Subzero. X86. Uses pshufb for shufflevector lowering.

Subzero. X86. Uses pshufb for shufflevector lowering.

This CL enables subzero to lower shufflevector instructions in x86 using
pshufb (only for SSE 4.1)

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4136
BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4077
R=stichnot@chromium.org

Committed: https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a=commit;h=de29f1201e8b77e58f5500ef4cfba51d547368b7

[John, 2016-04-26 13:33:04]

jpp@chromium.org changed reviewers:
+ eholk@chromium.org, kschimpf@google.com, sehr@chromium.org,
stichnot@chromium.org

[John, 2016-04-26 13:33:05]

NOTE: no tests added because there are no special case lowerings to be exercised
(and because the crosstests already exercise the new lowering.)

[Jim Stichnoth, 2016-04-26 16:42:57]

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
File src/IceTargetLoweringX86BaseImpl.h (right):

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:5690: static constexpr bool SuppressMangling
= false;
I think it doesn't matter whatsoever if mangling is used or not.  As such, I
would drop this line and use its default value for
VariableDeclaration::create().

Note that you are using the GlobalString version of setName() below, in which
case the SuppressMangling value is ignored.

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:5695: GlobalString MaskName =
This whole function makes me sad, but I guess what can you do?

In any case, can you make it so that a std::string is not instantiated in a
NODUMP build?  See the InstX86Label ctor for an example.

[John, 2016-04-26 16:52:28]

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
File src/IceTargetLoweringX86BaseImpl.h (right):

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:5690: static constexpr bool SuppressMangling
= false;
On 2016/04/26 16:42:56, stichnot wrote:
> I think it doesn't matter whatsoever if mangling is used or not.  As such, I
> would drop this line and use its default value for
> VariableDeclaration::create().
> 
> Note that you are using the GlobalString version of setName() below, in which
> case the SuppressMangling value is ignored.

Done.

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:5695: GlobalString MaskName =
On 2016/04/26 16:42:57, stichnot wrote:
> This whole function makes me sad, but I guess what can you do?
> 
Why? What's wrong with it?

> In any case, can you make it so that a std::string is not instantiated in a
> NODUMP build?  See the InstX86Label ctor for an example.

Done

[Jim Stichnoth, 2016-04-26 17:03:16]

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
File src/IceTargetLoweringX86BaseImpl.h (right):

https://codereview.chromium.org/1917863004/diff/1/src/IceTargetLoweringX86Bas...
src/IceTargetLoweringX86BaseImpl.h:5695: GlobalString MaskName =
On 2016/04/26 16:52:27, John wrote:
> On 2016/04/26 16:42:57, stichnot wrote:
> > This whole function makes me sad, but I guess what can you do?
> > 
> Why? What's wrong with it?

Ideally, you would call GlobalContext::getConstant_<foo>() for some vector type,
and all the unique-ifying/pooling would be handled for you.

Also, this just occurred to me: Have you considered whether you get
deterministic output with multithreaded translation?  I suspect it's not
deterministic.

[Jim Stichnoth, 2016-04-26 21:38:09]

lgtm

[John, 2016-04-27 02:16:09]

Description was changed from

==========
Subzero. X86. Uses pshufb for shufflevector lowering.

This CL enables subzero to lower shufflevector instructions in x86 using
pshufb (only for SSE 4.1)

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4136
BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4077
==========

to

==========
Subzero. X86. Uses pshufb for shufflevector lowering.

This CL enables subzero to lower shufflevector instructions in x86 using
pshufb (only for SSE 4.1)

BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4136
BUG= https://bugs.chromium.org/p/nativeclient/issues/detail?id=4077
R=stichnot@chromium.org

Committed:
https://gerrit.chromium.org/gerrit/gitweb?p=native_client/pnacl-subzero.git;a...
==========

[John, 2016-04-27 02:16:11]

Committed patchset #3 (id:40001) manually as
de29f1201e8b77e58f5500ef4cfba51d547368b7 (presubmit successful).