1 | // Copyright (C) 2016 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 QDRAWHELPER_P_H |
5 | #define QDRAWHELPER_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 <QtGui/private/qtguiglobal_p.h> |
19 | #include "QtCore/qmath.h" |
20 | #include "QtGui/qcolor.h" |
21 | #include "QtGui/qpainter.h" |
22 | #include "QtGui/qimage.h" |
23 | #include "QtGui/qrgba64.h" |
24 | #ifndef QT_FT_BEGIN_HEADER |
25 | #define |
26 | #define |
27 | #endif |
28 | #include "private/qpixellayout_p.h" |
29 | #include "private/qrasterdefs_p.h" |
30 | #include <private/qsimd_p.h> |
31 | |
32 | #include <memory> |
33 | |
34 | QT_BEGIN_NAMESPACE |
35 | |
36 | #if defined(Q_CC_GNU) |
37 | # define Q_DECL_RESTRICT __restrict__ |
38 | # if defined(Q_PROCESSOR_X86_32) && defined(Q_CC_GNU) && !defined(Q_CC_CLANG) |
39 | # define Q_DECL_VECTORCALL __attribute__((sseregparm,regparm(3))) |
40 | # else |
41 | # define Q_DECL_VECTORCALL |
42 | # endif |
43 | #elif defined(Q_CC_MSVC) |
44 | # define Q_DECL_RESTRICT __restrict |
45 | # define Q_DECL_VECTORCALL __vectorcall |
46 | #else |
47 | # define Q_DECL_RESTRICT |
48 | # define Q_DECL_VECTORCALL |
49 | #endif |
50 | |
51 | static const uint AMASK = 0xff000000; |
52 | static const uint RMASK = 0x00ff0000; |
53 | static const uint GMASK = 0x0000ff00; |
54 | static const uint BMASK = 0x000000ff; |
55 | |
56 | struct QSolidData; |
57 | struct QTextureData; |
58 | struct QGradientData; |
59 | struct QLinearGradientData; |
60 | struct QRadialGradientData; |
61 | struct QConicalGradientData; |
62 | struct QSpanData; |
63 | class QGradient; |
64 | class QRasterBuffer; |
65 | class QClipData; |
66 | class QRasterPaintEngineState; |
67 | |
68 | template<typename F> class QRgbaFloat; |
69 | typedef QRgbaFloat<float> QRgbaFloat32; |
70 | |
71 | typedef QT_FT_SpanFunc ProcessSpans; |
72 | typedef void (*BitmapBlitFunc)(QRasterBuffer *rasterBuffer, |
73 | int x, int y, const QRgba64 &color, |
74 | const uchar *bitmap, |
75 | int mapWidth, int mapHeight, int mapStride); |
76 | |
77 | typedef void (*AlphamapBlitFunc)(QRasterBuffer *rasterBuffer, |
78 | int x, int y, const QRgba64 &color, |
79 | const uchar *bitmap, |
80 | int mapWidth, int mapHeight, int mapStride, |
81 | const QClipData *clip, bool useGammaCorrection); |
82 | |
83 | typedef void (*AlphaRGBBlitFunc)(QRasterBuffer *rasterBuffer, |
84 | int x, int y, const QRgba64 &color, |
85 | const uint *rgbmask, |
86 | int mapWidth, int mapHeight, int mapStride, |
87 | const QClipData *clip, bool useGammaCorrection); |
88 | |
89 | typedef void (*RectFillFunc)(QRasterBuffer *rasterBuffer, |
90 | int x, int y, int width, int height, |
91 | const QRgba64 &color); |
92 | |
93 | typedef void (*SrcOverBlendFunc)(uchar *destPixels, int dbpl, |
94 | const uchar *src, int spbl, |
95 | int w, int h, |
96 | int const_alpha); |
97 | |
98 | typedef void (*SrcOverScaleFunc)(uchar *destPixels, int dbpl, |
99 | const uchar *src, int spbl, int srch, |
100 | const QRectF &targetRect, |
101 | const QRectF &sourceRect, |
102 | const QRect &clipRect, |
103 | int const_alpha); |
104 | |
105 | typedef void (*SrcOverTransformFunc)(uchar *destPixels, int dbpl, |
106 | const uchar *src, int spbl, |
107 | const QRectF &targetRect, |
108 | const QRectF &sourceRect, |
109 | const QRect &clipRect, |
110 | const QTransform &targetRectTransform, |
111 | int const_alpha); |
112 | |
113 | struct DrawHelper { |
114 | ProcessSpans blendColor; |
115 | BitmapBlitFunc bitmapBlit; |
116 | AlphamapBlitFunc alphamapBlit; |
117 | AlphaRGBBlitFunc alphaRGBBlit; |
118 | RectFillFunc fillRect; |
119 | }; |
120 | |
121 | extern SrcOverBlendFunc qBlendFunctions[QImage::NImageFormats][QImage::NImageFormats]; |
122 | extern SrcOverScaleFunc qScaleFunctions[QImage::NImageFormats][QImage::NImageFormats]; |
123 | extern SrcOverTransformFunc qTransformFunctions[QImage::NImageFormats][QImage::NImageFormats]; |
124 | |
125 | extern DrawHelper qDrawHelper[QImage::NImageFormats]; |
126 | |
127 | struct quint24 { |
128 | quint24() = default; |
129 | quint24(uint value) |
130 | { |
131 | data[0] = uchar(value >> 16); |
132 | data[1] = uchar(value >> 8); |
133 | data[2] = uchar(value); |
134 | } |
135 | operator uint() const |
136 | { |
137 | return data[2] | (data[1] << 8) | (data[0] << 16); |
138 | } |
139 | |
140 | uchar data[3]; |
141 | }; |
142 | |
143 | void qBlendGradient(int count, const QT_FT_Span *spans, void *userData); |
144 | void qBlendTexture(int count, const QT_FT_Span *spans, void *userData); |
145 | #ifdef Q_PROCESSOR_X86 |
146 | extern void (*qt_memfill64)(quint64 *dest, quint64 value, qsizetype count); |
147 | extern void (*qt_memfill32)(quint32 *dest, quint32 value, qsizetype count); |
148 | #else |
149 | extern void qt_memfill64(quint64 *dest, quint64 value, qsizetype count); |
150 | extern void qt_memfill32(quint32 *dest, quint32 value, qsizetype count); |
151 | #endif |
152 | extern void qt_memfill24(quint24 *dest, quint24 value, qsizetype count); |
153 | extern void qt_memfill16(quint16 *dest, quint16 value, qsizetype count); |
154 | |
155 | typedef void (QT_FASTCALL *CompositionFunction)(uint *Q_DECL_RESTRICT dest, const uint *Q_DECL_RESTRICT src, int length, uint const_alpha); |
156 | typedef void (QT_FASTCALL *CompositionFunction64)(QRgba64 *Q_DECL_RESTRICT dest, const QRgba64 *Q_DECL_RESTRICT src, int length, uint const_alpha); |
157 | typedef void (QT_FASTCALL *CompositionFunctionFP)(QRgbaFloat32 *Q_DECL_RESTRICT dest, const QRgbaFloat32 *Q_DECL_RESTRICT src, int length, uint const_alpha); |
158 | typedef void (QT_FASTCALL *CompositionFunctionSolid)(uint *dest, int length, uint color, uint const_alpha); |
159 | typedef void (QT_FASTCALL *CompositionFunctionSolid64)(QRgba64 *dest, int length, QRgba64 color, uint const_alpha); |
160 | typedef void (QT_FASTCALL *CompositionFunctionSolidFP)(QRgbaFloat32 *dest, int length, QRgbaFloat32 color, uint const_alpha); |
161 | |
162 | struct LinearGradientValues |
163 | { |
164 | qreal dx; |
165 | qreal dy; |
166 | qreal l; |
167 | qreal off; |
168 | }; |
169 | |
170 | struct RadialGradientValues |
171 | { |
172 | qreal dx; |
173 | qreal dy; |
174 | qreal dr; |
175 | qreal sqrfr; |
176 | qreal a; |
177 | bool extended; |
178 | }; |
179 | |
180 | struct Operator; |
181 | typedef uint* (QT_FASTCALL *DestFetchProc)(uint *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length); |
182 | typedef QRgba64* (QT_FASTCALL *DestFetchProc64)(QRgba64 *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length); |
183 | typedef QRgbaFloat32* (QT_FASTCALL *DestFetchProcFP)(QRgbaFloat32 *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length); |
184 | typedef void (QT_FASTCALL *DestStoreProc)(QRasterBuffer *rasterBuffer, int x, int y, const uint *buffer, int length); |
185 | typedef void (QT_FASTCALL *DestStoreProc64)(QRasterBuffer *rasterBuffer, int x, int y, const QRgba64 *buffer, int length); |
186 | typedef void (QT_FASTCALL *DestStoreProcFP)(QRasterBuffer *rasterBuffer, int x, int y, const QRgbaFloat32 *buffer, int length); |
187 | typedef const uint* (QT_FASTCALL *SourceFetchProc)(uint *buffer, const Operator *o, const QSpanData *data, int y, int x, int length); |
188 | typedef const QRgba64* (QT_FASTCALL *SourceFetchProc64)(QRgba64 *buffer, const Operator *o, const QSpanData *data, int y, int x, int length); |
189 | typedef const QRgbaFloat32* (QT_FASTCALL *SourceFetchProcFP)(QRgbaFloat32 *buffer, const Operator *o, const QSpanData *data, int y, int x, int length); |
190 | |
191 | struct Operator |
192 | { |
193 | QPainter::CompositionMode mode; |
194 | DestFetchProc destFetch; |
195 | DestStoreProc destStore; |
196 | SourceFetchProc srcFetch; |
197 | CompositionFunctionSolid funcSolid; |
198 | CompositionFunction func; |
199 | |
200 | DestFetchProc64 destFetch64; |
201 | DestStoreProc64 destStore64; |
202 | SourceFetchProc64 srcFetch64; |
203 | CompositionFunctionSolid64 funcSolid64; |
204 | CompositionFunction64 func64; |
205 | |
206 | DestFetchProcFP destFetchFP; |
207 | DestStoreProcFP destStoreFP; |
208 | SourceFetchProcFP srcFetchFP; |
209 | CompositionFunctionSolidFP funcSolidFP; |
210 | CompositionFunctionFP funcFP; |
211 | |
212 | union { |
213 | LinearGradientValues linear; |
214 | RadialGradientValues radial; |
215 | }; |
216 | }; |
217 | |
218 | class QRasterPaintEngine; |
219 | |
220 | struct QLinearGradientData |
221 | { |
222 | struct { |
223 | qreal x; |
224 | qreal y; |
225 | } origin; |
226 | struct { |
227 | qreal x; |
228 | qreal y; |
229 | } end; |
230 | }; |
231 | |
232 | struct QRadialGradientData |
233 | { |
234 | struct { |
235 | qreal x; |
236 | qreal y; |
237 | qreal radius; |
238 | } center; |
239 | struct { |
240 | qreal x; |
241 | qreal y; |
242 | qreal radius; |
243 | } focal; |
244 | }; |
245 | |
246 | struct QConicalGradientData |
247 | { |
248 | struct { |
249 | qreal x; |
250 | qreal y; |
251 | } center; |
252 | qreal angle; |
253 | }; |
254 | |
255 | struct QGradientData |
256 | { |
257 | QGradient::Spread spread; |
258 | |
259 | union { |
260 | QLinearGradientData linear; |
261 | QRadialGradientData radial; |
262 | QConicalGradientData conical; |
263 | }; |
264 | |
265 | #define GRADIENT_STOPTABLE_SIZE 1024 |
266 | #define GRADIENT_STOPTABLE_SIZE_SHIFT 10 |
267 | |
268 | #if QT_CONFIG(raster_64bit) || QT_CONFIG(raster_fp) |
269 | const QRgba64 *colorTable64; //[GRADIENT_STOPTABLE_SIZE]; |
270 | #endif |
271 | const QRgb *colorTable32; //[GRADIENT_STOPTABLE_SIZE]; |
272 | |
273 | uint alphaColor : 1; |
274 | }; |
275 | |
276 | struct QTextureData |
277 | { |
278 | const uchar *imageData; |
279 | const uchar *scanLine(int y) const { return imageData + y*bytesPerLine; } |
280 | |
281 | int width; |
282 | int height; |
283 | // clip rect |
284 | int x1; |
285 | int y1; |
286 | int x2; |
287 | int y2; |
288 | qsizetype bytesPerLine; |
289 | QImage::Format format; |
290 | const QList<QRgb> *colorTable; |
291 | bool hasAlpha; |
292 | enum Type { |
293 | Plain, |
294 | Tiled, |
295 | Pattern |
296 | }; |
297 | Type type; |
298 | int const_alpha; |
299 | }; |
300 | |
301 | struct QSpanData |
302 | { |
303 | QSpanData() : tempImage(nullptr) {} |
304 | ~QSpanData() { delete tempImage; } |
305 | |
306 | QRasterBuffer *rasterBuffer; |
307 | ProcessSpans blend; |
308 | ProcessSpans unclipped_blend; |
309 | BitmapBlitFunc bitmapBlit; |
310 | AlphamapBlitFunc alphamapBlit; |
311 | AlphaRGBBlitFunc alphaRGBBlit; |
312 | RectFillFunc fillRect; |
313 | qreal m11, m12, m13, m21, m22, m23, m33, dx, dy; // inverse xform matrix |
314 | const QClipData *clip; |
315 | enum Type { |
316 | None, |
317 | Solid, |
318 | LinearGradient, |
319 | RadialGradient, |
320 | ConicalGradient, |
321 | Texture |
322 | } type : 8; |
323 | signed int txop : 8; |
324 | uint fast_matrix : 1; |
325 | bool bilinear; |
326 | QImage *tempImage; |
327 | QColor solidColor; |
328 | union { |
329 | QGradientData gradient; |
330 | QTextureData texture; |
331 | }; |
332 | std::shared_ptr<const void> cachedGradient; |
333 | |
334 | |
335 | void init(QRasterBuffer *rb, const QRasterPaintEngine *pe); |
336 | void setup(const QBrush &brush, int alpha, QPainter::CompositionMode compositionMode, bool isCosmetic); |
337 | void setupMatrix(const QTransform &matrix, int bilinear); |
338 | void initTexture(const QImage *image, int alpha, QTextureData::Type = QTextureData::Plain, const QRect &sourceRect = QRect()); |
339 | void adjustSpanMethods(); |
340 | }; |
341 | |
342 | static inline uint qt_gradient_clamp(const QGradientData *data, int ipos) |
343 | { |
344 | if (ipos < 0 || ipos >= GRADIENT_STOPTABLE_SIZE) { |
345 | if (data->spread == QGradient::RepeatSpread) { |
346 | ipos = ipos % GRADIENT_STOPTABLE_SIZE; |
347 | ipos = ipos < 0 ? GRADIENT_STOPTABLE_SIZE + ipos : ipos; |
348 | } else if (data->spread == QGradient::ReflectSpread) { |
349 | const int limit = GRADIENT_STOPTABLE_SIZE * 2; |
350 | ipos = ipos % limit; |
351 | ipos = ipos < 0 ? limit + ipos : ipos; |
352 | ipos = ipos >= GRADIENT_STOPTABLE_SIZE ? limit - 1 - ipos : ipos; |
353 | } else { |
354 | if (ipos < 0) |
355 | ipos = 0; |
356 | else if (ipos >= GRADIENT_STOPTABLE_SIZE) |
357 | ipos = GRADIENT_STOPTABLE_SIZE-1; |
358 | } |
359 | } |
360 | |
361 | Q_ASSERT(ipos >= 0); |
362 | Q_ASSERT(ipos < GRADIENT_STOPTABLE_SIZE); |
363 | |
364 | return ipos; |
365 | } |
366 | |
367 | static inline uint qt_gradient_pixel(const QGradientData *data, qreal pos) |
368 | { |
369 | int ipos = int(pos * (GRADIENT_STOPTABLE_SIZE - 1) + qreal(0.5)); |
370 | return data->colorTable32[qt_gradient_clamp(data, ipos)]; |
371 | } |
372 | |
373 | #if QT_CONFIG(raster_64bit) |
374 | static inline const QRgba64& qt_gradient_pixel64(const QGradientData *data, qreal pos) |
375 | { |
376 | int ipos = int(pos * (GRADIENT_STOPTABLE_SIZE - 1) + qreal(0.5)); |
377 | return data->colorTable64[qt_gradient_clamp(data, ipos)]; |
378 | } |
379 | #endif |
380 | |
381 | static inline qreal qRadialDeterminant(qreal a, qreal b, qreal c) |
382 | { |
383 | return (b * b) - (4 * a * c); |
384 | } |
385 | |
386 | template <class RadialFetchFunc, typename BlendType> static |
387 | const BlendType * QT_FASTCALL qt_fetch_radial_gradient_template(BlendType *buffer, const Operator *op, |
388 | const QSpanData *data, int y, int x, int length) |
389 | { |
390 | // avoid division by zero |
391 | if (qFuzzyIsNull(d: op->radial.a)) { |
392 | RadialFetchFunc::memfill(buffer, RadialFetchFunc::null(), length); |
393 | return buffer; |
394 | } |
395 | |
396 | const BlendType *b = buffer; |
397 | qreal rx = data->m21 * (y + qreal(0.5)) |
398 | + data->dx + data->m11 * (x + qreal(0.5)); |
399 | qreal ry = data->m22 * (y + qreal(0.5)) |
400 | + data->dy + data->m12 * (x + qreal(0.5)); |
401 | bool affine = !data->m13 && !data->m23; |
402 | |
403 | BlendType *end = buffer + length; |
404 | qreal inv_a = 1 / qreal(2 * op->radial.a); |
405 | |
406 | if (affine) { |
407 | rx -= data->gradient.radial.focal.x; |
408 | ry -= data->gradient.radial.focal.y; |
409 | |
410 | const qreal delta_rx = data->m11; |
411 | const qreal delta_ry = data->m12; |
412 | |
413 | qreal b = 2*(op->radial.dr*data->gradient.radial.focal.radius + rx * op->radial.dx + ry * op->radial.dy); |
414 | qreal delta_b = 2*(delta_rx * op->radial.dx + delta_ry * op->radial.dy); |
415 | const qreal b_delta_b = 2 * b * delta_b; |
416 | const qreal delta_b_delta_b = 2 * delta_b * delta_b; |
417 | |
418 | const qreal bb = b * b; |
419 | const qreal delta_bb = delta_b * delta_b; |
420 | |
421 | b *= inv_a; |
422 | delta_b *= inv_a; |
423 | |
424 | const qreal rxrxryry = rx * rx + ry * ry; |
425 | const qreal delta_rxrxryry = delta_rx * delta_rx + delta_ry * delta_ry; |
426 | const qreal rx_plus_ry = 2*(rx * delta_rx + ry * delta_ry); |
427 | const qreal delta_rx_plus_ry = 2 * delta_rxrxryry; |
428 | |
429 | inv_a *= inv_a; |
430 | |
431 | qreal det = (bb - 4 * op->radial.a * (op->radial.sqrfr - rxrxryry)) * inv_a; |
432 | qreal delta_det = (b_delta_b + delta_bb + 4 * op->radial.a * (rx_plus_ry + delta_rxrxryry)) * inv_a; |
433 | const qreal delta_delta_det = (delta_b_delta_b + 4 * op->radial.a * delta_rx_plus_ry) * inv_a; |
434 | |
435 | RadialFetchFunc::fetch(buffer, end, op, data, det, delta_det, delta_delta_det, b, delta_b); |
436 | } else { |
437 | qreal rw = data->m23 * (y + qreal(0.5)) |
438 | + data->m33 + data->m13 * (x + qreal(0.5)); |
439 | |
440 | while (buffer < end) { |
441 | if (rw == 0) { |
442 | *buffer = RadialFetchFunc::null(); |
443 | } else { |
444 | qreal invRw = 1 / rw; |
445 | qreal gx = rx * invRw - data->gradient.radial.focal.x; |
446 | qreal gy = ry * invRw - data->gradient.radial.focal.y; |
447 | qreal b = 2*(op->radial.dr*data->gradient.radial.focal.radius + gx*op->radial.dx + gy*op->radial.dy); |
448 | qreal det = qRadialDeterminant(a: op->radial.a, b, c: op->radial.sqrfr - (gx*gx + gy*gy)); |
449 | |
450 | BlendType result = RadialFetchFunc::null(); |
451 | if (det >= 0) { |
452 | qreal detSqrt = qSqrt(v: det); |
453 | |
454 | qreal s0 = (-b - detSqrt) * inv_a; |
455 | qreal s1 = (-b + detSqrt) * inv_a; |
456 | |
457 | qreal s = qMax(a: s0, b: s1); |
458 | |
459 | if (data->gradient.radial.focal.radius + op->radial.dr * s >= 0) |
460 | result = RadialFetchFunc::fetchSingle(data->gradient, s); |
461 | } |
462 | |
463 | *buffer = result; |
464 | } |
465 | |
466 | rx += data->m11; |
467 | ry += data->m12; |
468 | rw += data->m13; |
469 | |
470 | ++buffer; |
471 | } |
472 | } |
473 | |
474 | return b; |
475 | } |
476 | |
477 | template <class Simd> |
478 | class QRadialFetchSimd |
479 | { |
480 | public: |
481 | static uint null() { return 0; } |
482 | static uint fetchSingle(const QGradientData& gradient, qreal v) |
483 | { |
484 | return qt_gradient_pixel(data: &gradient, pos: v); |
485 | } |
486 | static void memfill(uint *buffer, uint fill, int length) |
487 | { |
488 | qt_memfill32(buffer, fill, length); |
489 | } |
490 | static void fetch(uint *buffer, uint *end, const Operator *op, const QSpanData *data, qreal det, |
491 | qreal delta_det, qreal delta_delta_det, qreal b, qreal delta_b) |
492 | { |
493 | typename Simd::Vect_buffer_f det_vec; |
494 | typename Simd::Vect_buffer_f delta_det4_vec; |
495 | typename Simd::Vect_buffer_f b_vec; |
496 | |
497 | for (int i = 0; i < 4; ++i) { |
498 | det_vec.f[i] = det; |
499 | delta_det4_vec.f[i] = 4 * delta_det; |
500 | b_vec.f[i] = b; |
501 | |
502 | det += delta_det; |
503 | delta_det += delta_delta_det; |
504 | b += delta_b; |
505 | } |
506 | |
507 | const typename Simd::Float32x4 v_delta_delta_det16 = Simd::v_dup(16 * delta_delta_det); |
508 | const typename Simd::Float32x4 v_delta_delta_det6 = Simd::v_dup(6 * delta_delta_det); |
509 | const typename Simd::Float32x4 v_delta_b4 = Simd::v_dup(4 * delta_b); |
510 | |
511 | const typename Simd::Float32x4 v_r0 = Simd::v_dup(data->gradient.radial.focal.radius); |
512 | const typename Simd::Float32x4 v_dr = Simd::v_dup(op->radial.dr); |
513 | |
514 | #if defined(__ARM_NEON__) |
515 | // NEON doesn't have SIMD sqrt, but uses rsqrt instead that can't be taken of 0. |
516 | const typename Simd::Float32x4 v_min = Simd::v_dup(std::numeric_limits<float>::epsilon()); |
517 | #else |
518 | const typename Simd::Float32x4 v_min = Simd::v_dup(0.0f); |
519 | #endif |
520 | const typename Simd::Float32x4 v_max = Simd::v_dup(float(GRADIENT_STOPTABLE_SIZE-1)); |
521 | const typename Simd::Float32x4 v_half = Simd::v_dup(0.5f); |
522 | |
523 | const typename Simd::Int32x4 v_repeat_mask = Simd::v_dup(~(uint(0xffffff) << GRADIENT_STOPTABLE_SIZE_SHIFT)); |
524 | const typename Simd::Int32x4 v_reflect_mask = Simd::v_dup(~(uint(0xffffff) << (GRADIENT_STOPTABLE_SIZE_SHIFT+1))); |
525 | |
526 | const typename Simd::Int32x4 v_reflect_limit = Simd::v_dup(2 * GRADIENT_STOPTABLE_SIZE - 1); |
527 | |
528 | const int extended_mask = op->radial.extended ? 0x0 : ~0x0; |
529 | |
530 | #define FETCH_RADIAL_LOOP_PROLOGUE \ |
531 | while (buffer < end) { \ |
532 | typename Simd::Vect_buffer_i v_buffer_mask; \ |
533 | v_buffer_mask.v = Simd::v_greaterOrEqual(det_vec.v, v_min); \ |
534 | const typename Simd::Float32x4 v_index_local = Simd::v_sub(Simd::v_sqrt(Simd::v_max(v_min, det_vec.v)), b_vec.v); \ |
535 | const typename Simd::Float32x4 v_index = Simd::v_add(Simd::v_mul(v_index_local, v_max), v_half); \ |
536 | v_buffer_mask.v = Simd::v_and(v_buffer_mask.v, Simd::v_greaterOrEqual(Simd::v_add(v_r0, Simd::v_mul(v_dr, v_index_local)), v_min)); \ |
537 | typename Simd::Vect_buffer_i index_vec; |
538 | #define FETCH_RADIAL_LOOP_CLAMP_REPEAT \ |
539 | index_vec.v = Simd::v_and(v_repeat_mask, Simd::v_toInt(v_index)); |
540 | #define FETCH_RADIAL_LOOP_CLAMP_REFLECT \ |
541 | const typename Simd::Int32x4 v_index_i = Simd::v_and(v_reflect_mask, Simd::v_toInt(v_index)); \ |
542 | const typename Simd::Int32x4 v_index_i_inv = Simd::v_sub(v_reflect_limit, v_index_i); \ |
543 | index_vec.v = Simd::v_min_16(v_index_i, v_index_i_inv); |
544 | #define FETCH_RADIAL_LOOP_CLAMP_PAD \ |
545 | index_vec.v = Simd::v_toInt(Simd::v_min(v_max, Simd::v_max(v_min, v_index))); |
546 | #define FETCH_RADIAL_LOOP_EPILOGUE \ |
547 | det_vec.v = Simd::v_add(Simd::v_add(det_vec.v, delta_det4_vec.v), v_delta_delta_det6); \ |
548 | delta_det4_vec.v = Simd::v_add(delta_det4_vec.v, v_delta_delta_det16); \ |
549 | b_vec.v = Simd::v_add(b_vec.v, v_delta_b4); \ |
550 | for (int i = 0; i < 4; ++i) \ |
551 | *buffer++ = (extended_mask | v_buffer_mask.i[i]) & data->gradient.colorTable32[index_vec.i[i]]; \ |
552 | } |
553 | |
554 | #define FETCH_RADIAL_LOOP(FETCH_RADIAL_LOOP_CLAMP) \ |
555 | FETCH_RADIAL_LOOP_PROLOGUE \ |
556 | FETCH_RADIAL_LOOP_CLAMP \ |
557 | FETCH_RADIAL_LOOP_EPILOGUE |
558 | |
559 | switch (data->gradient.spread) { |
560 | case QGradient::RepeatSpread: |
561 | FETCH_RADIAL_LOOP(FETCH_RADIAL_LOOP_CLAMP_REPEAT) |
562 | break; |
563 | case QGradient::ReflectSpread: |
564 | FETCH_RADIAL_LOOP(FETCH_RADIAL_LOOP_CLAMP_REFLECT) |
565 | break; |
566 | case QGradient::PadSpread: |
567 | FETCH_RADIAL_LOOP(FETCH_RADIAL_LOOP_CLAMP_PAD) |
568 | break; |
569 | default: |
570 | Q_UNREACHABLE(); |
571 | } |
572 | } |
573 | }; |
574 | |
575 | static inline uint INTERPOLATE_PIXEL_255(uint x, uint a, uint y, uint b) { |
576 | uint t = (x & 0xff00ff) * a + (y & 0xff00ff) * b; |
577 | t = (t + ((t >> 8) & 0xff00ff) + 0x800080) >> 8; |
578 | t &= 0xff00ff; |
579 | |
580 | x = ((x >> 8) & 0xff00ff) * a + ((y >> 8) & 0xff00ff) * b; |
581 | x = (x + ((x >> 8) & 0xff00ff) + 0x800080); |
582 | x &= 0xff00ff00; |
583 | x |= t; |
584 | return x; |
585 | } |
586 | |
587 | #if Q_PROCESSOR_WORDSIZE == 8 // 64-bit versions |
588 | |
589 | static inline uint INTERPOLATE_PIXEL_256(uint x, uint a, uint y, uint b) { |
590 | quint64 t = (((quint64(x)) | ((quint64(x)) << 24)) & 0x00ff00ff00ff00ff) * a; |
591 | t += (((quint64(y)) | ((quint64(y)) << 24)) & 0x00ff00ff00ff00ff) * b; |
592 | t >>= 8; |
593 | t &= 0x00ff00ff00ff00ff; |
594 | return (uint(t)) | (uint(t >> 24)); |
595 | } |
596 | |
597 | static inline uint BYTE_MUL(uint x, uint a) { |
598 | quint64 t = (((quint64(x)) | ((quint64(x)) << 24)) & 0x00ff00ff00ff00ff) * a; |
599 | t = (t + ((t >> 8) & 0xff00ff00ff00ff) + 0x80008000800080) >> 8; |
600 | t &= 0x00ff00ff00ff00ff; |
601 | return (uint(t)) | (uint(t >> 24)); |
602 | } |
603 | |
604 | #else // 32-bit versions |
605 | |
606 | static inline uint INTERPOLATE_PIXEL_256(uint x, uint a, uint y, uint b) { |
607 | uint t = (x & 0xff00ff) * a + (y & 0xff00ff) * b; |
608 | t >>= 8; |
609 | t &= 0xff00ff; |
610 | |
611 | x = ((x >> 8) & 0xff00ff) * a + ((y >> 8) & 0xff00ff) * b; |
612 | x &= 0xff00ff00; |
613 | x |= t; |
614 | return x; |
615 | } |
616 | |
617 | static inline uint BYTE_MUL(uint x, uint a) { |
618 | uint t = (x & 0xff00ff) * a; |
619 | t = (t + ((t >> 8) & 0xff00ff) + 0x800080) >> 8; |
620 | t &= 0xff00ff; |
621 | |
622 | x = ((x >> 8) & 0xff00ff) * a; |
623 | x = (x + ((x >> 8) & 0xff00ff) + 0x800080); |
624 | x &= 0xff00ff00; |
625 | x |= t; |
626 | return x; |
627 | } |
628 | #endif |
629 | |
630 | static inline void blend_pixel(quint32 &dst, const quint32 src) |
631 | { |
632 | if (src >= 0xff000000) |
633 | dst = src; |
634 | else if (src != 0) |
635 | dst = src + BYTE_MUL(x: dst, a: qAlpha(rgb: ~src)); |
636 | } |
637 | |
638 | static inline void blend_pixel(quint32 &dst, const quint32 src, const int const_alpha) |
639 | { |
640 | if (const_alpha == 255) |
641 | return blend_pixel(dst, src); |
642 | if (src != 0) { |
643 | const quint32 s = BYTE_MUL(x: src, a: const_alpha); |
644 | dst = s + BYTE_MUL(x: dst, a: qAlpha(rgb: ~s)); |
645 | } |
646 | } |
647 | |
648 | #if defined(__SSE2__) |
649 | static inline uint Q_DECL_VECTORCALL interpolate_4_pixels_sse2(__m128i vt, __m128i vb, uint distx, uint disty) |
650 | { |
651 | // First interpolate top and bottom pixels in parallel. |
652 | vt = _mm_unpacklo_epi8(a: vt, b: _mm_setzero_si128()); |
653 | vb = _mm_unpacklo_epi8(a: vb, b: _mm_setzero_si128()); |
654 | vt = _mm_mullo_epi16(a: vt, b: _mm_set1_epi16(w: 256 - disty)); |
655 | vb = _mm_mullo_epi16(a: vb, b: _mm_set1_epi16(w: disty)); |
656 | __m128i vlr = _mm_add_epi16(a: vt, b: vb); |
657 | vlr = _mm_srli_epi16(a: vlr, count: 8); |
658 | // vlr now contains the result of the first two interpolate calls vlr = unpacked((xright << 64) | xleft) |
659 | |
660 | // Now the last interpolate between left and right.. |
661 | const __m128i vidistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(256 - distx), _MM_SHUFFLE(0, 0, 0, 0)); |
662 | const __m128i vdistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(distx), _MM_SHUFFLE(0, 0, 0, 0)); |
663 | const __m128i vmulx = _mm_unpacklo_epi16(a: vidistx, b: vdistx); |
664 | vlr = _mm_unpacklo_epi16(a: vlr, _mm_srli_si128(vlr, 8)); |
665 | // vlr now contains the colors of left and right interleaved { la, ra, lr, rr, lg, rg, lb, rb } |
666 | vlr = _mm_madd_epi16(a: vlr, b: vmulx); // Multiply and horizontal add. |
667 | vlr = _mm_srli_epi32(a: vlr, count: 8); |
668 | vlr = _mm_packs_epi32(a: vlr, b: vlr); |
669 | vlr = _mm_packus_epi16(a: vlr, b: vlr); |
670 | return _mm_cvtsi128_si32(a: vlr); |
671 | } |
672 | |
673 | static inline uint interpolate_4_pixels(uint tl, uint tr, uint bl, uint br, uint distx, uint disty) |
674 | { |
675 | __m128i vt = _mm_unpacklo_epi32(a: _mm_cvtsi32_si128(a: tl), b: _mm_cvtsi32_si128(a: tr)); |
676 | __m128i vb = _mm_unpacklo_epi32(a: _mm_cvtsi32_si128(a: bl), b: _mm_cvtsi32_si128(a: br)); |
677 | return interpolate_4_pixels_sse2(vt, vb, distx, disty); |
678 | } |
679 | |
680 | static inline uint interpolate_4_pixels(const uint t[], const uint b[], uint distx, uint disty) |
681 | { |
682 | __m128i vt = _mm_loadl_epi64(p: (const __m128i*)t); |
683 | __m128i vb = _mm_loadl_epi64(p: (const __m128i*)b); |
684 | return interpolate_4_pixels_sse2(vt, vb, distx, disty); |
685 | } |
686 | |
687 | static constexpr inline bool hasFastInterpolate4() { return true; } |
688 | |
689 | #elif defined(__ARM_NEON__) |
690 | static inline uint interpolate_4_pixels_neon(uint32x2_t vt32, uint32x2_t vb32, uint distx, uint disty) |
691 | { |
692 | uint16x8_t vt16 = vmovl_u8(vreinterpret_u8_u32(vt32)); |
693 | uint16x8_t vb16 = vmovl_u8(vreinterpret_u8_u32(vb32)); |
694 | vt16 = vmulq_n_u16(vt16, 256 - disty); |
695 | vt16 = vmlaq_n_u16(vt16, vb16, disty); |
696 | vt16 = vshrq_n_u16(vt16, 8); |
697 | uint16x4_t vl16 = vget_low_u16(vt16); |
698 | uint16x4_t vr16 = vget_high_u16(vt16); |
699 | vl16 = vmul_n_u16(vl16, 256 - distx); |
700 | vl16 = vmla_n_u16(vl16, vr16, distx); |
701 | vl16 = vshr_n_u16(vl16, 8); |
702 | uint8x8_t vr = vmovn_u16(vcombine_u16(vl16, vl16)); |
703 | return vget_lane_u32(vreinterpret_u32_u8(vr), 0); |
704 | } |
705 | |
706 | static inline uint interpolate_4_pixels(uint tl, uint tr, uint bl, uint br, uint distx, uint disty) |
707 | { |
708 | uint32x2_t vt32 = vmov_n_u32(tl); |
709 | uint32x2_t vb32 = vmov_n_u32(bl); |
710 | vt32 = vset_lane_u32(tr, vt32, 1); |
711 | vb32 = vset_lane_u32(br, vb32, 1); |
712 | return interpolate_4_pixels_neon(vt32, vb32, distx, disty); |
713 | } |
714 | |
715 | static inline uint interpolate_4_pixels(const uint t[], const uint b[], uint distx, uint disty) |
716 | { |
717 | uint32x2_t vt32 = vld1_u32(t); |
718 | uint32x2_t vb32 = vld1_u32(b); |
719 | return interpolate_4_pixels_neon(vt32, vb32, distx, disty); |
720 | } |
721 | |
722 | static constexpr inline bool hasFastInterpolate4() { return true; } |
723 | |
724 | #else |
725 | static inline uint interpolate_4_pixels(uint tl, uint tr, uint bl, uint br, uint distx, uint disty) |
726 | { |
727 | uint idistx = 256 - distx; |
728 | uint idisty = 256 - disty; |
729 | uint xtop = INTERPOLATE_PIXEL_256(tl, idistx, tr, distx); |
730 | uint xbot = INTERPOLATE_PIXEL_256(bl, idistx, br, distx); |
731 | return INTERPOLATE_PIXEL_256(xtop, idisty, xbot, disty); |
732 | } |
733 | |
734 | static inline uint interpolate_4_pixels(const uint t[], const uint b[], uint distx, uint disty) |
735 | { |
736 | return interpolate_4_pixels(t[0], t[1], b[0], b[1], distx, disty); |
737 | } |
738 | |
739 | static constexpr inline bool hasFastInterpolate4() { return false; } |
740 | |
741 | #endif |
742 | |
743 | static inline QRgba64 multiplyAlpha256(QRgba64 rgba64, uint alpha256) |
744 | { |
745 | return QRgba64::fromRgba64(red: (rgba64.red() * alpha256) >> 8, |
746 | green: (rgba64.green() * alpha256) >> 8, |
747 | blue: (rgba64.blue() * alpha256) >> 8, |
748 | alpha: (rgba64.alpha() * alpha256) >> 8); |
749 | } |
750 | static inline QRgba64 interpolate256(QRgba64 x, uint alpha1, QRgba64 y, uint alpha2) |
751 | { |
752 | return QRgba64::fromRgba64(c: multiplyAlpha256(rgba64: x, alpha256: alpha1) + multiplyAlpha256(rgba64: y, alpha256: alpha2)); |
753 | } |
754 | |
755 | #ifdef __SSE2__ |
756 | static inline QRgba64 interpolate_4_pixels_rgb64(const QRgba64 t[], const QRgba64 b[], uint distx, uint disty) |
757 | { |
758 | __m128i vt = _mm_loadu_si128(p: (const __m128i*)t); |
759 | if (disty) { |
760 | __m128i vb = _mm_loadu_si128(p: (const __m128i*)b); |
761 | vt = _mm_mulhi_epu16(a: vt, b: _mm_set1_epi16(w: 0x10000 - disty)); |
762 | vb = _mm_mulhi_epu16(a: vb, b: _mm_set1_epi16(w: disty)); |
763 | vt = _mm_add_epi16(a: vt, b: vb); |
764 | } |
765 | if (distx) { |
766 | const __m128i vdistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(distx), _MM_SHUFFLE(0, 0, 0, 0)); |
767 | const __m128i vidistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(0x10000 - distx), _MM_SHUFFLE(0, 0, 0, 0)); |
768 | vt = _mm_mulhi_epu16(a: vt, b: _mm_unpacklo_epi64(a: vidistx, b: vdistx)); |
769 | vt = _mm_add_epi16(a: vt, _mm_srli_si128(vt, 8)); |
770 | } |
771 | #ifdef Q_PROCESSOR_X86_64 |
772 | return QRgba64::fromRgba64(c: _mm_cvtsi128_si64(a: vt)); |
773 | #else |
774 | QRgba64 out; |
775 | _mm_storel_epi64((__m128i*)&out, vt); |
776 | return out; |
777 | #endif // Q_PROCESSOR_X86_64 |
778 | } |
779 | #elif defined(__ARM_NEON__) |
780 | static inline QRgba64 interpolate_4_pixels_rgb64(const QRgba64 t[], const QRgba64 b[], uint distx, uint disty) |
781 | { |
782 | uint64x1x2_t vt = vld2_u64(reinterpret_cast<const uint64_t *>(t)); |
783 | if (disty) { |
784 | uint64x1x2_t vb = vld2_u64(reinterpret_cast<const uint64_t *>(b)); |
785 | uint32x4_t vt0 = vmull_n_u16(vreinterpret_u16_u64(vt.val[0]), 0x10000 - disty); |
786 | uint32x4_t vt1 = vmull_n_u16(vreinterpret_u16_u64(vt.val[1]), 0x10000 - disty); |
787 | vt0 = vmlal_n_u16(vt0, vreinterpret_u16_u64(vb.val[0]), disty); |
788 | vt1 = vmlal_n_u16(vt1, vreinterpret_u16_u64(vb.val[1]), disty); |
789 | vt.val[0] = vreinterpret_u64_u16(vshrn_n_u32(vt0, 16)); |
790 | vt.val[1] = vreinterpret_u64_u16(vshrn_n_u32(vt1, 16)); |
791 | } |
792 | if (distx) { |
793 | uint32x4_t vt0 = vmull_n_u16(vreinterpret_u16_u64(vt.val[0]), 0x10000 - distx); |
794 | vt0 = vmlal_n_u16(vt0, vreinterpret_u16_u64(vt.val[1]), distx); |
795 | vt.val[0] = vreinterpret_u64_u16(vshrn_n_u32(vt0, 16)); |
796 | } |
797 | QRgba64 out; |
798 | vst1_u64(reinterpret_cast<uint64_t *>(&out), vt.val[0]); |
799 | return out; |
800 | } |
801 | #else |
802 | static inline QRgba64 interpolate_4_pixels_rgb64(const QRgba64 t[], const QRgba64 b[], uint distx, uint disty) |
803 | { |
804 | const uint dx = distx>>8; |
805 | const uint dy = disty>>8; |
806 | const uint idx = 256 - dx; |
807 | const uint idy = 256 - dy; |
808 | QRgba64 xtop = interpolate256(t[0], idx, t[1], dx); |
809 | QRgba64 xbot = interpolate256(b[0], idx, b[1], dx); |
810 | return interpolate256(xtop, idy, xbot, dy); |
811 | } |
812 | #endif // __SSE2__ |
813 | |
814 | #if QT_CONFIG(raster_fp) |
815 | static inline QRgbaFloat32 multiplyAlpha_rgba32f(QRgbaFloat32 c, float a) |
816 | { |
817 | return QRgbaFloat32 { .r: c.r * a, .g: c.g * a, .b: c.b * a, .a: c.a * a }; |
818 | } |
819 | |
820 | static inline QRgbaFloat32 interpolate_rgba32f(QRgbaFloat32 x, float alpha1, QRgbaFloat32 y, float alpha2) |
821 | { |
822 | x = multiplyAlpha_rgba32f(c: x, a: alpha1); |
823 | y = multiplyAlpha_rgba32f(c: y, a: alpha2); |
824 | return QRgbaFloat32 { .r: x.r + y.r, .g: x.g + y.g, .b: x.b + y.b, .a: x.a + y.a }; |
825 | } |
826 | #ifdef __SSE2__ |
827 | static inline __m128 Q_DECL_VECTORCALL interpolate_rgba32f(__m128 x, __m128 alpha1, __m128 y, __m128 alpha2) |
828 | { |
829 | return _mm_add_ps(a: _mm_mul_ps(a: x, b: alpha1), b: _mm_mul_ps(a: y, b: alpha2)); |
830 | } |
831 | #endif |
832 | |
833 | static inline QRgbaFloat32 interpolate_4_pixels_rgba32f(const QRgbaFloat32 t[], const QRgbaFloat32 b[], uint distx, uint disty) |
834 | { |
835 | constexpr float f = 1.0f / 65536.0f; |
836 | const float dx = distx * f; |
837 | const float dy = disty * f; |
838 | const float idx = 1.0f - dx; |
839 | const float idy = 1.0f - dy; |
840 | #ifdef __SSE2__ |
841 | const __m128 vtl = _mm_load_ps(p: (const float *)&t[0]); |
842 | const __m128 vtr = _mm_load_ps(p: (const float *)&t[1]); |
843 | const __m128 vbl = _mm_load_ps(p: (const float *)&b[0]); |
844 | const __m128 vbr = _mm_load_ps(p: (const float *)&b[1]); |
845 | |
846 | const __m128 vdx = _mm_set1_ps(w: dx); |
847 | const __m128 vidx = _mm_set1_ps(w: idx); |
848 | __m128 vt = interpolate_rgba32f(x: vtl, alpha1: vidx, y: vtr, alpha2: vdx); |
849 | __m128 vb = interpolate_rgba32f(x: vbl, alpha1: vidx, y: vbr, alpha2: vdx); |
850 | const __m128 vdy = _mm_set1_ps(w: dy); |
851 | const __m128 vidy = _mm_set1_ps(w: idy); |
852 | vt = interpolate_rgba32f(x: vt, alpha1: vidy, y: vb, alpha2: vdy); |
853 | QRgbaFloat32 res; |
854 | _mm_store_ps(p: (float*)&res, a: vt); |
855 | return res; |
856 | #else |
857 | QRgbaFloat32 xtop = interpolate_rgba32f(t[0], idx, t[1], dx); |
858 | QRgbaFloat32 xbot = interpolate_rgba32f(b[0], idx, b[1], dx); |
859 | xtop = interpolate_rgba32f(xtop, idy, xbot, dy); |
860 | return xtop; |
861 | #endif |
862 | } |
863 | #endif // QT_CONFIG(raster_fp) |
864 | |
865 | static inline uint BYTE_MUL_RGB16(uint x, uint a) { |
866 | a += 1; |
867 | uint t = (((x & 0x07e0)*a) >> 8) & 0x07e0; |
868 | t |= (((x & 0xf81f)*(a>>2)) >> 6) & 0xf81f; |
869 | return t; |
870 | } |
871 | |
872 | static inline uint BYTE_MUL_RGB16_32(uint x, uint a) { |
873 | uint t = (((x & 0xf81f07e0) >> 5)*a) & 0xf81f07e0; |
874 | t |= (((x & 0x07e0f81f)*a) >> 5) & 0x07e0f81f; |
875 | return t; |
876 | } |
877 | |
878 | // qt_div_255 is a fast rounded division by 255 using an approximation that is accurate for all positive 16-bit integers |
879 | static constexpr inline int qt_div_255(int x) { return (x + (x>>8) + 0x80) >> 8; } |
880 | static constexpr inline uint qt_div_257_floor(uint x) { return (x - (x >> 8)) >> 8; } |
881 | static constexpr inline uint qt_div_257(uint x) { return qt_div_257_floor(x: x + 128); } |
882 | static constexpr inline uint qt_div_65535(uint x) { return (x + (x>>16) + 0x8000U) >> 16; } |
883 | |
884 | template <class T> inline void qt_memfill_template(T *dest, T color, qsizetype count) |
885 | { |
886 | if (!count) |
887 | return; |
888 | |
889 | qsizetype n = (count + 7) / 8; |
890 | switch (count & 0x07) |
891 | { |
892 | case 0: do { *dest++ = color; Q_FALLTHROUGH(); |
893 | case 7: *dest++ = color; Q_FALLTHROUGH(); |
894 | case 6: *dest++ = color; Q_FALLTHROUGH(); |
895 | case 5: *dest++ = color; Q_FALLTHROUGH(); |
896 | case 4: *dest++ = color; Q_FALLTHROUGH(); |
897 | case 3: *dest++ = color; Q_FALLTHROUGH(); |
898 | case 2: *dest++ = color; Q_FALLTHROUGH(); |
899 | case 1: *dest++ = color; |
900 | } while (--n > 0); |
901 | } |
902 | } |
903 | |
904 | template <class T> inline void qt_memfill(T *dest, T value, qsizetype count) |
905 | { |
906 | qt_memfill_template(dest, value, count); |
907 | } |
908 | |
909 | template<> inline void qt_memfill(quint64 *dest, quint64 color, qsizetype count) |
910 | { |
911 | qt_memfill64(dest, color, count); |
912 | } |
913 | |
914 | template<> inline void qt_memfill(quint32 *dest, quint32 color, qsizetype count) |
915 | { |
916 | qt_memfill32(dest, color, count); |
917 | } |
918 | |
919 | template<> inline void qt_memfill(quint24 *dest, quint24 color, qsizetype count) |
920 | { |
921 | qt_memfill24(dest, value: color, count); |
922 | } |
923 | |
924 | template<> inline void qt_memfill(quint16 *dest, quint16 color, qsizetype count) |
925 | { |
926 | qt_memfill16(dest, value: color, count); |
927 | } |
928 | |
929 | template<> inline void qt_memfill(quint8 *dest, quint8 color, qsizetype count) |
930 | { |
931 | memset(s: dest, c: color, n: count); |
932 | } |
933 | |
934 | template <class T> static |
935 | inline void qt_rectfill(T *dest, T value, |
936 | int x, int y, int width, int height, qsizetype stride) |
937 | { |
938 | char *d = reinterpret_cast<char*>(dest + x) + y * stride; |
939 | if (uint(stride) == (width * sizeof(T))) { |
940 | qt_memfill(reinterpret_cast<T*>(d), value, qsizetype(width) * height); |
941 | } else { |
942 | for (int j = 0; j < height; ++j) { |
943 | dest = reinterpret_cast<T*>(d); |
944 | qt_memfill(dest, value, width); |
945 | d += stride; |
946 | } |
947 | } |
948 | } |
949 | |
950 | inline ushort qConvertRgb32To16(uint c) |
951 | { |
952 | return (((c) >> 3) & 0x001f) |
953 | | (((c) >> 5) & 0x07e0) |
954 | | (((c) >> 8) & 0xf800); |
955 | } |
956 | |
957 | inline QRgb qConvertRgb16To32(uint c) |
958 | { |
959 | return 0xff000000 |
960 | | ((((c) << 3) & 0xf8) | (((c) >> 2) & 0x7)) |
961 | | ((((c) << 5) & 0xfc00) | (((c) >> 1) & 0x300)) |
962 | | ((((c) << 8) & 0xf80000) | (((c) << 3) & 0x70000)); |
963 | } |
964 | |
965 | const uint qt_bayer_matrix[16][16] = { |
966 | { 0x1, 0xc0, 0x30, 0xf0, 0xc, 0xcc, 0x3c, 0xfc, |
967 | 0x3, 0xc3, 0x33, 0xf3, 0xf, 0xcf, 0x3f, 0xff}, |
968 | { 0x80, 0x40, 0xb0, 0x70, 0x8c, 0x4c, 0xbc, 0x7c, |
969 | 0x83, 0x43, 0xb3, 0x73, 0x8f, 0x4f, 0xbf, 0x7f}, |
970 | { 0x20, 0xe0, 0x10, 0xd0, 0x2c, 0xec, 0x1c, 0xdc, |
971 | 0x23, 0xe3, 0x13, 0xd3, 0x2f, 0xef, 0x1f, 0xdf}, |
972 | { 0xa0, 0x60, 0x90, 0x50, 0xac, 0x6c, 0x9c, 0x5c, |
973 | 0xa3, 0x63, 0x93, 0x53, 0xaf, 0x6f, 0x9f, 0x5f}, |
974 | { 0x8, 0xc8, 0x38, 0xf8, 0x4, 0xc4, 0x34, 0xf4, |
975 | 0xb, 0xcb, 0x3b, 0xfb, 0x7, 0xc7, 0x37, 0xf7}, |
976 | { 0x88, 0x48, 0xb8, 0x78, 0x84, 0x44, 0xb4, 0x74, |
977 | 0x8b, 0x4b, 0xbb, 0x7b, 0x87, 0x47, 0xb7, 0x77}, |
978 | { 0x28, 0xe8, 0x18, 0xd8, 0x24, 0xe4, 0x14, 0xd4, |
979 | 0x2b, 0xeb, 0x1b, 0xdb, 0x27, 0xe7, 0x17, 0xd7}, |
980 | { 0xa8, 0x68, 0x98, 0x58, 0xa4, 0x64, 0x94, 0x54, |
981 | 0xab, 0x6b, 0x9b, 0x5b, 0xa7, 0x67, 0x97, 0x57}, |
982 | { 0x2, 0xc2, 0x32, 0xf2, 0xe, 0xce, 0x3e, 0xfe, |
983 | 0x1, 0xc1, 0x31, 0xf1, 0xd, 0xcd, 0x3d, 0xfd}, |
984 | { 0x82, 0x42, 0xb2, 0x72, 0x8e, 0x4e, 0xbe, 0x7e, |
985 | 0x81, 0x41, 0xb1, 0x71, 0x8d, 0x4d, 0xbd, 0x7d}, |
986 | { 0x22, 0xe2, 0x12, 0xd2, 0x2e, 0xee, 0x1e, 0xde, |
987 | 0x21, 0xe1, 0x11, 0xd1, 0x2d, 0xed, 0x1d, 0xdd}, |
988 | { 0xa2, 0x62, 0x92, 0x52, 0xae, 0x6e, 0x9e, 0x5e, |
989 | 0xa1, 0x61, 0x91, 0x51, 0xad, 0x6d, 0x9d, 0x5d}, |
990 | { 0xa, 0xca, 0x3a, 0xfa, 0x6, 0xc6, 0x36, 0xf6, |
991 | 0x9, 0xc9, 0x39, 0xf9, 0x5, 0xc5, 0x35, 0xf5}, |
992 | { 0x8a, 0x4a, 0xba, 0x7a, 0x86, 0x46, 0xb6, 0x76, |
993 | 0x89, 0x49, 0xb9, 0x79, 0x85, 0x45, 0xb5, 0x75}, |
994 | { 0x2a, 0xea, 0x1a, 0xda, 0x26, 0xe6, 0x16, 0xd6, |
995 | 0x29, 0xe9, 0x19, 0xd9, 0x25, 0xe5, 0x15, 0xd5}, |
996 | { 0xaa, 0x6a, 0x9a, 0x5a, 0xa6, 0x66, 0x96, 0x56, |
997 | 0xa9, 0x69, 0x99, 0x59, 0xa5, 0x65, 0x95, 0x55} |
998 | }; |
999 | |
1000 | #define ARGB_COMBINE_ALPHA(argb, alpha) \ |
1001 | ((((argb >> 24) * alpha) >> 8) << 24) | (argb & 0x00ffffff) |
1002 | |
1003 | |
1004 | #if Q_PROCESSOR_WORDSIZE == 8 // 64-bit versions |
1005 | #define AMIX(mask) (qMin(((quint64(s)&mask) + (quint64(d)&mask)), quint64(mask))) |
1006 | #define MIX(mask) (qMin(((quint64(s)&mask) + (quint64(d)&mask)), quint64(mask))) |
1007 | #else // 32 bits |
1008 | // The mask for alpha can overflow over 32 bits |
1009 | #define AMIX(mask) quint32(qMin(((quint64(s)&mask) + (quint64(d)&mask)), quint64(mask))) |
1010 | #define MIX(mask) (qMin(((quint32(s)&mask) + (quint32(d)&mask)), quint32(mask))) |
1011 | #endif |
1012 | |
1013 | inline uint comp_func_Plus_one_pixel_const_alpha(uint d, const uint s, const uint const_alpha, const uint one_minus_const_alpha) |
1014 | { |
1015 | const uint result = uint(AMIX(AMASK) | MIX(RMASK) | MIX(GMASK) | MIX(BMASK)); |
1016 | return INTERPOLATE_PIXEL_255(x: result, a: const_alpha, y: d, b: one_minus_const_alpha); |
1017 | } |
1018 | |
1019 | inline uint comp_func_Plus_one_pixel(uint d, const uint s) |
1020 | { |
1021 | const uint result = uint(AMIX(AMASK) | MIX(RMASK) | MIX(GMASK) | MIX(BMASK)); |
1022 | return result; |
1023 | } |
1024 | |
1025 | #undef MIX |
1026 | #undef AMIX |
1027 | |
1028 | // must be multiple of 4 for easier SIMD implementations |
1029 | static constexpr int BufferSize = 2048; |
1030 | |
1031 | // A buffer of intermediate results used by simple bilinear scaling. |
1032 | struct IntermediateBuffer |
1033 | { |
1034 | // The idea is first to do the interpolation between the row s1 and the row s2 |
1035 | // into this intermediate buffer, then later interpolate between two pixel of this buffer. |
1036 | // |
1037 | // buffer_rb is a buffer of red-blue component of the pixel, in the form 0x00RR00BB |
1038 | // buffer_ag is the alpha-green component of the pixel, in the form 0x00AA00GG |
1039 | // +1 for the last pixel to interpolate with, and +1 for rounding errors. |
1040 | quint32 buffer_rb[BufferSize+2]; |
1041 | quint32 buffer_ag[BufferSize+2]; |
1042 | }; |
1043 | |
1044 | QT_END_NAMESPACE |
1045 | |
1046 | #endif // QDRAWHELPER_P_H |
1047 | |