Merge branch '4.7' of scm.dev.nokia.troll.no:qt/oslo-staging-2 into 4.7-integration

* '4.7' of scm.dev.nokia.troll.no:qt/oslo-staging-2:
  Normalize integers when calling glVertexAttribPointer()
  Add an implementation of comp_func_solid_SourceOver_neon() with Neon.
  Fix the casts of qdrawhelper_sse2
  Add a SSE2 implementation of comp_func_solid_SourceOver()
  Add a SSE2 version of comp_func_SourceOver()
  Fixed missing copy of raster pixmap data after change fb76a872e20bd.
  Fixed QPixmap::toImage() bug introduced in fb76a872e20bd.
  Optimized sub-rect copying / painting of QPixmaps.
  Fixed crash in the fast blend functions for raster

5 files changed, 205 insertions, 67 deletions

diff --git a/src/gui/painting/qdrawhelper.cpp b/src/gui/painting/qdrawhelper.cpp
index bfa1136..5727b3c 100644
--- a/src/gui/painting/qdrawhelper.cpp
+++ b/src/gui/painting/qdrawhelper.cpp
@@ -7817,6 +7817,15 @@ void qInitDrawhelperAsm()
 
 #ifdef QT_HAVE_SSE2
         if (features & SSE2) {
+            extern void QT_FASTCALL comp_func_SourceOver_sse2(uint *destPixels,
+                                                  const uint *srcPixels,
+                                                  int length,
+                                                  uint const_alpha);
+            extern void QT_FASTCALL comp_func_solid_SourceOver_sse2(uint *destPixels, int length, uint color, uint const_alpha);
+
+            functionForModeAsm[0] = comp_func_SourceOver_sse2;
+            functionForModeSolidAsm[0] = comp_func_solid_SourceOver_sse2;
+
             extern void qt_blend_rgb32_on_rgb32_sse2(uchar *destPixels, int dbpl,
                                                      const uchar *srcPixels, int sbpl,
                                                      int w, int h,
@@ -7826,7 +7835,6 @@ void qInitDrawhelperAsm()
                                                        int w, int h,
                                                        int const_alpha);
 
-
             qBlendFunctions[QImage::Format_RGB32][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_sse2;
             qBlendFunctions[QImage::Format_ARGB32_Premultiplied][QImage::Format_RGB32] = qt_blend_rgb32_on_rgb32_sse2;
             qBlendFunctions[QImage::Format_RGB32][QImage::Format_ARGB32_Premultiplied] = qt_blend_argb32_on_argb32_sse2;
@@ -7890,6 +7898,7 @@ void qInitDrawhelperAsm()
             qDrawHelper[QImage::Format_RGB16].alphamapBlit = qt_alphamapblit_quint16_neon;
 
             functionForMode_C[QPainter::CompositionMode_SourceOver] = qt_blend_argb32_on_argb32_scanline_neon;
+            functionForModeSolid_C[QPainter::CompositionMode_SourceOver] = comp_func_solid_SourceOver_neon;
             destFetchProc[QImage::Format_RGB16] = qt_destFetchRGB16_neon;
             destStoreProc[QImage::Format_RGB16] = qt_destStoreRGB16_neon;
         }
diff --git a/src/gui/painting/qdrawhelper_neon.cpp b/src/gui/painting/qdrawhelper_neon.cpp
index ee5f24a..3ce90d2 100644
--- a/src/gui/painting/qdrawhelper_neon.cpp
+++ b/src/gui/painting/qdrawhelper_neon.cpp
@@ -579,6 +579,49 @@ void QT_FASTCALL qt_destStoreRGB16_neon(QRasterBuffer *rasterBuffer, int x, int
     }
 }
 
+void QT_FASTCALL comp_func_solid_SourceOver_neon(uint *destPixels, int length, uint color, uint const_alpha)
+{
+    if ((const_alpha & qAlpha(color)) == 255) {
+        QT_MEMFILL_UINT(destPixels, length, color);
+    } else {
+        if (const_alpha != 255)
+            color = BYTE_MUL(color, const_alpha);
+
+        const quint32 minusAlphaOfColor = qAlpha(~color);
+        int x = 0;
+
+        uint32_t *dst = (uint32_t *) destPixels;
+        const uint32x4_t colorVector = vdupq_n_u32(color);
+        uint16x8_t half = vdupq_n_u16(0x80);
+        const uint16x8_t minusAlphaOfColorVector = vdupq_n_u16(minusAlphaOfColor);
+
+        for (; x < length-3; x += 4) {
+            uint32x4_t dstVector = vld1q_u32(&dst[x]);
+
+            const uint8x16_t dst8 = vreinterpretq_u8_u32(dstVector);
+
+            const uint8x8_t dst8_low = vget_low_u8(dst8);
+            const uint8x8_t dst8_high = vget_high_u8(dst8);
+
+            const uint16x8_t dst16_low = vmovl_u8(dst8_low);
+            const uint16x8_t dst16_high = vmovl_u8(dst8_high);
+
+            const uint16x8_t result16_low = qvbyte_mul_u16(dst16_low, minusAlphaOfColorVector, half);
+            const uint16x8_t result16_high = qvbyte_mul_u16(dst16_high, minusAlphaOfColorVector, half);
+
+            const uint32x2_t result32_low = vreinterpret_u32_u8(vmovn_u16(result16_low));
+            const uint32x2_t result32_high = vreinterpret_u32_u8(vmovn_u16(result16_high));
+
+            uint32x4_t blendedPixels = vcombine_u32(result32_low, result32_high);
+            uint32x4_t colorPlusBlendedPixels = vaddq_u32(colorVector, blendedPixels);
+            vst1q_u32(&dst[x], colorPlusBlendedPixels);
+        }
+
+        for (;x < length; ++x)
+            destPixels[x] = color + BYTE_MUL(destPixels[x], minusAlphaOfColor);
+    }
+}
+
 QT_END_NAMESPACE
 
 #endif // QT_HAVE_NEON
diff --git a/src/gui/painting/qdrawhelper_neon_p.h b/src/gui/painting/qdrawhelper_neon_p.h
index d6a4509..c054a1e 100644
--- a/src/gui/painting/qdrawhelper_neon_p.h
+++ b/src/gui/painting/qdrawhelper_neon_p.h
@@ -127,6 +127,8 @@ uint * QT_FASTCALL qt_destFetchRGB16_neon(uint *buffer,
 void QT_FASTCALL qt_destStoreRGB16_neon(QRasterBuffer *rasterBuffer,
                                         int x, int y, const uint *buffer, int length);
 
+void QT_FASTCALL comp_func_solid_SourceOver_neon(uint *destPixels, int length, uint color, uint const_alpha);
+
 #endif // QT_HAVE_NEON
 
 QT_END_NAMESPACE
diff --git a/src/gui/painting/qdrawhelper_sse2.cpp b/src/gui/painting/qdrawhelper_sse2.cpp
index 6ac64d3..e2a69ec 100644
--- a/src/gui/painting/qdrawhelper_sse2.cpp
+++ b/src/gui/painting/qdrawhelper_sse2.cpp
@@ -126,13 +126,100 @@ QT_BEGIN_NAMESPACE
     result = _mm_or_si128(finalAG, finalRB); \
 }
 
+// Basically blend src over dst with the const alpha defined as constAlphaVector.
+// nullVector, half, one, colorMask are constant accross the whole image/texture, and should be defined as:
+//const __m128i nullVector = _mm_set1_epi32(0);
+//const __m128i half = _mm_set1_epi16(0x80);
+//const __m128i one = _mm_set1_epi16(0xff);
+//const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
+//const __m128i alphaMask = _mm_set1_epi32(0xff000000);
+//
+// The computation being done is:
+// result = s + d * (1-alpha)
+// with shortcuts if fully opaque or fully transparent.
+#define BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, length, nullVector, half, one, colorMask, alphaMask) { \
+    int x = 0; \
+    for (; x < length-3; x += 4) { \
+        const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]); \
+        const __m128i srcVectorAlpha = _mm_and_si128(srcVector, alphaMask); \
+        if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, alphaMask)) == 0xffff) { \
+            /* all opaque */ \
+            _mm_storeu_si128((__m128i *)&dst[x], srcVector); \
+        } else if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, nullVector)) != 0xffff) { \
+            /* not fully transparent */ \
+            /* extract the alpha channel on 2 x 16 bits */ \
+            /* so we have room for the multiplication */ \
+            /* each 32 bits will be in the form 0x00AA00AA */ \
+            /* with A being the 1 - alpha */ \
+            __m128i alphaChannel = _mm_srli_epi32(srcVector, 24); \
+            alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16)); \
+            alphaChannel = _mm_sub_epi16(one, alphaChannel); \
+ \
+            const __m128i dstVector = _mm_loadu_si128((__m128i *)&dst[x]); \
+            __m128i destMultipliedByOneMinusAlpha; \
+            BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half); \
+ \
+            /* result = s + d * (1-alpha) */\
+            const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha); \
+            _mm_storeu_si128((__m128i *)&dst[x], result); \
+        } \
+    } \
+    for (; x < length; ++x) { \
+        uint s = src[x]; \
+        if (s >= 0xff000000) \
+            dst[x] = s; \
+        else if (s != 0) \
+            dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
+    } \
+}
+
+// Basically blend src over dst with the const alpha defined as constAlphaVector.
+// nullVector, half, one, colorMask are constant accross the whole image/texture, and should be defined as:
+//const __m128i nullVector = _mm_set1_epi32(0);
+//const __m128i half = _mm_set1_epi16(0x80);
+//const __m128i one = _mm_set1_epi16(0xff);
+//const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
+//
+// The computation being done is:
+// dest = (s + d * sia) * ca + d * cia
+//      = s * ca + d * (sia * ca + cia)
+//      = s * ca + d * (1 - sa*ca)
+#define BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, length, nullVector, half, one, colorMask, constAlphaVector) \
+{ \
+    int x = 0; \
+    for (; x < length-3; x += 4) { \
+        __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]); \
+        if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVector, nullVector)) != 0xffff) { \
+            BYTE_MUL_SSE2(srcVector, srcVector, constAlphaVector, colorMask, half); \
+\
+            __m128i alphaChannel = _mm_srli_epi32(srcVector, 24); \
+            alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16)); \
+            alphaChannel = _mm_sub_epi16(one, alphaChannel); \
+ \
+            const __m128i dstVector = _mm_loadu_si128((__m128i *)&dst[x]); \
+            __m128i destMultipliedByOneMinusAlpha; \
+            BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half); \
+ \
+            const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha); \
+            _mm_storeu_si128((__m128i *)&dst[x], result); \
+        } \
+    } \
+    for (; x < length; ++x) { \
+        quint32 s = src[x]; \
+        if (s != 0) { \
+            s = BYTE_MUL(s, const_alpha); \
+            dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
+        } \
+    } \
+}
+
 void qt_blend_argb32_on_argb32_sse2(uchar *destPixels, int dbpl,
                                     const uchar *srcPixels, int sbpl,
                                     int w, int h,
                                     int const_alpha)
 {
     const quint32 *src = (const quint32 *) srcPixels;
-    quint32 *dst = (uint *) destPixels;
+    quint32 *dst = (quint32 *) destPixels;
     if (const_alpha == 256) {
         const __m128i alphaMask = _mm_set1_epi32(0xff000000);
         const __m128i nullVector = _mm_set1_epi32(0);
@@ -140,41 +227,7 @@ void qt_blend_argb32_on_argb32_sse2(uchar *destPixels, int dbpl,
         const __m128i one = _mm_set1_epi16(0xff);
         const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
         for (int y = 0; y < h; ++y) {
-            int x = 0;
-            for (; x < w-3; x += 4) {
-                const __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]);
-                const __m128i srcVectorAlpha = _mm_and_si128(srcVector, alphaMask);
-                if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, alphaMask)) == 0xffff) {
-                    // all opaque
-                    _mm_storeu_si128((__m128i *)&dst[x], srcVector);
-                } else if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVectorAlpha, nullVector)) != 0xffff) {
-                    // not fully transparent
-                    // result = s + d * (1-alpha)
-
-                    // extract the alpha channel on 2 x 16 bits
-                    // so we have room for the multiplication
-                    // each 32 bits will be in the form 0x00AA00AA
-                    // with A being the 1 - alpha
-                    __m128i alphaChannel = _mm_srli_epi32(srcVector, 24);
-                    alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16));
-                    alphaChannel = _mm_sub_epi16(one, alphaChannel);
-
-                    const __m128i dstVector = _mm_loadu_si128((__m128i *)&dst[x]);
-                    __m128i destMultipliedByOneMinusAlpha;
-                    BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half);
-
-                    // result = s + d * (1-alpha)
-                    const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha);
-                    _mm_storeu_si128((__m128i *)&dst[x], result);
-                }
-            }
-            for (; x<w; ++x) {
-                uint s = src[x];
-                if (s >= 0xff000000)
-                    dst[x] = s;
-                else if (s != 0)
-                    dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
-            }
+            BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, w, nullVector, half, one, colorMask, alphaMask);
             dst = (quint32 *)(((uchar *) dst) + dbpl);
             src = (const quint32 *)(((const uchar *) src) + sbpl);
         }
@@ -189,31 +242,7 @@ void qt_blend_argb32_on_argb32_sse2(uchar *destPixels, int dbpl,
         const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
         const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
         for (int y = 0; y < h; ++y) {
-            int x = 0;
-            for (; x < w-3; x += 4) {
-                __m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]);
-                if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVector, nullVector)) != 0xffff) {
-                    BYTE_MUL_SSE2(srcVector, srcVector, constAlphaVector, colorMask, half);
-
-                    __m128i alphaChannel = _mm_srli_epi32(srcVector, 24);
-                    alphaChannel = _mm_or_si128(alphaChannel, _mm_slli_epi32(alphaChannel, 16));
-                    alphaChannel = _mm_sub_epi16(one, alphaChannel);
-
-                    const __m128i dstVector = _mm_loadu_si128((__m128i *)&dst[x]);
-                    __m128i destMultipliedByOneMinusAlpha;
-                    BYTE_MUL_SSE2(destMultipliedByOneMinusAlpha, dstVector, alphaChannel, colorMask, half);
-
-                    const __m128i result = _mm_add_epi8(srcVector, destMultipliedByOneMinusAlpha);
-                    _mm_storeu_si128((__m128i *)&dst[x], result);
-                }
-            }
-            for (; x<w; ++x) {
-                quint32 s = src[x];
-                if (s != 0) {
-                    s = BYTE_MUL(s, const_alpha);
-                    dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s));
-                }
-            }
+            BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, w, nullVector, half, one, colorMask, constAlphaVector)
             dst = (quint32 *)(((uchar *) dst) + dbpl);
             src = (const quint32 *)(((const uchar *) src) + sbpl);
         }
@@ -232,7 +261,7 @@ void qt_blend_rgb32_on_rgb32_sse2(uchar *destPixels, int dbpl,
                                  int const_alpha)
 {
     const quint32 *src = (const quint32 *) srcPixels;
-    quint32 *dst = (uint *) destPixels;
+    quint32 *dst = (quint32 *) destPixels;
     if (const_alpha != 256) {
         if (const_alpha != 0) {
             const __m128i nullVector = _mm_set1_epi32(0);
@@ -268,6 +297,27 @@ void qt_blend_rgb32_on_rgb32_sse2(uchar *destPixels, int dbpl,
     }
 }
 
+void QT_FASTCALL comp_func_SourceOver_sse2(uint *destPixels, const uint *srcPixels, int length, uint const_alpha)
+{
+    Q_ASSERT(const_alpha > 0); // if const_alpha == 0, this should never be called
+    Q_ASSERT(const_alpha < 256);
+
+    const quint32 *src = (const quint32 *) srcPixels;
+    quint32 *dst = (quint32 *) destPixels;
+
+    const __m128i nullVector = _mm_set1_epi32(0);
+    const __m128i half = _mm_set1_epi16(0x80);
+    const __m128i one = _mm_set1_epi16(0xff);
+    const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
+    if (const_alpha == 255) {
+        const __m128i alphaMask = _mm_set1_epi32(0xff000000);
+        BLEND_SOURCE_OVER_ARGB32_SSE2(dst, src, length, nullVector, half, one, colorMask, alphaMask);
+    } else {
+        const __m128i constAlphaVector = _mm_set1_epi16(const_alpha);
+        BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, length, nullVector, half, one, colorMask, constAlphaVector);
+    }
+}
+
 void qt_memfill32_sse2(quint32 *dest, quint32 value, int count)
 {
     if (count < 7) {
@@ -312,6 +362,34 @@ void qt_memfill32_sse2(quint32 *dest, quint32 value, int count)
     }
 }
 
+void QT_FASTCALL comp_func_solid_SourceOver_sse2(uint *destPixels, int length, uint color, uint const_alpha)
+{
+    if ((const_alpha & qAlpha(color)) == 255) {
+        qt_memfill32_sse2(destPixels, length, color);
+    } else {
+        if (const_alpha != 255)
+            color = BYTE_MUL(color, const_alpha);
+
+        const quint32 minusAlphaOfColor = qAlpha(~color);
+        int x = 0;
+
+        quint32 *dst = (quint32 *) destPixels;
+        const __m128i colorVector = _mm_set1_epi32(color);
+        const __m128i colorMask = _mm_set1_epi32(0x00ff00ff);
+        const __m128i half = _mm_set1_epi16(0x80);
+        const __m128i minusAlphaOfColorVector = _mm_set1_epi16(minusAlphaOfColor);
+
+        for (; x < length-3; x += 4) {
+            __m128i dstVector = _mm_loadu_si128((__m128i *)&dst[x]);
+            BYTE_MUL_SSE2(dstVector, dstVector, minusAlphaOfColorVector, colorMask, half);
+            dstVector = _mm_add_epi8(colorVector, dstVector);
+            _mm_storeu_si128((__m128i *)&dst[x], dstVector);
+        }
+        for (;x < length; ++x)
+            destPixels[x] = color + BYTE_MUL(destPixels[x], minusAlphaOfColor);
+    }
+}
+
 void qt_memfill16_sse2(quint16 *dest, quint16 value, int count)
 {
     if (count < 3) {
diff --git a/src/gui/painting/qpaintengine_raster.cpp b/src/gui/painting/qpaintengine_raster.cpp
index a212718..84b36c7 100644
--- a/src/gui/painting/qpaintengine_raster.cpp
+++ b/src/gui/painting/qpaintengine_raster.cpp
@@ -2419,7 +2419,9 @@ void QRasterPaintEngine::drawPixmap(const QRectF &r, const QPixmap &pixmap, cons
             drawImage(r, image, sr);
         }
     } else {
-        const QImage image = pixmap.toImage();
+        QRect clippedSource = sr.toAlignedRect().intersected(pixmap.rect());
+        const QImage image = pd->toImage(clippedSource);
+        QRectF translatedSource = sr.translated(-clippedSource.topLeft());
         if (image.depth() == 1) {
             Q_D(QRasterPaintEngine);
             QRasterPaintEngineState *s = state();
@@ -2430,10 +2432,10 @@ void QRasterPaintEngine::drawPixmap(const QRectF &r, const QPixmap &pixmap, cons
                 drawBitmap(r.topLeft() + QPointF(s->matrix.dx(), s->matrix.dy()), image, &s->penData);
                 return;
             } else {
-                drawImage(r, d->rasterBuffer->colorizeBitmap(image, s->pen.color()), sr);
+                drawImage(r, d->rasterBuffer->colorizeBitmap(image, s->pen.color()), translatedSource);
             }
         } else {
-            drawImage(r, image, sr);
+            drawImage(r, image, translatedSource);
         }
     }
 }
@@ -2688,7 +2690,11 @@ void QRasterPaintEngine::drawImage(const QRectF &r, const QImage &img, const QRe
 
     if (s->matrix.type() > QTransform::TxTranslate || stretch_sr) {
 
-        if (s->flags.fast_images) {
+        QRectF targetBounds = s->matrix.mapRect(r);
+        bool exceedsPrecision = targetBounds.width() > 0xffff
+                                || targetBounds.height() > 0xffff;
+
+        if (s->flags.fast_images && !exceedsPrecision) {
             if (s->matrix.type() > QTransform::TxScale) {
                 SrcOverTransformFunc func = qTransformFunctions[d->rasterBuffer->format][img.format()];
                 if (func && (!clip || clip->hasRectClip)) {