diff options
| author | Samuel Rødal <samuel.rodal@digia.com> | 2012-12-14 10:52:47 +0100 |
|---|---|---|
| committer | The Qt Project <gerrit-noreply@qt-project.org> | 2012-12-14 17:12:44 +0100 |
| commit | 52619ae7787b3c4febb73a02afa623b12edabc97 (patch) | |
| tree | 6a2d25bfbe59a1ca73c37bddf82ee11fbe8037c9 /src/gui/painting/qdrawhelper_ssse3.cpp | |
| parent | 7ca226c9512e8f1d11e428b24e9f91c81ad4cbe7 (diff) | |
Fixed invalid memory read in SSSE3 image blending code.
We need to do bounds comparison on the actual offset we're going to use
with _mm_load_si128 to read 16 bytes from memory (even though we won't
use the trailing bytes in the end).
Task-number: QTBUG-28324
Change-Id: Id0d6094da796ca67338d8ad225fa6b2f309bbe6e
Reviewed-by: Olivier Goffart <ogoffart@woboq.com>
Diffstat (limited to 'src/gui/painting/qdrawhelper_ssse3.cpp')
| -rw-r--r-- | src/gui/painting/qdrawhelper_ssse3.cpp | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/gui/painting/qdrawhelper_ssse3.cpp b/src/gui/painting/qdrawhelper_ssse3.cpp index 09e0516dcb..9f80c70fcc 100644 --- a/src/gui/painting/qdrawhelper_ssse3.cpp +++ b/src/gui/painting/qdrawhelper_ssse3.cpp @@ -60,7 +60,7 @@ inline static void blend_pixel(quint32 &dst, const quint32 src) shift (4, 8, 12). Checking the alignment inside the loop is unfortunatelly way too slow. */ #define BLENDING_LOOP(palignrOffset, length)\ - for (; x < length-3; x += 4) { \ + for (; x-minusOffsetToAlignSrcOn16Bytes < length-7; x += 4) { \ const __m128i srcVectorLastLoaded = _mm_load_si128((__m128i *)&src[x - minusOffsetToAlignSrcOn16Bytes + 4]);\ const __m128i srcVector = _mm_alignr_epi8(srcVectorLastLoaded, srcVectorPrevLoaded, palignrOffset); \ const __m128i srcVectorAlpha = _mm_and_si128(srcVector, alphaMask); \ |