Clean up SkXfermode_opts.h

It seems that MSVC + __vectorcall don't play well together,
so back ourselves out into a situation where we don't need it.

   - Inline transfermode functions.  This removes the need for SK_VECTORCALL.
   - Remove 565 destination specializations.
     Blending into 565 is not speed-critical enough to merit the code bloat.
   - Removing 565 specializations means a bunch of Sk4px code is now dead.

8888 xfermodes generally speed up a bit from inlining, smoothly ranging from no change down to 0.65x for the fastest functions like Plus or Modulate.

565 xfermodes generally slow down because we're doing 565 -> 8888 and 8888->565 conversion serially[1] and using the stack, smoothly ranging from no change up to 2x slower for the fastest functions like Plus and Modulate.

[1] the 565->8888 conversion is actually being autovectorized

BUG=skia:4765,skia:4776
GOLD_TRYBOT_URL= https://gold.skia.org/search2?unt=true&query=source_type%3Dgm&master=false&issue=1565223002
CQ_EXTRA_TRYBOTS=client.skia:Test-Ubuntu-GCC-GCE-CPU-AVX2-x86_64-Release-SKNX_NO_SIMD-Trybot

No public API changes.
TBR=reed@google.com

Review URL: https://codereview.chromium.org/1565223002

1 files changed, 0 insertions, 71 deletions

diff --git a/src/opts/Sk4px_NEON.h b/src/opts/Sk4px_NEON.h
index c8abff24b1..62f1deb4ac 100644
--- a/src/opts/Sk4px_NEON.h
+++ b/src/opts/Sk4px_NEON.h
@@ -95,76 +95,5 @@ inline Sk4px Sk4px::zeroAlphas() const {
     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) {
-    // RGB565 format:   |R....|G.....|B....|
-    //           Bit:  16    11      5     0
-
-    // First get each pixel into its own 32-bit lane.
-    //      v ==                       rgb3 rgb2  rgb1 rgb0
-    // spread == 0000 rgb3  0000 rgb2  0000 rgb1  0000 rgb0
-    uint32x4_t spread = vmovl_u16(v);
-
-    // Get each color independently, still in 565 precison but down at bit 0.
-    auto r5 = vshrq_n_u32(spread, 11),
-         g6 = vandq_u32(vdupq_n_u32(63), vshrq_n_u32(spread, 5)),
-         b5 = vandq_u32(vdupq_n_u32(31), spread);
-
-    // Scale 565 precision up to 8-bit each, filling low 323 bits with high bits of each component.
-    auto r8 = vorrq_u32(vshlq_n_u32(r5, 3), vshrq_n_u32(r5, 2)),
-         g8 = vorrq_u32(vshlq_n_u32(g6, 2), vshrq_n_u32(g6, 4)),
-         b8 = vorrq_u32(vshlq_n_u32(b5, 3), vshrq_n_u32(b5, 2));
-
-    // Now put all the 8-bit components into SkPMColor order.
-    return (uint8x16_t)vorrq_u32(vshlq_n_u32(r8, SK_R32_SHIFT),   // TODO: one shift is zero...
-                       vorrq_u32(vshlq_n_u32(g8, SK_G32_SHIFT),
-                       vorrq_u32(vshlq_n_u32(b8, SK_B32_SHIFT),
-                                 vdupq_n_u32(0xFF << SK_A32_SHIFT))));
-}
-
-static 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.
-    auto r5 = vandq_u32(vdupq_n_u32(31), vshrq_n_u32(w, SK_R32_SHIFT + 3)),
-         g6 = vandq_u32(vdupq_n_u32(63), vshrq_n_u32(w, SK_G32_SHIFT + 2)),
-         b5 = vandq_u32(vdupq_n_u32(31), vshrq_n_u32(w, SK_B32_SHIFT + 3));
-
-    // Now put the bits in place in the low 16-bits of each 32-bit lane.
-    auto spread = vorrq_u32(vshlq_n_u32(r5, 11),
-                  vorrq_u32(vshlq_n_u32(g6,  5),
-                            b5));
-
-    // Pack the low 16-bits of our 128-bit register down into a 64-bit register.
-    // spread == 0000 rgb3  0000 rgb2  0000 rgb1  0000 rgb0
-    //      v ==                       rgb3 rgb2  rgb1 rgb0
-    auto v = vmovn_u32(spread);
-    return v;
-}
-
-
-inline Sk4px Sk4px::Load4(const SkPMColor16 src[4]) {
-    return Sk16b(widen_to_8888(vld1_u16(src)));
-}
-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]) {
-    return Sk16b(widen_to_8888(vcreate_u16(src[0])));
-}
-
-inline void Sk4px::store4(SkPMColor16 dst[4]) const {
-    vst1_u16(dst, narrow_to_565(this->fVec));
-}
-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 {
-    dst[0] = vget_lane_u16(narrow_to_565(this->fVec), 0);
-}
-
 } // namespace