summaryrefslogtreecommitdiffstats
path: root/src/gui
diff options
context:
space:
mode:
authorQt Continuous Integration System <qt-info@nokia.com>2010-07-18 23:33:45 (GMT)
committerQt Continuous Integration System <qt-info@nokia.com>2010-07-18 23:33:45 (GMT)
commite7b78d2225c5338545852330c9160084f12ffe1f (patch)
tree663fa0923182e25b5ba613a831eb7c6a517ebaf7 /src/gui
parentc79ad5443835a5be9057de5e1553786dde46f65d (diff)
parent23ea4340a622cbfed81eb7afb2e09ec64b0ebef8 (diff)
downloadQt-e7b78d2225c5338545852330c9160084f12ffe1f.zip
Qt-e7b78d2225c5338545852330c9160084f12ffe1f.tar.gz
Qt-e7b78d2225c5338545852330c9160084f12ffe1f.tar.bz2
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: Corrected BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2 Used aligned load and store when possible for the blending of ARGB32
Diffstat (limited to 'src/gui')
-rw-r--r--src/gui/painting/qdrawingprimitive_sse2_p.h33
1 files changed, 28 insertions, 5 deletions
diff --git a/src/gui/painting/qdrawingprimitive_sse2_p.h b/src/gui/painting/qdrawingprimitive_sse2_p.h
index 3c96946..18355c2 100644
--- a/src/gui/painting/qdrawingprimitive_sse2_p.h
+++ b/src/gui/painting/qdrawingprimitive_sse2_p.h
@@ -141,12 +141,24 @@ QT_BEGIN_NAMESPACE
// 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; \
+\
+ /* First, get dst aligned. */ \
+ const int offsetToAlignOn16Bytes = (4 - ((reinterpret_cast<quintptr>(dst) >> 2) & 0x3)) & 0x3;\
+ const int prologLength = qMin(length, offsetToAlignOn16Bytes);\
+ for (; x < prologLength; ++x) { \
+ uint s = src[x]; \
+ if (s >= 0xff000000) \
+ dst[x] = s; \
+ else if (s != 0) \
+ dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
+ } \
+\
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); \
+ _mm_store_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 */ \
@@ -157,13 +169,13 @@ QT_BEGIN_NAMESPACE
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]); \
+ const __m128i dstVector = _mm_load_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); \
+ _mm_store_si128((__m128i *)&dst[x], result); \
} \
} \
for (; x < length; ++x) { \
@@ -189,6 +201,17 @@ QT_BEGIN_NAMESPACE
#define BLEND_SOURCE_OVER_ARGB32_WITH_CONST_ALPHA_SSE2(dst, src, length, nullVector, half, one, colorMask, constAlphaVector) \
{ \
int x = 0; \
+\
+ const int offsetToAlignOn16Bytes = (4 - ((reinterpret_cast<quintptr>(dst) >> 2) & 0x3)) & 0x3;\
+ const int prologLength = qMin(length, offsetToAlignOn16Bytes);\
+ for (; x < prologLength; ++x) { \
+ quint32 s = src[x]; \
+ if (s != 0) { \
+ s = BYTE_MUL(s, const_alpha); \
+ dst[x] = s + BYTE_MUL(dst[x], qAlpha(~s)); \
+ } \
+ } \
+\
for (; x < length-3; x += 4) { \
__m128i srcVector = _mm_loadu_si128((__m128i *)&src[x]); \
if (_mm_movemask_epi8(_mm_cmpeq_epi32(srcVector, nullVector)) != 0xffff) { \
@@ -198,12 +221,12 @@ QT_BEGIN_NAMESPACE
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]); \
+ const __m128i dstVector = _mm_load_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); \
+ _mm_store_si128((__m128i *)&dst[x], result); \
} \
} \
for (; x < length; ++x) { \