div255(x) as ((x+128)*257)>>16 with SSE
_mm_mulhi_epu16 makes the (...*257)>>16 part simple.
This seems to speed up every transfermode that uses div255(),
in the 7-25% range.
It even appears to obviate the need for approxMulDiv255() on SSE.
I'm not sure about NEON yet, so I'll keep approxMulDiv255() for now.
Should be no pixels change:
https://gold.skia.org/search2?issue=1452903004&unt=true&query=source_type%3Dgm&master=false
BUG=skia:
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot
Review URL: https://codereview.chromium.org/1452903004
diff --git a/src/opts/Sk4px_none.h b/src/opts/Sk4px_none.h
index 540edb8..efbd780 100644
--- a/src/opts/Sk4px_none.h
+++ b/src/opts/Sk4px_none.h
@@ -62,6 +62,12 @@
r.kth<12>(), r.kth<13>(), r.kth<14>(), r.kth<15>());
}
+inline Sk4px Sk4px::Wide::div255() const {
+ // Calculated as ((x+128) + ((x+128)>>8)) >> 8.
+ auto v = *this + Sk16h(128);
+ return v.addNarrowHi(v>>8);
+}
+
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>(),