Add AVX flag for Windows GN builds

Add AVX flag for Windows GN builds

R=ckocagil@chromium.org

Committed: 8dc6ea93de62df4fb4d8bcc6156c3ad4df158bb3

[Johann, 2015-01-22 18:33:37]

Only question I have is should it be AVX or AVX2? It seems like we are currently
using AVX in most places, but the instruction set we want is definitely AVX2

[Tom Finegan, 2015-01-22 18:43:49]

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

[Tom Finegan, 2015-01-22 18:43:49]

Not sure if I'm right or wrong on this... someone more familiar with the current
windows toolchain requirements can hopefully add more information.

https://codereview.chromium.org/817293008/diff/1/BUILD.gn
File BUILD.gn (right):

https://codereview.chromium.org/817293008/diff/1/BUILD.gn#newcode135
BUILD.gn:135: cflags = [ "/arch:AVX" ]
msdn vs2013 docs[1] say both are supported. I think vs2013 is the minimum
required compiler for windows chromium development, so AVX2 should be OK (and is
probably what you want since we're using -mavx2 everywhere else).

[1] https://msdn.microsoft.com/en-us/library/jj620901.aspx

[Johann, 2015-01-22 18:48:31]

On 2015/01/22 18:43:49, Tom Finegan wrote:
> Not sure if I'm right or wrong on this... someone more familiar with the
current
> windows toolchain requirements can hopefully add more information.

Yeah it's a little weird. Libvpx upstream uses just AVX:
if [[ $f =~ avx.?\.c$ ]]; then
  tag_content AdditionalOptions "/arch:AVX"

the gyp files use 'EnableEnhancedInstructionSet': '3' which isn't really clear
what it maps to, AVX or AVX2, but appears to be AVX

at the same time, the code builds fine in those cases. It's not rejecting the
avx2-only instructions, so .... not sure what's up.

[ckocagil, 2015-01-22 19:20:38]

lgtm other than the AVX vs AVX2 bit, which I don't know much about.

[Johann, 2015-01-22 19:22:06]

On 2015/01/22 19:20:38, ckocagil wrote:
> lgtm other than the AVX vs AVX2 bit, which I don't know much about.

thanks. i'll stick with AVX for this roll since that seems to be the status quo
and see what happens.

[Johann, 2015-01-22 19:22:29]

Committed patchset #1 (id:1) manually as
8dc6ea93de62df4fb4d8bcc6156c3ad4df158bb3 (presubmit successful).