MIPS: Fix 'MIPS: [turbofan] Optimize certain chains of Branch into a Switch.'

MIPS: Fix 'MIPS: [turbofan] Optimize certain chains of Branch into a Switch.'

Space and time constants fixed. Delay slot optimisations added.

BUG=

Committed: https://crrev.com/833e3e0b8ca7078dbfbcd3e52fd48eee83049039
Cr-Commit-Position: refs/heads/master@{#26729}

[balazs.kilvady, 2015-02-18 15:21:17]

balazs.kilvady@imgtec.com changed reviewers:
+ akos.palfi@mattakis.com, dusan.milosavljevic@imgtec.com,
gergely.kis@imgtec.com, paul.lind@imgtec.com

[balazs.kilvady, 2015-02-18 15:21:18]

PTAL: I am not sure if padding nop should be added to the code block "saved" by
BlockTrampolinePoolFor() on mip64el in function
CodeGenerator::AssembleArchTableSwitch(). Or if the code block moved then we
need the same padding?

[paul.l..., 2015-02-18 16:52:50]

LGTM, with one suggested small change.

The nop() should be included Block'd region. The padding won't change if/when
this code block is moved by gc, since the start of the new assembly buffer will
be aligned to 8 on 64-bit platforms. The offset within the buffer will stay the
same.

https://codereview.chromium.org/940453003/diff/1/src/compiler/mips64/code-gen...
File src/compiler/mips64/code-generator-mips64.cc (right):

https://codereview.chromium.org/940453003/diff/1/src/compiler/mips64/code-gen...
src/compiler/mips64/code-generator-mips64.cc:1068: }
I agree that you should include the nop() in the Block..For scope, but there is
no penalty for blocking for one additional instruction, so I suggest moving that
out of the if(), and always use (case_count * 2 + 7).

I think that will enhance readability, and make the adding the padding nop()
more clear.

[balazs.kilvady, 2015-02-18 17:41:35]

New patchsets have been uploaded after l-g-t-m from paul.lind@imgtec.com

[balazs.kilvady, 2015-02-18 17:48:09]

https://codereview.chromium.org/940453003/diff/1/src/compiler/mips64/code-gen...
File src/compiler/mips64/code-generator-mips64.cc (right):

https://codereview.chromium.org/940453003/diff/1/src/compiler/mips64/code-gen...
src/compiler/mips64/code-generator-mips64.cc:1068: }
On 2015/02/18 16:52:50, paul.l... wrote:
> I agree that you should include the nop() in the Block..For scope, but there
is
> no penalty for blocking for one additional instruction, so I suggest moving
that
> out of the if(), and always use (case_count * 2 + 7).
> 
> I think that will enhance readability, and make the adding the padding nop()
> more clear.

Done.

[balazs.kilvady, 2015-02-18 18:07:50]

The CQ bit was checked by balazs.kilvady@imgtec.com

[commit-bot: I haz the power, 2015-02-18 18:08:50]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/940453003/20001

[commit-bot: I haz the power, 2015-02-18 18:41:54]

Committed patchset #2 (id:20001)

[commit-bot: I haz the power, 2015-02-18 18:42:04]

Patchset 2 (id:??) landed as
https://crrev.com/833e3e0b8ca7078dbfbcd3e52fd48eee83049039
Cr-Commit-Position: refs/heads/master@{#26729}