Merge branch 'master' of scm.dev.nokia.troll.no:qt/oslo-staging-1 into master-integration

* 'master' of scm.dev.nokia.troll.no:qt/oslo-staging-1:
  Build fixes for the SIMD functions
  Implement QString::fromLatin1_helper() with SSE2
  Cache the result of qDetectCPUFeatures()
  Move the SIMD detection from QtGui to QtCore

7 files changed, 427 insertions, 208 deletions

diff --git a/src/corelib/corelib.pro b/src/corelib/corelib.pro
index efee610..df9a463 100644
--- a/src/corelib/corelib.pro
+++ b/src/corelib/corelib.pro
@@ -52,3 +52,25 @@ symbian: {
     partial_upgrade.path = c:/sys/bin
     DEPLOYMENT = partial_upgrade $$DEPLOYMENT
 }
+
+mmx {
+    DEFINES += QT_HAVE_MMX
+}
+3dnow {
+    DEFINES += QT_HAVE_3DNOW
+}
+sse {
+    DEFINES += QT_HAVE_SSE
+    DEFINES += QT_HAVE_MMXEXT
+}
+sse2 {
+    DEFINES += QT_HAVE_SSE2
+}
+iwmmxt {
+    DEFINES += QT_HAVE_IWMMXT
+}
+neon {
+    DEFINES += QT_HAVE_NEON
+    QMAKE_CXXFLAGS *= -mfpu=neon
+}
+
diff --git a/src/corelib/tools/qsimd.cpp b/src/corelib/tools/qsimd.cpp
new file mode 100644
index 0000000..52d2cea
--- /dev/null
+++ b/src/corelib/tools/qsimd.cpp
@@ -0,0 +1,246 @@
+/****************************************************************************
+**
+** Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies).
+** All rights reserved.
+** Contact: Nokia Corporation (qt-info@nokia.com)
+**
+** This file is part of the QtCore module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** No Commercial Usage
+** This file contains pre-release code and may not be distributed.
+** You may use this file in accordance with the terms and conditions
+** contained in the Technology Preview License Agreement accompanying
+** this package.
+**
+** 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, Nokia gives you certain additional
+** rights.  These rights are described in the Nokia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** If you have questions regarding the use of this file, please contact
+** Nokia at qt-info@nokia.com.
+**
+**
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#include "qsimd_p.h"
+#include <QByteArray>
+
+QT_BEGIN_NAMESPACE
+
+uint qDetectCPUFeatures()
+{
+    static uint features = 0xffffffff;
+    if (features != 0xffffffff)
+        return features;
+
+#if defined (Q_OS_WINCE)
+#if defined (ARM)
+    if (IsProcessorFeaturePresent(PF_ARM_INTEL_WMMX)) {
+        features = IWMMXT;
+        return features;
+    }
+#elif defined(_X86_)
+    features = 0;
+#if defined QT_HAVE_MMX
+    if (IsProcessorFeaturePresent(PF_MMX_INSTRUCTIONS_AVAILABLE))
+        features |= MMX;
+#endif
+#if defined QT_HAVE_3DNOW
+    if (IsProcessorFeaturePresent(PF_3DNOW_INSTRUCTIONS_AVAILABLE))
+        features |= MMX3DNOW;
+#endif
+    return features;
+#endif
+    features = 0;
+    return features;
+#elif defined(QT_HAVE_IWMMXT)
+    // runtime detection only available when running as a previlegied process
+    static const bool doIWMMXT = !qgetenv("QT_NO_IWMMXT").toInt();
+    features = doIWMMXT ? IWMMXT : 0;
+    return features;
+#elif defined(QT_HAVE_NEON)
+    static const bool doNEON = !qgetenv("QT_NO_NEON").toInt();
+    features = doNEON ? NEON : 0;
+    return features;
+#else
+    features = 0;
+#if defined(__x86_64__) || defined(Q_OS_WIN64)
+    features = MMX|SSE|SSE2|CMOV;
+#elif defined(__ia64__)
+    features = MMX|SSE|SSE2;
+#elif defined(__i386__) || defined(_M_IX86)
+    unsigned int extended_result = 0;
+    uint result = 0;
+    /* see p. 118 of amd64 instruction set manual Vol3 */
+#if defined(Q_CC_GNU)
+    asm ("push %%ebx\n"
+         "pushf\n"
+         "pop %%eax\n"
+         "mov %%eax, %%ebx\n"
+         "xor $0x00200000, %%eax\n"
+         "push %%eax\n"
+         "popf\n"
+         "pushf\n"
+         "pop %%eax\n"
+         "xor %%edx, %%edx\n"
+         "xor %%ebx, %%eax\n"
+         "jz 1f\n"
+
+         "mov $0x00000001, %%eax\n"
+         "cpuid\n"
+         "1:\n"
+         "pop %%ebx\n"
+         "mov %%edx, %0\n"
+        : "=r" (result)
+        :
+        : "%eax", "%ecx", "%edx"
+        );
+
+    asm ("push %%ebx\n"
+         "pushf\n"
+         "pop %%eax\n"
+         "mov %%eax, %%ebx\n"
+         "xor $0x00200000, %%eax\n"
+         "push %%eax\n"
+         "popf\n"
+         "pushf\n"
+         "pop %%eax\n"
+         "xor %%edx, %%edx\n"
+         "xor %%ebx, %%eax\n"
+         "jz 2f\n"
+
+         "mov $0x80000000, %%eax\n"
+         "cpuid\n"
+         "cmp $0x80000000, %%eax\n"
+         "jbe 2f\n"
+         "mov $0x80000001, %%eax\n"
+         "cpuid\n"
+         "2:\n"
+         "pop %%ebx\n"
+         "mov %%edx, %0\n"
+        : "=r" (extended_result)
+        :
+        : "%eax", "%ecx", "%edx"
+        );
+#elif defined (Q_OS_WIN)
+    _asm {
+        push eax
+        push ebx
+        push ecx
+        push edx
+        pushfd
+        pop eax
+        mov ebx, eax
+        xor eax, 00200000h
+        push eax
+        popfd
+        pushfd
+        pop eax
+        mov edx, 0
+        xor eax, ebx
+        jz skip
+
+        mov eax, 1
+        cpuid
+        mov result, edx
+    skip:
+        pop edx
+        pop ecx
+        pop ebx
+        pop eax
+    }
+
+    _asm {
+        push eax
+        push ebx
+        push ecx
+        push edx
+        pushfd
+        pop eax
+        mov ebx, eax
+        xor eax, 00200000h
+        push eax
+        popfd
+        pushfd
+        pop eax
+        mov edx, 0
+        xor eax, ebx
+        jz skip2
+
+        mov eax, 80000000h
+        cpuid
+        cmp eax, 80000000h
+        jbe skip2
+        mov eax, 80000001h
+        cpuid
+        mov extended_result, edx
+    skip2:
+        pop edx
+        pop ecx
+        pop ebx
+        pop eax
+    }
+#endif
+
+    // result now contains the standard feature bits
+    if (result & (1u << 15))
+        features |= CMOV;
+    if (result & (1u << 23))
+        features |= MMX;
+    if (extended_result & (1u << 22))
+        features |= MMXEXT;
+    if (extended_result & (1u << 31))
+        features |= MMX3DNOW;
+    if (extended_result & (1u << 30))
+        features |= MMX3DNOWEXT;
+    if (result & (1u << 25))
+        features |= SSE;
+    if (result & (1u << 26))
+        features |= SSE2;
+#endif // i386
+
+#if defined(QT_HAVE_MMX)
+    if (qgetenv("QT_NO_MMX").toInt())
+        features ^= MMX;
+#endif
+    if (qgetenv("QT_NO_MMXEXT").toInt())
+        features ^= MMXEXT;
+
+#if defined(QT_HAVE_3DNOW)
+    if (qgetenv("QT_NO_3DNOW").toInt())
+        features ^= MMX3DNOW;
+#endif
+    if (qgetenv("QT_NO_3DNOWEXT").toInt())
+        features ^= MMX3DNOWEXT;
+
+#if defined(QT_HAVE_SSE)
+    if (qgetenv("QT_NO_SSE").toInt())
+        features ^= SSE;
+#endif
+#if defined(QT_HAVE_SSE2)
+    if (qgetenv("QT_NO_SSE2").toInt())
+        features ^= SSE2;
+#endif
+
+    return features;
+#endif
+}
+
+QT_END_NAMESPACE
diff --git a/src/corelib/tools/qsimd_p.h b/src/corelib/tools/qsimd_p.h
new file mode 100644
index 0000000..07fae97
--- /dev/null
+++ b/src/corelib/tools/qsimd_p.h
@@ -0,0 +1,133 @@
+/****************************************************************************
+**
+** Copyright (C) 2010 Nokia Corporation and/or its subsidiary(-ies).
+** All rights reserved.
+** Contact: Nokia Corporation (qt-info@nokia.com)
+**
+** This file is part of the QtCore module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** No Commercial Usage
+** This file contains pre-release code and may not be distributed.
+** You may use this file in accordance with the terms and conditions
+** contained in the Technology Preview License Agreement accompanying
+** this package.
+**
+** 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, Nokia gives you certain additional
+** rights.  These rights are described in the Nokia Qt LGPL Exception
+** version 1.1, included in the file LGPL_EXCEPTION.txt in this package.
+**
+** If you have questions regarding the use of this file, please contact
+** Nokia at qt-info@nokia.com.
+**
+**
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#ifndef QSIMD_P_H
+#define QSIMD_P_H
+
+#include <qglobal.h>
+
+
+QT_BEGIN_HEADER
+
+QT_BEGIN_NAMESPACE
+
+QT_MODULE(Core)
+
+#if defined(QT_NO_MAC_XARCH) || (defined(Q_OS_DARWIN) && (defined(__ppc__) || defined(__ppc64__)))
+// Disable MMX and SSE on Mac/PPC builds, or if the compiler
+// does not support -Xarch argument passing
+#undef QT_HAVE_SSE2
+#undef QT_HAVE_SSE
+#undef QT_HAVE_3DNOW
+#undef QT_HAVE_MMX
+#endif
+
+
+#if defined(__x86_64__) || defined(Q_OS_WIN64) || defined(__ia64__) || defined(__SSE2__)
+#if defined(QT_HAVE_SSE2)
+// Defined for small fast functions that can take advantages of SSE2 intrinsics
+#define QT_ALWAYS_HAVE_SSE2
+#endif
+#endif
+
+
+// SSE intrinsics
+#if defined(QT_HAVE_SSE2) || defined(QT_HAVE_SSE) || defined(QT_HAVE_MMX)
+#if defined(QT_LINUXBASE)
+/// this is an evil hack - the posix_memalign declaration in LSB
+/// is wrong - see http://bugs.linuxbase.org/show_bug.cgi?id=2431
+#  define posix_memalign _lsb_hack_posix_memalign
+#  include <emmintrin.h>
+#  undef posix_memalign
+#else
+#  include <emmintrin.h>
+#endif
+#endif
+
+// NEON intrinsics
+#if defined(QT_HAVE_NEON)
+#include <arm_neon.h>
+#endif
+
+
+// IWMMXT intrinsics
+#if defined(QT_HAVE_IWMMXT)
+#include <mmintrin.h>
+#if defined(Q_OS_WINCE)
+#  include "qplatformdefs.h"
+#endif
+#endif
+
+#if defined(QT_HAVE_IWMMXT)
+#if !defined(__IWMMXT__) && !defined(Q_OS_WINCE)
+#  include <xmmintrin.h>
+#elif defined(Q_OS_WINCE_STD) && defined(_X86_)
+#  pragma warning(disable: 4391)
+#  include <xmmintrin.h>
+#endif
+#endif
+
+// 3D now intrinsics
+#if defined(QT_HAVE_3DNOW)
+#include <mm3dnow.h>
+#endif
+
+enum CPUFeatures {
+    None        = 0,
+    MMX         = 0x1,
+    MMXEXT      = 0x2,
+    MMX3DNOW    = 0x4,
+    MMX3DNOWEXT = 0x8,
+    SSE         = 0x10,
+    SSE2        = 0x20,
+    CMOV        = 0x40,
+    IWMMXT      = 0x80,
+    NEON        = 0x100
+};
+
+
+QT_END_NAMESPACE
+
+QT_END_HEADER
+
+Q_CORE_EXPORT uint qDetectCPUFeatures();
+
+#endif // QSIMD_P_H
diff --git a/src/corelib/tools/qstring.cpp b/src/corelib/tools/qstring.cpp
index ac1bee7..9431ef4 100644
--- a/src/corelib/tools/qstring.cpp
+++ b/src/corelib/tools/qstring.cpp
@@ -46,6 +46,7 @@
 #include <qtextcodec.h>
 #endif
 #include <private/qutfcodec_p.h>
+#include "qsimd_p.h"
 #include <qdatastream.h>
 #include <qlist.h>
 #include "qlocale.h"
@@ -3647,10 +3648,29 @@ QString::Data *QString::fromLatin1_helper(const char *str, int size)
         d->alloc = d->size = size;
         d->clean = d->asciiCache = d->simpletext = d->righttoleft = d->capacity = 0;
         d->data = d->array;
-        ushort *i = d->data;
         d->array[size] = '\0';
+        ushort *dst = d->data;
+#if defined(QT_ALWAYS_HAVE_SSE2)
+        if (size >= 16) {
+            int chunkCount = size >> 4; // divided by 16
+            const __m128i nullMask = _mm_set1_epi32(0);
+            for (int i = 0; i < chunkCount; ++i) {
+                const __m128i chunk = _mm_loadu_si128((__m128i*)str);
+                str += 16;
+
+                const __m128i firstHalf = _mm_unpacklo_epi8(chunk, nullMask);
+                _mm_storeu_si128((__m128i*)dst, firstHalf);
+                dst += 8;
+
+                const __m128i secondHalf = _mm_unpackhi_epi8 (chunk, nullMask);
+                _mm_storeu_si128((__m128i*)dst, secondHalf);
+                dst += 8;
+            }
+            size = size % 16;
+        }
+#endif
         while (size--)
-            *i++ = (uchar)*str++;
+            *dst++ = (uchar)*str++;
     }
     return d;
 }
diff --git a/src/corelib/tools/tools.pri b/src/corelib/tools/tools.pri
index 3406e41..6d64915 100644
--- a/src/corelib/tools/tools.pri
+++ b/src/corelib/tools/tools.pri
@@ -33,6 +33,7 @@ HEADERS +=  \
         tools/qsharedpointer.h \
         tools/qsharedpointer_impl.h \
         tools/qset.h \
+        tools/qsimd_p.h \
         tools/qsize.h \
         tools/qstack.h \
         tools/qstring.h \
@@ -68,6 +69,7 @@ SOURCES += \
         tools/qregexp.cpp \
         tools/qshareddata.cpp \
         tools/qsharedpointer.cpp \
+        tools/qsimd.cpp \
         tools/qsize.cpp \
         tools/qstring.cpp \
         tools/qstringbuilder.cpp \
diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index 070491d..c37e49a 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -46,6 +46,7 @@
 #include <private/qdrawhelper_armv6_p.h>
 #include <private/qdrawhelper_neon_p.h>
 #include <private/qmath_p.h>
+#include <private/qsimd_p.h>
 #include <qmath.h>
 
 QT_BEGIN_NAMESPACE
@@ -7720,199 +7721,6 @@ static void qt_memfill16_setup(quint16 *dest, quint16 value, int count);
 qt_memfill32_func qt_memfill32 = qt_memfill32_setup;
 qt_memfill16_func qt_memfill16 = qt_memfill16_setup;
 
-enum CPUFeatures {
-    None        = 0,
-    MMX         = 0x1,
-    MMXEXT      = 0x2,
-    MMX3DNOW    = 0x4,
-    MMX3DNOWEXT = 0x8,
-    SSE         = 0x10,
-    SSE2        = 0x20,
-    CMOV        = 0x40,
-    IWMMXT      = 0x80,
-    NEON        = 0x100
-};
-
-static uint detectCPUFeatures()
-{
-#if defined (Q_OS_WINCE)
-#if defined (ARM)
-    if (IsProcessorFeaturePresent(PF_ARM_INTEL_WMMX))
-        return IWMMXT;
-#elif defined(_X86_)
-    uint features = 0;
-#if defined QT_HAVE_MMX
-    if (IsProcessorFeaturePresent(PF_MMX_INSTRUCTIONS_AVAILABLE))
-        features |= MMX;
-#endif
-#if defined QT_HAVE_3DNOW
-    if (IsProcessorFeaturePresent(PF_3DNOW_INSTRUCTIONS_AVAILABLE))
-        features |= MMX3DNOW;
-#endif
-    return features;
-#endif
-    return 0;
-#elif defined(QT_HAVE_IWMMXT)
-    // runtime detection only available when running as a previlegied process
-    static const bool doIWMMXT = !qgetenv("QT_NO_IWMMXT").toInt();
-    return doIWMMXT ? IWMMXT : 0;
-#elif defined(QT_HAVE_NEON)
-    static const bool doNEON = !qgetenv("QT_NO_NEON").toInt();
-    return doNEON ? NEON : 0;
-#else
-    uint features = 0;
-#if defined(__x86_64__) || defined(Q_OS_WIN64)
-    features = MMX|SSE|SSE2|CMOV;
-#elif defined(__ia64__)
-    features = MMX|SSE|SSE2;
-#elif defined(__i386__) || defined(_M_IX86)
-    unsigned int extended_result = 0;
-    uint result = 0;
-    /* see p. 118 of amd64 instruction set manual Vol3 */
-#if defined(Q_CC_GNU)
-    asm ("push %%ebx\n"
-         "pushf\n"
-         "pop %%eax\n"
-         "mov %%eax, %%ebx\n"
-         "xor $0x00200000, %%eax\n"
-         "push %%eax\n"
-         "popf\n"
-         "pushf\n"
-         "pop %%eax\n"
-         "xor %%edx, %%edx\n"
-         "xor %%ebx, %%eax\n"
-         "jz 1f\n"
-
-         "mov $0x00000001, %%eax\n"
-         "cpuid\n"
-         "1:\n"
-         "pop %%ebx\n"
-         "mov %%edx, %0\n"
-        : "=r" (result)
-        :
-        : "%eax", "%ecx", "%edx"
-        );
-
-    asm ("push %%ebx\n"
-         "pushf\n"
-         "pop %%eax\n"
-         "mov %%eax, %%ebx\n"
-         "xor $0x00200000, %%eax\n"
-         "push %%eax\n"
-         "popf\n"
-         "pushf\n"
-         "pop %%eax\n"
-         "xor %%edx, %%edx\n"
-         "xor %%ebx, %%eax\n"
-         "jz 2f\n"
-
-         "mov $0x80000000, %%eax\n"
-         "cpuid\n"
-         "cmp $0x80000000, %%eax\n"
-         "jbe 2f\n"
-         "mov $0x80000001, %%eax\n"
-         "cpuid\n"
-         "2:\n"
-         "pop %%ebx\n"
-         "mov %%edx, %0\n"
-        : "=r" (extended_result)
-        :
-        : "%eax", "%ecx", "%edx"
-        );
-#elif defined (Q_OS_WIN)
-    _asm {
-        push eax
-        push ebx
-        push ecx
-        push edx
-        pushfd
-        pop eax
-        mov ebx, eax
-        xor eax, 00200000h
-        push eax
-        popfd
-        pushfd
-        pop eax
-        mov edx, 0
-        xor eax, ebx
-        jz skip
-
-        mov eax, 1
-        cpuid
-        mov result, edx
-    skip:
-        pop edx
-        pop ecx
-        pop ebx
-        pop eax
-    }
-
-    _asm {
-        push eax
-        push ebx
-        push ecx
-        push edx
-        pushfd
-        pop eax
-        mov ebx, eax
-        xor eax, 00200000h
-        push eax
-        popfd
-        pushfd
-        pop eax
-        mov edx, 0
-        xor eax, ebx
-        jz skip2
-
-        mov eax, 80000000h
-        cpuid
-        cmp eax, 80000000h
-        jbe skip2
-        mov eax, 80000001h
-        cpuid
-        mov extended_result, edx
-    skip2:
-        pop edx
-        pop ecx
-        pop ebx
-        pop eax
-    }
-#endif
-
-    // result now contains the standard feature bits
-    if (result & (1u << 15))
-        features |= CMOV;
-    if (result & (1u << 23))
-        features |= MMX;
-    if (extended_result & (1u << 22))
-        features |= MMXEXT;
-    if (extended_result & (1u << 31))
-        features |= MMX3DNOW;
-    if (extended_result & (1u << 30))
-        features |= MMX3DNOWEXT;
-    if (result & (1u << 25))
-        features |= SSE;
-    if (result & (1u << 26))
-        features |= SSE2;
-#endif // i386
-
-    if (qgetenv("QT_NO_MMX").toInt())
-        features ^= MMX;
-    if (qgetenv("QT_NO_MMXEXT").toInt())
-        features ^= MMXEXT;
-    if (qgetenv("QT_NO_3DNOW").toInt())
-        features ^= MMX3DNOW;
-    if (qgetenv("QT_NO_3DNOWEXT").toInt())
-        features ^= MMX3DNOWEXT;
-    if (qgetenv("QT_NO_SSE").toInt())
-        features ^= SSE;
-    if (qgetenv("QT_NO_SSE2").toInt())
-        features ^= SSE2;
-
-    return features;
-#endif
-}
-
 #if defined(Q_CC_RVCT) && defined(QT_HAVE_ARMV6)
 // Move these to qdrawhelper_arm.c when all
 // functions are implemented using arm assembly.
@@ -8005,10 +7813,7 @@ static void qt_blend_color_argb_armv6(int count, const QSpan *spans, void *userD
 
 void qInitDrawhelperAsm()
 {
-    static uint features = 0xffffffff;
-    if (features != 0xffffffff)
-        return;
-    features = detectCPUFeatures();
+    const uint features = qDetectCPUFeatures();
 
     qt_memfill32 = qt_memfill_template<quint32, quint32>;
     qt_memfill16 = qt_memfill_quint16; //qt_memfill_template<quint16, quint16>;
diff --git a/src/gui/painting/qdrawhelper_p.h b/src/gui/painting/qdrawhelper_p.h
index cb0db4f..df4d9c5 100644
--- a/src/gui/painting/qdrawhelper_p.h
+++ b/src/gui/painting/qdrawhelper_p.h
@@ -67,15 +67,6 @@
 #include "QtGui/qscreen_qws.h"
 #endif
 
-// Disable MMX and SSE on Mac/PPC builds, or if the compiler
-// does not support -Xarch argument passing
-#if defined(QT_NO_MAC_XARCH) || (defined(Q_OS_DARWIN) && (defined(__ppc__) || defined(__ppc64__)))
-#undef QT_HAVE_SSE2
-#undef QT_HAVE_SSE
-#undef QT_HAVE_3DNOW
-#undef QT_HAVE_MMX
-#endif
-
 QT_BEGIN_NAMESPACE
 
 #if defined(Q_CC_MSVC) && _MSCVER <= 1300 && !defined(Q_CC_INTEL)