tree 10ff03b0df80f99ad14f7f234560b89207d514a6
parent f80a78602e37d3236202465e0cdcad8800a79ff1
author Mike Klein <mtklein@google.com> 1591113765 -0500
committer Skia Commit-Bot <skia-commit-bot@chromium.org> 1591220045 +0000

apply SkOpts_skx approach to SkOpts_hsw

Very slightly different build flags, switching from -march=haswell to
-mavx2 -mfma -mf16c, but there are no diffs.

Left some TODOs for the next ones, but _hsw being so common I figured
I'd do this one stand alone and make sure it shakes down before doing
the rest.

clang-cl (but not clang) barfs when we've got a lambda using an AVX
intrinsic inside a templated static helper function. Luckily they're
all non-type template parameters, so we can just pass them as normal
arguments, and it'll optimize the same as the templated code anyway.

This was weird, but since we're covering almost all the SkFoo_opts.h
headers with HSW here, I don't anticipate this being a problem in the
future.  (And I'm sure I'll never look back on this statement as naive.)

Change-Id: I2f84db356cafa5e158bcc3724fb1c3f58aca7f1e
Reviewed-on: https://skia-review.googlesource.com/c/skia/+/293599
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@google.com>