Re-proc SkBlitRow::Color32 for ARM.

This is a spiritual revert of http://crrev.com/1104183004.

BUG=skia:

Review URL: https://codereview.chromium.org/1145283003

include/core/SkBlitRow.h

@@ -82,6 +82,9 @@ public:
     static ColorProc16 PlatformColorFactory565(unsigned flags);
 
 private:
+    typedef void (*Color32Proc)(SkPMColor[], const SkPMColor[], int, SkPMColor);
+    static Color32Proc PlatformColor32Proc();
+
     enum {
         kFlags16_Mask = 7,
         kFlags32_Mask = 3

src/core/SkBlitRow_D32.cpp

@@ -144,6 +144,11 @@ void SkBlitRow::Color32(SkPMColor dst[], const SkPMColor src[], int count, SkPMC
         case 255: sk_memset32(dst, color, count);               return;
     }
 
+    // This Sk4px impl works great on other platforms or when we have NEON.
+#if defined(SK_CPU_ARM32) && !defined(SK_ARM_HAS_NEON)
+    if (auto proc = PlatformColor32Proc()) { return proc(dst, src, count, color); }
+#endif
+
     unsigned invA = 255 - SkGetPackedA32(color);
     invA += invA >> 7;
     SkASSERT(invA < 256);  // We've already handled alpha == 0 above.

src/opts/SkBlitRow_opts_arm.cpp

@@ -390,3 +390,7 @@ SkBlitRow::Proc32 SkBlitRow::PlatformProcs32(unsigned flags) {
     return SK_ARM_NEON_WRAP(sk_blitrow_platform_32_procs_arm)[flags];
 }
 
+SkBlitRow::Color32Proc SkBlitRow::PlatformColor32Proc() {
+    #define sk_blitrow_color32_arm NULL
+    return SK_ARM_NEON_WRAP(sk_blitrow_color32_arm);
+}

src/opts/SkBlitRow_opts_arm_neon.cpp

@@ -1730,3 +1730,20 @@ const SkBlitRow::Proc32 sk_blitrow_platform_32_procs_arm_neon[] = {
     NULL
 #endif
 };
+
+#include "Sk4px.h"
+
+void sk_blitrow_color32_arm_neon(SkPMColor* dst, const SkPMColor* src, int count, SkPMColor color) {
+    // Until it becomes a little more reasonable to assume we'll be built with NEON,
+    // we copy our Sk4px implementation of SkBlitRow::Color32 here so it picks up NEON at runtime.
+    unsigned invA = 255 - SkGetPackedA32(color);
+    invA += invA >> 7;
+    SkASSERT(invA < 256);  // Our caller has already handled the alpha == 0 case.
+
+    Sk16h colorHighAndRound = Sk4px(color).widenHi() + Sk16h(128);
+    Sk16b invA_16x(invA);
+
+    Sk4px::MapSrc(count, dst, src, [&](const Sk4px& src4) -> Sk4px {
+        return src4.mulWiden(invA_16x).addNarrowHi(colorHighAndRound);
+    });
+}

src/opts/SkBlitRow_opts_arm_neon.h

@@ -13,4 +13,6 @@ extern const SkBlitRow::Proc16 sk_blitrow_platform_565_procs_arm_neon[];
 extern const SkBlitRow::ColorProc16 sk_blitrow_platform_565_colorprocs_arm_neon[];
 extern const SkBlitRow::Proc32 sk_blitrow_platform_32_procs_arm_neon[];
 
+extern void sk_blitrow_color32_arm_neon(SkPMColor[], const SkPMColor[], int, SkPMColor);
+
 #endif