Clarify iOS flush-to-zero status.

src/core/SkMathPriv.h

 /*
  * Copyright 2012 Google Inc.
  *
  * Use of this source code is governed by a BSD-style license that can be
  * found in the LICENSE file.
  */
 
 #ifndef SkMathPriv_DEFINED
 #define SkMathPriv_DEFINED
 
 #include "SkMath.h"
 
-#ifdef SK_BUILD_FOR_IOS
-// The iOS ARM processor discards small denormalized numbers to go faster.
+#if defined(SK_BUILD_FOR_IOS) && (defined(SK_BUILD_FOR_ARM32) || defined(SK_BUILD_FOR_ARM64))
+// iOS on ARM starts processes with the Flush-To-Zero (FTZ) and
+// Denormals-Are-Zero (DAZ) bits in the fpscr register set.
 // Algorithms that rely on denormalized numbers need alternative implementations.
-#define SK_DISCARD_DENORMALIZED_FOR_SPEED
+// This can also be controlled in SSE with the MXCSR register,
+// x87 with FSTCW/FLDCW, and mips with FCSR. This should be detected at runtime,
+// or the library built one way or the other more generally (by the build).
+#define SK_CPU_FLUSH_TO_ZERO
 #endif
 
 /** Returns -1 if n < 0, else returns 0
  */
 #define SkExtractSign(n)    ((int32_t)(n) >> 31)
 
 /** If sign == -1, returns -n, else sign must be 0, and returns n.
  Typically used in conjunction with SkExtractSign().
  */
 static inline int32_t SkApplySign(int32_t n, int32_t sign) {
@@ skipping 41 matching lines @@
 }
 
 /** Just the rounding step in SkDiv255Round: round(value / 255)
  */
 static inline unsigned SkDiv255Round(unsigned prod) {
     prod += 128;
     return (prod + (prod >> 8)) >> 8;
 }
 
 #endif