| commit | 84419b700809fcb6ff2655a79bdde477c11f90ac | [log] [tgz] |
|---|---|---|
| author | Kai Ninomiya <kainino@chromium.org> | Wed Oct 16 23:28:30 2024 +0000 |
| committer | Dawn LUCI CQ <dawn-scoped@luci-project-accounts.iam.gserviceaccount.com> | Wed Oct 16 23:28:30 2024 +0000 |
| tree | cdc5e6c9e94d7042027328b54047d6c6b8684259 | |
| parent | 5f3c2cabb2430396e97af778b09a4c6ac794d0cc [diff] |
Reduce maxSubcasesInFlight Right now we have a lot of OOMs caused by issues like creating 500 pipelines simultaneously on 32-bit and mobile systems. This should avoid that. The default of 500 comes from here (I will also change it): https://github.com/gpuweb/cts/blob/648e0564cc6bf26d17323db52848485654d64bcf/src/common/framework/test_config.ts#L43 The limit takes effect here: https://github.com/gpuweb/cts/blob/648e0564cc6bf26d17323db52848485654d64bcf/src/common/internal/test_group.ts#L667 Bug: 373485785 Change-Id: Idfc7a8513fcd5cab65911226b00ec6b7f227015e Reviewed-on: https://dawn-review.googlesource.com/c/dawn/+/211196 Commit-Queue: Kai Ninomiya <kainino@chromium.org> Reviewed-by: David Neto <dneto@google.com> Reviewed-by: Brian Sheedy <bsheedy@google.com>
diff --git a/webgpu-cts/test_runner.js b/webgpu-cts/test_runner.js index d02ae3f..0d8ad58 100644 --- a/webgpu-cts/test_runner.js +++ b/webgpu-cts/test_runner.js
@@ -166,6 +166,10 @@ globalTestConfig.testHeartbeatCallback = sendHeartbeat; globalTestConfig.noRaceWithRejectOnTimeout = true; +// Avoid doing too many things at once (e.g. creating 500 pipelines +// simultaneously on a 32-bit system easily runs out of memory). +globalTestConfig.maxSubcasesInFlight = 100; + // FXC is very slow to compile unrolled const-eval loops, where the metal shader // compiler (Intel GPU) is very slow to compile rolled loops. Intel drivers for // linux may also suffer the same performance issues, so unroll const-eval loops