static void srcover_n(const SkXfermode*, uint64_t dst[], const SkPM4f src[], int count,
const SkAlpha aa[]) {
for (int i = 0; i < count; ++i) {
Sk4f s = Sk4f::Load(src+i),
d = SkHalfToFloat_finite_ftz(dst[i]),
r = s + d*(1.0f - SkNx_shuffle<3,3,3,3>(s));
if (aa) {
r = lerp_by_coverage(r, d, aa[i]);
}
SkFloatToHalf_finite_ftz(r).store(&dst[i]);
}
}
template <DstType D> void src_1(const SkXfermode::U64State& state, uint64_t dst[],
const SkPM4f& src, int count, const SkAlpha aa[]) {
const Sk4f s4 = pm_to_rgba_order(unit_to_bias<D>(Sk4f::Load(src.fVec)));
if (aa) {
for (int i = 0; i < count; ++i) {
const Sk4f d4 = load_from_dst<D>(dst[i]);
dst[i] = store_to_dst<D>(lerp_by_coverage(s4, d4, aa[i]));
}
} else {
sk_memset64(dst, store_to_dst<D>(s4), count);
}
}
static void src_1(SkBlendMode, uint64_t dst[], const SkPM4f* src, int count, const SkAlpha aa[]) {
const Sk4f s4 = Sk4f::Load(src->fVec);
if (aa) {
for (int i = 0; i < count; ++i) {
const Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]);
SkFloatToHalf_finite_ftz(lerp_by_coverage(s4, d4, aa[i])).store(&dst[i]);
}
} else {
uint64_t s4h;
SkFloatToHalf_finite_ftz(s4).store(&s4h);
sk_memset64(dst, s4h, count);
}
}
static void srcover_1(const SkXfermode*, uint64_t dst[], const SkPM4f* src, int count,
const SkAlpha aa[]) {
const Sk4f s4 = Sk4f::Load(src->fVec);
const Sk4f dst_scale = Sk4f(1 - get_alpha(s4));
for (int i = 0; i < count; ++i) {
const Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]);
const Sk4f r4 = s4 + d4 * dst_scale;
if (aa) {
SkFloatToHalf_finite_ftz(lerp_by_coverage(r4, d4, aa[i])).store(&dst[i]);
} else {
SkFloatToHalf_finite_ftz(r4).store(&dst[i]);
}
}
}
static void src_n(SkBlendMode, uint64_t dst[], const SkPM4f src[], int count, const SkAlpha aa[]) {
if (aa) {
for (int i = 0; i < count; ++i) {
const Sk4f s4 = Sk4f::Load(src[i].fVec);
const Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]);
SkFloatToHalf_finite_ftz(lerp_by_coverage(s4, d4, aa[i])).store(&dst[i]);
}
} else {
for (int i = 0; i < count; ++i) {
const Sk4f s4 = Sk4f::Load(src[i].fVec);
SkFloatToHalf_finite_ftz(s4).store(&dst[i]);
}
}
}
template <DstType D> void srcover_n(const SkXfermode::U64State& state, uint64_t dst[],
const SkPM4f src[], int count, const SkAlpha aa[]) {
for (int i = 0; i < count; ++i) {
const Sk4f s4 = pm_to_rgba_order(Sk4f::Load(src[i].fVec));
const Sk4f dst_scale = Sk4f(1 - get_alpha(s4));
const Sk4f s4bias = unit_to_bias<D>(s4);
const Sk4f d4bias = load_from_dst<D>(dst[i]);
const Sk4f r4bias = s4bias + d4bias * dst_scale;
if (aa) {
dst[i] = store_to_dst<D>(lerp_by_coverage(r4bias, d4bias, aa[i]));
} else {
dst[i] = store_to_dst<D>(r4bias);
}
}
}
static void xfer_n(const SkXfermode* xfer, uint64_t dst[], const SkPM4f src[], int count,
const SkAlpha aa[]) {
SkXfermodeProc4f proc = xfer->getProc4f();
SkPM4f d;
if (aa) {
for (int i = 0; i < count; ++i) {
Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]);
d4.store(d.fVec);
Sk4f r4 = Sk4f::Load(proc(src[i], d).fVec);
SkFloatToHalf_finite_ftz(lerp_by_coverage(r4, d4, aa[i])).store(&dst[i]);
}
} else {
for (int i = 0; i < count; ++i) {
SkHalfToFloat_finite_ftz(dst[i]).store(d.fVec);
Sk4f r4 = Sk4f::Load(proc(src[i], d).fVec);
SkFloatToHalf_finite_ftz(r4).store(&dst[i]);
}
}
}
static void xfer_1(SkBlendMode mode, uint64_t dst[], const SkPM4f* src, int count,
const SkAlpha aa[]) {
SkXfermodeProc4f proc = SkXfermode::GetProc4f(mode);
SkPM4f d;
if (aa) {
for (int i = 0; i < count; ++i) {
Sk4f d4 = SkHalfToFloat_finite_ftz(dst[i]);
d4.store(d.fVec);
Sk4f r4 = Sk4f::Load(proc(*src, d).fVec);
SkFloatToHalf_finite_ftz(lerp_by_coverage(r4, d4, aa[i])).store(&dst[i]);
}
} else {
for (int i = 0; i < count; ++i) {
SkHalfToFloat_finite_ftz(dst[i]).store(d.fVec);
Sk4f r4 = Sk4f::Load(proc(*src, d).fVec);
SkFloatToHalf_finite_ftz(r4).store(&dst[i]);
}
}
}
template <DstType D> void xfer_u64_n(const SkXfermode::U64State& state, uint64_t dst[],
const SkPM4f src[], int count, const SkAlpha aa[]) {
SkXfermodeProc4f proc = state.fXfer->getProc4f();
SkPM4f d;
if (aa) {
for (int i = 0; i < count; ++i) {
Sk4f d4 = bias_to_unit<D>(load_from_dst<D>(dst[i]));
d4.store(d.fVec);
Sk4f r4 = unit_to_bias<D>(Sk4f::Load(proc(src[i], d).fVec));
dst[i] = store_to_dst<D>(lerp_by_coverage(r4, d4, aa[i]));
}
} else {
for (int i = 0; i < count; ++i) {
bias_to_unit<D>(load_from_dst<D>(dst[i])).store(d.fVec);
Sk4f r4 = unit_to_bias<D>(Sk4f::Load(proc(src[i], d).fVec));
dst[i] = store_to_dst<D>(r4);
}
}
}
