[BUG]: Extremely slow `cuda_core/tests/test_helpers.py::test_patterngen_seeds` on Windows

[rwgk]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this bug and that I agree to the Code of Conduct

Type of Bug

Performance

Component

cuda.core

Describe the bug

Using the log archive from https://github.com/NVIDIA/cuda-python/actions/runs/20871705999 (main @ acc78f7 ← #1368)

Maximum measured times per platform:

• linux-64: 14.93s
• linux-aarch64: 12.07s
• win-64: 945.29s (15m 45s)

The linux-64 maximum (14.93s) is from Test_linux-64___py3.13__13.0.2__local__rtxpro6000.txt.

The linux-aarch64 maximum (12.07s) is from Test_linux-aarch64___py3.14t__13.1.0__local__a100.txt.

The win-64 maximum (945.29s) is from Test_win-64___py3.14__13.1.0__local__l4__MCDM_.txt.

smc120-0004.ipp2a2.colossus.nvidia.com:/wrk/main_logs_20871705999 $ grep ' call.*test_helpers.py::test_patterngen_seeds' *.txt
Test_linux-64___py3.10__12.9.1__local__v100.txt:2026-01-10T03:19:17.6039085Z 14.63s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.10__13.0.2__wheels__l4.txt:2026-01-10T03:08:24.5435099Z 12.83s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.10__13.1.0__wheels__l4.txt:2026-01-10T03:08:21.9057669Z 13.24s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.11__12.9.1__wheels__rtxpro6000.txt:2026-01-10T03:19:08.2634608Z 13.43s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.11__13.0.2__local__l4.txt:2026-01-10T03:12:47.9856996Z 12.80s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.11__13.1.0__local__l4.txt:2026-01-10T03:09:34.1176450Z 12.16s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.12__13.0.2__wheels__l4.txt:2026-01-10T03:08:35.6711874Z 12.89s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.12__13.1.0__wheels__l4.txt:2026-01-10T03:08:41.5389490Z 13.49s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.13__12.9.1__wheels__v100.txt:2026-01-10T03:08:35.8080212Z 13.84s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.13__13.0.2__local__H100.txt:2026-01-10T03:10:39.8398063Z 12.63s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.13__13.0.2__local__rtxpro6000.txt:2026-01-10T03:25:03.8675910Z 14.93s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.13__13.1.0__local__H100.txt:2026-01-10T03:10:38.9082711Z 13.25s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.13__13.1.0__local__rtxpro6000.txt:2026-01-10T03:24:43.8860109Z 13.10s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.14__13.0.2__local__l4.txt:2026-01-10T03:09:43.3080030Z 12.62s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.14__13.1.0__local__l4.txt:2026-01-10T03:13:13.1600312Z 13.46s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.14t__13.0.2__local__l4.txt:2026-01-10T03:09:43.4362977Z 12.80s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-64___py3.14t__13.1.0__local__l4.txt:2026-01-10T03:13:38.5521286Z 14.57s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.10__12.9.1__local__a100.txt:2026-01-10T03:09:59.0165735Z 11.79s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.10__13.0.2__wheels__a100.txt:2026-01-10T03:20:22.5541190Z 12.06s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.10__13.1.0__wheels__a100.txt:2026-01-10T03:21:11.9424489Z 12.05s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.11__12.9.1__wheels__a100.txt:2026-01-10T03:15:14.1405192Z 11.53s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.11__13.0.2__local__a100.txt:2026-01-10T03:11:00.9066452Z 11.50s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.11__13.1.0__local__a100.txt:2026-01-10T03:10:54.0936687Z 11.52s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.12__13.0.2__wheels__a100.txt:2026-01-10T03:24:02.4632216Z 11.75s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.12__13.1.0__wheels__a100.txt:2026-01-10T03:17:44.4896939Z 11.64s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.13__12.9.1__wheels__a100.txt:2026-01-10T03:08:55.4501852Z 11.70s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.13__13.0.2__local__a100.txt:2026-01-10T03:17:36.9852261Z 11.58s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.13__13.1.0__local__a100.txt:2026-01-10T03:15:28.9082362Z 11.78s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.14__13.0.2__local__a100.txt:2026-01-10T03:10:45.3140851Z 11.76s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.14__13.1.0__local__a100.txt:2026-01-10T03:14:46.4410166Z 11.69s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.14t__13.0.2__local__a100.txt:2026-01-10T03:10:52.6658849Z 12.05s call tests/test_helpers.py::test_patterngen_seeds
Test_linux-aarch64___py3.14t__13.1.0__local__a100.txt:2026-01-10T03:13:01.4328526Z 12.07s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.10__12.9.1__wheels__rtx2080__WDDM_.txt:2026-01-10T03:26:48.1889136Z 506.60s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.10__13.0.2__local__rtxpro6000__TCC_.txt:2026-01-10T03:44:19.9462869Z 702.69s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.10__13.1.0__local__rtxpro6000__TCC_.txt:2026-01-10T03:44:51.8427769Z 685.93s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.11__12.9.1__local__v100__MCDM_.txt:2026-01-10T03:30:23.4031313Z 763.77s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.11__13.0.2__wheels__rtx4090__WDDM_.txt:2026-01-10T03:27:44.5518977Z 506.31s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.11__13.1.0__wheels__rtx4090__WDDM_.txt:2026-01-10T03:26:54.8081465Z 529.50s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.12__12.9.1__wheels__l4__MCDM_.txt:2026-01-10T03:30:36.6552595Z 886.23s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.12__13.0.2__local__a100__TCC_.txt:2026-01-10T03:26:41.4786805Z 522.65s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.12__13.1.0__local__a100__TCC_.txt:2026-01-10T03:27:27.1592087Z 521.73s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.13__12.9.1__local__l4__TCC_.txt:2026-01-10T03:29:59.8609283Z 730.19s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.13__13.0.2__wheels__rtxpro6000__MCDM_.txt:2026-01-10T03:47:03.2483243Z 714.83s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.13__13.1.0__wheels__rtxpro6000__MCDM_.txt:2026-01-10T03:40:54.4933333Z 743.69s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.14__12.9.1__wheels__v100__TCC_.txt:2026-01-10T03:24:08.3370060Z 577.15s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.14__13.0.2__local__l4__MCDM_.txt:2026-01-10T03:35:16.9448794Z 930.24s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.14__13.1.0__local__l4__MCDM_.txt:2026-01-10T03:34:33.8842361Z 945.29s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.14t__12.9.1__local__l4__TCC_.txt:2026-01-10T03:30:01.8941727Z 767.32s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.14t__13.0.2__wheels__a100__MCDM_.txt:2026-01-10T03:29:51.7362946Z 623.74s call tests/test_helpers.py::test_patterngen_seeds
Test_win-64___py3.14t__13.1.0__wheels__a100__MCDM_.txt:2026-01-10T03:27:48.3543004Z 757.88s call tests/test_helpers.py::test_patterngen_seeds

How to Reproduce

Download CI logs.

Expected behavior

win-64 runtimes comparable to linux-64

Operating System

No response

nvidia-smi output

No response

[rwgk]

@Andy-Jost I think it'd be great to get the runtime for this test to below one second on all platforms. I see you're nesting two loops over 256 iterations. I haven't looked at all at details here, but generally, corner cases are 0, 1, 2 and then any number higher than that doesn't test anything extra, although I often try larger numbers interactively JIC, and then use a small number permanently.

[Andy-Jost]

The test checks for collisions between the 256 possible seed values. There aren't any corner cases unless we assume something about the implementation. That said, I was not expecting that test to be a performance hotspot, given that it only tests 256^2 cases, but it obviously is. I think skipping it on Windows is fine, and we could also reduce it to a smaller sampling of values. It's not important enough to justify taking up engineering time like it has been.

A quick fix that comes to mind is to just pick two small prime numbers, p1 and p2, and then test the combinations i x j only when i is a multiple of p1 and j is a multiple of p2.