1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
|
/****************************************************************************
**
** Copyright (C) 2013 Digia Plc and/or its subsidiary(-ies).
** Contact: http://www.qt-project.org/legal
**
** This file is part of the QtGui module of the Qt Toolkit.
**
** $QT_BEGIN_LICENSE:LGPL$
** Commercial License Usage
** Licensees holding valid commercial Qt licenses may use this file in
** accordance with the commercial license agreement provided with the
** Software or, alternatively, in accordance with the terms contained in
** a written agreement between you and Digia. For licensing terms and
** conditions see http://qt.digia.com/licensing. For further information
** use the contact form at http://qt.digia.com/contact-us.
**
** GNU Lesser General Public License Usage
** Alternatively, this file may be used under the terms of the GNU Lesser
** General Public License version 2.1 as published by the Free Software
** Foundation and appearing in the file LICENSE.LGPL included in the
** packaging of this file. Please review the following information to
** ensure the GNU Lesser General Public License version 2.1 requirements
** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
**
** In addition, as a special exception, Digia gives you certain additional
** rights. These rights are described in the Digia Qt LGPL Exception
** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
**
** GNU General Public License Usage
** Alternatively, this file may be used under the terms of the GNU
** General Public License version 3.0 as published by the Free Software
** Foundation and appearing in the file LICENSE.GPL included in the
** packaging of this file. Please review the following information to
** ensure the GNU General Public License version 3.0 requirements will be
** met: http://www.gnu.org/copyleft/gpl.html.
**
**
** $QT_END_LICENSE$
**
****************************************************************************/
#ifndef QDRAWINGPRIMITIVE_SSE2_P_H
#define QDRAWINGPRIMITIVE_SSE2_P_H
#include <private/qsimd_p.h>
#ifdef QT_HAVE_SSE2
//
// W A R N I N G
// -------------
//
// This file is not part of the Qt API. It exists purely as an
// implementation detail. This header file may change from version to
// version without notice, or even be removed.
//
// We mean it.
//
QT_BEGIN_NAMESPACE
/*
* Multiply the components of pixelVector by alphaChannel
* Each 32bits components of alphaChannel must be in the form 0x00AA00AA
* colorMask must have 0x00ff00ff on each 32 bits component
* half must have the value 128 (0x80) for each 32 bits compnent
*/
#define BYTE_MUL_SSE2(result, pixelVector, alphaChannel, colorMask, half) \
{ \
/* 1. separate the colors in 2 vectors so each color is on 16 bits \
(in order to be multiplied by the alpha \
each 32 bit of dstVectorAG are in the form 0x00AA00GG \
each 32 bit of dstVectorRB are in the form 0x00RR00BB */\
__m128i pixelVectorAG = _mm_srli_epi16(pixelVector, 8); \
__m128i pixelVectorRB = _mm_and_si128(pixelVector, colorMask); \
\
/* 2. multiply the vectors by the alpha channel */\
pixelVectorAG = _mm_mullo_epi16(pixelVectorAG, alphaChannel); \
pixelVectorRB = _mm_mullo_epi16(pixelVectorRB, alphaChannel); \
\
/* 3. divide by 255, that's the tricky part. \
we do it like for BYTE_MUL(), with bit shift: X/255 ~= (X + X/256 + rounding)/256 */ \
/** so first (X + X/256 + rounding) */\
pixelVectorRB = _mm_add_epi16(pixelVectorRB, _mm_srli_epi16(pixelVectorRB, 8)); \
pixelVectorRB = _mm_add_epi16(pixelVectorRB, half); \
pixelVectorAG = _mm_add_epi16(pixelVectorAG, _mm_srli_epi16(pixelVectorAG, 8)); \
pixelVectorAG = _mm_add_epi16(pixelVectorAG, half); \
\
/** second divide by 256 */\
pixelVectorRB = _mm_srli_epi16(pixelVectorRB, 8); \
/** for AG, we could >> 8 to divide followed by << 8 to put the \
bytes in the correct position. By masking instead, we execute \
only one instruction */\
pixelVectorAG = _mm_andnot_si128(colorMask, pixelVectorAG); \
\
/* 4. combine the 2 pairs of colors */ \
result = _mm_or_si128(pixelVectorAG, pixelVectorRB); \
}
/*
* Each 32bits components of alphaChannel must be in the form 0x00AA00AA
* oneMinusAlphaChannel must be 255 - alpha for each 32 bits component
* colorMask must have 0x00ff00ff on each 32 bits component
* half must have the value 128 (0x80) for each 32 bits compnent
*/
#define INTERPOLATE_PIXEL_255_SSE2(result, srcVector, dstVector, alphaChannel, oneMinusAlphaChannel, colorMask, half) { \
/* interpolate AG */\
__m128i srcVectorAG = _mm_srli_epi16(srcVector, 8); \
__m128i dstVectorAG = _mm_srli_epi16(dstVector, 8); \
__m128i srcVectorAGalpha = _mm_mullo_epi16(srcVectorAG, alphaChannel); \
__m128i dstVectorAGoneMinusAlphalpha = _mm_mullo_epi16(dstVectorAG, oneMinusAlphaChannel); \
__m128i finalAG = _mm_add_epi16(srcVectorAGalpha, dstVectorAGoneMinusAlphalpha); \
finalAG = _mm_add_epi16(finalAG, _mm_srli_epi16(finalAG, 8)); \
finalAG = _mm_add_epi16(finalAG, half); \
finalAG = _mm_andnot_si128(colorMask, finalAG); \
\
/* interpolate RB */\
__m128i srcVectorRB = _mm_and_si128(srcVector, colorMask); \
__m128i dstVectorRB = _mm_and_si128(dstVector, colorMask); \
__m128i srcVectorRBalpha = _mm_mullo_epi16(srcVectorRB, alphaChannel); \
__m128i dstVectorRBoneMinusAlphalpha = _mm_mullo_epi16(dstVectorRB, oneMinusAlphaChannel); \
__m128i finalRB = _mm_add_epi16(srcVectorRBalpha, dstVectorRBoneMinusAlphalpha); \
finalRB = _mm_add_epi16(finalRB, _mm_srli_epi16(finalRB, 8)); \
finalRB = _mm_add_epi16(finalRB, half); \
finalRB = _mm_srli_epi16(finalRB, 8); \
\
/* combine */\
result = _mm_or_si128(finalAG, finalRB); \
}
// Basically blend src over dst with the const alpha defined as constAlphaVector.
// nullVector, half, one, colorMask are constant across the whole image/texture, and should be defined as:
//const __m128i nullVector = _mm_set1_epi32(0);
//const __m128i half = _mm_set1_epi16(0x80);
//const __m128i one = _mm_set1_epi16(0xff);
//const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
//const __m128i alphaMask = _mm_set1_epi32(0xff000000);
//
// The computation being done is:
// result = s + d * (1-alpha)
// with shortcuts if fully opaque or fully transparent.
#define BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, length, nullVector, half, one, colorMask, alphaMask) { \
int x = 0; \
\
/* First, get dst aligned. */ \
ALIGNMENT_PROLOGUE_16BYTES(dst, x, length) { \
uint s = src[x]; \
if (s >= 0xff000000) \
dst[x] = s; \
else if (s != 0) \
dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
} \
\
for (; x < length-3; x += 4) { \
const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]); \
const __m128i srcVectorAlpha = _mm_and_si128(srcVector, alphaMask); \
if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, alphaMask)) == 0xffff) { \
/* all opaque */ \
_mm_store_si128((__m128i *)&dst[x], srcVector); \
} else if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, nullVector)) != 0xffff) { \
/* not fully transparent */ \
/* extract the alpha channel on 2 x 16 bits */ \
/* so we have room for the multiplication */ \
/* each 32 bits will be in the form 0x00AA00AA */ \
/* with A being the 1 - alpha */ \
__m128i alphaChannel = _mm_srli_epi32(srcVector, 24); \
alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16)); \
alphaChannel = _mm_sub_epi16(one, alphaChannel); \
\
const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]); \
__m128i destMultipliedByOneMinusAlpha; \
BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half); \
\
/* result = s + d * (1-alpha) */\
const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha); \
_mm_store_si128((__m128i *)&dst[x], result); \
} \
} \
for (; x < length; ++x) { \
uint s = src[x]; \
if (s >= 0xff000000) \
dst[x] = s; \
else if (s != 0) \
dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
} \
}
// Basically blend src over dst with the const alpha defined as constAlphaVector.
// nullVector, half, one, colorMask are constant across the whole image/texture, and should be defined as:
//const __m128i nullVector = _mm_set1_epi32(0);
//const __m128i half = _mm_set1_epi16(0x80);
//const __m128i one = _mm_set1_epi16(0xff);
//const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
//
// The computation being done is:
// dest = (s + d * sia) * ca + d * cia
// = s * ca + d * (sia * ca + cia)
// = s * ca + d * (1 - sa*ca)
#define BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, length, nullVector, half, one, colorMask, constAlphaVector) \
{ \
int x = 0; \
\
ALIGNMENT_PROLOGUE_16BYTES(dst, x, length) { \
quint32 s = src[x]; \
if (s != 0) { \
s = BYTE_MUL(s, const_alpha); \
dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
} \
} \
\
for (; x < length-3; x += 4) { \
__m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]); \
if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVector, nullVector)) != 0xffff) { \
BYTE_MUL_SSE2(srcVector, srcVector, constAlphaVector, colorMask, half); \
\
__m128i alphaChannel = _mm_srli_epi32(srcVector, 24); \
alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16)); \
alphaChannel = _mm_sub_epi16(one, alphaChannel); \
\
const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]); \
__m128i destMultipliedByOneMinusAlpha; \
BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half); \
\
const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha); \
_mm_store_si128((__m128i *)&dst[x], result); \
} \
} \
for (; x < length; ++x) { \
quint32 s = src[x]; \
if (s != 0) { \
s = BYTE_MUL(s, const_alpha); \
dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
} \
} \
}
QT_END_NAMESPACE
#endif // QT_HAVE_SSE2
#endif // QDRAWINGPRIMITIVE_SSE2_P_H
|
