Actually it is DeltaBlue, not MandreelLatency - the dashboard had incorrectly labeled the graphs on the unified graph. Updated graph:

https://chromeperf.appspot.com/report?sid=d838945fe2c5421ed9cc1cfc5ee9c31dfbc14502e10f1a2b378114a5b5ddc1b6

I can somewhat flakily reproduce this locally. The "DeltaBlue" regresses from a stable score to bi-modal behavior. It seems unrelated to the change in the runtime profile, as I can reproduce by just toggling the --ignition-preserve-bytecode flag. Investigating.

I figured out what is causing the regression. It is the self-healing for function closures in the interpreter entry trampoline that is no longer kicking in. I am working on a fix.

The following revision refers to this bug:
  https://chromium.googlesource.com/v8/v8.git/+/4598d9139e5930543b55bca2154afc966b64c557

commit 4598d9139e5930543b55bca2154afc966b64c557
Author: mstarzinger <mstarzinger@chromium.org>
Date: Fri Aug 19 10:34:05 2016

[interpreter] Fix self-healing with preserved bytecode.

This fixes the self-healing mechanism for closures in the interpreter
entry trampoline not that bytecode can be preserved even when baseline
code is already available.

R=rmcilroy@chromium.org
TEST=cctest/test-compiler/IgnitionEntryTrampolineSelfHealing
BUG= chromium:638225 

Review-Url: https://codereview.chromium.org/2257143002
Cr-Commit-Position: refs/heads/master@{#38747}

[modify] https://crrev.com/4598d9139e5930543b55bca2154afc966b64c557/src/builtins/arm/builtins-arm.cc
[modify] https://crrev.com/4598d9139e5930543b55bca2154afc966b64c557/src/builtins/arm64/builtins-arm64.cc
[modify] https://crrev.com/4598d9139e5930543b55bca2154afc966b64c557/src/builtins/ia32/builtins-ia32.cc
[modify] https://crrev.com/4598d9139e5930543b55bca2154afc966b64c557/src/builtins/mips/builtins-mips.cc
[modify] https://crrev.com/4598d9139e5930543b55bca2154afc966b64c557/src/builtins/mips64/builtins-mips64.cc
[modify] https://crrev.com/4598d9139e5930543b55bca2154afc966b64c557/src/builtins/x64/builtins-x64.cc
[modify] https://crrev.com/4598d9139e5930543b55bca2154afc966b64c557/test/cctest/test-compiler.cc

The change in comment #4 should fix all regressions. I'll verify once the perf bots have picked it up and keep this issue open until then.

All graphs have recovered.

The following revision refers to this bug:
  https://chromium.googlesource.com/v8/v8.git/+/7b8d760457e48cfc2be6658b0e99f7b0a7793c78

commit 7b8d760457e48cfc2be6658b0e99f7b0a7793c78
Author: bjaideep <bjaideep@ca.ibm.com>
Date: Mon Aug 22 18:34:35 2016

PPC/s390: [interpreter] Fix self-healing with preserved bytecode.

Port 4598d9139e5930543b55bca2154afc966b64c557

Original commit message:

    This fixes the self-healing mechanism for closures in the interpreter
    entry trampoline not that bytecode can be preserved even when baseline
    code is already available.

R=mstarzinger@chromium.org, joransiu@ca.ibm.com, jyan@ca.ibm.com, michael_dawson@ca.ibm.com, mbrandy@us.ibm.com

BUG= chromium:638225 
LOG=N

Review-Url: https://codereview.chromium.org/2265193002
Cr-Commit-Position: refs/heads/master@{#38799}

[modify] https://crrev.com/7b8d760457e48cfc2be6658b0e99f7b0a7793c78/src/builtins/ppc/builtins-ppc.cc
[modify] https://crrev.com/7b8d760457e48cfc2be6658b0e99f7b0a7793c78/src/builtins/s390/builtins-s390.cc