1 | // Copyright (C) 2020 The Qt Company Ltd. |
2 | // SPDX-License-Identifier: LicenseRef-Qt-Commercial OR LGPL-3.0-only OR GPL-2.0-only OR GPL-3.0-only |
3 | |
4 | #ifndef QRGBA64_P_H |
5 | #define QRGBA64_P_H |
6 | |
7 | // |
8 | // W A R N I N G |
9 | // ------------- |
10 | // |
11 | // This file is not part of the Qt API. It exists purely as an |
12 | // implementation detail. This header file may change from version to |
13 | // version without notice, or even be removed. |
14 | // |
15 | // We mean it. |
16 | // |
17 | |
18 | #include "qrgba64.h" |
19 | #include "qdrawhelper_p.h" |
20 | |
21 | #include <QtCore/private/qsimd_p.h> |
22 | #include <QtGui/private/qtguiglobal_p.h> |
23 | |
24 | QT_BEGIN_NAMESPACE |
25 | |
26 | inline QRgba64 combineAlpha256(QRgba64 rgba64, uint alpha256) |
27 | { |
28 | return QRgba64::fromRgba64(red: rgba64.red(), green: rgba64.green(), blue: rgba64.blue(), alpha: (rgba64.alpha() * alpha256) >> 8); |
29 | } |
30 | |
31 | #if defined(__SSE2__) |
32 | static inline __m128i Q_DECL_VECTORCALL multiplyAlpha65535(__m128i rgba64, __m128i va) |
33 | { |
34 | __m128i vs = rgba64; |
35 | vs = _mm_unpacklo_epi16(a: _mm_mullo_epi16(a: vs, b: va), b: _mm_mulhi_epu16(a: vs, b: va)); |
36 | vs = _mm_add_epi32(a: vs, b: _mm_srli_epi32(a: vs, count: 16)); |
37 | vs = _mm_add_epi32(a: vs, b: _mm_set1_epi32(i: 0x8000)); |
38 | vs = _mm_srai_epi32(a: vs, count: 16); |
39 | vs = _mm_packs_epi32(a: vs, b: vs); |
40 | return vs; |
41 | } |
42 | static inline __m128i Q_DECL_VECTORCALL multiplyAlpha65535(__m128i rgba64, uint alpha65535) |
43 | { |
44 | const __m128i va = _mm_shufflelo_epi16(_mm_cvtsi32_si128(alpha65535), _MM_SHUFFLE(0, 0, 0, 0)); |
45 | return multiplyAlpha65535(rgba64, va); |
46 | } |
47 | #elif defined(__ARM_NEON__) |
48 | static inline uint16x4_t multiplyAlpha65535(uint16x4_t rgba64, uint16x4_t alpha65535) |
49 | { |
50 | uint32x4_t vs32 = vmull_u16(rgba64, alpha65535); // vs = vs * alpha |
51 | vs32 = vsraq_n_u32(vs32, vs32, 16); // vs = vs + (vs >> 16) |
52 | return vrshrn_n_u32(vs32, 16); // vs = (vs + 0x8000) >> 16 |
53 | } |
54 | static inline uint16x4_t multiplyAlpha65535(uint16x4_t rgba64, uint alpha65535) |
55 | { |
56 | uint32x4_t vs32 = vmull_n_u16(rgba64, alpha65535); // vs = vs * alpha |
57 | vs32 = vsraq_n_u32(vs32, vs32, 16); // vs = vs + (vs >> 16) |
58 | return vrshrn_n_u32(vs32, 16); // vs = (vs + 0x8000) >> 16 |
59 | } |
60 | #endif |
61 | |
62 | static inline QRgba64 multiplyAlpha65535(QRgba64 rgba64, uint alpha65535) |
63 | { |
64 | #if defined(__SSE2__) |
65 | const __m128i v = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&rgba64)); |
66 | const __m128i vr = multiplyAlpha65535(rgba64: v, alpha65535); |
67 | QRgba64 r; |
68 | _mm_storel_epi64(p: reinterpret_cast<__m128i *>(&r), a: vr); |
69 | return r; |
70 | #elif defined(__ARM_NEON__) |
71 | const uint16x4_t v = vreinterpret_u16_u64(vld1_u64(reinterpret_cast<const uint64_t *>(&rgba64))); |
72 | const uint16x4_t vr = multiplyAlpha65535(v, alpha65535); |
73 | QRgba64 r; |
74 | vst1_u64(reinterpret_cast<uint64_t *>(&r), vreinterpret_u64_u16(vr)); |
75 | return r; |
76 | #else |
77 | return QRgba64::fromRgba64(qt_div_65535(rgba64.red() * alpha65535), |
78 | qt_div_65535(rgba64.green() * alpha65535), |
79 | qt_div_65535(rgba64.blue() * alpha65535), |
80 | qt_div_65535(rgba64.alpha() * alpha65535)); |
81 | #endif |
82 | } |
83 | |
84 | #if defined(__SSE2__) || defined(__ARM_NEON__) |
85 | template<typename T> |
86 | static inline T Q_DECL_VECTORCALL multiplyAlpha255(T rgba64, uint alpha255) |
87 | { |
88 | return multiplyAlpha65535(rgba64, alpha255 * 257); |
89 | } |
90 | #else |
91 | template<typename T> |
92 | static inline T multiplyAlpha255(T rgba64, uint alpha255) |
93 | { |
94 | return QRgba64::fromRgba64(qt_div_255(rgba64.red() * alpha255), |
95 | qt_div_255(rgba64.green() * alpha255), |
96 | qt_div_255(rgba64.blue() * alpha255), |
97 | qt_div_255(rgba64.alpha() * alpha255)); |
98 | } |
99 | #endif |
100 | |
101 | #if defined __SSE2__ |
102 | static inline __m128i Q_DECL_VECTORCALL interpolate255(__m128i x, uint alpha1, __m128i y, uint alpha2) |
103 | { |
104 | return _mm_add_epi16(a: multiplyAlpha255(rgba64: x, alpha255: alpha1), b: multiplyAlpha255(rgba64: y, alpha255: alpha2)); |
105 | } |
106 | #endif |
107 | |
108 | #if defined __ARM_NEON__ |
109 | inline uint16x4_t interpolate255(uint16x4_t x, uint alpha1, uint16x4_t y, uint alpha2) |
110 | { |
111 | return vadd_u16(multiplyAlpha255(x, alpha1), multiplyAlpha255(y, alpha2)); |
112 | } |
113 | #endif |
114 | |
115 | static inline QRgba64 interpolate255(QRgba64 x, uint alpha1, QRgba64 y, uint alpha2) |
116 | { |
117 | #if defined(__SSE2__) |
118 | const __m128i vx = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&x)); |
119 | const __m128i vy = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&y)); |
120 | const __m128i vr = interpolate255(x: vx, alpha1, y: vy, alpha2); |
121 | QRgba64 r; |
122 | _mm_storel_epi64(p: reinterpret_cast<__m128i *>(&r), a: vr); |
123 | return r; |
124 | #elif defined(__ARM_NEON__) |
125 | const uint16x4_t vx = vreinterpret_u16_u64(vld1_u64(reinterpret_cast<const uint64_t *>(&x))); |
126 | const uint16x4_t vy = vreinterpret_u16_u64(vld1_u64(reinterpret_cast<const uint64_t *>(&y))); |
127 | const uint16x4_t vr = interpolate255(vx, alpha1, vy, alpha2); |
128 | QRgba64 r; |
129 | vst1_u64(reinterpret_cast<uint64_t *>(&r), vreinterpret_u64_u16(vr)); |
130 | return r; |
131 | #else |
132 | return QRgba64::fromRgba64(multiplyAlpha255(x, alpha1) + multiplyAlpha255(y, alpha2)); |
133 | #endif |
134 | } |
135 | |
136 | #if defined __SSE2__ |
137 | static inline __m128i Q_DECL_VECTORCALL interpolate65535(__m128i x, uint alpha1, __m128i y, uint alpha2) |
138 | { |
139 | return _mm_add_epi16(a: multiplyAlpha65535(rgba64: x, alpha65535: alpha1), b: multiplyAlpha65535(rgba64: y, alpha65535: alpha2)); |
140 | } |
141 | |
142 | static inline __m128i Q_DECL_VECTORCALL interpolate65535(__m128i x, __m128i alpha1, __m128i y, __m128i alpha2) |
143 | { |
144 | return _mm_add_epi16(a: multiplyAlpha65535(rgba64: x, va: alpha1), b: multiplyAlpha65535(rgba64: y, va: alpha2)); |
145 | } |
146 | #endif |
147 | |
148 | #if defined __ARM_NEON__ |
149 | inline uint16x4_t interpolate65535(uint16x4_t x, uint alpha1, uint16x4_t y, uint alpha2) |
150 | { |
151 | return vadd_u16(multiplyAlpha65535(x, alpha1), multiplyAlpha65535(y, alpha2)); |
152 | } |
153 | inline uint16x4_t interpolate65535(uint16x4_t x, uint16x4_t alpha1, uint16x4_t y, uint16x4_t alpha2) |
154 | { |
155 | return vadd_u16(multiplyAlpha65535(x, alpha1), multiplyAlpha65535(y, alpha2)); |
156 | } |
157 | #endif |
158 | |
159 | static inline QRgba64 interpolate65535(QRgba64 x, uint alpha1, QRgba64 y, uint alpha2) |
160 | { |
161 | #if defined(__SSE2__) |
162 | const __m128i vx = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&x)); |
163 | const __m128i vy = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&y)); |
164 | const __m128i vr = interpolate65535(x: vx, alpha1, y: vy, alpha2); |
165 | QRgba64 r; |
166 | _mm_storel_epi64(p: reinterpret_cast<__m128i *>(&r), a: vr); |
167 | return r; |
168 | #elif defined(__ARM_NEON__) |
169 | const uint16x4_t vx = vreinterpret_u16_u64(vld1_u64(reinterpret_cast<const uint64_t *>(&x))); |
170 | const uint16x4_t vy = vreinterpret_u16_u64(vld1_u64(reinterpret_cast<const uint64_t *>(&y))); |
171 | const uint16x4_t vr = interpolate65535(vx, alpha1, vy, alpha2); |
172 | QRgba64 r; |
173 | vst1_u64(reinterpret_cast<uint64_t *>(&r), vreinterpret_u64_u16(vr)); |
174 | return r; |
175 | #else |
176 | return QRgba64::fromRgba64(multiplyAlpha65535(x, alpha1) + multiplyAlpha65535(y, alpha2)); |
177 | #endif |
178 | } |
179 | |
180 | static inline QRgba64 addWithSaturation(QRgba64 a, QRgba64 b) |
181 | { |
182 | #if defined(__SSE2__) |
183 | const __m128i va = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&a)); |
184 | const __m128i vb = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&b)); |
185 | const __m128i vr = _mm_adds_epu16(a: va, b: vb); |
186 | QRgba64 r; |
187 | _mm_storel_epi64(p: reinterpret_cast<__m128i *>(&r), a: vr); |
188 | return r; |
189 | #else |
190 | return QRgba64::fromRgba64(qMin(a.red() + b.red(), 65535), |
191 | qMin(a.green() + b.green(), 65535), |
192 | qMin(a.blue() + b.blue(), 65535), |
193 | qMin(a.alpha() + b.alpha(), 65535)); |
194 | #endif |
195 | } |
196 | |
197 | #if QT_COMPILER_SUPPORTS_HERE(SSE2) |
198 | QT_FUNCTION_TARGET(SSE2) |
199 | static inline uint Q_DECL_VECTORCALL toArgb32(__m128i v) |
200 | { |
201 | v = _mm_unpacklo_epi16(a: v, b: _mm_setzero_si128()); |
202 | v = _mm_add_epi32(a: v, b: _mm_set1_epi32(i: 128)); |
203 | v = _mm_sub_epi32(a: v, b: _mm_srli_epi32(a: v, count: 8)); |
204 | v = _mm_srli_epi32(a: v, count: 8); |
205 | v = _mm_packs_epi32(a: v, b: v); |
206 | v = _mm_packus_epi16(a: v, b: v); |
207 | return _mm_cvtsi128_si32(a: v); |
208 | } |
209 | #elif defined __ARM_NEON__ |
210 | static inline uint toArgb32(uint16x4_t v) |
211 | { |
212 | v = vsub_u16(v, vrshr_n_u16(v, 8)); |
213 | v = vrshr_n_u16(v, 8); |
214 | uint8x8_t v8 = vmovn_u16(vcombine_u16(v, v)); |
215 | return vget_lane_u32(vreinterpret_u32_u8(v8), 0); |
216 | } |
217 | #endif |
218 | |
219 | static inline uint toArgb32(QRgba64 rgba64) |
220 | { |
221 | #if defined __SSE2__ |
222 | __m128i v = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&rgba64)); |
223 | v = _mm_shufflelo_epi16(v, _MM_SHUFFLE(3, 0, 1, 2)); |
224 | return toArgb32(v); |
225 | #elif defined __ARM_NEON__ |
226 | uint16x4_t v = vreinterpret_u16_u64(vld1_u64(reinterpret_cast<const uint64_t *>(&rgba64))); |
227 | #if Q_BYTE_ORDER == Q_LITTLE_ENDIAN |
228 | const uint8x8_t shuffleMask = { 4, 5, 2, 3, 0, 1, 6, 7 }; |
229 | v = vreinterpret_u16_u8(vtbl1_u8(vreinterpret_u8_u16(v), shuffleMask)); |
230 | #else |
231 | v = vext_u16(v, v, 3); |
232 | #endif |
233 | return toArgb32(v); |
234 | #else |
235 | return rgba64.toArgb32(); |
236 | #endif |
237 | } |
238 | |
239 | static inline uint toRgba8888(QRgba64 rgba64) |
240 | { |
241 | #if defined __SSE2__ |
242 | __m128i v = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&rgba64)); |
243 | return toArgb32(v); |
244 | #elif defined __ARM_NEON__ |
245 | uint16x4_t v = vreinterpret_u16_u64(vld1_u64(reinterpret_cast<const uint64_t *>(&rgba64))); |
246 | return toArgb32(v); |
247 | #else |
248 | return ARGB2RGBA(toArgb32(rgba64)); |
249 | #endif |
250 | } |
251 | |
252 | static inline QRgba64 rgbBlend(QRgba64 d, QRgba64 s, uint rgbAlpha) |
253 | { |
254 | QRgba64 blend; |
255 | #if defined(__SSE2__) |
256 | __m128i vd = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&d)); |
257 | __m128i vs = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&s)); |
258 | __m128i va = _mm_cvtsi32_si128(a: rgbAlpha); |
259 | va = _mm_unpacklo_epi8(a: va, b: va); |
260 | va = _mm_shufflelo_epi16(va, _MM_SHUFFLE(3, 0, 1, 2)); |
261 | __m128i vb = _mm_xor_si128(a: _mm_set1_epi16(w: -1), b: va); |
262 | |
263 | vs = _mm_unpacklo_epi16(a: _mm_mullo_epi16(a: vs, b: va), b: _mm_mulhi_epu16(a: vs, b: va)); |
264 | vd = _mm_unpacklo_epi16(a: _mm_mullo_epi16(a: vd, b: vb), b: _mm_mulhi_epu16(a: vd, b: vb)); |
265 | vd = _mm_add_epi32(a: vd, b: vs); |
266 | vd = _mm_add_epi32(a: vd, b: _mm_srli_epi32(a: vd, count: 16)); |
267 | vd = _mm_add_epi32(a: vd, b: _mm_set1_epi32(i: 0x8000)); |
268 | vd = _mm_srai_epi32(a: vd, count: 16); |
269 | vd = _mm_packs_epi32(a: vd, b: vd); |
270 | |
271 | _mm_storel_epi64(p: reinterpret_cast<__m128i *>(&blend), a: vd); |
272 | #elif defined(__ARM_NEON__) |
273 | uint16x4_t vd = vreinterpret_u16_u64(vmov_n_u64(d)); |
274 | uint16x4_t vs = vreinterpret_u16_u64(vmov_n_u64(s)); |
275 | uint8x8_t va8 = vreinterpret_u8_u32(vmov_n_u32(ARGB2RGBA(rgbAlpha))); |
276 | uint16x4_t va = vreinterpret_u16_u8(vzip_u8(va8, va8).val[0]); |
277 | uint16x4_t vb = vdup_n_u16(0xffff); |
278 | vb = vsub_u16(vb, va); |
279 | |
280 | uint32x4_t vs32 = vmull_u16(vs, va); |
281 | uint32x4_t vd32 = vmull_u16(vd, vb); |
282 | vd32 = vaddq_u32(vd32, vs32); |
283 | vd32 = vsraq_n_u32(vd32, vd32, 16); |
284 | vd = vrshrn_n_u32(vd32, 16); |
285 | vst1_u64(reinterpret_cast<uint64_t *>(&blend), vreinterpret_u64_u16(vd)); |
286 | #else |
287 | const int mr = qRed(rgbAlpha); |
288 | const int mg = qGreen(rgbAlpha); |
289 | const int mb = qBlue(rgbAlpha); |
290 | blend = qRgba64(qt_div_255(s.red() * mr + d.red() * (255 - mr)), |
291 | qt_div_255(s.green() * mg + d.green() * (255 - mg)), |
292 | qt_div_255(s.blue() * mb + d.blue() * (255 - mb)), |
293 | s.alpha()); |
294 | #endif |
295 | return blend; |
296 | } |
297 | |
298 | static inline void blend_pixel(QRgba64 &dst, QRgba64 src) |
299 | { |
300 | if (src.isOpaque()) |
301 | dst = src; |
302 | else if (!src.isTransparent()) { |
303 | #if defined(__SSE2__) |
304 | const __m128i vd = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&dst)); |
305 | const __m128i vs = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&src)); |
306 | const __m128i via = _mm_xor_si128(a: _mm_set1_epi16(w: -1), _mm_shufflelo_epi16(vs, _MM_SHUFFLE(3, 3, 3, 3))); |
307 | const __m128i vr = _mm_add_epi16(a: vs, b: multiplyAlpha65535(rgba64: vd, va: via)); |
308 | _mm_storel_epi64(p: reinterpret_cast<__m128i *>(&dst), a: vr); |
309 | #else |
310 | dst = src + multiplyAlpha65535(dst, 65535 - src.alpha()); |
311 | #endif |
312 | } |
313 | } |
314 | |
315 | static inline void blend_pixel(QRgba64 &dst, QRgba64 src, const int const_alpha) |
316 | { |
317 | if (const_alpha == 255) |
318 | return blend_pixel(dst, src); |
319 | if (!src.isTransparent()) { |
320 | #if defined(__SSE2__) |
321 | const __m128i vd = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&dst)); |
322 | __m128i vs = _mm_loadl_epi64(p: reinterpret_cast<const __m128i *>(&src)); |
323 | vs = multiplyAlpha255(rgba64: vs, alpha255: const_alpha); |
324 | const __m128i via = _mm_xor_si128(a: _mm_set1_epi16(w: -1), _mm_shufflelo_epi16(vs, _MM_SHUFFLE(3, 3, 3, 3))); |
325 | const __m128i vr = _mm_add_epi16(a: vs, b: multiplyAlpha65535(rgba64: vd, va: via)); |
326 | _mm_storel_epi64(p: reinterpret_cast<__m128i *>(&dst), a: vr); |
327 | #else |
328 | src = multiplyAlpha255(src, const_alpha); |
329 | dst = src + multiplyAlpha65535(dst, 65535 - src.alpha()); |
330 | #endif |
331 | } |
332 | } |
333 | |
334 | QT_END_NAMESPACE |
335 | |
336 | #endif // QRGBA64_P_H |
337 | |