1 | /****************************************************************************
|
---|
2 | **
|
---|
3 | ** Copyright (C) 2011 Nokia Corporation and/or its subsidiary(-ies).
|
---|
4 | ** All rights reserved.
|
---|
5 | ** Contact: Nokia Corporation ([email protected])
|
---|
6 | **
|
---|
7 | ** This file is part of the QtGui module of the Qt Toolkit.
|
---|
8 | **
|
---|
9 | ** $QT_BEGIN_LICENSE:LGPL$
|
---|
10 | ** Commercial Usage
|
---|
11 | ** Licensees holding valid Qt Commercial licenses may use this file in
|
---|
12 | ** accordance with the Qt Commercial License Agreement provided with the
|
---|
13 | ** Software or, alternatively, in accordance with the terms contained in
|
---|
14 | ** a written agreement between you and Nokia.
|
---|
15 | **
|
---|
16 | ** GNU Lesser General Public License Usage
|
---|
17 | ** Alternatively, this file may be used under the terms of the GNU Lesser
|
---|
18 | ** General Public License version 2.1 as published by the Free Software
|
---|
19 | ** Foundation and appearing in the file LICENSE.LGPL included in the
|
---|
20 | ** packaging of this file. Please review the following information to
|
---|
21 | ** ensure the GNU Lesser General Public License version 2.1 requirements
|
---|
22 | ** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
|
---|
23 | **
|
---|
24 | ** In addition, as a special exception, Nokia gives you certain additional
|
---|
25 | ** rights. These rights are described in the Nokia Qt LGPL Exception
|
---|
26 | ** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
|
---|
27 | **
|
---|
28 | ** GNU General Public License Usage
|
---|
29 | ** Alternatively, this file may be used under the terms of the GNU
|
---|
30 | ** General Public License version 3.0 as published by the Free Software
|
---|
31 | ** Foundation and appearing in the file LICENSE.GPL included in the
|
---|
32 | ** packaging of this file. Please review the following information to
|
---|
33 | ** ensure the GNU General Public License version 3.0 requirements will be
|
---|
34 | ** met: http://www.gnu.org/copyleft/gpl.html.
|
---|
35 | **
|
---|
36 | ** If you have questions regarding the use of this file, please contact
|
---|
37 | ** Nokia at [email protected].
|
---|
38 | ** $QT_END_LICENSE$
|
---|
39 | **
|
---|
40 | ****************************************************************************/
|
---|
41 |
|
---|
42 | #include <qimage.h>
|
---|
43 | #include <private/qimage_p.h>
|
---|
44 | #include <private/qsimd_p.h>
|
---|
45 |
|
---|
46 | #ifdef QT_HAVE_SSSE3
|
---|
47 |
|
---|
48 | QT_BEGIN_NAMESPACE
|
---|
49 |
|
---|
50 | // Convert a scanline of RGB888 (src) to RGB32 (dst)
|
---|
51 | // src must be at least len * 3 bytes
|
---|
52 | // dst must be at least len * 4 bytes
|
---|
53 | Q_GUI_EXPORT void QT_FASTCALL qt_convert_rgb888_to_rgb32_ssse3(quint32 *dst, const uchar *src, int len)
|
---|
54 | {
|
---|
55 | quint32 *const end = dst + len;
|
---|
56 |
|
---|
57 | // Prologue, align dst to 16 bytes. The alignment is done on dst because it has 4 store()
|
---|
58 | // for each 3 load() of src.
|
---|
59 | const int offsetToAlignOn16Bytes = (4 - ((reinterpret_cast<quintptr>(dst) >> 2) & 0x3)) & 0x3;
|
---|
60 | const int prologLength = qMin(len, offsetToAlignOn16Bytes);
|
---|
61 |
|
---|
62 | for (int i = 0; i < prologLength; ++i) {
|
---|
63 | *dst++ = qRgb(src[0], src[1], src[2]);
|
---|
64 | src += 3;
|
---|
65 | }
|
---|
66 |
|
---|
67 | // Mask the 4 first colors of the RGB888 vector
|
---|
68 | const __m128i shuffleMask = _mm_set_epi8(0xff, 9, 10, 11, 0xff, 6, 7, 8, 0xff, 3, 4, 5, 0xff, 0, 1, 2);
|
---|
69 |
|
---|
70 | // Mask the 4 last colors of a RGB888 vector with an offset of 1 (so the last 3 bytes are RGB)
|
---|
71 | const __m128i shuffleMaskEnd = _mm_set_epi8(0xff, 13, 14, 15, 0xff, 10, 11, 12, 0xff, 7, 8, 9, 0xff, 4, 5, 6);
|
---|
72 |
|
---|
73 | // Mask to have alpha = 0xff
|
---|
74 | const __m128i alphaMask = _mm_set1_epi32(0xff000000);
|
---|
75 |
|
---|
76 | __m128i *inVectorPtr = (__m128i *)src;
|
---|
77 | __m128i *dstVectorPtr = (__m128i *)dst;
|
---|
78 |
|
---|
79 | const int simdRoundCount = (len - prologLength) / 16; // one iteration in the loop converts 16 pixels
|
---|
80 | for (int i = 0; i < simdRoundCount; ++i) {
|
---|
81 | /*
|
---|
82 | RGB888 has 5 pixels per vector, + 1 byte from the next pixel. The idea here is
|
---|
83 | to load vectors of RGB888 and use palignr to select a vector out of two vectors.
|
---|
84 |
|
---|
85 | After 3 loads of RGB888 and 3 stores of RGB32, we have 4 pixels left in the last
|
---|
86 | vector of RGB888, we can mask it directly to get a last store or RGB32. After that,
|
---|
87 | the first next byte is a R, and we can loop for the next 16 pixels.
|
---|
88 |
|
---|
89 | The conversion itself is done with a byte permutation (pshufb).
|
---|
90 | */
|
---|
91 | __m128i firstSrcVector = _mm_lddqu_si128(inVectorPtr);
|
---|
92 | __m128i outputVector = _mm_shuffle_epi8(firstSrcVector, shuffleMask);
|
---|
93 | _mm_store_si128(dstVectorPtr, _mm_or_si128(outputVector, alphaMask));
|
---|
94 | ++inVectorPtr;
|
---|
95 | ++dstVectorPtr;
|
---|
96 |
|
---|
97 | // There are 4 unused bytes left in srcVector, we need to load the next 16 bytes
|
---|
98 | // and load the next input with palignr
|
---|
99 | __m128i secondSrcVector = _mm_lddqu_si128(inVectorPtr);
|
---|
100 | __m128i srcVector = _mm_alignr_epi8(secondSrcVector, firstSrcVector, 12);
|
---|
101 | outputVector = _mm_shuffle_epi8(srcVector, shuffleMask);
|
---|
102 | _mm_store_si128(dstVectorPtr, _mm_or_si128(outputVector, alphaMask));
|
---|
103 | ++inVectorPtr;
|
---|
104 | ++dstVectorPtr;
|
---|
105 | firstSrcVector = secondSrcVector;
|
---|
106 |
|
---|
107 | // We now have 8 unused bytes left in firstSrcVector
|
---|
108 | secondSrcVector = _mm_lddqu_si128(inVectorPtr);
|
---|
109 | srcVector = _mm_alignr_epi8(secondSrcVector, firstSrcVector, 8);
|
---|
110 | outputVector = _mm_shuffle_epi8(srcVector, shuffleMask);
|
---|
111 | _mm_store_si128(dstVectorPtr, _mm_or_si128(outputVector, alphaMask));
|
---|
112 | ++inVectorPtr;
|
---|
113 | ++dstVectorPtr;
|
---|
114 |
|
---|
115 | // There are now 12 unused bytes in firstSrcVector.
|
---|
116 | // We can mask them directly, almost there.
|
---|
117 | outputVector = _mm_shuffle_epi8(secondSrcVector, shuffleMaskEnd);
|
---|
118 | _mm_store_si128(dstVectorPtr, _mm_or_si128(outputVector, alphaMask));
|
---|
119 | ++dstVectorPtr;
|
---|
120 | }
|
---|
121 | src = (uchar *)inVectorPtr;
|
---|
122 | dst = (quint32 *)dstVectorPtr;
|
---|
123 |
|
---|
124 | while (dst != end) {
|
---|
125 | *dst++ = qRgb(src[0], src[1], src[2]);
|
---|
126 | src += 3;
|
---|
127 | }
|
---|
128 | }
|
---|
129 |
|
---|
130 | void convert_RGB888_to_RGB32_ssse3(QImageData *dest, const QImageData *src, Qt::ImageConversionFlags)
|
---|
131 | {
|
---|
132 | Q_ASSERT(src->format == QImage::Format_RGB888);
|
---|
133 | Q_ASSERT(dest->format == QImage::Format_RGB32 || dest->format == QImage::Format_ARGB32 || dest->format == QImage::Format_ARGB32_Premultiplied);
|
---|
134 | Q_ASSERT(src->width == dest->width);
|
---|
135 | Q_ASSERT(src->height == dest->height);
|
---|
136 |
|
---|
137 | const uchar *src_data = (uchar *) src->data;
|
---|
138 | quint32 *dest_data = (quint32 *) dest->data;
|
---|
139 |
|
---|
140 | for (int i = 0; i < src->height; ++i) {
|
---|
141 | qt_convert_rgb888_to_rgb32_ssse3(dest_data, src_data, src->width);
|
---|
142 | src_data += src->bytes_per_line;
|
---|
143 | dest_data = (quint32 *)((uchar*)dest_data + dest->bytes_per_line);
|
---|
144 | }
|
---|
145 | }
|
---|
146 |
|
---|
147 | QT_END_NAMESPACE
|
---|
148 |
|
---|
149 | #endif // QT_HAVE_SSSE3
|
---|