}
}
-/*
- I use less accurate approximation here by simply left-shifting the input
- value and filling the low order bits with zeroes. This method improves PNG
- compression but this scheme cannot reproduce white exactly, since it does
- not generate an all-ones maximum value; the net effect is to darken the
- image slightly.
-
- The better method should be "left bit replication":
-
- 4 3 2 1 0
- ---------
- 1 1 0 1 1
-
- 7 6 5 4 3 2 1 0
- ----------------
- 1 1 0 1 1 1 1 0
- |=======| |===|
- | leftmost bits repeated to fill open bits
- |
- original bits
-*/
static inline void RENAME(rgb15tobgr24)(const uint8_t *src, uint8_t *dst, int src_size)
{
const uint16_t *end;
"pand %2, %%mm0 \n\t"
"pand %3, %%mm1 \n\t"
"pand %4, %%mm2 \n\t"
- "psllq $3, %%mm0 \n\t"
- "psrlq $2, %%mm1 \n\t"
- "psrlq $7, %%mm2 \n\t"
+ "psllq $5, %%mm0 \n\t"
+ "pmulhw %6, %%mm0 \n\t"
+ "pmulhw %6, %%mm1 \n\t"
+ "pmulhw %7, %%mm2 \n\t"
"movq %%mm0, %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"movq %%mm2, %%mm5 \n\t"
"pand %2, %%mm0 \n\t"
"pand %3, %%mm1 \n\t"
"pand %4, %%mm2 \n\t"
- "psllq $3, %%mm0 \n\t"
- "psrlq $2, %%mm1 \n\t"
- "psrlq $7, %%mm2 \n\t"
+ "psllq $5, %%mm0 \n\t"
+ "pmulhw %6, %%mm0 \n\t"
+ "pmulhw %6, %%mm1 \n\t"
+ "pmulhw %7, %%mm2 \n\t"
"movq %%mm0, %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"movq %%mm2, %%mm5 \n\t"
"por %%mm5, %%mm3 \n\t"
:"=m"(*d)
- :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r), "m"(mmx_null)
+ :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r),"m"(mmx_null),"m"(mul15_mid),"m"(mul15_hi)
:"memory");
/* borrowed 32 to 24 */
__asm__ volatile(
while (s < end) {
register uint16_t bgr;
bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x3E0)>>2;
- *d++ = (bgr&0x7C00)>>7;
+ *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
+ *d++ = ((bgr&0x3E0)>>2) | ((bgr&0x3E0)>>7);
+ *d++ = ((bgr&0x7C00)>>7) | ((bgr&0x7C00)>>12);
}
}
"pand %2, %%mm0 \n\t"
"pand %3, %%mm1 \n\t"
"pand %4, %%mm2 \n\t"
- "psllq $3, %%mm0 \n\t"
- "psrlq $3, %%mm1 \n\t"
- "psrlq $8, %%mm2 \n\t"
+ "psllq $5, %%mm0 \n\t"
+ "psrlq $1, %%mm2 \n\t"
+ "pmulhw %6, %%mm0 \n\t"
+ "pmulhw %8, %%mm1 \n\t"
+ "pmulhw %7, %%mm2 \n\t"
"movq %%mm0, %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"movq %%mm2, %%mm5 \n\t"
"pand %2, %%mm0 \n\t"
"pand %3, %%mm1 \n\t"
"pand %4, %%mm2 \n\t"
- "psllq $3, %%mm0 \n\t"
- "psrlq $3, %%mm1 \n\t"
- "psrlq $8, %%mm2 \n\t"
+ "psllq $5, %%mm0 \n\t"
+ "psrlq $1, %%mm2 \n\t"
+ "pmulhw %6, %%mm0 \n\t"
+ "pmulhw %8, %%mm1 \n\t"
+ "pmulhw %7, %%mm2 \n\t"
"movq %%mm0, %%mm3 \n\t"
"movq %%mm1, %%mm4 \n\t"
"movq %%mm2, %%mm5 \n\t"
"por %%mm4, %%mm3 \n\t"
"por %%mm5, %%mm3 \n\t"
:"=m"(*d)
- :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null)
+ :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mmx_null),"m"(mul15_mid),"m"(mul15_hi),"m"(mul16_mid)
:"memory");
/* borrowed 32 to 24 */
__asm__ volatile(
while (s < end) {
register uint16_t bgr;
bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x7E0)>>3;
- *d++ = (bgr&0xF800)>>8;
+ *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
+ *d++ = ((bgr&0x7E0)>>3) | ((bgr&0x7E0)>>9);
+ *d++ = ((bgr&0xF800)>>8) | ((bgr&0xF800)>>13);
}
}
"pand %2, %%mm0 \n\t"
"pand %3, %%mm1 \n\t"
"pand %4, %%mm2 \n\t"
- "psllq $3, %%mm0 \n\t"
- "psrlq $2, %%mm1 \n\t"
- "psrlq $7, %%mm2 \n\t"
+ "psllq $5, %%mm0 \n\t"
+ "pmulhw %5, %%mm0 \n\t"
+ "pmulhw %5, %%mm1 \n\t"
+ "pmulhw %6, %%mm2 \n\t"
PACK_RGB32
:"=m"(*d)
- :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r)
+ :"m"(*s),"m"(mask15b),"m"(mask15g),"m"(mask15r),"m"(mul15_mid),"m"(mul15_hi)
:"memory");
d += 16;
s += 4;
while (s < end) {
register uint16_t bgr;
bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x3E0)>>2;
- *d++ = (bgr&0x7C00)>>7;
+ *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
+ *d++ = ((bgr&0x3E0)>>2) | ((bgr&0x3E0)>>7);
+ *d++ = ((bgr&0x7C00)>>7) | ((bgr&0x7C00)>>12);
*d++ = 255;
}
}
"pand %2, %%mm0 \n\t"
"pand %3, %%mm1 \n\t"
"pand %4, %%mm2 \n\t"
- "psllq $3, %%mm0 \n\t"
- "psrlq $3, %%mm1 \n\t"
- "psrlq $8, %%mm2 \n\t"
+ "psllq $5, %%mm0 \n\t"
+ "psrlq $1, %%mm2 \n\t"
+ "pmulhw %5, %%mm0 \n\t"
+ "pmulhw %7, %%mm1 \n\t"
+ "pmulhw %6, %%mm2 \n\t"
PACK_RGB32
:"=m"(*d)
- :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r)
+ :"m"(*s),"m"(mask16b),"m"(mask16g),"m"(mask16r),"m"(mul15_mid),"m"(mul15_hi),"m"(mul16_mid)
:"memory");
d += 16;
s += 4;
while (s < end) {
register uint16_t bgr;
bgr = *s++;
- *d++ = (bgr&0x1F)<<3;
- *d++ = (bgr&0x7E0)>>3;
- *d++ = (bgr&0xF800)>>8;
+ *d++ = ((bgr&0x1F)<<3) | ((bgr&0x1F)>>2);
+ *d++ = ((bgr&0x7E0)>>3) | ((bgr&0x7E0)>>9);
+ *d++ = ((bgr&0xF800)>>8) | ((bgr&0xF800)>>13);
*d++ = 255;
}
}