Add AVX2 detection

Add AVX2 detection.

Update unittest to test avx and avx2, including for win64 with clangcl.
Add AVX2 to uma for histograms.

BUG=chromium:537916, chromium:168866
Committed: https://crrev.com/0ce41ae0824b15da94b7ec5f3293f8889c56362a
Cr-Commit-Position: refs/heads/master@{#351949}

[fbarchard, 2015-10-01 03:45:00]

fbarchard@chromium.org changed reviewers:
+ mark@chromium.org, rkaplow@chromium.org

[rkaplow, 2015-10-01 15:04:14]

lgtm

histogram lgtm

[Mark Mentovai, 2015-10-01 15:50:07]

Frank, I didn’t see that you requested review on this. Is it ready?

[fbarchard, 2015-10-01 17:35:30]

fbarchard@chromium.org changed reviewers:
+ fbarchard@chromium.org

[fbarchard, 2015-10-01 17:35:30]

Ready for review, Mark.
Fixed typo in xgetbv

[Mark Mentovai, 2015-10-01 17:54:15]

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc
File base/cpu_unittest.cc (right):

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:29: #if  defined(__x86_64__) || defined(__i386__)
Can you use the build/build_config.h macros for this?

This should be defined(COMPILER_GCC). (And there’s a space between if and
defined that you should get rid of.)

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:61: __asm__ __volatile__("vzeroupper\n" : : : "xmm0");
This actually clobbers ymmh0 (but not the low xmm0 portion). Do clang/gcc not
allow you to express that?

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:66: __asm__ __volatile__("vpunpcklbw %%ymm0, %%ymm0,
%%ymm0\n" : : : "xmm0");
Same here, this clobbers all of ymm0.

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:70: #elif defined(_M_IX86)
defined(COPMILER_MSVC) && defined(ARCH_CPU_X86)

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:73: __asm emms;
Are there intrinsics that you can use so that this works for 64-bit Windows too?
That’s the easiest way to fix the TODO above.

[fbarchard, 2015-10-01 18:42:06]

caveat This version disables the test for win64.

Tested with drcpusim:

\bin\DynamoRIO-Windows-5.0.16708-0x5ee2a05\bin32\drrun -t drcpusim -continue
-no_fool_cpuid -cpu Pentium -- out\Debug\base_unittests.exe
--gtest_filter==CPU.RunExtendedInstructions
[ RUN      ] CPU.RunExtendedInstructions
<Invalid Pentium instruction "emms" @ base_unittests.exe+0x1a80da.  Continuing.>
<Invalid Pentium instruction "xorps" @ base_unittests.exe+0x1a80dc. 
Continuing.>
<Invalid Pentium instruction "psrldq" @ base_unittests.exe+0x1a80df. 
Continuing.>
<Invalid Pentium instruction "addsubpd" @ base_unittests.exe+0x1a80fb. 
Continuing.>
<Invalid Pentium instruction "psignb" @ base_unittests.exe+0x1a8116. 
Continuing.>
<Invalid Pentium instruction "pmuldq" @ base_unittests.exe+0x1a8132. 
Continuing.>
<Invalid Pentium instruction "crc32" @ base_unittests.exe+0x1a814e. 
Continuing.>
<Invalid Pentium instruction "vzeroupper" @ base_unittests.exe+0x1a816a. 
Continuing.>
[       OK ] CPU.RunExtendedInstructions (14 ms)

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc
File base/cpu_unittest.cc (right):

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:29: #if  defined(__x86_64__) || defined(__i386__)
On 2015/10/01 17:54:14, Mark Mentovai wrote:
> Can you use the build/build_config.h macros for this?
> 
> This should be defined(COMPILER_GCC). (And there’s a space between if and
> defined that you should get rid of.)

Done.
Caveat The intent here is clangcl support for intel cpus.
COMPILER_GCC is not set by clangcl.
Its not the compiler that matters, (icc and icl should also work), its the
instruction set syntax.

To test with clangcl 32 bit:
set GYP_DEFINES=clang=1 component=shared_library fastbuild=1 target_arch=ia32
test_isolation_mode=prepare
python build\gyp_chromium -fninja chrome\chrome.gyp

My goal was to add and test the AVX and AVX2 instructions.
The testing is complete, so I've changed to COMPILER_GCC but cant test with
clang or win64.
Can revisit the clang/win64 issue.

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:61: __asm__ __volatile__("vzeroupper\n" : : : "xmm0");
On 2015/10/01 17:54:15, Mark Mentovai wrote:
> This actually clobbers ymmh0 (but not the low xmm0 portion). Do clang/gcc not
> allow you to express that?

clang/gcc do not let you say clobber ymm0.

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:66: __asm__ __volatile__("vpunpcklbw %%ymm0, %%ymm0,
%%ymm0\n" : : : "xmm0");
On 2015/10/01 17:54:14, Mark Mentovai wrote:
> Same here, this clobbers all of ymm0.

Acknowledged.

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:70: #elif defined(_M_IX86)
On 2015/10/01 17:54:15, Mark Mentovai wrote:
> defined(COPMILER_MSVC) && defined(ARCH_CPU_X86)

Done.
I'd prefer not make it specific to Microsoft Visual C?
It works with other x86 compilers.

https://codereview.chromium.org/1380943002/diff/80001/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:73: __asm emms;
On 2015/10/01 17:54:15, Mark Mentovai wrote:
> Are there intrinsics that you can use so that this works for 64-bit Windows
too?
> That’s the easiest way to fix the TODO above.

Acknowledged.
Its possible, but not trivial.
purpose of this CL is avx2 detect/metrics, so I used the existing framerwork to
add the tests.

[Mark Mentovai, 2015-10-01 19:58:28]

LGTM

https://codereview.chromium.org/1380943002/diff/90005/base/cpu_unittest.cc
File base/cpu_unittest.cc (right):

https://codereview.chromium.org/1380943002/diff/90005/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:114: #endif  // defined(__x86_64__) || defined(__i386__)
Fix this to match.

[fbarchard, 2015-10-01 21:27:52]

avx2 tested with intel emulator

c:\intelsde\sde -ast -hsw -- out\Debug\base_unittests.exe --gtest_filter=*CPU*

Note: Google Test filter = CPU.RunExtendedInstructions
[==========] Running 1 test from 1 test case.
[----------] Global test environment set-up.
[----------] 1 test from CPU
[ RUN      ] CPU.RunExtendedInstructions
[       OK ] CPU.RunExtendedInstructions (61 ms)
[----------] 1 test from CPU (234 ms total)

[----------] Global test environment tear-down
[==========] 1 test from 1 test case ran. (416 ms total)
[  PASSED  ] 1 test.
[1/1] CPU.RunExtendedInstructions (61 ms)
SUCCESS: all tests passed.

https://codereview.chromium.org/1380943002/diff/90005/base/cpu_unittest.cc
File base/cpu_unittest.cc (right):

https://codereview.chromium.org/1380943002/diff/90005/base/cpu_unittest.cc#ne...
base/cpu_unittest.cc:114: #endif  // defined(__x86_64__) || defined(__i386__)
On 2015/10/01 19:58:28, Mark Mentovai wrote:
> Fix this to match.

Done.

[Mark Mentovai, 2015-10-01 21:32:06]

LGTM still

[fbarchard, 2015-10-01 22:59:22]

The CQ bit was checked by fbarchard@chromium.org

[fbarchard, 2015-10-01 22:59:22]

The patchset sent to the CQ was uploaded after l-g-t-m from rkaplow@chromium.org
Link to the patchset: https://codereview.chromium.org/1380943002/#ps130001
(title: "endif comment for gcc compiler")

[commit-bot: I haz the power, 2015-10-01 22:59:51]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1380943002/130001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1380943002/130001

[commit-bot: I haz the power, 2015-10-01 23:05:16]

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

[commit-bot: I haz the power, 2015-10-01 23:05:16]

Try jobs failed on following builders:
  win_chromium_x64_rel_ng on tryserver.chromium.win (JOB_FAILED,
http://build.chromium.org/p/tryserver.chromium.win/builders/win_chromium_x64_...)

[fbarchard, 2015-10-01 23:40:14]

The CQ bit was checked by fbarchard@chromium.org

[fbarchard, 2015-10-01 23:40:14]

The patchset sent to the CQ was uploaded after l-g-t-m from mark@chromium.org,
rkaplow@chromium.org
Link to the patchset: https://codereview.chromium.org/1380943002/#ps140001
(title: "no change... git pull to merge head")

[commit-bot: I haz the power, 2015-10-01 23:40:35]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1380943002/140001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1380943002/140001

[commit-bot: I haz the power, 2015-10-02 03:23:28]

Committed patchset #9 (id:140001)

[commit-bot: I haz the power, 2015-10-02 03:24:38]

Patchset 9 (id:??) landed as
https://crrev.com/0ce41ae0824b15da94b7ec5f3293f8889c56362a
Cr-Commit-Position: refs/heads/master@{#351949}