Spin off some fixes to land right away.

BUG=skia:

Review URL: https://codereview.chromium.org/960023002

6 files changed, 22 insertions, 24 deletions

diff --git a/include/core/SkPostConfig.h b/include/core/SkPostConfig.h
index 41ad930989..cf111af37c 100644
--- a/include/core/SkPostConfig.h
+++ b/include/core/SkPostConfig.h
@@ -68,6 +68,15 @@
 #  endif
 #endif
 
+// As usual, there are two ways to increase alignment... the MSVC way and the everyone-else way.
+#ifndef SK_STRUCT_ALIGN
+    #ifdef _MSC_VER
+        #define SK_STRUCT_ALIGN(N) __declspec(align(N))
+    #else
+        #define SK_STRUCT_ALIGN(N) __attribute__((aligned(N)))
+    #endif
+#endif
+
 #if !defined(SK_SUPPORT_GPU)
 #  define SK_SUPPORT_GPU 1
 #endif
diff --git a/src/core/Sk4x_sse.h b/src/core/Sk4x_sse.h
index 246abd31f9..10a0a830df 100644
--- a/src/core/Sk4x_sse.h
+++ b/src/core/Sk4x_sse.h
@@ -21,10 +21,10 @@
 
     // These are all free, zero instructions.
     // MSVC insists we use _mm_castA_B(a) instead of (B)a.
-    static __m128  as_4f(__m128i v) { return _mm_castsi128_ps(v); }
-    static __m128  as_4f(__m128  v) { return                  v ; }
-    static __m128i as_4i(__m128i v) { return                  v ; }
-    static __m128i as_4i(__m128  v) { return _mm_castps_si128(v); }
+    static inline __m128  as_4f(__m128i v) { return _mm_castsi128_ps(v); }
+    static inline __m128  as_4f(__m128  v) { return                  v ; }
+    static inline __m128i as_4i(__m128i v) { return                  v ; }
+    static inline __m128i as_4i(__m128  v) { return _mm_castps_si128(v); }
 
 #elif defined(SK4X_PRIVATE)
     // It'd be slightly faster to call _mm_cmpeq_epi32() on an unintialized register and itself,
@@ -83,11 +83,11 @@ M(Sk4f) LoadAligned(const float fs[4]) { return _mm_load_ps (fs); }
 M(void) store       (float fs[4]) const { _mm_storeu_ps(fs, fVec); }
 M(void) storeAligned(float fs[4]) const { _mm_store_ps (fs, fVec); }
 
-template <> template <>
-Sk4i Sk4f::reinterpret<Sk4i>() const { return as_4i(fVec); }
+template <>
+M(Sk4i) reinterpret<Sk4i>() const { return as_4i(fVec); }
 
-template <> template <>
-Sk4i Sk4f::cast<Sk4i>() const { return _mm_cvtps_epi32(fVec); }
+template <>
+M(Sk4i) cast<Sk4i>() const { return _mm_cvtps_epi32(fVec); }
 
 // We're going to try a little experiment here and skip allTrue(), anyTrue(), and bit-manipulators
 // for Sk4f.  Code that calls them probably does so accidentally.
@@ -120,11 +120,11 @@ M(Sk4i) LoadAligned(const int32_t is[4]) { return _mm_load_si128 ((const __m128i
 M(void) store       (int32_t is[4]) const { _mm_storeu_si128((__m128i*)is, fVec); }
 M(void) storeAligned(int32_t is[4]) const { _mm_store_si128 ((__m128i*)is, fVec); }
 
-template <> template <>
-Sk4f Sk4i::reinterpret<Sk4f>() const { return as_4f(fVec); }
+template <>
+M(Sk4f) reinterpret<Sk4f>() const { return as_4f(fVec); }
 
-template <> template <>
-Sk4f Sk4i::cast<Sk4f>() const { return _mm_cvtepi32_ps(fVec); }
+template <>
+M(Sk4f) cast<Sk4f>() const { return _mm_cvtepi32_ps(fVec); }
 
 M(bool) allTrue() const { return 0xf == _mm_movemask_ps(as_4f(fVec)); }
 M(bool) anyTrue() const { return 0x0 != _mm_movemask_ps(as_4f(fVec)); }
diff --git a/src/core/SkPMFloat.h b/src/core/SkPMFloat.h
index 26bc9cce07..d2d83d0068 100644
--- a/src/core/SkPMFloat.h
+++ b/src/core/SkPMFloat.h
@@ -4,15 +4,8 @@
 #include "SkTypes.h"
 #include "SkColor.h"
 
-// As usual, there are two ways to increase alignment... the MSVC way and the everyone-else way.
-#ifdef _MSC_VER
-    #define ALIGN(N) __declspec(align(N))
-#else
-    #define ALIGN(N) __attribute__((aligned(N)))
-#endif
-
 // A pre-multiplied color in the same order as SkPMColor storing each component as a float.
-struct ALIGN(16) SkPMFloat {
+struct SK_STRUCT_ALIGN(16) SkPMFloat {
     float fColor[4];
 
     float a() const { return fColor[SK_A32_SHIFT / 8]; }
@@ -37,7 +30,6 @@ struct ALIGN(16) SkPMFloat {
             && this->b() >= 0 && this->b() <= this->a();
     }
 };
-#undef ALIGN
 
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2
     #include "../opts/SkPMFloat_SSE2.h"
diff --git a/src/opts/SkPMFloat_SSE2.h b/src/opts/SkPMFloat_SSE2.h
index 7bacf56af7..c3dbbf2cb5 100644
--- a/src/opts/SkPMFloat_SSE2.h
+++ b/src/opts/SkPMFloat_SSE2.h
@@ -1,5 +1,4 @@
 #include "SkColorPriv.h"
-#include "SkPMFloat.h"
 #include <emmintrin.h>
 
 // For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bits (fix8_16),
diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h
index f3d265573d..004b6588dd 100644
--- a/src/opts/SkPMFloat_neon.h
+++ b/src/opts/SkPMFloat_neon.h
@@ -1,5 +1,4 @@
 #include "SkColorPriv.h"
-#include "SkPMFloat.h"
 #include <arm_neon.h>
 
 // For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bits (fix8_16),
diff --git a/src/opts/SkPMFloat_none.h b/src/opts/SkPMFloat_none.h
index dee7ce6862..a11fe241cf 100644
--- a/src/opts/SkPMFloat_none.h
+++ b/src/opts/SkPMFloat_none.h
@@ -1,5 +1,4 @@
 #include "SkColorPriv.h"
-#include "SkPMFloat.h"
 
 inline void SkPMFloat::set(SkPMColor c) {
     float scale = 1.0f / 255.0f;