finish up constants
authorMike Klein <mtklein@chromium.org>
Mon, 1 May 2017 20:01:38 +0000 (16:01 -0400)
committerMike Klein <mtklein@chromium.org>
Mon, 1 May 2017 20:59:45 +0000 (20:59 +0000)
commit5664e65eb1680a14eeaa6ca79ddf9e734518c822
treeee66d340b7acce5ef99a1296108981862dbf962d
parent879a08ac146acb2518363cc9d2d8aa6dce04528d
finish up constants

For whatever reason, if I swap the condition in the if_then_else tests
from < to >= and swap the then/else values, I can use constants in
hsl_to_rgb.  Still don't understand why, but I'll take it.  I suspect it
has something to do with SSE, IEEE, and NaN, but I don't care enough to
speculate any more concretely.

This does that, removes C() and _f, updates some comments, and adds a
guard in build_stages.py to yell if it sees trouble like LCPI40_4...

This reminds me to try -ffast-math soon.  I think that was mostly held
back by constants.

Change-Id: I3f8a37a4d4642f77422ce3261b750061e9e604a3
Reviewed-on: https://skia-review.googlesource.com/14942
Reviewed-by: Herb Derby <herb@google.com>
src/jumper/SkJumper.h
src/jumper/SkJumper_generated.S
src/jumper/SkJumper_generated_win.S
src/jumper/SkJumper_misc.h
src/jumper/SkJumper_stages.cpp
src/jumper/build_stages.py