SkPMFloat: avoid loads and stores where possible.

src/opts/SkPMFloat_neon.h

Index: src/opts/SkPMFloat_neon.h
diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h
index 036d10d0d6e152c2d4f1f7c9fd569db2f51cfaa3..6c9df37e5107b018847a09b71e7b6585eb2bc4d3 100644
--- a/src/opts/SkPMFloat_neon.h
+++ b/src/opts/SkPMFloat_neon.h
@@ -1,5 +1,7 @@
-#include "SkColorPriv.h"
-#include <arm_neon.h>
+inline SkPMFloat& SkPMFloat::operator=(const SkPMFloat& that) {
+    fColors = that.fColors;
+    return *this;
+}
 
 // For SkPMFloat(SkPMFColor), we widen our 8 bit components (fix8) to 8-bit components in 16 bits
 // (fix8_16), then widen those to 8-bit-in-32-bits (fix8_32), and finally convert those to floats.
@@ -13,13 +15,13 @@ inline SkPMFloat::SkPMFloat(SkPMColor c) {
     uint8x8_t   fix8    = (uint8x8_t)vdup_n_u32(c);
     uint16x8_t  fix8_16 = vmovl_u8(fix8);
     uint32x4_t  fix8_32 = vmovl_u16(vget_low_u16(fix8_16));
-    vst1q_f32(fColor, vcvtq_f32_u32(fix8_32));
+    fColors = vcvtq_f32_u32(fix8_32);
     SkASSERT(this->isValid());
 }
 
 inline SkPMColor SkPMFloat::get() const {
     SkASSERT(this->isValid());
-    float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f));
+    float32x4_t add_half = vaddq_f32(fColors, vdupq_n_f32(0.5f));
     uint32x4_t  fix8_32  = vcvtq_u32_f32(add_half);  // vcvtq_u32_f32 truncates, so round manually
     uint16x4_t  fix8_16  = vmovn_u32(fix8_32);
     uint8x8_t   fix8     = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
@@ -29,7 +31,7 @@ inline SkPMColor SkPMFloat::get() const {
 }
 
 inline SkPMColor SkPMFloat::clamped() const {
-    float32x4_t add_half = vaddq_f32(vld1q_f32(fColor), vdupq_n_f32(0.5f));
+    float32x4_t add_half = vaddq_f32(fColors, vdupq_n_f32(0.5f));
     uint32x4_t  fix8_32  = vcvtq_u32_f32(add_half);  // vcvtq_u32_f32 truncates, so round manually
     uint16x4_t  fix8_16  = vqmovn_u32(fix8_32);
     uint8x8_t   fix8     = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));