Flush denormals to zero on arm and arm64.

Flush denormals to zero on arm and arm64.

We also slightly refactor the code to make it a little easier to read.

The impact of denormals on arm appear to be much less pronounced than
on x86, but flushing denormals will make arm and x86 results more similar.

This was tested manually by creating a test to check the response of a
lowpass filter to a very small impulse. All values after a certain
point should be zero when denormals are flushed to zero. If denormals
are not flushed, the non-zero output would last much longer.

Verified on Linux, Windows, OSX, and Android that denormals are flushed.

The test is not included; it seems not that important to ensure
denormals are flushed.

BUG=395254
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=178620

[Raymond Toy, 2014-07-17 23:08:48]

PTAL

[Ken Russell (switch to Gerrit), 2014-07-18 18:46:57]

Needs more work. I also suggest you file a tracking bug in case this has to be
reverted and re-landed.

https://codereview.chromium.org/402803003/diff/20001/Source/platform/audio/De...
File Source/platform/audio/DenormalDisabler.h (right):

https://codereview.chromium.org/402803003/diff/20001/Source/platform/audio/De...
Source/platform/audio/DenormalDisabler.h:39: // Windows compiled using MSVC with
SSE2 (_M_IX86_FP == 2) can flush denormals.
A little searching indicates that _M_IX86_FP isn't defined at all on 64-bit
Windows builds, but they implicitly have SSE2 support.

https://codereview.chromium.org/402803003/diff/20001/Source/platform/audio/De...
Source/platform/audio/DenormalDisabler.h:99: #elif OS(WIN) && COMPILER(MSVC)
The conditions in this #elif don't match those in the #ifdef above.

[Raymond Toy, 2014-07-18 20:08:34]

https://codereview.chromium.org/402803003/diff/20001/Source/platform/audio/De...
File Source/platform/audio/DenormalDisabler.h (right):

https://codereview.chromium.org/402803003/diff/20001/Source/platform/audio/De...
Source/platform/audio/DenormalDisabler.h:39: // Windows compiled using MSVC with
SSE2 (_M_IX86_FP == 2) can flush denormals.
On 2014/07/18 18:46:56, Ken Russell wrote:
> A little searching indicates that _M_IX86_FP isn't defined at all on 64-bit
> Windows builds, but they implicitly have SSE2 support.

Should just probably get rid of this check, now that windows always uses sse2.

[Raymond Toy, 2014-07-21 20:56:14]

PTAL.

[Ken Russell (switch to Gerrit), 2014-07-21 22:11:17]

lgtm

[Raymond Toy, 2014-07-22 03:19:51]

The CQ bit was checked by rtoy@chromium.org

[commit-bot: I haz the power, 2014-07-22 03:20:29]

CQ is trying da patch. Follow status at
 https://chromium-status.appspot.com/cq/rtoy@chromium.org/402803003/40001

[commit-bot: I haz the power, 2014-07-22 04:23:09]

Change committed as 178620