[sfnt] Fix `premultiply_data' (#52092).
* src/sfnt/pngshim.c (premultiply_data): Don't use vector extension if we have less than 16 bytes of data.
This commit is contained in:
parent
e720206925
commit
0d1262a41e
@ -1,3 +1,10 @@
|
||||
2017-09-24 Jonathan Kew <jfkthame@gmail.com>
|
||||
|
||||
[sfnt] Fix `premultiply_data' (#52092).
|
||||
|
||||
* src/sfnt/pngshim.c (premultiply_data): Don't use vector extension
|
||||
if we have less than 16 bytes of data.
|
||||
|
||||
2017-09-24 Werner Lemberg <wl@gnu.org>
|
||||
|
||||
[otvalid] Fix handling of ValueRecords.
|
||||
|
@ -82,42 +82,45 @@
|
||||
typedef unsigned short v82 __attribute__(( vector_size( 16 ) ));
|
||||
|
||||
|
||||
/* process blocks of 16 bytes in one rush, which gives a nice speed-up */
|
||||
limit = row_info->rowbytes - 16 + 1;
|
||||
for ( ; i < limit; i += 16 )
|
||||
if ( row_info->rowbytes > 15 )
|
||||
{
|
||||
unsigned char* base = &data[i];
|
||||
/* process blocks of 16 bytes in one rush, which gives a nice speed-up */
|
||||
limit = row_info->rowbytes - 16 + 1;
|
||||
for ( ; i < limit; i += 16 )
|
||||
{
|
||||
unsigned char* base = &data[i];
|
||||
|
||||
v82 s, s0, s1, a;
|
||||
v82 s, s0, s1, a;
|
||||
|
||||
/* clang <= 3.9 can't apply scalar values to vectors */
|
||||
/* (or rather, it needs a different syntax) */
|
||||
v82 n0x80 = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
|
||||
v82 n0xFF = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
|
||||
v82 n8 = { 8, 8, 8, 8, 8, 8, 8, 8 };
|
||||
/* clang <= 3.9 can't apply scalar values to vectors */
|
||||
/* (or rather, it needs a different syntax) */
|
||||
v82 n0x80 = { 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80, 0x80 };
|
||||
v82 n0xFF = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
|
||||
v82 n8 = { 8, 8, 8, 8, 8, 8, 8, 8 };
|
||||
|
||||
v82 ma = { 1, 1, 3, 3, 5, 5, 7, 7 };
|
||||
v82 o1 = { 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF };
|
||||
v82 m0 = { 1, 0, 3, 2, 5, 4, 7, 6 };
|
||||
v82 ma = { 1, 1, 3, 3, 5, 5, 7, 7 };
|
||||
v82 o1 = { 0, 0xFF, 0, 0xFF, 0, 0xFF, 0, 0xFF };
|
||||
v82 m0 = { 1, 0, 3, 2, 5, 4, 7, 6 };
|
||||
|
||||
|
||||
memcpy( &s, base, 16 ); /* RGBA RGBA RGBA RGBA */
|
||||
s0 = s & n0xFF; /* R B R B R B R B */
|
||||
s1 = s >> n8; /* G A G A G A G A */
|
||||
memcpy( &s, base, 16 ); /* RGBA RGBA RGBA RGBA */
|
||||
s0 = s & n0xFF; /* R B R B R B R B */
|
||||
s1 = s >> n8; /* G A G A G A G A */
|
||||
|
||||
a = vector_shuffle( s1, ma ); /* A A A A A A A A */
|
||||
s1 |= o1; /* G 1 G 1 G 1 G 1 */
|
||||
s0 = vector_shuffle( s0, m0 ); /* B R B R B R B R */
|
||||
a = vector_shuffle( s1, ma ); /* A A A A A A A A */
|
||||
s1 |= o1; /* G 1 G 1 G 1 G 1 */
|
||||
s0 = vector_shuffle( s0, m0 ); /* B R B R B R B R */
|
||||
|
||||
s0 *= a;
|
||||
s1 *= a;
|
||||
s0 += n0x80;
|
||||
s1 += n0x80;
|
||||
s0 = ( s0 + ( s0 >> n8 ) ) >> n8;
|
||||
s1 = ( s1 + ( s1 >> n8 ) ) >> n8;
|
||||
s0 *= a;
|
||||
s1 *= a;
|
||||
s0 += n0x80;
|
||||
s1 += n0x80;
|
||||
s0 = ( s0 + ( s0 >> n8 ) ) >> n8;
|
||||
s1 = ( s1 + ( s1 >> n8 ) ) >> n8;
|
||||
|
||||
s = s0 | ( s1 << n8 );
|
||||
memcpy( base, &s, 16 );
|
||||
s = s0 | ( s1 << n8 );
|
||||
memcpy( base, &s, 16 );
|
||||
}
|
||||
}
|
||||
#endif /* use `vector_size' */
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user