1 /****************************************************************************

     2 **

     3 ** Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies).

     4 ** All rights reserved.

     5 ** Contact: Nokia Corporation (qt-info@nokia.com)

     6 **

     7 ** This file is part of the QtGui module of the Qt Toolkit.

     8 **

     9 ** $QT_BEGIN_LICENSE:LGPL$

    10 ** No Commercial Usage

    11 ** This file contains pre-release code and may not be distributed.

    12 ** You may use this file in accordance with the terms and conditions

    13 ** contained in the Technology Preview License Agreement accompanying

    14 ** this package.

    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 ** If you have questions regarding the use of this file, please contact

    29 ** Nokia at qt-info@nokia.com.

    30 **

    31 **

    32 **

    33 **

    34 **

    35 **

    36 **

    37 **

    38 ** $QT_END_LICENSE$

    39 **

    40 ****************************************************************************/

    41 

    42 #ifndef QDRAWINGPRIMITIVE_SSE2_P_H

    43 #define QDRAWINGPRIMITIVE_SSE2_P_H

    44 

    45 #include <private/qsimd_p.h>

    46 

    47 #ifdef QT_HAVE_SSE2

    48 

    49 //

    50 //  W A R N I N G

    51 //  -------------

    52 //

    53 // This file is not part of the Qt API.  It exists purely as an

    54 // implementation detail.  This header file may change from version to

    55 // version without notice, or even be removed.

    56 //

    57 // We mean it.

    58 //

    59 

    60 QT_BEGIN_NAMESPACE

    61 

    62 /*

    63  * Multiply the components of pixelVector by alphaChannel

    64  * Each 32bits components of alphaChannel must be in the form 0x00AA00AA

    65  * colorMask must have 0x00ff00ff on each 32 bits component

    66  * half must have the value 128 (0x80) for each 32 bits compnent

    67  */

    68 #define BYTE_MUL_SSE2(result, pixelVector, alphaChannel, colorMask, half) \

    69 { \

    70     /* 1. separate the colors in 2 vectors so each color is on 16 bits \

    71        (in order to be multiplied by the alpha \

    72        each 32 bit of dstVectorAG are in the form 0x00AA00GG \

    73        each 32 bit of dstVectorRB are in the form 0x00RR00BB */\

    74     __m128i pixelVectorAG = _mm_srli_epi16(pixelVector, 8); \

    75     __m128i pixelVectorRB = _mm_and_si128(pixelVector, colorMask); \

    76  \

    77     /* 2. multiply the vectors by the alpha channel */\

    78     pixelVectorAG = _mm_mullo_epi16(pixelVectorAG, alphaChannel); \

    79     pixelVectorRB = _mm_mullo_epi16(pixelVectorRB, alphaChannel); \

    80  \

    81     /* 3. devide by 255, that's the tricky part. \

    82        we do it like for BYTE_MUL(), with bit shift: X/255 ~= (X + X/256 + rounding)/256 */ \

    83     /** so first (X + X/256 + rounding) */\

    84     pixelVectorRB = _mm_add_epi16(pixelVectorRB, _mm_srli_epi16(pixelVectorRB, 8)); \

    85     pixelVectorRB = _mm_add_epi16(pixelVectorRB, half); \

    86     pixelVectorAG = _mm_add_epi16(pixelVectorAG, _mm_srli_epi16(pixelVectorAG, 8)); \

    87     pixelVectorAG = _mm_add_epi16(pixelVectorAG, half); \

    88  \

    89     /** second devide by 256 */\

    90     pixelVectorRB = _mm_srli_epi16(pixelVectorRB, 8); \

    91     /** for AG, we could >> 8 to divide followed by << 8 to put the \

    92         bytes in the correct position. By masking instead, we execute \

    93         only one instruction */\

    94     pixelVectorAG = _mm_andnot_si128(colorMask, pixelVectorAG); \

    95  \

    96     /* 4. combine the 2 pairs of colors */ \

    97     result = _mm_or_si128(pixelVectorAG, pixelVectorRB); \

    98 }

    99 

   100 /*

   101  * Each 32bits components of alphaChannel must be in the form 0x00AA00AA

   102  * oneMinusAlphaChannel must be 255 - alpha for each 32 bits component

   103  * colorMask must have 0x00ff00ff on each 32 bits component

   104  * half must have the value 128 (0x80) for each 32 bits compnent

   105  */

   106 #define INTERPOLATE_PIXEL_255_SSE2(result, srcVector, dstVector, alphaChannel, oneMinusAlphaChannel, colorMask, half) { \

   107     /* interpolate AG */\

   108     __m128i srcVectorAG = _mm_srli_epi16(srcVector, 8); \

   109     __m128i dstVectorAG = _mm_srli_epi16(dstVector, 8); \

   110     __m128i srcVectorAGalpha = _mm_mullo_epi16(srcVectorAG, alphaChannel); \

   111     __m128i dstVectorAGoneMinusAlphalpha = _mm_mullo_epi16(dstVectorAG, oneMinusAlphaChannel); \

   112     __m128i finalAG = _mm_add_epi16(srcVectorAGalpha, dstVectorAGoneMinusAlphalpha); \

   113     finalAG = _mm_add_epi16(finalAG, _mm_srli_epi16(finalAG, 8)); \

   114     finalAG = _mm_add_epi16(finalAG, half); \

   115     finalAG = _mm_andnot_si128(colorMask, finalAG); \

   116  \

   117     /* interpolate RB */\

   118     __m128i srcVectorRB = _mm_and_si128(srcVector, colorMask); \

   119     __m128i dstVectorRB = _mm_and_si128(dstVector, colorMask); \

   120     __m128i srcVectorRBalpha = _mm_mullo_epi16(srcVectorRB, alphaChannel); \

   121     __m128i dstVectorRBoneMinusAlphalpha = _mm_mullo_epi16(dstVectorRB, oneMinusAlphaChannel); \

   122     __m128i finalRB = _mm_add_epi16(srcVectorRBalpha, dstVectorRBoneMinusAlphalpha); \

   123     finalRB = _mm_add_epi16(finalRB, _mm_srli_epi16(finalRB, 8)); \

   124     finalRB = _mm_add_epi16(finalRB, half); \

   125     finalRB = _mm_srli_epi16(finalRB, 8); \

   126  \

   127     /* combine */\

   128     result = _mm_or_si128(finalAG, finalRB); \

   129 }

   130 

   131 // Basically blend src over dst with the const alpha defined as constAlphaVector.

   132 // nullVector, half, one, colorMask are constant accross the whole image/texture, and should be defined as:

   133 //const __m128i nullVector = _mm_set1_epi32(0);

   134 //const __m128i half = _mm_set1_epi16(0x80);

   135 //const __m128i one = _mm_set1_epi16(0xff);

   136 //const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);

   137 //const __m128i alphaMask = _mm_set1_epi32(0xff000000);

   138 //

   139 // The computation being done is:

   140 // result = s + d * (1-alpha)

   141 // with shortcuts if fully opaque or fully transparent.

   142 #define BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, length, nullVector, half, one, colorMask, alphaMask) { \

   143     int x = 0; \

   144 \

   145     /* First, get dst aligned. */ \

   146     const int offsetToAlignOn16Bytes = (4 - ((reinterpret_cast<quintptr>(dst) >> 2) & 0x3)) & 0x3;\

   147     const int prologLength = qMin(length, offsetToAlignOn16Bytes);\

   148     for (; x < prologLength; ++x) { \

   149         uint s = src[x]; \

   150         if (s >= 0xff000000) \

   151             dst[x] = s; \

   152         else if (s != 0) \

   153             dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \

   154     } \

   155 \

   156     for (; x < length-3; x += 4) { \

   157         const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]); \

   158         const __m128i srcVectorAlpha = _mm_and_si128(srcVector, alphaMask); \

   159         if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, alphaMask)) == 0xffff) { \

   160             /* all opaque */ \

   161             _mm_store_si128((__m128i *)&dst[x], srcVector); \

   162         } else if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, nullVector)) != 0xffff) { \

   163             /* not fully transparent */ \

   164             /* extract the alpha channel on 2 x 16 bits */ \

   165             /* so we have room for the multiplication */ \

   166             /* each 32 bits will be in the form 0x00AA00AA */ \

   167             /* with A being the 1 - alpha */ \

   168             __m128i alphaChannel = _mm_srli_epi32(srcVector, 24); \

   169             alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16)); \

   170             alphaChannel = _mm_sub_epi16(one, alphaChannel); \

   171  \

   172             const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]); \

   173             __m128i destMultipliedByOneMinusAlpha; \

   174             BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half); \

   175  \

   176             /* result = s + d * (1-alpha) */\

   177             const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha); \

   178             _mm_store_si128((__m128i *)&dst[x], result); \

   179         } \

   180     } \

   181     for (; x < length; ++x) { \

   182         uint s = src[x]; \

   183         if (s >= 0xff000000) \

   184             dst[x] = s; \

   185         else if (s != 0) \

   186             dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \

   187     } \

   188 }

   189 

   190 // Basically blend src over dst with the const alpha defined as constAlphaVector.

   191 // nullVector, half, one, colorMask are constant accross the whole image/texture, and should be defined as:

   192 //const __m128i nullVector = _mm_set1_epi32(0);

   193 //const __m128i half = _mm_set1_epi16(0x80);

   194 //const __m128i one = _mm_set1_epi16(0xff);

   195 //const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);

   196 //

   197 // The computation being done is:

   198 // dest = (s + d * sia) * ca + d * cia

   199 //      = s * ca + d * (sia * ca + cia)

   200 //      = s * ca + d * (1 - sa*ca)

   201 #define BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, length, nullVector, half, one, colorMask, constAlphaVector) \

   202 { \

   203     int x = 0; \

   204 \

   205     const int offsetToAlignOn16Bytes = (4 - ((reinterpret_cast<quintptr>(dst) >> 2) & 0x3)) & 0x3;\

   206     const int prologLength = qMin(length, offsetToAlignOn16Bytes);\

   207     for (; x < prologLength; ++x) { \

   208         quint32 s = src[x]; \

   209         if (s != 0) { \

   210             s = BYTE_MUL(s, const_alpha); \

   211             dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \

   212         } \

   213     } \

   214 \

   215     for (; x < length-3; x += 4) { \

   216         __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]); \

   217         if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVector, nullVector)) != 0xffff) { \

   218             BYTE_MUL_SSE2(srcVector, srcVector, constAlphaVector, colorMask, half); \

   219 \

   220             __m128i alphaChannel = _mm_srli_epi32(srcVector, 24); \

   221             alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16)); \

   222             alphaChannel = _mm_sub_epi16(one, alphaChannel); \

   223  \

   224             const __m128i dstVector = _mm_load_si128((__m128i *)&dst[x]); \

   225             __m128i destMultipliedByOneMinusAlpha; \

   226             BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half); \

   227  \

   228             const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha); \

   229             _mm_store_si128((__m128i *)&dst[x], result); \

   230         } \

   231     } \

   232     for (; x < length; ++x) { \

   233         quint32 s = src[x]; \

   234         if (s != 0) { \

   235             s = BYTE_MUL(s, const_alpha); \

   236             dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \

   237         } \

   238     } \

   239 }

   240 

   241 QT_END_NAMESPACE

   242 

   243 #endif // QT_HAVE_SSE2

   244 

   245 #endif // QDRAWINGPRIMITIVE_SSE2_P_H