Require SSE2 for all 32-bit Linux builds and remove disable_sse2 option.

Require SSE2 for all 32-bit Linux builds and remove disable_sse2 option.

Previously, only 32-bit Google Chrome Linux was being built with x87 floating point math, which led to bugs caused by 80-to-{32,64}-bit rounding errors that weren't caught on the main Chromium builders. UMA data tells us there are ~0 Google Chrome Linux users without SSE2, making a strong case to require SSE2 across the board and align the Google Chrome build closer to the Chromium build.

As for disable_sse2 option, it was added in r45777 but it's unclear if anyone is using it outside of some other GYP code that flips it to 1 for ARM and MIPS architectures. Instead replace the few instances that don't support SSE2 with specific checks for the target architecture and remove the option entirely.

BUG=348761
R=jamesr@chromium.org
TBR=cpu

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=256366

[Paweł Hajdan Jr., 2014-03-10 23:49:54]

Drive-by.

https://codereview.chromium.org/187423002/diff/20001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/187423002/diff/20001/build/common.gypi#newcod...
build/common.gypi:3207: '-march=pentium4',
Do we need to force -march=pentium4 here?

[scherkus (not reviewing), 2014-03-11 00:08:52]

https://codereview.chromium.org/187423002/diff/20001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/187423002/diff/20001/build/common.gypi#newcod...
build/common.gypi:3207: '-march=pentium4',
On 2014/03/10 23:49:54, Paweł Hajdan Jr. wrote:
> Do we need to force -march=pentium4 here?

As far as I can tell it was added in r12228 with most of the original Linux gyp
support.

agl@ questioned it in
https://codereview.chromium.org/1702009/diff/1/2#newcode925 as well.

I'm OK removing it.

[Paweł Hajdan Jr., 2014-03-11 00:28:15]

https://codereview.chromium.org/187423002/diff/20001/build/common.gypi
File build/common.gypi (right):

https://codereview.chromium.org/187423002/diff/20001/build/common.gypi#newcod...
build/common.gypi:3207: '-march=pentium4',
On 2014/03/11 00:08:52, scherkus wrote:
> On 2014/03/10 23:49:54, Paweł Hajdan Jr. wrote:
> > Do we need to force -march=pentium4 here?
> 
> As far as I can tell it was added in r12228 with most of the original Linux
gyp
> support.
> 
> agl@ questioned it in
> https://codereview.chromium.org/1702009/diff/1/2#newcode925 as well.
> 
> I'm OK removing it.

Thank you, looks good to me now.

[scherkus (not reviewing), 2014-03-11 00:34:59]

jamesr: feel comfortable reviewing this? if not -- know who should?

[jamesr, 2014-03-11 00:45:57]

lgtm

https://codereview.chromium.org/187423002/diff/40001/third_party/qcms/qcms.gyp
File third_party/qcms/qcms.gyp (right):

https://codereview.chromium.org/187423002/diff/40001/third_party/qcms/qcms.gy...
third_party/qcms/qcms.gyp:43: # MSVC x64 doesn't support the MMX intrinsics
present in the SSE1 code,
when would we ever use sse1?  we use sse2 on all platforms with this patch (sse2
has been enabled on windows since the vs2013 switchover)

[scherkus (not reviewing), 2014-03-11 18:31:01]

PTAL -- I tweaked qcms.gyp

https://codereview.chromium.org/187423002/diff/40001/third_party/qcms/qcms.gyp
File third_party/qcms/qcms.gyp (right):

https://codereview.chromium.org/187423002/diff/40001/third_party/qcms/qcms.gy...
third_party/qcms/qcms.gyp:43: # MSVC x64 doesn't support the MMX intrinsics
present in the SSE1 code,
On 2014/03/11 00:45:57, jamesr wrote:
> when would we ever use sse1?  we use sse2 on all platforms with this patch
(sse2
> has been enabled on windows since the vs2013 switchover)

ah right -- I forgot that bit about vs2013 and assumed we still might be hitting
sse1 machines on win32 OSes

I made this change yesterday in r256183 so I'm cool simplifying this to always
include SSE2

[jamesr, 2014-03-11 18:43:48]

lgtm

[scherkus (not reviewing), 2014-03-11 20:05:00]

looks like QCMS requires SSE1 to be compiled in as we're getting linker errors

I'll revert my qcms.gyp changes but update the comment to better reflect what's
actually going on (qcms intentionally #ifdef's out the SSE1 code for win64 but
leaves it in for all other platforms).

[scherkus (not reviewing), 2014-03-11 20:13:10]

TBR=cpu for third_party/* OWNERS

[scherkus (not reviewing), 2014-03-11 20:13:14]

The CQ bit was checked by scherkus@chromium.org

[commit-bot: I haz the power, 2014-03-11 21:34:01]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/scherkus@chromium.org/187423002/110001

[commit-bot: I haz the power, 2014-03-11 23:41:58]

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

[commit-bot: I haz the power, 2014-03-11 23:41:58]

Try jobs failed on following builders: mac_chromium_compile_dbg,
mac_chromium_rel

[scherkus (not reviewing), 2014-03-12 00:32:14]

Committed patchset #5 manually as r256366 (presubmit successful).