Add AVX2 versions of qt_memfill32 and qt_memfill64

The implementation is almost the same 4-way-unrolled loop, but because
of the wider registers, we fill 128 bytes per loop. Unlike the SSE2
implementation, the AVX2 version uses unaligned stores and won't try to
align in the prologue, matching glibc's __memset_avx2 (also unaligned).

Change-Id: Iba4b5c183776497d8ee1fffd15637ccb2a7b83bc
Reviewed-by: Allan Sandfeld Jensen <allan.jensen@qt.io>

1 files changed, 5 insertions, 2 deletions

diff --git a/src/gui/painting/qdrawhelper_x86_p.h b/src/gui/painting/qdrawhelper_x86_p.h
index 6f97242199..5749d8c9fb 100644
--- a/src/gui/painting/qdrawhelper_x86_p.h
+++ b/src/gui/painting/qdrawhelper_x86_p.h
@@ -57,8 +57,8 @@
 QT_BEGIN_NAMESPACE
 
 #ifdef __SSE2__
-void qt_memfill64(quint64 *dest, quint64 value, qsizetype count);
-void qt_memfill32(quint32 *dest, quint32 value, qsizetype count);
+void qt_memfill64_sse2(quint64 *dest, quint64 value, qsizetype count);
+void qt_memfill32_sse2(quint32 *dest, quint32 value, qsizetype count);
 void qt_bitmapblit32_sse2(QRasterBuffer *rasterBuffer, int x, int y,
                           const QRgba64 &color,
                           const uchar *src, int width, int height, int stride);
@@ -79,6 +79,9 @@ void qt_blend_rgb32_on_rgb32_sse2(uchar *destPixels, int dbpl,
 
 extern CompositionFunction qt_functionForMode_SSE2[];
 extern CompositionFunctionSolid qt_functionForModeSolid_SSE2[];
+
+void qt_memfill64_avx2(quint64 *dest, quint64 value, qsizetype count);
+void qt_memfill32_avx2(quint32 *dest, quint32 value, qsizetype count);
 #endif // __SSE2__
 
 static const int numCompositionFunctions = 38;