1 /****************************************************************************
2 **
3 ** Copyright (C) 2016 The Qt Company Ltd.
4 ** Contact: https://www.qt.io/licensing/
5 **
6 ** This file is part of the QtGui module of the Qt Toolkit.
7 **
8 ** $QT_BEGIN_LICENSE:LGPL$
9 ** Commercial License Usage
10 ** Licensees holding valid commercial Qt licenses may use this file in
11 ** accordance with the commercial license agreement provided with the
12 ** Software or, alternatively, in accordance with the terms contained in
13 ** a written agreement between you and The Qt Company. For licensing terms
14 ** and conditions see https://www.qt.io/terms-conditions. For further
15 ** information use the contact form at https://www.qt.io/contact-us.
16 **
17 ** GNU Lesser General Public License Usage
18 ** Alternatively, this file may be used under the terms of the GNU Lesser
19 ** General Public License version 3 as published by the Free Software
20 ** Foundation and appearing in the file LICENSE.LGPL3 included in the
21 ** packaging of this file. Please review the following information to
22 ** ensure the GNU Lesser General Public License version 3 requirements
23 ** will be met: https://www.gnu.org/licenses/lgpl-3.0.html.
24 **
25 ** GNU General Public License Usage
26 ** Alternatively, this file may be used under the terms of the GNU
27 ** General Public License version 2.0 or (at your option) the GNU General
28 ** Public license version 3 or any later version approved by the KDE Free
29 ** Qt Foundation. The licenses are as published by the Free Software
30 ** Foundation and appearing in the file LICENSE.GPL2 and LICENSE.GPL3
31 ** included in the packaging of this file. Please review the following
32 ** information to ensure the GNU General Public License requirements will
33 ** be met: https://www.gnu.org/licenses/gpl-2.0.html and
34 ** https://www.gnu.org/licenses/gpl-3.0.html.
35 **
36 ** $QT_END_LICENSE$
37 **
38 ****************************************************************************/
39
40 #include "qimagescale_p.h"
41 #include "qimage.h"
42 #include <private/qsimd_p.h>
43
44 #if QT_CONFIG(thread) && !defined(Q_OS_WASM)
45 #include "qsemaphore.h"
46 #include "qthreadpool.h"
47 #endif
48
49 #if defined(__ARM_NEON__)
50
51 QT_BEGIN_NAMESPACE
52
53 using namespace QImageScale;
54
55 template<typename T>
multithread_pixels_function(QImageScaleInfo * isi,int dh,const T & scaleSection)56 static inline void multithread_pixels_function(QImageScaleInfo *isi, int dh, const T &scaleSection)
57 {
58 #if QT_CONFIG(thread) && !defined(Q_OS_WASM)
59 int segments = (qsizetype(isi->sh) * isi->sw) / (1<<16);
60 segments = std::min(segments, dh);
61 QThreadPool *threadPool = QThreadPool::globalInstance();
62 if (segments > 1 && threadPool && !threadPool->contains(QThread::currentThread())) {
63 QSemaphore semaphore;
64 int y = 0;
65 for (int i = 0; i < segments; ++i) {
66 int yn = (dh - y) / (segments - i);
67 threadPool->start([&, y, yn]() {
68 scaleSection(y, y + yn);
69 semaphore.release(1);
70 });
71 y += yn;
72 }
73 semaphore.acquire(segments);
74 return;
75 }
76 #endif
77 scaleSection(0, dh);
78 }
79
qt_qimageScaleAARGBA_helper(const unsigned int * pix,int xyap,int Cxy,int step)80 inline static uint32x4_t qt_qimageScaleAARGBA_helper(const unsigned int *pix, int xyap, int Cxy, int step)
81 {
82 uint32x2_t vpix32 = vmov_n_u32(*pix);
83 uint16x4_t vpix16 = vget_low_u16(vmovl_u8(vreinterpret_u8_u32(vpix32)));
84 uint32x4_t vx = vmull_n_u16(vpix16, xyap);
85 int i;
86 for (i = (1 << 14) - xyap; i > Cxy; i -= Cxy) {
87 pix += step;
88 vpix32 = vmov_n_u32(*pix);
89 vpix16 = vget_low_u16(vmovl_u8(vreinterpret_u8_u32(vpix32)));
90 vx = vaddq_u32(vx, vmull_n_u16(vpix16, Cxy));
91 }
92 pix += step;
93 vpix32 = vmov_n_u32(*pix);
94 vpix16 = vget_low_u16(vmovl_u8(vreinterpret_u8_u32(vpix32)));
95 vx = vaddq_u32(vx, vmull_n_u16(vpix16, i));
96 return vx;
97 }
98
99 template<bool RGB>
qt_qimageScaleAARGBA_up_x_down_y_neon(QImageScaleInfo * isi,unsigned int * dest,int dw,int dh,int dow,int sow)100 void qt_qimageScaleAARGBA_up_x_down_y_neon(QImageScaleInfo *isi, unsigned int *dest,
101 int dw, int dh, int dow, int sow)
102 {
103 const unsigned int **ypoints = isi->ypoints;
104 int *xpoints = isi->xpoints;
105 int *xapoints = isi->xapoints;
106 int *yapoints = isi->yapoints;
107
108 /* go through every scanline in the output buffer */
109 auto scaleSection = [&] (int yStart, int yEnd) {
110 for (int y = yStart; y < yEnd; ++y) {
111 int Cy = yapoints[y] >> 16;
112 int yap = yapoints[y] & 0xffff;
113
114 unsigned int *dptr = dest + (y * dow);
115 for (int x = 0; x < dw; x++) {
116 const unsigned int *sptr = ypoints[y] + xpoints[x];
117 uint32x4_t vx = qt_qimageScaleAARGBA_helper(sptr, yap, Cy, sow);
118
119 int xap = xapoints[x];
120 if (xap > 0) {
121 uint32x4_t vr = qt_qimageScaleAARGBA_helper(sptr + 1, yap, Cy, sow);
122
123 vx = vmulq_n_u32(vx, 256 - xap);
124 vr = vmulq_n_u32(vr, xap);
125 vx = vaddq_u32(vx, vr);
126 vx = vshrq_n_u32(vx, 8);
127 }
128 vx = vshrq_n_u32(vx, 14);
129 const uint16x4_t vx16 = vmovn_u32(vx);
130 const uint8x8_t vx8 = vmovn_u16(vcombine_u16(vx16, vx16));
131 *dptr = vget_lane_u32(vreinterpret_u32_u8(vx8), 0);
132 if (RGB)
133 *dptr |= 0xff000000;
134 dptr++;
135 }
136 }
137 };
138 multithread_pixels_function(isi, dh, scaleSection);
139 }
140
141 template<bool RGB>
qt_qimageScaleAARGBA_down_x_up_y_neon(QImageScaleInfo * isi,unsigned int * dest,int dw,int dh,int dow,int sow)142 void qt_qimageScaleAARGBA_down_x_up_y_neon(QImageScaleInfo *isi, unsigned int *dest,
143 int dw, int dh, int dow, int sow)
144 {
145 const unsigned int **ypoints = isi->ypoints;
146 int *xpoints = isi->xpoints;
147 int *xapoints = isi->xapoints;
148 int *yapoints = isi->yapoints;
149
150 /* go through every scanline in the output buffer */
151 auto scaleSection = [&] (int yStart, int yEnd) {
152 for (int y = yStart; y < yEnd; ++y) {
153 unsigned int *dptr = dest + (y * dow);
154 for (int x = 0; x < dw; x++) {
155 int Cx = xapoints[x] >> 16;
156 int xap = xapoints[x] & 0xffff;
157
158 const unsigned int *sptr = ypoints[y] + xpoints[x];
159 uint32x4_t vx = qt_qimageScaleAARGBA_helper(sptr, xap, Cx, 1);
160
161 int yap = yapoints[y];
162 if (yap > 0) {
163 uint32x4_t vr = qt_qimageScaleAARGBA_helper(sptr + sow, xap, Cx, 1);
164
165 vx = vmulq_n_u32(vx, 256 - yap);
166 vr = vmulq_n_u32(vr, yap);
167 vx = vaddq_u32(vx, vr);
168 vx = vshrq_n_u32(vx, 8);
169 }
170 vx = vshrq_n_u32(vx, 14);
171 const uint16x4_t vx16 = vmovn_u32(vx);
172 const uint8x8_t vx8 = vmovn_u16(vcombine_u16(vx16, vx16));
173 *dptr = vget_lane_u32(vreinterpret_u32_u8(vx8), 0);
174 if (RGB)
175 *dptr |= 0xff000000;
176 dptr++;
177 }
178 }
179 };
180 multithread_pixels_function(isi, dh, scaleSection);
181 }
182
183 template<bool RGB>
qt_qimageScaleAARGBA_down_xy_neon(QImageScaleInfo * isi,unsigned int * dest,int dw,int dh,int dow,int sow)184 void qt_qimageScaleAARGBA_down_xy_neon(QImageScaleInfo *isi, unsigned int *dest,
185 int dw, int dh, int dow, int sow)
186 {
187 const unsigned int **ypoints = isi->ypoints;
188 int *xpoints = isi->xpoints;
189 int *xapoints = isi->xapoints;
190 int *yapoints = isi->yapoints;
191
192 auto scaleSection = [&] (int yStart, int yEnd) {
193 for (int y = yStart; y < yEnd; ++y) {
194 int Cy = yapoints[y] >> 16;
195 int yap = yapoints[y] & 0xffff;
196
197 unsigned int *dptr = dest + (y * dow);
198 for (int x = 0; x < dw; x++) {
199 const int Cx = xapoints[x] >> 16;
200 const int xap = xapoints[x] & 0xffff;
201
202 const unsigned int *sptr = ypoints[y] + xpoints[x];
203 uint32x4_t vx = qt_qimageScaleAARGBA_helper(sptr, xap, Cx, 1);
204 vx = vshrq_n_u32(vx, 4);
205 uint32x4_t vr = vmulq_n_u32(vx, yap);
206
207 int j;
208 for (j = (1 << 14) - yap; j > Cy; j -= Cy) {
209 sptr += sow;
210 vx = qt_qimageScaleAARGBA_helper(sptr, xap, Cx, 1);
211 vx = vshrq_n_u32(vx, 4);
212 vx = vmulq_n_u32(vx, Cy);
213 vr = vaddq_u32(vr, vx);
214 }
215 sptr += sow;
216 vx = qt_qimageScaleAARGBA_helper(sptr, xap, Cx, 1);
217 vx = vshrq_n_u32(vx, 4);
218 vx = vmulq_n_u32(vx, j);
219 vr = vaddq_u32(vr, vx);
220
221 vx = vshrq_n_u32(vr, 24);
222 const uint16x4_t vx16 = vmovn_u32(vx);
223 const uint8x8_t vx8 = vmovn_u16(vcombine_u16(vx16, vx16));
224 *dptr = vget_lane_u32(vreinterpret_u32_u8(vx8), 0);
225 if (RGB)
226 *dptr |= 0xff000000;
227 dptr++;
228 }
229 }
230 };
231 multithread_pixels_function(isi, dh, scaleSection);
232 }
233
234 template void qt_qimageScaleAARGBA_up_x_down_y_neon<false>(QImageScaleInfo *isi, unsigned int *dest,
235 int dw, int dh, int dow, int sow);
236
237 template void qt_qimageScaleAARGBA_up_x_down_y_neon<true>(QImageScaleInfo *isi, unsigned int *dest,
238 int dw, int dh, int dow, int sow);
239
240 template void qt_qimageScaleAARGBA_down_x_up_y_neon<false>(QImageScaleInfo *isi, unsigned int *dest,
241 int dw, int dh, int dow, int sow);
242
243 template void qt_qimageScaleAARGBA_down_x_up_y_neon<true>(QImageScaleInfo *isi, unsigned int *dest,
244 int dw, int dh, int dow, int sow);
245
246 template void qt_qimageScaleAARGBA_down_xy_neon<false>(QImageScaleInfo *isi, unsigned int *dest,
247 int dw, int dh, int dow, int sow);
248
249 template void qt_qimageScaleAARGBA_down_xy_neon<true>(QImageScaleInfo *isi, unsigned int *dest,
250 int dw, int dh, int dow, int sow);
251
252 QT_END_NAMESPACE
253
254 #endif
255