Use __rdtsc on Windows.

Use __rdtsc on Windows.

This seems to be ~100x higher resolution than QueryPerformanceCounter.  AFAIK, all our Windows perf bots have constant_tsc, so we can be a bit more direct about using rdtsc directly: it'll always tick at the max CPU frequency.

Now, the question remains, what is the max CPU frequency to divide through by?  It looks like QueryPerformanceFrequency actually gives the CPU frequency in kHz, suspiciously exactly what we need to divide through to get elapsed milliseconds.  That was a freebie.

I did some before/after comparison on slow benchmarks.  Timings look the same.  Going to land this without review tonight to see what happens on the bots; happy to review carefully tomorrow.

R=mtklein@google.com
TBR=bungeman


BUG=skia:

Committed: https://skia.googlesource.com/skia/+/9129477

[mtklein, 2014-07-16 23:44:14]

lgtm

[commit-bot: I haz the power, 2014-07-16 23:44:35]

CQ is trying da patch. Follow status at
 https://skia-tree-status.appspot.com/cq/mtklein@chromium.org/394363003/1

[mtklein, 2014-07-16 23:59:26]

The CQ bit was unchecked by mtklein@google.com

[mtklein_C, 2014-07-16 23:59:37]

Committed patchset #1 manually as r9129477 (presubmit successful).

[bungeman-skia, 2014-07-17 14:07:14]

Other than being totally magic and stuff, lgtm. Is there any way to know if the
assumptions here hold true, so at least we know if they ever become bogus?

[mtklein, 2014-07-17 14:12:39]

On 2014/07/17 14:07:14, bungeman1 wrote:
> Other than being totally magic and stuff, lgtm. Is there any way to know if
the
> assumptions here hold true, so at least we know if they ever become bogus?

Yeah, there might be.  

As far as I've read up, constant_tsc will always be true for future Intel and
AMD chips, and it's true of every machine we'll ever run Windows performance
tests on (even, run Windows on).  The only part I'm wary about is reading the
processor frequency, but all our numbers seem to line up correctly, so I'd say,
so far so good.  Worst case we can time a long loop of this against a long loop
of a lower-precision timer and assert they're close.