Always inline SkFloatToFixed_arm.
We have "inline" assembly for speed on ARM, but the compiler when
told to optimize for space wasn't inlining it, destroying any
possible performance improvement.
BUG=skbug:2550
R=mtklein@google.com, reed@google.com, tomhudson@google.com
Author: tomhudson@chromium.org
Review URL: https://codereview.chromium.org/281143002
git-svn-id: http://skia.googlecode.com/svn/trunk@14745 2bbb7eff-a529-9590-31e7-b0007b416f81
diff --git a/include/core/SkFixed.h b/include/core/SkFixed.h
index b5987fc..a16adac 100644
--- a/include/core/SkFixed.h
+++ b/include/core/SkFixed.h
@@ -107,9 +107,10 @@
#if defined(SK_CPU_ARM)
/* This guy does not handle NaN or other obscurities, but is faster than
- than (int)(x*65536)
+ than (int)(x*65536). When built on Android with -Os, needs forcing
+ to inline or we lose the speed benefit.
*/
- inline SkFixed SkFloatToFixed_arm(float x)
+ SK_ALWAYS_INLINE SkFixed SkFloatToFixed_arm(float x)
{
int32_t y, z;
asm("movs %1, %3, lsl #1 \n"