Add AVX support for the painting and image code.

There are no new routines, this is just the old SSE2 and SSSE3 code
compiled in AVX mode, meaning the instructions use the VEX prefix.

Change-Id: I79a8bfaf6b30a050618db899f5a3bbc220449f0b
Reviewed-by: Samuel Rødal <samuel.rodal@nokia.com>

9 files changed, 246 insertions, 11 deletions

diff --git a/src/gui/gui.pro b/src/gui/gui.pro
index 2d21a7d7b2..3940705b50 100644
--- a/src/gui/gui.pro
+++ b/src/gui/gui.pro
@@ -84,6 +84,18 @@ win32:!contains(QT_CONFIG, directwrite) {
             silent:ssse3_compiler.commands = @echo compiling[ssse3] ${QMAKE_FILE_IN} && $$ssse3_compiler.commands
             QMAKE_EXTRA_COMPILERS += ssse3_compiler
         }
+        avx {
+            avx_compiler.commands = $$QMAKE_CXX -c -Winline
+            avx_compiler.commands += -mavx
+            avx_compiler.commands += $(CXXFLAGS) $(INCPATH) ${QMAKE_FILE_IN} -o ${QMAKE_FILE_OUT}
+            avx_compiler.dependency_type = TYPE_C
+            avx_compiler.output = ${QMAKE_VAR_OBJECTS_DIR}${QMAKE_FILE_BASE}$${first(QMAKE_EXT_OBJ)}
+            avx_compiler.input = AVX_SOURCES
+            avx_compiler.variable_out = OBJECTS
+            avx_compiler.name = compiling[avx] ${QMAKE_FILE_IN}
+            silent:avx_compiler.commands = @echo compiling[avx] ${QMAKE_FILE_IN} && $$avx_compiler.commands
+            QMAKE_EXTRA_COMPILERS += avx_compiler
+        }
         iwmmxt {
             iwmmxt_compiler.commands = $$QMAKE_CXX -c -Winline
             iwmmxt_compiler.commands += -mcpu=iwmmxt
diff --git a/src/gui/image/image.pri b/src/gui/image/image.pri
index 6b306ddaad..f29385d0a5 100644
--- a/src/gui/image/image.pri
+++ b/src/gui/image/image.pri
@@ -77,3 +77,4 @@ contains(QT_CONFIG, gif):include($$PWD/qgifhandler.pri)
 NEON_SOURCES += image/qimage_neon.cpp
 SSE2_SOURCES += image/qimage_sse2.cpp
 SSSE3_SOURCES += image/qimage_ssse3.cpp
+AVX_SOURCES += image/qimage_avx.cpp
diff --git a/src/gui/image/qimage.cpp b/src/gui/image/qimage.cpp
index 89060cfbb4..2d70b28923 100644
--- a/src/gui/image/qimage.cpp
+++ b/src/gui/image/qimage.cpp
@@ -3273,19 +3273,32 @@ void qInitImageConversions()
     Q_UNUSED(features);
 
 #ifdef QT_HAVE_SSE2
-    if (features & SSE2) {
-        extern bool convert_ARGB_to_ARGB_PM_inplace_sse2(QImageData *data, Qt::ImageConversionFlags);
-        inplace_converter_map[QImage::Format_ARGB32][QImage::Format_ARGB32_Premultiplied] = convert_ARGB_to_ARGB_PM_inplace_sse2;
-    }
+#ifdef QT_HAVE_AVX
+    if (features & AVX) {
+        extern bool convert_ARGB_to_ARGB_PM_inplace_avx(QImageData *data, Qt::ImageConversionFlags);
+        inplace_converter_map[QImage::Format_ARGB32][QImage::Format_ARGB32_Premultiplied] = convert_ARGB_to_ARGB_PM_inplace_avx;
+
+        extern void convert_RGB888_to_RGB32_avx(QImageData *dest, const QImageData *src, Qt::ImageConversionFlags);
+        converter_map[QImage::Format_RGB888][QImage::Format_RGB32] = convert_RGB888_to_RGB32_avx;
+        converter_map[QImage::Format_RGB888][QImage::Format_ARGB32] = convert_RGB888_to_RGB32_avx;
+        converter_map[QImage::Format_RGB888][QImage::Format_ARGB32_Premultiplied] = convert_RGB888_to_RGB32_avx;
+    } else
 #endif
+
+        if (features & SSE2) {
+            extern bool convert_ARGB_to_ARGB_PM_inplace_sse2(QImageData *data, Qt::ImageConversionFlags);
+            inplace_converter_map[QImage::Format_ARGB32][QImage::Format_ARGB32_Premultiplied] = convert_ARGB_to_ARGB_PM_inplace_sse2;
 #ifdef QT_HAVE_SSSE3
-    if (features & SSSE3) {
-        extern void convert_RGB888_to_RGB32_ssse3(QImageData *dest, const QImageData *src, Qt::ImageConversionFlags);
-        converter_map[QImage::Format_RGB888][QImage::Format_RGB32] = convert_RGB888_to_RGB32_ssse3;
-        converter_map[QImage::Format_RGB888][QImage::Format_ARGB32] = convert_RGB888_to_RGB32_ssse3;
-        converter_map[QImage::Format_RGB888][QImage::Format_ARGB32_Premultiplied] = convert_RGB888_to_RGB32_ssse3;
-    }
+            if (features & SSSE3) {
+                extern void convert_RGB888_to_RGB32_ssse3(QImageData *dest, const QImageData *src, Qt::ImageConversionFlags);
+                converter_map[QImage::Format_RGB888][QImage::Format_RGB32] = convert_RGB888_to_RGB32_ssse3;
+                converter_map[QImage::Format_RGB888][QImage::Format_ARGB32] = convert_RGB888_to_RGB32_ssse3;
+                converter_map[QImage::Format_RGB888][QImage::Format_ARGB32_Premultiplied] = convert_RGB888_to_RGB32_ssse3;
+            }
 #endif
+        }
+#endif // SSE2
+
 #ifdef QT_HAVE_NEON
     if (features & NEON) {
         extern void convert_RGB888_to_RGB32_neon(QImageData *dest, const QImageData *src, Qt::ImageConversionFlags);
diff --git a/src/gui/image/qimage_avx.cpp b/src/gui/image/qimage_avx.cpp
new file mode 100644
index 0000000000..f112a46d96
--- /dev/null
+++ b/src/gui/image/qimage_avx.cpp
@@ -0,0 +1,57 @@
+/****************************************************************************
+**
+** Copyright (C) 2012 Intel Corporation
+** Contact: http://www.qt-project.org/
+**
+** This file is part of the QtGui module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** GNU Lesser General Public License Usage
+** 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.
+**
+** 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.
+**
+** Other Usage
+** Alternatively, this file may be used in accordance with the terms and
+** conditions contained in a signed written agreement between you and Nokia.
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#include <private/qsimd_p.h>
+
+#ifdef QT_HAVE_AVX
+
+#ifndef __AVX__
+#error "AVX not enabled in this file, cannot proceed"
+#endif
+
+#define convert_ARGB_to_ARGB_PM_inplace_sse2 convert_ARGB_to_ARGB_PM_inplace_avx
+#include "qimage_sse2.cpp"
+
+#define qt_convert_rgb888_to_rgb32_ssse3 qt_convert_rgb888_to_rgb32_avx
+#define convert_RGB888_to_RGB32_ssse3 convert_RGB888_to_RGB32_avx
+#include "qimage_ssse3.cpp"
+
+#endif
diff --git a/src/gui/painting/painting.pri b/src/gui/painting/painting.pri
index 7a20f35b33..e4b3f5e568 100644
--- a/src/gui/painting/painting.pri
+++ b/src/gui/painting/painting.pri
@@ -91,12 +91,13 @@ SOURCES += \
 
 
 
-if(mmx|3dnow|sse|sse2|iwmmxt) {
+if(mmx|3dnow|sse|sse2|iwmmxt|avx) {
     HEADERS += painting/qdrawhelper_x86_p.h \
                painting/qdrawingprimitive_sse2_p.h
     SSE2_SOURCES += painting/qdrawhelper_sse2.cpp
     SSSE3_SOURCES += painting/qdrawhelper_ssse3.cpp
     IWMMXT_SOURCES += painting/qdrawhelper_iwmmxt.cpp
+    AVX_SOURCES += painting/qdrawhelper_avx.cpp
 }
 
 NEON_SOURCES += painting/qdrawhelper_neon.cpp
diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index c21cc61ec5..23a1853314 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -5805,6 +5805,24 @@ void qInitDrawhelperAsm()
 
     const uint features = qDetectCPUFeatures();
     if (false) {
+#ifdef QT_HAVE_AVX
+    } else if (features & AVX) {
+        qt_memfill32 = qt_memfill32_avx;
+        qt_memfill16 = qt_memfill16_avx;
+        qDrawHelper[QImage::Format_RGB32].bitmapBlit = qt_bitmapblit32_avx;
+        qDrawHelper[QImage::Format_ARGB32].bitmapBlit = qt_bitmapblit32_avx;
+        qDrawHelper[QImage::Format_ARGB32_Premultiplied].bitmapBlit = qt_bitmapblit32_avx;
+        qDrawHelper[QImage::Format_RGB16].bitmapBlit = qt_bitmapblit16_avx;
+
+        extern void qt_scale_image_argb32_on_argb32_avx(uchar *destPixels, int dbpl,
+                                                         const uchar *srcPixels, int sbpl,
+                                                         const QRectF &targetRect,
+                                                         const QRectF &sourceRect,
+                                                         const QRect &clip,
+                                                         int const_alpha);
+        qScaleFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_scale_image_argb32_on_argb32_avx;
+        qScaleFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_scale_image_argb32_on_argb32_avx;
+#endif
 #ifdef QT_HAVE_SSE2
     } else if (features & SSE2) {
         qt_memfill32 = qt_memfill32_sse2;
@@ -5859,12 +5877,51 @@ void qInitDrawhelperAsm()
     }
 #endif // SSSE3
 
+#ifdef QT_HAVE_AVX
+    if (features & AVX) {
+        extern void qt_blend_rgb32_on_rgb32_avx(uchar *destPixels, int dbpl,
+                                                const uchar *srcPixels, int sbpl,
+                                                int w, int h,
+                                                int const_alpha);
+        extern void qt_blend_argb32_on_argb32_avx(uchar *destPixels, int dbpl,
+                                                  const uchar *srcPixels, int sbpl,
+                                                  int w, int h,
+                                                  int const_alpha);
+
+        qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_avx;
+        qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_avx;
+        qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_avx;
+        qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_avx;
+
+        extern const uint * QT_FASTCALL qt_fetch_radial_gradient_avx(uint *buffer, const Operator *op, const QSpanData *data,
+                                                                     int y, int x, int length);
+
+        qt_fetch_radial_gradient = qt_fetch_radial_gradient_avx;
+    }
+#endif // AVX
+
 #endif // SSE2
 
 #ifdef QT_HAVE_SSE2
     if (features & SSE2) {
         functionForModeAsm = qt_functionForMode_SSE2;
         functionForModeSolidAsm = qt_functionForModeSolid_SSE2;
+        }
+#endif
+#ifdef QT_HAVE_AVX
+        if (features & AVX) {
+            extern void QT_FASTCALL comp_func_SourceOver_avx(uint *destPixels,
+                                                             const uint *srcPixels,
+                                                             int length,
+                                                             uint const_alpha);
+            extern void QT_FASTCALL comp_func_solid_SourceOver_avx(uint *destPixels, int length, uint color, uint const_alpha);
+            extern void QT_FASTCALL comp_func_Plus_avx(uint *dst, const uint *src, int length, uint const_alpha);
+            extern void QT_FASTCALL comp_func_Source_avx(uint *dst, const uint *src, int length, uint const_alpha);
+
+            functionForModeAsm[0] = comp_func_SourceOver_avx;
+            functionForModeAsm[QPainter::CompositionMode_Source] = comp_func_Source_avx;
+            functionForModeAsm[QPainter::CompositionMode_Plus] = comp_func_Plus_avx;
+            functionForModeSolidAsm[0] = comp_func_solid_SourceOver_avx;
     }
 #endif // SSE2
 
diff --git a/src/gui/painting/qdrawhelper_avx.cpp b/src/gui/painting/qdrawhelper_avx.cpp
new file mode 100644
index 0000000000..28ba5f8aa3
--- /dev/null
+++ b/src/gui/painting/qdrawhelper_avx.cpp
@@ -0,0 +1,70 @@
+/****************************************************************************
+**
+** Copyright (C) 2012 Intel Corporation
+** Contact: http://www.qt-project.org/
+**
+** This file is part of the QtGui module of the Qt Toolkit.
+**
+** $QT_BEGIN_LICENSE:LGPL$
+** GNU Lesser General Public License Usage
+** 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.
+**
+** 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.
+**
+** Other Usage
+** Alternatively, this file may be used in accordance with the terms and
+** conditions contained in a signed written agreement between you and Nokia.
+**
+**
+**
+**
+**
+**
+** $QT_END_LICENSE$
+**
+****************************************************************************/
+
+#include <private/qsimd_p.h>
+
+#ifdef QT_HAVE_AVX
+#define QDRAWHELPER_AVX
+
+#ifndef __AVX__
+#error "AVX not enabled in this file, cannot proceed"
+#endif
+
+#define qt_blend_argb32_on_argb32_ssse3 qt_blend_argb32_on_argb32_avx
+#include "qdrawhelper_ssse3.cpp"
+
+//#define qt_blend_argb32_on_argb32_sse2 qt_blend_argb32_on_argb32_avx
+#define qt_blend_rgb32_on_rgb32_sse2 qt_blend_rgb32_on_rgb32_avx
+#define comp_func_SourceOver_sse2 comp_func_SourceOver_avx
+#define comp_func_Plus_sse2 comp_func_Plus_avx
+#define comp_func_Source_sse2 comp_func_Source_avx
+#define comp_func_solid_SourceOver_sse2 comp_func_solid_SourceOver_avx
+#define qt_memfill32_sse2 qt_memfill32_avx
+#define qt_memfill16_sse2 qt_memfill16_avx
+#define qt_bitmapblit32_sse2 qt_bitmapblit32_avx
+#define qt_bitmapblit16_sse2 qt_bitmapblit16_avx
+#define QSimdSse2 QSimdAvx
+#define qt_fetch_radial_gradient_sse2 qt_fetch_radial_gradient_avx
+#define qt_scale_image_argb32_on_argb32_sse2 qt_scale_image_argb32_on_argb32_avx
+
+#include "qdrawhelper_sse2.cpp"
+
+#endif
diff --git a/src/gui/painting/qdrawhelper_sse2.cpp b/src/gui/painting/qdrawhelper_sse2.cpp
index 2c87aabe93..e44116570d 100644
--- a/src/gui/painting/qdrawhelper_sse2.cpp
+++ b/src/gui/painting/qdrawhelper_sse2.cpp
@@ -48,6 +48,8 @@
 
 QT_BEGIN_NAMESPACE
 
+#ifndef QDRAWHELPER_AVX
+// in AVX mode, we'll use the SSSE3 code
 void qt_blend_argb32_on_argb32_sse2(uchar *destPixels, int dbpl,
                                     const uchar *srcPixels, int sbpl,
                                     int w, int h,
@@ -83,6 +85,7 @@ void qt_blend_argb32_on_argb32_sse2(uchar *destPixels, int dbpl,
         }
     }
 }
+#endif
 
 // qblendfunctions.cpp
 void qt_blend_rgb32_on_rgb32(uchar *destPixels, int dbpl,
@@ -310,6 +313,7 @@ void QT_FASTCALL comp_func_solid_SourceOver_sse2(uint *destPixels, int length, u
     }
 }
 
+#ifndef QDRAWHELPER_AVX
 CompositionFunctionSolid qt_functionForModeSolid_SSE2[numCompositionFunctions] = {
     comp_func_solid_SourceOver_sse2,
     comp_func_solid_DestinationOver,
@@ -381,6 +385,7 @@ CompositionFunction qt_functionForMode_SSE2[numCompositionFunctions] = {
     rasterop_NotSourceAndDestination,
     rasterop_SourceAndNotDestination
 };
+#endif
 
 void qt_memfill16_sse2(quint16 *dest, quint16 value, int count)
 {
diff --git a/src/gui/painting/qdrawhelper_x86_p.h b/src/gui/painting/qdrawhelper_x86_p.h
index 93abaf4fff..eb434e5fda 100644
--- a/src/gui/painting/qdrawhelper_x86_p.h
+++ b/src/gui/painting/qdrawhelper_x86_p.h
@@ -79,6 +79,25 @@ extern CompositionFunction qt_functionForMode_SSE2[];
 extern CompositionFunctionSolid qt_functionForModeSolid_SSE2[];
 #endif // QT_HAVE_SSE2
 
+#ifdef QT_HAVE_AVX
+void qt_memfill32_avx(quint32 *dest, quint32 value, int count);
+void qt_memfill16_avx(quint16 *dest, quint16 value, int count);
+void qt_bitmapblit32_avx(QRasterBuffer *rasterBuffer, int x, int y,
+                         quint32 color,
+                         const uchar *src, int width, int height, int stride);
+void qt_bitmapblit16_avx(QRasterBuffer *rasterBuffer, int x, int y,
+                         quint32 color,
+                         const uchar *src, int width, int height, int stride);
+void qt_blend_argb32_on_argb32_avx(uchar *destPixels, int dbpl,
+                                   const uchar *srcPixels, int sbpl,
+                                   int w, int h,
+                                   int const_alpha);
+void qt_blend_rgb32_on_rgb32_avx(uchar *destPixels, int dbpl,
+                                 const uchar *srcPixels, int sbpl,
+                                 int w, int h,
+                                 int const_alpha);
+#endif // QT_HAVE_AVX
+
 #ifdef QT_HAVE_IWMMXT
 void qt_blend_color_argb_iwmmxt(int count, const QSpan *spans, void *userData);