diff --git a/src/core/SkMipMap.cpp b/src/core/SkMipMap.cpp
index 987949fea0..ba3cb18749 100644
--- a/src/core/SkMipMap.cpp
+++ b/src/core/SkMipMap.cpp
@@ -23,6 +23,7 @@
 
 struct ColorTypeFilter_8888 {
     typedef uint32_t Type;
+#if defined(SKNX_IS_FAST)
     static Sk4h Expand(uint32_t x) {
         return SkNx_cast<uint16_t>(Sk4b::Load(&x));
     }
@@ -31,6 +32,14 @@ struct ColorTypeFilter_8888 {
         SkNx_cast<uint8_t>(x).store(&r);
         return r;
     }
+#else
+    static uint64_t Expand(uint32_t x) {
+        return (x & 0xFF00FF) | ((uint64_t)(x & 0xFF00FF00) << 24);
+    }
+    static uint32_t Compact(uint64_t x) {
+        return (uint32_t)((x & 0xFF00FF) | ((x >> 24) & 0xFF00FF00));
+    }
+#endif
 };
 
 struct ColorTypeFilter_S32 {
diff --git a/src/effects/gradients/SkLinearGradient.cpp b/src/effects/gradients/SkLinearGradient.cpp
index 4bb640de91..5994fd6d0c 100644
--- a/src/effects/gradients/SkLinearGradient.cpp
+++ b/src/effects/gradients/SkLinearGradient.cpp
@@ -276,12 +276,15 @@ void SkLinearGradient::LinearGradientContext::shadeSpan(int x, int y, SkPMColor*
     SkASSERT(count > 0);
     const SkLinearGradient& linearGradient = static_cast<const SkLinearGradient&>(fShader);
 
+// Only use the Sk4f impl when known to be fast.
+#if defined(SKNX_IS_FAST)
     if (SkShader::kClamp_TileMode == linearGradient.fTileMode &&
         kLinear_MatrixClass == fDstToIndexClass)
     {
         this->shade4_clamp(x, y, dstC, count);
         return;
     }
+#endif
 
     SkPoint             srcPt;
     SkMatrix::MapXYProc dstProc = fDstToIndexProc;
diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h
index 6836a527e7..83873a76e1 100644
--- a/src/opts/SkNx_neon.h
+++ b/src/opts/SkNx_neon.h
@@ -10,6 +10,8 @@
 
 #include <arm_neon.h>
 
+#define SKNX_IS_FAST
+
 namespace {
 
 // ARMv8 has vrndmq_f32 to floor 4 floats.  Here we emulate it:
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index 394115626e..554d65d5ce 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -13,6 +13,8 @@
 // This file may assume <= SSE2, but must check SK_CPU_SSE_LEVEL for anything more recent.
 // If you do, make sure this is in a static inline function... anywhere else risks violating ODR.
 
+#define SKNX_IS_FAST
+
 namespace {
 
 template <>