Refactor to put SkXfermode_opts inside SK_OPTS_NS.

Without this refactor I was getting warnings previously about having code
inside namespace SK_OPTS_NS (e.g. namespace sse2, namespace neon) referring to
code inside an anonymous namespace (Sk4px, SkPMFloat, Sk4f, etc) [1].

That low-level code was in an anonymous namespace to allow multiple independent
copies of its methods to be instantiated without the linker getting confused /
offended about violating the One Definition Rule.  This was only happening in
Debug mode where the methods were not being inlined.

To fix this all, I've force-inlined the methods of the low-level code and
removed the anonymous namespace.

BUG=skia:4117

[1] Here is what those errors looked like:

In file included from ../../../../src/core/SkOpts.cpp:18:0:
../../../../src/opts/SkXfermode_opts.h:193:7: error: 'portable::Sk4pxXfermode' has a field 'portable::Sk4pxXfermode::fProc4' whose type uses the anonymous namespace [-Werror]
 class Sk4pxXfermode : public SkProcCoeffXfermode {
       ^
../../../../src/opts/SkXfermode_opts.h:193:7: error: 'portable::Sk4pxXfermode' has a field 'portable::Sk4pxXfermode::fAAProc4' whose type uses the anonymous namespace [-Werror]
../../../../src/opts/SkXfermode_opts.h:235:7: error: 'portable::SkPMFloatXfermode' has a field 'portable::SkPMFloatXfermode::fProcF' whose type uses the anonymous namespace [-Werror]
 class SkPMFloatXfermode : public SkProcCoeffXfermode {
       ^
cc1plus: all warnings being treated as errors

Review URL: https://codereview.chromium.org/1286093004

15 files changed, 97 insertions, 149 deletions

diff --git a/src/core/Sk4px.h b/src/core/Sk4px.h
index ffde1af504..1c8ed0df71 100644
--- a/src/core/Sk4px.h
+++ b/src/core/Sk4px.h
@@ -12,13 +12,6 @@
 #include "SkColor.h"
 #include "SkColorPriv.h"
 
-// This file may be included multiple times by .cpp files with different flags, leading
-// to different definitions.  Usually that doesn't matter because it's all inlined, but
-// in Debug modes the compilers may not inline everything.  So wrap everything in an
-// anonymous namespace to give each includer their own silo of this code (or the linker
-// will probably pick one randomly for us, which is rarely correct).
-namespace {
-
 // 1, 2 or 4 SkPMColors, generally vectorized.
 class Sk4px : public Sk16b {
 public:
@@ -226,8 +219,6 @@ private:
     typedef Sk16b INHERITED;
 };
 
-}  // namespace
-
 #ifdef SKNX_NO_SIMD
     #include "../opts/Sk4px_none.h"
 #else
diff --git a/src/core/SkNx.h b/src/core/SkNx.h
index 84f9b69353..728e450ebb 100644
--- a/src/core/SkNx.h
+++ b/src/core/SkNx.h
@@ -17,13 +17,6 @@
 #include <math.h>
 #define REQUIRE(x) static_assert(x, #x)
 
-// This file may be included multiple times by .cpp files with different flags, leading
-// to different definitions.  Usually that doesn't matter because it's all inlined, but
-// in Debug modes the compilers may not inline everything.  So wrap everything in an
-// anonymous namespace to give each includer their own silo of this code (or the linker
-// will probably pick one randomly for us, which is rarely correct).
-namespace {
-
 // The default implementations just fall back on a pair of size N/2.
 
 template <int N, typename T>
@@ -252,8 +245,6 @@ protected:
     T fVal;
 };
 
-}  // namespace
-
 // Include platform specific specializations if available.
 #ifndef SKNX_NO_SIMD
     #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSE2
diff --git a/src/core/SkOpts.cpp b/src/core/SkOpts.cpp
index 2bfc1af899..986c7cf6a8 100644
--- a/src/core/SkOpts.cpp
+++ b/src/core/SkOpts.cpp
@@ -37,7 +37,7 @@ namespace SkOpts {
     decltype(rsqrt)                     rsqrt = portable::rsqrt;
     decltype(memset16)               memset16 = portable::memset16;
     decltype(memset32)               memset32 = portable::memset32;
-    decltype(create_xfermode) create_xfermode = SkCreate4pxXfermode;
+    decltype(create_xfermode) create_xfermode = portable::create_xfermode;
 
     decltype(box_blur_xx) box_blur_xx = portable::box_blur_xx;
     decltype(box_blur_xy) box_blur_xy = portable::box_blur_xy;
diff --git a/src/core/SkPMFloat.h b/src/core/SkPMFloat.h
index f97f25c9c6..7db689917e 100644
--- a/src/core/SkPMFloat.h
+++ b/src/core/SkPMFloat.h
@@ -13,13 +13,6 @@
 #include "SkColorPriv.h"
 #include "SkNx.h"
 
-// This file may be included multiple times by .cpp files with different flags, leading
-// to different definitions.  Usually that doesn't matter because it's all inlined, but
-// in Debug modes the compilers may not inline everything.  So wrap everything in an
-// anonymous namespace to give each includer their own silo of this code (or the linker
-// will probably pick one randomly for us, which is rarely correct).
-namespace {
-
 // A pre-multiplied color storing each component in the same order as SkPMColor,
 // but as a float in the range [0, 1].
 class SkPMFloat : public Sk4f {
@@ -59,8 +52,6 @@ private:
     typedef Sk4f INHERITED;
 };
 
-}  // namespace
-
 #ifdef SKNX_NO_SIMD
     // Platform implementations of SkPMFloat assume Sk4f uses SSE or NEON.  _none is generic.
     #include "../opts/SkPMFloat_none.h"
diff --git a/src/opts/Sk4px_NEON.h b/src/opts/Sk4px_NEON.h
index 89841d927e..e88771d2a4 100644
--- a/src/opts/Sk4px_NEON.h
+++ b/src/opts/Sk4px_NEON.h
@@ -5,64 +5,64 @@
  * found in the LICENSE file.
  */
 
-namespace { // See Sk4px.h
-
-inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b((uint8x16_t)vdupq_n_u32(px)); }
+SK_ALWAYS_INLINE Sk4px Sk4px::DupPMColor(SkPMColor px) {
+    return Sk16b((uint8x16_t)vdupq_n_u32(px));
+}
 
-inline Sk4px Sk4px::Load4(const SkPMColor px[4]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load4(const SkPMColor px[4]) {
     return Sk16b((uint8x16_t)vld1q_u32(px));
 }
-inline Sk4px Sk4px::Load2(const SkPMColor px[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2(const SkPMColor px[2]) {
     uint32x2_t px2 = vld1_u32(px);
     return Sk16b((uint8x16_t)vcombine_u32(px2, px2));
 }
-inline Sk4px Sk4px::Load1(const SkPMColor px[1]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load1(const SkPMColor px[1]) {
     return Sk16b((uint8x16_t)vdupq_n_u32(*px));
 }
 
-inline void Sk4px::store4(SkPMColor px[4]) const {
+SK_ALWAYS_INLINE void Sk4px::store4(SkPMColor px[4]) const {
     vst1q_u32(px, (uint32x4_t)this->fVec);
 }
-inline void Sk4px::store2(SkPMColor px[2]) const {
+SK_ALWAYS_INLINE void Sk4px::store2(SkPMColor px[2]) const {
     vst1_u32(px, (uint32x2_t)vget_low_u8(this->fVec));
 }
-inline void Sk4px::store1(SkPMColor px[1]) const {
+SK_ALWAYS_INLINE void Sk4px::store1(SkPMColor px[1]) const {
     vst1q_lane_u32(px, (uint32x4_t)this->fVec, 0);
 }
 
-inline Sk4px::Wide Sk4px::widenLo() const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenLo() const {
     return Sk16h(vmovl_u8(vget_low_u8 (this->fVec)),
                  vmovl_u8(vget_high_u8(this->fVec)));
 }
 
-inline Sk4px::Wide Sk4px::widenHi() const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenHi() const {
     return Sk16h(vshll_n_u8(vget_low_u8 (this->fVec), 8),
                  vshll_n_u8(vget_high_u8(this->fVec), 8));
 }
 
-inline Sk4px::Wide Sk4px::widenLoHi() const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenLoHi() const {
     auto zipped = vzipq_u8(this->fVec, this->fVec);
     return Sk16h((uint16x8_t)zipped.val[0],
                  (uint16x8_t)zipped.val[1]);
 }
 
-inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
     return Sk16h(vmull_u8(vget_low_u8 (this->fVec), vget_low_u8 (other.fVec)),
                  vmull_u8(vget_high_u8(this->fVec), vget_high_u8(other.fVec)));
 }
 
-inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
+SK_ALWAYS_INLINE Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
     const Sk4px::Wide o(other);  // Should be no code, but allows us to access fLo, fHi.
     return Sk16b(vcombine_u8(vaddhn_u16(this->fLo.fVec, o.fLo.fVec),
                              vaddhn_u16(this->fHi.fVec, o.fHi.fVec)));
 }
 
-inline Sk4px Sk4px::alphas() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::alphas() const {
     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]) {
+SK_ALWAYS_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
@@ -72,7 +72,7 @@ inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
     return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101));  // 3333 2222 1111 0000
 }
 
-inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
+SK_ALWAYS_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
@@ -80,16 +80,16 @@ inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
     return Sk16b((uint8x16_t)vmulq_n_u32(a32, 0x01010101));  // ____ ____ 1111 0000
 }
 
-inline Sk4px Sk4px::zeroColors() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::zeroColors() const {
     return Sk16b(vandq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
 }
 
-inline Sk4px Sk4px::zeroAlphas() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::zeroAlphas() const {
     // vbic(a,b) == a & ~b
     return Sk16b(vbicq_u8(this->fVec, (uint8x16_t)vdupq_n_u32(0xFF << SK_A32_SHIFT)));
 }
 
-static inline uint8x16_t widen_to_8888(uint16x4_t v) {
+static SK_ALWAYS_INLINE uint8x16_t widen_to_8888(uint16x4_t v) {
     // RGB565 format:   |R....|G.....|B....|
     //           Bit:  16    11      5     0
 
@@ -115,7 +115,7 @@ static inline uint8x16_t widen_to_8888(uint16x4_t v) {
                                  vdupq_n_u32(0xFF << SK_A32_SHIFT))));
 }
 
-static inline uint16x4_t narrow_to_565(uint8x16_t w8x16) {
+static SK_ALWAYS_INLINE uint16x4_t narrow_to_565(uint8x16_t w8x16) {
     uint32x4_t w = (uint32x4_t)w8x16;
 
     // Extract out top RGB 565 bits of each pixel, with no rounding.
@@ -136,29 +136,27 @@ static inline uint16x4_t narrow_to_565(uint8x16_t w8x16) {
 }
 
 
-inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
     return Sk16b(widen_to_8888(vld1_u16(src)));
 }
-inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
     auto src2 = ((uint32_t)src[0]      )
               | ((uint32_t)src[1] << 16);
     return Sk16b(widen_to_8888(vcreate_u16(src2)));
 }
-inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
     return Sk16b(widen_to_8888(vcreate_u16(src[0])));
 }
 
-inline void Sk4px::store4(SkPMColor16 dst[4]) const {
+SK_ALWAYS_INLINE void Sk4px::store4(SkPMColor16 dst[4]) const {
     vst1_u16(dst, narrow_to_565(this->fVec));
 }
-inline void Sk4px::store2(SkPMColor16 dst[2]) const {
+SK_ALWAYS_INLINE void Sk4px::store2(SkPMColor16 dst[2]) const {
     auto v = narrow_to_565(this->fVec);
     dst[0] = vget_lane_u16(v, 0);
     dst[1] = vget_lane_u16(v, 1);
 }
-inline void Sk4px::store1(SkPMColor16 dst[1]) const {
+SK_ALWAYS_INLINE void Sk4px::store1(SkPMColor16 dst[1]) const {
     dst[0] = vget_lane_u16(narrow_to_565(this->fVec), 0);
 }
 
-} // namespace
-
diff --git a/src/opts/Sk4px_SSE2.h b/src/opts/Sk4px_SSE2.h
index 5e97abf308..f235a1560e 100644
--- a/src/opts/Sk4px_SSE2.h
+++ b/src/opts/Sk4px_SSE2.h
@@ -5,42 +5,46 @@
  * found in the LICENSE file.
  */
 
-namespace { // See Sk4px.h
+SK_ALWAYS_INLINE Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b(_mm_set1_epi32(px)); }
 
-inline Sk4px Sk4px::DupPMColor(SkPMColor px) { return Sk16b(_mm_set1_epi32(px)); }
-
-inline Sk4px Sk4px::Load4(const SkPMColor px[4]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load4(const SkPMColor px[4]) {
     return Sk16b(_mm_loadu_si128((const __m128i*)px));
 }
-inline Sk4px Sk4px::Load2(const SkPMColor px[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2(const SkPMColor px[2]) {
     return Sk16b(_mm_loadl_epi64((const __m128i*)px));
 }
-inline Sk4px Sk4px::Load1(const SkPMColor px[1]) { return Sk16b(_mm_cvtsi32_si128(*px)); }
+SK_ALWAYS_INLINE Sk4px Sk4px::Load1(const SkPMColor px[1]) { return Sk16b(_mm_cvtsi32_si128(*px)); }
 
-inline void Sk4px::store4(SkPMColor px[4]) const { _mm_storeu_si128((__m128i*)px, this->fVec); }
-inline void Sk4px::store2(SkPMColor px[2]) const { _mm_storel_epi64((__m128i*)px, this->fVec); }
-inline void Sk4px::store1(SkPMColor px[1]) const { *px = _mm_cvtsi128_si32(this->fVec); }
+SK_ALWAYS_INLINE void Sk4px::store4(SkPMColor px[4]) const {
+    _mm_storeu_si128((__m128i*)px, this->fVec);
+}
+SK_ALWAYS_INLINE void Sk4px::store2(SkPMColor px[2]) const {
+    _mm_storel_epi64((__m128i*)px, this->fVec);
+}
+SK_ALWAYS_INLINE void Sk4px::store1(SkPMColor px[1]) const {
+    *px = _mm_cvtsi128_si32(this->fVec);
+}
 
-inline Sk4px::Wide Sk4px::widenLo() const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenLo() const {
     return Sk16h(_mm_unpacklo_epi8(this->fVec, _mm_setzero_si128()),
                  _mm_unpackhi_epi8(this->fVec, _mm_setzero_si128()));
 }
 
-inline Sk4px::Wide Sk4px::widenHi() const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenHi() const {
     return Sk16h(_mm_unpacklo_epi8(_mm_setzero_si128(), this->fVec),
                  _mm_unpackhi_epi8(_mm_setzero_si128(), this->fVec));
 }
 
-inline Sk4px::Wide Sk4px::widenLoHi() const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenLoHi() const {
     return Sk16h(_mm_unpacklo_epi8(this->fVec, this->fVec),
                  _mm_unpackhi_epi8(this->fVec, this->fVec));
 }
 
-inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
     return this->widenLo() * Sk4px(other).widenLo();
 }
 
-inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
+SK_ALWAYS_INLINE Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
     Sk4px::Wide r = (*this + other) >> 8;
     return Sk4px(_mm_packus_epi16(r.fLo.fVec, r.fHi.fVec));
 }
@@ -49,19 +53,19 @@ inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
 // These are safe on x86, often with no speed penalty.
 
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
-    inline Sk4px Sk4px::alphas() const {
+    SK_ALWAYS_INLINE Sk4px Sk4px::alphas() const {
         static_assert(SK_A32_SHIFT == 24, "Intel's always little-endian.");
         __m128i splat = _mm_set_epi8(15,15,15,15, 11,11,11,11, 7,7,7,7, 3,3,3,3);
         return Sk16b(_mm_shuffle_epi8(this->fVec, splat));
     }
 
-    inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
+    SK_ALWAYS_INLINE Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
         uint32_t as = *(const uint32_t*)a;
         __m128i splat = _mm_set_epi8(3,3,3,3, 2,2,2,2, 1,1,1,1, 0,0,0,0);
         return Sk16b(_mm_shuffle_epi8(_mm_cvtsi32_si128(as), splat));
     }
 #else
-    inline Sk4px Sk4px::alphas() const {
+    SK_ALWAYS_INLINE Sk4px Sk4px::alphas() const {
         static_assert(SK_A32_SHIFT == 24, "Intel's always little-endian.");
         __m128i as = _mm_srli_epi32(this->fVec, 24);   // ___3 ___2 ___1 ___0
         as = _mm_or_si128(as, _mm_slli_si128(as, 1));  // __33 __22 __11 __00
@@ -69,7 +73,7 @@ inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
         return Sk16b(as);
     }
 
-    inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
+    SK_ALWAYS_INLINE Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
         __m128i as = _mm_cvtsi32_si128(*(const uint32_t*)a);  // ____ ____ ____ 3210
         as = _mm_unpacklo_epi8 (as, _mm_setzero_si128());     // ____ ____ _3_2 _1_0
         as = _mm_unpacklo_epi16(as, _mm_setzero_si128());     // ___3 ___2 ___1 ___0
@@ -79,21 +83,21 @@ inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
     }
 #endif
 
-inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
     uint32_t as = *(const uint16_t*)a;   // Aa -> Aa00
     return Load4Alphas((const SkAlpha*)&as);
 }
 
-inline Sk4px Sk4px::zeroColors() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::zeroColors() const {
     return Sk16b(_mm_and_si128(_mm_set1_epi32(0xFF << SK_A32_SHIFT), this->fVec));
 }
 
-inline Sk4px Sk4px::zeroAlphas() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::zeroAlphas() const {
     // andnot(a,b) == ~a & b
     return Sk16b(_mm_andnot_si128(_mm_set1_epi32(0xFF << SK_A32_SHIFT), this->fVec));
 }
 
-static inline __m128i widen_low_half_to_8888(__m128i v) {
+static SK_ALWAYS_INLINE __m128i widen_low_half_to_8888(__m128i v) {
     // RGB565 format:   |R....|G.....|B....|
     //           Bit:  16    11      5     0
 
@@ -119,7 +123,7 @@ static inline __m128i widen_low_half_to_8888(__m128i v) {
                         _mm_set1_epi32(0xFF << SK_A32_SHIFT))));
 }
 
-static inline __m128i narrow_to_565(__m128i w) {
+static SK_ALWAYS_INLINE __m128i narrow_to_565(__m128i w) {
     // Extract out top RGB 565 bits of each pixel, with no rounding.
     auto r5 = _mm_and_si128(_mm_set1_epi32(31), _mm_srli_epi32(w, SK_R32_SHIFT + 3)),
          g6 = _mm_and_si128(_mm_set1_epi32(63), _mm_srli_epi32(w, SK_G32_SHIFT + 2)),
@@ -143,29 +147,27 @@ static inline __m128i narrow_to_565(__m128i w) {
     return v;
 }
 
-inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
     return Sk16b(widen_low_half_to_8888(_mm_loadl_epi64((const __m128i*)src)));
 }
-inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
     auto src2 = ((uint32_t)src[0]      )
               | ((uint32_t)src[1] << 16);
     return Sk16b(widen_low_half_to_8888(_mm_cvtsi32_si128(src2)));
 }
-inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
     return Sk16b(widen_low_half_to_8888(_mm_insert_epi16(_mm_setzero_si128(), src[0], 0)));
 }
 
-inline void Sk4px::store4(SkPMColor16 dst[4]) const {
+SK_ALWAYS_INLINE void Sk4px::store4(SkPMColor16 dst[4]) const {
     _mm_storel_epi64((__m128i*)dst, narrow_to_565(this->fVec));
 }
-inline void Sk4px::store2(SkPMColor16 dst[2]) const {
+SK_ALWAYS_INLINE void Sk4px::store2(SkPMColor16 dst[2]) const {
     uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec));
     dst[0] = dst2;
     dst[1] = dst2 >> 16;
 }
-inline void Sk4px::store1(SkPMColor16 dst[1]) const {
+SK_ALWAYS_INLINE void Sk4px::store1(SkPMColor16 dst[1]) const {
     uint32_t dst2 = _mm_cvtsi128_si32(narrow_to_565(this->fVec));
     dst[0] = dst2;
 }
-
-}  // namespace
diff --git a/src/opts/Sk4px_none.h b/src/opts/Sk4px_none.h
index 540edb821d..d2231c9ba1 100644
--- a/src/opts/Sk4px_none.h
+++ b/src/opts/Sk4px_none.h
@@ -7,54 +7,52 @@
 
 #include "SkUtils.h"
 
-namespace { // See Sk4px.h
-
 static_assert(sizeof(Sk4px) == 16, "This file uses memcpy / sk_memset32, so exact size matters.");
 
-inline Sk4px Sk4px::DupPMColor(SkPMColor px) {
+SK_ALWAYS_INLINE Sk4px Sk4px::DupPMColor(SkPMColor px) {
     Sk4px px4 = Sk16b();
     sk_memset32((uint32_t*)&px4, px, 4);
     return px4;
 }
 
-inline Sk4px Sk4px::Load4(const SkPMColor px[4]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load4(const SkPMColor px[4]) {
     Sk4px px4 = Sk16b();
     memcpy(&px4, px, 16);
     return px4;
 }
 
-inline Sk4px Sk4px::Load2(const SkPMColor px[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2(const SkPMColor px[2]) {
     Sk4px px2 = Sk16b();
     memcpy(&px2, px, 8);
     return px2;
 }
 
-inline Sk4px Sk4px::Load1(const SkPMColor px[1]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load1(const SkPMColor px[1]) {
     Sk4px px1 = Sk16b();
     memcpy(&px1, px, 4);
     return px1;
 }
 
-inline void Sk4px::store4(SkPMColor px[4]) const { memcpy(px, this, 16); }
-inline void Sk4px::store2(SkPMColor px[2]) const { memcpy(px, this,  8); }
-inline void Sk4px::store1(SkPMColor px[1]) const { memcpy(px, this,  4); }
+SK_ALWAYS_INLINE void Sk4px::store4(SkPMColor px[4]) const { memcpy(px, this, 16); }
+SK_ALWAYS_INLINE void Sk4px::store2(SkPMColor px[2]) const { memcpy(px, this,  8); }
+SK_ALWAYS_INLINE void Sk4px::store1(SkPMColor px[1]) const { memcpy(px, this,  4); }
 
-inline Sk4px::Wide Sk4px::widenLo() const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenLo() const {
     return Sk16h(this->kth< 0>(), this->kth< 1>(), this->kth< 2>(), this->kth< 3>(),
                  this->kth< 4>(), this->kth< 5>(), this->kth< 6>(), this->kth< 7>(),
                  this->kth< 8>(), this->kth< 9>(), this->kth<10>(), this->kth<11>(),
                  this->kth<12>(), this->kth<13>(), this->kth<14>(), this->kth<15>());
 }
 
-inline Sk4px::Wide Sk4px::widenHi() const { return this->widenLo() << 8; }
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenHi() const { return this->widenLo() << 8; }
 
-inline Sk4px::Wide Sk4px::widenLoHi() const { return this->widenLo() + this->widenHi(); }
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::widenLoHi() const { return this->widenLo() + this->widenHi(); }
 
-inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
+SK_ALWAYS_INLINE Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
     return this->widenLo() * Sk4px(other).widenLo();
 }
 
-inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
+SK_ALWAYS_INLINE Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
     Sk4px::Wide r = (*this + other) >> 8;
     return Sk16b(r.kth< 0>(), r.kth< 1>(), r.kth< 2>(), r.kth< 3>(),
                  r.kth< 4>(), r.kth< 5>(), r.kth< 6>(), r.kth< 7>(),
@@ -62,7 +60,7 @@ inline Sk4px Sk4px::Wide::addNarrowHi(const Sk16h& other) const {
                  r.kth<12>(), r.kth<13>(), r.kth<14>(), r.kth<15>());
 }
 
-inline Sk4px Sk4px::alphas() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::alphas() const {
     static_assert(SK_A32_SHIFT == 24, "This method assumes little-endian.");
     return Sk16b(this->kth< 3>(), this->kth< 3>(), this->kth< 3>(), this->kth< 3>(),
                  this->kth< 7>(), this->kth< 7>(), this->kth< 7>(), this->kth< 7>(),
@@ -70,21 +68,21 @@ inline Sk4px Sk4px::alphas() const {
                  this->kth<15>(), this->kth<15>(), this->kth<15>(), this->kth<15>());
 }
 
-inline Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load4Alphas(const SkAlpha a[4]) {
     return Sk16b(a[0], a[0], a[0], a[0],
                  a[1], a[1], a[1], a[1],
                  a[2], a[2], a[2], a[2],
                  a[3], a[3], a[3], a[3]);
 }
 
-inline Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2Alphas(const SkAlpha a[2]) {
     return Sk16b(a[0], a[0], a[0], a[0],
                  a[1], a[1], a[1], a[1],
                  0,0,0,0,
                  0,0,0,0);
 }
 
-inline Sk4px Sk4px::zeroAlphas() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::zeroAlphas() const {
     static_assert(SK_A32_SHIFT == 24, "This method assumes little-endian.");
     return Sk16b(this->kth< 0>(), this->kth< 1>(), this->kth< 2>(), 0,
                  this->kth< 4>(), this->kth< 5>(), this->kth< 6>(), 0,
@@ -92,7 +90,7 @@ inline Sk4px Sk4px::zeroAlphas() const {
                  this->kth<12>(), this->kth<13>(), this->kth<14>(), 0);
 }
 
-inline Sk4px Sk4px::zeroColors() const {
+SK_ALWAYS_INLINE Sk4px Sk4px::zeroColors() const {
     static_assert(SK_A32_SHIFT == 24, "This method assumes little-endian.");
     return Sk16b(0,0,0, this->kth< 3>(),
                  0,0,0, this->kth< 7>(),
@@ -100,35 +98,33 @@ inline Sk4px Sk4px::zeroColors() const {
                  0,0,0, this->kth<15>());
 }
 
-inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
     SkPMColor src32[4];
     for (int i = 0; i < 4; i++) { src32[i] = SkPixel16ToPixel32(src[i]); }
     return Load4(src32);
 }
-inline Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load2(const SkPMColor16 src[2]) {
     SkPMColor src32[2];
     for (int i = 0; i < 2; i++) { src32[i] = SkPixel16ToPixel32(src[i]); }
     return Load2(src32);
 }
-inline Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
+SK_ALWAYS_INLINE Sk4px Sk4px::Load1(const SkPMColor16 src[1]) {
     SkPMColor src32 = SkPixel16ToPixel32(src[0]);
     return Load1(&src32);
 }
 
-inline void Sk4px::store4(SkPMColor16 dst[4]) const {
+SK_ALWAYS_INLINE void Sk4px::store4(SkPMColor16 dst[4]) const {
     SkPMColor dst32[4];
     this->store4(dst32);
     for (int i = 0; i < 4; i++) { dst[i] = SkPixel32ToPixel16(dst32[i]); }
 }
-inline void Sk4px::store2(SkPMColor16 dst[2]) const {
+SK_ALWAYS_INLINE void Sk4px::store2(SkPMColor16 dst[2]) const {
     SkPMColor dst32[2];
     this->store2(dst32);
     for (int i = 0; i < 2; i++) { dst[i] = SkPixel32ToPixel16(dst32[i]); }
 }
-inline void Sk4px::store1(SkPMColor16 dst[1]) const {
+SK_ALWAYS_INLINE void Sk4px::store1(SkPMColor16 dst[1]) const {
     SkPMColor dst32;
     this->store1(&dst32);
     dst[0] = SkPixel32ToPixel16(dst32);
 }
-
-}  // namespace
diff --git a/src/opts/SkNx_neon.h b/src/opts/SkNx_neon.h
index 660b92c6b0..e9e7bf5230 100644
--- a/src/opts/SkNx_neon.h
+++ b/src/opts/SkNx_neon.h
@@ -8,8 +8,6 @@
 #ifndef SkNx_neon_DEFINED
 #define SkNx_neon_DEFINED
 
-namespace {  // See SkNx.h
-
 // Well, this is absurd.  The shifts require compile-time constant arguments.
 
 #define SHIFT8(op, v, bits) switch(bits) { \
@@ -383,6 +381,4 @@ public:
 #undef SHIFT16
 #undef SHIFT8
 
-}  // namespace
-
 #endif//SkNx_neon_DEFINED
diff --git a/src/opts/SkNx_sse.h b/src/opts/SkNx_sse.h
index e165f58737..75b35845a8 100644
--- a/src/opts/SkNx_sse.h
+++ b/src/opts/SkNx_sse.h
@@ -10,9 +10,6 @@
 
 // This file may assume <= SSE2, but must check SK_CPU_SSE_LEVEL for anything more recent.
 
-namespace {  // See SkNx.h
-
-
 template <>
 class SkNf<2, float> {
 public:
@@ -313,6 +310,4 @@ public:
     __m128i fVec;
 };
 
-}  // namespace
-
 #endif//SkNx_sse_DEFINED
diff --git a/src/opts/SkOpts_neon.cpp b/src/opts/SkOpts_neon.cpp
index 789a977238..a6d3f00d20 100644
--- a/src/opts/SkOpts_neon.cpp
+++ b/src/opts/SkOpts_neon.cpp
@@ -21,7 +21,7 @@ namespace SkOpts {
         rsqrt           = neon::rsqrt;
         memset16        = neon::memset16;
         memset32        = neon::memset32;
-        create_xfermode = SkCreate4pxXfermode;
+        create_xfermode = neon::create_xfermode;
 
         box_blur_xx = neon::box_blur_xx;
         box_blur_xy = neon::box_blur_xy;
diff --git a/src/opts/SkOpts_sse2.cpp b/src/opts/SkOpts_sse2.cpp
index 3440676935..b2f0262263 100644
--- a/src/opts/SkOpts_sse2.cpp
+++ b/src/opts/SkOpts_sse2.cpp
@@ -18,7 +18,7 @@ namespace SkOpts {
     void Init_sse2() {
         memset16        = sse2::memset16;
         memset32        = sse2::memset32;
-        create_xfermode = SkCreate4pxXfermode;
+        create_xfermode = sse2::create_xfermode;
 
         box_blur_xx = sse2::box_blur_xx;
         box_blur_xy = sse2::box_blur_xy;
diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h
index 8bee5b551a..edbbc58b5a 100644
--- a/src/opts/SkPMFloat_neon.h
+++ b/src/opts/SkPMFloat_neon.h
@@ -5,9 +5,7 @@
  * found in the LICENSE file.
  */
 
-namespace { // See SkPMFloat.h
-
-inline SkPMFloat::SkPMFloat(SkPMColor c) {
+SK_ALWAYS_INLINE SkPMFloat::SkPMFloat(SkPMColor c) {
     SkPMColorAssert(c);
     uint8x8_t   fix8    = (uint8x8_t)vdup_n_u32(c);
     uint16x8_t  fix8_16 = vmovl_u8(fix8);
@@ -16,7 +14,7 @@ inline SkPMFloat::SkPMFloat(SkPMColor c) {
     SkASSERT(this->isValid());
 }
 
-inline SkPMColor SkPMFloat::round() const {
+SK_ALWAYS_INLINE SkPMColor SkPMFloat::round() const {
     // vcvt_u32_f32 truncates, so we round manually by adding a half before converting.
     float32x4_t rounded = vmlaq_f32(vdupq_n_f32(0.5f), fVec, vdupq_n_f32(255));
     uint32x4_t  fix8_32 = vcvtq_u32_f32(rounded);
@@ -27,9 +25,7 @@ inline SkPMColor SkPMFloat::round() const {
     return c;
 }
 
-inline Sk4f SkPMFloat::alphas() const {
+SK_ALWAYS_INLINE Sk4f SkPMFloat::alphas() const {
     static_assert(SK_A32_SHIFT == 24, "Assuming little-endian.");
     return vdupq_lane_f32(vget_high_f32(fVec), 1);  // Duplicate high lane of high half i.e. lane 3.
 }
-
-}  // namespace
diff --git a/src/opts/SkPMFloat_none.h b/src/opts/SkPMFloat_none.h
index 518ad159ff..ca98b1ab7e 100644
--- a/src/opts/SkPMFloat_none.h
+++ b/src/opts/SkPMFloat_none.h
@@ -5,9 +5,7 @@
  * found in the LICENSE file.
  */
 
-namespace {  // See SkPMFloat.h
-
-inline SkPMFloat::SkPMFloat(SkPMColor c) {
+SK_ALWAYS_INLINE SkPMFloat::SkPMFloat(SkPMColor c) {
     float inv255 = 1.0f/255;
     *this = SkPMFloat::FromARGB(SkGetPackedA32(c) * inv255,
                                 SkGetPackedR32(c) * inv255,
@@ -16,7 +14,7 @@ inline SkPMFloat::SkPMFloat(SkPMColor c) {
     SkASSERT(this->isValid());
 }
 
-inline SkPMColor SkPMFloat::round() const {
+SK_ALWAYS_INLINE SkPMColor SkPMFloat::round() const {
     float a = this->a(),
           r = this->r(),
           g = this->g(),
@@ -30,8 +28,6 @@ inline SkPMColor SkPMFloat::round() const {
     return c;
 }
 
-inline Sk4f SkPMFloat::alphas() const {
+SK_ALWAYS_INLINE Sk4f SkPMFloat::alphas() const {
     return Sk4f(this->a());
 }
-
-}  // namespace
diff --git a/src/opts/SkPMFloat_sse.h b/src/opts/SkPMFloat_sse.h
index 28aa90bf29..016644e052 100644
--- a/src/opts/SkPMFloat_sse.h
+++ b/src/opts/SkPMFloat_sse.h
@@ -5,9 +5,7 @@
  * found in the LICENSE file.
  */
 
-namespace {  // See SkPMFloat.h
-
-inline SkPMFloat::SkPMFloat(SkPMColor c) {
+SK_ALWAYS_INLINE SkPMFloat::SkPMFloat(SkPMColor c) {
     SkPMColorAssert(c);
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
     const int _ = 255;  // Zero these bytes.
@@ -22,7 +20,7 @@ inline SkPMFloat::SkPMFloat(SkPMColor c) {
     SkASSERT(this->isValid());
 }
 
-inline SkPMColor SkPMFloat::round() const {
+SK_ALWAYS_INLINE SkPMColor SkPMFloat::round() const {
     // We don't use _mm_cvtps_epi32, because we want precise control over how 0.5 rounds (up).
     __m128 scaled = _mm_mul_ps(_mm_set1_ps(255), fVec);
     __m128i fix8_32 = _mm_cvttps_epi32(_mm_add_ps(_mm_set1_ps(0.5f), scaled)),
@@ -33,9 +31,7 @@ inline SkPMColor SkPMFloat::round() const {
     return c;
 }
 
-inline Sk4f SkPMFloat::alphas() const {
+SK_ALWAYS_INLINE Sk4f SkPMFloat::alphas() const {
     static_assert(SK_A32_SHIFT == 24, "");
     return _mm_shuffle_ps(fVec, fVec, 0xff);  // Read as 11 11 11 11, copying lane 3 to all lanes.
 }
-
-}  // namespace
diff --git a/src/opts/SkXfermode_opts.h b/src/opts/SkXfermode_opts.h
index e5ca257a28..5394e636e7 100644
--- a/src/opts/SkXfermode_opts.h
+++ b/src/opts/SkXfermode_opts.h
@@ -12,7 +12,7 @@
 #include "SkPMFloat.h"
 #include "SkXfermode_proccoeff.h"
 
-namespace /* TODO: SK_OPTS_NS */ {
+namespace SK_OPTS_NS {
 
 // Most xfermodes can be done most efficiently 4 pixels at a time in 8 or 16-bit fixed point.
 #define XFERMODE(Name) static Sk4px SK_VECTORCALL Name(Sk4px s, Sk4px d)
@@ -274,7 +274,7 @@ private:
     typedef SkProcCoeffXfermode INHERITED;
 };
 
-static SkXfermode* SkCreate4pxXfermode(const ProcCoeff& rec, SkXfermode::Mode mode) {
+static SkXfermode* create_xfermode(const ProcCoeff& rec, SkXfermode::Mode mode) {
     switch (mode) {
     #define CASE(Mode) case SkXfermode::k##Mode##_Mode: \
         return SkNEW_ARGS(Sk4pxXfermode, (rec, mode, &Mode, &xfer_aa<Mode>))