[ARM] Improve VFP register moves.

[ARM] Improve VFP register moves.
- Adds vdup.<size> Dd/Qd, Dm[i] instruction.
- Adds vsli, vsri instructions.
- Changes VMovExtended to use these to avoid moves to core registers.

LOG=N
BUG=v8:6020
Review-Url: https://codereview.chromium.org/2868603002
Cr-Commit-Position: refs/heads/master@{#45351}
Committed: https://chromium.googlesource.com/v8/v8/+/b3acc272654c08d7c1bb875496b84667349b0288

[bbudge, 2017-05-09 17:21:59]

The CQ bit was checked by bbudge@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-05-09 17:22:09]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-05-09 17:49:59]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-05-09 17:50:00]

Dry run: This issue passed the CQ dry run.

[bbudge, 2017-05-09 22:30:30]

The CQ bit was checked by bbudge@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-05-09 22:30:37]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-05-09 22:41:52]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-05-09 22:41:53]

Dry run: Try jobs failed on following builders:
  v8_android_arm_compile_rel on master.tryserver.v8 (JOB_FAILED,
http://build.chromium.org/p/tryserver.v8/builders/v8_android_arm_compile_rel/...)

[bbudge, 2017-05-09 22:56:23]

The CQ bit was checked by bbudge@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-05-09 22:56:33]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-05-09 23:24:29]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-05-09 23:24:30]

Dry run: This issue passed the CQ dry run.

[bbudge, 2017-05-10 00:21:18]

The CQ bit was checked by bbudge@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-05-10 00:21:21]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-05-10 00:52:59]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-05-10 00:53:00]

Dry run: This issue passed the CQ dry run.

[bbudge, 2017-05-10 01:31:54]

bbudge@chromium.org changed reviewers:
+ martyn.capewell@arm.com

[martyn.capewell, 2017-05-10 12:38:06]

https://codereview.chromium.org/2868603002/diff/80001/src/arm/assembler-arm.cc
File src/arm/assembler-arm.cc (right):

https://codereview.chromium.org/2868603002/diff/80001/src/arm/assembler-arm.c...
src/arm/assembler-arm.cc:3927: DCHECK_NE(Neon64, size);
Assert index is in range.

https://codereview.chromium.org/2868603002/diff/80001/src/arm/disasm-arm.cc
File src/arm/disasm-arm.cc (right):

https://codereview.chromium.org/2868603002/diff/80001/src/arm/disasm-arm.cc#n...
src/arm/disasm-arm.cc:2338: } else if (instr->Bits(11, 8) == 0x5 &&
instr->Bit(6) == 0 &&
Consider combining these - bit 8 determines how shift is calculated and whether
to print 'l' or 'r', otherwise the code is the same.

https://codereview.chromium.org/2868603002/diff/80001/src/arm/macro-assembler...
File src/arm/macro-assembler-arm.cc (right):

https://codereview.chromium.org/2868603002/diff/80001/src/arm/macro-assembler...
src/arm/macro-assembler-arm.cc:1101: }
There's a tiny optimisation here: as vdup is slightly cheaper than vs[rl]i, if
the src and dst are in the same d register, use vdup and return.

https://codereview.chromium.org/2868603002/diff/80001/src/arm/macro-assembler...
src/arm/macro-assembler-arm.cc:1140: int dst_s_code =
kScratchDoubleReg.low().code() + (dst_code & 1);
When supported, this can use the "single element to one lane" form (vld1.32
{dx[y]}, [r]) to load directly, and similarly for store. Add a TODO?

[bbudge, 2017-05-10 17:53:50]

The CQ bit was checked by bbudge@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-05-10 17:53:56]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[bbudge, 2017-05-10 17:54:51]

bbudge@chromium.org changed reviewers:
+ jarin@chromium.org

[bbudge, 2017-05-10 17:54:52]

Thanks Martyn, for all the clever ways to move.

+Jarin for compiler

https://codereview.chromium.org/2868603002/diff/80001/src/arm/assembler-arm.cc
File src/arm/assembler-arm.cc (right):

https://codereview.chromium.org/2868603002/diff/80001/src/arm/assembler-arm.c...
src/arm/assembler-arm.cc:3927: DCHECK_NE(Neon64, size);
On 2017/05/10 12:38:06, martyn.capewell wrote:
> Assert index is in range.

Done.

https://codereview.chromium.org/2868603002/diff/80001/src/arm/disasm-arm.cc
File src/arm/disasm-arm.cc (right):

https://codereview.chromium.org/2868603002/diff/80001/src/arm/disasm-arm.cc#n...
src/arm/disasm-arm.cc:2338: } else if (instr->Bits(11, 8) == 0x5 &&
instr->Bit(6) == 0 &&
On 2017/05/10 12:38:06, martyn.capewell wrote:
> Consider combining these - bit 8 determines how shift is calculated and
whether
> to print 'l' or 'r', otherwise the code is the same.

That occurred to me too. I think it's a good idea. Done.

https://codereview.chromium.org/2868603002/diff/80001/src/arm/macro-assembler...
File src/arm/macro-assembler-arm.cc (right):

https://codereview.chromium.org/2868603002/diff/80001/src/arm/macro-assembler...
src/arm/macro-assembler-arm.cc:1101: }
On 2017/05/10 12:38:06, martyn.capewell wrote:
> There's a tiny optimisation here: as vdup is slightly cheaper than vs[rl]i, if
> the src and dst are in the same d register, use vdup and return.

Nice. Done.

https://codereview.chromium.org/2868603002/diff/80001/src/arm/macro-assembler...
src/arm/macro-assembler-arm.cc:1140: int dst_s_code =
kScratchDoubleReg.low().code() + (dst_code & 1);
On 2017/05/10 12:38:06, martyn.capewell wrote:
> When supported, this can use the "single element to one lane" form (vld1.32
> {dx[y]}, [r]) to load directly, and similarly for store. Add a TODO?

Added TODO. Thanks.

[bbudge, 2017-05-10 18:00:21]

The CQ bit was checked by bbudge@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-05-10 18:00:26]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-05-10 18:31:44]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-05-10 18:31:44]

Dry run: This issue passed the CQ dry run.

[martyn.capewell, 2017-05-12 10:32:47]

lgtm

[bbudge, 2017-05-16 19:39:43]

bbudge@chromium.org changed reviewers:
+ mtrofin@chromium.org
- jarin@chromium.org

[bbudge, 2017-05-16 19:39:44]

- Jarin
+ Mircea

for compiler OWNERS

[Mircea Trofin, 2017-05-16 19:45:27]

On 2017/05/16 19:39:44, bbudge wrote:
> - Jarin
> + Mircea
> 
> for compiler OWNERS

compiler lgtm

[bbudge, 2017-05-16 20:02:53]

The CQ bit was checked by bbudge@chromium.org to run a CQ dry run

[commit-bot: I haz the power, 2017-05-16 20:02:58]

Dry run: CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-05-16 20:33:59]

The CQ bit was unchecked by commit-bot@chromium.org

[commit-bot: I haz the power, 2017-05-16 20:34:00]

Dry run: This issue passed the CQ dry run.

[bbudge, 2017-05-16 20:35:12]

The CQ bit was checked by bbudge@chromium.org

[bbudge, 2017-05-16 20:35:13]

The patchset sent to the CQ was uploaded after l-g-t-m from
mtrofin@chromium.org, martyn.capewell@arm.com
Link to the patchset: https://codereview.chromium.org/2868603002/#ps160001
(title: "Rebase.")

[commit-bot: I haz the power, 2017-05-16 20:35:19]

CQ is trying da patch.

Follow status at:
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2017-05-16 20:37:16]

CQ is committing da patch.

Bot data: {"patchset_id": 160001, "attempt_start_ts": 1494966912923820,
"parent_rev": "211cc5856516329c0b2f0bafa3a5cfb2b4f5363d", "commit_rev":
"b3acc272654c08d7c1bb875496b84667349b0288"}

[commit-bot: I haz the power, 2017-05-16 20:37:24]

Description was changed from

==========
[ARM] Improve VFP register moves.
- Adds vdup.<size> Dd/Qd, Dm[i] instruction.
- Adds vsli, vsri instructions.
- Changes VMovExtended to use these to avoid moves to core registers.

LOG=N
BUG=v8:6020
==========

to

==========
[ARM] Improve VFP register moves.
- Adds vdup.<size> Dd/Qd, Dm[i] instruction.
- Adds vsli, vsri instructions.
- Changes VMovExtended to use these to avoid moves to core registers.

LOG=N
BUG=v8:6020

Review-Url: https://codereview.chromium.org/2868603002
Cr-Commit-Position: refs/heads/master@{#45351}
Committed:
https://chromium.googlesource.com/v8/v8/+/b3acc272654c08d7c1bb875496b84667349...
==========

[commit-bot: I haz the power, 2017-05-16 20:37:25]

Committed patchset #9 (id:160001) as
https://chromium.googlesource.com/v8/v8/+/b3acc272654c08d7c1bb875496b84667349...