recipes/qt4/qt-4.6.0/1165-NEON-configure-detection-and-initial-blend-function-.patch

   1
   2 Adds new NEON configure test and -no-neon configure option. NEON
   3 implementations can also be turned off by setting the QT_NO_NEON
   4 environment variable.
   5
   6 Performance improvements (in frames per second):
   7 - Blending ARGB32 on RGB32/ARGB32, mostly opaque: 71 %
   8 - Blending ARGB32 on RGB32/ARGB32, no opaque pixels: 108 %
   9 - Blending ARGB32 on RGB32/ARGB32, with 0.5 opacity: 158 %
  10 - Blending RGB32 on RGB32/ARGB32, with 0.5 opacity: 189 %
  11
  12 Task-number: QTBUG-6684
  13 Reviewed-by: Gunnar Sletta
  14 Reviewed-by: Paul Olav Tvete
  15 ---
  16  config.tests/unix/neon/neon.cpp       |   51 +++++++
  17  config.tests/unix/neon/neon.pro       |    2 +
  18  configure                             |   22 +++-
  19  src/gui/painting/painting.pri         |    7 +
  20  src/gui/painting/qblendfunctions.cpp  |    2 +-
  21  src/gui/painting/qdrawhelper.cpp      |   16 ++-
  22  src/gui/painting/qdrawhelper_neon.cpp |  260 +++++++++++++++++++++++++++++++++
  23  src/gui/painting/qdrawhelper_neon_p.h |   76 ++++++++++
  24  8 files changed, 432 insertions(+), 4 deletions(-)
  25  create mode 100644 config.tests/unix/neon/neon.cpp
  26  create mode 100644 config.tests/unix/neon/neon.pro
  27  create mode 100644 src/gui/painting/qdrawhelper_neon.cpp
  28  create mode 100644 src/gui/painting/qdrawhelper_neon_p.h
  29
  30 diff --git a/config.tests/unix/neon/neon.cpp b/config.tests/unix/neon/neon.cpp
  31 new file mode 100644
  32 index 0000000..c31a9fd
  33 --- /dev/null
  34 +++ b/config.tests/unix/neon/neon.cpp
  35 @@ -0,0 +1,51 @@
  36 +/****************************************************************************
  37 +**
  38 +** Copyright (C) 2009 Nokia Corporation and/or its subsidiary(-ies).
  39 +** All rights reserved.
  40 +** Contact: Nokia Corporation (qt-info@nokia.com)
  41 +**
  42 +** This file is part of the config.tests of the Qt Toolkit.
  43 +**
  44 +** $QT_BEGIN_LICENSE:LGPL$
  45 +** No Commercial Usage
  46 +** This file contains pre-release code and may not be distributed.
  47 +** You may use this file in accordance with the terms and conditions
  48 +** contained in the Technology Preview License Agreement accompanying
  49 +** this package.
  50 +**
  51 +** GNU Lesser General Public License Usage
  52 +** Alternatively, this file may be used under the terms of the GNU Lesser
  53 +** General Public License version 2.1 as published by the Free Software
  54 +** Foundation and appearing in the file LICENSE.LGPL included in the
  55 +** packaging of this file.  Please review the following information to
  56 +** ensure the GNU Lesser General Public License version 2.1 requirements
  57 +** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
  58 +**
  59 +** In addition, as a special exception, Nokia gives you certain additional
  60 +** rights.  These rights are described in the Nokia Qt LGPL Exception
  61 +** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
  62 +**
  63 +** If you have questions regarding the use of this file, please contact
  64 +** Nokia at qt-info@nokia.com.
  65 +**
  66 +**
  67 +**
  68 +**
  69 +**
  70 +**
  71 +**
  72 +**
  73 +** $QT_END_LICENSE$
  74 +**
  75 +****************************************************************************/
  76 +
  77 +#include <arm_neon.h>
  78 +
  79 +int main(int, char**)
  80 +{
  81 +    int32x4_t null = vdupq_n_s32(0x0);
  82 +
  83 +    int result;
  84 +    vst1q_lane_s32(&result, null, 0);
  85 +    return result;
  86 +}
  87 diff --git a/config.tests/unix/neon/neon.pro b/config.tests/unix/neon/neon.pro
  88 new file mode 100644
  89 index 0000000..de20c4e
  90 --- /dev/null
  91 +++ b/config.tests/unix/neon/neon.pro
  92 @@ -0,0 +1,2 @@
  93 +SOURCES = neon.cpp
  94 +CONFIG -= x11 qt
  95 diff --git a/configure b/configure
  96 index 2114863..22e6bd4 100755
  97 --- a/configure
  98 +++ b/configure
  99 @@ -745,6 +745,7 @@ CFG_HOST_ENDIAN=auto
 100  CFG_DOUBLEFORMAT=auto
 101  CFG_ARMFPA=auto
 102  CFG_IWMMXT=no
 103 +CFG_NEON=auto
 104  CFG_CLOCK_GETTIME=auto
 105  CFG_CLOCK_MONOTONIC=auto
 106  CFG_MREMAP=auto
 107 @@ -1604,6 +1605,13 @@ while [ "$#" -gt 0 ]; do
 108      iwmmxt)
 109         CFG_IWMMXT="yes"
 110         ;;
 111 +    neon)
 112 +        if [ "$VAL" = "no" ]; then
 113 +            CFG_NEON="$VAL"
 114 +        else
 115 +            UNKNOWN_OPT=yes
 116 +        fi
 117 +        ;;
 118      reduce-relocations)
 119          if [ "$VAL" = "yes" ] || [ "$VAL" = "no" ]; then
 120              CFG_REDUCE_RELOCATIONS="$VAL"
 121 @@ -3889,6 +3897,7 @@ Qt for Embedded Linux only:
 122      -iwmmxt ............ Compile using the iWMMXt instruction set
 123                           (available on some XScale CPUs).
 124
 125 +    -no-neon ........... Do not compile with use of NEON instructions.
 126  EOF
 127  fi
 128
 129 @@ -4509,6 +4518,15 @@ if [ "$CFG_IWMMXT" = "yes" ]; then
 130      fi
 131  fi
 132
 133 +# detect neon support
 134 +if ([ "${CFG_ARCH}" = "arm" ] || [ "${CFG_ARCH}" = "armv6" ]) && [ "${CFG_NEON}" = "auto" ]; then
 135 +    if "$unixtests/compile.test" "$XQMAKESPEC" "$QMAKE_CONFIG" $OPT_VERBOSE "$relpath" "$outpath" config.tests/unix/neon "neon" $L_FLAGS $I_FLAGS $l_FLAGS "-mfpu=neon"; then
 136 +       CFG_NEON=yes
 137 +    else
 138 +       CFG_NEON=no
 139 +    fi
 140 +fi
 141 +
 142  # detect zlib
 143  if [ "$CFG_ZLIB" = "no" ]; then
 144      # Note: Qt no longer support builds without zlib
 145 @@ -6124,6 +6142,7 @@ fi
 146  [ "$CFG_SSE" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG sse"
 147  [ "$CFG_SSE2" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG sse2"
 148  [ "$CFG_IWMMXT" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG iwmmxt"
 149 +[ "$CFG_NEON" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG neon"
 150  [ "$PLATFORM_MAC" = "yes" ] && QMAKE_CONFIG="$QMAKE_CONFIG $CFG_MAC_ARCHS"
 151  if [ "$CFG_IPV6" = "yes" ]; then
 152      QT_CONFIG="$QT_CONFIG ipv6"
 153 @@ -7416,8 +7435,9 @@ echo "Declarative module .. $CFG_DECLARATIVE"
 154  echo "STL support ......... $CFG_STL"
 155  echo "PCH support ......... $CFG_PRECOMPILE"
 156  echo "MMX/3DNOW/SSE/SSE2..  ${CFG_MMX}/${CFG_3DNOW}/${CFG_SSE}/${CFG_SSE2}"
 157 -if [ "${CFG_ARCH}" = "arm" ]; then
 158 +if [ "${CFG_ARCH}" = "arm" ] || [ "${CFG_ARCH}" = "armv6" ]; then
 159      echo "iWMMXt support ...... ${CFG_IWMMXT}"
 160 +    echo "NEON support ........ ${CFG_NEON}"
 161  fi
 162  [ "${PLATFORM_QWS}" != "yes" ] && echo "Graphics System ..... $CFG_GRAPHICS_SYSTEM"
 163  echo "IPv6 support ........ $CFG_IPV6"
 164 diff --git a/src/gui/painting/painting.pri b/src/gui/painting/painting.pri
 165 index 628a109..0b1e79a 100644
 166 --- a/src/gui/painting/painting.pri
 167 +++ b/src/gui/painting/painting.pri
 168 @@ -379,6 +379,13 @@ symbian {
 169          QMAKE_CXXFLAGS.ARMCC *= -O3
 170  }
 171
 172 +neon {
 173 +    DEFINES += QT_HAVE_NEON
 174 +    HEADERS += painting/qdrawhelper_neon_p.h
 175 +    SOURCES += painting/qdrawhelper_neon.cpp
 176 +
 177 +}
 178 +
 179  contains(QT_CONFIG, zlib) {
 180     INCLUDEPATH += ../3rdparty/zlib
 181  } else:!contains(QT_CONFIG, no-zlib) {
 182 diff --git a/src/gui/painting/qblendfunctions.cpp b/src/gui/painting/qblendfunctions.cpp
 183 index 1d15dac..81d1515 100644
 184 --- a/src/gui/painting/qblendfunctions.cpp
 185 +++ b/src/gui/painting/qblendfunctions.cpp
 186 @@ -605,7 +605,7 @@ static void qt_blend_argb32_on_argb32(uchar *destPixels, int dbpl,
 187  }
 188
 189
 190 -static void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
 191 +void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
 192                               const uchar *srcPixels, int sbpl,
 193                               int w, int h,
 194                               int const_alpha)
 195 diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
 196 index 23236ec..84cf5cc 100644
 197 --- a/src/gui/painting/qdrawhelper.cpp
 198 +++ b/src/gui/painting/qdrawhelper.cpp
 199 @@ -44,6 +44,7 @@
 200  #include <private/qpainter_p.h>
 201  #include <private/qdrawhelper_x86_p.h>
 202  #include <private/qdrawhelper_armv6_p.h>
 203 +#include <private/qdrawhelper_neon_p.h>
 204  #include <private/qmath_p.h>
 205  #include <qmath.h>
 206
 207 @@ -7725,7 +7726,8 @@ enum CPUFeatures {
 208      SSE         = 0x10,
 209      SSE2        = 0x20,
 210      CMOV        = 0x40,
 211 -    IWMMXT      = 0x80
 212 +    IWMMXT      = 0x80,
 213 +    NEON        = 0x100
 214  };
 215
 216  static uint detectCPUFeatures()
 217 @@ -7751,6 +7753,9 @@ static uint detectCPUFeatures()
 218      // runtime detection only available when running as a previlegied process
 219      static const bool doIWMMXT = !qgetenv("QT_NO_IWMMXT").toInt();
 220      return doIWMMXT ? IWMMXT : 0;
 221 +#elif defined(QT_HAVE_NEON)
 222 +    static const bool doNEON = !qgetenv("QT_NO_NEON").toInt();
 223 +    return doNEON ? NEON : 0;
 224  #else
 225      uint features = 0;
 226  #if defined(__x86_64__) || defined(Q_OS_WIN64)
 227 @@ -8122,7 +8127,14 @@ void qInitDrawhelperAsm()
 228          qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_armv6;
 229          qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_armv6;
 230          qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_armv6;
 231 -#endif // Q_CC_RVCT && QT_HAVE_ARMV6
 232 +#elif defined(QT_HAVE_NEON)
 233 +        if (features & NEON) {
 234 +            qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_neon;
 235 +            qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_neon;
 236 +            qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_neon;
 237 +            qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_neon;
 238 +        }
 239 +#endif
 240
 241      if (functionForModeSolidAsm) {
 242          const int destinationMode = QPainter::CompositionMode_Destination;
 243 diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp
 244 new file mode 100644
 245 index 0000000..7fe11bf
 246 --- /dev/null
 247 +++ b/src/gui/painting/qdrawhelper_neon.cpp
 248 @@ -0,0 +1,260 @@
 249 +/****************************************************************************
 250 +**
 251 +** Copyright (C) 2009 Nokia Corporation and/or its subsidiary(-ies).
 252 +** All rights reserved.
 253 +** Contact: Nokia Corporation (qt-info@nokia.com)
 254 +**
 255 +** This file is part of the QtGui module of the Qt Toolkit.
 256 +**
 257 +** $QT_BEGIN_LICENSE:LGPL$
 258 +** No Commercial Usage
 259 +** This file contains pre-release code and may not be distributed.
 260 +** You may use this file in accordance with the terms and conditions
 261 +** contained in the Technology Preview License Agreement accompanying
 262 +** this package.
 263 +**
 264 +** GNU Lesser General Public License Usage
 265 +** Alternatively, this file may be used under the terms of the GNU Lesser
 266 +** General Public License version 2.1 as published by the Free Software
 267 +** Foundation and appearing in the file LICENSE.LGPL included in the
 268 +** packaging of this file.  Please review the following information to
 269 +** ensure the GNU Lesser General Public License version 2.1 requirements
 270 +** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
 271 +**
 272 +** In addition, as a special exception, Nokia gives you certain additional
 273 +** rights.  These rights are described in the Nokia Qt LGPL Exception
 274 +** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
 275 +**
 276 +** If you have questions regarding the use of this file, please contact
 277 +** Nokia at qt-info@nokia.com.
 278 +**
 279 +**
 280 +**
 281 +**
 282 +**
 283 +**
 284 +**
 285 +**
 286 +** $QT_END_LICENSE$
 287 +**
 288 +****************************************************************************/
 289 +
 290 +#include <private/qdrawhelper_p.h>
 291 +
 292 +#ifdef QT_HAVE_NEON
 293 +
 294 +#include <private/qdrawhelper_neon_p.h>
 295 +#include <arm_neon.h>
 296 +
 297 +QT_BEGIN_NAMESPACE
 298 +
 299 +static inline int16x8_t qvdiv_255_s16(int16x8_t x, int16x8_t half)
 300 +{
 301 +    // result = (x + (x >> 8) + 0x80) >> 8
 302 +
 303 +    const int16x8_t temp = vshrq_n_s16(x, 8); // x >> 8
 304 +    const int16x8_t sum_part = vaddq_s16(x, half); // x + 0x80
 305 +    const int16x8_t sum = vaddq_s16(temp, sum_part);
 306 +
 307 +    return vreinterpretq_s16_u16(vshrq_n_u16(vreinterpretq_u16_s16(sum), 8));
 308 +}
 309 +
 310 +static inline int16x8_t qvbyte_mul_s16(int16x8_t x, int16x8_t alpha, int16x8_t half)
 311 +{
 312 +    // t = qRound(x * alpha / 255.0)
 313 +
 314 +    const int16x8_t t = vmulq_s16(x, alpha); // t
 315 +    return qvdiv_255_s16(t, half);
 316 +}
 317 +
 318 +static inline int16x8_t qvinterpolate_pixel_255(int16x8_t x, int16x8_t a, int16x8_t y, int16x8_t b, int16x8_t half)
 319 +{
 320 +    // t = x * a + y * b
 321 +
 322 +    const int16x8_t ta = vmulq_s16(x, a);
 323 +    const int16x8_t tb = vmulq_s16(y, b);
 324 +
 325 +    return qvdiv_255_s16(vaddq_s16(ta, tb), half);
 326 +}
 327 +
 328 +static inline int16x8_t qvsource_over_s16(int16x8_t src16, int16x8_t dst16, int16x8_t half, int16x8_t full)
 329 +{
 330 +    const int16x4_t alpha16_high = vdup_lane_s16(vget_high_s16(src16), 3);
 331 +    const int16x4_t alpha16_low = vdup_lane_s16(vget_low_s16(src16), 3);
 332 +
 333 +    const int16x8_t alpha16 = vsubq_s16(full, vcombine_s16(alpha16_low, alpha16_high));
 334 +
 335 +    return vaddq_s16(src16, qvbyte_mul_s16(dst16, alpha16, half));
 336 +}
 337 +
 338 +void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
 339 +                                    const uchar *srcPixels, int sbpl,
 340 +                                    int w, int h,
 341 +                                    int const_alpha)
 342 +{
 343 +    const uint *src = (const uint *) srcPixels;
 344 +    uint *dst = (uint *) destPixels;
 345 +    int16x8_t half = vdupq_n_s16(0x80);
 346 +    int16x8_t full = vdupq_n_s16(0xff);
 347 +    if (const_alpha == 256) {
 348 +        for (int y = 0; y < h; ++y) {
 349 +            int x = 0;
 350 +            for (; x < w-3; x += 4) {
 351 +                int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
 352 +                if ((src[x] & src[x+1] & src[x+2] & src[x+3]) >= 0xff000000) {
 353 +                    // all opaque
 354 +                    vst1q_s32((int32_t *)&dst[x], src32);
 355 +                } else if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
 356 +                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
 357 +
 358 +                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
 359 +                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
 360 +
 361 +                    const uint8x8_t src8_low = vget_low_u8(src8);
 362 +                    const uint8x8_t dst8_low = vget_low_u8(dst8);
 363 +
 364 +                    const uint8x8_t src8_high = vget_high_u8(src8);
 365 +                    const uint8x8_t dst8_high = vget_high_u8(dst8);
 366 +
 367 +                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
 368 +                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
 369 +
 370 +                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
 371 +                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
 372 +
 373 +                    const int16x8_t result16_low = qvsource_over_s16(src16_low, dst16_low, half, full);
 374 +                    const int16x8_t result16_high = qvsource_over_s16(src16_high, dst16_high, half, full);
 375 +
 376 +                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
 377 +                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
 378 +
 379 +                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
 380 +                }
 381 +            }
 382 +            for (; x<w; ++x) {
 383 +                uint s = src[x];
 384 +                if (s >= 0xff000000)
 385 +                    dst[x] = s;
 386 +                else if (s != 0)
 387 +                    dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
 388 +            }
 389 +            dst = (quint32 *)(((uchar *) dst) + dbpl);
 390 +            src = (const quint32 *)(((const uchar *) src) + sbpl);
 391 +        }
 392 +    } else if (const_alpha != 0) {
 393 +        const_alpha = (const_alpha * 255) >> 8;
 394 +        int16x8_t const_alpha16 = vdupq_n_s16(const_alpha);
 395 +        for (int y = 0; y < h; ++y) {
 396 +            int x = 0;
 397 +            for (; x < w-3; x += 4) {
 398 +                if (src[x] | src[x+1] | src[x+2] | src[x+3]) {
 399 +                    int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
 400 +                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
 401 +
 402 +                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
 403 +                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
 404 +
 405 +                    const uint8x8_t src8_low = vget_low_u8(src8);
 406 +                    const uint8x8_t dst8_low = vget_low_u8(dst8);
 407 +
 408 +                    const uint8x8_t src8_high = vget_high_u8(src8);
 409 +                    const uint8x8_t dst8_high = vget_high_u8(dst8);
 410 +
 411 +                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
 412 +                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
 413 +
 414 +                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
 415 +                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
 416 +
 417 +                    const int16x8_t srcalpha16_low = qvbyte_mul_s16(src16_low, const_alpha16, half);
 418 +                    const int16x8_t srcalpha16_high = qvbyte_mul_s16(src16_high, const_alpha16, half);
 419 +
 420 +                    const int16x8_t result16_low = qvsource_over_s16(srcalpha16_low, dst16_low, half, full);
 421 +                    const int16x8_t result16_high = qvsource_over_s16(srcalpha16_high, dst16_high, half, full);
 422 +
 423 +                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
 424 +                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
 425 +
 426 +                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
 427 +                }
 428 +            }
 429 +            for (; x<w; ++x) {
 430 +                uint s = src[x];
 431 +                if (s != 0) {
 432 +                    s = BYTE_MUL(s, const_alpha);
 433 +                    dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
 434 +                }
 435 +            }
 436 +            dst = (quint32 *)(((uchar *) dst) + dbpl);
 437 +            src = (const quint32 *)(((const uchar *) src) + sbpl);
 438 +        }
 439 +    }
 440 +}
 441 +
 442 +// qblendfunctions.cpp
 443 +void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
 444 +                             const uchar *srcPixels, int sbpl,
 445 +                             int w, int h,
 446 +                             int const_alpha);
 447 +
 448 +void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl,
 449 +                                  const uchar *srcPixels, int sbpl,
 450 +                                  int w, int h,
 451 +                                  int const_alpha)
 452 +{
 453 +    if (const_alpha != 256) {
 454 +        if (const_alpha != 0) {
 455 +            const uint *src = (const uint *) srcPixels;
 456 +            uint *dst = (uint *) destPixels;
 457 +            int16x8_t half = vdupq_n_s16(0x80);
 458 +            const_alpha = (const_alpha * 255) >> 8;
 459 +            int one_minus_const_alpha = 255 - const_alpha;
 460 +            int16x8_t const_alpha16 = vdupq_n_s16(const_alpha);
 461 +            int16x8_t one_minus_const_alpha16 = vdupq_n_s16(255 - const_alpha);
 462 +            for (int y = 0; y < h; ++y) {
 463 +                int x = 0;
 464 +                for (; x < w-3; x += 4) {
 465 +                    int32x4_t src32 = vld1q_s32((int32_t *)&src[x]);
 466 +                    int32x4_t dst32 = vld1q_s32((int32_t *)&dst[x]);
 467 +
 468 +                    const uint8x16_t src8 = vreinterpretq_u8_s32(src32);
 469 +                    const uint8x16_t dst8 = vreinterpretq_u8_s32(dst32);
 470 +
 471 +                    const uint8x8_t src8_low = vget_low_u8(src8);
 472 +                    const uint8x8_t dst8_low = vget_low_u8(dst8);
 473 +
 474 +                    const uint8x8_t src8_high = vget_high_u8(src8);
 475 +                    const uint8x8_t dst8_high = vget_high_u8(dst8);
 476 +
 477 +                    const int16x8_t src16_low = vreinterpretq_s16_u16(vmovl_u8(src8_low));
 478 +                    const int16x8_t dst16_low = vreinterpretq_s16_u16(vmovl_u8(dst8_low));
 479 +
 480 +                    const int16x8_t src16_high = vreinterpretq_s16_u16(vmovl_u8(src8_high));
 481 +                    const int16x8_t dst16_high = vreinterpretq_s16_u16(vmovl_u8(dst8_high));
 482 +
 483 +                    const int16x8_t result16_low = qvinterpolate_pixel_255(src16_low, const_alpha16, dst16_low, one_minus_const_alpha16, half);
 484 +                    const int16x8_t result16_high = qvinterpolate_pixel_255(src16_high, const_alpha16, dst16_high, one_minus_const_alpha16, half);
 485 +
 486 +                    const int32x2_t result32_low = vreinterpret_s32_s8(vmovn_s16(result16_low));
 487 +                    const int32x2_t result32_high = vreinterpret_s32_s8(vmovn_s16(result16_high));
 488 +
 489 +                    vst1q_s32((int32_t *)&dst[x], vcombine_s32(result32_low, result32_high));
 490 +                }
 491 +                for (; x<w; ++x) {
 492 +                    uint s = src[x];
 493 +                    s = BYTE_MUL(s, const_alpha);
 494 +                    dst[x] = INTERPOLATE_PIXEL_255(src[x], const_alpha, dst[x], one_minus_const_alpha);
 495 +                }
 496 +                dst = (quint32 *)(((uchar *) dst) + dbpl);
 497 +                src = (const quint32 *)(((const uchar *) src) + sbpl);
 498 +            }
 499 +        }
 500 +    } else {
 501 +        qt_blend_rgb32_on_rgb32(destPixels, dbpl, srcPixels, sbpl, w, h, const_alpha);
 502 +    }
 503 +}
 504 +
 505 +QT_END_NAMESPACE
 506 +
 507 +#endif // QT_HAVE_NEON
 508 +
 509 diff --git a/src/gui/painting/qdrawhelper_neon_p.h b/src/gui/painting/qdrawhelper_neon_p.h
 510 new file mode 100644
 511 index 0000000..cb9a0d6
 512 --- /dev/null
 513 +++ b/src/gui/painting/qdrawhelper_neon_p.h
 514 @@ -0,0 +1,76 @@
 515 +/****************************************************************************
 516 +**
 517 +** Copyright (C) 2009 Nokia Corporation and/or its subsidiary(-ies).
 518 +** All rights reserved.
 519 +** Contact: Nokia Corporation (qt-info@nokia.com)
 520 +**
 521 +** This file is part of the QtGui module of the Qt Toolkit.
 522 +**
 523 +** $QT_BEGIN_LICENSE:LGPL$
 524 +** No Commercial Usage
 525 +** This file contains pre-release code and may not be distributed.
 526 +** You may use this file in accordance with the terms and conditions
 527 +** contained in the Technology Preview License Agreement accompanying
 528 +** this package.
 529 +**
 530 +** GNU Lesser General Public License Usage
 531 +** Alternatively, this file may be used under the terms of the GNU Lesser
 532 +** General Public License version 2.1 as published by the Free Software
 533 +** Foundation and appearing in the file LICENSE.LGPL included in the
 534 +** packaging of this file.  Please review the following information to
 535 +** ensure the GNU Lesser General Public License version 2.1 requirements
 536 +** will be met: http://www.gnu.org/licenses/old-licenses/lgpl-2.1.html.
 537 +**
 538 +** In addition, as a special exception, Nokia gives you certain additional
 539 +** rights.  These rights are described in the Nokia Qt LGPL Exception
 540 +** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
 541 +**
 542 +** If you have questions regarding the use of this file, please contact
 543 +** Nokia at qt-info@nokia.com.
 544 +**
 545 +**
 546 +**
 547 +**
 548 +**
 549 +**
 550 +**
 551 +**
 552 +** $QT_END_LICENSE$
 553 +**
 554 +****************************************************************************/
 555 +
 556 +#ifndef QDRAWHELPER_NEON_P_H
 557 +#define QDRAWHELPER_NEON_P_H
 558 +
 559 +//
 560 +//  W A R N I N G
 561 +//  -------------
 562 +//
 563 +// This file is not part of the Qt API.  It exists purely as an
 564 +// implementation detail.  This header file may change from version to
 565 +// version without notice, or even be removed.
 566 +//
 567 +// We mean it.
 568 +//
 569 +
 570 +#include <private/qdrawhelper_p.h>
 571 +
 572 +QT_BEGIN_NAMESPACE
 573 +
 574 +#ifdef QT_HAVE_NEON
 575 +
 576 +void qt_blend_argb32_on_argb32_neon(uchar *destPixels, int dbpl,
 577 +                                            const uchar *srcPixels, int sbpl,
 578 +                                            int w, int h,
 579 +                                            int const_alpha);
 580 +
 581 +void qt_blend_rgb32_on_rgb32_neon(uchar *destPixels, int dbpl,
 582 +                                  const uchar *srcPixels, int sbpl,
 583 +                                  int w, int h,
 584 +                                  int const_alpha);
 585 +
 586 +#endif // QT_HAVE_NEON
 587 +
 588 +QT_END_NAMESPACE
 589 +
 590 +#endif // QDRAWHELPER_NEON_P_H
 591 --
 592 1.6.5
 593