Sk4px

Xfermode_SrcOver:
   SSE: 2.08ms -> 2.03ms  (~2% faster)
   NEON: my N5 is noisy, but there appears to be no perf change

BUG=skia:

Review URL: https://codereview.chromium.org/1132273004

1 files changed, 57 insertions, 0 deletions

diff --git a/src/opts/Sk4px_none.h b/src/opts/Sk4px_none.h
new file mode 100644
index 0000000000..c8c33a0d16
--- /dev/null
+++ b/src/opts/Sk4px_none.h
@@ -0,0 +1,57 @@
+/*
+ * Copyright 2015 Google Inc.
+ *
+ * Use of this source code is governed by a BSD-style license that can be
+ * found in the LICENSE file.
+ */
+
+#include "SkUtils.h"
+
+static_assert(sizeof(Sk4px) == 16, "This file uses memcpy / sk_memset32, so exact size matters.");
+
+inline Sk4px::Sk4px(SkPMColor px) {
+    sk_memset32((uint32_t*)this, px, 4);
+}
+
+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]) {
+    Sk4px px2 = Sk16b();
+    memcpy(&px2, px, 8);
+    return px2;
+}
+
+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); }
+
+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; }
+
+inline Sk4px::Wide Sk4px::mulWiden(const Sk16b& other) const {
+    return this->widenLo() * Sk4px(other).widenLo();
+}
+
+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>(),
+                 r.kth< 8>(), r.kth< 9>(), r.kth<10>(), r.kth<11>(),
+                 r.kth<12>(), r.kth<13>(), r.kth<14>(), r.kth<15>());
+}