[turbofan] Avoid unnecessary write barriers and improve code generation.

[turbofan] Avoid unnecessary write barriers and improve code generation.

Avoid write barriers when storing values in the root set, and use
cheaper write barriers for storing maps or tagged pointers. Also
improve the generated code for write barriers, utilizing the out
of line code mechanism that is available to TurboFan backends,
which moves the unlikely case out of the hot path.

R=jarin@chromium.org, mstarzinger@chromium.org

Committed: https://crrev.com/1e2770123bd5c1cf7631bf11be846f0fbd884c08
Cr-Commit-Position: refs/heads/master@{#31914}

[Benedikt Meurer, 2015-11-10 09:11:22]

Hey Michi, Jaro,

Here's an initial patch to make us less stupid wrt. write barriers in TurboFan.
There's still more to do to be competitive with Crankshaft, tho.
The changes are mostly c&p, but I expect that arm64, mips and mips64 can be
improved quite a bit (also arm has some potential), so I didn't want to
prematurely abstractify it (yet).

Michi: Please look at heap.
Jaro: Please check rest.

Thanks,
Benedikt

[Michael Starzinger, 2015-11-10 09:23:17]

LGTM on "heap", didn't look at the rest at all.

https://codereview.chromium.org/1414183006/diff/1/src/heap/heap.cc
File src/heap/heap.cc (right):

https://codereview.chromium.org/1414183006/diff/1/src/heap/heap.cc#newcode4384
src/heap/heap.cc:4384: IMMORTAL_IMMOVABLE_ROOT_LIST(IMMORTAL_IMMOVABLE_ROOT);
nit: Trailing semicolon is not needed.

[Benedikt Meurer, 2015-11-10 09:30:10]

https://codereview.chromium.org/1414183006/diff/1/src/heap/heap.cc
File src/heap/heap.cc (right):

https://codereview.chromium.org/1414183006/diff/1/src/heap/heap.cc#newcode4384
src/heap/heap.cc:4384: IMMORTAL_IMMOVABLE_ROOT_LIST(IMMORTAL_IMMOVABLE_ROOT);
On 2015/11/10 09:23:17, Michael Starzinger wrote:
> nit: Trailing semicolon is not needed.

Done.

[Jarin, 2015-11-10 10:07:55]

lgtm

[Benedikt Meurer, 2015-11-10 10:30:48]

The CQ bit was checked by bmeurer@chromium.org

[Benedikt Meurer, 2015-11-10 10:30:48]

The patchset sent to the CQ was uploaded after l-g-t-m from
mstarzinger@chromium.org, jarin@chromium.org
Link to the patchset: https://codereview.chromium.org/1414183006/#ps80001
(title: "Fix typo.")

[commit-bot: I haz the power, 2015-11-10 10:31:10]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1414183006/80001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1414183006/80001

[commit-bot: I haz the power, 2015-11-10 11:05:25]

Committed patchset #5 (id:80001)

[commit-bot: I haz the power, 2015-11-10 11:06:08]

Patchset 5 (id:??) landed as
https://crrev.com/1e2770123bd5c1cf7631bf11be846f0fbd884c08
Cr-Commit-Position: refs/heads/master@{#31914}

[ulan, 2016-02-19 15:02:31]

ulan@chromium.org changed reviewers:
+ ulan@chromium.org

[ulan, 2016-02-19 15:02:32]

https://codereview.chromium.org/1414183006/diff/80001/src/compiler/x64/code-g...
File src/compiler/x64/code-generator-x64.cc (right):

https://codereview.chromium.org/1414183006/diff/80001/src/compiler/x64/code-g...
src/compiler/x64/code-generator-x64.cc:213: __ CheckPageFlag(value_, scratch0_,
Eliding kPointersToHereAreInterestingMask check for maps is not correct.

There are two issues:
1. Incremental marker dynamically sets and clears this flag for map space pages,
so we don't know the value of the flag at code generation time.

2. Even assuming that there is no incremental marking, the intention of the code
was probably to not emit the write barrier because we know that the flag is 0
for maps? But this code actually forces the write barrier because it removes the
check. So we end up with old -> map slots in old -> new remembered set.

Context: https://bugs.chromium.org/p/chromium/issues/detail?id=587004#c13