[WASM SIMD] Replace primitive shuffles with general Shuffle.

[WASM SIMD] Replace primitive shuffles with general Shuffle.
- Removes primitive shuffle opcodes.
- Adds Shuffle opcode for S32x4, S16x8, S8x16.
- Adds code to ARM instruction selector to pick best opcodes for some
  common shuffle patterns.

LOG=N
BUG=v8:6020
Review-Url: https://codereview.chromium.org/2847663005
Cr-Commit-Position: refs/heads/master@{#45104}
Committed: https://chromium.googlesource.com/v8/v8/+/0cd0fa3b98a06126681f4ce091555a0aa2e35859

[bbudge, 2017-04-28 21:05:25]

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

[commit-bot: I haz the power, 2017-04-28 21:05:35]

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, 2017-04-28 21:11:25]

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

[commit-bot: I haz the power, 2017-04-28 21:11:26]

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, 2017-04-28 21:17:57]

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

[commit-bot: I haz the power, 2017-04-28 21:18:05]

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, 2017-04-28 21:39:11]

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

[commit-bot: I haz the power, 2017-04-28 21:39:12]

Dry run: Try jobs failed on following builders:
  v8_linux64_gyp_rel_ng on master.tryserver.v8 (JOB_FAILED,
http://build.chromium.org/p/tryserver.v8/builders/v8_linux64_gyp_rel_ng/build...)
  v8_linux64_gyp_rel_ng_triggered on master.tryserver.v8 (JOB_FAILED,
http://build.chromium.org/p/tryserver.v8/builders/v8_linux64_gyp_rel_ng_trigg...)

[bbudge, 2017-04-28 21:49:28]

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

[commit-bot: I haz the power, 2017-04-28 21:49:31]

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, 2017-04-28 22:22:18]

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

[commit-bot: I haz the power, 2017-04-28 22:22:18]

Dry run: This issue passed the CQ dry run.

[bbudge, 2017-04-30 04:18:21]

bbudge@chromium.org changed reviewers:
+ bmeurer@chromium.org, gdeepti@chromium.org, georgia.kouveli@arm.com

[martyn.capewell, 2017-05-02 14:56:55]

martyn.capewell@arm.com changed reviewers:
+ martyn.capewell@arm.com

[martyn.capewell, 2017-05-02 14:56:56]

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
File src/compiler/arm/instruction-selector-arm.cc (right):

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
src/compiler/arm/instruction-selector-arm.cc:2580: {{0, 4, 1, 5},
kArmS32x4ZipLeft},
Odd suggestion: I wonder if the shuffle operation is better expressed as a
string, then use existing (fast) string operations for comparison and
manipulation. It makes the code layout easier, too.

For src0 versus src1, you could use upper and lower case - the masks become an
application of tolower or toupper.

Later, it might allow some of the more complex matching to be applied through
regular expressions...

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
src/compiler/arm/instruction-selector-arm.cc:2660: bool src0 = false;
This would be clearer named src0_is_unary, etc.

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
src/compiler/arm/instruction-selector-arm.cc:2678:
selector->GetVirtualRegister(node->InputAt(1))) {
This case could be moved earlier, avoiding the shuffle checking loop.

[bbudge, 2017-05-02 22:59:27]

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
File src/compiler/arm/instruction-selector-arm.cc (right):

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
src/compiler/arm/instruction-selector-arm.cc:2580: {{0, 4, 1, 5},
kArmS32x4ZipLeft},
On 2017/05/02 14:56:56, martyn.capewell wrote:
> Odd suggestion: I wonder if the shuffle operation is better expressed as a
> string, then use existing (fast) string operations for comparison and
> manipulation. It makes the code layout easier, too.
> 
> For src0 versus src1, you could use upper and lower case - the masks become an
> application of tolower or toupper.
> 
> Later, it might allow some of the more complex matching to be applied through
> regular expressions...

I think these are the only patterns we'll be matching on ARM. Everything else
will use explicit moves or vtbl, so I think it's better to keep these as raw
bytes.

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
src/compiler/arm/instruction-selector-arm.cc:2660: bool src0 = false;
On 2017/05/02 14:56:56, martyn.capewell wrote:
> This would be clearer named src0_is_unary, etc.

How about src0_is_used? Done.

https://codereview.chromium.org/2847663005/diff/60001/src/compiler/arm/instru...
src/compiler/arm/instruction-selector-arm.cc:2678:
selector->GetVirtualRegister(node->InputAt(1))) {
On 2017/05/02 14:56:56, martyn.capewell wrote:
> This case could be moved earlier, avoiding the shuffle checking loop.

Done.

[martyn.capewell, 2017-05-03 10:49:23]

lgtm

[bbudge, 2017-05-04 10:11:39]

Ready for your review Benedikt.

[Benedikt Meurer, 2017-05-04 10:37:13]

LGTM on the compiler bits.

[bbudge, 2017-05-04 12:32:45]

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

[commit-bot: I haz the power, 2017-05-04 12:32:55]

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, 2017-05-04 13:03:18]

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

[commit-bot: I haz the power, 2017-05-04 13:03:19]

Dry run: This issue passed the CQ dry run.

[bbudge, 2017-05-04 16:49:07]

The CQ bit was checked by bbudge@chromium.org

[bbudge, 2017-05-04 16:49:07]

The patchset sent to the CQ was uploaded after l-g-t-m from
bmeurer@chromium.org, martyn.capewell@arm.com
Link to the patchset: https://codereview.chromium.org/2847663005/#ps100001
(title: "Rebase.")

[commit-bot: I haz the power, 2017-05-04 16:49:14]

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, 2017-05-04 16:50:52]

CQ is committing da patch.

Bot data: {"patchset_id": 100001, "attempt_start_ts": 1493916547037710,
"parent_rev": "ec619cbd89ca8e16b41ec573ee9cb216609c47d6", "commit_rev":
"0cd0fa3b98a06126681f4ce091555a0aa2e35859"}

[commit-bot: I haz the power, 2017-05-04 16:50:59]

Description was changed from

==========
[WASM SIMD] Replace primitive shuffles with general Shuffle.
- Removes primitive shuffle opcodes.
- Adds Shuffle opcode for S32x4, S16x8, S8x16.
- Adds code to ARM instruction selector to pick best opcodes for some
  common shuffle patterns.

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

to

==========
[WASM SIMD] Replace primitive shuffles with general Shuffle.
- Removes primitive shuffle opcodes.
- Adds Shuffle opcode for S32x4, S16x8, S8x16.
- Adds code to ARM instruction selector to pick best opcodes for some
  common shuffle patterns.

LOG=N
BUG=v8:6020

Review-Url: https://codereview.chromium.org/2847663005
Cr-Commit-Position: refs/heads/master@{#45104}
Committed:
https://chromium.googlesource.com/v8/v8/+/0cd0fa3b98a06126681f4ce091555a0aa2e...
==========

[commit-bot: I haz the power, 2017-05-04 16:51:00]

Committed patchset #6 (id:100001) as
https://chromium.googlesource.com/v8/v8/+/0cd0fa3b98a06126681f4ce091555a0aa2e...