All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=906284

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=6791dab8a3adb866a8b3d16fd04f79e4f563edbcae7de5fb38a95c83ec709bb2


Bot(s) for this bug's original alert(s):

Win 7 Perf

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14a944f7e40000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/14a944f7e40000

Constexpr-eliminate GetCategoryGroupEnabled calls for popular categories by alexilin@chromium.org
https://chromium.googlesource.com/chromium/src/+/331266377d03b51dc3e2cf2c26b896e21f8f0391
vector_math_fmac: 652.5 → 607.6 (-44.87)

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

Alex, any chance that this regression is because of your code?
(Seems a bit unlikely to me.)

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/16a89924140000

😿 Pinpoint job stopped with an error.
https://pinpoint-dot-chromeperf.appspot.com/job/16a89924140000

All of the runs failed. The most common error (10/20 runs) was:
HTTPException: HTTP status code 400: {"error": {"message": "CIPD package path is required. Use \".\" to install to run dir."}}

Yeah, it looks unlikely to me as well, given that media_perftests doesn't produce any trace events.

I started another bisect, let's look at what it will find.

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/171c458fe40000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/171c458fe40000

Constexpr-eliminate GetCategoryGroupEnabled calls for popular categories by alexilin@chromium.org
https://chromium.googlesource.com/chromium/src/+/331266377d03b51dc3e2cf2c26b896e21f8f0391
vector_math_fmac: 652 → 607.6 (-44.39)

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/17c6bd38140000

😿 Pinpoint job stopped with an error.
https://pinpoint-dot-chromeperf.appspot.com/job/17c6bd38140000

All of the runs failed. The most common error (20/20 runs) was:
SwarmingTaskError: The swarming task failed with state "BOT_DIED".

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/10d5f34be40000

😿 Pinpoint job stopped with an error.
https://pinpoint-dot-chromeperf.appspot.com/job/10d5f34be40000

All of the runs failed. The most common error (20/20 runs) was:
SwarmingTaskError: The swarming task failed with state "BOT_DIED".

I checked that the code generated for the function being tested (media::vector_math::FMAC_SSE) is identical before and after the patch % addresses.

Have no idea where the performance difference comes from. It may be related to less effective cache utilization. Understanding the reason of the previous improvement from this graph may help. I filed the Issue 907871.

I took a look at this, but I'm as stumped as you are.
My best guess is that the tested code ends up crossing a page boundary with your code, and the extra page faults is causing it to execute slower.

Either way, I'm going to assign it to you since your Cl does seem to be involved somehow, if perhaps only incidentally. Maybe closing it as WontFix would be better though...