1 | /**************************************************************************** |
2 | ** |
3 | ** Copyright (C) 2018 The Qt Company Ltd. |
4 | ** Copyright (C) 2018 Intel Corporation. |
5 | ** Contact: https://www.qt.io/licensing/ |
6 | ** |
7 | ** This file is part of the QtGui module of the Qt Toolkit. |
8 | ** |
9 | ** $QT_BEGIN_LICENSE:LGPL$ |
10 | ** Commercial License Usage |
11 | ** Licensees holding valid commercial Qt licenses may use this file in |
12 | ** accordance with the commercial license agreement provided with the |
13 | ** Software or, alternatively, in accordance with the terms contained in |
14 | ** a written agreement between you and The Qt Company. For licensing terms |
15 | ** and conditions see https://www.qt.io/terms-conditions. For further |
16 | ** information use the contact form at https://www.qt.io/contact-us. |
17 | ** |
18 | ** GNU Lesser General Public License Usage |
19 | ** Alternatively, this file may be used under the terms of the GNU Lesser |
20 | ** General Public License version 3 as published by the Free Software |
21 | ** Foundation and appearing in the file LICENSE.LGPL3 included in the |
22 | ** packaging of this file. Please review the following information to |
23 | ** ensure the GNU Lesser General Public License version 3 requirements |
24 | ** will be met: https://www.gnu.org/licenses/lgpl-3.0.html. |
25 | ** |
26 | ** GNU General Public License Usage |
27 | ** Alternatively, this file may be used under the terms of the GNU |
28 | ** General Public License version 2.0 or (at your option) the GNU General |
29 | ** Public license version 3 or any later version approved by the KDE Free |
30 | ** Qt Foundation. The licenses are as published by the Free Software |
31 | ** Foundation and appearing in the file LICENSE.GPL2 and LICENSE.GPL3 |
32 | ** included in the packaging of this file. Please review the following |
33 | ** information to ensure the GNU General Public License requirements will |
34 | ** be met: https://www.gnu.org/licenses/gpl-2.0.html and |
35 | ** https://www.gnu.org/licenses/gpl-3.0.html. |
36 | ** |
37 | ** $QT_END_LICENSE$ |
38 | ** |
39 | ****************************************************************************/ |
40 | |
41 | #include <qglobal.h> |
42 | |
43 | #include <qstylehints.h> |
44 | #include <qguiapplication.h> |
45 | #include <qatomic.h> |
46 | #include <private/qcolortrclut_p.h> |
47 | #include <private/qdrawhelper_p.h> |
48 | #include <private/qdrawhelper_x86_p.h> |
49 | #include <private/qdrawingprimitive_sse2_p.h> |
50 | #include <private/qdrawhelper_neon_p.h> |
51 | #if defined(QT_COMPILER_SUPPORTS_MIPS_DSP) || defined(QT_COMPILER_SUPPORTS_MIPS_DSPR2) |
52 | #include <private/qdrawhelper_mips_dsp_p.h> |
53 | #endif |
54 | #include <private/qguiapplication_p.h> |
55 | #include <private/qpaintengine_raster_p.h> |
56 | #include <private/qpainter_p.h> |
57 | #include <private/qpixellayout_p.h> |
58 | #include <private/qrgba64_p.h> |
59 | #include <qendian.h> |
60 | #include <qloggingcategory.h> |
61 | #include <qmath.h> |
62 | |
63 | QT_BEGIN_NAMESPACE |
64 | |
65 | Q_LOGGING_CATEGORY(lcQtGuiDrawHelper, "qt.gui.drawhelper" ) |
66 | |
67 | #define MASK(src, a) src = BYTE_MUL(src, a) |
68 | |
69 | /* |
70 | constants and structures |
71 | */ |
72 | |
73 | enum { |
74 | fixed_scale = 1 << 16, |
75 | half_point = 1 << 15 |
76 | }; |
77 | |
78 | #if QT_CONFIG(raster_64bit) |
79 | static void convertRGBA64ToRGBA64PM(QRgba64 *buffer, int count) |
80 | { |
81 | for (int i = 0; i < count; ++i) |
82 | buffer[i] = buffer[i].premultiplied(); |
83 | } |
84 | |
85 | static void convertRGBA64PMToRGBA64PM(QRgba64 *, int) |
86 | { |
87 | } |
88 | #endif |
89 | |
90 | /* |
91 | Destination fetch. This is simple as we don't have to do bounds checks or |
92 | transformations |
93 | */ |
94 | |
95 | static uint * QT_FASTCALL destFetchMono(uint *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length) |
96 | { |
97 | uchar *Q_DECL_RESTRICT data = (uchar *)rasterBuffer->scanLine(y); |
98 | uint *start = buffer; |
99 | const uint *end = buffer + length; |
100 | while (buffer < end) { |
101 | *buffer = data[x>>3] & (0x80 >> (x & 7)) ? rasterBuffer->destColor1 : rasterBuffer->destColor0; |
102 | ++buffer; |
103 | ++x; |
104 | } |
105 | return start; |
106 | } |
107 | |
108 | static uint * QT_FASTCALL destFetchMonoLsb(uint *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length) |
109 | { |
110 | uchar *Q_DECL_RESTRICT data = (uchar *)rasterBuffer->scanLine(y); |
111 | uint *start = buffer; |
112 | const uint *end = buffer + length; |
113 | while (buffer < end) { |
114 | *buffer = data[x>>3] & (0x1 << (x & 7)) ? rasterBuffer->destColor1 : rasterBuffer->destColor0; |
115 | ++buffer; |
116 | ++x; |
117 | } |
118 | return start; |
119 | } |
120 | |
121 | static uint * QT_FASTCALL destFetchARGB32P(uint *, QRasterBuffer *rasterBuffer, int x, int y, int) |
122 | { |
123 | return (uint *)rasterBuffer->scanLine(y) + x; |
124 | } |
125 | |
126 | static uint * QT_FASTCALL destFetchRGB16(uint *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length) |
127 | { |
128 | const ushort *Q_DECL_RESTRICT data = (const ushort *)rasterBuffer->scanLine(y) + x; |
129 | for (int i = 0; i < length; ++i) |
130 | buffer[i] = qConvertRgb16To32(data[i]); |
131 | return buffer; |
132 | } |
133 | |
134 | static uint *QT_FASTCALL destFetch(uint *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length) |
135 | { |
136 | const QPixelLayout *layout = &qPixelLayouts[rasterBuffer->format]; |
137 | return const_cast<uint *>(layout->fetchToARGB32PM(buffer, rasterBuffer->scanLine(y), x, length, nullptr, nullptr)); |
138 | } |
139 | |
140 | static uint *QT_FASTCALL destFetchUndefined(uint *buffer, QRasterBuffer *, int, int, int) |
141 | { |
142 | return buffer; |
143 | } |
144 | |
145 | static DestFetchProc destFetchProc[QImage::NImageFormats] = |
146 | { |
147 | nullptr, // Format_Invalid |
148 | destFetchMono, // Format_Mono, |
149 | destFetchMonoLsb, // Format_MonoLSB |
150 | nullptr, // Format_Indexed8 |
151 | destFetchARGB32P, // Format_RGB32 |
152 | destFetch, // Format_ARGB32, |
153 | destFetchARGB32P, // Format_ARGB32_Premultiplied |
154 | destFetchRGB16, // Format_RGB16 |
155 | destFetch, // Format_ARGB8565_Premultiplied |
156 | destFetch, // Format_RGB666 |
157 | destFetch, // Format_ARGB6666_Premultiplied |
158 | destFetch, // Format_RGB555 |
159 | destFetch, // Format_ARGB8555_Premultiplied |
160 | destFetch, // Format_RGB888 |
161 | destFetch, // Format_RGB444 |
162 | destFetch, // Format_ARGB4444_Premultiplied |
163 | destFetch, // Format_RGBX8888 |
164 | destFetch, // Format_RGBA8888 |
165 | destFetch, // Format_RGBA8888_Premultiplied |
166 | destFetch, // Format_BGR30 |
167 | destFetch, // Format_A2BGR30_Premultiplied |
168 | destFetch, // Format_RGB30 |
169 | destFetch, // Format_A2RGB30_Premultiplied |
170 | destFetch, // Format_Alpha8 |
171 | destFetch, // Format_Grayscale8 |
172 | destFetch, // Format_RGBX64 |
173 | destFetch, // Format_RGBA64 |
174 | destFetch, // Format_RGBA64_Premultiplied |
175 | destFetch, // Format_Grayscale16 |
176 | destFetch, // Format_BGR888 |
177 | }; |
178 | |
179 | #if QT_CONFIG(raster_64bit) |
180 | static QRgba64 *QT_FASTCALL destFetch64(QRgba64 *buffer, QRasterBuffer *rasterBuffer, int x, int y, int length) |
181 | { |
182 | const QPixelLayout *layout = &qPixelLayouts[rasterBuffer->format]; |
183 | return const_cast<QRgba64 *>(layout->fetchToRGBA64PM(buffer, rasterBuffer->scanLine(y), x, length, nullptr, nullptr)); |
184 | } |
185 | |
186 | static QRgba64 * QT_FASTCALL destFetchRGB64(QRgba64 *, QRasterBuffer *rasterBuffer, int x, int y, int) |
187 | { |
188 | return (QRgba64 *)rasterBuffer->scanLine(y) + x; |
189 | } |
190 | |
191 | static QRgba64 * QT_FASTCALL destFetch64Undefined(QRgba64 *buffer, QRasterBuffer *, int, int, int) |
192 | { |
193 | return buffer; |
194 | } |
195 | |
196 | static DestFetchProc64 destFetchProc64[QImage::NImageFormats] = |
197 | { |
198 | nullptr, // Format_Invalid |
199 | nullptr, // Format_Mono, |
200 | nullptr, // Format_MonoLSB |
201 | nullptr, // Format_Indexed8 |
202 | destFetch64, // Format_RGB32 |
203 | destFetch64, // Format_ARGB32, |
204 | destFetch64, // Format_ARGB32_Premultiplied |
205 | destFetch64, // Format_RGB16 |
206 | destFetch64, // Format_ARGB8565_Premultiplied |
207 | destFetch64, // Format_RGB666 |
208 | destFetch64, // Format_ARGB6666_Premultiplied |
209 | destFetch64, // Format_RGB555 |
210 | destFetch64, // Format_ARGB8555_Premultiplied |
211 | destFetch64, // Format_RGB888 |
212 | destFetch64, // Format_RGB444 |
213 | destFetch64, // Format_ARGB4444_Premultiplied |
214 | destFetch64, // Format_RGBX8888 |
215 | destFetch64, // Format_RGBA8888 |
216 | destFetch64, // Format_RGBA8888_Premultiplied |
217 | destFetch64, // Format_BGR30 |
218 | destFetch64, // Format_A2BGR30_Premultiplied |
219 | destFetch64, // Format_RGB30 |
220 | destFetch64, // Format_A2RGB30_Premultiplied |
221 | destFetch64, // Format_Alpha8 |
222 | destFetch64, // Format_Grayscale8 |
223 | destFetchRGB64, // Format_RGBX64 |
224 | destFetch64, // Format_RGBA64 |
225 | destFetchRGB64, // Format_RGBA64_Premultiplied |
226 | destFetch64, // Format_Grayscale16 |
227 | destFetch64, // Format_BGR888 |
228 | }; |
229 | #endif |
230 | |
231 | /* |
232 | Returns the color in the mono destination color table |
233 | that is the "nearest" to /color/. |
234 | */ |
235 | static inline QRgb findNearestColor(QRgb color, QRasterBuffer *rbuf) |
236 | { |
237 | QRgb color_0 = qPremultiply(rbuf->destColor0); |
238 | QRgb color_1 = qPremultiply(rbuf->destColor1); |
239 | color = qPremultiply(color); |
240 | |
241 | int r = qRed(color); |
242 | int g = qGreen(color); |
243 | int b = qBlue(color); |
244 | int rx, gx, bx; |
245 | int dist_0, dist_1; |
246 | |
247 | rx = r - qRed(color_0); |
248 | gx = g - qGreen(color_0); |
249 | bx = b - qBlue(color_0); |
250 | dist_0 = rx*rx + gx*gx + bx*bx; |
251 | |
252 | rx = r - qRed(color_1); |
253 | gx = g - qGreen(color_1); |
254 | bx = b - qBlue(color_1); |
255 | dist_1 = rx*rx + gx*gx + bx*bx; |
256 | |
257 | if (dist_0 < dist_1) |
258 | return color_0; |
259 | return color_1; |
260 | } |
261 | |
262 | /* |
263 | Destination store. |
264 | */ |
265 | |
266 | static void QT_FASTCALL destStoreMono(QRasterBuffer *rasterBuffer, int x, int y, const uint *buffer, int length) |
267 | { |
268 | uchar *Q_DECL_RESTRICT data = (uchar *)rasterBuffer->scanLine(y); |
269 | if (rasterBuffer->monoDestinationWithClut) { |
270 | for (int i = 0; i < length; ++i) { |
271 | if (buffer[i] == rasterBuffer->destColor0) { |
272 | data[x >> 3] &= ~(0x80 >> (x & 7)); |
273 | } else if (buffer[i] == rasterBuffer->destColor1) { |
274 | data[x >> 3] |= 0x80 >> (x & 7); |
275 | } else if (findNearestColor(buffer[i], rasterBuffer) == rasterBuffer->destColor0) { |
276 | data[x >> 3] &= ~(0x80 >> (x & 7)); |
277 | } else { |
278 | data[x >> 3] |= 0x80 >> (x & 7); |
279 | } |
280 | ++x; |
281 | } |
282 | } else { |
283 | for (int i = 0; i < length; ++i) { |
284 | if (qGray(buffer[i]) < int(qt_bayer_matrix[y & 15][x & 15])) |
285 | data[x >> 3] |= 0x80 >> (x & 7); |
286 | else |
287 | data[x >> 3] &= ~(0x80 >> (x & 7)); |
288 | ++x; |
289 | } |
290 | } |
291 | } |
292 | |
293 | static void QT_FASTCALL destStoreMonoLsb(QRasterBuffer *rasterBuffer, int x, int y, const uint *buffer, int length) |
294 | { |
295 | uchar *Q_DECL_RESTRICT data = (uchar *)rasterBuffer->scanLine(y); |
296 | if (rasterBuffer->monoDestinationWithClut) { |
297 | for (int i = 0; i < length; ++i) { |
298 | if (buffer[i] == rasterBuffer->destColor0) { |
299 | data[x >> 3] &= ~(1 << (x & 7)); |
300 | } else if (buffer[i] == rasterBuffer->destColor1) { |
301 | data[x >> 3] |= 1 << (x & 7); |
302 | } else if (findNearestColor(buffer[i], rasterBuffer) == rasterBuffer->destColor0) { |
303 | data[x >> 3] &= ~(1 << (x & 7)); |
304 | } else { |
305 | data[x >> 3] |= 1 << (x & 7); |
306 | } |
307 | ++x; |
308 | } |
309 | } else { |
310 | for (int i = 0; i < length; ++i) { |
311 | if (qGray(buffer[i]) < int(qt_bayer_matrix[y & 15][x & 15])) |
312 | data[x >> 3] |= 1 << (x & 7); |
313 | else |
314 | data[x >> 3] &= ~(1 << (x & 7)); |
315 | ++x; |
316 | } |
317 | } |
318 | } |
319 | |
320 | static void QT_FASTCALL destStoreRGB16(QRasterBuffer *rasterBuffer, int x, int y, const uint *buffer, int length) |
321 | { |
322 | quint16 *data = (quint16*)rasterBuffer->scanLine(y) + x; |
323 | for (int i = 0; i < length; ++i) |
324 | data[i] = qConvertRgb32To16(buffer[i]); |
325 | } |
326 | |
327 | static void QT_FASTCALL destStore(QRasterBuffer *rasterBuffer, int x, int y, const uint *buffer, int length) |
328 | { |
329 | const QPixelLayout *layout = &qPixelLayouts[rasterBuffer->format]; |
330 | ConvertAndStorePixelsFunc store = layout->storeFromARGB32PM; |
331 | if (!layout->premultiplied && !layout->hasAlphaChannel) |
332 | store = layout->storeFromRGB32; |
333 | uchar *dest = rasterBuffer->scanLine(y); |
334 | store(dest, buffer, x, length, nullptr, nullptr); |
335 | } |
336 | |
337 | static DestStoreProc destStoreProc[QImage::NImageFormats] = |
338 | { |
339 | nullptr, // Format_Invalid |
340 | destStoreMono, // Format_Mono, |
341 | destStoreMonoLsb, // Format_MonoLSB |
342 | nullptr, // Format_Indexed8 |
343 | nullptr, // Format_RGB32 |
344 | destStore, // Format_ARGB32, |
345 | nullptr, // Format_ARGB32_Premultiplied |
346 | destStoreRGB16, // Format_RGB16 |
347 | destStore, // Format_ARGB8565_Premultiplied |
348 | destStore, // Format_RGB666 |
349 | destStore, // Format_ARGB6666_Premultiplied |
350 | destStore, // Format_RGB555 |
351 | destStore, // Format_ARGB8555_Premultiplied |
352 | destStore, // Format_RGB888 |
353 | destStore, // Format_RGB444 |
354 | destStore, // Format_ARGB4444_Premultiplied |
355 | destStore, // Format_RGBX8888 |
356 | destStore, // Format_RGBA8888 |
357 | destStore, // Format_RGBA8888_Premultiplied |
358 | destStore, // Format_BGR30 |
359 | destStore, // Format_A2BGR30_Premultiplied |
360 | destStore, // Format_RGB30 |
361 | destStore, // Format_A2RGB30_Premultiplied |
362 | destStore, // Format_Alpha8 |
363 | destStore, // Format_Grayscale8 |
364 | destStore, // Format_RGBX64 |
365 | destStore, // Format_RGBA64 |
366 | destStore, // Format_RGBA64_Premultiplied |
367 | destStore, // Format_Grayscale16 |
368 | destStore, // Format_BGR888 |
369 | }; |
370 | |
371 | #if QT_CONFIG(raster_64bit) |
372 | static void QT_FASTCALL destStore64(QRasterBuffer *rasterBuffer, int x, int y, const QRgba64 *buffer, int length) |
373 | { |
374 | auto store = qStoreFromRGBA64PM[rasterBuffer->format]; |
375 | uchar *dest = rasterBuffer->scanLine(y); |
376 | store(dest, buffer, x, length, nullptr, nullptr); |
377 | } |
378 | |
379 | static void QT_FASTCALL destStore64RGBA64(QRasterBuffer *rasterBuffer, int x, int y, const QRgba64 *buffer, int length) |
380 | { |
381 | QRgba64 *dest = reinterpret_cast<QRgba64*>(rasterBuffer->scanLine(y)) + x; |
382 | for (int i = 0; i < length; ++i) { |
383 | dest[i] = buffer[i].unpremultiplied(); |
384 | } |
385 | } |
386 | |
387 | static DestStoreProc64 destStoreProc64[QImage::NImageFormats] = |
388 | { |
389 | nullptr, // Format_Invalid |
390 | nullptr, // Format_Mono, |
391 | nullptr, // Format_MonoLSB |
392 | nullptr, // Format_Indexed8 |
393 | destStore64, // Format_RGB32 |
394 | destStore64, // Format_ARGB32, |
395 | destStore64, // Format_ARGB32_Premultiplied |
396 | destStore64, // Format_RGB16 |
397 | destStore64, // Format_ARGB8565_Premultiplied |
398 | destStore64, // Format_RGB666 |
399 | destStore64, // Format_ARGB6666_Premultiplied |
400 | destStore64, // Format_RGB555 |
401 | destStore64, // Format_ARGB8555_Premultiplied |
402 | destStore64, // Format_RGB888 |
403 | destStore64, // Format_RGB444 |
404 | destStore64, // Format_ARGB4444_Premultiplied |
405 | destStore64, // Format_RGBX8888 |
406 | destStore64, // Format_RGBA8888 |
407 | destStore64, // Format_RGBA8888_Premultiplied |
408 | destStore64, // Format_BGR30 |
409 | destStore64, // Format_A2BGR30_Premultiplied |
410 | destStore64, // Format_RGB30 |
411 | destStore64, // Format_A2RGB30_Premultiplied |
412 | destStore64, // Format_Alpha8 |
413 | destStore64, // Format_Grayscale8 |
414 | nullptr, // Format_RGBX64 |
415 | destStore64RGBA64, // Format_RGBA64 |
416 | nullptr, // Format_RGBA64_Premultiplied |
417 | destStore64, // Format_Grayscale16 |
418 | destStore64, // Format_BGR888 |
419 | }; |
420 | #endif |
421 | |
422 | /* |
423 | Source fetches |
424 | |
425 | This is a bit more complicated, as we need several fetch routines for every surface type |
426 | |
427 | We need 5 fetch methods per surface type: |
428 | untransformed |
429 | transformed (tiled and not tiled) |
430 | transformed bilinear (tiled and not tiled) |
431 | |
432 | We don't need bounds checks for untransformed, but we need them for the other ones. |
433 | |
434 | The generic implementation does pixel by pixel fetches |
435 | */ |
436 | |
437 | enum TextureBlendType { |
438 | BlendUntransformed, |
439 | BlendTiled, |
440 | BlendTransformed, |
441 | BlendTransformedTiled, |
442 | BlendTransformedBilinear, |
443 | BlendTransformedBilinearTiled, |
444 | NBlendTypes |
445 | }; |
446 | |
447 | static const uint *QT_FASTCALL fetchUntransformed(uint *buffer, const Operator *, |
448 | const QSpanData *data, int y, int x, int length) |
449 | { |
450 | const QPixelLayout *layout = &qPixelLayouts[data->texture.format]; |
451 | return layout->fetchToARGB32PM(buffer, data->texture.scanLine(y), x, length, data->texture.colorTable, nullptr); |
452 | } |
453 | |
454 | static const uint *QT_FASTCALL fetchUntransformedARGB32PM(uint *, const Operator *, |
455 | const QSpanData *data, int y, int x, int) |
456 | { |
457 | const uchar *scanLine = data->texture.scanLine(y); |
458 | return reinterpret_cast<const uint *>(scanLine) + x; |
459 | } |
460 | |
461 | static const uint *QT_FASTCALL fetchUntransformedRGB16(uint *buffer, const Operator *, |
462 | const QSpanData *data, int y, int x, |
463 | int length) |
464 | { |
465 | const quint16 *scanLine = (const quint16 *)data->texture.scanLine(y) + x; |
466 | for (int i = 0; i < length; ++i) |
467 | buffer[i] = qConvertRgb16To32(scanLine[i]); |
468 | return buffer; |
469 | } |
470 | |
471 | #if QT_CONFIG(raster_64bit) |
472 | static const QRgba64 *QT_FASTCALL fetchUntransformed64(QRgba64 *buffer, const Operator *, |
473 | const QSpanData *data, int y, int x, int length) |
474 | { |
475 | const QPixelLayout *layout = &qPixelLayouts[data->texture.format]; |
476 | return layout->fetchToRGBA64PM(buffer, data->texture.scanLine(y), x, length, data->texture.colorTable, nullptr); |
477 | } |
478 | |
479 | static const QRgba64 *QT_FASTCALL fetchUntransformedRGBA64PM(QRgba64 *, const Operator *, |
480 | const QSpanData *data, int y, int x, int) |
481 | { |
482 | const uchar *scanLine = data->texture.scanLine(y); |
483 | return reinterpret_cast<const QRgba64 *>(scanLine) + x; |
484 | } |
485 | #endif |
486 | |
487 | template<TextureBlendType blendType> |
488 | inline void fetchTransformed_pixelBounds(int max, int l1, int l2, int &v) |
489 | { |
490 | static_assert(blendType == BlendTransformed || blendType == BlendTransformedTiled); |
491 | if (blendType == BlendTransformedTiled) { |
492 | if (v < 0 || v >= max) { |
493 | v %= max; |
494 | if (v < 0) v += max; |
495 | } |
496 | } else { |
497 | v = qBound(l1, v, l2); |
498 | } |
499 | } |
500 | |
501 | static inline bool canUseFastMatrixPath(const qreal cx, const qreal cy, const qsizetype length, const QSpanData *data) |
502 | { |
503 | if (Q_UNLIKELY(!data->fast_matrix)) |
504 | return false; |
505 | |
506 | qreal fx = (data->m21 * cy + data->m11 * cx + data->dx) * fixed_scale; |
507 | qreal fy = (data->m22 * cy + data->m12 * cx + data->dy) * fixed_scale; |
508 | qreal minc = std::min(fx, fy); |
509 | qreal maxc = std::max(fx, fy); |
510 | fx += std::trunc(data->m11 * fixed_scale) * length; |
511 | fy += std::trunc(data->m12 * fixed_scale) * length; |
512 | minc = std::min(minc, std::min(fx, fy)); |
513 | maxc = std::max(maxc, std::max(fx, fy)); |
514 | |
515 | return minc >= std::numeric_limits<int>::min() && maxc <= std::numeric_limits<int>::max(); |
516 | } |
517 | |
518 | template<TextureBlendType blendType, QPixelLayout::BPP bpp, typename T> |
519 | static void QT_FASTCALL fetchTransformed_fetcher(T *buffer, const QSpanData *data, |
520 | int y, int x, int length) |
521 | { |
522 | static_assert(blendType == BlendTransformed || blendType == BlendTransformedTiled); |
523 | const QTextureData &image = data->texture; |
524 | |
525 | const qreal cx = x + qreal(0.5); |
526 | const qreal cy = y + qreal(0.5); |
527 | |
528 | constexpr bool useFetch = (bpp < QPixelLayout::BPP32) && sizeof(T) == sizeof(uint); |
529 | const QPixelLayout *layout = &qPixelLayouts[data->texture.format]; |
530 | if (!useFetch) |
531 | Q_ASSERT(layout->bpp == bpp); |
532 | // When templated 'fetch' should be inlined at compile time: |
533 | const FetchPixelFunc fetch = (bpp == QPixelLayout::BPPNone) ? qFetchPixelTable[layout->bpp] : FetchPixelFunc(qFetchPixel<bpp>); |
534 | |
535 | if (canUseFastMatrixPath(cx, cy, length, data)) { |
536 | // The increment pr x in the scanline |
537 | int fdx = (int)(data->m11 * fixed_scale); |
538 | int fdy = (int)(data->m12 * fixed_scale); |
539 | |
540 | int fx = int((data->m21 * cy |
541 | + data->m11 * cx + data->dx) * fixed_scale); |
542 | int fy = int((data->m22 * cy |
543 | + data->m12 * cx + data->dy) * fixed_scale); |
544 | |
545 | if (fdy == 0) { // simple scale, no rotation or shear |
546 | int py = (fy >> 16); |
547 | fetchTransformed_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, py); |
548 | const uchar *src = image.scanLine(py); |
549 | |
550 | int i = 0; |
551 | if (blendType == BlendTransformed) { |
552 | int fastLen = length; |
553 | if (fdx > 0) |
554 | fastLen = qMin(fastLen, int((qint64(image.x2 - 1) * fixed_scale - fx) / fdx)); |
555 | else if (fdx < 0) |
556 | fastLen = qMin(fastLen, int((qint64(image.x1) * fixed_scale - fx) / fdx)); |
557 | |
558 | for (; i < fastLen; ++i) { |
559 | int x1 = (fx >> 16); |
560 | int x2 = x1; |
561 | fetchTransformed_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1); |
562 | if (x1 == x2) |
563 | break; |
564 | if (useFetch) |
565 | buffer[i] = fetch(src, x1); |
566 | else |
567 | buffer[i] = reinterpret_cast<const T*>(src)[x1]; |
568 | fx += fdx; |
569 | } |
570 | |
571 | for (; i < fastLen; ++i) { |
572 | int px = (fx >> 16); |
573 | if (useFetch) |
574 | buffer[i] = fetch(src, px); |
575 | else |
576 | buffer[i] = reinterpret_cast<const T*>(src)[px]; |
577 | fx += fdx; |
578 | } |
579 | } |
580 | |
581 | for (; i < length; ++i) { |
582 | int px = (fx >> 16); |
583 | fetchTransformed_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, px); |
584 | if (useFetch) |
585 | buffer[i] = fetch(src, px); |
586 | else |
587 | buffer[i] = reinterpret_cast<const T*>(src)[px]; |
588 | fx += fdx; |
589 | } |
590 | } else { // rotation or shear |
591 | int i = 0; |
592 | if (blendType == BlendTransformed) { |
593 | int fastLen = length; |
594 | if (fdx > 0) |
595 | fastLen = qMin(fastLen, int((qint64(image.x2 - 1) * fixed_scale - fx) / fdx)); |
596 | else if (fdx < 0) |
597 | fastLen = qMin(fastLen, int((qint64(image.x1) * fixed_scale - fx) / fdx)); |
598 | if (fdy > 0) |
599 | fastLen = qMin(fastLen, int((qint64(image.y2 - 1) * fixed_scale - fy) / fdy)); |
600 | else if (fdy < 0) |
601 | fastLen = qMin(fastLen, int((qint64(image.y1) * fixed_scale - fy) / fdy)); |
602 | |
603 | for (; i < fastLen; ++i) { |
604 | int x1 = (fx >> 16); |
605 | int y1 = (fy >> 16); |
606 | int x2 = x1; |
607 | int y2 = y1; |
608 | fetchTransformed_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1); |
609 | fetchTransformed_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1); |
610 | if (x1 == x2 && y1 == y2) |
611 | break; |
612 | if (useFetch) |
613 | buffer[i] = fetch(image.scanLine(y1), x1); |
614 | else |
615 | buffer[i] = reinterpret_cast<const T*>(image.scanLine(y1))[x1]; |
616 | fx += fdx; |
617 | fy += fdy; |
618 | } |
619 | |
620 | for (; i < fastLen; ++i) { |
621 | int px = (fx >> 16); |
622 | int py = (fy >> 16); |
623 | if (useFetch) |
624 | buffer[i] = fetch(image.scanLine(py), px); |
625 | else |
626 | buffer[i] = reinterpret_cast<const T*>(image.scanLine(py))[px]; |
627 | fx += fdx; |
628 | fy += fdy; |
629 | } |
630 | } |
631 | |
632 | for (; i < length; ++i) { |
633 | int px = (fx >> 16); |
634 | int py = (fy >> 16); |
635 | fetchTransformed_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, px); |
636 | fetchTransformed_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, py); |
637 | if (useFetch) |
638 | buffer[i] = fetch(image.scanLine(py), px); |
639 | else |
640 | buffer[i] = reinterpret_cast<const T*>(image.scanLine(py))[px]; |
641 | fx += fdx; |
642 | fy += fdy; |
643 | } |
644 | } |
645 | } else { |
646 | const qreal fdx = data->m11; |
647 | const qreal fdy = data->m12; |
648 | const qreal fdw = data->m13; |
649 | |
650 | qreal fx = data->m21 * cy + data->m11 * cx + data->dx; |
651 | qreal fy = data->m22 * cy + data->m12 * cx + data->dy; |
652 | qreal fw = data->m23 * cy + data->m13 * cx + data->m33; |
653 | |
654 | T *const end = buffer + length; |
655 | T *b = buffer; |
656 | while (b < end) { |
657 | const qreal iw = fw == 0 ? 1 : 1 / fw; |
658 | const qreal tx = fx * iw; |
659 | const qreal ty = fy * iw; |
660 | int px = qFloor(tx); |
661 | int py = qFloor(ty); |
662 | |
663 | fetchTransformed_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, py); |
664 | fetchTransformed_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, px); |
665 | if (useFetch) |
666 | *b = fetch(image.scanLine(py), px); |
667 | else |
668 | *b = reinterpret_cast<const T*>(image.scanLine(py))[px]; |
669 | |
670 | fx += fdx; |
671 | fy += fdy; |
672 | fw += fdw; |
673 | //force increment to avoid /0 |
674 | if (!fw) { |
675 | fw += fdw; |
676 | } |
677 | ++b; |
678 | } |
679 | } |
680 | } |
681 | |
682 | template<TextureBlendType blendType, QPixelLayout::BPP bpp> |
683 | static const uint *QT_FASTCALL fetchTransformed(uint *buffer, const Operator *, const QSpanData *data, |
684 | int y, int x, int length) |
685 | { |
686 | static_assert(blendType == BlendTransformed || blendType == BlendTransformedTiled); |
687 | const QPixelLayout *layout = &qPixelLayouts[data->texture.format]; |
688 | fetchTransformed_fetcher<blendType, bpp, uint>(buffer, data, y, x, length); |
689 | layout->convertToARGB32PM(buffer, length, data->texture.colorTable); |
690 | return buffer; |
691 | } |
692 | |
693 | #if QT_CONFIG(raster_64bit) |
694 | template<TextureBlendType blendType> /* either BlendTransformed or BlendTransformedTiled */ |
695 | static const QRgba64 *QT_FASTCALL fetchTransformed64(QRgba64 *buffer, const Operator *, const QSpanData *data, |
696 | int y, int x, int length) |
697 | { |
698 | const QPixelLayout *layout = &qPixelLayouts[data->texture.format]; |
699 | if (layout->bpp != QPixelLayout::BPP64) { |
700 | uint buffer32[BufferSize]; |
701 | Q_ASSERT(length <= BufferSize); |
702 | if (layout->bpp == QPixelLayout::BPP32) |
703 | fetchTransformed_fetcher<blendType, QPixelLayout::BPP32, uint>(buffer32, data, y, x, length); |
704 | else |
705 | fetchTransformed_fetcher<blendType, QPixelLayout::BPPNone, uint>(buffer32, data, y, x, length); |
706 | return layout->convertToRGBA64PM(buffer, buffer32, length, data->texture.colorTable, nullptr); |
707 | } |
708 | |
709 | fetchTransformed_fetcher<blendType, QPixelLayout::BPP64, QRgba64>(buffer, data, y, x, length); |
710 | if (data->texture.format == QImage::Format_RGBA64) |
711 | convertRGBA64ToRGBA64PM(buffer, length); |
712 | return buffer; |
713 | } |
714 | #endif |
715 | |
716 | /** \internal |
717 | interpolate 4 argb pixels with the distx and disty factor. |
718 | distx and disty must be between 0 and 16 |
719 | */ |
720 | static inline uint interpolate_4_pixels_16(uint tl, uint tr, uint bl, uint br, uint distx, uint disty) |
721 | { |
722 | uint distxy = distx * disty; |
723 | //idistx * disty = (16-distx) * disty = 16*disty - distxy |
724 | //idistx * idisty = (16-distx) * (16-disty) = 16*16 - 16*distx -16*disty + distxy |
725 | uint tlrb = (tl & 0x00ff00ff) * (16*16 - 16*distx - 16*disty + distxy); |
726 | uint tlag = ((tl & 0xff00ff00) >> 8) * (16*16 - 16*distx - 16*disty + distxy); |
727 | uint trrb = ((tr & 0x00ff00ff) * (distx*16 - distxy)); |
728 | uint trag = (((tr & 0xff00ff00) >> 8) * (distx*16 - distxy)); |
729 | uint blrb = ((bl & 0x00ff00ff) * (disty*16 - distxy)); |
730 | uint blag = (((bl & 0xff00ff00) >> 8) * (disty*16 - distxy)); |
731 | uint brrb = ((br & 0x00ff00ff) * (distxy)); |
732 | uint brag = (((br & 0xff00ff00) >> 8) * (distxy)); |
733 | return (((tlrb + trrb + blrb + brrb) >> 8) & 0x00ff00ff) | ((tlag + trag + blag + brag) & 0xff00ff00); |
734 | } |
735 | |
736 | #if defined(__SSE2__) |
737 | #define interpolate_4_pixels_16_sse2(tl, tr, bl, br, distx, disty, colorMask, v_256, b) \ |
738 | { \ |
739 | const __m128i dxdy = _mm_mullo_epi16 (distx, disty); \ |
740 | const __m128i distx_ = _mm_slli_epi16(distx, 4); \ |
741 | const __m128i disty_ = _mm_slli_epi16(disty, 4); \ |
742 | const __m128i idxidy = _mm_add_epi16(dxdy, _mm_sub_epi16(v_256, _mm_add_epi16(distx_, disty_))); \ |
743 | const __m128i dxidy = _mm_sub_epi16(distx_, dxdy); \ |
744 | const __m128i idxdy = _mm_sub_epi16(disty_, dxdy); \ |
745 | \ |
746 | __m128i tlAG = _mm_srli_epi16(tl, 8); \ |
747 | __m128i tlRB = _mm_and_si128(tl, colorMask); \ |
748 | __m128i trAG = _mm_srli_epi16(tr, 8); \ |
749 | __m128i trRB = _mm_and_si128(tr, colorMask); \ |
750 | __m128i blAG = _mm_srli_epi16(bl, 8); \ |
751 | __m128i blRB = _mm_and_si128(bl, colorMask); \ |
752 | __m128i brAG = _mm_srli_epi16(br, 8); \ |
753 | __m128i brRB = _mm_and_si128(br, colorMask); \ |
754 | \ |
755 | tlAG = _mm_mullo_epi16(tlAG, idxidy); \ |
756 | tlRB = _mm_mullo_epi16(tlRB, idxidy); \ |
757 | trAG = _mm_mullo_epi16(trAG, dxidy); \ |
758 | trRB = _mm_mullo_epi16(trRB, dxidy); \ |
759 | blAG = _mm_mullo_epi16(blAG, idxdy); \ |
760 | blRB = _mm_mullo_epi16(blRB, idxdy); \ |
761 | brAG = _mm_mullo_epi16(brAG, dxdy); \ |
762 | brRB = _mm_mullo_epi16(brRB, dxdy); \ |
763 | \ |
764 | /* Add the values, and shift to only keep 8 significant bits per colors */ \ |
765 | __m128i rAG =_mm_add_epi16(_mm_add_epi16(tlAG, trAG), _mm_add_epi16(blAG, brAG)); \ |
766 | __m128i rRB =_mm_add_epi16(_mm_add_epi16(tlRB, trRB), _mm_add_epi16(blRB, brRB)); \ |
767 | rAG = _mm_andnot_si128(colorMask, rAG); \ |
768 | rRB = _mm_srli_epi16(rRB, 8); \ |
769 | _mm_storeu_si128((__m128i*)(b), _mm_or_si128(rAG, rRB)); \ |
770 | } |
771 | #endif |
772 | |
773 | #if defined(__ARM_NEON__) |
774 | #define interpolate_4_pixels_16_neon(tl, tr, bl, br, distx, disty, disty_, colorMask, invColorMask, v_256, b) \ |
775 | { \ |
776 | const int16x8_t dxdy = vmulq_s16(distx, disty); \ |
777 | const int16x8_t distx_ = vshlq_n_s16(distx, 4); \ |
778 | const int16x8_t idxidy = vaddq_s16(dxdy, vsubq_s16(v_256, vaddq_s16(distx_, disty_))); \ |
779 | const int16x8_t dxidy = vsubq_s16(distx_, dxdy); \ |
780 | const int16x8_t idxdy = vsubq_s16(disty_, dxdy); \ |
781 | \ |
782 | int16x8_t tlAG = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(tl), 8)); \ |
783 | int16x8_t tlRB = vandq_s16(tl, colorMask); \ |
784 | int16x8_t trAG = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(tr), 8)); \ |
785 | int16x8_t trRB = vandq_s16(tr, colorMask); \ |
786 | int16x8_t blAG = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(bl), 8)); \ |
787 | int16x8_t blRB = vandq_s16(bl, colorMask); \ |
788 | int16x8_t brAG = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(br), 8)); \ |
789 | int16x8_t brRB = vandq_s16(br, colorMask); \ |
790 | \ |
791 | int16x8_t rAG = vmulq_s16(tlAG, idxidy); \ |
792 | int16x8_t rRB = vmulq_s16(tlRB, idxidy); \ |
793 | rAG = vmlaq_s16(rAG, trAG, dxidy); \ |
794 | rRB = vmlaq_s16(rRB, trRB, dxidy); \ |
795 | rAG = vmlaq_s16(rAG, blAG, idxdy); \ |
796 | rRB = vmlaq_s16(rRB, blRB, idxdy); \ |
797 | rAG = vmlaq_s16(rAG, brAG, dxdy); \ |
798 | rRB = vmlaq_s16(rRB, brRB, dxdy); \ |
799 | \ |
800 | rAG = vandq_s16(invColorMask, rAG); \ |
801 | rRB = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(rRB), 8)); \ |
802 | vst1q_s16((int16_t*)(b), vorrq_s16(rAG, rRB)); \ |
803 | } |
804 | #endif |
805 | |
806 | template<TextureBlendType blendType> |
807 | void fetchTransformedBilinear_pixelBounds(int max, int l1, int l2, int &v1, int &v2); |
808 | |
809 | template<> |
810 | inline void fetchTransformedBilinear_pixelBounds<BlendTransformedBilinearTiled>(int max, int, int, int &v1, int &v2) |
811 | { |
812 | v1 %= max; |
813 | if (v1 < 0) |
814 | v1 += max; |
815 | v2 = v1 + 1; |
816 | if (v2 == max) |
817 | v2 = 0; |
818 | Q_ASSERT(v1 >= 0 && v1 < max); |
819 | Q_ASSERT(v2 >= 0 && v2 < max); |
820 | } |
821 | |
822 | template<> |
823 | inline void fetchTransformedBilinear_pixelBounds<BlendTransformedBilinear>(int, int l1, int l2, int &v1, int &v2) |
824 | { |
825 | if (v1 < l1) |
826 | v2 = v1 = l1; |
827 | else if (v1 >= l2) |
828 | v2 = v1 = l2; |
829 | else |
830 | v2 = v1 + 1; |
831 | Q_ASSERT(v1 >= l1 && v1 <= l2); |
832 | Q_ASSERT(v2 >= l1 && v2 <= l2); |
833 | } |
834 | |
835 | enum FastTransformTypes { |
836 | SimpleScaleTransform, |
837 | UpscaleTransform, |
838 | DownscaleTransform, |
839 | RotateTransform, |
840 | FastRotateTransform, |
841 | NFastTransformTypes |
842 | }; |
843 | |
844 | // Completes the partial interpolation stored in IntermediateBuffer. |
845 | // by performing the x-axis interpolation and joining the RB and AG buffers. |
846 | static void QT_FASTCALL intermediate_adder(uint *b, uint *end, const IntermediateBuffer &intermediate, int offset, int &fx, int fdx) |
847 | { |
848 | #if defined(QT_COMPILER_SUPPORTS_AVX2) |
849 | extern void QT_FASTCALL intermediate_adder_avx2(uint *b, uint *end, const IntermediateBuffer &intermediate, int offset, int &fx, int fdx); |
850 | if (qCpuHasFeature(ArchHaswell)) |
851 | return intermediate_adder_avx2(b, end, intermediate, offset, fx, fdx); |
852 | #endif |
853 | |
854 | // Switch to intermediate buffer coordinates |
855 | fx -= offset * fixed_scale; |
856 | |
857 | while (b < end) { |
858 | const int x = (fx >> 16); |
859 | |
860 | const uint distx = (fx & 0x0000ffff) >> 8; |
861 | const uint idistx = 256 - distx; |
862 | const uint rb = (intermediate.buffer_rb[x] * idistx + intermediate.buffer_rb[x + 1] * distx) & 0xff00ff00; |
863 | const uint ag = (intermediate.buffer_ag[x] * idistx + intermediate.buffer_ag[x + 1] * distx) & 0xff00ff00; |
864 | *b = (rb >> 8) | ag; |
865 | b++; |
866 | fx += fdx; |
867 | } |
868 | fx += offset * fixed_scale; |
869 | } |
870 | |
871 | typedef void (QT_FASTCALL *BilinearFastTransformHelper)(uint *b, uint *end, const QTextureData &image, int &fx, int &fy, int fdx, int fdy); |
872 | |
873 | template<TextureBlendType blendType> |
874 | static void QT_FASTCALL fetchTransformedBilinearARGB32PM_simple_scale_helper(uint *b, uint *end, const QTextureData &image, |
875 | int &fx, int &fy, int fdx, int /*fdy*/) |
876 | { |
877 | int y1 = (fy >> 16); |
878 | int y2; |
879 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
880 | const uint *s1 = (const uint *)image.scanLine(y1); |
881 | const uint *s2 = (const uint *)image.scanLine(y2); |
882 | |
883 | const int disty = (fy & 0x0000ffff) >> 8; |
884 | const int idisty = 256 - disty; |
885 | const int length = end - b; |
886 | |
887 | // The intermediate buffer is generated in the positive direction |
888 | const int adjust = (fdx < 0) ? fdx * length : 0; |
889 | const int offset = (fx + adjust) >> 16; |
890 | int x = offset; |
891 | |
892 | IntermediateBuffer intermediate; |
893 | // count is the size used in the intermediate.buffer. |
894 | int count = (qint64(length) * qAbs(fdx) + fixed_scale - 1) / fixed_scale + 2; |
895 | // length is supposed to be <= BufferSize either because data->m11 < 1 or |
896 | // data->m11 < 2, and any larger buffers split |
897 | Q_ASSERT(count <= BufferSize + 2); |
898 | int f = 0; |
899 | int lim = count; |
900 | if (blendType == BlendTransformedBilinearTiled) { |
901 | x %= image.width; |
902 | if (x < 0) x += image.width; |
903 | } else { |
904 | lim = qMin(count, image.x2 - x); |
905 | if (x < image.x1) { |
906 | Q_ASSERT(x < image.x2); |
907 | uint t = s1[image.x1]; |
908 | uint b = s2[image.x1]; |
909 | quint32 rb = (((t & 0xff00ff) * idisty + (b & 0xff00ff) * disty) >> 8) & 0xff00ff; |
910 | quint32 ag = ((((t>>8) & 0xff00ff) * idisty + ((b>>8) & 0xff00ff) * disty) >> 8) & 0xff00ff; |
911 | do { |
912 | intermediate.buffer_rb[f] = rb; |
913 | intermediate.buffer_ag[f] = ag; |
914 | f++; |
915 | x++; |
916 | } while (x < image.x1 && f < lim); |
917 | } |
918 | } |
919 | |
920 | if (blendType != BlendTransformedBilinearTiled) { |
921 | #if defined(__SSE2__) |
922 | const __m128i disty_ = _mm_set1_epi16(disty); |
923 | const __m128i idisty_ = _mm_set1_epi16(idisty); |
924 | const __m128i colorMask = _mm_set1_epi32(0x00ff00ff); |
925 | |
926 | lim -= 3; |
927 | for (; f < lim; x += 4, f += 4) { |
928 | // Load 4 pixels from s1, and split the alpha-green and red-blue component |
929 | __m128i top = _mm_loadu_si128((const __m128i*)((const uint *)(s1)+x)); |
930 | __m128i topAG = _mm_srli_epi16(top, 8); |
931 | __m128i topRB = _mm_and_si128(top, colorMask); |
932 | // Multiplies each color component by idisty |
933 | topAG = _mm_mullo_epi16 (topAG, idisty_); |
934 | topRB = _mm_mullo_epi16 (topRB, idisty_); |
935 | |
936 | // Same for the s2 vector |
937 | __m128i bottom = _mm_loadu_si128((const __m128i*)((const uint *)(s2)+x)); |
938 | __m128i bottomAG = _mm_srli_epi16(bottom, 8); |
939 | __m128i bottomRB = _mm_and_si128(bottom, colorMask); |
940 | bottomAG = _mm_mullo_epi16 (bottomAG, disty_); |
941 | bottomRB = _mm_mullo_epi16 (bottomRB, disty_); |
942 | |
943 | // Add the values, and shift to only keep 8 significant bits per colors |
944 | __m128i rAG =_mm_add_epi16(topAG, bottomAG); |
945 | rAG = _mm_srli_epi16(rAG, 8); |
946 | _mm_storeu_si128((__m128i*)(&intermediate.buffer_ag[f]), rAG); |
947 | __m128i rRB =_mm_add_epi16(topRB, bottomRB); |
948 | rRB = _mm_srli_epi16(rRB, 8); |
949 | _mm_storeu_si128((__m128i*)(&intermediate.buffer_rb[f]), rRB); |
950 | } |
951 | #elif defined(__ARM_NEON__) |
952 | const int16x8_t disty_ = vdupq_n_s16(disty); |
953 | const int16x8_t idisty_ = vdupq_n_s16(idisty); |
954 | const int16x8_t colorMask = vdupq_n_s16(0x00ff); |
955 | |
956 | lim -= 3; |
957 | for (; f < lim; x += 4, f += 4) { |
958 | // Load 4 pixels from s1, and split the alpha-green and red-blue component |
959 | int16x8_t top = vld1q_s16((int16_t*)((const uint *)(s1)+x)); |
960 | int16x8_t topAG = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(top), 8)); |
961 | int16x8_t topRB = vandq_s16(top, colorMask); |
962 | // Multiplies each color component by idisty |
963 | topAG = vmulq_s16(topAG, idisty_); |
964 | topRB = vmulq_s16(topRB, idisty_); |
965 | |
966 | // Same for the s2 vector |
967 | int16x8_t bottom = vld1q_s16((int16_t*)((const uint *)(s2)+x)); |
968 | int16x8_t bottomAG = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(bottom), 8)); |
969 | int16x8_t bottomRB = vandq_s16(bottom, colorMask); |
970 | bottomAG = vmulq_s16(bottomAG, disty_); |
971 | bottomRB = vmulq_s16(bottomRB, disty_); |
972 | |
973 | // Add the values, and shift to only keep 8 significant bits per colors |
974 | int16x8_t rAG = vaddq_s16(topAG, bottomAG); |
975 | rAG = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(rAG), 8)); |
976 | vst1q_s16((int16_t*)(&intermediate.buffer_ag[f]), rAG); |
977 | int16x8_t rRB = vaddq_s16(topRB, bottomRB); |
978 | rRB = vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(rRB), 8)); |
979 | vst1q_s16((int16_t*)(&intermediate.buffer_rb[f]), rRB); |
980 | } |
981 | #endif |
982 | } |
983 | for (; f < count; f++) { // Same as above but without simd |
984 | if (blendType == BlendTransformedBilinearTiled) { |
985 | if (x >= image.width) x -= image.width; |
986 | } else { |
987 | x = qMin(x, image.x2 - 1); |
988 | } |
989 | |
990 | uint t = s1[x]; |
991 | uint b = s2[x]; |
992 | |
993 | intermediate.buffer_rb[f] = (((t & 0xff00ff) * idisty + (b & 0xff00ff) * disty) >> 8) & 0xff00ff; |
994 | intermediate.buffer_ag[f] = ((((t>>8) & 0xff00ff) * idisty + ((b>>8) & 0xff00ff) * disty) >> 8) & 0xff00ff; |
995 | x++; |
996 | } |
997 | |
998 | // Now interpolate the values from the intermediate.buffer to get the final result. |
999 | intermediate_adder(b, end, intermediate, offset, fx, fdx); |
1000 | } |
1001 | |
1002 | template<TextureBlendType blendType> |
1003 | static void QT_FASTCALL fetchTransformedBilinearARGB32PM_upscale_helper(uint *b, uint *end, const QTextureData &image, |
1004 | int &fx, int &fy, int fdx, int /*fdy*/) |
1005 | { |
1006 | int y1 = (fy >> 16); |
1007 | int y2; |
1008 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1009 | const uint *s1 = (const uint *)image.scanLine(y1); |
1010 | const uint *s2 = (const uint *)image.scanLine(y2); |
1011 | const int disty = (fy & 0x0000ffff) >> 8; |
1012 | |
1013 | if (blendType != BlendTransformedBilinearTiled) { |
1014 | const qint64 min_fx = qint64(image.x1) * fixed_scale; |
1015 | const qint64 max_fx = qint64(image.x2 - 1) * fixed_scale; |
1016 | while (b < end) { |
1017 | int x1 = (fx >> 16); |
1018 | int x2; |
1019 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1020 | if (x1 != x2) |
1021 | break; |
1022 | uint top = s1[x1]; |
1023 | uint bot = s2[x1]; |
1024 | *b = INTERPOLATE_PIXEL_256(top, 256 - disty, bot, disty); |
1025 | fx += fdx; |
1026 | ++b; |
1027 | } |
1028 | uint *boundedEnd = end; |
1029 | if (fdx > 0) |
1030 | boundedEnd = qMin(boundedEnd, b + (max_fx - fx) / fdx); |
1031 | else if (fdx < 0) |
1032 | boundedEnd = qMin(boundedEnd, b + (min_fx - fx) / fdx); |
1033 | |
1034 | // A fast middle part without boundary checks |
1035 | while (b < boundedEnd) { |
1036 | int x = (fx >> 16); |
1037 | int distx = (fx & 0x0000ffff) >> 8; |
1038 | *b = interpolate_4_pixels(s1 + x, s2 + x, distx, disty); |
1039 | fx += fdx; |
1040 | ++b; |
1041 | } |
1042 | } |
1043 | |
1044 | while (b < end) { |
1045 | int x1 = (fx >> 16); |
1046 | int x2; |
1047 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1 , x1, x2); |
1048 | uint tl = s1[x1]; |
1049 | uint tr = s1[x2]; |
1050 | uint bl = s2[x1]; |
1051 | uint br = s2[x2]; |
1052 | int distx = (fx & 0x0000ffff) >> 8; |
1053 | *b = interpolate_4_pixels(tl, tr, bl, br, distx, disty); |
1054 | |
1055 | fx += fdx; |
1056 | ++b; |
1057 | } |
1058 | } |
1059 | |
1060 | template<TextureBlendType blendType> |
1061 | static void QT_FASTCALL fetchTransformedBilinearARGB32PM_downscale_helper(uint *b, uint *end, const QTextureData &image, |
1062 | int &fx, int &fy, int fdx, int /*fdy*/) |
1063 | { |
1064 | int y1 = (fy >> 16); |
1065 | int y2; |
1066 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1067 | const uint *s1 = (const uint *)image.scanLine(y1); |
1068 | const uint *s2 = (const uint *)image.scanLine(y2); |
1069 | const int disty8 = (fy & 0x0000ffff) >> 8; |
1070 | const int disty4 = (disty8 + 0x08) >> 4; |
1071 | |
1072 | if (blendType != BlendTransformedBilinearTiled) { |
1073 | const qint64 min_fx = qint64(image.x1) * fixed_scale; |
1074 | const qint64 max_fx = qint64(image.x2 - 1) * fixed_scale; |
1075 | while (b < end) { |
1076 | int x1 = (fx >> 16); |
1077 | int x2; |
1078 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1079 | if (x1 != x2) |
1080 | break; |
1081 | uint top = s1[x1]; |
1082 | uint bot = s2[x1]; |
1083 | *b = INTERPOLATE_PIXEL_256(top, 256 - disty8, bot, disty8); |
1084 | fx += fdx; |
1085 | ++b; |
1086 | } |
1087 | uint *boundedEnd = end; |
1088 | if (fdx > 0) |
1089 | boundedEnd = qMin(boundedEnd, b + (max_fx - fx) / fdx); |
1090 | else if (fdx < 0) |
1091 | boundedEnd = qMin(boundedEnd, b + (min_fx - fx) / fdx); |
1092 | // A fast middle part without boundary checks |
1093 | #if defined(__SSE2__) |
1094 | const __m128i colorMask = _mm_set1_epi32(0x00ff00ff); |
1095 | const __m128i v_256 = _mm_set1_epi16(256); |
1096 | const __m128i v_disty = _mm_set1_epi16(disty4); |
1097 | const __m128i v_fdx = _mm_set1_epi32(fdx*4); |
1098 | const __m128i v_fx_r = _mm_set1_epi32(0x8); |
1099 | __m128i v_fx = _mm_setr_epi32(fx, fx + fdx, fx + fdx + fdx, fx + fdx + fdx + fdx); |
1100 | |
1101 | while (b < boundedEnd - 3) { |
1102 | __m128i offset = _mm_srli_epi32(v_fx, 16); |
1103 | const int offset0 = _mm_cvtsi128_si32(offset); offset = _mm_srli_si128(offset, 4); |
1104 | const int offset1 = _mm_cvtsi128_si32(offset); offset = _mm_srli_si128(offset, 4); |
1105 | const int offset2 = _mm_cvtsi128_si32(offset); offset = _mm_srli_si128(offset, 4); |
1106 | const int offset3 = _mm_cvtsi128_si32(offset); |
1107 | const __m128i tl = _mm_setr_epi32(s1[offset0], s1[offset1], s1[offset2], s1[offset3]); |
1108 | const __m128i tr = _mm_setr_epi32(s1[offset0 + 1], s1[offset1 + 1], s1[offset2 + 1], s1[offset3 + 1]); |
1109 | const __m128i bl = _mm_setr_epi32(s2[offset0], s2[offset1], s2[offset2], s2[offset3]); |
1110 | const __m128i br = _mm_setr_epi32(s2[offset0 + 1], s2[offset1 + 1], s2[offset2 + 1], s2[offset3 + 1]); |
1111 | |
1112 | __m128i v_distx = _mm_srli_epi16(v_fx, 8); |
1113 | v_distx = _mm_srli_epi16(_mm_add_epi32(v_distx, v_fx_r), 4); |
1114 | v_distx = _mm_shufflehi_epi16(v_distx, _MM_SHUFFLE(2,2,0,0)); |
1115 | v_distx = _mm_shufflelo_epi16(v_distx, _MM_SHUFFLE(2,2,0,0)); |
1116 | |
1117 | interpolate_4_pixels_16_sse2(tl, tr, bl, br, v_distx, v_disty, colorMask, v_256, b); |
1118 | b += 4; |
1119 | v_fx = _mm_add_epi32(v_fx, v_fdx); |
1120 | } |
1121 | fx = _mm_cvtsi128_si32(v_fx); |
1122 | #elif defined(__ARM_NEON__) |
1123 | const int16x8_t colorMask = vdupq_n_s16(0x00ff); |
1124 | const int16x8_t invColorMask = vmvnq_s16(colorMask); |
1125 | const int16x8_t v_256 = vdupq_n_s16(256); |
1126 | const int16x8_t v_disty = vdupq_n_s16(disty4); |
1127 | const int16x8_t v_disty_ = vshlq_n_s16(v_disty, 4); |
1128 | int32x4_t v_fdx = vdupq_n_s32(fdx*4); |
1129 | |
1130 | int32x4_t v_fx = vmovq_n_s32(fx); |
1131 | v_fx = vsetq_lane_s32(fx + fdx, v_fx, 1); |
1132 | v_fx = vsetq_lane_s32(fx + fdx * 2, v_fx, 2); |
1133 | v_fx = vsetq_lane_s32(fx + fdx * 3, v_fx, 3); |
1134 | |
1135 | const int32x4_t v_ffff_mask = vdupq_n_s32(0x0000ffff); |
1136 | const int32x4_t v_fx_r = vdupq_n_s32(0x0800); |
1137 | |
1138 | while (b < boundedEnd - 3) { |
1139 | uint32x4x2_t v_top, v_bot; |
1140 | |
1141 | int x1 = (fx >> 16); |
1142 | fx += fdx; |
1143 | v_top = vld2q_lane_u32(s1 + x1, v_top, 0); |
1144 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 0); |
1145 | x1 = (fx >> 16); |
1146 | fx += fdx; |
1147 | v_top = vld2q_lane_u32(s1 + x1, v_top, 1); |
1148 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 1); |
1149 | x1 = (fx >> 16); |
1150 | fx += fdx; |
1151 | v_top = vld2q_lane_u32(s1 + x1, v_top, 2); |
1152 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 2); |
1153 | x1 = (fx >> 16); |
1154 | fx += fdx; |
1155 | v_top = vld2q_lane_u32(s1 + x1, v_top, 3); |
1156 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 3); |
1157 | |
1158 | int32x4_t v_distx = vshrq_n_s32(vaddq_s32(vandq_s32(v_fx, v_ffff_mask), v_fx_r), 12); |
1159 | v_distx = vorrq_s32(v_distx, vshlq_n_s32(v_distx, 16)); |
1160 | |
1161 | interpolate_4_pixels_16_neon( |
1162 | vreinterpretq_s16_u32(v_top.val[0]), vreinterpretq_s16_u32(v_top.val[1]), |
1163 | vreinterpretq_s16_u32(v_bot.val[0]), vreinterpretq_s16_u32(v_bot.val[1]), |
1164 | vreinterpretq_s16_s32(v_distx), v_disty, v_disty_, |
1165 | colorMask, invColorMask, v_256, b); |
1166 | b+=4; |
1167 | v_fx = vaddq_s32(v_fx, v_fdx); |
1168 | } |
1169 | #endif |
1170 | while (b < boundedEnd) { |
1171 | int x = (fx >> 16); |
1172 | if (hasFastInterpolate4()) { |
1173 | int distx8 = (fx & 0x0000ffff) >> 8; |
1174 | *b = interpolate_4_pixels(s1 + x, s2 + x, distx8, disty8); |
1175 | } else { |
1176 | int distx4 = ((fx & 0x0000ffff) + 0x0800) >> 12; |
1177 | *b = interpolate_4_pixels_16(s1[x], s1[x + 1], s2[x], s2[x + 1], distx4, disty4); |
1178 | } |
1179 | fx += fdx; |
1180 | ++b; |
1181 | } |
1182 | } |
1183 | |
1184 | while (b < end) { |
1185 | int x1 = (fx >> 16); |
1186 | int x2; |
1187 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1188 | uint tl = s1[x1]; |
1189 | uint tr = s1[x2]; |
1190 | uint bl = s2[x1]; |
1191 | uint br = s2[x2]; |
1192 | if (hasFastInterpolate4()) { |
1193 | int distx8 = (fx & 0x0000ffff) >> 8; |
1194 | *b = interpolate_4_pixels(tl, tr, bl, br, distx8, disty8); |
1195 | } else { |
1196 | int distx4 = ((fx & 0x0000ffff) + 0x0800) >> 12; |
1197 | *b = interpolate_4_pixels_16(tl, tr, bl, br, distx4, disty4); |
1198 | } |
1199 | fx += fdx; |
1200 | ++b; |
1201 | } |
1202 | } |
1203 | |
1204 | template<TextureBlendType blendType> |
1205 | static void QT_FASTCALL fetchTransformedBilinearARGB32PM_rotate_helper(uint *b, uint *end, const QTextureData &image, |
1206 | int &fx, int &fy, int fdx, int fdy) |
1207 | { |
1208 | // if we are zooming more than 8 times, we use 8bit precision for the position. |
1209 | while (b < end) { |
1210 | int x1 = (fx >> 16); |
1211 | int x2; |
1212 | int y1 = (fy >> 16); |
1213 | int y2; |
1214 | |
1215 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1216 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1217 | |
1218 | const uint *s1 = (const uint *)image.scanLine(y1); |
1219 | const uint *s2 = (const uint *)image.scanLine(y2); |
1220 | |
1221 | uint tl = s1[x1]; |
1222 | uint tr = s1[x2]; |
1223 | uint bl = s2[x1]; |
1224 | uint br = s2[x2]; |
1225 | |
1226 | int distx = (fx & 0x0000ffff) >> 8; |
1227 | int disty = (fy & 0x0000ffff) >> 8; |
1228 | |
1229 | *b = interpolate_4_pixels(tl, tr, bl, br, distx, disty); |
1230 | |
1231 | fx += fdx; |
1232 | fy += fdy; |
1233 | ++b; |
1234 | } |
1235 | } |
1236 | |
1237 | template<TextureBlendType blendType> |
1238 | static void QT_FASTCALL fetchTransformedBilinearARGB32PM_fast_rotate_helper(uint *b, uint *end, const QTextureData &image, |
1239 | int &fx, int &fy, int fdx, int fdy) |
1240 | { |
1241 | //we are zooming less than 8x, use 4bit precision |
1242 | if (blendType != BlendTransformedBilinearTiled) { |
1243 | const qint64 min_fx = qint64(image.x1) * fixed_scale; |
1244 | const qint64 max_fx = qint64(image.x2 - 1) * fixed_scale; |
1245 | const qint64 min_fy = qint64(image.y1) * fixed_scale; |
1246 | const qint64 max_fy = qint64(image.y2 - 1) * fixed_scale; |
1247 | // first handle the possibly bounded part in the beginning |
1248 | while (b < end) { |
1249 | int x1 = (fx >> 16); |
1250 | int x2; |
1251 | int y1 = (fy >> 16); |
1252 | int y2; |
1253 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1254 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1255 | if (x1 != x2 && y1 != y2) |
1256 | break; |
1257 | const uint *s1 = (const uint *)image.scanLine(y1); |
1258 | const uint *s2 = (const uint *)image.scanLine(y2); |
1259 | uint tl = s1[x1]; |
1260 | uint tr = s1[x2]; |
1261 | uint bl = s2[x1]; |
1262 | uint br = s2[x2]; |
1263 | if (hasFastInterpolate4()) { |
1264 | int distx = (fx & 0x0000ffff) >> 8; |
1265 | int disty = (fy & 0x0000ffff) >> 8; |
1266 | *b = interpolate_4_pixels(tl, tr, bl, br, distx, disty); |
1267 | } else { |
1268 | int distx = ((fx & 0x0000ffff) + 0x0800) >> 12; |
1269 | int disty = ((fy & 0x0000ffff) + 0x0800) >> 12; |
1270 | *b = interpolate_4_pixels_16(tl, tr, bl, br, distx, disty); |
1271 | } |
1272 | fx += fdx; |
1273 | fy += fdy; |
1274 | ++b; |
1275 | } |
1276 | uint *boundedEnd = end; |
1277 | if (fdx > 0) |
1278 | boundedEnd = qMin(boundedEnd, b + (max_fx - fx) / fdx); |
1279 | else if (fdx < 0) |
1280 | boundedEnd = qMin(boundedEnd, b + (min_fx - fx) / fdx); |
1281 | if (fdy > 0) |
1282 | boundedEnd = qMin(boundedEnd, b + (max_fy - fy) / fdy); |
1283 | else if (fdy < 0) |
1284 | boundedEnd = qMin(boundedEnd, b + (min_fy - fy) / fdy); |
1285 | |
1286 | // until boundedEnd we can now have a fast middle part without boundary checks |
1287 | #if defined(__SSE2__) |
1288 | const __m128i colorMask = _mm_set1_epi32(0x00ff00ff); |
1289 | const __m128i v_256 = _mm_set1_epi16(256); |
1290 | const __m128i v_fdx = _mm_set1_epi32(fdx*4); |
1291 | const __m128i v_fdy = _mm_set1_epi32(fdy*4); |
1292 | const __m128i v_fxy_r = _mm_set1_epi32(0x8); |
1293 | __m128i v_fx = _mm_setr_epi32(fx, fx + fdx, fx + fdx + fdx, fx + fdx + fdx + fdx); |
1294 | __m128i v_fy = _mm_setr_epi32(fy, fy + fdy, fy + fdy + fdy, fy + fdy + fdy + fdy); |
1295 | |
1296 | const uchar *textureData = image.imageData; |
1297 | const qsizetype bytesPerLine = image.bytesPerLine; |
1298 | const __m128i vbpl = _mm_shufflelo_epi16(_mm_cvtsi32_si128(bytesPerLine/4), _MM_SHUFFLE(0, 0, 0, 0)); |
1299 | |
1300 | while (b < boundedEnd - 3) { |
1301 | const __m128i vy = _mm_packs_epi32(_mm_srli_epi32(v_fy, 16), _mm_setzero_si128()); |
1302 | // 4x16bit * 4x16bit -> 4x32bit |
1303 | __m128i offset = _mm_unpacklo_epi16(_mm_mullo_epi16(vy, vbpl), _mm_mulhi_epi16(vy, vbpl)); |
1304 | offset = _mm_add_epi32(offset, _mm_srli_epi32(v_fx, 16)); |
1305 | const int offset0 = _mm_cvtsi128_si32(offset); offset = _mm_srli_si128(offset, 4); |
1306 | const int offset1 = _mm_cvtsi128_si32(offset); offset = _mm_srli_si128(offset, 4); |
1307 | const int offset2 = _mm_cvtsi128_si32(offset); offset = _mm_srli_si128(offset, 4); |
1308 | const int offset3 = _mm_cvtsi128_si32(offset); |
1309 | const uint *topData = (const uint *)(textureData); |
1310 | const __m128i tl = _mm_setr_epi32(topData[offset0], topData[offset1], topData[offset2], topData[offset3]); |
1311 | const __m128i tr = _mm_setr_epi32(topData[offset0 + 1], topData[offset1 + 1], topData[offset2 + 1], topData[offset3 + 1]); |
1312 | const uint *bottomData = (const uint *)(textureData + bytesPerLine); |
1313 | const __m128i bl = _mm_setr_epi32(bottomData[offset0], bottomData[offset1], bottomData[offset2], bottomData[offset3]); |
1314 | const __m128i br = _mm_setr_epi32(bottomData[offset0 + 1], bottomData[offset1 + 1], bottomData[offset2 + 1], bottomData[offset3 + 1]); |
1315 | |
1316 | __m128i v_distx = _mm_srli_epi16(v_fx, 8); |
1317 | __m128i v_disty = _mm_srli_epi16(v_fy, 8); |
1318 | v_distx = _mm_srli_epi16(_mm_add_epi32(v_distx, v_fxy_r), 4); |
1319 | v_disty = _mm_srli_epi16(_mm_add_epi32(v_disty, v_fxy_r), 4); |
1320 | v_distx = _mm_shufflehi_epi16(v_distx, _MM_SHUFFLE(2,2,0,0)); |
1321 | v_distx = _mm_shufflelo_epi16(v_distx, _MM_SHUFFLE(2,2,0,0)); |
1322 | v_disty = _mm_shufflehi_epi16(v_disty, _MM_SHUFFLE(2,2,0,0)); |
1323 | v_disty = _mm_shufflelo_epi16(v_disty, _MM_SHUFFLE(2,2,0,0)); |
1324 | |
1325 | interpolate_4_pixels_16_sse2(tl, tr, bl, br, v_distx, v_disty, colorMask, v_256, b); |
1326 | b += 4; |
1327 | v_fx = _mm_add_epi32(v_fx, v_fdx); |
1328 | v_fy = _mm_add_epi32(v_fy, v_fdy); |
1329 | } |
1330 | fx = _mm_cvtsi128_si32(v_fx); |
1331 | fy = _mm_cvtsi128_si32(v_fy); |
1332 | #elif defined(__ARM_NEON__) |
1333 | const int16x8_t colorMask = vdupq_n_s16(0x00ff); |
1334 | const int16x8_t invColorMask = vmvnq_s16(colorMask); |
1335 | const int16x8_t v_256 = vdupq_n_s16(256); |
1336 | int32x4_t v_fdx = vdupq_n_s32(fdx * 4); |
1337 | int32x4_t v_fdy = vdupq_n_s32(fdy * 4); |
1338 | |
1339 | const uchar *textureData = image.imageData; |
1340 | const qsizetype bytesPerLine = image.bytesPerLine; |
1341 | |
1342 | int32x4_t v_fx = vmovq_n_s32(fx); |
1343 | int32x4_t v_fy = vmovq_n_s32(fy); |
1344 | v_fx = vsetq_lane_s32(fx + fdx, v_fx, 1); |
1345 | v_fy = vsetq_lane_s32(fy + fdy, v_fy, 1); |
1346 | v_fx = vsetq_lane_s32(fx + fdx * 2, v_fx, 2); |
1347 | v_fy = vsetq_lane_s32(fy + fdy * 2, v_fy, 2); |
1348 | v_fx = vsetq_lane_s32(fx + fdx * 3, v_fx, 3); |
1349 | v_fy = vsetq_lane_s32(fy + fdy * 3, v_fy, 3); |
1350 | |
1351 | const int32x4_t v_ffff_mask = vdupq_n_s32(0x0000ffff); |
1352 | const int32x4_t v_round = vdupq_n_s32(0x0800); |
1353 | |
1354 | while (b < boundedEnd - 3) { |
1355 | uint32x4x2_t v_top, v_bot; |
1356 | |
1357 | int x1 = (fx >> 16); |
1358 | int y1 = (fy >> 16); |
1359 | fx += fdx; fy += fdy; |
1360 | const uchar *sl = textureData + bytesPerLine * y1; |
1361 | const uint *s1 = reinterpret_cast<const uint *>(sl); |
1362 | const uint *s2 = reinterpret_cast<const uint *>(sl + bytesPerLine); |
1363 | v_top = vld2q_lane_u32(s1 + x1, v_top, 0); |
1364 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 0); |
1365 | x1 = (fx >> 16); |
1366 | y1 = (fy >> 16); |
1367 | fx += fdx; fy += fdy; |
1368 | sl = textureData + bytesPerLine * y1; |
1369 | s1 = reinterpret_cast<const uint *>(sl); |
1370 | s2 = reinterpret_cast<const uint *>(sl + bytesPerLine); |
1371 | v_top = vld2q_lane_u32(s1 + x1, v_top, 1); |
1372 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 1); |
1373 | x1 = (fx >> 16); |
1374 | y1 = (fy >> 16); |
1375 | fx += fdx; fy += fdy; |
1376 | sl = textureData + bytesPerLine * y1; |
1377 | s1 = reinterpret_cast<const uint *>(sl); |
1378 | s2 = reinterpret_cast<const uint *>(sl + bytesPerLine); |
1379 | v_top = vld2q_lane_u32(s1 + x1, v_top, 2); |
1380 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 2); |
1381 | x1 = (fx >> 16); |
1382 | y1 = (fy >> 16); |
1383 | fx += fdx; fy += fdy; |
1384 | sl = textureData + bytesPerLine * y1; |
1385 | s1 = reinterpret_cast<const uint *>(sl); |
1386 | s2 = reinterpret_cast<const uint *>(sl + bytesPerLine); |
1387 | v_top = vld2q_lane_u32(s1 + x1, v_top, 3); |
1388 | v_bot = vld2q_lane_u32(s2 + x1, v_bot, 3); |
1389 | |
1390 | int32x4_t v_distx = vshrq_n_s32(vaddq_s32(vandq_s32(v_fx, v_ffff_mask), v_round), 12); |
1391 | int32x4_t v_disty = vshrq_n_s32(vaddq_s32(vandq_s32(v_fy, v_ffff_mask), v_round), 12); |
1392 | v_distx = vorrq_s32(v_distx, vshlq_n_s32(v_distx, 16)); |
1393 | v_disty = vorrq_s32(v_disty, vshlq_n_s32(v_disty, 16)); |
1394 | int16x8_t v_disty_ = vshlq_n_s16(vreinterpretq_s16_s32(v_disty), 4); |
1395 | |
1396 | interpolate_4_pixels_16_neon( |
1397 | vreinterpretq_s16_u32(v_top.val[0]), vreinterpretq_s16_u32(v_top.val[1]), |
1398 | vreinterpretq_s16_u32(v_bot.val[0]), vreinterpretq_s16_u32(v_bot.val[1]), |
1399 | vreinterpretq_s16_s32(v_distx), vreinterpretq_s16_s32(v_disty), |
1400 | v_disty_, colorMask, invColorMask, v_256, b); |
1401 | b += 4; |
1402 | v_fx = vaddq_s32(v_fx, v_fdx); |
1403 | v_fy = vaddq_s32(v_fy, v_fdy); |
1404 | } |
1405 | #endif |
1406 | while (b < boundedEnd) { |
1407 | int x = (fx >> 16); |
1408 | int y = (fy >> 16); |
1409 | |
1410 | const uint *s1 = (const uint *)image.scanLine(y); |
1411 | const uint *s2 = (const uint *)image.scanLine(y + 1); |
1412 | |
1413 | if (hasFastInterpolate4()) { |
1414 | int distx = (fx & 0x0000ffff) >> 8; |
1415 | int disty = (fy & 0x0000ffff) >> 8; |
1416 | *b = interpolate_4_pixels(s1 + x, s2 + x, distx, disty); |
1417 | } else { |
1418 | int distx = ((fx & 0x0000ffff) + 0x0800) >> 12; |
1419 | int disty = ((fy & 0x0000ffff) + 0x0800) >> 12; |
1420 | *b = interpolate_4_pixels_16(s1[x], s1[x + 1], s2[x], s2[x + 1], distx, disty); |
1421 | } |
1422 | |
1423 | fx += fdx; |
1424 | fy += fdy; |
1425 | ++b; |
1426 | } |
1427 | } |
1428 | |
1429 | while (b < end) { |
1430 | int x1 = (fx >> 16); |
1431 | int x2; |
1432 | int y1 = (fy >> 16); |
1433 | int y2; |
1434 | |
1435 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1436 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1437 | |
1438 | const uint *s1 = (const uint *)image.scanLine(y1); |
1439 | const uint *s2 = (const uint *)image.scanLine(y2); |
1440 | |
1441 | uint tl = s1[x1]; |
1442 | uint tr = s1[x2]; |
1443 | uint bl = s2[x1]; |
1444 | uint br = s2[x2]; |
1445 | |
1446 | if (hasFastInterpolate4()) { |
1447 | int distx = (fx & 0x0000ffff) >> 8; |
1448 | int disty = (fy & 0x0000ffff) >> 8; |
1449 | *b = interpolate_4_pixels(tl, tr, bl, br, distx, disty); |
1450 | } else { |
1451 | int distx = ((fx & 0x0000ffff) + 0x0800) >> 12; |
1452 | int disty = ((fy & 0x0000ffff) + 0x0800) >> 12; |
1453 | *b = interpolate_4_pixels_16(tl, tr, bl, br, distx, disty); |
1454 | } |
1455 | |
1456 | fx += fdx; |
1457 | fy += fdy; |
1458 | ++b; |
1459 | } |
1460 | } |
1461 | |
1462 | |
1463 | static BilinearFastTransformHelper bilinearFastTransformHelperARGB32PM[2][NFastTransformTypes] = { |
1464 | { |
1465 | fetchTransformedBilinearARGB32PM_simple_scale_helper<BlendTransformedBilinear>, |
1466 | fetchTransformedBilinearARGB32PM_upscale_helper<BlendTransformedBilinear>, |
1467 | fetchTransformedBilinearARGB32PM_downscale_helper<BlendTransformedBilinear>, |
1468 | fetchTransformedBilinearARGB32PM_rotate_helper<BlendTransformedBilinear>, |
1469 | fetchTransformedBilinearARGB32PM_fast_rotate_helper<BlendTransformedBilinear> |
1470 | }, |
1471 | { |
1472 | fetchTransformedBilinearARGB32PM_simple_scale_helper<BlendTransformedBilinearTiled>, |
1473 | fetchTransformedBilinearARGB32PM_upscale_helper<BlendTransformedBilinearTiled>, |
1474 | fetchTransformedBilinearARGB32PM_downscale_helper<BlendTransformedBilinearTiled>, |
1475 | fetchTransformedBilinearARGB32PM_rotate_helper<BlendTransformedBilinearTiled>, |
1476 | fetchTransformedBilinearARGB32PM_fast_rotate_helper<BlendTransformedBilinearTiled> |
1477 | } |
1478 | }; |
1479 | |
1480 | template<TextureBlendType blendType> /* blendType = BlendTransformedBilinear or BlendTransformedBilinearTiled */ |
1481 | static const uint * QT_FASTCALL fetchTransformedBilinearARGB32PM(uint *buffer, const Operator *, |
1482 | const QSpanData *data, int y, int x, |
1483 | int length) |
1484 | { |
1485 | const qreal cx = x + qreal(0.5); |
1486 | const qreal cy = y + qreal(0.5); |
1487 | constexpr int tiled = (blendType == BlendTransformedBilinearTiled) ? 1 : 0; |
1488 | |
1489 | uint *end = buffer + length; |
1490 | uint *b = buffer; |
1491 | if (canUseFastMatrixPath(cx, cy, length, data)) { |
1492 | // The increment pr x in the scanline |
1493 | int fdx = (int)(data->m11 * fixed_scale); |
1494 | int fdy = (int)(data->m12 * fixed_scale); |
1495 | |
1496 | int fx = int((data->m21 * cy |
1497 | + data->m11 * cx + data->dx) * fixed_scale); |
1498 | int fy = int((data->m22 * cy |
1499 | + data->m12 * cx + data->dy) * fixed_scale); |
1500 | |
1501 | fx -= half_point; |
1502 | fy -= half_point; |
1503 | |
1504 | if (fdy == 0) { // simple scale, no rotation or shear |
1505 | if (qAbs(fdx) <= fixed_scale) { |
1506 | // simple scale up on X |
1507 | bilinearFastTransformHelperARGB32PM[tiled][SimpleScaleTransform](b, end, data->texture, fx, fy, fdx, fdy); |
1508 | } else if (qAbs(fdx) <= 2 * fixed_scale) { |
1509 | // simple scale down on X, less than 2x |
1510 | const int mid = (length * 2 < BufferSize) ? length : ((length + 1) / 2); |
1511 | bilinearFastTransformHelperARGB32PM[tiled][SimpleScaleTransform](buffer, buffer + mid, data->texture, fx, fy, fdx, fdy); |
1512 | if (mid != length) |
1513 | bilinearFastTransformHelperARGB32PM[tiled][SimpleScaleTransform](buffer + mid, buffer + length, data->texture, fx, fy, fdx, fdy); |
1514 | } else if (qAbs(data->m22) < qreal(1./8.)) { |
1515 | // scale up more than 8x (on Y) |
1516 | bilinearFastTransformHelperARGB32PM[tiled][UpscaleTransform](b, end, data->texture, fx, fy, fdx, fdy); |
1517 | } else { |
1518 | // scale down on X |
1519 | bilinearFastTransformHelperARGB32PM[tiled][DownscaleTransform](b, end, data->texture, fx, fy, fdx, fdy); |
1520 | } |
1521 | } else { // rotation or shear |
1522 | if (qAbs(data->m11) < qreal(1./8.) || qAbs(data->m22) < qreal(1./8.) ) { |
1523 | // if we are zooming more than 8 times, we use 8bit precision for the position. |
1524 | bilinearFastTransformHelperARGB32PM[tiled][RotateTransform](b, end, data->texture, fx, fy, fdx, fdy); |
1525 | } else { |
1526 | // we are zooming less than 8x, use 4bit precision |
1527 | bilinearFastTransformHelperARGB32PM[tiled][FastRotateTransform](b, end, data->texture, fx, fy, fdx, fdy); |
1528 | } |
1529 | } |
1530 | } else { |
1531 | const QTextureData &image = data->texture; |
1532 | |
1533 | const qreal fdx = data->m11; |
1534 | const qreal fdy = data->m12; |
1535 | const qreal fdw = data->m13; |
1536 | |
1537 | qreal fx = data->m21 * cy + data->m11 * cx + data->dx; |
1538 | qreal fy = data->m22 * cy + data->m12 * cx + data->dy; |
1539 | qreal fw = data->m23 * cy + data->m13 * cx + data->m33; |
1540 | |
1541 | while (b < end) { |
1542 | const qreal iw = fw == 0 ? 1 : 1 / fw; |
1543 | const qreal px = fx * iw - qreal(0.5); |
1544 | const qreal py = fy * iw - qreal(0.5); |
1545 | |
1546 | int x1 = int(px) - (px < 0); |
1547 | int x2; |
1548 | int y1 = int(py) - (py < 0); |
1549 | int y2; |
1550 | |
1551 | int distx = int((px - x1) * 256); |
1552 | int disty = int((py - y1) * 256); |
1553 | |
1554 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1555 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1556 | |
1557 | const uint *s1 = (const uint *)data->texture.scanLine(y1); |
1558 | const uint *s2 = (const uint *)data->texture.scanLine(y2); |
1559 | |
1560 | uint tl = s1[x1]; |
1561 | uint tr = s1[x2]; |
1562 | uint bl = s2[x1]; |
1563 | uint br = s2[x2]; |
1564 | |
1565 | *b = interpolate_4_pixels(tl, tr, bl, br, distx, disty); |
1566 | |
1567 | fx += fdx; |
1568 | fy += fdy; |
1569 | fw += fdw; |
1570 | //force increment to avoid /0 |
1571 | if (!fw) { |
1572 | fw += fdw; |
1573 | } |
1574 | ++b; |
1575 | } |
1576 | } |
1577 | |
1578 | return buffer; |
1579 | } |
1580 | |
1581 | template<TextureBlendType blendType> |
1582 | static void QT_FASTCALL fetchTransformedBilinear_simple_scale_helper(uint *b, uint *end, const QTextureData &image, |
1583 | int &fx, int &fy, int fdx, int /*fdy*/) |
1584 | { |
1585 | const QPixelLayout *layout = &qPixelLayouts[image.format]; |
1586 | const QList<QRgb> *clut = image.colorTable; |
1587 | const FetchAndConvertPixelsFunc fetch = layout->fetchToARGB32PM; |
1588 | |
1589 | int y1 = (fy >> 16); |
1590 | int y2; |
1591 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1592 | const uchar *s1 = image.scanLine(y1); |
1593 | const uchar *s2 = image.scanLine(y2); |
1594 | |
1595 | const int disty = (fy & 0x0000ffff) >> 8; |
1596 | const int idisty = 256 - disty; |
1597 | const int length = end - b; |
1598 | |
1599 | // The intermediate buffer is generated in the positive direction |
1600 | const int adjust = (fdx < 0) ? fdx * length : 0; |
1601 | const int offset = (fx + adjust) >> 16; |
1602 | int x = offset; |
1603 | |
1604 | IntermediateBuffer intermediate; |
1605 | uint *buf1 = intermediate.buffer_rb; |
1606 | uint *buf2 = intermediate.buffer_ag; |
1607 | const uint *ptr1; |
1608 | const uint *ptr2; |
1609 | |
1610 | int count = (qint64(length) * qAbs(fdx) + fixed_scale - 1) / fixed_scale + 2; |
1611 | Q_ASSERT(count <= BufferSize + 2); |
1612 | |
1613 | if (blendType == BlendTransformedBilinearTiled) { |
1614 | x %= image.width; |
1615 | if (x < 0) |
1616 | x += image.width; |
1617 | int len1 = qMin(count, image.width - x); |
1618 | int len2 = qMin(x, count - len1); |
1619 | |
1620 | ptr1 = fetch(buf1, s1, x, len1, clut, nullptr); |
1621 | ptr2 = fetch(buf2, s2, x, len1, clut, nullptr); |
1622 | for (int i = 0; i < len1; ++i) { |
1623 | uint t = ptr1[i]; |
1624 | uint b = ptr2[i]; |
1625 | buf1[i] = (((t & 0xff00ff) * idisty + (b & 0xff00ff) * disty) >> 8) & 0xff00ff; |
1626 | buf2[i] = ((((t >> 8) & 0xff00ff) * idisty + ((b >> 8) & 0xff00ff) * disty) >> 8) & 0xff00ff; |
1627 | } |
1628 | |
1629 | if (len2) { |
1630 | ptr1 = fetch(buf1 + len1, s1, 0, len2, clut, nullptr); |
1631 | ptr2 = fetch(buf2 + len1, s2, 0, len2, clut, nullptr); |
1632 | for (int i = 0; i < len2; ++i) { |
1633 | uint t = ptr1[i]; |
1634 | uint b = ptr2[i]; |
1635 | buf1[i + len1] = (((t & 0xff00ff) * idisty + (b & 0xff00ff) * disty) >> 8) & 0xff00ff; |
1636 | buf2[i + len1] = ((((t >> 8) & 0xff00ff) * idisty + ((b >> 8) & 0xff00ff) * disty) >> 8) & 0xff00ff; |
1637 | } |
1638 | } |
1639 | // Generate the rest by repeatedly repeating the previous set of pixels |
1640 | for (int i = image.width; i < count; ++i) { |
1641 | buf1[i] = buf1[i - image.width]; |
1642 | buf2[i] = buf2[i - image.width]; |
1643 | } |
1644 | } else { |
1645 | int start = qMax(x, image.x1); |
1646 | int end = qMin(x + count, image.x2); |
1647 | int len = qMax(1, end - start); |
1648 | int leading = start - x; |
1649 | |
1650 | ptr1 = fetch(buf1 + leading, s1, start, len, clut, nullptr); |
1651 | ptr2 = fetch(buf2 + leading, s2, start, len, clut, nullptr); |
1652 | |
1653 | for (int i = 0; i < len; ++i) { |
1654 | uint t = ptr1[i]; |
1655 | uint b = ptr2[i]; |
1656 | buf1[i + leading] = (((t & 0xff00ff) * idisty + (b & 0xff00ff) * disty) >> 8) & 0xff00ff; |
1657 | buf2[i + leading] = ((((t >> 8) & 0xff00ff) * idisty + ((b >> 8) & 0xff00ff) * disty) >> 8) & 0xff00ff; |
1658 | } |
1659 | |
1660 | for (int i = 0; i < leading; ++i) { |
1661 | buf1[i] = buf1[leading]; |
1662 | buf2[i] = buf2[leading]; |
1663 | } |
1664 | for (int i = leading + len; i < count; ++i) { |
1665 | buf1[i] = buf1[i - 1]; |
1666 | buf2[i] = buf2[i - 1]; |
1667 | } |
1668 | } |
1669 | |
1670 | // Now interpolate the values from the intermediate.buffer to get the final result. |
1671 | intermediate_adder(b, end, intermediate, offset, fx, fdx); |
1672 | } |
1673 | |
1674 | |
1675 | template<TextureBlendType blendType, QPixelLayout::BPP bpp, typename T> |
1676 | static void QT_FASTCALL fetchTransformedBilinear_fetcher(T *buf1, T *buf2, const int len, const QTextureData &image, |
1677 | int fx, int fy, const int fdx, const int fdy) |
1678 | { |
1679 | const QPixelLayout &layout = qPixelLayouts[image.format]; |
1680 | constexpr bool useFetch = (bpp < QPixelLayout::BPP32); |
1681 | if (useFetch) |
1682 | Q_ASSERT(sizeof(T) == sizeof(uint)); |
1683 | else |
1684 | Q_ASSERT(layout.bpp == bpp); |
1685 | const FetchPixelFunc fetch1 = (bpp == QPixelLayout::BPPNone) ? qFetchPixelTable[layout.bpp] : qFetchPixel<bpp>; |
1686 | if (fdy == 0) { |
1687 | int y1 = (fy >> 16); |
1688 | int y2; |
1689 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1690 | const uchar *s1 = image.scanLine(y1); |
1691 | const uchar *s2 = image.scanLine(y2); |
1692 | |
1693 | int i = 0; |
1694 | if (blendType == BlendTransformedBilinear) { |
1695 | for (; i < len; ++i) { |
1696 | int x1 = (fx >> 16); |
1697 | int x2; |
1698 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1699 | if (x1 != x2) |
1700 | break; |
1701 | if (useFetch) { |
1702 | buf1[i * 2 + 0] = buf1[i * 2 + 1] = fetch1(s1, x1); |
1703 | buf2[i * 2 + 0] = buf2[i * 2 + 1] = fetch1(s2, x1); |
1704 | } else { |
1705 | buf1[i * 2 + 0] = buf1[i * 2 + 1] = reinterpret_cast<const T *>(s1)[x1]; |
1706 | buf2[i * 2 + 0] = buf2[i * 2 + 1] = reinterpret_cast<const T *>(s2)[x1]; |
1707 | } |
1708 | fx += fdx; |
1709 | } |
1710 | int fastLen = len; |
1711 | if (fdx > 0) |
1712 | fastLen = qMin(fastLen, int((qint64(image.x2 - 1) * fixed_scale - fx) / fdx)); |
1713 | else if (fdx < 0) |
1714 | fastLen = qMin(fastLen, int((qint64(image.x1) * fixed_scale - fx) / fdx)); |
1715 | |
1716 | for (; i < fastLen; ++i) { |
1717 | int x = (fx >> 16); |
1718 | if (useFetch) { |
1719 | buf1[i * 2 + 0] = fetch1(s1, x); |
1720 | buf1[i * 2 + 1] = fetch1(s1, x + 1); |
1721 | buf2[i * 2 + 0] = fetch1(s2, x); |
1722 | buf2[i * 2 + 1] = fetch1(s2, x + 1); |
1723 | } else { |
1724 | buf1[i * 2 + 0] = reinterpret_cast<const T *>(s1)[x]; |
1725 | buf1[i * 2 + 1] = reinterpret_cast<const T *>(s1)[x + 1]; |
1726 | buf2[i * 2 + 0] = reinterpret_cast<const T *>(s2)[x]; |
1727 | buf2[i * 2 + 1] = reinterpret_cast<const T *>(s2)[x + 1]; |
1728 | } |
1729 | fx += fdx; |
1730 | } |
1731 | } |
1732 | |
1733 | for (; i < len; ++i) { |
1734 | int x1 = (fx >> 16); |
1735 | int x2; |
1736 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1737 | if (useFetch) { |
1738 | buf1[i * 2 + 0] = fetch1(s1, x1); |
1739 | buf1[i * 2 + 1] = fetch1(s1, x2); |
1740 | buf2[i * 2 + 0] = fetch1(s2, x1); |
1741 | buf2[i * 2 + 1] = fetch1(s2, x2); |
1742 | } else { |
1743 | buf1[i * 2 + 0] = reinterpret_cast<const T *>(s1)[x1]; |
1744 | buf1[i * 2 + 1] = reinterpret_cast<const T *>(s1)[x2]; |
1745 | buf2[i * 2 + 0] = reinterpret_cast<const T *>(s2)[x1]; |
1746 | buf2[i * 2 + 1] = reinterpret_cast<const T *>(s2)[x2]; |
1747 | } |
1748 | fx += fdx; |
1749 | } |
1750 | } else { |
1751 | int i = 0; |
1752 | if (blendType == BlendTransformedBilinear) { |
1753 | for (; i < len; ++i) { |
1754 | int x1 = (fx >> 16); |
1755 | int x2; |
1756 | int y1 = (fy >> 16); |
1757 | int y2; |
1758 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1759 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1760 | if (x1 != x2 && y1 != y2) |
1761 | break; |
1762 | const uchar *s1 = image.scanLine(y1); |
1763 | const uchar *s2 = image.scanLine(y2); |
1764 | if (useFetch) { |
1765 | buf1[i * 2 + 0] = fetch1(s1, x1); |
1766 | buf1[i * 2 + 1] = fetch1(s1, x2); |
1767 | buf2[i * 2 + 0] = fetch1(s2, x1); |
1768 | buf2[i * 2 + 1] = fetch1(s2, x2); |
1769 | } else { |
1770 | buf1[i * 2 + 0] = reinterpret_cast<const T *>(s1)[x1]; |
1771 | buf1[i * 2 + 1] = reinterpret_cast<const T *>(s1)[x2]; |
1772 | buf2[i * 2 + 0] = reinterpret_cast<const T *>(s2)[x1]; |
1773 | buf2[i * 2 + 1] = reinterpret_cast<const T *>(s2)[x2]; |
1774 | } |
1775 | fx += fdx; |
1776 | fy += fdy; |
1777 | } |
1778 | int fastLen = len; |
1779 | if (fdx > 0) |
1780 | fastLen = qMin(fastLen, int((qint64(image.x2 - 1) * fixed_scale - fx) / fdx)); |
1781 | else if (fdx < 0) |
1782 | fastLen = qMin(fastLen, int((qint64(image.x1) * fixed_scale - fx) / fdx)); |
1783 | if (fdy > 0) |
1784 | fastLen = qMin(fastLen, int((qint64(image.y2 - 1) * fixed_scale - fy) / fdy)); |
1785 | else if (fdy < 0) |
1786 | fastLen = qMin(fastLen, int((qint64(image.y1) * fixed_scale - fy) / fdy)); |
1787 | |
1788 | for (; i < fastLen; ++i) { |
1789 | int x = (fx >> 16); |
1790 | int y = (fy >> 16); |
1791 | const uchar *s1 = image.scanLine(y); |
1792 | const uchar *s2 = s1 + image.bytesPerLine; |
1793 | if (useFetch) { |
1794 | buf1[i * 2 + 0] = fetch1(s1, x); |
1795 | buf1[i * 2 + 1] = fetch1(s1, x + 1); |
1796 | buf2[i * 2 + 0] = fetch1(s2, x); |
1797 | buf2[i * 2 + 1] = fetch1(s2, x + 1); |
1798 | } else { |
1799 | buf1[i * 2 + 0] = reinterpret_cast<const T *>(s1)[x]; |
1800 | buf1[i * 2 + 1] = reinterpret_cast<const T *>(s1)[x + 1]; |
1801 | buf2[i * 2 + 0] = reinterpret_cast<const T *>(s2)[x]; |
1802 | buf2[i * 2 + 1] = reinterpret_cast<const T *>(s2)[x + 1]; |
1803 | } |
1804 | fx += fdx; |
1805 | fy += fdy; |
1806 | } |
1807 | } |
1808 | |
1809 | for (; i < len; ++i) { |
1810 | int x1 = (fx >> 16); |
1811 | int x2; |
1812 | int y1 = (fy >> 16); |
1813 | int y2; |
1814 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1815 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1816 | |
1817 | const uchar *s1 = image.scanLine(y1); |
1818 | const uchar *s2 = image.scanLine(y2); |
1819 | if (useFetch) { |
1820 | buf1[i * 2 + 0] = fetch1(s1, x1); |
1821 | buf1[i * 2 + 1] = fetch1(s1, x2); |
1822 | buf2[i * 2 + 0] = fetch1(s2, x1); |
1823 | buf2[i * 2 + 1] = fetch1(s2, x2); |
1824 | } else { |
1825 | buf1[i * 2 + 0] = reinterpret_cast<const T *>(s1)[x1]; |
1826 | buf1[i * 2 + 1] = reinterpret_cast<const T *>(s1)[x2]; |
1827 | buf2[i * 2 + 0] = reinterpret_cast<const T *>(s2)[x1]; |
1828 | buf2[i * 2 + 1] = reinterpret_cast<const T *>(s2)[x2]; |
1829 | } |
1830 | fx += fdx; |
1831 | fy += fdy; |
1832 | } |
1833 | } |
1834 | } |
1835 | |
1836 | template<TextureBlendType blendType, QPixelLayout::BPP bpp, typename T> |
1837 | static void QT_FASTCALL fetchTransformedBilinear_slow_fetcher(T *buf1, T *buf2, ushort *distxs, ushort *distys, |
1838 | const int len, const QTextureData &image, |
1839 | qreal &fx, qreal &fy, qreal &fw, |
1840 | const qreal fdx, const qreal fdy, const qreal fdw) |
1841 | { |
1842 | const QPixelLayout &layout = qPixelLayouts[image.format]; |
1843 | constexpr bool useFetch = (bpp < QPixelLayout::BPP32); |
1844 | if (useFetch) |
1845 | Q_ASSERT(sizeof(T) == sizeof(uint)); |
1846 | else |
1847 | Q_ASSERT(layout.bpp == bpp); |
1848 | |
1849 | const FetchPixelFunc fetch1 = (bpp == QPixelLayout::BPPNone) ? qFetchPixelTable[layout.bpp] : qFetchPixel<bpp>; |
1850 | |
1851 | for (int i = 0; i < len; ++i) { |
1852 | const qreal iw = fw == 0 ? 16384 : 1 / fw; |
1853 | const qreal px = fx * iw - qreal(0.5); |
1854 | const qreal py = fy * iw - qreal(0.5); |
1855 | |
1856 | int x1 = qFloor(px); |
1857 | int x2; |
1858 | int y1 = qFloor(py); |
1859 | int y2; |
1860 | |
1861 | distxs[i] = ushort((px - x1) * (1<<16)); |
1862 | distys[i] = ushort((py - y1) * (1<<16)); |
1863 | |
1864 | fetchTransformedBilinear_pixelBounds<blendType>(image.width, image.x1, image.x2 - 1, x1, x2); |
1865 | fetchTransformedBilinear_pixelBounds<blendType>(image.height, image.y1, image.y2 - 1, y1, y2); |
1866 | |
1867 | const uchar *s1 = image.scanLine(y1); |
1868 | const uchar *s2 = image.scanLine(y2); |
1869 | if (useFetch) { |
1870 | buf1[i * 2 + 0] = fetch1(s1, x1); |
1871 | buf1[i * 2 + 1] = fetch1(s1, x2); |
1872 | buf2[i * 2 + 0] = fetch1(s2, x1); |
1873 | buf2[i * 2 + 1] = fetch1(s2, x2); |
1874 | } else { |
1875 | buf1[i * 2 + 0] = reinterpret_cast<const T *>(s1)[x1]; |
1876 | buf1[i * 2 + 1] = reinterpret_cast<const T *>(s1)[x2]; |
1877 | buf2[i * 2 + 0] = reinterpret_cast<const T *>(s2)[x1]; |
1878 | buf2[i * 2 + 1] = reinterpret_cast<const T *>(s2)[x2]; |
1879 | } |
1880 | |
1881 | fx += fdx; |
1882 | fy += fdy; |
1883 | fw += fdw; |
1884 | } |
1885 | } |
1886 | |
1887 | // blendType = BlendTransformedBilinear or BlendTransformedBilinearTiled |
1888 | template<TextureBlendType blendType, QPixelLayout::BPP bpp> |
1889 | static const uint *QT_FASTCALL fetchTransformedBilinear(uint *buffer, const Operator *, |
1890 | const QSpanData *data, int y, int x, int length) |
1891 | { |
1892 | const QPixelLayout *layout = &qPixelLayouts[data->texture.format]; |
1893 | const QList<QRgb> *clut = data->texture.colorTable; |
1894 | Q_ASSERT(bpp == QPixelLayout::BPPNone || layout->bpp == bpp); |
1895 | |
1896 | const qreal cx = x + qreal(0.5); |
1897 | const qreal cy = y + qreal(0.5); |
1898 | |
1899 | if (canUseFastMatrixPath(cx, cy, length, data)) { |
1900 | // The increment pr x in the scanline |
1901 | int fdx = (int)(data->m11 * fixed_scale); |
1902 | int fdy = (int)(data->m12 * fixed_scale); |
1903 | |
1904 | int fx = int((data->m21 * cy + data->m11 * cx + data->dx) * fixed_scale); |
1905 | int fy = int((data->m22 * cy + data->m12 * cx + data->dy) * fixed_scale); |
1906 | |
1907 | fx -= half_point; |
1908 | fy -= half_point; |
1909 | |
1910 | if (fdy == 0) { // simple scale, no rotation or shear |
1911 | if (qAbs(fdx) <= fixed_scale) { // scale up on X |
1912 | fetchTransformedBilinear_simple_scale_helper<blendType>(buffer, buffer + length, data->texture, fx, fy, fdx, fdy); |
1913 | } else if (qAbs(fdx) <= 2 * fixed_scale) { // scale down on X less than 2x |
1914 | const int mid = (length * 2 < BufferSize) ? length : ((length + 1) / 2); |
1915 | fetchTransformedBilinear_simple_scale_helper<blendType>(buffer, buffer + mid, data->texture, fx, fy, fdx, fdy); |
1916 | if (mid != length) |
1917 | fetchTransformedBilinear_simple_scale_helper<blendType>(buffer + mid, buffer + length, data->texture, fx, fy, fdx, fdy); |
1918 | } else { |
1919 | const auto fetcher = fetchTransformedBilinear_fetcher<blendType,bpp,uint>; |
1920 | |
1921 | uint buf1[BufferSize]; |
1922 | uint buf2[BufferSize]; |
1923 | uint *b = buffer; |
1924 | while (length) { |
1925 | int len = qMin(length, BufferSize / 2); |
1926 | fetcher(buf1, buf2, len, data->texture, fx, fy, fdx, 0); |
1927 | layout->convertToARGB32PM(buf1, len * 2, clut); |
1928 | layout->convertToARGB32PM(buf2, len * 2, clut); |
1929 | |
1930 | if (hasFastInterpolate4() || qAbs(data->m22) < qreal(1./8.)) { // scale up more than 8x (on Y) |
1931 | int disty = (fy & 0x0000ffff) >> 8; |
1932 | for (int i = 0; i < len; ++i) { |
1933 | int distx = (fx & 0x0000ffff) >> 8; |
1934 | b[i] = interpolate_4_pixels(buf1 + i * 2, buf2 + i * 2, distx, disty); |
1935 | fx += fdx; |
1936 | } |
1937 | } else { |
1938 | int disty = ((fy & 0x0000ffff) + 0x0800) >> 12; |
1939 | for (int i = 0; i < len; ++i) { |
1940 | uint tl = buf1[i * 2 + 0]; |
1941 | uint tr = buf1[i * 2 + 1]; |
1942 | uint bl = buf2[i * 2 + 0]; |
1943 | uint br = buf2[i * 2 + 1]; |
1944 | int distx = ((fx & 0x0000ffff) + 0x0800) >> 12; |
1945 | b[i] = interpolate_4_pixels_16(tl, tr, bl, br, distx, disty); |
1946 | fx += fdx; |
1947 | } |
1948 | } |
1949 | length -= len; |
1950 | b += len; |
1951 | } |
1952 | } |
1953 | } else { // rotation or shear |
1954 | const auto fetcher = fetchTransformedBilinear_fetcher<blendType,bpp,uint>; |
1955 | |
1956 | uint buf1[BufferSize]; |
1957 | uint buf2[BufferSize]; |
1958 | uint *b = buffer; |
1959 | while (length) { |
1960 | int len = qMin(length, BufferSize / 2); |
1961 | fetcher(buf1, buf2, len, data->texture, fx, fy, fdx, fdy); |
1962 | layout->convertToARGB32PM(buf1, len * 2, clut); |
1963 | layout->convertToARGB32PM(buf2, len * 2, clut); |
1964 | |
1965 | if (hasFastInterpolate4() || qAbs(data->m11) < qreal(1./8.) || qAbs(data->m22) < qreal(1./8.)) { |
1966 | // If we are zooming more than 8 times, we use 8bit precision for the position. |
1967 | for (int i = 0; i < len; ++i) { |
1968 | int distx = (fx & 0x0000ffff) >> 8; |
1969 | int disty = (fy & 0x0000ffff) >> 8; |
1970 | |
1971 | b[i] = interpolate_4_pixels(buf1 + i * 2, buf2 + i * 2, distx, disty); |
1972 | fx += fdx; |
1973 | fy += fdy; |
1974 | } |
1975 | } else { |
1976 | // We are zooming less than 8x, use 4bit precision |
1977 | for (int i = 0; i < len; ++i) { |
1978 | uint tl = buf1[i * 2 + 0]; |
1979 | uint tr = buf1[i * 2 + 1]; |
1980 | uint bl = buf2[i * 2 + 0]; |
1981 | uint br = buf2[i * 2 + 1]; |
1982 | |
1983 | int distx = ((fx & 0x0000ffff) + 0x0800) >> 12; |
1984 | int disty = ((fy & 0x0000ffff) + 0x0800) >> 12; |
1985 | |
1986 | b[i] = interpolate_4_pixels_16(tl, tr, bl, br, distx, disty); |
1987 | fx += fdx; |
1988 | fy += fdy; |
1989 | } |
1990 | } |
1991 | |
1992 | length -= len; |
1993 | b += len; |
1994 | } |
1995 | } |
1996 | } else { |
1997 | const auto fetcher = fetchTransformedBilinear_slow_fetcher<blendType,bpp,uint>; |
1998 | |
1999 | const qreal fdx = data->m11; |
2000 | const qreal fdy = data->m12; |
2001 | const qreal fdw = data->m13; |
2002 | |
2003 | qreal fx = data->m21 * cy + data->m11 * cx + data->dx; |
2004 | qreal fy = data->m22 * cy + data->m12 * cx + data->dy; |
2005 | qreal fw = data->m23 * cy + data->m13 * cx + data->m33; |
2006 | |
2007 | uint buf1[BufferSize]; |
2008 | uint buf2[BufferSize]; |
2009 | uint *b = buffer; |
2010 | |
2011 | ushort distxs[BufferSize / 2]; |
2012 | ushort distys[BufferSize / 2]; |
2013 | |
2014 | while (length) { |
2015 | const int len = qMin(length, BufferSize / 2); |
2016 | fetcher(buf1, buf2, distxs, distys, len, data->texture, fx, fy, fw, fdx, fdy, fdw); |
2017 | |
2018 | layout->convertToARGB32PM(buf1, len * 2, clut); |
2019 | layout->convertToARGB32PM(buf2, len * 2, clut); |
2020 | |
2021 | for (int i = 0; i < len; ++i) { |
2022 | const int distx = distxs[i] >> 8; |
2023 | const int disty = distys[i] >> 8; |
2024 | |
2025 | b[i] = interpolate_4_pixels(buf1 + i * 2, buf2 + i * 2, distx, disty); |
2026 | } |
2027 | length -= len; |
2028 | b += len; |
2029 | } |
2030 | } |
2031 | |
2032 | return buffer; |
2033 | } |
2034 | |
2035 | #if QT_CONFIG(raster_64bit) |
2036 | template<TextureBlendType blendType> |
2037 | static const QRgba64 *QT_FASTCALL fetchTransformedBilinear64_uint32(QRgba64 *buffer, const QSpanData *data, |
2038 | int y, int x, int length) |
2039 | { |
2040 | const QPixelLayout *layout = &qPixelLayouts[data->texture.format]; |
2041 | const QList<QRgb> *clut = data->texture.colorTable; |
2042 | |
2043 | const qreal cx = x + qreal(0.5); |
2044 | const qreal cy = y + qreal(0.5); |
2045 | |
2046 | uint sbuf1[BufferSize]; |
2047 | uint sbuf2[BufferSize]; |
2048 | alignas(8) QRgba64 buf1[BufferSize]; |
2049 | alignas(8) QRgba64 buf2[BufferSize]; |
2050 | QRgba64 *end = buffer + length; |
2051 | QRgba64 *b = buffer; |
2052 | |
2053 | if (canUseFastMatrixPath(cx, cy, length, data)) { |
2054 | // The increment pr x in the scanline |
2055 | const int fdx = (int)(data->m11 * fixed_scale); |
2056 | const int fdy = (int)(data->m12 * fixed_scale); |
2057 | |
2058 | int fx = int((data->m21 * cy + data->m11 * cx + data->dx) * fixed_scale); |
2059 | int fy = int((data->m22 * cy + data->m12 * cx + data->dy) * fixed_scale); |
2060 | |
2061 | fx -= half_point; |
2062 | fy -= half_point; |
2063 | |
2064 | const auto fetcher = |
2065 | (layout->bpp == QPixelLayout::BPP32) |
2066 | ? fetchTransformedBilinear_fetcher<blendType, QPixelLayout::BPP32, uint> |
2067 | : fetchTransformedBilinear_fetcher<blendType, QPixelLayout::BPPNone, uint>; |
2068 | |
2069 | if (fdy == 0) { //simple scale, no rotation |
2070 | while (length) { |
2071 | int len = qMin(length, BufferSize / 2); |
2072 | int disty = (fy & 0x0000ffff); |
2073 | #if defined(__SSE2__) |
2074 | const __m128i vdy = _mm_set1_epi16(disty); |
2075 | const __m128i vidy = _mm_set1_epi16(0x10000 - disty); |
2076 | #endif |
2077 | fetcher(sbuf1, sbuf2, len, data->texture, fx, fy, fdx, fdy); |
2078 | |
2079 | layout->convertToRGBA64PM(buf1, sbuf1, len * 2, clut, nullptr); |
2080 | if (disty) |
2081 | layout->convertToRGBA64PM(buf2, sbuf2, len * 2, clut, nullptr); |
2082 | |
2083 | for (int i = 0; i < len; ++i) { |
2084 | int distx = (fx & 0x0000ffff); |
2085 | #if defined(__SSE2__) |
2086 | __m128i vt = _mm_loadu_si128((const __m128i*)(buf1 + i*2)); |
2087 | if (disty) { |
2088 | __m128i vb = _mm_loadu_si128((const __m128i*)(buf2 + i*2)); |
2089 | vt = _mm_mulhi_epu16(vt, vidy); |
2090 | vb = _mm_mulhi_epu16(vb, vdy); |
2091 | vt = _mm_add_epi16(vt, vb); |
2092 | } |
2093 | if (distx) { |
2094 | const __m128i vdistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(distx), _MM_SHUFFLE(0, 0, 0, 0)); |
2095 | const __m128i vidistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(0x10000 - distx), _MM_SHUFFLE(0, 0, 0, 0)); |
2096 | vt = _mm_mulhi_epu16(vt, _mm_unpacklo_epi64(vidistx, vdistx)); |
2097 | vt = _mm_add_epi16(vt, _mm_srli_si128(vt, 8)); |
2098 | } |
2099 | _mm_storel_epi64((__m128i*)(b+i), vt); |
2100 | #else |
2101 | b[i] = interpolate_4_pixels_rgb64(buf1 + i*2, buf2 + i*2, distx, disty); |
2102 | #endif |
2103 | fx += fdx; |
2104 | } |
2105 | length -= len; |
2106 | b += len; |
2107 | } |
2108 | } else { // rotation or shear |
2109 | while (b < end) { |
2110 | int len = qMin(length, BufferSize / 2); |
2111 | |
2112 | fetcher(sbuf1, sbuf2, len, data->texture, fx, fy, fdx, fdy); |
2113 | |
2114 | layout->convertToRGBA64PM(buf1, sbuf1, len * 2, clut, nullptr); |
2115 | layout->convertToRGBA64PM(buf2, sbuf2, len * 2, clut, nullptr); |
2116 | |
2117 | for (int i = 0; i < len; ++i) { |
2118 | int distx = (fx & 0x0000ffff); |
2119 | int disty = (fy & 0x0000ffff); |
2120 | b[i] = interpolate_4_pixels_rgb64(buf1 + i*2, buf2 + i*2, distx, disty); |
2121 | fx += fdx; |
2122 | fy += fdy; |
2123 | } |
2124 | |
2125 | length -= len; |
2126 | b += len; |
2127 | } |
2128 | } |
2129 | } else { // !(data->fast_matrix) |
2130 | const auto fetcher = |
2131 | (layout->bpp == QPixelLayout::BPP32) |
2132 | ? fetchTransformedBilinear_slow_fetcher<blendType, QPixelLayout::BPP32, uint> |
2133 | : fetchTransformedBilinear_slow_fetcher<blendType, QPixelLayout::BPPNone, uint>; |
2134 | |
2135 | const qreal fdx = data->m11; |
2136 | const qreal fdy = data->m12; |
2137 | const qreal fdw = data->m13; |
2138 | |
2139 | qreal fx = data->m21 * cy + data->m11 * cx + data->dx; |
2140 | qreal fy = data->m22 * cy + data->m12 * cx + data->dy; |
2141 | qreal fw = data->m23 * cy + data->m13 * cx + data->m33; |
2142 | |
2143 | ushort distxs[BufferSize / 2]; |
2144 | ushort distys[BufferSize / 2]; |
2145 | |
2146 | while (length) { |
2147 | const int len = qMin(length, BufferSize / 2); |
2148 | fetcher(sbuf1, sbuf2, distxs, distys, len, data->texture, fx, fy, fw, fdx, fdy, fdw); |
2149 | |
2150 | layout->convertToRGBA64PM(buf1, sbuf1, len * 2, clut, nullptr); |
2151 | layout->convertToRGBA64PM(buf2, sbuf2, len * 2, clut, nullptr); |
2152 | |
2153 | for (int i = 0; i < len; ++i) { |
2154 | const int distx = distxs[i]; |
2155 | const int disty = distys[i]; |
2156 | b[i] = interpolate_4_pixels_rgb64(buf1 + i*2, buf2 + i*2, distx, disty); |
2157 | } |
2158 | |
2159 | length -= len; |
2160 | b += len; |
2161 | } |
2162 | } |
2163 | return buffer; |
2164 | } |
2165 | |
2166 | template<TextureBlendType blendType> |
2167 | static const QRgba64 *QT_FASTCALL fetchTransformedBilinear64_uint64(QRgba64 *buffer, const QSpanData *data, |
2168 | int y, int x, int length) |
2169 | { |
2170 | Q_ASSERT(qPixelLayouts[data->texture.format].bpp == QPixelLayout::BPP64); |
2171 | const auto convert = (data->texture.format == QImage::Format_RGBA64) ? convertRGBA64ToRGBA64PM : convertRGBA64PMToRGBA64PM; |
2172 | |
2173 | const qreal cx = x + qreal(0.5); |
2174 | const qreal cy = y + qreal(0.5); |
2175 | |
2176 | alignas(8) QRgba64 buf1[BufferSize]; |
2177 | alignas(8) QRgba64 buf2[BufferSize]; |
2178 | QRgba64 *end = buffer + length; |
2179 | QRgba64 *b = buffer; |
2180 | |
2181 | if (canUseFastMatrixPath(cx, cy, length, data)) { |
2182 | // The increment pr x in the scanline |
2183 | const int fdx = (int)(data->m11 * fixed_scale); |
2184 | const int fdy = (int)(data->m12 * fixed_scale); |
2185 | |
2186 | int fx = int((data->m21 * cy + data->m11 * cx + data->dx) * fixed_scale); |
2187 | int fy = int((data->m22 * cy + data->m12 * cx + data->dy) * fixed_scale); |
2188 | |
2189 | fx -= half_point; |
2190 | fy -= half_point; |
2191 | const auto fetcher = fetchTransformedBilinear_fetcher<blendType, QPixelLayout::BPP64, QRgba64>; |
2192 | |
2193 | if (fdy == 0) { //simple scale, no rotation |
2194 | while (length) { |
2195 | int len = qMin(length, BufferSize / 2); |
2196 | int disty = (fy & 0x0000ffff); |
2197 | #if defined(__SSE2__) |
2198 | const __m128i vdy = _mm_set1_epi16(disty); |
2199 | const __m128i vidy = _mm_set1_epi16(0x10000 - disty); |
2200 | #endif |
2201 | fetcher(buf1, buf2, len, data->texture, fx, fy, fdx, fdy); |
2202 | |
2203 | convert(buf1, len * 2); |
2204 | if (disty) |
2205 | convert(buf2, len * 2); |
2206 | |
2207 | for (int i = 0; i < len; ++i) { |
2208 | int distx = (fx & 0x0000ffff); |
2209 | #if defined(__SSE2__) |
2210 | __m128i vt = _mm_loadu_si128((const __m128i*)(buf1 + i*2)); |
2211 | if (disty) { |
2212 | __m128i vb = _mm_loadu_si128((const __m128i*)(buf2 + i*2)); |
2213 | vt = _mm_mulhi_epu16(vt, vidy); |
2214 | vb = _mm_mulhi_epu16(vb, vdy); |
2215 | vt = _mm_add_epi16(vt, vb); |
2216 | } |
2217 | if (distx) { |
2218 | const __m128i vdistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(distx), _MM_SHUFFLE(0, 0, 0, 0)); |
2219 | const __m128i vidistx = _mm_shufflelo_epi16(_mm_cvtsi32_si128(0x10000 - distx), _MM_SHUFFLE(0, 0, 0, 0)); |
2220 | vt = _mm_mulhi_epu16(vt, _mm_unpacklo_epi64(vidistx, vdistx)); |
2221 | vt = _mm_add_epi16(vt, _mm_srli_si128(vt, 8)); |
2222 | } |
2223 | _mm_storel_epi64((__m128i*)(b+i), vt); |
2224 | #else |
2225 | b[i] = interpolate_4_pixels_rgb64(buf1 + i*2, buf2 + i*2, distx, disty); |
2226 | #endif |
2227 | fx += fdx; |
2228 | } |
2229 | length -= len; |
2230 | b += len; |
2231 | } |
2232 | } else { // rotation or shear |
2233 | while (b < end) { |
2234 | int len = qMin(length, BufferSize / 2); |
2235 | |
2236 | fetcher(buf1, buf2, len, data->texture, fx, fy, fdx, fdy); |
2237 | |
2238 | convert(buf1, len * 2); |
2239 | convert(buf2, len * 2); |
2240 | |
2241 | for (int i = 0; i < len; ++i) { |
2242 | int distx = (fx & 0x0000ffff); |
2243 | int disty = (fy & 0x0000ffff); |
2244 | b[i] = interpolate_4_pixels_rgb64(buf1 + i*2, buf2 + i*2, distx, disty); |
2245 | fx += fdx; |
2246 | fy += fdy; |
2247 | } |
2248 | |
2249 | length -= len; |
2250 | b += len; |
2251 | } |
2252 | } |
2253 | } else { // !(data->fast_matrix) |
2254 | const auto fetcher = fetchTransformedBilinear_slow_fetcher<blendType, QPixelLayout::BPP64, QRgba64>; |
2255 | |
2256 | const qreal fdx = data->m11; |
2257 | const qreal fdy = data->m12; |
2258 | const qreal fdw = data->m13; |
2259 | |
2260 | qreal fx = data->m21 * cy + data->m11 * cx + data->dx; |
2261 | qreal fy = data->m22 * cy + data->m12 * cx + data->dy; |
2262 | qreal fw = data->m23 * cy + data->m13 * cx + data->m33; |
2263 | |
2264 | ushort distxs[BufferSize / 2]; |
2265 | ushort distys[BufferSize / 2]; |
2266 | |
2267 | while (length) { |
2268 | const int len = qMin(length, BufferSize / 2); |
2269 | fetcher(buf1, buf2, distxs, distys, len, data->texture, fx, fy, fw, fdx, fdy, fdw); |
2270 | |
2271 | convert(buf1, len * 2); |
2272 | convert(buf2, len * 2); |
2273 | |
2274 | for (int i = 0; i < len; ++i) { |
2275 | const int distx = distxs[i]; |
2276 | const int disty = distys[i]; |
2277 | b[i] = interpolate_4_pixels_rgb64(buf1 + i*2, buf2 + i*2, distx, disty); |
2278 | } |
2279 | |
2280 | length -= len; |
2281 | b += len; |
2282 | } |
2283 | } |
2284 | return buffer; |
2285 | } |
2286 | |
2287 | template<TextureBlendType blendType> |
2288 | static const QRgba64 *QT_FASTCALL fetchTransformedBilinear64(QRgba64 *buffer, const Operator *, |
2289 | const QSpanData *data, int y, int x, int length) |
2290 | { |
2291 | if (qPixelLayouts[data->texture.format].bpp == QPixelLayout::BPP64) |
2292 | return fetchTransformedBilinear64_uint64<blendType>(buffer, data, y, x, length); |
2293 | return fetchTransformedBilinear64_uint32<blendType>(buffer, data, y, x, length); |
2294 | } |
2295 | #endif |
2296 | |
2297 | // FetchUntransformed can have more specialized methods added depending on SIMD features. |
2298 | static SourceFetchProc sourceFetchUntransformed[QImage::NImageFormats] = { |
2299 | nullptr, // Invalid |
2300 | fetchUntransformed, // Mono |
2301 | fetchUntransformed, // MonoLsb |
2302 | fetchUntransformed, // Indexed8 |
2303 | fetchUntransformedARGB32PM, // RGB32 |
2304 | fetchUntransformed, // ARGB32 |
2305 | fetchUntransformedARGB32PM, // ARGB32_Premultiplied |
2306 | fetchUntransformedRGB16, // RGB16 |
2307 | fetchUntransformed, // ARGB8565_Premultiplied |
2308 | fetchUntransformed, // RGB666 |
2309 | fetchUntransformed, // ARGB6666_Premultiplied |
2310 | fetchUntransformed, // RGB555 |
2311 | fetchUntransformed, // ARGB8555_Premultiplied |
2312 | fetchUntransformed, // RGB888 |
2313 | fetchUntransformed, // RGB444 |
2314 | fetchUntransformed, // ARGB4444_Premultiplied |
2315 | fetchUntransformed, // RGBX8888 |
2316 | fetchUntransformed, // RGBA8888 |
2317 | fetchUntransformed, // RGBA8888_Premultiplied |
2318 | fetchUntransformed, // Format_BGR30 |
2319 | fetchUntransformed, // Format_A2BGR30_Premultiplied |
2320 | fetchUntransformed, // Format_RGB30 |
2321 | fetchUntransformed, // Format_A2RGB30_Premultiplied |
2322 | fetchUntransformed, // Alpha8 |
2323 | fetchUntransformed, // Grayscale8 |
2324 | fetchUntransformed, // RGBX64 |
2325 | fetchUntransformed, // RGBA64 |
2326 | fetchUntransformed, // RGBA64_Premultiplied |
2327 | fetchUntransformed, // Grayscale16 |
2328 | fetchUntransformed, // BGR888 |
2329 | }; |
2330 | |
2331 | static const SourceFetchProc sourceFetchGeneric[NBlendTypes] = { |
2332 | fetchUntransformed, // Untransformed |
2333 | fetchUntransformed, // Tiled |
2334 | fetchTransformed<BlendTransformed, QPixelLayout::BPPNone>, // Transformed |
2335 | fetchTransformed<BlendTransformedTiled, QPixelLayout::BPPNone>, // TransformedTiled |
2336 | fetchTransformedBilinear<BlendTransformedBilinear, QPixelLayout::BPPNone>, // TransformedBilinear |
2337 | fetchTransformedBilinear<BlendTransformedBilinearTiled, QPixelLayout::BPPNone> // TransformedBilinearTiled |
2338 | }; |
2339 | |
2340 | static SourceFetchProc sourceFetchARGB32PM[NBlendTypes] = { |
2341 | fetchUntransformedARGB32PM, // Untransformed |
2342 | fetchUntransformedARGB32PM, // Tiled |
2343 | fetchTransformed<BlendTransformed, QPixelLayout::BPP32>, // Transformed |
2344 | fetchTransformed<BlendTransformedTiled, QPixelLayout::BPP32>, // TransformedTiled |
2345 | fetchTransformedBilinearARGB32PM<BlendTransformedBilinear>, // Bilinear |
2346 | fetchTransformedBilinearARGB32PM<BlendTransformedBilinearTiled> // BilinearTiled |
2347 | }; |
2348 | |
2349 | static SourceFetchProc sourceFetchAny16[NBlendTypes] = { |
2350 | fetchUntransformed, // Untransformed |
2351 | fetchUntransformed, // Tiled |
2352 | fetchTransformed<BlendTransformed, QPixelLayout::BPP16>, // Transformed |
2353 | fetchTransformed<BlendTransformedTiled, QPixelLayout::BPP16>, // TransformedTiled |
2354 | fetchTransformedBilinear<BlendTransformedBilinear, QPixelLayout::BPP16>, // TransformedBilinear |
2355 | fetchTransformedBilinear<BlendTransformedBilinearTiled, QPixelLayout::BPP16> // TransformedBilinearTiled |
2356 | }; |
2357 | |
2358 | static SourceFetchProc sourceFetchAny32[NBlendTypes] = { |
2359 | fetchUntransformed, // Untransformed |
2360 | fetchUntransformed, // Tiled |
2361 | fetchTransformed<BlendTransformed, QPixelLayout::BPP32>, // Transformed |
2362 | fetchTransformed<BlendTransformedTiled, QPixelLayout::BPP32>, // TransformedTiled |
2363 | fetchTransformedBilinear<BlendTransformedBilinear, QPixelLayout::BPP32>, // TransformedBilinear |
2364 | fetchTransformedBilinear<BlendTransformedBilinearTiled, QPixelLayout::BPP32> // TransformedBilinearTiled |
2365 | }; |
2366 | |
2367 | static inline SourceFetchProc getSourceFetch(TextureBlendType blendType, QImage::Format format) |
2368 | { |
2369 | if (format == QImage::Format_RGB32 || format == QImage::Format_ARGB32_Premultiplied) |
2370 | return sourceFetchARGB32PM[blendType]; |
2371 | if (blendType == BlendUntransformed || blendType == BlendTiled) |
2372 | return sourceFetchUntransformed[format]; |
2373 | if (qPixelLayouts[format].bpp == QPixelLayout::BPP16) |
2374 | return sourceFetchAny16[blendType]; |
2375 | if (qPixelLayouts[format].bpp == QPixelLayout::BPP32) |
2376 | return sourceFetchAny32[blendType]; |
2377 | return sourceFetchGeneric[blendType]; |
2378 | } |
2379 | |
2380 | #if QT_CONFIG(raster_64bit) |
2381 | static const SourceFetchProc64 sourceFetchGeneric64[NBlendTypes] = { |
2382 | fetchUntransformed64, // Untransformed |
2383 | fetchUntransformed64, // Tiled |
2384 | fetchTransformed64<BlendTransformed>, // Transformed |
2385 | fetchTransformed64<BlendTransformedTiled>, // TransformedTiled |
2386 | fetchTransformedBilinear64<BlendTransformedBilinear>, // Bilinear |
2387 | fetchTransformedBilinear64<BlendTransformedBilinearTiled> // BilinearTiled |
2388 | }; |
2389 | |
2390 | static const SourceFetchProc64 sourceFetchRGBA64PM[NBlendTypes] = { |
2391 | fetchUntransformedRGBA64PM, // Untransformed |
2392 | fetchUntransformedRGBA64PM, // Tiled |
2393 | fetchTransformed64<BlendTransformed>, // Transformed |
2394 | fetchTransformed64<BlendTransformedTiled>, // TransformedTiled |
2395 | fetchTransformedBilinear64<BlendTransformedBilinear>, // Bilinear |
2396 | fetchTransformedBilinear64<BlendTransformedBilinearTiled> // BilinearTiled |
2397 | }; |
2398 | |
2399 | static inline SourceFetchProc64 getSourceFetch64(TextureBlendType blendType, QImage::Format format) |
2400 | { |
2401 | if (format == QImage::Format_RGBX64 || format == QImage::Format_RGBA64_Premultiplied) |
2402 | return sourceFetchRGBA64PM[blendType]; |
2403 | return sourceFetchGeneric64[blendType]; |
2404 | } |
2405 | #endif |
2406 | |
2407 | |
2408 | #define FIXPT_BITS 8 |
2409 | #define FIXPT_SIZE (1<<FIXPT_BITS) |
2410 | |
2411 | static uint qt_gradient_pixel_fixed(const QGradientData *data, int fixed_pos) |
2412 | { |
2413 | int ipos = (fixed_pos + (FIXPT_SIZE / 2)) >> FIXPT_BITS; |
2414 | return data->colorTable32[qt_gradient_clamp(data, ipos)]; |
2415 | } |
2416 | |
2417 | #if QT_CONFIG(raster_64bit) |
2418 | static const QRgba64& qt_gradient_pixel64_fixed(const QGradientData *data, int fixed_pos) |
2419 | { |
2420 | int ipos = (fixed_pos + (FIXPT_SIZE / 2)) >> FIXPT_BITS; |
2421 | return data->colorTable64[qt_gradient_clamp(data, ipos)]; |
2422 | } |
2423 | #endif |
2424 | |
2425 | static void QT_FASTCALL getLinearGradientValues(LinearGradientValues *v, const QSpanData *data) |
2426 | { |
2427 | v->dx = data->gradient.linear.end.x - data->gradient.linear.origin.x; |
2428 | v->dy = data->gradient.linear.end.y - data->gradient.linear.origin.y; |
2429 | v->l = v->dx * v->dx + v->dy * v->dy; |
2430 | v->off = 0; |
2431 | if (v->l != 0) { |
2432 | v->dx /= v->l; |
2433 | v->dy /= v->l; |
2434 | v->off = -v->dx * data->gradient.linear.origin.x - v->dy * data->gradient.linear.origin.y; |
2435 | } |
2436 | } |
2437 | |
2438 | class GradientBase32 |
2439 | { |
2440 | public: |
2441 | typedef uint Type; |
2442 | static Type null() { return 0; } |
2443 | static Type fetchSingle(const QGradientData& gradient, qreal v) |
2444 | { |
2445 | return qt_gradient_pixel(&gradient, v); |
2446 | } |
2447 | static Type fetchSingle(const QGradientData& gradient, int v) |
2448 | { |
2449 | return qt_gradient_pixel_fixed(&gradient, v); |
2450 | } |
2451 | static void memfill(Type *buffer, Type fill, int length) |
2452 | { |
2453 | qt_memfill32(buffer, fill, length); |
2454 | } |
2455 | }; |
2456 | |
2457 | #if QT_CONFIG(raster_64bit) |
2458 | class GradientBase64 |
2459 | { |
2460 | public: |
2461 | typedef QRgba64 Type; |
2462 | static Type null() { return QRgba64::fromRgba64(0); } |
2463 | static Type fetchSingle(const QGradientData& gradient, qreal v) |
2464 | { |
2465 | return qt_gradient_pixel64(&gradient, v); |
2466 | } |
2467 | static Type fetchSingle(const QGradientData& gradient, int v) |
2468 | { |
2469 | return qt_gradient_pixel64_fixed(&gradient, v); |
2470 | } |
2471 | static void memfill(Type *buffer, Type fill, int length) |
2472 | { |
2473 | qt_memfill64((quint64*)buffer, fill, length); |
2474 | } |
2475 | }; |
2476 | #endif |
2477 | |
2478 | template<class GradientBase, typename BlendType> |
2479 | static inline const BlendType * QT_FASTCALL qt_fetch_linear_gradient_template( |
2480 | BlendType *buffer, const Operator *op, const QSpanData *data, |
2481 | int y, int x, int length) |
2482 | { |
2483 | const BlendType *b = buffer; |
2484 | qreal t, inc; |
2485 | |
2486 | bool affine = true; |
2487 | qreal rx=0, ry=0; |
2488 | if (op->linear.l == 0) { |
2489 | t = inc = 0; |
2490 | } else { |
2491 | rx = data->m21 * (y + qreal(0.5)) + data->m11 * (x + qreal(0.5)) + data->dx; |
2492 | ry = data->m22 * (y + qreal(0.5)) + data->m12 * (x + qreal(0.5)) + data->dy; |
2493 | t = op->linear.dx*rx + op->linear.dy*ry + op->linear.off; |
2494 | inc = op->linear.dx * data->m11 + op->linear.dy * data->m12; |
2495 | affine = !data->m13 && !data->m23; |
2496 | |
2497 | if (affine) { |
2498 | t *= (GRADIENT_STOPTABLE_SIZE - 1); |
2499 | inc *= (GRADIENT_STOPTABLE_SIZE - 1); |
2500 | } |
2501 | } |
2502 | |
2503 | const BlendType *end = buffer + length; |
2504 | if (affine) { |
2505 | if (inc > qreal(-1e-5) && inc < qreal(1e-5)) { |
2506 | GradientBase::memfill(buffer, GradientBase::fetchSingle(data->gradient, int(t * FIXPT_SIZE)), length); |
2507 | } else { |
2508 | if (t+inc*length < qreal(INT_MAX >> (FIXPT_BITS + 1)) && |
2509 | t+inc*length > qreal(INT_MIN >> (FIXPT_BITS + 1))) { |
2510 | // we can use fixed point math |
2511 | int t_fixed = int(t * FIXPT_SIZE); |
2512 | int inc_fixed = int(inc * FIXPT_SIZE); |
2513 | while (buffer < end) { |
2514 | *buffer = GradientBase::fetchSingle(data->gradient, t_fixed); |
2515 | t_fixed += inc_fixed; |
2516 | ++buffer; |
2517 | } |
2518 | } else { |
2519 | // we have to fall back to float math |
2520 | while (buffer < end) { |
2521 | *buffer = GradientBase::fetchSingle(data->gradient, t/GRADIENT_STOPTABLE_SIZE); |
2522 | t += inc; |
2523 | ++buffer; |
2524 | } |
2525 | } |
2526 | } |
2527 | } else { // fall back to float math here as well |
2528 | qreal rw = data->m23 * (y + qreal(0.5)) + data->m13 * (x + qreal(0.5)) + data->m33; |
2529 | while (buffer < end) { |
2530 | qreal x = rx/rw; |
2531 | qreal y = ry/rw; |
2532 | t = (op->linear.dx*x + op->linear.dy *y) + op->linear.off; |
2533 | |
2534 | *buffer = GradientBase::fetchSingle(data->gradient, t); |
2535 | rx += data->m11; |
2536 | ry += data->m12; |
2537 | rw += data->m13; |
2538 | if (!rw) { |
2539 | rw += data->m13; |
2540 | } |
2541 | ++buffer; |
2542 | } |
2543 | } |
2544 | |
2545 | return b; |
2546 | } |
2547 | |
2548 | static const uint * QT_FASTCALL qt_fetch_linear_gradient(uint *buffer, const Operator *op, const QSpanData *data, |
2549 | int y, int x, int length) |
2550 | { |
2551 | return qt_fetch_linear_gradient_template<GradientBase32, uint>(buffer, op, data, y, x, length); |
2552 | } |
2553 | |
2554 | #if QT_CONFIG(raster_64bit) |
2555 | static const QRgba64 * QT_FASTCALL qt_fetch_linear_gradient_rgb64(QRgba64 *buffer, const Operator *op, const QSpanData *data, |
2556 | int y, int x, int length) |
2557 | { |
2558 | return qt_fetch_linear_gradient_template<GradientBase64, QRgba64>(buffer, op, data, y, x, length); |
2559 | } |
2560 | #endif |
2561 | |
2562 | static void QT_FASTCALL getRadialGradientValues(RadialGradientValues *v, const QSpanData *data) |
2563 | { |
2564 | v->dx = data->gradient.radial.center.x - data->gradient.radial.focal.x; |
2565 | v->dy = data->gradient.radial.center.y - data->gradient.radial.focal.y; |
2566 | |
2567 | v->dr = data->gradient.radial.center.radius - data->gradient.radial.focal.radius; |
2568 | v->sqrfr = data->gradient.radial.focal.radius * data->gradient.radial.focal.radius; |
2569 | |
2570 | v->a = v->dr * v->dr - v->dx*v->dx - v->dy*v->dy; |
2571 | v->inv2a = 1 / (2 * v->a); |
2572 | |
2573 | v->extended = !qFuzzyIsNull(data->gradient.radial.focal.radius) || v->a <= 0; |
2574 | } |
2575 | |
2576 | template <class GradientBase> |
2577 | class RadialFetchPlain : public GradientBase |
2578 | { |
2579 | public: |
2580 | typedef typename GradientBase::Type BlendType; |
2581 | static void fetch(BlendType *buffer, BlendType *end, |
2582 | const Operator *op, const QSpanData *data, qreal det, |
2583 | qreal delta_det, qreal delta_delta_det, qreal b, qreal delta_b) |
2584 | { |
2585 | if (op->radial.extended) { |
2586 | while (buffer < end) { |
2587 | BlendType result = GradientBase::null(); |
2588 | if (det >= 0) { |
2589 | qreal w = qSqrt(det) - b; |
2590 | if (data->gradient.radial.focal.radius + op->radial.dr * w >= 0) |
2591 | result = GradientBase::fetchSingle(data->gradient, w); |
2592 | } |
2593 | |
2594 | *buffer = result; |
2595 | |
2596 | det += delta_det; |
2597 | delta_det += delta_delta_det; |
2598 | b += delta_b; |
2599 | |
2600 | ++buffer; |
2601 | } |
2602 | } else { |
2603 | while (buffer < end) { |
2604 | *buffer++ = GradientBase::fetchSingle(data->gradient, qSqrt(det) - b); |
2605 | |
2606 | det += delta_det; |
2607 | delta_det += delta_delta_det; |
2608 | b += delta_b; |
2609 | } |
2610 | } |
2611 | } |
2612 | }; |
2613 | |
2614 | const uint * QT_FASTCALL qt_fetch_radial_gradient_plain(uint *buffer, const Operator *op, const QSpanData *data, |
2615 | int y, int x, int length) |
2616 | { |
2617 | return qt_fetch_radial_gradient_template<RadialFetchPlain<GradientBase32>, uint>(buffer, op, data, y, x, length); |
2618 | } |
2619 | |
2620 | static SourceFetchProc qt_fetch_radial_gradient = qt_fetch_radial_gradient_plain; |
2621 | |
2622 | #if QT_CONFIG(raster_64bit) |
2623 | const QRgba64 * QT_FASTCALL qt_fetch_radial_gradient_rgb64(QRgba64 *buffer, const Operator *op, const QSpanData *data, |
2624 | int y, int x, int length) |
2625 | { |
2626 | return qt_fetch_radial_gradient_template<RadialFetchPlain<GradientBase64>, QRgba64>(buffer, op, data, y, x, length); |
2627 | } |
2628 | #endif |
2629 | |
2630 | template <class GradientBase, typename BlendType> |
2631 | static inline const BlendType * QT_FASTCALL qt_fetch_conical_gradient_template( |
2632 | BlendType *buffer, const QSpanData *data, |
2633 | int y, int x, int length) |
2634 | { |
2635 | const BlendType *b = buffer; |
2636 | qreal rx = data->m21 * (y + qreal(0.5)) |
2637 | + data->dx + data->m11 * (x + qreal(0.5)); |
2638 | qreal ry = data->m22 * (y + qreal(0.5)) |
2639 | + data->dy + data->m12 * (x + qreal(0.5)); |
2640 | bool affine = !data->m13 && !data->m23; |
2641 | |
2642 | const qreal inv2pi = M_1_PI / 2.0; |
2643 | |
2644 | const BlendType *end = buffer + length; |
2645 | if (affine) { |
2646 | rx -= data->gradient.conical.center.x; |
2647 | ry -= data->gradient.conical.center.y; |
2648 | while (buffer < end) { |
2649 | qreal angle = qAtan2(ry, rx) + data->gradient.conical.angle; |
2650 | |
2651 | *buffer = GradientBase::fetchSingle(data->gradient, 1 - angle * inv2pi); |
2652 | |
2653 | rx += data->m11; |
2654 | ry += data->m12; |
2655 | ++buffer; |
2656 | } |
2657 | } else { |
2658 | qreal rw = data->m23 * (y + qreal(0.5)) |
2659 | + data->m33 + data->m13 * (x + qreal(0.5)); |
2660 | if (!rw) |
2661 | rw = 1; |
2662 | while (buffer < end) { |
2663 | qreal angle = qAtan2(ry/rw - data->gradient.conical.center.x, |
2664 | rx/rw - data->gradient.conical.center.y) |
2665 | + data->gradient.conical.angle; |
2666 | |
2667 | *buffer = GradientBase::fetchSingle(data->gradient, 1 - angle * inv2pi); |
2668 | |
2669 | rx += data->m11; |
2670 | ry += data->m12; |
2671 | rw += data->m13; |
2672 | if (!rw) { |
2673 | rw += data->m13; |
2674 | } |
2675 | ++buffer; |
2676 | } |
2677 | } |
2678 | return b; |
2679 | } |
2680 | |
2681 | static const uint * QT_FASTCALL qt_fetch_conical_gradient(uint *buffer, const Operator *, const QSpanData *data, |
2682 | int y, int x, int length) |
2683 | { |
2684 | return qt_fetch_conical_gradient_template<GradientBase32, uint>(buffer, data, y, x, length); |
2685 | } |
2686 | |
2687 | #if QT_CONFIG(raster_64bit) |
2688 | static const QRgba64 * QT_FASTCALL qt_fetch_conical_gradient_rgb64(QRgba64 *buffer, const Operator *, const QSpanData *data, |
2689 | int y, int x, int length) |
2690 | { |
2691 | return qt_fetch_conical_gradient_template<GradientBase64, QRgba64>(buffer, data, y, x, length); |
2692 | } |
2693 | #endif |
2694 | |
2695 | extern CompositionFunctionSolid qt_functionForModeSolid_C[]; |
2696 | extern CompositionFunctionSolid64 qt_functionForModeSolid64_C[]; |
2697 | |
2698 | static const CompositionFunctionSolid *functionForModeSolid = qt_functionForModeSolid_C; |
2699 | #if QT_CONFIG(raster_64bit) |
2700 | static const CompositionFunctionSolid64 *functionForModeSolid64 = qt_functionForModeSolid64_C; |
2701 | #endif |
2702 | |
2703 | extern CompositionFunction qt_functionForMode_C[]; |
2704 | extern CompositionFunction64 qt_functionForMode64_C[]; |
2705 | |
2706 | static const CompositionFunction *functionForMode = qt_functionForMode_C; |
2707 | #if QT_CONFIG(raster_64bit) |
2708 | static const CompositionFunction64 *functionForMode64 = qt_functionForMode64_C; |
2709 | #endif |
2710 | |
2711 | static TextureBlendType getBlendType(const QSpanData *data) |
2712 | { |
2713 | TextureBlendType ft; |
2714 | if (data->txop <= QTransform::TxTranslate) |
2715 | if (data->texture.type == QTextureData::Tiled) |
2716 | ft = BlendTiled; |
2717 | else |
2718 | ft = BlendUntransformed; |
2719 | else if (data->bilinear) |
2720 | if (data->texture.type == QTextureData::Tiled) |
2721 | ft = BlendTransformedBilinearTiled; |
2722 | else |
2723 | ft = BlendTransformedBilinear; |
2724 | else |
2725 | if (data->texture.type == QTextureData::Tiled) |
2726 | ft = BlendTransformedTiled; |
2727 | else |
2728 | ft = BlendTransformed; |
2729 | return ft; |
2730 | } |
2731 | |
2732 | static inline Operator getOperator(const QSpanData *data, const QSpan *spans, int spanCount) |
2733 | { |
2734 | Operator op; |
2735 | bool solidSource = false; |
2736 | |
2737 | switch(data->type) { |
2738 | case QSpanData::Solid: |
2739 | solidSource = data->solidColor.isOpaque(); |
2740 | op.srcFetch = nullptr; |
2741 | #if QT_CONFIG(raster_64bit) |
2742 | op.srcFetch64 = nullptr; |
2743 | #endif |
2744 | break; |
2745 | case QSpanData::LinearGradient: |
2746 | solidSource = !data->gradient.alphaColor; |
2747 | getLinearGradientValues(&op.linear, data); |
2748 | op.srcFetch = qt_fetch_linear_gradient; |
2749 | #if QT_CONFIG(raster_64bit) |
2750 | op.srcFetch64 = qt_fetch_linear_gradient_rgb64; |
2751 | #endif |
2752 | break; |
2753 | case QSpanData::RadialGradient: |
2754 | solidSource = !data->gradient.alphaColor; |
2755 | getRadialGradientValues(&op.radial, data); |
2756 | op.srcFetch = qt_fetch_radial_gradient; |
2757 | #if QT_CONFIG(raster_64bit) |
2758 | op.srcFetch64 = qt_fetch_radial_gradient_rgb64; |
2759 | #endif |
2760 | break; |
2761 | case QSpanData::ConicalGradient: |
2762 | solidSource = !data->gradient.alphaColor; |
2763 | op.srcFetch = qt_fetch_conical_gradient; |
2764 | #if QT_CONFIG(raster_64bit) |
2765 | op.srcFetch64 = qt_fetch_conical_gradient_rgb64; |
2766 | #endif |
2767 | break; |
2768 | case QSpanData::Texture: |
2769 | solidSource = !data->texture.hasAlpha; |
2770 | op.srcFetch = getSourceFetch(getBlendType(data), data->texture.format); |
2771 | #if QT_CONFIG(raster_64bit) |
2772 | op.srcFetch64 = getSourceFetch64(getBlendType(data), data->texture.format);; |
2773 | #endif |
2774 | break; |
2775 | default: |
2776 | Q_UNREACHABLE(); |
2777 | break; |
2778 | } |
2779 | #if !QT_CONFIG(raster_64bit) |
2780 | op.srcFetch64 = 0; |
2781 | #endif |
2782 | |
2783 | op.mode = data->rasterBuffer->compositionMode; |
2784 | if (op.mode == QPainter::CompositionMode_SourceOver && solidSource) |
2785 | op.mode = QPainter::CompositionMode_Source; |
2786 | |
2787 | op.destFetch = destFetchProc[data->rasterBuffer->format]; |
2788 | #if QT_CONFIG(raster_64bit) |
2789 | op.destFetch64 = destFetchProc64[data->rasterBuffer->format]; |
2790 | #else |
2791 | op.destFetch64 = 0; |
2792 | #endif |
2793 | if (op.mode == QPainter::CompositionMode_Source && |
2794 | (data->type != QSpanData::Texture || data->texture.const_alpha == 256)) { |
2795 | const QSpan *lastSpan = spans + spanCount; |
2796 | bool alphaSpans = false; |
2797 | while (spans < lastSpan) { |
2798 | if (spans->coverage != 255) { |
2799 | alphaSpans = true; |
2800 | break; |
2801 | } |
2802 | ++spans; |
2803 | } |
2804 | if (!alphaSpans && spanCount > 0) { |
2805 | // If all spans are opaque we do not need to fetch dest. |
2806 | // But don't clear passthrough destFetch as they are just as fast and save destStore. |
2807 | if (op.destFetch != destFetchARGB32P) |
2808 | op.destFetch = destFetchUndefined; |
2809 | #if QT_CONFIG(raster_64bit) |
2810 | if (op.destFetch64 != destFetchRGB64) |
2811 | op.destFetch64 = destFetch64Undefined; |
2812 | #endif |
2813 | } |
2814 | } |
2815 | |
2816 | op.destStore = destStoreProc[data->rasterBuffer->format]; |
2817 | op.funcSolid = functionForModeSolid[op.mode]; |
2818 | op.func = functionForMode[op.mode]; |
2819 | #if QT_CONFIG(raster_64bit) |
2820 | op.destStore64 = destStoreProc64[data->rasterBuffer->format]; |
2821 | op.funcSolid64 = functionForModeSolid64[op.mode]; |
2822 | op.func64 = functionForMode64[op.mode]; |
2823 | #else |
2824 | op.destStore64 = 0; |
2825 | op.funcSolid64 = 0; |
2826 | op.func64 = 0; |
2827 | #endif |
2828 | |
2829 | return op; |
2830 | } |
2831 | |
2832 | static void spanfill_from_first(QRasterBuffer *rasterBuffer, QPixelLayout::BPP bpp, int x, int y, int length) |
2833 | { |
2834 | switch (bpp) { |
2835 | case QPixelLayout::BPP64: { |
2836 | quint64 *dest = reinterpret_cast<quint64 *>(rasterBuffer->scanLine(y)) + x; |
2837 | qt_memfill_template(dest + 1, dest[0], length - 1); |
2838 | break; |
2839 | } |
2840 | case QPixelLayout::BPP32: { |
2841 | quint32 *dest = reinterpret_cast<quint32 *>(rasterBuffer->scanLine(y)) + x; |
2842 | qt_memfill_template(dest + 1, dest[0], length - 1); |
2843 | break; |
2844 | } |
2845 | case QPixelLayout::BPP24: { |
2846 | quint24 *dest = reinterpret_cast<quint24 *>(rasterBuffer->scanLine(y)) + x; |
2847 | qt_memfill_template(dest + 1, dest[0], length - 1); |
2848 | break; |
2849 | } |
2850 | case QPixelLayout::BPP16: { |
2851 | quint16 *dest = reinterpret_cast<quint16 *>(rasterBuffer->scanLine(y)) + x; |
2852 | qt_memfill_template(dest + 1, dest[0], length - 1); |
2853 | break; |
2854 | } |
2855 | case QPixelLayout::BPP8: { |
2856 | uchar *dest = rasterBuffer->scanLine(y) + x; |
2857 | memset(dest + 1, dest[0], length - 1); |
2858 | break; |
2859 | } |
2860 | default: |
2861 | Q_UNREACHABLE(); |
2862 | } |
2863 | } |
2864 | |
2865 | |
2866 | // -------------------- blend methods --------------------- |
2867 | |
2868 | static void blend_color_generic(int count, const QSpan *spans, void *userData) |
2869 | { |
2870 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
2871 | uint buffer[BufferSize]; |
2872 | Operator op = getOperator(data, nullptr, 0); |
2873 | const uint color = data->solidColor.toArgb32(); |
2874 | const bool solidFill = op.mode == QPainter::CompositionMode_Source; |
2875 | const QPixelLayout::BPP bpp = qPixelLayouts[data->rasterBuffer->format].bpp; |
2876 | |
2877 | while (count--) { |
2878 | int x = spans->x; |
2879 | int length = spans->len; |
2880 | if (solidFill && bpp >= QPixelLayout::BPP8 && spans->coverage == 255 && length) { |
2881 | // If dest doesn't matter we don't need to bother with blending or converting all the identical pixels |
2882 | op.destStore(data->rasterBuffer, x, spans->y, &color, 1); |
2883 | spanfill_from_first(data->rasterBuffer, bpp, x, spans->y, length); |
2884 | length = 0; |
2885 | } |
2886 | |
2887 | while (length) { |
2888 | int l = qMin(BufferSize, length); |
2889 | uint *dest = op.destFetch(buffer, data->rasterBuffer, x, spans->y, l); |
2890 | op.funcSolid(dest, l, color, spans->coverage); |
2891 | if (op.destStore) |
2892 | op.destStore(data->rasterBuffer, x, spans->y, dest, l); |
2893 | length -= l; |
2894 | x += l; |
2895 | } |
2896 | ++spans; |
2897 | } |
2898 | } |
2899 | |
2900 | static void blend_color_argb(int count, const QSpan *spans, void *userData) |
2901 | { |
2902 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
2903 | |
2904 | const Operator op = getOperator(data, nullptr, 0); |
2905 | const uint color = data->solidColor.toArgb32(); |
2906 | |
2907 | if (op.mode == QPainter::CompositionMode_Source) { |
2908 | // inline for performance |
2909 | while (count--) { |
2910 | uint *target = ((uint *)data->rasterBuffer->scanLine(spans->y)) + spans->x; |
2911 | if (spans->coverage == 255) { |
2912 | qt_memfill(target, color, spans->len); |
2913 | #ifdef __SSE2__ |
2914 | } else if (spans->len > 16) { |
2915 | op.funcSolid(target, spans->len, color, spans->coverage); |
2916 | #endif |
2917 | } else { |
2918 | uint c = BYTE_MUL(color, spans->coverage); |
2919 | int ialpha = 255 - spans->coverage; |
2920 | for (int i = 0; i < spans->len; ++i) |
2921 | target[i] = c + BYTE_MUL(target[i], ialpha); |
2922 | } |
2923 | ++spans; |
2924 | } |
2925 | return; |
2926 | } |
2927 | |
2928 | while (count--) { |
2929 | uint *target = ((uint *)data->rasterBuffer->scanLine(spans->y)) + spans->x; |
2930 | op.funcSolid(target, spans->len, color, spans->coverage); |
2931 | ++spans; |
2932 | } |
2933 | } |
2934 | |
2935 | void blend_color_generic_rgb64(int count, const QSpan *spans, void *userData) |
2936 | { |
2937 | #if QT_CONFIG(raster_64bit) |
2938 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
2939 | Operator op = getOperator(data, nullptr, 0); |
2940 | if (!op.funcSolid64) { |
2941 | qCDebug(lcQtGuiDrawHelper, "blend_color_generic_rgb64: unsupported 64bit blend attempted, falling back to 32-bit" ); |
2942 | return blend_color_generic(count, spans, userData); |
2943 | } |
2944 | |
2945 | alignas(8) QRgba64 buffer[BufferSize]; |
2946 | const QRgba64 color = data->solidColor; |
2947 | const bool solidFill = op.mode == QPainter::CompositionMode_Source; |
2948 | const QPixelLayout::BPP bpp = qPixelLayouts[data->rasterBuffer->format].bpp; |
2949 | |
2950 | while (count--) { |
2951 | int x = spans->x; |
2952 | int length = spans->len; |
2953 | if (solidFill && bpp >= QPixelLayout::BPP8 && spans->coverage == 255 && length && op.destStore64) { |
2954 | // If dest doesn't matter we don't need to bother with blending or converting all the identical pixels |
2955 | op.destStore64(data->rasterBuffer, x, spans->y, &color, 1); |
2956 | spanfill_from_first(data->rasterBuffer, bpp, x, spans->y, length); |
2957 | length = 0; |
2958 | } |
2959 | |
2960 | while (length) { |
2961 | int l = qMin(BufferSize, length); |
2962 | QRgba64 *dest = op.destFetch64(buffer, data->rasterBuffer, x, spans->y, l); |
2963 | op.funcSolid64(dest, l, color, spans->coverage); |
2964 | if (op.destStore64) |
2965 | op.destStore64(data->rasterBuffer, x, spans->y, dest, l); |
2966 | length -= l; |
2967 | x += l; |
2968 | } |
2969 | ++spans; |
2970 | } |
2971 | #else |
2972 | blend_color_generic(count, spans, userData); |
2973 | #endif |
2974 | } |
2975 | |
2976 | static void blend_color_rgb16(int count, const QSpan *spans, void *userData) |
2977 | { |
2978 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
2979 | |
2980 | /* |
2981 | We duplicate a little logic from getOperator() and calculate the |
2982 | composition mode directly. This allows blend_color_rgb16 to be used |
2983 | from qt_gradient_quint16 with minimal overhead. |
2984 | */ |
2985 | QPainter::CompositionMode mode = data->rasterBuffer->compositionMode; |
2986 | if (mode == QPainter::CompositionMode_SourceOver && data->solidColor.isOpaque()) |
2987 | mode = QPainter::CompositionMode_Source; |
2988 | |
2989 | if (mode == QPainter::CompositionMode_Source) { |
2990 | // inline for performance |
2991 | ushort c = data->solidColor.toRgb16(); |
2992 | for (; count--; spans++) { |
2993 | if (!spans->len) |
2994 | continue; |
2995 | ushort *target = ((ushort *)data->rasterBuffer->scanLine(spans->y)) + spans->x; |
2996 | if (spans->coverage == 255) { |
2997 | qt_memfill(target, c, spans->len); |
2998 | } else { |
2999 | ushort color = BYTE_MUL_RGB16(c, spans->coverage); |
3000 | int ialpha = 255 - spans->coverage; |
3001 | const ushort *end = target + spans->len; |
3002 | while (target < end) { |
3003 | *target = color + BYTE_MUL_RGB16(*target, ialpha); |
3004 | ++target; |
3005 | } |
3006 | } |
3007 | } |
3008 | return; |
3009 | } |
3010 | |
3011 | if (mode == QPainter::CompositionMode_SourceOver) { |
3012 | for (; count--; spans++) { |
3013 | if (!spans->len) |
3014 | continue; |
3015 | uint color = BYTE_MUL(data->solidColor.toArgb32(), spans->coverage); |
3016 | int ialpha = qAlpha(~color); |
3017 | ushort c = qConvertRgb32To16(color); |
3018 | ushort *target = ((ushort *)data->rasterBuffer->scanLine(spans->y)) + spans->x; |
3019 | int len = spans->len; |
3020 | bool pre = (((quintptr)target) & 0x3) != 0; |
3021 | bool post = false; |
3022 | if (pre) { |
3023 | // skip to word boundary |
3024 | *target = c + BYTE_MUL_RGB16(*target, ialpha); |
3025 | ++target; |
3026 | --len; |
3027 | } |
3028 | if (len & 0x1) { |
3029 | post = true; |
3030 | --len; |
3031 | } |
3032 | uint *target32 = (uint*)target; |
3033 | uint c32 = c | (c<<16); |
3034 | len >>= 1; |
3035 | uint salpha = (ialpha+1) >> 3; // calculate here rather than in loop |
3036 | while (len--) { |
3037 | // blend full words |
3038 | *target32 = c32 + BYTE_MUL_RGB16_32(*target32, salpha); |
3039 | ++target32; |
3040 | target += 2; |
3041 | } |
3042 | if (post) { |
3043 | // one last pixel beyond a full word |
3044 | *target = c + BYTE_MUL_RGB16(*target, ialpha); |
3045 | } |
3046 | } |
3047 | return; |
3048 | } |
3049 | |
3050 | blend_color_generic(count, spans, userData); |
3051 | } |
3052 | |
3053 | template <typename T> |
3054 | void handleSpans(int count, const QSpan *spans, const QSpanData *data, T &handler) |
3055 | { |
3056 | uint const_alpha = 256; |
3057 | if (data->type == QSpanData::Texture) |
3058 | const_alpha = data->texture.const_alpha; |
3059 | |
3060 | int coverage = 0; |
3061 | while (count) { |
3062 | if (!spans->len) { |
3063 | ++spans; |
3064 | --count; |
3065 | continue; |
3066 | } |
3067 | int x = spans->x; |
3068 | const int y = spans->y; |
3069 | int right = x + spans->len; |
3070 | |
3071 | // compute length of adjacent spans |
3072 | for (int i = 1; i < count && spans[i].y == y && spans[i].x == right; ++i) |
3073 | right += spans[i].len; |
3074 | int length = right - x; |
3075 | |
3076 | while (length) { |
3077 | int l = qMin(BufferSize, length); |
3078 | length -= l; |
3079 | |
3080 | int process_length = l; |
3081 | int process_x = x; |
3082 | |
3083 | const typename T::BlendType *src = handler.fetch(process_x, y, process_length); |
3084 | int offset = 0; |
3085 | while (l > 0) { |
3086 | if (x == spans->x) // new span? |
3087 | coverage = (spans->coverage * const_alpha) >> 8; |
3088 | |
3089 | int right = spans->x + spans->len; |
3090 | int len = qMin(l, right - x); |
3091 | |
3092 | handler.process(x, y, len, coverage, src, offset); |
3093 | |
3094 | l -= len; |
3095 | x += len; |
3096 | offset += len; |
3097 | |
3098 | if (x == right) { // done with current span? |
3099 | ++spans; |
3100 | --count; |
3101 | } |
3102 | } |
3103 | handler.store(process_x, y, process_length); |
3104 | } |
3105 | } |
3106 | } |
3107 | |
3108 | template<typename T> |
3109 | struct QBlendBase |
3110 | { |
3111 | typedef T BlendType; |
3112 | QBlendBase(QSpanData *d, const Operator &o) |
3113 | : data(d) |
3114 | , op(o) |
3115 | , dest(nullptr) |
3116 | { |
3117 | } |
3118 | |
3119 | QSpanData *data; |
3120 | Operator op; |
3121 | |
3122 | BlendType *dest; |
3123 | |
3124 | alignas(8) BlendType buffer[BufferSize]; |
3125 | alignas(8) BlendType src_buffer[BufferSize]; |
3126 | }; |
3127 | |
3128 | class BlendSrcGeneric : public QBlendBase<uint> |
3129 | { |
3130 | public: |
3131 | BlendSrcGeneric(QSpanData *d, const Operator &o) |
3132 | : QBlendBase<uint>(d, o) |
3133 | { |
3134 | } |
3135 | |
3136 | const uint *fetch(int x, int y, int len) |
3137 | { |
3138 | dest = op.destFetch(buffer, data->rasterBuffer, x, y, len); |
3139 | return op.srcFetch(src_buffer, &op, data, y, x, len); |
3140 | } |
3141 | |
3142 | void process(int, int, int len, int coverage, const uint *src, int offset) |
3143 | { |
3144 | op.func(dest + offset, src + offset, len, coverage); |
3145 | } |
3146 | |
3147 | void store(int x, int y, int len) |
3148 | { |
3149 | if (op.destStore) |
3150 | op.destStore(data->rasterBuffer, x, y, dest, len); |
3151 | } |
3152 | }; |
3153 | |
3154 | #if QT_CONFIG(raster_64bit) |
3155 | class BlendSrcGenericRGB64 : public QBlendBase<QRgba64> |
3156 | { |
3157 | public: |
3158 | BlendSrcGenericRGB64(QSpanData *d, const Operator &o) |
3159 | : QBlendBase<QRgba64>(d, o) |
3160 | { |
3161 | } |
3162 | |
3163 | bool isSupported() const |
3164 | { |
3165 | return op.func64 && op.destFetch64; |
3166 | } |
3167 | |
3168 | const QRgba64 *fetch(int x, int y, int len) |
3169 | { |
3170 | dest = op.destFetch64(buffer, data->rasterBuffer, x, y, len); |
3171 | return op.srcFetch64(src_buffer, &op, data, y, x, len); |
3172 | } |
3173 | |
3174 | void process(int, int, int len, int coverage, const QRgba64 *src, int offset) |
3175 | { |
3176 | op.func64(dest + offset, src + offset, len, coverage); |
3177 | } |
3178 | |
3179 | void store(int x, int y, int len) |
3180 | { |
3181 | if (op.destStore64) |
3182 | op.destStore64(data->rasterBuffer, x, y, dest, len); |
3183 | } |
3184 | }; |
3185 | #endif |
3186 | |
3187 | static void blend_src_generic(int count, const QSpan *spans, void *userData) |
3188 | { |
3189 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3190 | BlendSrcGeneric blend(data, getOperator(data, spans, count)); |
3191 | handleSpans(count, spans, data, blend); |
3192 | } |
3193 | |
3194 | #if QT_CONFIG(raster_64bit) |
3195 | static void blend_src_generic_rgb64(int count, const QSpan *spans, void *userData) |
3196 | { |
3197 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3198 | Operator op = getOperator(data, spans, count); |
3199 | BlendSrcGenericRGB64 blend64(data, op); |
3200 | if (blend64.isSupported()) |
3201 | handleSpans(count, spans, data, blend64); |
3202 | else { |
3203 | qCDebug(lcQtGuiDrawHelper, "blend_src_generic_rgb64: unsupported 64-bit blend attempted, falling back to 32-bit" ); |
3204 | BlendSrcGeneric blend32(data, op); |
3205 | handleSpans(count, spans, data, blend32); |
3206 | } |
3207 | } |
3208 | #endif |
3209 | |
3210 | static void blend_untransformed_generic(int count, const QSpan *spans, void *userData) |
3211 | { |
3212 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3213 | |
3214 | uint buffer[BufferSize]; |
3215 | uint src_buffer[BufferSize]; |
3216 | Operator op = getOperator(data, spans, count); |
3217 | |
3218 | const int image_width = data->texture.width; |
3219 | const int image_height = data->texture.height; |
3220 | int xoff = -qRound(-data->dx); |
3221 | int yoff = -qRound(-data->dy); |
3222 | |
3223 | for (; count--; spans++) { |
3224 | if (!spans->len) |
3225 | continue; |
3226 | int x = spans->x; |
3227 | int length = spans->len; |
3228 | int sx = xoff + x; |
3229 | int sy = yoff + spans->y; |
3230 | if (sy >= 0 && sy < image_height && sx < image_width) { |
3231 | if (sx < 0) { |
3232 | x -= sx; |
3233 | length += sx; |
3234 | sx = 0; |
3235 | } |
3236 | if (sx + length > image_width) |
3237 | length = image_width - sx; |
3238 | if (length > 0) { |
3239 | const int coverage = (spans->coverage * data->texture.const_alpha) >> 8; |
3240 | while (length) { |
3241 | int l = qMin(BufferSize, length); |
3242 | const uint *src = op.srcFetch(src_buffer, &op, data, sy, sx, l); |
3243 | uint *dest = op.destFetch(buffer, data->rasterBuffer, x, spans->y, l); |
3244 | op.func(dest, src, l, coverage); |
3245 | if (op.destStore) |
3246 | op.destStore(data->rasterBuffer, x, spans->y, dest, l); |
3247 | x += l; |
3248 | sx += l; |
3249 | length -= l; |
3250 | } |
3251 | } |
3252 | } |
3253 | } |
3254 | } |
3255 | |
3256 | #if QT_CONFIG(raster_64bit) |
3257 | static void blend_untransformed_generic_rgb64(int count, const QSpan *spans, void *userData) |
3258 | { |
3259 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3260 | |
3261 | Operator op = getOperator(data, spans, count); |
3262 | if (!op.func64) { |
3263 | qCDebug(lcQtGuiDrawHelper, "blend_untransformed_generic_rgb64: unsupported 64-bit blend attempted, falling back to 32-bit" ); |
3264 | return blend_untransformed_generic(count, spans, userData); |
3265 | } |
3266 | alignas(8) QRgba64 buffer[BufferSize]; |
3267 | alignas(8) QRgba64 src_buffer[BufferSize]; |
3268 | |
3269 | const int image_width = data->texture.width; |
3270 | const int image_height = data->texture.height; |
3271 | int xoff = -qRound(-data->dx); |
3272 | int yoff = -qRound(-data->dy); |
3273 | |
3274 | for (; count--; spans++) { |
3275 | if (!spans->len) |
3276 | continue; |
3277 | int x = spans->x; |
3278 | int length = spans->len; |
3279 | int sx = xoff + x; |
3280 | int sy = yoff + spans->y; |
3281 | if (sy >= 0 && sy < image_height && sx < image_width) { |
3282 | if (sx < 0) { |
3283 | x -= sx; |
3284 | length += sx; |
3285 | sx = 0; |
3286 | } |
3287 | if (sx + length > image_width) |
3288 | length = image_width - sx; |
3289 | if (length > 0) { |
3290 | const int coverage = (spans->coverage * data->texture.const_alpha) >> 8; |
3291 | while (length) { |
3292 | int l = qMin(BufferSize, length); |
3293 | const QRgba64 *src = op.srcFetch64(src_buffer, &op, data, sy, sx, l); |
3294 | QRgba64 *dest = op.destFetch64(buffer, data->rasterBuffer, x, spans->y, l); |
3295 | op.func64(dest, src, l, coverage); |
3296 | if (op.destStore64) |
3297 | op.destStore64(data->rasterBuffer, x, spans->y, dest, l); |
3298 | x += l; |
3299 | sx += l; |
3300 | length -= l; |
3301 | } |
3302 | } |
3303 | } |
3304 | } |
3305 | } |
3306 | #endif |
3307 | |
3308 | static void blend_untransformed_argb(int count, const QSpan *spans, void *userData) |
3309 | { |
3310 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3311 | if (data->texture.format != QImage::Format_ARGB32_Premultiplied |
3312 | && data->texture.format != QImage::Format_RGB32) { |
3313 | blend_untransformed_generic(count, spans, userData); |
3314 | return; |
3315 | } |
3316 | |
3317 | Operator op = getOperator(data, spans, count); |
3318 | |
3319 | const int image_width = data->texture.width; |
3320 | const int image_height = data->texture.height; |
3321 | int xoff = -qRound(-data->dx); |
3322 | int yoff = -qRound(-data->dy); |
3323 | |
3324 | for (; count--; spans++) { |
3325 | if (!spans->len) |
3326 | continue; |
3327 | int x = spans->x; |
3328 | int length = spans->len; |
3329 | int sx = xoff + x; |
3330 | int sy = yoff + spans->y; |
3331 | if (sy >= 0 && sy < image_height && sx < image_width) { |
3332 | if (sx < 0) { |
3333 | x -= sx; |
3334 | length += sx; |
3335 | sx = 0; |
3336 | } |
3337 | if (sx + length > image_width) |
3338 | length = image_width - sx; |
3339 | if (length > 0) { |
3340 | const int coverage = (spans->coverage * data->texture.const_alpha) >> 8; |
3341 | const uint *src = (const uint *)data->texture.scanLine(sy) + sx; |
3342 | uint *dest = ((uint *)data->rasterBuffer->scanLine(spans->y)) + x; |
3343 | op.func(dest, src, length, coverage); |
3344 | } |
3345 | } |
3346 | } |
3347 | } |
3348 | |
3349 | static inline quint16 interpolate_pixel_rgb16_255(quint16 x, quint8 a, |
3350 | quint16 y, quint8 b) |
3351 | { |
3352 | quint16 t = ((((x & 0x07e0) * a) + ((y & 0x07e0) * b)) >> 5) & 0x07e0; |
3353 | t |= ((((x & 0xf81f) * a) + ((y & 0xf81f) * b)) >> 5) & 0xf81f; |
3354 | |
3355 | return t; |
3356 | } |
3357 | |
3358 | static inline quint32 interpolate_pixel_rgb16x2_255(quint32 x, quint8 a, |
3359 | quint32 y, quint8 b) |
3360 | { |
3361 | uint t; |
3362 | t = ((((x & 0xf81f07e0) >> 5) * a) + (((y & 0xf81f07e0) >> 5) * b)) & 0xf81f07e0; |
3363 | t |= ((((x & 0x07e0f81f) * a) + ((y & 0x07e0f81f) * b)) >> 5) & 0x07e0f81f; |
3364 | return t; |
3365 | } |
3366 | |
3367 | static inline void blend_sourceOver_rgb16_rgb16(quint16 *Q_DECL_RESTRICT dest, |
3368 | const quint16 *Q_DECL_RESTRICT src, |
3369 | int length, |
3370 | const quint8 alpha, |
3371 | const quint8 ialpha) |
3372 | { |
3373 | const int dstAlign = ((quintptr)dest) & 0x3; |
3374 | if (dstAlign) { |
3375 | *dest = interpolate_pixel_rgb16_255(*src, alpha, *dest, ialpha); |
3376 | ++dest; |
3377 | ++src; |
3378 | --length; |
3379 | } |
3380 | const int srcAlign = ((quintptr)src) & 0x3; |
3381 | int length32 = length >> 1; |
3382 | if (length32 && srcAlign == 0) { |
3383 | while (length32--) { |
3384 | const quint32 *src32 = reinterpret_cast<const quint32*>(src); |
3385 | quint32 *dest32 = reinterpret_cast<quint32*>(dest); |
3386 | *dest32 = interpolate_pixel_rgb16x2_255(*src32, alpha, |
3387 | *dest32, ialpha); |
3388 | dest += 2; |
3389 | src += 2; |
3390 | } |
3391 | length &= 0x1; |
3392 | } |
3393 | while (length--) { |
3394 | *dest = interpolate_pixel_rgb16_255(*src, alpha, *dest, ialpha); |
3395 | ++dest; |
3396 | ++src; |
3397 | } |
3398 | } |
3399 | |
3400 | static void blend_untransformed_rgb565(int count, const QSpan *spans, void *userData) |
3401 | { |
3402 | QSpanData *data = reinterpret_cast<QSpanData*>(userData); |
3403 | QPainter::CompositionMode mode = data->rasterBuffer->compositionMode; |
3404 | |
3405 | if (data->texture.format != QImage::Format_RGB16 |
3406 | || (mode != QPainter::CompositionMode_SourceOver |
3407 | && mode != QPainter::CompositionMode_Source)) |
3408 | { |
3409 | blend_untransformed_generic(count, spans, userData); |
3410 | return; |
3411 | } |
3412 | |
3413 | const int image_width = data->texture.width; |
3414 | const int image_height = data->texture.height; |
3415 | int xoff = -qRound(-data->dx); |
3416 | int yoff = -qRound(-data->dy); |
3417 | |
3418 | const QSpan *end = spans + count; |
3419 | while (spans < end) { |
3420 | if (!spans->len) { |
3421 | ++spans; |
3422 | continue; |
3423 | } |
3424 | const quint8 coverage = (data->texture.const_alpha * spans->coverage) >> 8; |
3425 | if (coverage == 0) { |
3426 | ++spans; |
3427 | continue; |
3428 | } |
3429 | |
3430 | int x = spans->x; |
3431 | int length = spans->len; |
3432 | int sx = xoff + x; |
3433 | int sy = yoff + spans->y; |
3434 | if (sy >= 0 && sy < image_height && sx < image_width) { |
3435 | if (sx < 0) { |
3436 | x -= sx; |
3437 | length += sx; |
3438 | sx = 0; |
3439 | } |
3440 | if (sx + length > image_width) |
3441 | length = image_width - sx; |
3442 | if (length > 0) { |
3443 | quint16 *dest = (quint16 *)data->rasterBuffer->scanLine(spans->y) + x; |
3444 | const quint16 *src = (const quint16 *)data->texture.scanLine(sy) + sx; |
3445 | if (coverage == 255) { |
3446 | memcpy(dest, src, length * sizeof(quint16)); |
3447 | } else { |
3448 | const quint8 alpha = (coverage + 1) >> 3; |
3449 | const quint8 ialpha = 0x20 - alpha; |
3450 | if (alpha > 0) |
3451 | blend_sourceOver_rgb16_rgb16(dest, src, length, alpha, ialpha); |
3452 | } |
3453 | } |
3454 | } |
3455 | ++spans; |
3456 | } |
3457 | } |
3458 | |
3459 | static void blend_tiled_generic(int count, const QSpan *spans, void *userData) |
3460 | { |
3461 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3462 | |
3463 | uint buffer[BufferSize]; |
3464 | uint src_buffer[BufferSize]; |
3465 | Operator op = getOperator(data, spans, count); |
3466 | |
3467 | const int image_width = data->texture.width; |
3468 | const int image_height = data->texture.height; |
3469 | int xoff = -qRound(-data->dx) % image_width; |
3470 | int yoff = -qRound(-data->dy) % image_height; |
3471 | |
3472 | if (xoff < 0) |
3473 | xoff += image_width; |
3474 | if (yoff < 0) |
3475 | yoff += image_height; |
3476 | |
3477 | while (count--) { |
3478 | int x = spans->x; |
3479 | int length = spans->len; |
3480 | int sx = (xoff + spans->x) % image_width; |
3481 | int sy = (spans->y + yoff) % image_height; |
3482 | if (sx < 0) |
3483 | sx += image_width; |
3484 | if (sy < 0) |
3485 | sy += image_height; |
3486 | |
3487 | const int coverage = (spans->coverage * data->texture.const_alpha) >> 8; |
3488 | while (length) { |
3489 | int l = qMin(image_width - sx, length); |
3490 | if (BufferSize < l) |
3491 | l = BufferSize; |
3492 | const uint *src = op.srcFetch(src_buffer, &op, data, sy, sx, l); |
3493 | uint *dest = op.destFetch(buffer, data->rasterBuffer, x, spans->y, l); |
3494 | op.func(dest, src, l, coverage); |
3495 | if (op.destStore) |
3496 | op.destStore(data->rasterBuffer, x, spans->y, dest, l); |
3497 | x += l; |
3498 | sx += l; |
3499 | length -= l; |
3500 | if (sx >= image_width) |
3501 | sx = 0; |
3502 | } |
3503 | ++spans; |
3504 | } |
3505 | } |
3506 | |
3507 | #if QT_CONFIG(raster_64bit) |
3508 | static void blend_tiled_generic_rgb64(int count, const QSpan *spans, void *userData) |
3509 | { |
3510 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3511 | |
3512 | Operator op = getOperator(data, spans, count); |
3513 | if (!op.func64) { |
3514 | qCDebug(lcQtGuiDrawHelper, "blend_tiled_generic_rgb64: unsupported 64-bit blend attempted, falling back to 32-bit" ); |
3515 | return blend_tiled_generic(count, spans, userData); |
3516 | } |
3517 | alignas(8) QRgba64 buffer[BufferSize]; |
3518 | alignas(8) QRgba64 src_buffer[BufferSize]; |
3519 | |
3520 | const int image_width = data->texture.width; |
3521 | const int image_height = data->texture.height; |
3522 | int xoff = -qRound(-data->dx) % image_width; |
3523 | int yoff = -qRound(-data->dy) % image_height; |
3524 | |
3525 | if (xoff < 0) |
3526 | xoff += image_width; |
3527 | if (yoff < 0) |
3528 | yoff += image_height; |
3529 | |
3530 | bool isBpp32 = qPixelLayouts[data->rasterBuffer->format].bpp == QPixelLayout::BPP32; |
3531 | bool isBpp64 = qPixelLayouts[data->rasterBuffer->format].bpp == QPixelLayout::BPP64; |
3532 | if (op.destFetch64 == destFetch64Undefined && image_width <= BufferSize && (isBpp32 || isBpp64)) { |
3533 | // If destination isn't blended into the result, we can do the tiling directly on destination pixels. |
3534 | while (count--) { |
3535 | int x = spans->x; |
3536 | int y = spans->y; |
3537 | int length = spans->len; |
3538 | int sx = (xoff + spans->x) % image_width; |
3539 | int sy = (spans->y + yoff) % image_height; |
3540 | if (sx < 0) |
3541 | sx += image_width; |
3542 | if (sy < 0) |
3543 | sy += image_height; |
3544 | |
3545 | int sl = qMin(image_width, length); |
3546 | if (sx > 0 && sl > 0) { |
3547 | int l = qMin(image_width - sx, sl); |
3548 | const QRgba64 *src = op.srcFetch64(src_buffer, &op, data, sy, sx, l); |
3549 | op.destStore64(data->rasterBuffer, x, y, src, l); |
3550 | x += l; |
3551 | sx += l; |
3552 | sl -= l; |
3553 | if (sx >= image_width) |
3554 | sx = 0; |
3555 | } |
3556 | if (sl > 0) { |
3557 | Q_ASSERT(sx == 0); |
3558 | const QRgba64 *src = op.srcFetch64(src_buffer, &op, data, sy, sx, sl); |
3559 | op.destStore64(data->rasterBuffer, x, y, src, sl); |
3560 | x += sl; |
3561 | sx += sl; |
3562 | sl -= sl; |
3563 | if (sx >= image_width) |
3564 | sx = 0; |
3565 | } |
3566 | if (isBpp32) { |
3567 | uint *dest = reinterpret_cast<uint *>(data->rasterBuffer->scanLine(y)) + x - image_width; |
3568 | for (int i = image_width; i < length; ++i) |
3569 | dest[i] = dest[i - image_width]; |
3570 | } else { |
3571 | quint64 *dest = reinterpret_cast<quint64 *>(data->rasterBuffer->scanLine(y)) + x - image_width; |
3572 | for (int i = image_width; i < length; ++i) |
3573 | dest[i] = dest[i - image_width]; |
3574 | } |
3575 | ++spans; |
3576 | } |
3577 | return; |
3578 | } |
3579 | |
3580 | while (count--) { |
3581 | int x = spans->x; |
3582 | int length = spans->len; |
3583 | int sx = (xoff + spans->x) % image_width; |
3584 | int sy = (spans->y + yoff) % image_height; |
3585 | if (sx < 0) |
3586 | sx += image_width; |
3587 | if (sy < 0) |
3588 | sy += image_height; |
3589 | |
3590 | const int coverage = (spans->coverage * data->texture.const_alpha) >> 8; |
3591 | while (length) { |
3592 | int l = qMin(image_width - sx, length); |
3593 | if (BufferSize < l) |
3594 | l = BufferSize; |
3595 | const QRgba64 *src = op.srcFetch64(src_buffer, &op, data, sy, sx, l); |
3596 | QRgba64 *dest = op.destFetch64(buffer, data->rasterBuffer, x, spans->y, l); |
3597 | op.func64(dest, src, l, coverage); |
3598 | if (op.destStore64) |
3599 | op.destStore64(data->rasterBuffer, x, spans->y, dest, l); |
3600 | x += l; |
3601 | sx += l; |
3602 | length -= l; |
3603 | if (sx >= image_width) |
3604 | sx = 0; |
3605 | } |
3606 | ++spans; |
3607 | } |
3608 | } |
3609 | #endif |
3610 | |
3611 | static void blend_tiled_argb(int count, const QSpan *spans, void *userData) |
3612 | { |
3613 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3614 | if (data->texture.format != QImage::Format_ARGB32_Premultiplied |
3615 | && data->texture.format != QImage::Format_RGB32) { |
3616 | blend_tiled_generic(count, spans, userData); |
3617 | return; |
3618 | } |
3619 | |
3620 | Operator op = getOperator(data, spans, count); |
3621 | |
3622 | int image_width = data->texture.width; |
3623 | int image_height = data->texture.height; |
3624 | int xoff = -qRound(-data->dx) % image_width; |
3625 | int yoff = -qRound(-data->dy) % image_height; |
3626 | |
3627 | if (xoff < 0) |
3628 | xoff += image_width; |
3629 | if (yoff < 0) |
3630 | yoff += image_height; |
3631 | |
3632 | while (count--) { |
3633 | int x = spans->x; |
3634 | int length = spans->len; |
3635 | int sx = (xoff + spans->x) % image_width; |
3636 | int sy = (spans->y + yoff) % image_height; |
3637 | if (sx < 0) |
3638 | sx += image_width; |
3639 | if (sy < 0) |
3640 | sy += image_height; |
3641 | |
3642 | const int coverage = (spans->coverage * data->texture.const_alpha) >> 8; |
3643 | while (length) { |
3644 | int l = qMin(image_width - sx, length); |
3645 | if (BufferSize < l) |
3646 | l = BufferSize; |
3647 | const uint *src = (const uint *)data->texture.scanLine(sy) + sx; |
3648 | uint *dest = ((uint *)data->rasterBuffer->scanLine(spans->y)) + x; |
3649 | op.func(dest, src, l, coverage); |
3650 | x += l; |
3651 | sx += l; |
3652 | length -= l; |
3653 | if (sx >= image_width) |
3654 | sx = 0; |
3655 | } |
3656 | ++spans; |
3657 | } |
3658 | } |
3659 | |
3660 | static void blend_tiled_rgb565(int count, const QSpan *spans, void *userData) |
3661 | { |
3662 | QSpanData *data = reinterpret_cast<QSpanData*>(userData); |
3663 | QPainter::CompositionMode mode = data->rasterBuffer->compositionMode; |
3664 | |
3665 | if (data->texture.format != QImage::Format_RGB16 |
3666 | || (mode != QPainter::CompositionMode_SourceOver |
3667 | && mode != QPainter::CompositionMode_Source)) |
3668 | { |
3669 | blend_tiled_generic(count, spans, userData); |
3670 | return; |
3671 | } |
3672 | |
3673 | const int image_width = data->texture.width; |
3674 | const int image_height = data->texture.height; |
3675 | int xoff = -qRound(-data->dx) % image_width; |
3676 | int yoff = -qRound(-data->dy) % image_height; |
3677 | |
3678 | if (xoff < 0) |
3679 | xoff += image_width; |
3680 | if (yoff < 0) |
3681 | yoff += image_height; |
3682 | |
3683 | while (count--) { |
3684 | const quint8 coverage = (data->texture.const_alpha * spans->coverage) >> 8; |
3685 | if (coverage == 0) { |
3686 | ++spans; |
3687 | continue; |
3688 | } |
3689 | |
3690 | int x = spans->x; |
3691 | int length = spans->len; |
3692 | int sx = (xoff + spans->x) % image_width; |
3693 | int sy = (spans->y + yoff) % image_height; |
3694 | if (sx < 0) |
3695 | sx += image_width; |
3696 | if (sy < 0) |
3697 | sy += image_height; |
3698 | |
3699 | if (coverage == 255) { |
3700 | // Copy the first texture block |
3701 | length = qMin(image_width,length); |
3702 | int tx = x; |
3703 | while (length) { |
3704 | int l = qMin(image_width - sx, length); |
3705 | if (BufferSize < l) |
3706 | l = BufferSize; |
3707 | quint16 *dest = ((quint16 *)data->rasterBuffer->scanLine(spans->y)) + tx; |
3708 | const quint16 *src = (const quint16 *)data->texture.scanLine(sy) + sx; |
3709 | memcpy(dest, src, l * sizeof(quint16)); |
3710 | length -= l; |
3711 | tx += l; |
3712 | sx += l; |
3713 | if (sx >= image_width) |
3714 | sx = 0; |
3715 | } |
3716 | |
3717 | // Now use the rasterBuffer as the source of the texture, |
3718 | // We can now progressively copy larger blocks |
3719 | // - Less cpu time in code figuring out what to copy |
3720 | // We are dealing with one block of data |
3721 | // - More likely to fit in the cache |
3722 | // - can use memcpy |
3723 | int copy_image_width = qMin(image_width, int(spans->len)); |
3724 | length = spans->len - copy_image_width; |
3725 | quint16 *src = ((quint16 *)data->rasterBuffer->scanLine(spans->y)) + x; |
3726 | quint16 *dest = src + copy_image_width; |
3727 | while (copy_image_width < length) { |
3728 | memcpy(dest, src, copy_image_width * sizeof(quint16)); |
3729 | dest += copy_image_width; |
3730 | length -= copy_image_width; |
3731 | copy_image_width *= 2; |
3732 | } |
3733 | if (length > 0) |
3734 | memcpy(dest, src, length * sizeof(quint16)); |
3735 | } else { |
3736 | const quint8 alpha = (coverage + 1) >> 3; |
3737 | const quint8 ialpha = 0x20 - alpha; |
3738 | if (alpha > 0) { |
3739 | while (length) { |
3740 | int l = qMin(image_width - sx, length); |
3741 | if (BufferSize < l) |
3742 | l = BufferSize; |
3743 | quint16 *dest = ((quint16 *)data->rasterBuffer->scanLine(spans->y)) + x; |
3744 | const quint16 *src = (const quint16 *)data->texture.scanLine(sy) + sx; |
3745 | blend_sourceOver_rgb16_rgb16(dest, src, l, alpha, ialpha); |
3746 | x += l; |
3747 | sx += l; |
3748 | length -= l; |
3749 | if (sx >= image_width) |
3750 | sx = 0; |
3751 | } |
3752 | } |
3753 | } |
3754 | ++spans; |
3755 | } |
3756 | } |
3757 | |
3758 | /* Image formats here are target formats */ |
3759 | static const ProcessSpans processTextureSpansARGB32PM[NBlendTypes] = { |
3760 | blend_untransformed_argb, // Untransformed |
3761 | blend_tiled_argb, // Tiled |
3762 | blend_src_generic, // Transformed |
3763 | blend_src_generic, // TransformedTiled |
3764 | blend_src_generic, // TransformedBilinear |
3765 | blend_src_generic // TransformedBilinearTiled |
3766 | }; |
3767 | |
3768 | static const ProcessSpans processTextureSpansRGB16[NBlendTypes] = { |
3769 | blend_untransformed_rgb565, // Untransformed |
3770 | blend_tiled_rgb565, // Tiled |
3771 | blend_src_generic, // Transformed |
3772 | blend_src_generic, // TransformedTiled |
3773 | blend_src_generic, // TransformedBilinear |
3774 | blend_src_generic // TransformedBilinearTiled |
3775 | }; |
3776 | |
3777 | static const ProcessSpans processTextureSpansGeneric[NBlendTypes] = { |
3778 | blend_untransformed_generic, // Untransformed |
3779 | blend_tiled_generic, // Tiled |
3780 | blend_src_generic, // Transformed |
3781 | blend_src_generic, // TransformedTiled |
3782 | blend_src_generic, // TransformedBilinear |
3783 | blend_src_generic // TransformedBilinearTiled |
3784 | }; |
3785 | |
3786 | #if QT_CONFIG(raster_64bit) |
3787 | static const ProcessSpans processTextureSpansGeneric64[NBlendTypes] = { |
3788 | blend_untransformed_generic_rgb64, // Untransformed |
3789 | blend_tiled_generic_rgb64, // Tiled |
3790 | blend_src_generic_rgb64, // Transformed |
3791 | blend_src_generic_rgb64, // TransformedTiled |
3792 | blend_src_generic_rgb64, // TransformedBilinear |
3793 | blend_src_generic_rgb64 // TransformedBilinearTiled |
3794 | }; |
3795 | #endif |
3796 | |
3797 | void qBlendTexture(int count, const QSpan *spans, void *userData) |
3798 | { |
3799 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3800 | TextureBlendType blendType = getBlendType(data); |
3801 | ProcessSpans proc; |
3802 | switch (data->rasterBuffer->format) { |
3803 | case QImage::Format_ARGB32_Premultiplied: |
3804 | proc = processTextureSpansARGB32PM[blendType]; |
3805 | break; |
3806 | case QImage::Format_RGB16: |
3807 | proc = processTextureSpansRGB16[blendType]; |
3808 | break; |
3809 | #if QT_CONFIG(raster_64bit) |
3810 | #if defined(__SSE2__) || defined(__ARM_NEON__) || (Q_PROCESSOR_WORDSIZE == 8) |
3811 | case QImage::Format_ARGB32: |
3812 | case QImage::Format_RGBA8888: |
3813 | #endif |
3814 | case QImage::Format_BGR30: |
3815 | case QImage::Format_A2BGR30_Premultiplied: |
3816 | case QImage::Format_RGB30: |
3817 | case QImage::Format_A2RGB30_Premultiplied: |
3818 | case QImage::Format_RGBX64: |
3819 | case QImage::Format_RGBA64: |
3820 | case QImage::Format_RGBA64_Premultiplied: |
3821 | case QImage::Format_Grayscale16: |
3822 | proc = processTextureSpansGeneric64[blendType]; |
3823 | break; |
3824 | #endif // QT_CONFIG(raster_64bit) |
3825 | case QImage::Format_Invalid: |
3826 | Q_UNREACHABLE(); |
3827 | return; |
3828 | default: |
3829 | proc = processTextureSpansGeneric[blendType]; |
3830 | break; |
3831 | } |
3832 | proc(count, spans, userData); |
3833 | } |
3834 | |
3835 | static void blend_vertical_gradient_argb(int count, const QSpan *spans, void *userData) |
3836 | { |
3837 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3838 | |
3839 | LinearGradientValues linear; |
3840 | getLinearGradientValues(&linear, data); |
3841 | |
3842 | CompositionFunctionSolid funcSolid = |
3843 | functionForModeSolid[data->rasterBuffer->compositionMode]; |
3844 | |
3845 | /* |
3846 | The logic for vertical gradient calculations is a mathematically |
3847 | reduced copy of that in fetchLinearGradient() - which is basically: |
3848 | |
3849 | qreal ry = data->m22 * (y + 0.5) + data->dy; |
3850 | qreal t = linear.dy*ry + linear.off; |
3851 | t *= (GRADIENT_STOPTABLE_SIZE - 1); |
3852 | quint32 color = |
3853 | qt_gradient_pixel_fixed(&data->gradient, |
3854 | int(t * FIXPT_SIZE)); |
3855 | |
3856 | This has then been converted to fixed point to improve performance. |
3857 | */ |
3858 | const int gss = GRADIENT_STOPTABLE_SIZE - 1; |
3859 | int yinc = int((linear.dy * data->m22 * gss) * FIXPT_SIZE); |
3860 | int off = int((((linear.dy * (data->m22 * qreal(0.5) + data->dy) + linear.off) * gss) * FIXPT_SIZE)); |
3861 | |
3862 | while (count--) { |
3863 | int y = spans->y; |
3864 | int x = spans->x; |
3865 | |
3866 | quint32 *dst = (quint32 *)(data->rasterBuffer->scanLine(y)) + x; |
3867 | quint32 color = |
3868 | qt_gradient_pixel_fixed(&data->gradient, yinc * y + off); |
3869 | |
3870 | funcSolid(dst, spans->len, color, spans->coverage); |
3871 | ++spans; |
3872 | } |
3873 | } |
3874 | |
3875 | template<ProcessSpans blend_color> |
3876 | static void blend_vertical_gradient(int count, const QSpan *spans, void *userData) |
3877 | { |
3878 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3879 | |
3880 | LinearGradientValues linear; |
3881 | getLinearGradientValues(&linear, data); |
3882 | |
3883 | // Based on the same logic as blend_vertical_gradient_argb. |
3884 | |
3885 | const int gss = GRADIENT_STOPTABLE_SIZE - 1; |
3886 | int yinc = int((linear.dy * data->m22 * gss) * FIXPT_SIZE); |
3887 | int off = int((((linear.dy * (data->m22 * qreal(0.5) + data->dy) + linear.off) * gss) * FIXPT_SIZE)); |
3888 | |
3889 | while (count--) { |
3890 | int y = spans->y; |
3891 | |
3892 | #if QT_CONFIG(raster_64bit) |
3893 | data->solidColor = qt_gradient_pixel64_fixed(&data->gradient, yinc * y + off); |
3894 | #else |
3895 | data->solidColor = QRgba64::fromArgb32(qt_gradient_pixel_fixed(&data->gradient, yinc * y + off)); |
3896 | #endif |
3897 | blend_color(1, spans, userData); |
3898 | ++spans; |
3899 | } |
3900 | } |
3901 | |
3902 | void qBlendGradient(int count, const QSpan *spans, void *userData) |
3903 | { |
3904 | QSpanData *data = reinterpret_cast<QSpanData *>(userData); |
3905 | bool isVerticalGradient = |
3906 | data->txop <= QTransform::TxScale && |
3907 | data->type == QSpanData::LinearGradient && |
3908 | data->gradient.linear.end.x == data->gradient.linear.origin.x; |
3909 | switch (data->rasterBuffer->format) { |
3910 | case QImage::Format_RGB16: |
3911 | if (isVerticalGradient) |
3912 | return blend_vertical_gradient<blend_color_rgb16>(count, spans, userData); |
3913 | return blend_src_generic(count, spans, userData); |
3914 | case QImage::Format_RGB32: |
3915 | case QImage::Format_ARGB32_Premultiplied: |
3916 | if (isVerticalGradient) |
3917 | return blend_vertical_gradient_argb(count, spans, userData); |
3918 | return blend_src_generic(count, spans, userData); |
3919 | #if QT_CONFIG(raster_64bit) |
3920 | #if defined(__SSE2__) || defined(__ARM_NEON__) || (Q_PROCESSOR_WORDSIZE == 8) |
3921 | case QImage::Format_ARGB32: |
3922 | case QImage::Format_RGBA8888: |
3923 | #endif |
3924 | case QImage::Format_BGR30: |
3925 | case QImage::Format_A2BGR30_Premultiplied: |
3926 | case QImage::Format_RGB30: |
3927 | case QImage::Format_A2RGB30_Premultiplied: |
3928 | case QImage::Format_RGBX64: |
3929 | case QImage::Format_RGBA64: |
3930 | case QImage::Format_RGBA64_Premultiplied: |
3931 | if (isVerticalGradient) |
3932 | return blend_vertical_gradient<blend_color_generic_rgb64>(count, spans, userData); |
3933 | return blend_src_generic_rgb64(count, spans, userData); |
3934 | #endif // QT_CONFIG(raster_64bit) |
3935 | case QImage::Format_Invalid: |
3936 | break; |
3937 | default: |
3938 | if (isVerticalGradient) |
3939 | return blend_vertical_gradient<blend_color_generic>(count, spans, userData); |
3940 | return blend_src_generic(count, spans, userData); |
3941 | } |
3942 | Q_UNREACHABLE(); |
3943 | } |
3944 | |
3945 | template <class DST> static |
3946 | inline void qt_bitmapblit_template(QRasterBuffer *rasterBuffer, |
3947 | int x, int y, DST color, |
3948 | const uchar *map, |
3949 | int mapWidth, int mapHeight, int mapStride) |
3950 | { |
3951 | DST *dest = reinterpret_cast<DST *>(rasterBuffer->scanLine(y)) + x; |
3952 | const int destStride = rasterBuffer->stride<DST>(); |
3953 | |
3954 | if (mapWidth > 8) { |
3955 | while (mapHeight--) { |
3956 | int x0 = 0; |
3957 | int n = 0; |
3958 | for (int x = 0; x < mapWidth; x += 8) { |
3959 | uchar s = map[x >> 3]; |
3960 | for (int i = 0; i < 8; ++i) { |
3961 | if (s & 0x80) { |
3962 | ++n; |
3963 | } else { |
3964 | if (n) { |
3965 | qt_memfill(dest + x0, color, n); |
3966 | x0 += n + 1; |
3967 | n = 0; |
3968 | } else { |
3969 | ++x0; |
3970 | } |
3971 | if (!s) { |
3972 | x0 += 8 - 1 - i; |
3973 | break; |
3974 | } |
3975 | } |
3976 | s <<= 1; |
3977 | } |
3978 | } |
3979 | if (n) |
3980 | qt_memfill(dest + x0, color, n); |
3981 | dest += destStride; |
3982 | map += mapStride; |
3983 | } |
3984 | } else { |
3985 | while (mapHeight--) { |
3986 | int x0 = 0; |
3987 | int n = 0; |
3988 | for (uchar s = *map; s; s <<= 1) { |
3989 | if (s & 0x80) { |
3990 | ++n; |
3991 | } else if (n) { |
3992 | qt_memfill(dest + x0, color, n); |
3993 | x0 += n + 1; |
3994 | n = 0; |
3995 | } else { |
3996 | ++x0; |
3997 | } |
3998 | } |
3999 | if (n) |
4000 | qt_memfill(dest + x0, color, n); |
4001 | dest += destStride; |
4002 | map += mapStride; |
4003 | } |
4004 | } |
4005 | } |
4006 | |
4007 | inline static void qt_bitmapblit_argb32(QRasterBuffer *rasterBuffer, |
4008 | int x, int y, const QRgba64 &color, |
4009 | const uchar *map, |
4010 | int mapWidth, int mapHeight, int mapStride) |
4011 | { |
4012 | qt_bitmapblit_template<quint32>(rasterBuffer, x, y, color.toArgb32(), |
4013 | map, mapWidth, mapHeight, mapStride); |
4014 | } |
4015 | |
4016 | inline static void qt_bitmapblit_rgba8888(QRasterBuffer *rasterBuffer, |
4017 | int x, int y, const QRgba64 &color, |
4018 | const uchar *map, |
4019 | int mapWidth, int mapHeight, int mapStride) |
4020 | { |
4021 | qt_bitmapblit_template<quint32>(rasterBuffer, x, y, ARGB2RGBA(color.toArgb32()), |
4022 | map, mapWidth, mapHeight, mapStride); |
4023 | } |
4024 | |
4025 | template<QtPixelOrder PixelOrder> |
4026 | inline static void qt_bitmapblit_rgb30(QRasterBuffer *rasterBuffer, |
4027 | int x, int y, const QRgba64 &color, |
4028 | const uchar *map, |
4029 | int mapWidth, int mapHeight, int mapStride) |
4030 | { |
4031 | qt_bitmapblit_template<quint32>(rasterBuffer, x, y, qConvertRgb64ToRgb30<PixelOrder>(color), |
4032 | map, mapWidth, mapHeight, mapStride); |
4033 | } |
4034 | |
4035 | inline static void qt_bitmapblit_quint16(QRasterBuffer *rasterBuffer, |
4036 | int x, int y, const QRgba64 &color, |
4037 | const uchar *map, |
4038 | int mapWidth, int mapHeight, int mapStride) |
4039 | { |
4040 | qt_bitmapblit_template<quint16>(rasterBuffer, x, y, color.toRgb16(), |
4041 | map, mapWidth, mapHeight, mapStride); |
4042 | } |
4043 | |
4044 | static inline void grayBlendPixel(quint32 *dst, int coverage, QRgba64 srcLinear, const QColorTrcLut *colorProfile) |
4045 | { |
4046 | // Do a gammacorrected gray alphablend... |
4047 | const QRgba64 dstLinear = colorProfile ? colorProfile->toLinear64(*dst) : QRgba64::fromArgb32(*dst); |
4048 | |
4049 | QRgba64 blend = interpolate255(srcLinear, coverage, dstLinear, 255 - coverage); |
4050 | |
4051 | *dst = colorProfile ? colorProfile->fromLinear64(blend) : toArgb32(blend); |
4052 | } |
4053 | |
4054 | static inline void alphamapblend_argb32(quint32 *dst, int coverage, QRgba64 srcLinear, quint32 src, const QColorTrcLut *colorProfile) |
4055 | { |
4056 | if (coverage == 0) { |
4057 | // nothing |
4058 | } else if (coverage == 255 || !colorProfile) { |
4059 | blend_pixel(*dst, src, coverage); |
4060 | } else if (*dst < 0xff000000) { |
4061 | // Give up and do a naive gray alphablend. Needed to deal with ARGB32 and invalid ARGB32_premultiplied, see QTBUG-60571 |
4062 | blend_pixel(*dst, src, coverage); |
4063 | } else if (src >= 0xff000000) { |
4064 | grayBlendPixel(dst, coverage, srcLinear, colorProfile); |
4065 | } else { |
4066 | // First do naive blend with text-color |
4067 | QRgb s = *dst; |
4068 | blend_pixel(s, src); |
4069 | // Then gamma-corrected blend with glyph shape |
4070 | QRgba64 s64 = colorProfile ? colorProfile->toLinear64(s) : QRgba64::fromArgb32(s); |
4071 | grayBlendPixel(dst, coverage, s64, colorProfile); |
4072 | } |
4073 | } |
4074 | |
4075 | #if QT_CONFIG(raster_64bit) |
4076 | |
4077 | static inline void grayBlendPixel(QRgba64 &dst, int coverage, QRgba64 srcLinear, const QColorTrcLut *colorProfile) |
4078 | { |
4079 | // Do a gammacorrected gray alphablend... |
4080 | QRgba64 dstColor = dst; |
4081 | if (colorProfile) { |
4082 | if (dstColor.isOpaque()) |
4083 | dstColor = colorProfile->toLinear(dstColor); |
4084 | else if (!dstColor.isTransparent()) |
4085 | dstColor = colorProfile->toLinear(dstColor.unpremultiplied()).premultiplied(); |
4086 | } |
4087 | |
4088 | blend_pixel(dstColor, srcLinear, coverage); |
4089 | |
4090 | if (colorProfile) { |
4091 | if (dstColor.isOpaque()) |
4092 | dstColor = colorProfile->fromLinear(dstColor); |
4093 | else if (!dstColor.isTransparent()) |
4094 | dstColor = colorProfile->fromLinear(dstColor.unpremultiplied()).premultiplied(); |
4095 | } |
4096 | dst = dstColor; |
4097 | } |
4098 | |
4099 | static inline void alphamapblend_generic(int coverage, QRgba64 *dest, int x, const QRgba64 &srcLinear, const QRgba64 &src, const QColorTrcLut *colorProfile) |
4100 | { |
4101 | if (coverage == 0) { |
4102 | // nothing |
4103 | } else if (coverage == 255) { |
4104 | blend_pixel(dest[x], src); |
4105 | } else if (src.isOpaque()) { |
4106 | grayBlendPixel(dest[x], coverage, srcLinear, colorProfile); |
4107 | } else { |
4108 | // First do naive blend with text-color |
4109 | QRgba64 s = dest[x]; |
4110 | blend_pixel(s, src); |
4111 | // Then gamma-corrected blend with glyph shape |
4112 | if (colorProfile) |
4113 | s = colorProfile->toLinear(s); |
4114 | grayBlendPixel(dest[x], coverage, s, colorProfile); |
4115 | } |
4116 | } |
4117 | |
4118 | static void qt_alphamapblit_generic(QRasterBuffer *rasterBuffer, |
4119 | int x, int y, const QRgba64 &color, |
4120 | const uchar *map, |
4121 | int mapWidth, int mapHeight, int mapStride, |
4122 | const QClipData *clip, bool useGammaCorrection) |
4123 | { |
4124 | if (color.isTransparent()) |
4125 | return; |
4126 | |
4127 | const QColorTrcLut *colorProfile = nullptr; |
4128 | |
4129 | if (useGammaCorrection) |
4130 | colorProfile = QGuiApplicationPrivate::instance()->colorProfileForA8Text(); |
4131 | |
4132 | QRgba64 srcColor = color; |
4133 | if (colorProfile && color.isOpaque()) |
4134 | srcColor = colorProfile->toLinear(srcColor); |
4135 | |
4136 | alignas(8) QRgba64 buffer[BufferSize]; |
4137 | const DestFetchProc64 destFetch64 = destFetchProc64[rasterBuffer->format]; |
4138 | const DestStoreProc64 destStore64 = destStoreProc64[rasterBuffer->format]; |
4139 | |
4140 | if (!clip) { |
4141 | for (int ly = 0; ly < mapHeight; ++ly) { |
4142 | int i = x; |
4143 | int length = mapWidth; |
4144 | while (length > 0) { |
4145 | int l = qMin(BufferSize, length); |
4146 | QRgba64 *dest = destFetch64(buffer, rasterBuffer, i, y + ly, l); |
4147 | for (int j=0; j < l; ++j) { |
4148 | const int coverage = map[j + (i - x)]; |
4149 | alphamapblend_generic(coverage, dest, j, srcColor, color, colorProfile); |
4150 | } |
4151 | if (destStore64) |
4152 | destStore64(rasterBuffer, i, y + ly, dest, l); |
4153 | length -= l; |
4154 | i += l; |
4155 | } |
4156 | map += mapStride; |
4157 | } |
4158 | } else { |
4159 | int bottom = qMin(y + mapHeight, rasterBuffer->height()); |
4160 | |
4161 | int top = qMax(y, 0); |
4162 | map += (top - y) * mapStride; |
4163 | |
4164 | const_cast<QClipData *>(clip)->initialize(); |
4165 | for (int yp = top; yp<bottom; ++yp) { |
4166 | const QClipData::ClipLine &line = clip->m_clipLines[yp]; |
4167 | |
4168 | for (int i=0; i<line.count; ++i) { |
4169 | const QSpan &clip = line.spans[i]; |
4170 | |
4171 | int start = qMax<int>(x, clip.x); |
4172 | int end = qMin<int>(x + mapWidth, clip.x + clip.len); |
4173 | if (end <= start) |
4174 | continue; |
4175 | Q_ASSERT(end - start <= BufferSize); |
4176 | QRgba64 *dest = destFetch64(buffer, rasterBuffer, start, clip.y, end - start); |
4177 | |
4178 | for (int xp=start; xp<end; ++xp) { |
4179 | const int coverage = map[xp - x]; |
4180 | alphamapblend_generic(coverage, dest, xp - start, srcColor, color, colorProfile); |
4181 | } |
4182 | if (destStore64) |
4183 | destStore64(rasterBuffer, start, clip.y, dest, end - start); |
4184 | } // for (i -> line.count) |
4185 | map += mapStride; |
4186 | } // for (yp -> bottom) |
4187 | } |
4188 | } |
4189 | #else |
4190 | static void qt_alphamapblit_generic(QRasterBuffer *rasterBuffer, |
4191 | int x, int y, const QRgba64 &color, |
4192 | const uchar *map, |
4193 | int mapWidth, int mapHeight, int mapStride, |
4194 | const QClipData *clip, bool useGammaCorrection) |
4195 | { |
4196 | if (color.isTransparent()) |
4197 | return; |
4198 | |
4199 | const quint32 c = color.toArgb32(); |
4200 | |
4201 | const QColorTrcLut *colorProfile = nullptr; |
4202 | |
4203 | if (useGammaCorrection) |
4204 | colorProfile = QGuiApplicationPrivate::instance()->colorProfileForA8Text(); |
4205 | |
4206 | QRgba64 srcColor = color; |
4207 | if (colorProfile && color.isOpaque()) |
4208 | srcColor = colorProfile->toLinear(srcColor); |
4209 | |
4210 | quint32 buffer[BufferSize]; |
4211 | const DestFetchProc destFetch = destFetchProc[rasterBuffer->format]; |
4212 | const DestStoreProc destStore = destStoreProc[rasterBuffer->format]; |
4213 | |
4214 | if (!clip) { |
4215 | for (int ly = 0; ly < mapHeight; ++ly) { |
4216 | int i = x; |
4217 | int length = mapWidth; |
4218 | while (length > 0) { |
4219 | int l = qMin(BufferSize, length); |
4220 | quint32 *dest = destFetch(buffer, rasterBuffer, i, y + ly, l); |
4221 | for (int j=0; j < l; ++j) { |
4222 | const int coverage = map[j + (i - x)]; |
4223 | alphamapblend_argb32(dest + j, coverage, srcColor, c, colorProfile); |
4224 | } |
4225 | if (destStore) |
4226 | destStore(rasterBuffer, i, y + ly, dest, l); |
4227 | length -= l; |
4228 | i += l; |
4229 | } |
4230 | map += mapStride; |
4231 | } |
4232 | } else { |
4233 | int bottom = qMin(y + mapHeight, rasterBuffer->height()); |
4234 | |
4235 | int top = qMax(y, 0); |
4236 | map += (top - y) * mapStride; |
4237 | |
4238 | const_cast<QClipData *>(clip)->initialize(); |
4239 | for (int yp = top; yp<bottom; ++yp) { |
4240 | const QClipData::ClipLine &line = clip->m_clipLines[yp]; |
4241 | |
4242 | for (int i=0; i<line.count; ++i) { |
4243 | const QSpan &clip = line.spans[i]; |
4244 | |
4245 | int start = qMax<int>(x, clip.x); |
4246 | int end = qMin<int>(x + mapWidth, clip.x + clip.len); |
4247 | if (end <= start) |
4248 | continue; |
4249 | Q_ASSERT(end - start <= BufferSize); |
4250 | quint32 *dest = destFetch(buffer, rasterBuffer, start, clip.y, end - start); |
4251 | |
4252 | for (int xp=start; xp<end; ++xp) { |
4253 | const int coverage = map[xp - x]; |
4254 | alphamapblend_argb32(dest + xp - x, coverage, srcColor, color, colorProfile); |
4255 | } |
4256 | if (destStore) |
4257 | destStore(rasterBuffer, start, clip.y, dest, end - start); |
4258 | } // for (i -> line.count) |
4259 | map += mapStride; |
4260 | } // for (yp -> bottom) |
4261 | } |
4262 | } |
4263 | #endif |
4264 | |
4265 | static inline void alphamapblend_quint16(int coverage, quint16 *dest, int x, const quint16 srcColor) |
4266 | { |
4267 | if (coverage == 0) { |
4268 | // nothing |
4269 | } else if (coverage == 255) { |
4270 | dest[x] = srcColor; |
4271 | } else { |
4272 | dest[x] = BYTE_MUL_RGB16(srcColor, coverage) |
4273 | + BYTE_MUL_RGB16(dest[x], 255 - coverage); |
4274 | } |
4275 | } |
4276 | |
4277 | void qt_alphamapblit_quint16(QRasterBuffer *rasterBuffer, |
4278 | int x, int y, const QRgba64 &color, |
4279 | const uchar *map, |
4280 | int mapWidth, int mapHeight, int mapStride, |
4281 | const QClipData *clip, bool useGammaCorrection) |
4282 | { |
4283 | if (useGammaCorrection || !color.isOpaque()) { |
4284 | qt_alphamapblit_generic(rasterBuffer, x, y, color, map, mapWidth, mapHeight, mapStride, clip, useGammaCorrection); |
4285 | return; |
4286 | } |
4287 | |
4288 | const quint16 c = color.toRgb16(); |
4289 | |
4290 | if (!clip) { |
4291 | quint16 *dest = reinterpret_cast<quint16*>(rasterBuffer->scanLine(y)) + x; |
4292 | const int destStride = rasterBuffer->stride<quint16>(); |
4293 | while (mapHeight--) { |
4294 | for (int i = 0; i < mapWidth; ++i) |
4295 | alphamapblend_quint16(map[i], dest, i, c); |
4296 | dest += destStride; |
4297 | map += mapStride; |
4298 | } |
4299 | } else { |
4300 | int top = qMax(y, 0); |
4301 | int bottom = qMin(y + mapHeight, rasterBuffer->height()); |
4302 | map += (top - y) * mapStride; |
4303 | |
4304 | const_cast<QClipData *>(clip)->initialize(); |
4305 | for (int yp = top; yp<bottom; ++yp) { |
4306 | const QClipData::ClipLine &line = clip->m_clipLines[yp]; |
4307 | |
4308 | quint16 *dest = reinterpret_cast<quint16*>(rasterBuffer->scanLine(yp)); |
4309 | |
4310 | for (int i=0; i<line.count; ++i) { |
4311 | const QSpan &clip = line.spans[i]; |
4312 | |
4313 | int start = qMax<int>(x, clip.x); |
4314 | int end = qMin<int>(x + mapWidth, clip.x + clip.len); |
4315 | |
4316 | for (int xp=start; xp<end; ++xp) |
4317 | alphamapblend_quint16(map[xp - x], dest, xp, c); |
4318 | } // for (i -> line.count) |
4319 | map += mapStride; |
4320 | } // for (yp -> bottom) |
4321 | } |
4322 | } |
4323 | |
4324 | static void qt_alphamapblit_argb32(QRasterBuffer *rasterBuffer, |
4325 | int x, int y, const QRgba64 &color, |
4326 | const uchar *map, |
4327 | int mapWidth, int mapHeight, int mapStride, |
4328 | const QClipData *clip, bool useGammaCorrection) |
4329 | { |
4330 | const quint32 c = color.toArgb32(); |
4331 | const int destStride = rasterBuffer->stride<quint32>(); |
4332 | |
4333 | if (color.isTransparent()) |
4334 | return; |
4335 | |
4336 | const QColorTrcLut *colorProfile = nullptr; |
4337 | |
4338 | if (useGammaCorrection) |
4339 | colorProfile = QGuiApplicationPrivate::instance()->colorProfileForA8Text(); |
4340 | |
4341 | QRgba64 srcColor = color; |
4342 | if (colorProfile && color.isOpaque()) |
4343 | srcColor = colorProfile->toLinear(srcColor); |
4344 | |
4345 | if (!clip) { |
4346 | quint32 *dest = reinterpret_cast<quint32*>(rasterBuffer->scanLine(y)) + x; |
4347 | while (mapHeight--) { |
4348 | for (int i = 0; i < mapWidth; ++i) { |
4349 | const int coverage = map[i]; |
4350 | alphamapblend_argb32(dest + i, coverage, srcColor, c, colorProfile); |
4351 | } |
4352 | dest += destStride; |
4353 | map += mapStride; |
4354 | } |
4355 | } else { |
4356 | int bottom = qMin(y + mapHeight, rasterBuffer->height()); |
4357 | |
4358 | int top = qMax(y, 0); |
4359 | map += (top - y) * mapStride; |
4360 | |
4361 | const_cast<QClipData *>(clip)->initialize(); |
4362 | for (int yp = top; yp<bottom; ++yp) { |
4363 | const QClipData::ClipLine &line = clip->m_clipLines[yp]; |
4364 | |
4365 | quint32 *dest = reinterpret_cast<quint32 *>(rasterBuffer->scanLine(yp)); |
4366 | |
4367 | for (int i=0; i<line.count; ++i) { |
4368 | const QSpan &clip = line.spans[i]; |
4369 | |
4370 | int start = qMax<int>(x, clip.x); |
4371 | int end = qMin<int>(x + mapWidth, clip.x + clip.len); |
4372 | |
4373 | for (int xp=start; xp<end; ++xp) { |
4374 | const int coverage = map[xp - x]; |
4375 | alphamapblend_argb32(dest + xp, coverage, srcColor, c, colorProfile); |
4376 | } // for (i -> line.count) |
4377 | } // for (yp -> bottom) |
4378 | map += mapStride; |
4379 | } |
4380 | } |
4381 | } |
4382 | |
4383 | static inline int qRgbAvg(QRgb rgb) |
4384 | { |
4385 | return (qRed(rgb) * 5 + qGreen(rgb) * 6 + qBlue(rgb) * 5) / 16; |
4386 | } |
4387 | |
4388 | static inline void rgbBlendPixel(quint32 *dst, int coverage, QRgba64 slinear, const QColorTrcLut *colorProfile) |
4389 | { |
4390 | // Do a gammacorrected RGB alphablend... |
4391 | const QRgba64 dlinear = colorProfile ? colorProfile->toLinear64(*dst) : QRgba64::fromArgb32(*dst); |
4392 | |
4393 | QRgba64 blend = rgbBlend(dlinear, slinear, coverage); |
4394 | |
4395 | *dst = colorProfile ? colorProfile->fromLinear64(blend) : toArgb32(blend); |
4396 | } |
4397 | |
4398 | static inline QRgb rgbBlend(QRgb d, QRgb s, uint rgbAlpha) |
4399 | { |
4400 | #if defined(__SSE2__) |
4401 | __m128i vd = _mm_cvtsi32_si128(d); |
4402 | __m128i vs = _mm_cvtsi32_si128(s); |
4403 | __m128i va = _mm_cvtsi32_si128(rgbAlpha); |
4404 | const __m128i vz = _mm_setzero_si128(); |
4405 | vd = _mm_unpacklo_epi8(vd, vz); |
4406 | vs = _mm_unpacklo_epi8(vs, vz); |
4407 | va = _mm_unpacklo_epi8(va, vz); |
4408 | __m128i vb = _mm_xor_si128(_mm_set1_epi16(255), va); |
4409 | vs = _mm_mullo_epi16(vs, va); |
4410 | vd = _mm_mullo_epi16(vd, vb); |
4411 | vd = _mm_add_epi16(vd, vs); |
4412 | vd = _mm_add_epi16(vd, _mm_srli_epi16(vd, 8)); |
4413 | vd = _mm_add_epi16(vd, _mm_set1_epi16(0x80)); |
4414 | vd = _mm_srli_epi16(vd, 8); |
4415 | vd = _mm_packus_epi16(vd, vd); |
4416 | return _mm_cvtsi128_si32(vd); |
4417 | #else |
4418 | const int dr = qRed(d); |
4419 | const int dg = qGreen(d); |
4420 | const int db = qBlue(d); |
4421 | |
4422 | const int sr = qRed(s); |
4423 | const int sg = qGreen(s); |
4424 | const int sb = qBlue(s); |
4425 | |
4426 | const int mr = qRed(rgbAlpha); |
4427 | const int mg = qGreen(rgbAlpha); |
4428 | const int mb = qBlue(rgbAlpha); |
4429 | |
4430 | const int nr = qt_div_255(sr * mr + dr * (255 - mr)); |
4431 | const int ng = qt_div_255(sg * mg + dg * (255 - mg)); |
4432 | const int nb = qt_div_255(sb * mb + db * (255 - mb)); |
4433 | |
4434 | return 0xff000000 | (nr << 16) | (ng << 8) | nb; |
4435 | #endif |
4436 | } |
4437 | |
4438 | static inline void alphargbblend_argb32(quint32 *dst, uint coverage, const QRgba64 &srcLinear, quint32 src, const QColorTrcLut *colorProfile) |
4439 | { |
4440 | if (coverage == 0xff000000) { |
4441 | // nothing |
4442 | } else if (coverage == 0xffffffff && qAlpha(src) == 255) { |
4443 | blend_pixel(*dst, src); |
4444 | } else if (*dst < 0xff000000) { |
4445 | // Give up and do a naive gray alphablend. Needed to deal with ARGB32 and invalid ARGB32_premultiplied, see QTBUG-60571 |
4446 | blend_pixel(*dst, src, qRgbAvg(coverage)); |
4447 | } else if (!colorProfile) { |
4448 | // First do naive blend with text-color |
4449 | QRgb s = *dst; |
4450 | blend_pixel(s, src); |
4451 | // Then a naive blend with glyph shape |
4452 | *dst = rgbBlend(*dst, s, coverage); |
4453 | } else if (srcLinear.isOpaque()) { |
4454 | rgbBlendPixel(dst, coverage, srcLinear, colorProfile); |
4455 | } else { |
4456 | // First do naive blend with text-color |
4457 | QRgb s = *dst; |
4458 | blend_pixel(s, src); |
4459 | // Then gamma-corrected blend with glyph shape |
4460 | QRgba64 s64 = colorProfile ? colorProfile->toLinear64(s) : QRgba64::fromArgb32(s); |
4461 | rgbBlendPixel(dst, coverage, s64, colorProfile); |
4462 | } |
4463 | } |
4464 | |
4465 | #if QT_CONFIG(raster_64bit) |
4466 | static inline void rgbBlendPixel(QRgba64 &dst, int coverage, QRgba64 slinear, const QColorTrcLut *colorProfile) |
4467 | { |
4468 | // Do a gammacorrected RGB alphablend... |
4469 | const QRgba64 dlinear = colorProfile ? colorProfile->toLinear64(dst) : dst; |
4470 | |
4471 | QRgba64 blend = rgbBlend(dlinear, slinear, coverage); |
4472 | |
4473 | dst = colorProfile ? colorProfile->fromLinear(blend) : blend; |
4474 | } |
4475 | |
4476 | static inline void alphargbblend_generic(uint coverage, QRgba64 *dest, int x, const QRgba64 &srcLinear, const QRgba64 &src, const QColorTrcLut *colorProfile) |
4477 | { |
4478 | if (coverage == 0xff000000) { |
4479 | // nothing |
4480 | } else if (coverage == 0xffffffff) { |
4481 | blend_pixel(dest[x], src); |
4482 | } else if (!dest[x].isOpaque()) { |
4483 | // Do a gray alphablend. |
4484 | alphamapblend_generic(qRgbAvg(coverage), dest, x, srcLinear, src, colorProfile); |
4485 | } else if (src.isOpaque()) { |
4486 | rgbBlendPixel(dest[x], coverage, srcLinear, colorProfile); |
4487 | } else { |
4488 | // First do naive blend with text-color |
4489 | QRgba64 s = dest[x]; |
4490 | blend_pixel(s, src); |
4491 | // Then gamma-corrected blend with glyph shape |
4492 | if (colorProfile) |
4493 | s = colorProfile->toLinear(s); |
4494 | rgbBlendPixel(dest[x], coverage, s, colorProfile); |
4495 | } |
4496 | } |
4497 | |
4498 | static void qt_alphargbblit_generic(QRasterBuffer *rasterBuffer, |
4499 | int x, int y, const QRgba64 &color, |
4500 | const uint *src, int mapWidth, int mapHeight, int srcStride, |
4501 | const QClipData *clip, bool useGammaCorrection) |
4502 | { |
4503 | if (color.isTransparent()) |
4504 | return; |
4505 | |
4506 | const QColorTrcLut *colorProfile = nullptr; |
4507 | |
4508 | if (useGammaCorrection) |
4509 | colorProfile = QGuiApplicationPrivate::instance()->colorProfileForA32Text(); |
4510 | |
4511 | QRgba64 srcColor = color; |
4512 | if (colorProfile && color.isOpaque()) |
4513 | srcColor = colorProfile->toLinear(srcColor); |
4514 | |
4515 | alignas(8) QRgba64 buffer[BufferSize]; |
4516 | const DestFetchProc64 destFetch64 = destFetchProc64[rasterBuffer->format]; |
4517 | const DestStoreProc64 destStore64 = destStoreProc64[rasterBuffer->format]; |
4518 | |
4519 | if (!clip) { |
4520 | for (int ly = 0; ly < mapHeight; ++ly) { |
4521 | int i = x; |
4522 | int length = mapWidth; |
4523 | while (length > 0) { |
4524 | int l = qMin(BufferSize, length); |
4525 | QRgba64 *dest = destFetch64(buffer, rasterBuffer, i, y + ly, l); |
4526 | for (int j=0; j < l; ++j) { |
4527 | const uint coverage = src[j + (i - x)]; |
4528 | alphargbblend_generic(coverage, dest, j, srcColor, color, colorProfile); |
4529 | } |
4530 | if (destStore64) |
4531 | destStore64(rasterBuffer, i, y + ly, dest, l); |
4532 | length -= l; |
4533 | i += l; |
4534 | } |
4535 | src += srcStride; |
4536 | } |
4537 | } else { |
4538 | int bottom = qMin(y + mapHeight, rasterBuffer->height()); |
4539 | |
4540 | int top = qMax(y, 0); |
4541 | src += (top - y) * srcStride; |
4542 | |
4543 | const_cast<QClipData *>(clip)->initialize(); |
4544 | for (int yp = top; yp<bottom; ++yp) { |
4545 | const QClipData::ClipLine &line = clip->m_clipLines[yp]; |
4546 | |
4547 | for (int i=0; i<line.count; ++i) { |
4548 | const QSpan &clip = line.spans[i]; |
4549 | |
4550 | int start = qMax<int>(x, clip.x); |
4551 | int end = qMin<int>(x + mapWidth, clip.x + clip.len); |
4552 | if (end <= start) |
4553 | continue; |
4554 | Q_ASSERT(end - start <= BufferSize); |
4555 | QRgba64 *dest = destFetch64(buffer, rasterBuffer, start, clip.y, end - start); |
4556 | |
4557 | for (int xp=start; xp<end; ++xp) { |
4558 | const uint coverage = src[xp - x]; |
4559 | alphargbblend_generic(coverage, dest, xp - start, srcColor, color, colorProfile); |
4560 | } |
4561 | if (destStore64) |
4562 | destStore64(rasterBuffer, start, clip.y, dest, end - start); |
4563 | } // for (i -> line.count) |
4564 | src += srcStride; |
4565 | } // for (yp -> bottom) |
4566 | } |
4567 | } |
4568 | #else |
4569 | static void qt_alphargbblit_generic(QRasterBuffer *rasterBuffer, |
4570 | int x, int y, const QRgba64 &color, |
4571 | const uint *src, int mapWidth, int mapHeight, int srcStride, |
4572 | const QClipData *clip, bool useGammaCorrection) |
4573 | { |
4574 | if (color.isTransparent()) |
4575 | return; |
4576 | |
4577 | const quint32 c = color.toArgb32(); |
4578 | |
4579 | const QColorTrcLut *colorProfile = nullptr; |
4580 | |
4581 | if (useGammaCorrection) |
4582 | colorProfile = QGuiApplicationPrivate::instance()->colorProfileForA32Text(); |
4583 | |
4584 | QRgba64 srcColor = color; |
4585 | if (colorProfile && color.isOpaque()) |
4586 | srcColor = colorProfile->toLinear(srcColor); |
4587 | |
4588 | quint32 buffer[BufferSize]; |
4589 | const DestFetchProc destFetch = destFetchProc[rasterBuffer->format]; |
4590 | const DestStoreProc destStore = destStoreProc[rasterBuffer->format]; |
4591 | |
4592 | if (!clip) { |
4593 | for (int ly = 0; ly < mapHeight; ++ly) { |
4594 | int i = x; |
4595 | int length = mapWidth; |
4596 | while (length > 0) { |
4597 | int l = qMin(BufferSize, length); |
4598 | quint32 *dest = destFetch(buffer, rasterBuffer, i, y + ly, l); |
4599 | for (int j=0; j < l; ++j) { |
4600 | const uint coverage = src[j + (i - x)]; |
4601 | alphargbblend_argb32(dest + j, coverage, srcColor, c, colorProfile); |
4602 | } |
4603 | if (destStore) |
4604 | destStore(rasterBuffer, i, y + ly, dest, l); |
4605 | length -= l; |
4606 | i += l; |
4607 | } |
4608 | src += srcStride; |
4609 | } |
4610 | } else { |
4611 | int bottom = qMin(y + mapHeight, rasterBuffer->height()); |
4612 | |
4613 | int top = qMax(y, 0); |
4614 | src += (top - y) * srcStride; |
4615 | |
4616 | const_cast<QClipData *>(clip)->initialize(); |
4617 | for (int yp = top; yp<bottom; ++yp) { |
4618 | const QClipData::ClipLine &line = clip->m_clipLines[yp]; |
4619 | |
4620 | for (int i=0; i<line.count; ++i) { |
4621 | const QSpan &clip = line.spans[i]; |
4622 | |
4623 | int start = qMax<int>(x, clip.x); |
4624 | int end = qMin<int>(x + mapWidth, clip.x + clip.len); |
4625 | if (end <= start) |
4626 | continue; |
4627 | Q_ASSERT(end - start <= BufferSize); |
4628 | quint32 *dest = destFetch(buffer, rasterBuffer, start, clip.y, end - start); |
4629 | |
4630 | for (int xp=start; xp<end; ++xp) { |
4631 | const uint coverage = src[xp - x]; |
4632 | alphargbblend_argb32(dest + xp - start, coverage, srcColor, c, colorProfile); |
4633 | } |
4634 | if (destStore) |
4635 | destStore(rasterBuffer, start, clip.y, dest, end - start); |
4636 | } // for (i -> line.count) |
4637 | src += srcStride; |
4638 | } // for (yp -> bottom) |
4639 | } |
4640 | } |
4641 | #endif |
4642 | |
4643 | static void qt_alphargbblit_argb32(QRasterBuffer *rasterBuffer, |
4644 | int x, int y, const QRgba64 &color, |
4645 | const uint *src, int mapWidth, int mapHeight, int srcStride, |
4646 | const QClipData *clip, bool useGammaCorrection) |
4647 | { |
4648 | if (color.isTransparent()) |
4649 | return; |
4650 | |
4651 | const quint32 c = color.toArgb32(); |
4652 | |
4653 | const QColorTrcLut *colorProfile = nullptr; |
4654 | |
4655 | if (useGammaCorrection) |
4656 | colorProfile = QGuiApplicationPrivate::instance()->colorProfileForA32Text(); |
4657 | |
4658 | QRgba64 srcColor = color; |
4659 | if (colorProfile && color.isOpaque()) |
4660 | srcColor = colorProfile->toLinear(srcColor); |
4661 | |
4662 | if (!clip) { |
4663 | quint32 *dst = reinterpret_cast<quint32*>(rasterBuffer->scanLine(y)) + x; |
4664 | const int destStride = rasterBuffer->stride<quint32>(); |
4665 | while (mapHeight--) { |
4666 | for (int i = 0; i < mapWidth; ++i) { |
4667 | const uint coverage = src[i]; |
4668 | alphargbblend_argb32(dst + i, coverage, srcColor, c, colorProfile); |
4669 | } |
4670 | |
4671 | dst += destStride; |
4672 | src += srcStride; |
4673 | } |
4674 | } else { |
4675 | int bottom = qMin(y + mapHeight, rasterBuffer->height()); |
4676 | |
4677 | int top = qMax(y, 0); |
4678 | src += (top - y) * srcStride; |
4679 | |
4680 | const_cast<QClipData *>(clip)->initialize(); |
4681 | for (int yp = top; yp<bottom; ++yp) { |
4682 | const QClipData::ClipLine &line = clip->m_clipLines[yp]; |
4683 | |
4684 | quint32 *dst = reinterpret_cast<quint32 *>(rasterBuffer->scanLine(yp)); |
4685 | |
4686 | for (int i=0; i<line.count; ++i) { |
4687 | const QSpan &clip = line.spans[i]; |
4688 | |
4689 | int start = qMax<int>(x, clip.x); |
4690 | int end = qMin<int>(x + mapWidth, clip.x + clip.len); |
4691 | |
4692 | for (int xp=start; xp<end; ++xp) { |
4693 | const uint coverage = src[xp - x]; |
4694 | alphargbblend_argb32(dst + xp, coverage, srcColor, c, colorProfile); |
4695 | } |
4696 | } // for (i -> line.count) |
4697 | src += srcStride; |
4698 | } // for (yp -> bottom) |
4699 | |
4700 | } |
4701 | } |
4702 | |
4703 | static void qt_rectfill_argb32(QRasterBuffer *rasterBuffer, |
4704 | int x, int y, int width, int height, |
4705 | const QRgba64 &color) |
4706 | { |
4707 | qt_rectfill<quint32>(reinterpret_cast<quint32 *>(rasterBuffer->buffer()), |
4708 | color.toArgb32(), x, y, width, height, rasterBuffer->bytesPerLine()); |
4709 | } |
4710 | |
4711 | static void qt_rectfill_quint16(QRasterBuffer *rasterBuffer, |
4712 | int x, int y, int width, int height, |
4713 | const QRgba64 &color) |
4714 | { |
4715 | const QPixelLayout &layout = qPixelLayouts[rasterBuffer->format]; |
4716 | quint32 c32 = color.toArgb32(); |
4717 | quint16 c16; |
4718 | layout.storeFromARGB32PM(reinterpret_cast<uchar *>(&c16), &c32, 0, 1, nullptr, nullptr); |
4719 | qt_rectfill<quint16>(reinterpret_cast<quint16 *>(rasterBuffer->buffer()), |
4720 | c16, x, y, width, height, rasterBuffer->bytesPerLine()); |
4721 | } |
4722 | |
4723 | static void qt_rectfill_quint24(QRasterBuffer *rasterBuffer, |
4724 | int x, int y, int width, int height, |
4725 | const QRgba64 &color) |
4726 | { |
4727 | const QPixelLayout &layout = qPixelLayouts[rasterBuffer->format]; |
4728 | quint32 c32 = color.toArgb32(); |
4729 | quint24 c24; |
4730 | layout.storeFromARGB32PM(reinterpret_cast<uchar *>(&c24), &c32, 0, 1, nullptr, nullptr); |
4731 | qt_rectfill<quint24>(reinterpret_cast<quint24 *>(rasterBuffer->buffer()), |
4732 | c24, x, y, width, height, rasterBuffer->bytesPerLine()); |
4733 | } |
4734 | |
4735 | static void qt_rectfill_nonpremul_argb32(QRasterBuffer *rasterBuffer, |
4736 | int x, int y, int width, int height, |
4737 | const QRgba64 &color) |
4738 | { |
4739 | qt_rectfill<quint32>(reinterpret_cast<quint32 *>(rasterBuffer->buffer()), |
4740 | color.unpremultiplied().toArgb32(), x, y, width, height, rasterBuffer->bytesPerLine()); |
4741 | } |
4742 | |
4743 | static void qt_rectfill_rgba(QRasterBuffer *rasterBuffer, |
4744 | int x, int y, int width, int height, |
4745 | const QRgba64 &color) |
4746 | { |
4747 | qt_rectfill<quint32>(reinterpret_cast<quint32 *>(rasterBuffer->buffer()), |
4748 | ARGB2RGBA(color.toArgb32()), x, y, width, height, rasterBuffer->bytesPerLine()); |
4749 | } |
4750 | |
4751 | static void qt_rectfill_nonpremul_rgba(QRasterBuffer *rasterBuffer, |
4752 | int x, int y, int width, int height, |
4753 | const QRgba64 &color) |
4754 | { |
4755 | qt_rectfill<quint32>(reinterpret_cast<quint32 *>(rasterBuffer->buffer()), |
4756 | ARGB2RGBA(color.unpremultiplied().toArgb32()), x, y, width, height, rasterBuffer->bytesPerLine()); |
4757 | } |
4758 | |
4759 | template<QtPixelOrder PixelOrder> |
4760 | static void qt_rectfill_rgb30(QRasterBuffer *rasterBuffer, |
4761 | int x, int y, int width, int height, |
4762 | const QRgba64 &color) |
4763 | { |
4764 | qt_rectfill<quint32>(reinterpret_cast<quint32 *>(rasterBuffer->buffer()), |
4765 | qConvertRgb64ToRgb30<PixelOrder>(color), x, y, width, height, rasterBuffer->bytesPerLine()); |
4766 | } |
4767 | |
4768 | static void qt_rectfill_alpha(QRasterBuffer *rasterBuffer, |
4769 | int x, int y, int width, int height, |
4770 | const QRgba64 &color) |
4771 | { |
4772 | qt_rectfill<quint8>(reinterpret_cast<quint8 *>(rasterBuffer->buffer()), |
4773 | color.alpha() >> 8, x, y, width, height, rasterBuffer->bytesPerLine()); |
4774 | } |
4775 | |
4776 | static void qt_rectfill_gray(QRasterBuffer *rasterBuffer, |
4777 | int x, int y, int width, int height, |
4778 | const QRgba64 &color) |
4779 | { |
4780 | qt_rectfill<quint8>(reinterpret_cast<quint8 *>(rasterBuffer->buffer()), |
4781 | qGray(color.toArgb32()), x, y, width, height, rasterBuffer->bytesPerLine()); |
4782 | } |
4783 | |
4784 | static void qt_rectfill_quint64(QRasterBuffer *rasterBuffer, |
4785 | int x, int y, int width, int height, |
4786 | const QRgba64 &color) |
4787 | { |
4788 | const auto store = qStoreFromRGBA64PM[rasterBuffer->format]; |
4789 | quint64 c64; |
4790 | store(reinterpret_cast<uchar *>(&c64), &color, 0, 1, nullptr, nullptr); |
4791 | qt_rectfill<quint64>(reinterpret_cast<quint64 *>(rasterBuffer->buffer()), |
4792 | c64, x, y, width, height, rasterBuffer->bytesPerLine()); |
4793 | } |
4794 | |
4795 | // Map table for destination image format. Contains function pointers |
4796 | // for blends of various types unto the destination |
4797 | |
4798 | DrawHelper qDrawHelper[QImage::NImageFormats] = |
4799 | { |
4800 | // Format_Invalid, |
4801 | { nullptr, nullptr, nullptr, nullptr, nullptr }, |
4802 | // Format_Mono, |
4803 | { |
4804 | blend_color_generic, |
4805 | nullptr, nullptr, nullptr, nullptr |
4806 | }, |
4807 | // Format_MonoLSB, |
4808 | { |
4809 | blend_color_generic, |
4810 | nullptr, nullptr, nullptr, nullptr |
4811 | }, |
4812 | // Format_Indexed8, |
4813 | { |
4814 | blend_color_generic, |
4815 | nullptr, nullptr, nullptr, nullptr |
4816 | }, |
4817 | // Format_RGB32, |
4818 | { |
4819 | blend_color_argb, |
4820 | qt_bitmapblit_argb32, |
4821 | qt_alphamapblit_argb32, |
4822 | qt_alphargbblit_argb32, |
4823 | qt_rectfill_argb32 |
4824 | }, |
4825 | // Format_ARGB32, |
4826 | { |
4827 | blend_color_generic, |
4828 | qt_bitmapblit_argb32, |
4829 | qt_alphamapblit_argb32, |
4830 | qt_alphargbblit_argb32, |
4831 | qt_rectfill_nonpremul_argb32 |
4832 | }, |
4833 | // Format_ARGB32_Premultiplied |
4834 | { |
4835 | blend_color_argb, |
4836 | qt_bitmapblit_argb32, |
4837 | qt_alphamapblit_argb32, |
4838 | qt_alphargbblit_argb32, |
4839 | qt_rectfill_argb32 |
4840 | }, |
4841 | // Format_RGB16 |
4842 | { |
4843 | blend_color_rgb16, |
4844 | qt_bitmapblit_quint16, |
4845 | qt_alphamapblit_quint16, |
4846 | qt_alphargbblit_generic, |
4847 | qt_rectfill_quint16 |
4848 | }, |
4849 | // Format_ARGB8565_Premultiplied |
4850 | { |
4851 | blend_color_generic, |
4852 | nullptr, |
4853 | qt_alphamapblit_generic, |
4854 | qt_alphargbblit_generic, |
4855 | qt_rectfill_quint24 |
4856 | }, |
4857 | // Format_RGB666 |
4858 | { |
4859 | blend_color_generic, |
4860 | nullptr, |
4861 | qt_alphamapblit_generic, |
4862 | qt_alphargbblit_generic, |
4863 | qt_rectfill_quint24 |
4864 | }, |
4865 | // Format_ARGB6666_Premultiplied |
4866 | { |
4867 | blend_color_generic, |
4868 | nullptr, |
4869 | qt_alphamapblit_generic, |
4870 | qt_alphargbblit_generic, |
4871 | qt_rectfill_quint24 |
4872 | }, |
4873 | // Format_RGB555 |
4874 | { |
4875 | blend_color_generic, |
4876 | nullptr, |
4877 | qt_alphamapblit_generic, |
4878 | qt_alphargbblit_generic, |
4879 | qt_rectfill_quint16 |
4880 | }, |
4881 | // Format_ARGB8555_Premultiplied |
4882 | { |
4883 | blend_color_generic, |
4884 | nullptr, |
4885 | qt_alphamapblit_generic, |
4886 | qt_alphargbblit_generic, |
4887 | qt_rectfill_quint24 |
4888 | }, |
4889 | // Format_RGB888 |
4890 | { |
4891 | blend_color_generic, |
4892 | nullptr, |
4893 | qt_alphamapblit_generic, |
4894 | qt_alphargbblit_generic, |
4895 | qt_rectfill_quint24 |
4896 | }, |
4897 | // Format_RGB444 |
4898 | { |
4899 | blend_color_generic, |
4900 | nullptr, |
4901 | qt_alphamapblit_generic, |
4902 | qt_alphargbblit_generic, |
4903 | qt_rectfill_quint16 |
4904 | }, |
4905 | // Format_ARGB4444_Premultiplied |
4906 | { |
4907 | blend_color_generic, |
4908 | nullptr, |
4909 | qt_alphamapblit_generic, |
4910 | qt_alphargbblit_generic, |
4911 | qt_rectfill_quint16 |
4912 | }, |
4913 | // Format_RGBX8888 |
4914 | { |
4915 | blend_color_generic, |
4916 | qt_bitmapblit_rgba8888, |
4917 | qt_alphamapblit_generic, |
4918 | qt_alphargbblit_generic, |
4919 | qt_rectfill_rgba |
4920 | }, |
4921 | // Format_RGBA8888 |
4922 | { |
4923 | blend_color_generic, |
4924 | qt_bitmapblit_rgba8888, |
4925 | qt_alphamapblit_generic, |
4926 | qt_alphargbblit_generic, |
4927 | qt_rectfill_nonpremul_rgba |
4928 | }, |
4929 | // Format_RGB8888_Premultiplied |
4930 | { |
4931 | blend_color_generic, |
4932 | qt_bitmapblit_rgba8888, |
4933 | qt_alphamapblit_generic, |
4934 | qt_alphargbblit_generic, |
4935 | qt_rectfill_rgba |
4936 | }, |
4937 | // Format_BGR30 |
4938 | { |
4939 | blend_color_generic_rgb64, |
4940 | qt_bitmapblit_rgb30<PixelOrderBGR>, |
4941 | qt_alphamapblit_generic, |
4942 | qt_alphargbblit_generic, |
4943 | qt_rectfill_rgb30<PixelOrderBGR> |
4944 | }, |
4945 | // Format_A2BGR30_Premultiplied |
4946 | { |
4947 | blend_color_generic_rgb64, |
4948 | qt_bitmapblit_rgb30<PixelOrderBGR>, |
4949 | qt_alphamapblit_generic, |
4950 | qt_alphargbblit_generic, |
4951 | qt_rectfill_rgb30<PixelOrderBGR> |
4952 | }, |
4953 | // Format_RGB30 |
4954 | { |
4955 | blend_color_generic_rgb64, |
4956 | qt_bitmapblit_rgb30<PixelOrderRGB>, |
4957 | qt_alphamapblit_generic, |
4958 | qt_alphargbblit_generic, |
4959 | qt_rectfill_rgb30<PixelOrderRGB> |
4960 | }, |
4961 | // Format_A2RGB30_Premultiplied |
4962 | { |
4963 | blend_color_generic_rgb64, |
4964 | qt_bitmapblit_rgb30<PixelOrderRGB>, |
4965 | qt_alphamapblit_generic, |
4966 | qt_alphargbblit_generic, |
4967 | qt_rectfill_rgb30<PixelOrderRGB> |
4968 | }, |
4969 | // Format_Alpha8 |
4970 | { |
4971 | blend_color_generic, |
4972 | nullptr, |
4973 | qt_alphamapblit_generic, |
4974 | qt_alphargbblit_generic, |
4975 | qt_rectfill_alpha |
4976 | }, |
4977 | // Format_Grayscale8 |
4978 | { |
4979 | blend_color_generic, |
4980 | nullptr, |
4981 | qt_alphamapblit_generic, |
4982 | qt_alphargbblit_generic, |
4983 | qt_rectfill_gray |
4984 | }, |
4985 | // Format_RGBX64 |
4986 | { |
4987 | blend_color_generic_rgb64, |
4988 | nullptr, |
4989 | qt_alphamapblit_generic, |
4990 | qt_alphargbblit_generic, |
4991 | qt_rectfill_quint64 |
4992 | }, |
4993 | // Format_RGBA64 |
4994 | { |
4995 | blend_color_generic_rgb64, |
4996 | nullptr, |
4997 | qt_alphamapblit_generic, |
4998 | qt_alphargbblit_generic, |
4999 | qt_rectfill_quint64 |
5000 | }, |
5001 | // Format_RGBA64_Premultiplied |
5002 | { |
5003 | blend_color_generic_rgb64, |
5004 | nullptr, |
5005 | qt_alphamapblit_generic, |
5006 | qt_alphargbblit_generic, |
5007 | qt_rectfill_quint64 |
5008 | }, |
5009 | // Format_Grayscale16 |
5010 | { |
5011 | blend_color_generic_rgb64, |
5012 | nullptr, |
5013 | qt_alphamapblit_generic, |
5014 | qt_alphargbblit_generic, |
5015 | qt_rectfill_quint16 |
5016 | }, |
5017 | // Format_BGR888 |
5018 | { |
5019 | blend_color_generic, |
5020 | nullptr, |
5021 | qt_alphamapblit_generic, |
5022 | qt_alphargbblit_generic, |
5023 | qt_rectfill_quint24 |
5024 | }, |
5025 | }; |
5026 | |
5027 | #if !defined(__SSE2__) |
5028 | void qt_memfill64(quint64 *dest, quint64 color, qsizetype count) |
5029 | { |
5030 | qt_memfill_template<quint64>(dest, color, count); |
5031 | } |
5032 | #endif |
5033 | |
5034 | #if defined(QT_COMPILER_SUPPORTS_SSSE3) && defined(Q_CC_GNU) && !defined(Q_CC_INTEL) && !defined(Q_CC_CLANG) |
5035 | __attribute__((optimize("no-tree-vectorize" ))) |
5036 | #endif |
5037 | void qt_memfill24(quint24 *dest, quint24 color, qsizetype count) |
5038 | { |
5039 | # ifdef QT_COMPILER_SUPPORTS_SSSE3 |
5040 | extern void qt_memfill24_ssse3(quint24 *, quint24, qsizetype); |
5041 | if (qCpuHasFeature(SSSE3)) |
5042 | return qt_memfill24_ssse3(dest, color, count); |
5043 | # endif |
5044 | |
5045 | const quint32 v = color; |
5046 | quint24 *end = dest + count; |
5047 | |
5048 | // prolog: align dest to 32bit |
5049 | while ((quintptr(dest) & 0x3) && dest < end) { |
5050 | *dest++ = v; |
5051 | } |
5052 | if (dest >= end) |
5053 | return; |
5054 | |
5055 | const uint val1 = qFromBigEndian((v << 8) | (v >> 16)); |
5056 | const uint val2 = qFromBigEndian((v << 16) | (v >> 8)); |
5057 | const uint val3 = qFromBigEndian((v << 24) | (v >> 0)); |
5058 | |
5059 | for ( ; dest <= (end - 4); dest += 4) { |
5060 | quint32 *dst = reinterpret_cast<quint32 *>(dest); |
5061 | dst[0] = val1; |
5062 | dst[1] = val2; |
5063 | dst[2] = val3; |
5064 | } |
5065 | |
5066 | // less than 4px left |
5067 | switch (end - dest) { |
5068 | case 3: |
5069 | *dest++ = v; |
5070 | Q_FALLTHROUGH(); |
5071 | case 2: |
5072 | *dest++ = v; |
5073 | Q_FALLTHROUGH(); |
5074 | case 1: |
5075 | *dest++ = v; |
5076 | } |
5077 | } |
5078 | |
5079 | void qt_memfill16(quint16 *dest, quint16 value, qsizetype count) |
5080 | { |
5081 | const int align = quintptr(dest) & 0x3; |
5082 | if (align) { |
5083 | *dest++ = value; |
5084 | --count; |
5085 | } |
5086 | |
5087 | if (count & 0x1) |
5088 | dest[count - 1] = value; |
5089 | |
5090 | const quint32 value32 = (value << 16) | value; |
5091 | qt_memfill32(reinterpret_cast<quint32*>(dest), value32, count / 2); |
5092 | } |
5093 | |
5094 | #if !defined(__SSE2__) && !defined(__ARM_NEON__) && !defined(__MIPS_DSP__) |
5095 | void qt_memfill32(quint32 *dest, quint32 color, qsizetype count) |
5096 | { |
5097 | qt_memfill_template<quint32>(dest, color, count); |
5098 | } |
5099 | #endif |
5100 | #ifdef __SSE2__ |
5101 | decltype(qt_memfill32_sse2) *qt_memfill32 = nullptr; |
5102 | decltype(qt_memfill64_sse2) *qt_memfill64 = nullptr; |
5103 | #endif |
5104 | |
5105 | #ifdef QT_COMPILER_SUPPORTS_SSE4_1 |
5106 | template<QtPixelOrder> void QT_FASTCALL storeA2RGB30PMFromARGB32PM_sse4(uchar *dest, const uint *src, int index, int count, const QList<QRgb> *, QDitherInfo *); |
5107 | #endif |
5108 | |
5109 | extern void qInitBlendFunctions(); |
5110 | |
5111 | static void qInitDrawhelperFunctions() |
5112 | { |
5113 | // Set up basic blend function tables. |
5114 | qInitBlendFunctions(); |
5115 | |
5116 | #ifdef __SSE2__ |
5117 | # ifndef __AVX2__ |
5118 | qt_memfill32 = qt_memfill32_sse2; |
5119 | qt_memfill64 = qt_memfill64_sse2; |
5120 | # endif |
5121 | qDrawHelper[QImage::Format_RGB32].bitmapBlit = qt_bitmapblit32_sse2; |
5122 | qDrawHelper[QImage::Format_ARGB32].bitmapBlit = qt_bitmapblit32_sse2; |
5123 | qDrawHelper[QImage::Format_ARGB32_Premultiplied].bitmapBlit = qt_bitmapblit32_sse2; |
5124 | qDrawHelper[QImage::Format_RGB16].bitmapBlit = qt_bitmapblit16_sse2; |
5125 | qDrawHelper[QImage::Format_RGBX8888].bitmapBlit = qt_bitmapblit8888_sse2; |
5126 | qDrawHelper[QImage::Format_RGBA8888].bitmapBlit = qt_bitmapblit8888_sse2; |
5127 | qDrawHelper[QImage::Format_RGBA8888_Premultiplied].bitmapBlit = qt_bitmapblit8888_sse2; |
5128 | |
5129 | extern void qt_scale_image_argb32_on_argb32_sse2(uchar *destPixels, int dbpl, |
5130 | const uchar *srcPixels, int sbpl, int srch, |
5131 | const QRectF &targetRect, |
5132 | const QRectF &sourceRect, |
5133 | const QRect &clip, |
5134 | int const_alpha); |
5135 | qScaleFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_scale_image_argb32_on_argb32_sse2; |
5136 | qScaleFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_scale_image_argb32_on_argb32_sse2; |
5137 | qScaleFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBA8888_Premultiplied] = qt_scale_image_argb32_on_argb32_sse2; |
5138 | qScaleFunctions[QImage::Format_RGBX8888][QImage::Format_RGBA8888_Premultiplied] = qt_scale_image_argb32_on_argb32_sse2; |
5139 | |
5140 | extern void qt_blend_rgb32_on_rgb32_sse2(uchar *destPixels, int dbpl, |
5141 | const uchar *srcPixels, int sbpl, |
5142 | int w, int h, |
5143 | int const_alpha); |
5144 | extern void qt_blend_argb32_on_argb32_sse2(uchar *destPixels, int dbpl, |
5145 | const uchar *srcPixels, int sbpl, |
5146 | int w, int h, |
5147 | int const_alpha); |
5148 | |
5149 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_sse2; |
5150 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_sse2; |
5151 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_sse2; |
5152 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_sse2; |
5153 | qBlendFunctions[QImage::Format_RGBX8888][QImage::Format_RGBX8888] = qt_blend_rgb32_on_rgb32_sse2; |
5154 | qBlendFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBX8888] = qt_blend_rgb32_on_rgb32_sse2; |
5155 | qBlendFunctions[QImage::Format_RGBX8888][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_sse2; |
5156 | qBlendFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_sse2; |
5157 | |
5158 | extern const uint * QT_FASTCALL qt_fetch_radial_gradient_sse2(uint *buffer, const Operator *op, const QSpanData *data, |
5159 | int y, int x, int length); |
5160 | |
5161 | qt_fetch_radial_gradient = qt_fetch_radial_gradient_sse2; |
5162 | |
5163 | extern void QT_FASTCALL comp_func_SourceOver_sse2(uint *destPixels, const uint *srcPixels, int length, uint const_alpha); |
5164 | extern void QT_FASTCALL comp_func_solid_SourceOver_sse2(uint *destPixels, int length, uint color, uint const_alpha); |
5165 | extern void QT_FASTCALL comp_func_Source_sse2(uint *destPixels, const uint *srcPixels, int length, uint const_alpha); |
5166 | extern void QT_FASTCALL comp_func_solid_Source_sse2(uint *destPixels, int length, uint color, uint const_alpha); |
5167 | extern void QT_FASTCALL comp_func_Plus_sse2(uint *destPixels, const uint *srcPixels, int length, uint const_alpha); |
5168 | qt_functionForMode_C[QPainter::CompositionMode_SourceOver] = comp_func_SourceOver_sse2; |
5169 | qt_functionForModeSolid_C[QPainter::CompositionMode_SourceOver] = comp_func_solid_SourceOver_sse2; |
5170 | qt_functionForMode_C[QPainter::CompositionMode_Source] = comp_func_Source_sse2; |
5171 | qt_functionForModeSolid_C[QPainter::CompositionMode_Source] = comp_func_solid_Source_sse2; |
5172 | qt_functionForMode_C[QPainter::CompositionMode_Plus] = comp_func_Plus_sse2; |
5173 | |
5174 | #ifdef QT_COMPILER_SUPPORTS_SSSE3 |
5175 | if (qCpuHasFeature(SSSE3)) { |
5176 | extern void qt_blend_argb32_on_argb32_ssse3(uchar *destPixels, int dbpl, |
5177 | const uchar *srcPixels, int sbpl, |
5178 | int w, int h, |
5179 | int const_alpha); |
5180 | |
5181 | extern const uint * QT_FASTCALL qt_fetchUntransformed_888_ssse3(uint *buffer, const Operator *, const QSpanData *data, |
5182 | int y, int x, int length); |
5183 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_ssse3; |
5184 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_ssse3; |
5185 | qBlendFunctions[QImage::Format_RGBX8888][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_ssse3; |
5186 | qBlendFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_ssse3; |
5187 | sourceFetchUntransformed[QImage::Format_RGB888] = qt_fetchUntransformed_888_ssse3; |
5188 | extern void QT_FASTCALL rbSwap_888_ssse3(uchar *dst, const uchar *src, int count); |
5189 | qPixelLayouts[QImage::Format_RGB888].rbSwap = rbSwap_888_ssse3; |
5190 | qPixelLayouts[QImage::Format_BGR888].rbSwap = rbSwap_888_ssse3; |
5191 | } |
5192 | #endif // SSSE3 |
5193 | |
5194 | #if defined(QT_COMPILER_SUPPORTS_SSE4_1) |
5195 | if (qCpuHasFeature(SSE4_1)) { |
5196 | extern void QT_FASTCALL convertARGB32ToARGB32PM_sse4(uint *buffer, int count, const QList<QRgb> *); |
5197 | extern void QT_FASTCALL convertRGBA8888ToARGB32PM_sse4(uint *buffer, int count, const QList<QRgb> *); |
5198 | extern const uint *QT_FASTCALL fetchARGB32ToARGB32PM_sse4(uint *buffer, const uchar *src, int index, int count, |
5199 | const QList<QRgb> *, QDitherInfo *); |
5200 | extern const uint *QT_FASTCALL fetchRGBA8888ToARGB32PM_sse4(uint *buffer, const uchar *src, int index, int count, |
5201 | const QList<QRgb> *, QDitherInfo *); |
5202 | extern const QRgba64 * QT_FASTCALL convertARGB32ToRGBA64PM_sse4(QRgba64 *buffer, const uint *src, int count, |
5203 | const QList<QRgb> *, QDitherInfo *); |
5204 | extern const QRgba64 * QT_FASTCALL convertRGBA8888ToRGBA64PM_sse4(QRgba64 *buffer, const uint *src, int count, |
5205 | const QList<QRgb> *, QDitherInfo *); |
5206 | extern const QRgba64 *QT_FASTCALL fetchARGB32ToRGBA64PM_sse4(QRgba64 *buffer, const uchar *src, int index, int count, |
5207 | const QList<QRgb> *, QDitherInfo *); |
5208 | extern const QRgba64 *QT_FASTCALL fetchRGBA8888ToRGBA64PM_sse4(QRgba64 *buffer, const uchar *src, int index, int count, |
5209 | const QList<QRgb> *, QDitherInfo *); |
5210 | extern void QT_FASTCALL storeARGB32FromARGB32PM_sse4(uchar *dest, const uint *src, int index, int count, |
5211 | const QList<QRgb> *, QDitherInfo *); |
5212 | extern void QT_FASTCALL storeRGBA8888FromARGB32PM_sse4(uchar *dest, const uint *src, int index, int count, |
5213 | const QList<QRgb> *, QDitherInfo *); |
5214 | extern void QT_FASTCALL storeRGBXFromARGB32PM_sse4(uchar *dest, const uint *src, int index, int count, |
5215 | const QList<QRgb> *, QDitherInfo *); |
5216 | extern void QT_FASTCALL storeARGB32FromRGBA64PM_sse4(uchar *dest, const QRgba64 *src, int index, int count, |
5217 | const QList<QRgb> *, QDitherInfo *); |
5218 | extern void QT_FASTCALL storeRGBA8888FromRGBA64PM_sse4(uchar *dest, const QRgba64 *src, int index, int count, |
5219 | const QList<QRgb> *, QDitherInfo *); |
5220 | extern void QT_FASTCALL destStore64ARGB32_sse4(QRasterBuffer *rasterBuffer, int x, int y, const QRgba64 *buffer, int length); |
5221 | extern void QT_FASTCALL destStore64RGBA8888_sse4(QRasterBuffer *rasterBuffer, int x, int y, const QRgba64 *buffer, int length); |
5222 | # ifndef __AVX2__ |
5223 | qPixelLayouts[QImage::Format_ARGB32].fetchToARGB32PM = fetchARGB32ToARGB32PM_sse4; |
5224 | qPixelLayouts[QImage::Format_ARGB32].convertToARGB32PM = convertARGB32ToARGB32PM_sse4; |
5225 | qPixelLayouts[QImage::Format_RGBA8888].fetchToARGB32PM = fetchRGBA8888ToARGB32PM_sse4; |
5226 | qPixelLayouts[QImage::Format_RGBA8888].convertToARGB32PM = convertRGBA8888ToARGB32PM_sse4; |
5227 | qPixelLayouts[QImage::Format_ARGB32].fetchToRGBA64PM = fetchARGB32ToRGBA64PM_sse4; |
5228 | qPixelLayouts[QImage::Format_ARGB32].convertToRGBA64PM = convertARGB32ToRGBA64PM_sse4; |
5229 | qPixelLayouts[QImage::Format_RGBA8888].fetchToRGBA64PM = fetchRGBA8888ToRGBA64PM_sse4; |
5230 | qPixelLayouts[QImage::Format_RGBA8888].convertToRGBA64PM = convertRGBA8888ToRGBA64PM_sse4; |
5231 | qPixelLayouts[QImage::Format_RGBX8888].fetchToRGBA64PM = fetchRGBA8888ToRGBA64PM_sse4; |
5232 | qPixelLayouts[QImage::Format_RGBX8888].convertToRGBA64PM = convertRGBA8888ToRGBA64PM_sse4; |
5233 | # endif |
5234 | qPixelLayouts[QImage::Format_ARGB32].storeFromARGB32PM = storeARGB32FromARGB32PM_sse4; |
5235 | qPixelLayouts[QImage::Format_RGBA8888].storeFromARGB32PM = storeRGBA8888FromARGB32PM_sse4; |
5236 | qPixelLayouts[QImage::Format_RGBX8888].storeFromARGB32PM = storeRGBXFromARGB32PM_sse4; |
5237 | qPixelLayouts[QImage::Format_A2BGR30_Premultiplied].storeFromARGB32PM = storeA2RGB30PMFromARGB32PM_sse4<PixelOrderBGR>; |
5238 | qPixelLayouts[QImage::Format_A2RGB30_Premultiplied].storeFromARGB32PM = storeA2RGB30PMFromARGB32PM_sse4<PixelOrderRGB>; |
5239 | qStoreFromRGBA64PM[QImage::Format_ARGB32] = storeARGB32FromRGBA64PM_sse4; |
5240 | qStoreFromRGBA64PM[QImage::Format_RGBA8888] = storeRGBA8888FromRGBA64PM_sse4; |
5241 | #if QT_CONFIG(raster_64bit) |
5242 | destStoreProc64[QImage::Format_ARGB32] = destStore64ARGB32_sse4; |
5243 | destStoreProc64[QImage::Format_RGBA8888] = destStore64RGBA8888_sse4; |
5244 | #endif |
5245 | } |
5246 | #endif |
5247 | |
5248 | #if defined(QT_COMPILER_SUPPORTS_AVX2) |
5249 | if (qCpuHasFeature(ArchHaswell)) { |
5250 | qt_memfill32 = qt_memfill32_avx2; |
5251 | qt_memfill64 = qt_memfill64_avx2; |
5252 | extern void qt_blend_rgb32_on_rgb32_avx2(uchar *destPixels, int dbpl, |
5253 | const uchar *srcPixels, int sbpl, |
5254 | int w, int h, int const_alpha); |
5255 | extern void qt_blend_argb32_on_argb32_avx2(uchar *destPixels, int dbpl, |
5256 | const uchar *srcPixels, int sbpl, |
5257 | int w, int h, int const_alpha); |
5258 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_avx2; |
5259 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_avx2; |
5260 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_avx2; |
5261 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_avx2; |
5262 | qBlendFunctions[QImage::Format_RGBX8888][QImage::Format_RGBX8888] = qt_blend_rgb32_on_rgb32_avx2; |
5263 | qBlendFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBX8888] = qt_blend_rgb32_on_rgb32_avx2; |
5264 | qBlendFunctions[QImage::Format_RGBX8888][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_avx2; |
5265 | qBlendFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_avx2; |
5266 | |
5267 | extern void QT_FASTCALL comp_func_Source_avx2(uint *destPixels, const uint *srcPixels, int length, uint const_alpha); |
5268 | extern void QT_FASTCALL comp_func_SourceOver_avx2(uint *destPixels, const uint *srcPixels, int length, uint const_alpha); |
5269 | extern void QT_FASTCALL comp_func_solid_SourceOver_avx2(uint *destPixels, int length, uint color, uint const_alpha); |
5270 | qt_functionForMode_C[QPainter::CompositionMode_Source] = comp_func_Source_avx2; |
5271 | qt_functionForMode_C[QPainter::CompositionMode_SourceOver] = comp_func_SourceOver_avx2; |
5272 | qt_functionForModeSolid_C[QPainter::CompositionMode_SourceOver] = comp_func_solid_SourceOver_avx2; |
5273 | #if QT_CONFIG(raster_64bit) |
5274 | extern void QT_FASTCALL comp_func_Source_rgb64_avx2(QRgba64 *destPixels, const QRgba64 *srcPixels, int length, uint const_alpha); |
5275 | extern void QT_FASTCALL comp_func_SourceOver_rgb64_avx2(QRgba64 *destPixels, const QRgba64 *srcPixels, int length, uint const_alpha); |
5276 | extern void QT_FASTCALL comp_func_solid_SourceOver_rgb64_avx2(QRgba64 *destPixels, int length, QRgba64 color, uint const_alpha); |
5277 | qt_functionForMode64_C[QPainter::CompositionMode_Source] = comp_func_Source_rgb64_avx2; |
5278 | qt_functionForMode64_C[QPainter::CompositionMode_SourceOver] = comp_func_SourceOver_rgb64_avx2; |
5279 | qt_functionForModeSolid64_C[QPainter::CompositionMode_SourceOver] = comp_func_solid_SourceOver_rgb64_avx2; |
5280 | #endif |
5281 | |
5282 | extern void QT_FASTCALL fetchTransformedBilinearARGB32PM_simple_scale_helper_avx2(uint *b, uint *end, const QTextureData &image, |
5283 | int &fx, int &fy, int fdx, int /*fdy*/); |
5284 | extern void QT_FASTCALL fetchTransformedBilinearARGB32PM_downscale_helper_avx2(uint *b, uint *end, const QTextureData &image, |
5285 | int &fx, int &fy, int fdx, int /*fdy*/); |
5286 | extern void QT_FASTCALL fetchTransformedBilinearARGB32PM_fast_rotate_helper_avx2(uint *b, uint *end, const QTextureData &image, |
5287 | int &fx, int &fy, int fdx, int fdy); |
5288 | |
5289 | bilinearFastTransformHelperARGB32PM[0][SimpleScaleTransform] = fetchTransformedBilinearARGB32PM_simple_scale_helper_avx2; |
5290 | bilinearFastTransformHelperARGB32PM[0][DownscaleTransform] = fetchTransformedBilinearARGB32PM_downscale_helper_avx2; |
5291 | bilinearFastTransformHelperARGB32PM[0][FastRotateTransform] = fetchTransformedBilinearARGB32PM_fast_rotate_helper_avx2; |
5292 | |
5293 | extern void QT_FASTCALL convertARGB32ToARGB32PM_avx2(uint *buffer, int count, const QList<QRgb> *); |
5294 | extern void QT_FASTCALL convertRGBA8888ToARGB32PM_avx2(uint *buffer, int count, const QList<QRgb> *); |
5295 | extern const uint *QT_FASTCALL fetchARGB32ToARGB32PM_avx2(uint *buffer, const uchar *src, int index, int count, |
5296 | const QList<QRgb> *, QDitherInfo *); |
5297 | extern const uint *QT_FASTCALL fetchRGBA8888ToARGB32PM_avx2(uint *buffer, const uchar *src, int index, int count, |
5298 | const QList<QRgb> *, QDitherInfo *); |
5299 | qPixelLayouts[QImage::Format_ARGB32].fetchToARGB32PM = fetchARGB32ToARGB32PM_avx2; |
5300 | qPixelLayouts[QImage::Format_ARGB32].convertToARGB32PM = convertARGB32ToARGB32PM_avx2; |
5301 | qPixelLayouts[QImage::Format_RGBA8888].fetchToARGB32PM = fetchRGBA8888ToARGB32PM_avx2; |
5302 | qPixelLayouts[QImage::Format_RGBA8888].convertToARGB32PM = convertRGBA8888ToARGB32PM_avx2; |
5303 | |
5304 | #if QT_CONFIG(raster_64bit) |
5305 | extern const QRgba64 * QT_FASTCALL convertARGB32ToRGBA64PM_avx2(QRgba64 *, const uint *, int, const QList<QRgb> *, QDitherInfo *); |
5306 | extern const QRgba64 * QT_FASTCALL convertRGBA8888ToRGBA64PM_avx2(QRgba64 *, const uint *, int count, const QList<QRgb> *, QDitherInfo *); |
5307 | extern const QRgba64 *QT_FASTCALL fetchARGB32ToRGBA64PM_avx2(QRgba64 *, const uchar *, int, int, const QList<QRgb> *, QDitherInfo *); |
5308 | extern const QRgba64 *QT_FASTCALL fetchRGBA8888ToRGBA64PM_avx2(QRgba64 *, const uchar *, int, int, const QList<QRgb> *, QDitherInfo *); |
5309 | qPixelLayouts[QImage::Format_ARGB32].convertToRGBA64PM = convertARGB32ToRGBA64PM_avx2; |
5310 | qPixelLayouts[QImage::Format_RGBX8888].convertToRGBA64PM = convertRGBA8888ToRGBA64PM_avx2; |
5311 | qPixelLayouts[QImage::Format_ARGB32].fetchToRGBA64PM = fetchARGB32ToRGBA64PM_avx2; |
5312 | qPixelLayouts[QImage::Format_RGBX8888].fetchToRGBA64PM = fetchRGBA8888ToRGBA64PM_avx2; |
5313 | #endif |
5314 | } |
5315 | #endif |
5316 | |
5317 | #endif // SSE2 |
5318 | |
5319 | #if defined(__ARM_NEON__) |
5320 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_neon; |
5321 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_neon; |
5322 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_neon; |
5323 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_neon; |
5324 | #if Q_BYTE_ORDER == Q_LITTLE_ENDIAN |
5325 | qBlendFunctions[QImage::Format_RGBX8888][QImage::Format_RGBX8888] = qt_blend_rgb32_on_rgb32_neon; |
5326 | qBlendFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBX8888] = qt_blend_rgb32_on_rgb32_neon; |
5327 | qBlendFunctions[QImage::Format_RGBX8888][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_neon; |
5328 | qBlendFunctions[QImage::Format_RGBA8888_Premultiplied][QImage::Format_RGBA8888_Premultiplied] = qt_blend_argb32_on_argb32_neon; |
5329 | #endif |
5330 | |
5331 | qt_functionForMode_C[QPainter::CompositionMode_SourceOver] = qt_blend_argb32_on_argb32_scanline_neon; |
5332 | qt_functionForModeSolid_C[QPainter::CompositionMode_SourceOver] = comp_func_solid_SourceOver_neon; |
5333 | qt_functionForMode_C[QPainter::CompositionMode_Plus] = comp_func_Plus_neon; |
5334 | |
5335 | extern const uint * QT_FASTCALL qt_fetch_radial_gradient_neon(uint *buffer, const Operator *op, const QSpanData *data, |
5336 | int y, int x, int length); |
5337 | |
5338 | qt_fetch_radial_gradient = qt_fetch_radial_gradient_neon; |
5339 | |
5340 | sourceFetchUntransformed[QImage::Format_RGB888] = qt_fetchUntransformed_888_neon; |
5341 | |
5342 | #if Q_BYTE_ORDER == Q_LITTLE_ENDIAN |
5343 | extern void QT_FASTCALL convertARGB32ToARGB32PM_neon(uint *buffer, int count, const QList<QRgb> *); |
5344 | extern void QT_FASTCALL convertRGBA8888ToARGB32PM_neon(uint *buffer, int count, const QList<QRgb> *); |
5345 | extern const uint *QT_FASTCALL fetchARGB32ToARGB32PM_neon(uint *buffer, const uchar *src, int index, int count, |
5346 | const QList<QRgb> *, QDitherInfo *); |
5347 | extern const uint *QT_FASTCALL fetchRGBA8888ToARGB32PM_neon(uint *buffer, const uchar *src, int index, int count, |
5348 | const QList<QRgb> *, QDitherInfo *); |
5349 | extern const QRgba64 * QT_FASTCALL convertARGB32ToRGBA64PM_neon(QRgba64 *buffer, const uint *src, int count, |
5350 | const QList<QRgb> *, QDitherInfo *); |
5351 | extern const QRgba64 * QT_FASTCALL convertRGBA8888ToRGBA64PM_neon(QRgba64 *buffer, const uint *src, int count, |
5352 | const QList<QRgb> *, QDitherInfo *); |
5353 | extern const QRgba64 *QT_FASTCALL fetchARGB32ToRGBA64PM_neon(QRgba64 *buffer, const uchar *src, int index, int count, |
5354 | const QList<QRgb> *, QDitherInfo *); |
5355 | extern const QRgba64 *QT_FASTCALL fetchRGBA8888ToRGBA64PM_neon(QRgba64 *buffer, const uchar *src, int index, int count, |
5356 | const QList<QRgb> *, QDitherInfo *); |
5357 | extern void QT_FASTCALL storeARGB32FromARGB32PM_neon(uchar *dest, const uint *src, int index, int count, |
5358 | const QList<QRgb> *, QDitherInfo *); |
5359 | extern void QT_FASTCALL storeRGBA8888FromARGB32PM_neon(uchar *dest, const uint *src, int index, int count, |
5360 | const QList<QRgb> *, QDitherInfo *); |
5361 | extern void QT_FASTCALL storeRGBXFromARGB32PM_neon(uchar *dest, const uint *src, int index, int count, |
5362 | const QList<QRgb> *, QDitherInfo *); |
5363 | qPixelLayouts[QImage::Format_ARGB32].fetchToARGB32PM = fetchARGB32ToARGB32PM_neon; |
5364 | qPixelLayouts[QImage::Format_ARGB32].convertToARGB32PM = convertARGB32ToARGB32PM_neon; |
5365 | qPixelLayouts[QImage::Format_ARGB32].storeFromARGB32PM = storeARGB32FromARGB32PM_neon; |
5366 | qPixelLayouts[QImage::Format_ARGB32].fetchToRGBA64PM = fetchARGB32ToRGBA64PM_neon; |
5367 | qPixelLayouts[QImage::Format_ARGB32].convertToRGBA64PM = convertARGB32ToRGBA64PM_neon; |
5368 | qPixelLayouts[QImage::Format_RGBA8888].fetchToARGB32PM = fetchRGBA8888ToARGB32PM_neon; |
5369 | qPixelLayouts[QImage::Format_RGBA8888].convertToARGB32PM = convertRGBA8888ToARGB32PM_neon; |
5370 | qPixelLayouts[QImage::Format_RGBA8888].storeFromARGB32PM = storeRGBA8888FromARGB32PM_neon; |
5371 | qPixelLayouts[QImage::Format_RGBA8888].fetchToRGBA64PM = fetchRGBA8888ToRGBA64PM_neon; |
5372 | qPixelLayouts[QImage::Format_RGBA8888].convertToRGBA64PM = convertRGBA8888ToRGBA64PM_neon; |
5373 | qPixelLayouts[QImage::Format_RGBX8888].storeFromARGB32PM = storeRGBXFromARGB32PM_neon; |
5374 | qPixelLayouts[QImage::Format_RGBX8888].fetchToRGBA64PM = fetchRGBA8888ToRGBA64PM_neon; |
5375 | qPixelLayouts[QImage::Format_RGBX8888].convertToRGBA64PM = convertRGBA8888ToRGBA64PM_neon; |
5376 | #endif |
5377 | |
5378 | #if defined(ENABLE_PIXMAN_DRAWHELPERS) |
5379 | // The RGB16 helpers are using Arm32 assemblythat has not been ported to AArch64 |
5380 | qBlendFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_rgb16_neon; |
5381 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB16] = qt_blend_rgb16_on_argb32_neon; |
5382 | qBlendFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_blend_rgb16_on_rgb16_neon; |
5383 | |
5384 | qScaleFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_scale_image_argb32_on_rgb16_neon; |
5385 | qScaleFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_scale_image_rgb16_on_rgb16_neon; |
5386 | |
5387 | qTransformFunctions[QImage::Format_RGB16][QImage::Format_ARGB32_Premultiplied] = qt_transform_image_argb32_on_rgb16_neon; |
5388 | qTransformFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_transform_image_rgb16_on_rgb16_neon; |
5389 | |
5390 | qDrawHelper[QImage::Format_RGB16].alphamapBlit = qt_alphamapblit_quint16_neon; |
5391 | |
5392 | destFetchProc[QImage::Format_RGB16] = qt_destFetchRGB16_neon; |
5393 | destStoreProc[QImage::Format_RGB16] = qt_destStoreRGB16_neon; |
5394 | |
5395 | qMemRotateFunctions[QPixelLayout::BPP16][0] = qt_memrotate90_16_neon; |
5396 | qMemRotateFunctions[QPixelLayout::BPP16][2] = qt_memrotate270_16_neon; |
5397 | #endif |
5398 | #endif // defined(__ARM_NEON__) |
5399 | |
5400 | #if defined(__MIPS_DSP__) |
5401 | // Composition functions are all DSP r1 |
5402 | qt_functionForMode_C[QPainter::CompositionMode_SourceOver] = comp_func_SourceOver_asm_mips_dsp; |
5403 | qt_functionForMode_C[QPainter::CompositionMode_Source] = comp_func_Source_mips_dsp; |
5404 | qt_functionForMode_C[QPainter::CompositionMode_DestinationOver] = comp_func_DestinationOver_mips_dsp; |
5405 | qt_functionForMode_C[QPainter::CompositionMode_SourceIn] = comp_func_SourceIn_mips_dsp; |
5406 | qt_functionForMode_C[QPainter::CompositionMode_DestinationIn] = comp_func_DestinationIn_mips_dsp; |
5407 | qt_functionForMode_C[QPainter::CompositionMode_DestinationOut] = comp_func_DestinationOut_mips_dsp; |
5408 | qt_functionForMode_C[QPainter::CompositionMode_SourceAtop] = comp_func_SourceAtop_mips_dsp; |
5409 | qt_functionForMode_C[QPainter::CompositionMode_DestinationAtop] = comp_func_DestinationAtop_mips_dsp; |
5410 | qt_functionForMode_C[QPainter::CompositionMode_Xor] = comp_func_XOR_mips_dsp; |
5411 | qt_functionForMode_C[QPainter::CompositionMode_SourceOut] = comp_func_SourceOut_mips_dsp; |
5412 | |
5413 | qt_functionForModeSolid_C[QPainter::CompositionMode_SourceOver] = comp_func_solid_SourceOver_mips_dsp; |
5414 | qt_functionForModeSolid_C[QPainter::CompositionMode_DestinationOver] = comp_func_solid_DestinationOver_mips_dsp; |
5415 | qt_functionForModeSolid_C[QPainter::CompositionMode_SourceIn] = comp_func_solid_SourceIn_mips_dsp; |
5416 | qt_functionForModeSolid_C[QPainter::CompositionMode_DestinationIn] = comp_func_solid_DestinationIn_mips_dsp; |
5417 | qt_functionForModeSolid_C[QPainter::CompositionMode_SourceAtop] = comp_func_solid_SourceAtop_mips_dsp; |
5418 | qt_functionForModeSolid_C[QPainter::CompositionMode_DestinationAtop] = comp_func_solid_DestinationAtop_mips_dsp; |
5419 | qt_functionForModeSolid_C[QPainter::CompositionMode_Xor] = comp_func_solid_XOR_mips_dsp; |
5420 | qt_functionForModeSolid_C[QPainter::CompositionMode_SourceOut] = comp_func_solid_SourceOut_mips_dsp; |
5421 | |
5422 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_mips_dsp; |
5423 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_mips_dsp; |
5424 | qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_mips_dsp; |
5425 | qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_mips_dsp; |
5426 | |
5427 | destFetchProc[QImage::Format_ARGB32] = qt_destFetchARGB32_mips_dsp; |
5428 | |
5429 | destStoreProc[QImage::Format_ARGB32] = qt_destStoreARGB32_mips_dsp; |
5430 | |
5431 | sourceFetchUntransformed[QImage::Format_RGB888] = qt_fetchUntransformed_888_mips_dsp; |
5432 | sourceFetchUntransformed[QImage::Format_RGB444] = qt_fetchUntransformed_444_mips_dsp; |
5433 | sourceFetchUntransformed[QImage::Format_ARGB8565_Premultiplied] = qt_fetchUntransformed_argb8565_premultiplied_mips_dsp; |
5434 | |
5435 | #if defined(__MIPS_DSPR2__) |
5436 | qBlendFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_blend_rgb16_on_rgb16_mips_dspr2; |
5437 | sourceFetchUntransformed[QImage::Format_RGB16] = qt_fetchUntransformedRGB16_mips_dspr2; |
5438 | #else |
5439 | qBlendFunctions[QImage::Format_RGB16][QImage::Format_RGB16] = qt_blend_rgb16_on_rgb16_mips_dsp; |
5440 | #endif // defined(__MIPS_DSPR2__) |
5441 | #endif // defined(__MIPS_DSP__) |
5442 | } |
5443 | |
5444 | // Ensure initialization if this object file is linked. |
5445 | Q_CONSTRUCTOR_FUNCTION(qInitDrawhelperFunctions); |
5446 | |
5447 | QT_END_NAMESPACE |
5448 | |