Templatize SkPMFloat to support both 1 and 255 biases.

src/opts/SkPMFloat_sse.h

Index: src/opts/SkPMFloat_sse.h
diff --git a/src/opts/SkPMFloat_sse.h b/src/opts/SkPMFloat_sse.h
index 6a4d5b6b5c1dbbfc9893223ba2ef17e05e9725bc..cb1904a9b2b5d008dd19219fc8e825c4e6046e0a 100644
--- a/src/opts/SkPMFloat_sse.h
+++ b/src/opts/SkPMFloat_sse.h
@@ -7,7 +7,8 @@
 
 namespace {  // See SkPMFloat.h
 
-inline SkPMFloat::SkPMFloat(SkPMColor c) {
+template <int kBias>
+inline SkPMFloat<kBias>::SkPMFloat(SkPMColor c) {
     SkPMColorAssert(c);
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
     const char _ = ~0;  // Zero these bytes.
@@ -18,13 +19,20 @@ inline SkPMFloat::SkPMFloat(SkPMColor c) {
             fix8_16 = _mm_unpacklo_epi8 (fix8,    _mm_setzero_si128()),
             fix8_32 = _mm_unpacklo_epi16(fix8_16, _mm_setzero_si128());
 #endif
-    fVec = _mm_mul_ps(_mm_cvtepi32_ps(fix8_32), _mm_set1_ps(1.0f / 255));
+    fVec = _mm_cvtepi32_ps(fix8_32);
+    if (kBias == 1) {
+        fVec = _mm_mul_ps(fVec, _mm_set1_ps(1.0f / 255));
+    }
     SkASSERT(this->isValid());
 }
 
-inline SkPMColor SkPMFloat::round() const {
+template <int kBias>
+inline SkPMColor SkPMFloat<kBias>::round() const {
+    __m128 scaled = fVec;
+    if (kBias == 1) {
+        scaled = _mm_mul_ps(scaled, _mm_set1_ps(255));
+    }
     // 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)),
             fix8_16 = _mm_packus_epi16(fix8_32, fix8_32),
             fix8    = _mm_packus_epi16(fix8_16, fix8_16);
@@ -33,12 +41,14 @@ inline SkPMColor SkPMFloat::round() const {
     return c;
 }
 
-inline Sk4f SkPMFloat::alphas() const {
+template <int kBias>
+inline Sk4f SkPMFloat<kBias>::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.
 }
 
-inline SkPMFloat SkPMFloat::FromOpaqueColor(SkColor c) {
+template <int kBias>
+inline SkPMFloat<kBias> SkPMFloat<kBias>::FromOpaqueColor(SkColor c) {
     SkASSERT(SkColorGetA(c) == 0xFF);
     __m128i fix8 = _mm_cvtsi32_si128((int)c);
 #if SK_CPU_SSE_LEVEL >= SK_CPU_SSE_LEVEL_SSSE3
@@ -57,7 +67,11 @@ inline SkPMFloat SkPMFloat::FromOpaqueColor(SkColor c) {
         fix8_32 = _mm_shuffle_epi32(fix8_32, 0xC6);  // C6 == 11 00 01 10, i.e swap lanes 0 and 2.
     #endif
 #endif
-    SkPMFloat pmf = Sk4f(_mm_mul_ps(_mm_cvtepi32_ps(fix8_32), _mm_set1_ps(1.0f/255)));
+    __m128 floats = _mm_cvtepi32_ps(fix8_32);
+    if (kBias == 1) {
+        floats = _mm_mul_ps(floats, _mm_set1_ps(1.0f/255));
+    }
+    SkPMFloat pmf = Sk4f(floats);
     SkASSERT(pmf.isValid());
     return pmf;
 }