[win/build] Enable /prefetch-inputs by default for lld-link [chromium/src : main]
0 views
Skip to first unread message
Ryosuke Miyake (Gerrit)
Apr 17, 2026, 2:33:36 AM (14 days ago) Apr 17
to Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Junji Watanabe and Takuto Ikuta
Ryosuke Miyake voted Commit-Queue+1![]( "Open in Gerrit")
| Commit-Queue | +1 |
Related details
Attention is currently required from:
- Junji Watanabe
- Takuto Ikuta
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Takuto Ikuta (Gerrit)
Apr 17, 2026, 2:40:58 AM (14 days ago) Apr 17
to Ryosuke Miyake, Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Junji Watanabe and Ryosuke Miyake
Takuto Ikuta added 1 comment![]( "Open in Gerrit")
Commit Message
Line 13, Patchset 1 (Latest):
Following the initial safe rollout, this flag is now enabled by default.Takuto Ikuta . unresolved
We haven't enable the flag, so no rollout happened actually?
Related details
Attention is currently required from:
- Junji Watanabe
- Ryosuke Miyake
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Ryosuke Miyake (Gerrit)
Apr 17, 2026, 2:54:54 AM (14 days ago) Apr 17
to Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Junji Watanabe and Takuto Ikuta
Ryosuke Miyake added 1 comment![]( "Open in Gerrit")
Commit Message
Line 13, Patchset 1:
Following the initial safe rollout, this flag is now enabled by default.Takuto Ikuta . resolved
We haven't enable the flag, so no rollout happened actually?
Ryosuke Miyake
I meant opt-in rollout, but I agree the wording is ambiguous. I have updated the CL description to clarify this. Thanks!
Related details
Attention is currently required from:
- Junji Watanabe
- Takuto Ikuta
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Junji Watanabe (Gerrit)
Apr 17, 2026, 3:43:56 AM (14 days ago) Apr 17
to Ryosuke Miyake, Takuto Ikuta, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Ryosuke Miyake and Takuto Ikuta
Junji Watanabe voted and added 1 comment![]( "Open in Gerrit")
Votes added by Junji Watanabe
| Code-Review | +1 |
1 comment
Patchset-level comments
File-level comment, Patchset 2 (Latest):
Junji Watanabe . unresolved
Can you try triggering LED jobs for the following benchmark builders to confirm it doesn't cause an obvious regression?
- CQ build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-siso
- CI build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-ci-siso
- Developer build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-developer
Related details
Attention is currently required from:
- Ryosuke Miyake
- Takuto Ikuta
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Takuto Ikuta (Gerrit)
Apr 17, 2026, 5:57:21 AM (14 days ago) Apr 17
to Ryosuke Miyake, Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Ryosuke Miyake
Takuto Ikuta voted Code-Review+1![]( "Open in Gerrit")
| Code-Review | +1 |
Related details
Attention is currently required from:
- Ryosuke Miyake
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Ryosuke Miyake (Gerrit)
Apr 19, 2026, 9:41:26 PM (11 days ago) Apr 19
to Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Junji Watanabe and Ryosuke Miyake
Ryosuke Miyake added 1 comment![]( "Open in Gerrit")
Patchset-level comments
Junji Watanabe . unresolved
Can you try triggering LED jobs for the following benchmark builders to confirm it doesn't cause an obvious regression?
- CQ build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-siso
- CI build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-ci-siso
- Developer build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-developer
Ryosuke Miyake
I triggered LED jobs for these builders on Apr 17:
- win-build-perf-siso: https://ci.chromium.org/b/8684287086096037137
- win-build-perf-ci-siso: https://ci.chromium.org/b/8684287064994165361 (Failed)
- win-build-perf-developer: https://ci.chromium.org/b/8684287046213061345
The failure in `win-build-perf-ci-siso` is likely an unrelated flake or a race condition exposed by timing?
I have tried running another job to confirm. https://ci.chromium.org/b/8684042132551856785
Related details
Attention is currently required from:
- Junji Watanabe
- Ryosuke Miyake
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Junji Watanabe (Gerrit)
Apr 19, 2026, 9:54:35 PM (11 days ago) Apr 19
to Ryosuke Miyake, Takuto Ikuta, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Ryosuke Miyake
Junji Watanabe added 1 comment![]( "Open in Gerrit")
Patchset-level comments
Junji Watanabe . unresolved
Ryosuke MiyakeCan you try triggering LED jobs for the following benchmark builders to confirm it doesn't cause an obvious regression?
- CQ build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-siso
- CI build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-ci-siso
- Developer build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-developer
I triggered LED jobs for these builders on Apr 17:
- win-build-perf-siso: https://ci.chromium.org/b/8684287086096037137
- win-build-perf-ci-siso: https://ci.chromium.org/b/8684287064994165361 (Failed)
- win-build-perf-developer: https://ci.chromium.org/b/8684287046213061345
The failure in `win-build-perf-ci-siso` is likely an unrelated flake or a race condition exposed by timing?
I have tried running another job to confirm. https://ci.chromium.org/b/8684042132551856785
Junji Watanabe
The failure in win-build-perf-ci-siso is likely an unrelated flake or a race condition exposed by timing?
Attention is currently required from:
- Ryosuke Miyake
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Ryosuke Miyake (Gerrit)
Apr 20, 2026, 9:09:34 PM (10 days ago) Apr 20
to Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Junji Watanabe, Ryosuke Miyake and Takuto Ikuta
Ryosuke Miyake added 1 comment![]( "Open in Gerrit")
Patchset-level comments
Junji Watanabe . unresolved
Ryosuke MiyakeCan you try triggering LED jobs for the following benchmark builders to confirm it doesn't cause an obvious regression?
- CQ build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-siso
- CI build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-ci-siso
- Developer build benchmark: https://ci.chromium.org/ui/p/chromium/builders/build/win-build-perf-developer
Junji WatanabeI triggered LED jobs for these builders on Apr 17:
- win-build-perf-siso: https://ci.chromium.org/b/8684287086096037137
- win-build-perf-ci-siso: https://ci.chromium.org/b/8684287064994165361 (Failed)
- win-build-perf-developer: https://ci.chromium.org/b/8684287046213061345
The failure in `win-build-perf-ci-siso` is likely an unrelated flake or a race condition exposed by timing?
I have tried running another job to confirm. https://ci.chromium.org/b/8684042132551856785
The failure in win-build-perf-ci-siso is likely an unrelated flake or a race condition exposed by timing?
Yeah, I assume so, too. Thank you for retrying it.
Ryosuke Miyake
The previous failure in `win-build-perf-ci-siso` did not occur on retry, so it was likely due to unrelated flakiness.
Regarding the build timing, there was no obvious regression observed. However, I noticed that the baseline was not strictly excluding the patch. To confirm the performance impact precisely, I have triggered LED jobs comparing the exact baseline (`351e6b42e2ab7`) and the patched state (`916d0abae39749d823a31`):
- *Baseline (`351e6b42e2ab7`):**
- `win-build-perf-siso`: https://ci.chromium.org/b/8683953674282332353
- `win-build-perf-ci-siso`: https://ci.chromium.org/b/8683953671922308001
- `win-build-perf-developer`: https://ci.chromium.org/b/8683953669900646529
- *With Patch (`916d0abae39749d823a31`):**
- `win-build-perf-siso`: https://ci.chromium.org/b/8683953668305090801
- `win-build-perf-ci-siso`: https://ci.chromium.org/b/8683953666692882529
- `win-build-perf-developer`: https://ci.chromium.org/b/8683953664776888385
Related details
Attention is currently required from:
- Junji Watanabe
- Ryosuke Miyake
- Takuto Ikuta
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Takuto Ikuta (Gerrit)
Apr 20, 2026, 11:02:33 PM (10 days ago) Apr 20
to Ryosuke Miyake, Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Junji Watanabe and Ryosuke Miyake
Takuto Ikuta voted and added 1 comment![]( "Open in Gerrit")
Votes added by Takuto Ikuta
| Code-Review | +1 |
Takuto Ikuta
Thanks, could you update description with results from led jobs too when it finished?
Related details
Attention is currently required from:
- Junji Watanabe
- Ryosuke Miyake
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Junji Watanabe (Gerrit)
Apr 21, 2026, 1:13:30 AM (10 days ago) Apr 21
to Ryosuke Miyake, Takuto Ikuta, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Ryosuke Miyake
Junji Watanabe added 1 comment![]( "Open in Gerrit")
Patchset-level comments
Junji Watanabe
With Patch (916d0abae39749d823a31):
How did you trigger the jobs? Did you use `led edit-gerrit-url`?
e.g.
```
led get-builder "chromium/build/win-build-perf-siso" | led edit-gerrit-url {gerrit url} | led launch
```
Related details
Attention is currently required from:
- Ryosuke Miyake
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
No-Unresolved-Comments
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Junji Watanabe (Gerrit)
Apr 21, 2026, 1:19:09 AM (10 days ago) Apr 21
to Ryosuke Miyake, Takuto Ikuta, Chromium LUCI CQ, chromium...@chromium.org
Patchset-level comments
Junji Watanabe
Baseline (351e6b42e2ab7):
With Patch (916d0abae39749d823a31):
Looking at the builds, those hashes don't look used at all.
Ryosuke Miyake (Gerrit)
Apr 21, 2026, 2:59:47 AM (10 days ago) Apr 21
to Takuto Ikuta, Junji Watanabe, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Junji Watanabe
Ryosuke Miyake added 1 comment![]( "Open in Gerrit")
Patchset-level comments
Ryosuke Miyake
You are right, I misunderstood how `led` handles local commit hashes and the revision properties.
To ensure accurate results, I compared the performance of the patched builds against the historical baseline data from the Chrome Build Benchmarks dashboard.
Here are the clean build times for the patchset: ```
- *win-build-perf-siso (Target=all):**
- Without remote cache: 13m 58.81s
- With remote cache: 12m 23.46s
win-build-perf-ci-siso (Target=all):
- Without remote cache: 2h 9m 12.80s
- With remote cache: 36m 31.88s
win-build-perf-developer (Target=chrome):
- Without remote cache: 6m 1.48s
- With remote cache: 5m 07.21s
- ```
- Comparing these with the dashboard trends, there is no obvious regression.
- I have updated the CL description with these results.
Related details
Attention is currently required from:
- Junji Watanabe
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Junji Watanabe (Gerrit)
Apr 21, 2026, 3:04:21 AM (10 days ago) Apr 21
to Ryosuke Miyake, Takuto Ikuta, Chromium LUCI CQ, chromium...@chromium.org
Attention needed from Ryosuke Miyake
Junji Watanabe voted![]( "Open in Gerrit")
Attention is currently required from:
- Ryosuke Miyake
Submit Requirements:
Code-Coverage
Code-Owners
Code-Review
Review-Enforcement
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Chromium LUCI CQ (Gerrit)
Apr 21, 2026, 3:08:02 AM (10 days ago) Apr 21
to Ryosuke Miyake, Junji Watanabe, Takuto Ikuta, chromium...@chromium.org
Chromium LUCI CQ submitted the change![]( "Open in Gerrit")
Change information
Commit message:
[win/build] Enable /prefetch-inputs by default for lld-link
This CL removes the `use_lld_prefetch_inputs` GN argument and enables
the `/prefetch-inputs` flag unconditionally for `lld-link.exe` on
Windows.
Following the initial introduction of the flag, it is verified that the
memory impact is acceptable. It improves linking performance for large
targets like chrome.dll during cold cache builds, while maintaining
parity during hot cache builds. The associated peak memory increase is
approximately 3%.
**Benchmark Results (on gWindows Cloudtop):**
Cold Cache:
- lld-baseline: 163.674 s ± 3.499 s
- lld-prefetch: 43.960 s ± 0.893 s
- **Summary**: lld-prefetch ran 3.72x faster than baseline.
Hot Cache:
- lld-baseline: 39.506 s ± 2.149 s
- lld-prefetch: 37.949 s ± 0.493 s
- **Summary**: lld-prefetch ran 1.04x faster (essentially on par).
**Memory Results (Peak Working Set):**
Target: chrome (is_component_build=false)
- Baseline (without flag): 20,567 MB (~20.57 GB)
- With /prefetch-inputs: 21,166 MB (~21.17 GB)
- Delta: +599 MB (approx. 3.1% increase)
**Verification Results (on LUCI Builders with Patch):**
No obvious regressions were observed in clean build times compared to
historical trends.
win-build-perf-siso (Target: all):
- Without remote cache: 13m 58.81s
- With remote cache: 12m 23.46s
win-build-perf-ci-siso (Target: all):
- Without remote cache: 2h 9m 12.80s
- With remote cache: 36m 31.88s
win-build-perf-developer (Target: chrome):
- Without remote cache: 6m 1.48s
- With remote cache: 5m 07.21s
Bug: 501965477
Change-Id: I0215b4d5121688c8c58d522e27e04d80bc8d3fca
Commit-Queue: Junji Watanabe <jw...@google.com>
Reviewed-by: Takuto Ikuta <tik...@chromium.org>
Reviewed-by: Junji Watanabe <jw...@google.com>
Cr-Commit-Position: refs/heads/main@{#1618009}
Files:
- M build/config/win/BUILD.gn
Change size: S
Delta: 1 file changed, 2 insertions(+), 8 deletions(-)
Branch: refs/heads/main
Submit Requirements:
Code-Review: +1 by Takuto Ikuta, +1 by Junji Watanabe
| Inspect html for hidden footers to help with email filtering. To unsubscribe visit settings. |
Reply all
Reply to author
Forward
0 new messages