3 files changed, 106 insertions, 0 deletions
diff --git a/src/opts/SkPMFloat_SSE2.h b/src/opts/SkPMFloat_SSE2.h
new file mode 100644
index 0000000000..7bacf56af7
--- /dev/null
+++ b/src/opts/SkPMFloat_SSE2.h
@@ -0,0 +1,36 @@
+#include "SkColorPriv.h"
+#include "SkPMFloat.h"
+#include <emmintrin.h>
+
+// For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bits (fix8_16),
+// then widen those to 8-bit-in-32-bits (fix8_32), convert those to floats (scaled),
+// then finally scale those down from [0.0f, 255.0f] to [0.0f, 1.0f] into fColor.
+
+// get() and clamped() do the opposite, working from [0.0f, 1.0f] floats to [0.0f, 255.0f],
+// to 8-bit-in-32-bit, to 8-bit-in-16-bit, back down to 8-bit components.
+// _mm_packus_epi16() gives us clamping for free while narrowing.
+
+inline void SkPMFloat::set(SkPMColor c) {
+    SkPMColorAssert(c);
+    __m128i fix8    = _mm_set_epi32(0,0,0,c),
+            fix8_16 = _mm_unpacklo_epi8 (fix8,    _mm_setzero_si128()),
+            fix8_32 = _mm_unpacklo_epi16(fix8_16, _mm_setzero_si128());
+    __m128  scaled  = _mm_cvtepi32_ps(fix8_32);
+    _mm_store_ps(fColor, _mm_mul_ps(scaled, _mm_set1_ps(1.0f/255.0f)));
+    SkASSERT(this->isValid());
+}
+
+inline SkPMColor SkPMFloat::get() const {
+    SkASSERT(this->isValid());
+    return this->clamped();  // At the moment, we don't know anything faster.
+}
+
+inline SkPMColor SkPMFloat::clamped() const {
+    __m128  scaled  = _mm_mul_ps(_mm_load_ps(fColor), _mm_set1_ps(255.0f));
+    __m128i fix8_32 = _mm_cvtps_epi32(scaled),
+            fix8_16 = _mm_packus_epi16(fix8_32, fix8_32),
+            fix8    = _mm_packus_epi16(fix8_16, fix8_16);
+    SkPMColor c = _mm_cvtsi128_si32(fix8);
+    SkPMColorAssert(c);
+    return c;
+}
diff --git a/src/opts/SkPMFloat_neon.h b/src/opts/SkPMFloat_neon.h
new file mode 100644
index 0000000000..f3d265573d
--- /dev/null
+++ b/src/opts/SkPMFloat_neon.h
@@ -0,0 +1,42 @@
+#include "SkColorPriv.h"
+#include "SkPMFloat.h"
+#include <arm_neon.h>
+
+// For set(), we widen our 8 bit components (fix8) to 8-bit components in 16 bits (fix8_16),
+// then widen those to 8-bit-in-32-bits (fix8_32), convert those to floats (scaled),
+// then finally scale those down from [0.0f, 255.0f] to [0.0f, 1.0f] into fColor.
+
+// get() and clamped() do the opposite, working from [0.0f, 1.0f] floats to [0.0f, 255.0f],
+// to 8-bit-in-32-bit, to 8-bit-in-16-bit, back down to 8-bit components.
+// clamped() uses vqmovn to clamp while narrowing instead of just narrowing with vmovn.
+
+inline void SkPMFloat::set(SkPMColor c) {
+    SkPMColorAssert(c);
+    uint8x8_t   fix8    = (uint8x8_t)vdup_n_u32(c);
+    uint16x8_t  fix8_16 = vmovl_u8(fix8);
+    uint32x4_t  fix8_32 = vmovl_u16(vget_low_u16(fix8_16));
+    float32x4_t scaled  = vcvtq_f32_u32(fix8_32);
+    vst1q_f32(fColor, vmulq_f32(scaled, vdupq_n_f32(1.0f/255.0f)));
+    SkASSERT(this->isValid());
+}
+
+inline SkPMColor SkPMFloat::get() const {
+    SkASSERT(this->isValid());
+    float32x4_t scaled  = vmulq_f32(vld1q_f32(fColor), vdupq_n_f32(255.0f));
+    uint32x4_t  fix8_32 = vcvtq_u32_f32(scaled);
+    uint16x4_t  fix8_16 = vmovn_u32(fix8_32);
+    uint8x8_t   fix8    = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
+    SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0);
+    SkPMColorAssert(c);
+    return c;
+}
+
+inline SkPMColor SkPMFloat::clamped() const {
+    float32x4_t scaled  = vmulq_f32(vld1q_f32(fColor), vdupq_n_f32(255.0f));
+    uint32x4_t  fix8_32 = vcvtq_u32_f32(scaled);
+    uint16x4_t  fix8_16 = vqmovn_u32(fix8_32);
+    uint8x8_t   fix8    = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
+    SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0);
+    SkPMColorAssert(c);
+    return c;
+}
diff --git a/src/opts/SkPMFloat_none.h b/src/opts/SkPMFloat_none.h
new file mode 100644
index 0000000000..dee7ce6862
--- /dev/null
+++ b/src/opts/SkPMFloat_none.h
@@ -0,0 +1,28 @@
+#include "SkColorPriv.h"
+#include "SkPMFloat.h"
+
+inline void SkPMFloat::set(SkPMColor c) {
+    float scale = 1.0f / 255.0f;
+    this->setA(SkGetPackedA32(c) * scale);
+    this->setR(SkGetPackedR32(c) * scale);
+    this->setG(SkGetPackedG32(c) * scale);
+    this->setB(SkGetPackedB32(c) * scale);
+    SkASSERT(this->isValid());
+}
+
+inline SkPMColor SkPMFloat::get() const {
+    SkASSERT(this->isValid());
+    return SkPackARGB32(this->a() * 255, this->r() * 255, this->g() * 255, this->b() * 255);
+}
+
+inline SkPMColor SkPMFloat::clamped() const {
+    float a = this->a(),
+          r = this->r(),
+          g = this->g(),
+          b = this->b();
+    a = a < 0 ? 0 : (a > 1 ? 1 : a);
+    r = r < 0 ? 0 : (r > 1 ? 1 : r);
+    g = g < 0 ? 0 : (g > 1 ? 1 : g);
+    b = b < 0 ? 0 : (b > 1 ? 1 : b);
+    return SkPackARGB32(a * 255, r * 255, g * 255, b * 255);
+}