1 /*
2 * Copyright 2015 Google Inc.
3 *
4 * Use of this source code is governed by a BSD-style license that can be
5 * found in the LICENSE file.
6 */
7
8 // For SkPMFloat(SkPMFColor), we widen our 8 bit components (fix8) to 8-bit components in 16 bits
9 // (fix8_16), then widen those to 8-bit-in-32-bits (fix8_32), and finally convert those to floats.
10
11 // round() and roundClamp() do the opposite, working from floats to 8-bit-in-32-bit,
12 // to 8-bit-in-16-bit, back down to 8-bit components.
13 // roundClamp() uses vqmovn to clamp while narrowing instead of just narrowing with vmovn.
14
SkPMFloat(SkPMColor c)15 inline SkPMFloat::SkPMFloat(SkPMColor c) {
16 SkPMColorAssert(c);
17 uint8x8_t fix8 = (uint8x8_t)vdup_n_u32(c);
18 uint16x8_t fix8_16 = vmovl_u8(fix8);
19 uint32x4_t fix8_32 = vmovl_u16(vget_low_u16(fix8_16));
20 fVec = vcvtq_f32_u32(fix8_32);
21 SkASSERT(this->isValid());
22 }
23
trunc()24 inline SkPMColor SkPMFloat::trunc() const {
25 uint32x4_t fix8_32 = vcvtq_u32_f32(fVec); // vcvtq_u32_f32 truncates
26 uint16x4_t fix8_16 = vmovn_u32(fix8_32);
27 uint8x8_t fix8 = vmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
28 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0);
29 SkPMColorAssert(c);
30 return c;
31 }
32
round()33 inline SkPMColor SkPMFloat::round() const {
34 return SkPMFloat(Sk4f(0.5f) + *this).trunc();
35 }
36
roundClamp()37 inline SkPMColor SkPMFloat::roundClamp() const {
38 float32x4_t add_half = vaddq_f32(fVec, vdupq_n_f32(0.5f));
39 uint32x4_t fix8_32 = vcvtq_u32_f32(add_half); // vcvtq_u32_f32 truncates, so round manually
40 uint16x4_t fix8_16 = vqmovn_u32(fix8_32);
41 uint8x8_t fix8 = vqmovn_u16(vcombine_u16(fix8_16, vdup_n_u16(0)));
42 SkPMColor c = vget_lane_u32((uint32x2_t)fix8, 0);
43 SkPMColorAssert(c);
44 return c;
45 }
46
47 // TODO: we should be able to beat these loops on all three methods.
From4PMColors(const SkPMColor colors[4],SkPMFloat * a,SkPMFloat * b,SkPMFloat * c,SkPMFloat * d)48 inline void SkPMFloat::From4PMColors(const SkPMColor colors[4],
49 SkPMFloat* a, SkPMFloat* b, SkPMFloat* c, SkPMFloat* d) {
50 *a = FromPMColor(colors[0]);
51 *b = FromPMColor(colors[1]);
52 *c = FromPMColor(colors[2]);
53 *d = FromPMColor(colors[3]);
54 }
55
RoundTo4PMColors(const SkPMFloat & a,const SkPMFloat & b,const SkPMFloat & c,const SkPMFloat & d,SkPMColor colors[4])56 inline void SkPMFloat::RoundTo4PMColors(
57 const SkPMFloat& a, const SkPMFloat& b, const SkPMFloat&c, const SkPMFloat& d,
58 SkPMColor colors[4]) {
59 colors[0] = a.round();
60 colors[1] = b.round();
61 colors[2] = c.round();
62 colors[3] = d.round();
63 }
64
RoundClampTo4PMColors(const SkPMFloat & a,const SkPMFloat & b,const SkPMFloat & c,const SkPMFloat & d,SkPMColor colors[4])65 inline void SkPMFloat::RoundClampTo4PMColors(
66 const SkPMFloat& a, const SkPMFloat& b, const SkPMFloat&c, const SkPMFloat& d,
67 SkPMColor colors[4]) {
68 colors[0] = a.roundClamp();
69 colors[1] = b.roundClamp();
70 colors[2] = c.roundClamp();
71 colors[3] = d.roundClamp();
72 }
73