f6c1345f62d3529eb5c3507b5f
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard-stats.log	eeb6179539155e9f674e11a8abda8108721c6bee5c06198681175729f411a7e2
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.json	35fac0d4171450af287e112583bb595c9202df492a9cf82c53f21d2041fffbbf
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.log	007e516e3e755ece67916f7ba2a6559ec497e6e764245dfdf1d658225120b373
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.nsys-rep	adaf460e8ed825865be906b5e67463949218a2fecd9f111a0c9ebfd1c4012a93
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.sqlite	42b1fd1fc75333528fc14653bafe5cc358d3a67c77187e86d680dad6714d4277
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	0e22d69d4ecc8a6a64d0121936cc7e96b8ee48b4467831022ded4d0be2a52d18
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	cfa3cddab963ec934ce920858f9c12468384c06e7ee15d02cb9bbcf89d1b8719
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	3a52c7708e05274042ded24efa979e349aa007a15c2947d88616c8c0520710d5
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit-stats.log	3233b12e6e371a7430c4ae920ec8c797c7d96527c83c849befed034f73ddbecd
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.json	fc6bce4e5978e197e6af854ea8c8eefd934c5ead0c22e04eed90242771118816
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.log	17cbbd401377faaa784c33ec83936846d524fd36a5105f006421670f56acb1de
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.nsys-rep	4536319c464e0a268b187c9486169c04abf61aa4feafac438a2f537aeaaf787d
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.sqlite	cc996bdec8132b6309ae1c8670c3ee303181c97753d8c050bbb392ca1f0a44aa
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_api_sum.csv	d37e93f5666c535bd012388736cb99c7f93fea040ceb6f1a025df9b1aa894247
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_gpu_kern_sum.csv	9f71501e94181cb8684cb7df93d7936a04f780b43694f1307be6f6ff859ea01d
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_nvtx_pushpop_sum.csv	31b1d57111e1e8655bf5480baf08357746a2c52355745766c5a1b582d63a4016

To reproduce a capture, use the public source-install and project-local CUDA instructions in Installation and CUDA, build the recorded source, then run the matching public scripts/profile_solve.py inside Nsight Systems with capture-range=cudaProfilerApi and cuda,cublas,nvtx tracing. The ignored orchestrator is not needed. Nsight installation, WSL permissions and GPU availability are external prerequisites. Exact frozen command argv and all original summary CSV rows are retained in the analysis JSON.

profiles-fresh CUDA follow-up diagnostics

These instrumented captures are separate from the uninstrumented baseline and are not framework scores. All ten-repetition captures retain passing before/after independent true-residual gates with identical known iterations, matvecs and direction restarts. Resource probes gate a solve outside the resource-only probe; their gated work does not describe operations performed inside Workspace/prepare construction and destruction.

All CUDA API counts below are totals over each complete ten-call cudaProfiler capture. The driver synchronizes before every NVTX range as well as within the measured phase. Full-capture stream-sync counts therefore include those inter-call fences and must not be interpreted as solver-only counts. Launch API counts are distinct from the recorded GPU kernel instances.

NVTX total sums only the ten exact outer iteration ranges, which do not overlap each other. NVTX phase time, Python completed-wall medians, CUDA API durations and GPU kernel durations have different boundaries and can overlap. They must not be added into latency fractions or inferred by subtracting different scope medians. API calls can wait for prior GPU work. Captures have three warmups, seed 20261017, one host thread, affinity [0] and ten repetitions. Clocks, thermal state, Windows display activity and physical host isolation are uncontrolled. Single-process instrumentation does not establish a speedup.

Workspace probes include construction and destruction. Prepare probes include the operator snapshot, optional Jacobi setup and destruction. Session workspace construction is outside its solve capture; output and requested Jacobi setup remain per-call. No historical pinned-host-scalar allocation is claimed as a current cost. Empty workspace kernel exports record zero instances only when the unchanged Nsight stats log explicitly confirms no CUDA kernel data. An absent export remains unreported; its absence is not converted into a fabricated zero.

All 17 full-capture diagnostics

Capture	Instrumented wall median ms	10 outer NVTX total ms	Malloc/free	Launch APIs	GPU kernels	Kernel CSV	Memcpy/async	Stream/device sync	Gated I/M/R	Before TrueR	Before threshold
cuda-cublas-dense-64-api-device-solve-jacobi-standard	3.093505	33.890284	70/80	670	670	recorded	0/120	140/30	5/7/0	2.43445e-07	3.05762e-06
cuda-cublas-dense-256-api-device-solve-jacobi-standard	2.596909	28.483948	70/80	560	560	recorded	0/110	130/30	4/6/0	1.05477e-06	5.02623e-06
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	2.770620	31.182909	70/80	560	560	recorded	0/110	130/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	3.400825	39.899452	70/80	450	450	recorded	0/100	120/30	3/5/0	1.23878e-05	1.48107e-05
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	3.357289	36.867620	60/70	400	400	recorded	0/180	200/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	1.680262	17.124619	40/50	0	0	empty; no kernel events	0/0	20/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	1.739318	18.479692	50/60	0	0	empty; no kernel events	0/0	20/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	0.844008	11.860357	20/20	10	10	recorded	0/10	40/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-api-device-solve-jacobi-standard	1.632817	18.383212	50/50	500	500	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	3.170574	38.196746	0/0	720	720	recorded	0/210	190/20	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-session-device-solve-jacobi-standard	1.151059	13.300396	20/20	500	500	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	1.251014	14.144323	20/20	560	560	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	2.998172	35.489561	0/0	720	720	recorded	0/210	190/20	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	1.235602	16.460001	0/0	490	490	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	1.128499	14.080175	0/0	550	550	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	19.728952	206.913200	0/0	11990	11990	recorded	0/1320	1300/0	118/122/2	1.21108e-06	1.44144e-06
cuda-native-dense-1024-prepared-device-solve-none-explicit	22.918593	223.527569	0/0	9030	9030	recorded	0/1410	1440/0	69/139/0	8.92006e-06	1.04058e-05

All CUDA API calls remain in the table below, including profiler and completion fences outside the NVTX ranges. Raw original CSV rows and API durations remain in the analysis JSON; durations are not combined into additive attribution.

Complete observed API count inventory

Capture	API	Calls per ten-call capture
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaLaunchKernel	670
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMemcpyAsync	120
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaStreamSynchronize	140
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaLaunchKernel	450
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaFree	70
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaLaunchKernel	400
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMalloc	60
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMemcpyAsync	180
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaStreamSynchronize	200
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaFree	50
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaMalloc	40
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaStreamSynchronize	20
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaFree	60
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaMalloc	50
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaStreamSynchronize	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaFree	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaLaunchKernel	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMalloc	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemcpy2DAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemcpyAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaStreamSynchronize	40
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaFree	50
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	500
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMalloc	50
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cuCtxSynchronize	1
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaDeviceSynchronize	20
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	720
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	210
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	190
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaFree	20
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	500
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMalloc	20
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaFree	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMalloc	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	130
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaDeviceSynchronize	20
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	720
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	210
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	190
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	490
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	550
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaLaunchKernel	11990
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	1320
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	1300
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaLaunchKernel	9030
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaMemcpyAsync	1410
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamSynchronize	1440

Compiled source and retained evidence

Actual compiled source: e5c587889906511b0f9e74f42d88d8e857dea3d0; tree: 241772b9eb7a0ed25bfbf8e82bbe4118386991ea.

Profile driver and campaign runner are different files and are independently hashed below. Raw JSON, CSVs, logs, Nsight reports and frozen source/binaries remain local ignored evidence; these paths are not public artifact downloads.

Source identities

Role	Path	SHA256
profile_driver	build/restart-optimization-20261017/fresh-stages/cuda-release/tools/profile_solve.py	20b805f3bd95a9f0c1a5d23105c17888e4bd285b754575c1540661df87b7e0c4
runner	build/restart-optimization-20261017/profile_current.py	eda0e6d1779698c6cc85d419c269cfb58150ef2c4cd8aad317ce066ea7f11d5e
stage_identity	build/restart-optimization-20261017/fresh-stages/cuda-release/identity.json	aba137408ee38d8d66ed6e90c092ad5cbe0afe31193999fd19cd936fa0d23d77
manifest	build/restart-optimization-20261017/profiles-fresh/manifest.json	74ad38680414aa28fbe09aa9f2cde6401c03b2461d98f85ab45b9a2912a01704
plan	build/restart-optimization-20261017/profiles-fresh/plan.json	a706526d5bb51d09eb10121ab7b8b7c70183bec34a8a8e5f93f1631bf05ce3a7
renderer	build/restart-optimization-20261017/render_followup_profiles.py	f74a687c7499ab92d6919368405045c780eb93974c217c57978e82e3f58bc488
render_helpers	build/restart-optimization-20261017/render_baseline.py	af180cd5ecb8e916450024b0d5fabd0f6b7088f9f41e0e720b95ea5552154ab0
baseline_inputs	build/restart-optimization-20261017/baseline-analysis.json	8d27f15818e58182a7c113e38024049849fc2ebc44048de4362cbd08e8385a9d

Every retained diagnostic artifact

Capture	Filename	SHA256
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard-stats.log	c7b513a79847793b1d797cc6968b0fee42984f442c65c1da17b39b9f2e26b811
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.json	e1353f93209f17d50d34f59081451646be0997219efb26ed876005ed53d170d3
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.log	90f142ce0269d4f5d9cdb3597cba088becb2b6b1c22c21534755f21b16c88e08
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.nsys-rep	2979ce9f09ef622b02d416b7a215009c83c7a45c3a1d21d32d91f5582e2d3fd6
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.sqlite	47f84a980cd781b3826ae5ccca08624113ea5f5b65f7be378a94ceef62eda0e9
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_cuda_api_sum.csv	c1d851b62123f227171e0a2daf0c6deb1c175e592ad94ba50a80852b1c635f6e
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	8d38c4ca7fae318529e1143cf14654c51467ccb137f42f195fc2549486d6d6e0
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	955272d77aeff3d7b7ab2e011d006ec269402e76b56d1f6c87a6393a0acabaf0
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard-stats.log	586bbff50cb8317bad4748d9539116fbab0256bb19efc3aebd21b93cc2d22cd7
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.json	fa44af7a1407067a01392c9f08bd7e26e33450892e03be38e233818a5a76c2d8
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.log	dacf3338ca206361623380101ebc00919476585e1a06a6e5d1ecb321dcb5cef5
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.nsys-rep	01bb9c325fa9602b822c72b87060d3bec5edd5a78b0e0bd4399f2ed0596d55c2
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.sqlite	e1e649910f6dffe176281506af41fefc3f543e63e965b0bdde07d8cfc6518317
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_cuda_api_sum.csv	9620ee59814a88e99203c3563726fb3e12b76f38662f88e10be95a42100e948c
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	ce8fddf8c2fe33e3d796d54473200d0b356b990397e3ce5ddeca8a1d7392ed2f
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	0b629918f09de93f3c47ddb38030480684cb3d72fe77a306658ba28b7ba5c345
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard-stats.log	51ad1141d1d7758155f6ae58cd041eac60bf488dcc912aebf8d07ddc7314ec87
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.json	b1b980aaddffcf243e7fb5ce4f3a088de2da7db4ccc27650a73c081848812544
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.log	62b40fc20c5cf2876e7b224895d3886f8f6bf376113528691b86b616b8882fb5
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.nsys-rep	db03b8d8c56affac0a21ff4103c67f7a18c2461d87e9420a5e58e37a9ce3b435
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.sqlite	a2af36e607925d26b7cb43019b4b27c272230c4e3ccb09aad7403f59ecad7138
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	94ef7eb020c95c1e2cb3f4731bd51909f90a6711907ea4198b6e933146078325
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	6f9180f87a795a8c627038c995cf0fa33ed21b16a0fc07042687e1f776490489
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	72df202e7ae289dd9bd941e2bec95359bd28bbc73cf7258b21fd865c38d801d9
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard-stats.log	007a80e01064fa41f83d773e18f1a428ec150d0f005763d20d9f4a99e6eea7fb
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.json	4351ccdfde96b7791301f9b9df22f235984c40f8570aee0cc9bb2fe2cf616fd6
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.log	830090ebfdcbce1402b91a653574e7dcb56823f0a0f5beebd683e9090cdc3694
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.nsys-rep	4ec7b256db4cbff4c6fc753cd3b3f0c7d9fbc9722ea27c3514d4ed2f8a31fa53
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.sqlite	e782109e19fc49837d8a08be76ab0e5df10f60ccbffa96238cf8306451129452
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_cuda_api_sum.csv	c303f7e67979c654618e6888636dac4b38acc20c21c9111153203126441237c4
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	36a2b599df06d5a3aa0c1db337e434b14327140e990a2470af42608e8a07cfe3
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	d179577c62963707261a9a1d0bb2c30e4246220e147a734b1ce58fd620ea626a
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard-stats.log	b7932c28076bcae8cc07df19b743f68d0799fc909ae227d41d65f7b15f73e4d0
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.json	b30dfaa223e553dda7a1c0ff0d2040c6b67f1840bbe5456fe3bef3d14b8b229f
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.log	63d73fca767e71591454e3499eac31e2c470507b98db4c752f380e8ae87b0072
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.nsys-rep	cbbc262bcfb590fe8833808e10c8ae881bd435bc760b0e4e3e6e2166813f5aa6
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.sqlite	4bb92d0b2e6483444903cde46fc577f738069069357e88f23b6ea46e8c083945
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_cuda_api_sum.csv	cae559c39be6ecb50fc1810aeebde3797a522237cf316b788ce2b6682857d1aa
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_cuda_gpu_kern_sum.csv	d6bdf7bb05b1c0316e51ab7a4740b315ed8323d3ec8f4eb6a5e04f7585895eb8
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_nvtx_pushpop_sum.csv	5446a582de06025ee2705fb9db2f0f43303e131e63c95ee8e15a7d16a70d6bcc
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard-stats.log	d8f6f1a0070cd2ff584b2c0894a260c64e47f6419c61f5cbe22c97473b30b89f
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.json	7889a42df83e8f3f74383bc3e8d5226e5566f27787ac29c74bb0f0fc16ca441a
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.log	c6734f036d3a3dc8dbe334875c82103ba4473efb3498f6b4360c30819f29ef29
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.nsys-rep	aa3cab5d96420c0d051e0eaa8adc6b404b9cf951730efb3fd1cbc938990959e7
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.sqlite	51decaaeeb35d7102a3b580dfeb8ff63719ee7e889598692510fee291d6eeb8b
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_cuda_api_sum.csv	7a50922cedd89c941f2a86e8f42cc59dc9aa30d518a26433901b5a7bcd86a61a
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_cuda_gpu_kern_sum.csv	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_nvtx_pushpop_sum.csv	70e6d335c243f7dc92623954bd8c666679542bc346fe644544f0cc2c2a5959eb
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard-stats.log	2babd65b794c2a71b3c753d47d480d614a379983958de1c79166f68b25f0947d
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.json	604104a76814d616ca61e4a9dd55f34b1dfd254842f26fe6f6dee16a7cfbe535
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.log	6721ec27984551316a0262bfb97a97d015138c65f0c5ad07d6e09ec4719e58f6
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.nsys-rep	8819aad0fcf6ed93124bf9f5529d55708305cfee71baa725b6f102f226c7a339
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.sqlite	92af46f73ff8106741f1832baa737768586cbbc14aac4dbb933cdced32113378
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_cuda_api_sum.csv	b9a3af5102ed361c4bb78d8664258ca1eb1747433e5f49e32ab04a50c9f2f2a6
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_cuda_gpu_kern_sum.csv	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_nvtx_pushpop_sum.csv	14b09c7ad9774e68f6bfed0220fca824be02d798b2e2ce3f3835d656a2313964
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard-stats.log	2d211975b8f0030431ad421f6331ea9a60c0038a471e807f617ebde8778e2fe2
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.json	0cd0625ff8a03dbdc2c10f37fd1eabe5b1283b924729559ce23a20f06c89c4f3
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.log	83a4d2a00b62477c03af9450d58e992dbd92589b03c38978ff6657c22d229959
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.nsys-rep	8b51778df50d0e57fc6399704cc7fdcd6e21ebfa6980090fbd2965a0132fe7d4
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.sqlite	763ecdf08aaaf27b297a517317da5c38e07f46f19568b7db8a006667378f0fbb
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_cuda_api_sum.csv	dbb12a5deb8adc64a9a7d9337d1b7ac48870f3a50ca20b0d2c638eb1e8367e3a
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_cuda_gpu_kern_sum.csv	972a61f520ecc5414490d595fc9eb306e4a753648967c8336be3df5cdd5fc5eb
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_nvtx_pushpop_sum.csv	b38f1c3d4b971accd4ef760dc4b64bce0d858da6e86bd5248ca3ec1e2e285574
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard-stats.log	c37f80ec205c0be2e0c82c7efa7f0d588812c30f9670fc5c1291f58764679b83
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.json	fd08e6ab98ce4478d6596ceca7c1c1997e392a54df364f0d1f9bfa280b0f5010
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.log	3b0ed3b1c033eb34cb0fccd6c2610334247dbd08a9c93972597435c2ac2693fa
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.nsys-rep	a6da292558c8331928cd5ed55c1457bca1b96e124dcc3610a6d981a8f465b3ef
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.sqlite	af1ffea242dc65ebe108ce01c20dd981afdbd40bc3805efc2405612db2128636
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	3a3fd08f48ae0993247c6f5b7bc9ad843a404e3743b9b54b791e6573b244110b
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	3d5bb1a2d7c75a5ca30315ba0c64e7b9566755d7cfbd610be0b963d888f734c2
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	511c012e638771b149acae2ccbbcc53aab03ede1d2e2db759cbe935770b1230a
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard-stats.log	920b3293c662c855f99583a978ce39dfc7e7e9a934721ae02cfed6fec1303be2
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.json	98dc5a4b496d3b93a2ecfc033718794a16c2dec4c6ebbcc28f16783dd66f16c0
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.log	9510ce599233e525390b3114c6882cc29b87bca239e6606a266db9923d7c82ac
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.nsys-rep	ad0d185c54956a5b873b4fbfa74fd78e36af4e45b90cbf2dbbc1cf0a9acf1a77
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.sqlite	f3b3066601854906aefc971396e44f5fd96279471bd5261aa2b05336b4be2744
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	68e282edbfaac2e09b70aa36e5fc927e2ffae71983051455ed57b7308f150f92
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	70174bb5df26af2ab02ba21d089f162c7da28bb9d2aaa839d0edaac7a96b3f1f
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	aef7896336f3066c6d5f65a49f661a3d8e65ec11e7604132e0b2cad590781ade
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard-stats.log	8c71fe02e1e0f57d0627687553e05c23b28e654fd94a44375f76a93af03c31d7
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.json	8f6a65813cf25685bed50c2efd21a8a5645982e71d0c954c89f1041541fbac32
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.log	d08c57695ca52126553162a888cc95865d9569ce9aef0bbc1ec410b13da1a314
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.nsys-rep	428638aa1c0b170bd7200d7742516c51bc0bd5a31cd69ea62566d8c16035d221
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.sqlite	17642be722aaad33b15e87656a56270689e2dcf0c3730f9c80d08bfc81bc4a01
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	4349a70b8690e641b89f29aae03aaf7110d08b6dcb5eedb48a0cd8f68cf7d5f8
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	4dbbc6db38d99fde533bbf6b771aa26542d3348f263319a557434d533b64edf1
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	8ff6007f0d9b1585b83d2c6f86b1b1e330e59f78e5ba77fcc06c35dd1554a1a7
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard-stats.log	5b1ed58e5479c792c81ed7fceedc98bfc983e01acb764566e7cbe33038358164
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.json	9e4ff34dbb38c11f74810f26c1db9c1d59c37f2efc2a2656870bcf105b636515
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.log	89a9d0269499d4ce2643339a9fcfdd80867b6f7c84040d491084b11b9dc9eb0d
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.nsys-rep	ee7a0ee30d5986812bd0becaeacaf2636fe9684fa4d246ddb32138f83d5f9e1b
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.sqlite	5aaf1d1bfcfae4acd70e4fed28be48538daeeff268a5d07cb8e96df418484fa6
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	67cb09c648f5378bd0d5484a0acf77b81110bdfc427bd7b2c4381d44a2db2c77
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	99c4a46f8e082d4727a761aabc0338dd53f22a44384c71a0a8b0d732e107ad87
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	254f7dc5cde5bfb87bcc50f6c2f980150d8c17055053e1de869dad520b272b48
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard-stats.log	478602accfb5e3acac0a21bdf44d5919a04c2830f92140a940022e8f72065538
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.json	a558c12d4a266b0e1df6475b5e90d0d1edde4d00af3429ecf8b4ccd31e21c862
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.log	47ebd31dafb0e3c23a660ea04d2621169d777030713fa956e4d47f8a9623fe32
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.nsys-rep	aca6e7b3c1dd0682427dbb04ac385eac2d1ff6185ac0cebf36e3827ea9ad67ea
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.sqlite	9eb95232204065d0cea24683c45e302e0ba628543ce117bfcc56ea66482c0b85
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	649d27422871e4e2fb30045b6395ea70862c2fe276b35603fa5213dd5b6b1f94
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	e12a76c459e5fe77205152ca96e81eb6f7ada5a6ae92f334108e7637728dda46
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	916115625a23a4b3c7e0edce5e715ec988c0c58652e7afe5ce16e0c7f1147d4d
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard-stats.log	4476c3e2544b2659188fd2cfe1fab1e9d2a2a7f1313cfd2674b8fda4dcd5885c
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.json	974ec7312921c284bbb25aa9d1107e0f9400ad0c8a93613eab4bb9437f74d22a
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.log	169b0d5e7a151e6bbd9f9b95500873b0feeab03b3ab92d6fd17815d0666e50b4
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	8faab530e73ec886fd960ff9575b7bc833290ef7103ccbb279a8629c66eab9a6
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.sqlite	cba80202817dd6415f6b053afc4eb29d15ab48dd0973fc32a68a55322d653c7e
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	f988bf95d97d874b33593b5ce4084c86e12563a83347213e02d187d28412f5c0
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	65351ffb7b5b5710b959b209c2cce91768026dd04312f2774ab88401586685a2
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	7d220bbb5d3dc5961a2ef2dd96d7a927ca0f05645501434e9491c8462c847c4e
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard-stats.log	06377512679ed27d927eb7fb96ade7b77b108ee33c99cdd6ec4037f0d2cf5965
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.json	e0782f94d6aa284e28459abebc8f63b1cbed64d31ca297a6b293426251b039f8
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.log	dc20e7947062cd5a6109ff5338403783c994c5db599d14f2e808c2a30550183b
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	e21536b816428f48b81a908b2e3657a7b6e45e2385793f8acb8a6b1a707be37d
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.sqlite	cfee4ae9528f3185d79051263671bf07f61f25ec39609f96c81d12baf05ffa9e
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	375b9c001992b9e3e9397e11b6c618245d893e0c8149b3e408e2338fcc573a2e
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	91cc2550486a1d8357bc5ace636583e21854791c94ea0b39a16d36429e06ea85
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	b0cef57a990f1f76cf620258f0d0723f90276e5a3d5315a516565d0e107e13c5
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard-stats.log	51672d78db947521d4b98dfa03f3ad9e7c1c984b1133eb7f8e5fd5449993a8a6
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.json	18c69f384560ff45f79b59917d3c6a73c360c3137bd601f677bcdb8cf1ce1a4a
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.log	e2028edf00d97f7911164c2c6b109b68c14c6c539ebf43bb289e35cc206eca39
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.nsys-rep	240ee404b72a594da95e135783d633dac20b6ffe405e888930bbb04e3209118f
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.sqlite	12af363875dd84cc3025c51662b42a5a9f9cc128e81c83da50c6d0dd03087c3d
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	0406e509d65fe966564df8efe011f3497a3f3e17b3530ca8cc9d1e968b6608e6
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	fb463cdafb87b1dbd1e6c6082912268ec6c770f15f7dacbece545001e209cf96
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	362e23b53d17ad9a421b21258c9990cc733c3beb8fef88f5dbae914abca2b96f
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit-stats.log	74d044de916356ff92b393263a91738c06eaaec26a2031394ba5fedc7c0ae652
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.json	701dfa7df7dd0ed652c353ec80b3dd9d555cb9cc0910bc66f2a098786ee09cb0
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.log	029335c72f8eae62ca0744265687074e3bbcf08a043482b83ef9c2734e34d4d8
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.nsys-rep	b1a8de18ff6d276d6bbb84a41944c4299754165ac7d1140a0cf45b7b9ea94e41
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.sqlite	4964be50aa90f8b90c40faa74d547e0702b70127391689e31ed7a3cca70f6d08
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_api_sum.csv	c3dc69f6914c2315687c503e932eb078435cb8815f116efe721ecc962d1900e0
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_gpu_kern_sum.csv	8c6efe794252bc207ea8feae008c7e46e0215617d535b909dfbf9b3d94f30c23
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_nvtx_pushpop_sum.csv	498d2e6bbc78e3eea8d517ab3e80603728d64c5f3cae3d0791db095a709cfafd

To reproduce a capture, use the public source-install and project-local CUDA instructions in Installation and CUDA, build the recorded source, then run the matching public scripts/profile_solve.py inside Nsight Systems with capture-range=cudaProfilerApi and cuda,cublas,nvtx tracing. The ignored orchestrator is not needed. Nsight installation, WSL permissions and GPU availability are external prerequisites. Exact frozen command argv and all original summary CSV rows are retained in the analysis JSON.

profiles-fresh-span CUDA follow-up diagnostics

These instrumented captures are separate from the uninstrumented baseline and are not framework scores. All ten-repetition captures retain passing before/after independent true-residual gates with identical known iterations, matvecs and direction restarts. Resource probes gate a solve outside the resource-only probe; their gated work does not describe operations performed inside Workspace/prepare construction and destruction.

All CUDA API counts below are totals over each complete ten-call cudaProfiler capture. The driver synchronizes before every NVTX range as well as within the measured phase. Full-capture stream-sync counts therefore include those inter-call fences and must not be interpreted as solver-only counts. Launch API counts are distinct from the recorded GPU kernel instances.

NVTX total sums only the ten exact outer iteration ranges, which do not overlap each other. NVTX phase time, Python completed-wall medians, CUDA API durations and GPU kernel durations have different boundaries and can overlap. They must not be added into latency fractions or inferred by subtracting different scope medians. API calls can wait for prior GPU work. Captures have three warmups, seed 20261017, one host thread, affinity [0] and ten repetitions. Clocks, thermal state, Windows display activity and physical host isolation are uncontrolled. Single-process instrumentation does not establish a speedup.

Workspace probes include construction and destruction. Prepare probes include the operator snapshot, optional Jacobi setup and destruction. Session workspace construction is outside its solve capture; output and requested Jacobi setup remain per-call. No historical pinned-host-scalar allocation is claimed as a current cost. Empty workspace kernel exports record zero instances only when the unchanged Nsight stats log explicitly confirms no CUDA kernel data. An absent export remains unreported; its absence is not converted into a fabricated zero.

This diagnostic variant was selected after the initial rejected resource trial. Both original and follow-up captures remain separately retained. Counts can verify a changed allocation/launch inventory under the same capture protocol; selected instrumented timings do not prove that layout caused an earlier regression or establish an independent speedup.

All 17 full-capture diagnostics

Capture	Instrumented wall median ms	10 outer NVTX total ms	Malloc/free	Launch APIs	GPU kernels	Kernel CSV	Memcpy/async	Stream/device sync	Gated I/M/R	Before TrueR	Before threshold
cuda-cublas-dense-64-api-device-solve-jacobi-standard	5.360959	61.858510	70/80	670	670	recorded	0/120	140/30	5/7/0	2.43445e-07	3.05762e-06
cuda-cublas-dense-256-api-device-solve-jacobi-standard	2.914120	31.078624	70/80	560	560	recorded	0/110	130/30	4/6/0	1.05477e-06	5.02623e-06
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	2.944275	36.146044	70/80	560	560	recorded	0/110	130/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	6.407979	65.298137	70/80	450	450	recorded	0/100	120/30	3/5/0	1.23878e-05	1.48107e-05
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	3.474406	38.588699	60/70	400	400	recorded	0/180	200/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	1.789818	18.873907	40/50	0	0	empty; no kernel events	0/0	20/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	1.668506	19.023418	50/60	0	0	empty; no kernel events	0/0	20/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	2.750163	30.154661	20/20	10	10	recorded	0/10	40/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-api-device-solve-jacobi-standard	1.210140	16.928656	50/50	500	500	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	3.483608	39.227683	0/0	720	720	recorded	0/210	190/20	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-session-device-solve-jacobi-standard	1.176751	16.560042	20/20	500	500	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	1.157695	15.934467	20/20	560	560	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	2.897893	38.092489	0/0	720	720	recorded	0/210	190/20	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	1.110623	15.968249	0/0	490	490	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	1.062854	14.052306	0/0	550	550	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	18.128611	174.960432	0/0	11990	11990	recorded	0/1320	1300/0	118/122/2	1.21108e-06	1.44144e-06
cuda-native-dense-1024-prepared-device-solve-none-explicit	22.997159	229.690452	0/0	9030	9030	recorded	0/1410	1440/0	69/139/0	8.92006e-06	1.04058e-05

All CUDA API calls remain in the table below, including profiler and completion fences outside the NVTX ranges. Raw original CSV rows and API durations remain in the analysis JSON; durations are not combined into additive attribution.

Complete observed API count inventory

Capture	API	Calls per ten-call capture
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaLaunchKernel	670
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMemcpyAsync	120
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cudaStreamSynchronize	140
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaLaunchKernel	450
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaFree	70
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaLaunchKernel	400
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMalloc	60
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMemcpyAsync	180
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cudaStreamSynchronize	200
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaFree	50
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaMalloc	40
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cudaStreamSynchronize	20
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaFree	60
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaMalloc	50
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cudaStreamSynchronize	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaFree	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaLaunchKernel	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMalloc	20
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemcpy2DAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemcpyAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cudaStreamSynchronize	40
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuCtxSynchronize	1
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaFree	50
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	500
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMalloc	50
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-native-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaDeviceSynchronize	20
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	720
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	210
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	190
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaFree	20
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	500
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMalloc	20
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaFree	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	560
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMalloc	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	130
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaDeviceSynchronize	20
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	720
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	210
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	190
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	490
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	550
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaLaunchKernel	11990
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	1320
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	1300
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaLaunchKernel	9030
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaMemcpyAsync	1410
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamSynchronize	1440

Compiled source and retained evidence

Actual compiled source: ce45ea9cc6d38cea05ef04cff8a321507e9684d5; tree: 6cf00b82bac6f75e89093d1eb627f36d4fcc7d9b.

Profile driver and campaign runner are different files and are independently hashed below. Raw JSON, CSVs, logs, Nsight reports and frozen source/binaries remain local ignored evidence; these paths are not public artifact downloads.

Source identities

Role	Path	SHA256
profile_driver	build/restart-optimization-20261017/fresh-span-stages/cuda-release/tools/profile_solve.py	20b805f3bd95a9f0c1a5d23105c17888e4bd285b754575c1540661df87b7e0c4
runner	build/restart-optimization-20261017/profile_current.py	eda0e6d1779698c6cc85d419c269cfb58150ef2c4cd8aad317ce066ea7f11d5e
stage_identity	build/restart-optimization-20261017/fresh-span-stages/cuda-release/identity.json	e7232ead6d14f51f9f6081b982c63c3fc3c9c553876d2f4daa052ba5c8fc3d47
manifest	build/restart-optimization-20261017/profiles-fresh-span/manifest.json	4f9a131d2fca093c80d105dbe3994c9acb3d283a1483f566f2d4c8aead3ceba0
plan	build/restart-optimization-20261017/profiles-fresh-span/plan.json	7af94555376b7bccafc077f5402d67c82b683ddf186914cc712fcc5bb223d8a6
renderer	build/restart-optimization-20261017/render_followup_profiles.py	f74a687c7499ab92d6919368405045c780eb93974c217c57978e82e3f58bc488
render_helpers	build/restart-optimization-20261017/render_baseline.py	af180cd5ecb8e916450024b0d5fabd0f6b7088f9f41e0e720b95ea5552154ab0
baseline_inputs	build/restart-optimization-20261017/baseline-analysis.json	8d27f15818e58182a7c113e38024049849fc2ebc44048de4362cbd08e8385a9d

Every retained diagnostic artifact

Capture	Filename	SHA256
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard-stats.log	758d3c3f59d3db8e3e1d5582814c0bc277cee35709faf670c566b8a0618b02b2
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.json	408174c01f7a68c44ee5fefd2aa212fff86a5a1929a1d9fb02facd2c57a33f55
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.log	1266bf83ba74614a68d74e54219a2882a69ccf4a342a55dbdc5314c42dee4ab0
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.nsys-rep	8447995f24eb39302d11910f0ec822479b48455e2d11f2853f279d721a231369
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard.sqlite	401e48c69dd125d21288c7b11af2420458a22b0b89b27532c9fcd54d92bf1cd4
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_cuda_api_sum.csv	beac281a113a112a04753c219dbe37b428ff59072442b3dcfb65d8cba05b25a0
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	6dddeec49e19d386bb1b7b81495e3fa793a3d7204184448ef14bd6707e928c14
cuda-cublas-dense-64-api-device-solve-jacobi-standard	cuda-cublas-dense-64-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	0936df7a07fc0fb40db406a08093849642d33c56de9b1a89bf271e20eb615f0a
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard-stats.log	fa1bc7f2c2ccb618e45f0001ff407758df70a632963a15bde47e5c174a43d3cf
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.json	1a88750b26515f4878aa05132a93a915e6713a04321fedaba571228f20473fa5
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.log	5ef79c12456abc26f342165f545278d7ebc9bc64854ec2c44bd2f298eafce79b
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.nsys-rep	bf55f5648e4cd170ec01bb4a75b2e3038ff3987939aff5a9a9a4cea03a5b4ed6
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard.sqlite	d95013a249da1f05a3329aac00c710e9710fc01002464e6f01a8ca8cf592b0d5
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_cuda_api_sum.csv	5b2a2785a2bdd2f83ab1ef641fabf94f91a26a370f797889630d1c716d3efca3
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	80589f2d823af4f38cbd4f82e39173d1b6d77c917f948599a0d935a28aab2a8f
cuda-cublas-dense-256-api-device-solve-jacobi-standard	cuda-cublas-dense-256-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	237541be06f1c720a3fb60429dda6090afd0193df176dd3c4f36d856cce87b96
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard-stats.log	f3ff255df13d03e11dd96b70ce6fa2521f2ba25ad7d8039164f9563e8ddd7cf9
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.json	527aedb31eb333dfd1942f2b3f55008f1b1419ee227ec386caee0e95db54f94e
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.log	12f1e5a3c3b201cc72bca18f50e8db0ded722c6c686404f267bf24f8d2929960
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.nsys-rep	ba4ebeae1314fb20d4e483c0101e4f006174862454adf42d1bc732a51c87e777
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.sqlite	4b61928d4fec91bc3c978b7dcb1d5a2b41f869c6dd0ef652195db41b5d72f10c
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	f0da5efc407332c53de1e115ae4a4cf9e0f90021487e8c1b4cab60a3032a1eb8
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	ed8e5fdef85217cbe2bf9c962bf2eeb55981cc32b9811c1647e4e085290eab33
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	564c0effd1789448241194ae205852bccac304b75613c5be773e5404e2b0f105
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard-stats.log	080ef485d8c24aaf2e312375a23b224fedf28738afc51dee31efda6fa5f270c2
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.json	44d96ea6d59d32f9ffc3050d7c97003c25cd92cbe6bedf1cfc5b2105033c44fd
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.log	989ac9c9b80c4546130e62a9401270afea5a3eadad1c832de48b7f47708b7f0b
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.nsys-rep	93a8cf0100612108d451e3d42078e56eab46fd4ed9aaad32de4312522dc24682
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard.sqlite	b0e3ff8126e19f60be7cc3fa0cebe5ace0e72920d1d2b4edfbfd31a54b1fcaec
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_cuda_api_sum.csv	b2df62c9c21e63bd93e9ffe57b3b37bed4e231b1a97590cd3d986b7ca76b5387
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	3fbc819100cf980dcbdc2c837b0c1c1f5aead95cb20ae11f0bc728fe32083616
cuda-cublas-dense-2048-api-device-solve-jacobi-standard	cuda-cublas-dense-2048-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	d62b5292caffb4cd24139bebf2785526e3ea9eb33880f1834023956f67301a79
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard-stats.log	b9fc3d35ab6535486dbe69e3494ba18aa622929b06fb0aac8a84b5d6dccf82c4
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.json	fddd8e9e4e7ee3c241bcdf3caf13dfb5f02109fc31cc9a22e9cc98ecd56c8f81
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.log	d9efd1cba2c0c67f17fc0f6c3a595b4c4f2291b0c5d29add4873b2531171ecbf
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.nsys-rep	0115b102f5748771b45314153dfd69b11fbeeab7ee9a7d10ec4354415b8070f7
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard.sqlite	92af8e8df7195d7d29ef360545b490db6322954baf0808b6b58ab99b415833f0
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_cuda_api_sum.csv	5c7b67b293fde244524ce33af7719fdbba694446f7de5126b07e65411ed4ae77
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_cuda_gpu_kern_sum.csv	5b39440121440b99665817a19713ed6a64a96ef1579bf7528b5b94816bb1fc66
cuda-cublas-dense-1024-api-host-solve-jacobi-standard	cuda-cublas-dense-1024-api-host-solve-jacobi-standard_nvtx_pushpop_sum.csv	4f9c56693f5cc9f33398e574c9f2083136aadeba3b2298855e4cfa3f03f8d61a
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard-stats.log	fab63d170518d364150d077f9a6dac3cbcc02d4de3bbeb9a26ead43795a51ae0
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.json	e6823d63cdacf5fd7e14eeda0c05106726651c42a9c2839cd63242362004f55c
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.log	d5cc871e98b06b3aa6ec81f22eecd4dbcfc94c0e9b24efd961cd80b0a4d5d45a
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.nsys-rep	3cf0c5ba47d0da4fe335c53311704c1f2cc132708fa71a35a62a211b37b198fe
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard.sqlite	f23b8a71e4e87e34aff0824af9b94bda25e3fcbae4a4d9544be9a265429d66ee
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_cuda_api_sum.csv	9af0fe5bf2556ecf0596e7c736865718892e8e5826405ec6201b9cee1637e6ee
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_cuda_gpu_kern_sum.csv	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
cuda-cublas-dense-1024-api-host-workspace-jacobi-standard	cuda-cublas-dense-1024-api-host-workspace-jacobi-standard_nvtx_pushpop_sum.csv	3b05a52e726fbe0e7144b1d224c8c7880d9629c05166ba095f009b143539a25a
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard-stats.log	67fc1619568208fd14b6ce494a149a217500d647672877a7a80e40fc6ba5ecd6
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.json	828f44ab8c0dcf43c369f796d6628f7dfac2f02fe910f273ccc4ad37deade42b
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.log	da0dd97d1232641d0fa3c0ac0c179ed78b7518a40a201fcd5590d24a99a0c46c
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.nsys-rep	7ec549f7345edf8a0dca3d1f139e5f68d5517cb91d1d9b4e12705562bf5a542e
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard.sqlite	00c50b8246940d17e404a828a99cdfde645e6ab97e3ec6399dd25e85aee33377
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_cuda_api_sum.csv	dc62e1dc3e53a7c196ad70ebfe4c85eb1aa2c4e0a457784b96e025d4797e8030
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_cuda_gpu_kern_sum.csv	e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855
cuda-cublas-dense-1024-api-device-workspace-jacobi-standard	cuda-cublas-dense-1024-api-device-workspace-jacobi-standard_nvtx_pushpop_sum.csv	f550fdfadc398e9a2c933e72306acb9a01ab1a77a579e8a0a1062d8562fe219c
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard-stats.log	2527a2cb54d8b15b68c9271990916d6f6f0d9f239b898efdd4ecf3d41315aa35
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.json	cf05c882fc7d33bd140f8fd9339ab16b9d7ef5aa76e4e957341a99f0903eee1b
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.log	0209876c599e6068d38b42b0ea730526b22a84c0369229fef3316ebe4f784f27
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.nsys-rep	b8f20d293b9aa1327e312b09cb969277b979c08b4139c721bf2464d54cb0317f
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard.sqlite	69efbbfe95d44cf59fdb0251ae8d932fe4f068e6099b37e6aa3cce4484e34ab2
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_cuda_api_sum.csv	6f08b3d8becb17bd6c42d160e69f0ab94dc0d3573020ca2e3f303c071cfe07d6
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_cuda_gpu_kern_sum.csv	ad0cef3934d7aa2621f290b963cc1934f8239dda189454ffee834dac6d5e8f76
cuda-cublas-dense-1024-api-device-prepare-jacobi-standard	cuda-cublas-dense-1024-api-device-prepare-jacobi-standard_nvtx_pushpop_sum.csv	df2ee411e79a9495655c9643e65b3cdb80c0f3c1f5226bbeefd62a1f86f2117e
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard-stats.log	171485c190eee0372a94df0fd8ffd22f5c4e6b859ec56210f2efc1f166806985
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.json	0d822aecd29cb023bc0e7f002faf7549febb79cf782ed881904e9c71bfaa06bb
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.log	a3485b4777285cc98222a31bb53569b27fff49a3dd237f4594c4bc7bd40774ff
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.nsys-rep	9c96ac8fe2bf3a5c37de64a07e9308a331dc09be2050beb3a64c0520088b0b3b
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard.sqlite	61ab856778645ac8707c1e66fd4f6a48f01c7043363bae0a0908761aa9f5c6f7
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	a8b269499fb16a3098f62bb6ea621f02548114597e55aaeb063734d2bbf393e9
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	1ebaee2f1fccc43ee1ce4c4f3563268fd2fe20abf7de699c57f31a041c52fe84
cuda-native-dense-1024-api-device-solve-jacobi-standard	cuda-native-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	eff0b7d19f70670f2e9161f205fab4c52051b18185731d391908d89333fe0702
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard-stats.log	972e7aa7294623d33615b83ed99f910970905ee63ce32fafea46ea2164ab9baa
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.json	7cb23b96c4c1f1b09dbfb66015e497b4c8f9341d88c3ea273568d4c31795c2c5
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.log	fe860cb8a0557933de6db3e7fc9879e8dc656e9b17c020bff0ad40f4a8f7e504
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.nsys-rep	cced1981dd2b5447a81437e47e5c0a7f09d950b7776ab71b0e3f99a151f1691c
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard.sqlite	8c82fc90afb52b2ed32daf60dc4fadb3ad1a78c4dc1521164d4217678ebdd6f2
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	c7de15649e47245a171b7235a13608f94d20b64b2bb21e8c4eeb8d6e79785646
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	bd85223965b7b69e12311bd486bf1286a31a02a38219ba242d5ba342fa2b87b2
torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	8a153e8743f92c82809458399dcb1e0ed29a595eb64d85e72baf59caa76f709d
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard-stats.log	73dad88faf66da19e51bd5edff6b442a3909bc99e4e090e006338e641fa7feeb
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.json	34c1cd29e4e7a2b489382b7e843991628c7f5f9da74db3b6a3da95e182039866
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.log	986e9d81aa25e9907fd1847aebd97ed92ee67a74e1b75637ecccbe2d2ccfb79f
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.nsys-rep	62211563dea0f28c661680435c033f39787aff86dd634c60198fe2e961c71c96
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard.sqlite	3acf0b5c56da5a91f80a3fa8dedfbe7036b2201c08480c82bd5c4e26629a9210
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	e2cbd15b74fcc1e9f9faf9e3ff3f987c78d0044e03abf2c3ef8c9416cb8c619a
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	1f45588ed00eb5e9b9c522629cbd3528c155fd0807879d533a16592bf6888604
cuda-native-dense-1024-session-device-solve-jacobi-standard	cuda-native-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	b71600c7cfcb87eb00879dce0f779174b444ff68694a24433a50d7067c86c246
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard-stats.log	c982ab592cf8a1dda9a0359dbe9403bd81bbe82e2aa39c02328829dee96b1cd5
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.json	77f7bb196cd11219478485b28049978e2bb0a1e000bd705028e724495ec28e4b
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.log	4c8489998c4a3cc2a2b78b45a3b000de8b1fe76b0ec53f2500f69ecece173120
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.nsys-rep	ed65502a17777261e9ecaecccb9afe35a99bdc9f6b9d94edf9a64868df72058e
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.sqlite	2bc68327b6e141b3f80da9311d198fc5f8f06f5a08473142c33d624f9ad4684c
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	b9e7b8b8bff1e81f352de3b37a132018666398a318cc9a26b0f253a817e3e386
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	198a43ad570caaa658a107dd0ceffb1ae1cd1990db4b21a79d3cc7efa88f0934
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	d6901925f7428a1c143438fc081d3c88bb4552ede40bdcfd8ad687ddc06948d6
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard-stats.log	17cb2aa0f8001047082280d84230c8b6f3c3d599f988f2abbd084ab0dcf88a95
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.json	7e9a5fb9e8a089a291860518af68cdc491eafebf33c6735bf8138892204cc1ff
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.log	b4ac1aac605d473f336ac8fb1eb9a03273eb4b4ae34e5345454c6c123c273b2b
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.nsys-rep	76c690110e6ba1fe7b40573e2344509ac2d3f4e1f70b05fce40380e032ad602b
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard.sqlite	261914f3e78a12e19938ad6149788dfeabfa202efd018cf4eeb7a8a8a5e73f1b
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	977d86364623b01ac0e7da263c09f905d7005f415d927709a207b48b30f1c4bb
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	89184d98107510d66800295591d864fa4396826457f09632cecd9bd9e6bb8589
torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard	torch-cuda-eager-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	252232c20746b5accacf293d9bc73df89e284c8659ea1e432053393e45cf2506
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard-stats.log	ca63f06a237407de6b479594e6f37526351384b0eee9dddc4b57aa5ad27c90d9
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.json	4582a86dd48df5913461b1667db80176367188f427ec8fb95231e0d6dc11d3cc
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.log	7b6050cfffa3d131c6ce77230cc681c9a9dd9b7717faa9f8979df70253134e9e
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	419303509fc71ce45d9c9b2efef33e14b561052a619f658e8f968a18b8d21d76
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.sqlite	5fddf5116938999b57b863bbb6313d3bf3808072344ebaccbcfbb40fc2f0fdc2
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	a9fca7ba4119a88d5dc98461bd9bdc1d133ac0700547063979457e8a62a9e663
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	8d18f652cf477cf81149f33cd25d9f2ae320afc2650126f62a759673b584cec0
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	d7569ebc847193be2254a7d7f5c16fd847a18b0d64e6f2481a59a278b730520a
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard-stats.log	71528dec7b0667cf80b5d6cdf8ca08ce57bc1085b13451f262063cb9a33c4451
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.json	4b0d59cf86bf41b000e28c4f3578289be32889f4dd896c7e424001e65ef1fb69
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.log	ea95168401233c8ec19c5e428946ffe89f69552d205970509025aa8ac2d5363f
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	7fc5f33151d9245bba16c5f3f5c62549760489b4c861fa56a62a71447edf0d53
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.sqlite	3dda5b4c2fdd05d987d162980075c9a18547975f4629b76721119fa20cfe3172
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	3d4d25dcf11d1e680adf11380c9230476f47b4ab277c8ebcda3879248d095353
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	9d0bb611bf7c5f9ec6a7d9f2ad91516c6074463c42985bd4133c11ba47c3a987
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	dd361d753e6c313c1069ca53eb6e2f62d8b49c19f7598527a98feb512529030b
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard-stats.log	fb99010fc5199d5ad7c3114b99342c4be2fd2ddf783a94c16422ee70382df004
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.json	a548a8a1103d344a2825bc407f60f935dd658117f3164e2907bb1f893837db00
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.log	1ed2824beef87ff20991f0099aadc6fa0229c3390661237cb3a121c1722ce4a0
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.nsys-rep	969d3a7ce60746577b2f009a5d108aaeac6c82fc1151193031b41ee19e1ef08f
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.sqlite	3a85976c266b5f2b11612da668def7b2357d4ca8a735867c3a7fccb442c893ac
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	430660193e173b22f0b841779930f95ad2f95fcdcc5b7ce1557eca62acea769f
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	9ed58e1eef1e5664f13fff213ad8b819c4f3de254606171d4cc2bdbe5054690f
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	c3040979196dae82881acb124c2c36388d2feb71f3252370007e8133db425366
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit-stats.log	7ec1d2c97f40fadd5b6e4117d3eb7d8da11270e78433a23a0fecf540573e3f36
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.json	49b221369fd7947b4451e2df4f85c5a007fa53d80aff72c64233b21552cdb25b
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.log	1720ffdd2cec7c49ea084d06036081a82d418ac789b8b3c1269f77b644742b46
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.nsys-rep	43279645cc59d3085040bfde873610dcaa0acc3ccec6da1ac71875ad88054b18
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.sqlite	4dde39fb4459334500af43eb355d6013a21b362a30b0b2dbef3cae6f09116342
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_api_sum.csv	af064c0000457356e68e2b967aba44e7600efe35b44ff97844722f49d8a76bcd
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_gpu_kern_sum.csv	79cef5bd91e23b03b5f412669cd9b0042ee6ab467b1700a2244190f8a035a8be
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_nvtx_pushpop_sum.csv	a8b5a59f3fb362df90e724d7a9941d07a7e7aae58a148b97ffded5f71c57efef

To reproduce a capture, use the public source-install and project-local CUDA instructions in Installation and CUDA, build the recorded source, then run the matching public scripts/profile_solve.py inside Nsight Systems with capture-range=cudaProfilerApi and cuda,cublas,nvtx tracing. The ignored orchestrator is not needed. Nsight installation, WSL permissions and GPU availability are external prerequisites. Exact frozen command argv and all original summary CSV rows are retained in the analysis JSON.

profiles-launch CUDA follow-up diagnostics

These instrumented captures are separate from the uninstrumented baseline and are not framework scores. All ten-repetition captures retain passing before/after independent true-residual gates with identical known iterations, matvecs and direction restarts. Resource probes gate a solve outside the resource-only probe; their gated work does not describe operations performed inside Workspace/prepare construction and destruction.

All CUDA API counts below are totals over each complete ten-call cudaProfiler capture. The driver synchronizes before every NVTX range as well as within the measured phase. Full-capture stream-sync counts therefore include those inter-call fences and must not be interpreted as solver-only counts. Launch API counts are distinct from the recorded GPU kernel instances.

NVTX total sums only the ten exact outer iteration ranges, which do not overlap each other. NVTX phase time, Python completed-wall medians, CUDA API durations and GPU kernel durations have different boundaries and can overlap. They must not be added into latency fractions or inferred by subtracting different scope medians. API calls can wait for prior GPU work. Captures have three warmups, seed 20261017, one host thread, affinity [0] and ten repetitions. Clocks, thermal state, Windows display activity and physical host isolation are uncontrolled. Single-process instrumentation does not establish a speedup.

Workspace probes include construction and destruction. Prepare probes include the operator snapshot, optional Jacobi setup and destruction. Session workspace construction is outside its solve capture; output and requested Jacobi setup remain per-call. No historical pinned-host-scalar allocation is claimed as a current cost. Empty workspace kernel exports record zero instances only when the unchanged Nsight stats log explicitly confirms no CUDA kernel data. An absent export remains unreported; its absence is not converted into a fabricated zero.

All 6 full-capture diagnostics

Capture	Instrumented wall median ms	10 outer NVTX total ms	Malloc/free	Launch APIs	GPU kernels	Kernel CSV	Memcpy/async	Stream/device sync	Gated I/M/R	Before TrueR	Before threshold
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	2.933460	34.083211	70/80	520	520	recorded	0/110	130/30	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	0.945598	12.621146	20/20	520	520	recorded	0/110	130/0	4/6/0	1.42392e-07	1.04058e-05
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	1.111044	12.685238	0/0	510	510	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	1.032236	12.163074	0/0	450	450	recorded	0/100	120/0	4/6/0	1.42392e-07	1.04058e-05
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	16.512261	161.154874	0/0	10810	10810	recorded	0/1320	1300/0	118/122/2	1.21108e-06	1.44144e-06
cuda-native-dense-1024-prepared-device-solve-none-explicit	21.339047	208.854444	0/0	8340	8340	recorded	0/1410	1440/0	69/139/0	8.92006e-06	1.04058e-05

All CUDA API calls remain in the table below, including profiler and completion fences outside the NVTX ranges. Raw original CSV rows and API durations remain in the analysis JSON; durations are not combined into additive attribution.

Complete observed API count inventory

Capture	API	Calls per ten-call capture
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaDeviceSynchronize	30
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventCreateWithFlags	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaEventDestroy	160
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaFree	80
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaLaunchKernel	520
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMalloc	70
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	40
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaFree	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaLaunchKernel	520
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMalloc	20
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemcpyAsync	110
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaMemsetAsync	10
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cudaStreamSynchronize	130
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	510
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaLaunchKernel	450
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	100
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	120
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuProfilerStart	1
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaLaunchKernel	10810
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaMemcpyAsync	1320
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamGetDevice_v12080	30
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cudaStreamSynchronize	1300
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuProfilerStart	1
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaLaunchKernel	8340
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaMemcpyAsync	1410
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamGetDevice_v12080	30
cuda-native-dense-1024-prepared-device-solve-none-explicit	cudaStreamSynchronize	1440

Compiled source and retained evidence

Actual compiled source: 14c37277de16182fdc7f4cfcf9193e0b352791d1; tree: 26fcec1817f59f8dd680980fffcdf8fe0907e628.

Profile driver and campaign runner are different files and are independently hashed below. Raw JSON, CSVs, logs, Nsight reports and frozen source/binaries remain local ignored evidence; these paths are not public artifact downloads.

Source identities

Role	Path	SHA256
profile_driver	build/restart-optimization-20261017/launch-stages/cuda-release/tools/profile_solve.py	20b805f3bd95a9f0c1a5d23105c17888e4bd285b754575c1540661df87b7e0c4
runner	build/restart-optimization-20261017/profile_current.py	eda0e6d1779698c6cc85d419c269cfb58150ef2c4cd8aad317ce066ea7f11d5e
stage_identity	build/restart-optimization-20261017/launch-stages/cuda-release/identity.json	daf34907780c5b6fd366347ff81161904289228e9cf5960e85c0c669887e6466
manifest	build/restart-optimization-20261017/profiles-launch/manifest.json	8b2d0e4586843cf69e0992e03774a035095707df648812c686939cb38790bec6
plan	build/restart-optimization-20261017/profiles-launch/plan.json	773336c15492930d13b476caebccbcd9dc1d861276e29cfe484505755754f1cb
renderer	build/restart-optimization-20261017/render_followup_profiles.py	f74a687c7499ab92d6919368405045c780eb93974c217c57978e82e3f58bc488
render_helpers	build/restart-optimization-20261017/render_baseline.py	af180cd5ecb8e916450024b0d5fabd0f6b7088f9f41e0e720b95ea5552154ab0
baseline_inputs	build/restart-optimization-20261017/baseline-analysis.json	8d27f15818e58182a7c113e38024049849fc2ebc44048de4362cbd08e8385a9d

Every retained diagnostic artifact

Capture	Filename	SHA256
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard-stats.log	0237dfcedc8c0c33740f2770e874b8d5b9cd357a3ebc0e99072b50cac284fe5d
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.json	29bdb85f55c1d94d136b0628425fc17ad03a9af66b5f5167bc2caad773fce74c
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.log	06552195c8404d10f288953764f4b851b38d3da8f8e2b73f46ade227f3852fab
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.nsys-rep	a4b4d4905cf9191e8f039762383320b46c8558e35f0dbabb055816ad394ec60e
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard.sqlite	00a4cbc6b76fc86a3181412aea2616843e2771ed28a199d937fba32d58b3a6fb
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_api_sum.csv	f26a9d0a96ac7050e2e3d561baf25a9073e363ca6c8c70bcb24d1a054435f4d9
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	80174311ff3f1a8b5a53bab517d9718894d2dd5d359abf7d117c2c2d92b696be
cuda-cublas-dense-1024-api-device-solve-jacobi-standard	cuda-cublas-dense-1024-api-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	bfccb2a4517748b517e51628807b597a10aa451500d3a9133d0f8d2eaf0203d0
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard-stats.log	b75a68e741804e40eb9e1d76cbac97cc4c255a305960431c2306e0114c69d457
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.json	97c32aeeb805022514dd46f00c5e6bb11c74477f7187f8a22f0da68be83f5ba1
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.log	fedc514536dc2315823f53114457a5aa985f99db36399d729388ba757c478bce
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.nsys-rep	9b4ea57b733667226f20182132074f86cd3cdc7ebd0843fb27a5978ec9a4fb73
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard.sqlite	96a07e8660bc336ef36f6cf8269a006b2a7357723bb30b53af43c5cc4987f7fb
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_api_sum.csv	99a7c4c67b716fdd84dfda320889a28bda1b661d2453afb9e9358dcf9ca78477
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	c3e9509300efe2838f7e38f5684ab77799579c168183d63d0636b864827953c5
cuda-cublas-dense-1024-session-device-solve-jacobi-standard	cuda-cublas-dense-1024-session-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	22bcd966f70c5b0d2aa37ec6585be5632027e70aca17db196034e9e9ed2e3bfd
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard-stats.log	b89a1554610a0969f5f3181269612dae51ba7ba9d4d25c598365216bea82d9fd
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.json	38b7cff6f16494056ed02a8f631c1a76be46fe13f50b3af73a89ff42de05ef8b
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.log	746c98b3d46688071105d209be76c5921b389fe534341df2c531318f8685cce3
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	706b6a7336a6b69d46b3c9bc442343cd458c98d277ea07f552af6e832ba6b4c5
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard.sqlite	0e0ca6d39951561fe1853a78eeae61c8df35264469f8d0282ebdc670fbeb3772
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	c589369f2f4d8137d34fb208973508efa71e1af13465a425b06b9d49b7d956eb
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	2e5e27abe917795748d7eb4892ff01cc999ad24f7a6562498dd45eb9320868fa
cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard	cuda-cublas-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	0296b6cf815ebb917cac6fa3c6f2d4db22b6027e0ac51c65716e57495024cea6
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard-stats.log	063db06752ce18ab3510b08d800d05cc7f3b77c7f8351097847c73cffde6be63
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.json	ce46e1d719ea00e4b0ddebc70ff3fd992ac484c3144e3cd85915236ceed0e73d
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.log	27045d881c1bc54e8f76577819ee3201600f4d8d0b43b409ad4e4b31447c17f2
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.nsys-rep	6b6af14a9f6313285500ac0fa3853eabdf425f7535183a14363d9a78926fc23a
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard.sqlite	66a59140fe160df62a0086d5205efa0ad730bd8621b8d523836ec1589740e3eb
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	2e3daf6332d694f0ac483c65b74b28acc9bf48383c7c5428b65484821c3127df
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	f47308f265751cd907c9b0ef8e64474c47bbfcee6ef0d641431dac80717932d8
cuda-native-dense-1024-prepared-device-solve-jacobi-standard	cuda-native-dense-1024-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	00704490963ffadda5e312bb7b0d196dc7d41a599b6a4a10be0519fc653b64c7
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard-stats.log	b7d375c1db280f845f90292b7a5e0a3dbb7ef58ade50e8e95cb0a7ab3bf5f24f
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.json	5c39377c11f8724d42ecd7bce2512de8457b484b3658b98204f2d73ea085cbbf
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.log	66238007c17356bd533073b8a3c55e08a09cdff951ae70b7c103204e6208beef
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.nsys-rep	e0e7dbf5ec4c49ae42baa827fd78f427b204c75c2d2b12c5a2237bb3770662c0
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard.sqlite	a98bd6828772abc2625d6321ea501d891bd7a1dea17e6673f53fa41bccf5d84d
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_api_sum.csv	d47dd33af3e46c6589948e73a271308756b42fb619438c46d4867d9ab810ce40
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_cuda_gpu_kern_sum.csv	a3e04747ed12459a1b1565fb865a25dd1cc153327ebbe767b0523664ebcff809
cuda-native-poisson-32-prepared-device-solve-jacobi-standard	cuda-native-poisson-32-prepared-device-solve-jacobi-standard_nvtx_pushpop_sum.csv	c770ece2aa428549c9ddd0cd0350e45aef5908e5ac35bd7b32bbf391075624a0
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit-stats.log	0a15081ad9838faa6ed99a41342a92857395c60cf02a11d9c35ff9657aeb4433
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.json	6a1f9c728f61f99657550d68ae7d239014177022a30a9ba07fec9621a2c63931
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.log	8e4ceab33dca51d54953cf1f0e130603076ff47a315008a09baa89e3d8fdc915
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.nsys-rep	0c1c8e74c9c424eed53d0568d313b3533f6fd2ca5ea0f97d99f1c5008855a2b8
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit.sqlite	e5a73ff12a00a266ed25a360367e00db261aaf210298813f9ec2dead805c7406
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_api_sum.csv	549aca636d878b643af0b54629ef2a40df6e26d52e7ae87255c1d14c01085396
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_cuda_gpu_kern_sum.csv	cdf675184ef359756a9dd3cbc2f4e824e817dc4975a4e65f80a92bd11aec7c59
cuda-native-dense-1024-prepared-device-solve-none-explicit	cuda-native-dense-1024-prepared-device-solve-none-explicit_nvtx_pushpop_sum.csv	5b65754db66f6226801bfb481bb1d0517567280feea6b6dad7be666c21aab04a

To reproduce a capture, use the public source-install and project-local CUDA instructions in Installation and CUDA, build the recorded source, then run the matching public scripts/profile_solve.py inside Nsight Systems with capture-range=cudaProfilerApi and cuda,cublas,nvtx tracing. The ignored orchestrator is not needed. Nsight installation, WSL permissions and GPU availability are external prerequisites. Exact frozen command argv and all original summary CSV rows are retained in the analysis JSON.