Use vmulq_n_u32(..., 0x01010101) to distribute alphas.

src/opts/Sk4px_NEON.h

Index: src/opts/Sk4px_NEON.h
diff --git a/src/opts/Sk4px_NEON.h b/src/opts/Sk4px_NEON.h
index 644a71f783fb93e1e50b1ae2c10f1992def5aae0..9401864697afff86f351c0cf551ec342665ff19a 100644
--- a/src/opts/Sk4px_NEON.h
+++ b/src/opts/Sk4px_NEON.h
@@ -52,33 +52,26 @@ inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
 }
 
 inline Sk4px Sk4px::alphas() const {
-    static_assert(SK_A32_SHIFT == 24, "This method assumes little-endian.");
-    auto as = vshrq_n_u32((uint32x4_t)this->fVec, 24);  // ___3 ___2 ___1 ___0
-    as = vorrq_u32(as, vshlq_n_u32(as,  8));            // __33 __22 __11 __11
-    as = vorrq_u32(as, vshlq_n_u32(as, 16));            // 3333 2222 1111 1111
-    return Sk16b((uint8x16_t)as);
+    auto as = vshrq_n_u32((uint32x4_t)fVec, SK_A32_SHIFT);  // ___3 ___2 ___1 ___0
+    return Sk16b((uint8x16_t)vmulq_n_u32(as, 0x01010101));  // 3333 2222 1111 0000
 }
 
 inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
-    uint8x16_t a8 = vdupq_n_u8(0);                        // ____ ____ ____ ____
-    a8 = vld1q_lane_u8(a+0, a8,  0);                      // ____ ____ ____ ___0
-    a8 = vld1q_lane_u8(a+1, a8,  4);                      // ____ ____ ___1 ___0
-    a8 = vld1q_lane_u8(a+2, a8,  8);                      // ____ ___2 ___1 ___0
-    a8 = vld1q_lane_u8(a+3, a8, 12);                      // ___3 ___2 ___1 ___0
-    auto a32 = (uint32x4_t)a8;                            //
-    a32 = vorrq_u32(a32, vshlq_n_u32(a32,  8));           // __33 __22 __11 __00
-    a32 = vorrq_u32(a32, vshlq_n_u32(a32, 16));           // 3333 2222 1111 0000
-    return Sk16b((uint8x16_t)a32);
+    uint8x16_t a8 = vdupq_n_u8(0);                           // ____ ____ ____ ____
+    a8 = vld1q_lane_u8(a+0, a8,  0);                         // ____ ____ ____ ___0
+    a8 = vld1q_lane_u8(a+1, a8,  4);                         // ____ ____ ___1 ___0
+    a8 = vld1q_lane_u8(a+2, a8,  8);                         // ____ ___2 ___1 ___0
+    a8 = vld1q_lane_u8(a+3, a8, 12);                         // ___3 ___2 ___1 ___0
+    auto a32 = (uint32x4_t)a8;                               //
+    return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101));  // 3333 2222 1111 0000
 }
 
 inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
-    uint8x16_t a8 = vdupq_n_u8(0);                        // ____ ____ ____ ____
-    a8 = vld1q_lane_u8(a+0, a8,  0);                      // ____ ____ ____ ___0
-    a8 = vld1q_lane_u8(a+1, a8,  4);                      // ____ ____ ___1 ___0
-    auto a32 = (uint32x4_t)a8;                            //
-    a32 = vorrq_u32(a32, vshlq_n_u32(a32,  8));           // ____ ____ __11 __00
-    a32 = vorrq_u32(a32, vshlq_n_u32(a32, 16));           // ____ ____ 1111 0000
-    return Sk16b((uint8x16_t)a32);
+    uint8x16_t a8 = vdupq_n_u8(0);                           // ____ ____ ____ ____
+    a8 = vld1q_lane_u8(a+0, a8,  0);                         // ____ ____ ____ ___0
+    a8 = vld1q_lane_u8(a+1, a8,  4);                         // ____ ____ ___1 ___0
+    auto a32 = (uint32x4_t)a8;                               //
+    return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101));  // ____ ____ 1111 0000
 }
 
 inline Sk4px Sk4px::zeroColors() const {