evas: implement _op_blend_mas_can_dp_neon in NEON intrinsics.
Reviewers: raster, cedric Reviewed By: cedric Projects: #efl Differential Revision: https://phab.enlightenment.org/D2369 Signed-off-by: Cedric BAIL <cedric@osg.samsung.com>
This commit is contained in:
parent
be7c7c2c77
commit
970afe9bea
|
@ -279,25 +279,123 @@ _op_blend_mas_c_dp_neon(DATA32 *s EINA_UNUSED, DATA8 *m, DATA32 c, DATA32 *d, in
|
|||
static void
|
||||
_op_blend_mas_can_dp_neon(DATA32 *s EINA_UNUSED, DATA8 *m, DATA32 c, DATA32 *d, int l) {
|
||||
#ifdef BUILD_NEON_INTRINSICS
|
||||
DATA32 *e;
|
||||
int alpha;
|
||||
UNROLL8_PLD_WHILE(d, l, e,
|
||||
{
|
||||
alpha = *m;
|
||||
switch(alpha)
|
||||
{
|
||||
case 0:
|
||||
break;
|
||||
case 255:
|
||||
*d = c;
|
||||
break;
|
||||
default:
|
||||
alpha++;
|
||||
*d = INTERP_256(alpha, c, *d);
|
||||
break;
|
||||
}
|
||||
m++; d++;
|
||||
});
|
||||
int16x8_t c_i16x8;
|
||||
int16x8_t d0_i16x8;
|
||||
int16x8_t d1_i16x8;
|
||||
int16x8_t dc0_i16x8;
|
||||
int16x8_t dc1_i16x8;
|
||||
int16x8_t m0_i16x8;
|
||||
int16x8_t m1_i16x8;
|
||||
int8x16_t dc_i8x16;
|
||||
int8x8_t dc0_i8x8;
|
||||
int8x8_t dc1_i8x8;
|
||||
uint16x8_t c_16x8;
|
||||
uint16x8_t d0_16x8;
|
||||
uint16x8_t d1_16x8;
|
||||
uint16x8_t m0_16x8;
|
||||
uint16x8_t m1_16x8;
|
||||
uint16x8_t m_16x8;
|
||||
uint32x2_t c_32x2;
|
||||
uint32x2_t m_32x2;
|
||||
uint32x4_t d_32x4;
|
||||
uint32x4_t dc_32x4;
|
||||
uint32x4_t m_32x4;
|
||||
uint32x4_t x1_32x4;
|
||||
uint8x16_t d_8x16;
|
||||
uint8x16_t m_8x16;
|
||||
uint8x16_t x1_8x16;
|
||||
uint8x8_t c_8x8;
|
||||
uint8x8_t d0_8x8;
|
||||
uint8x8_t d1_8x8;
|
||||
uint8x8_t m0_8x8;
|
||||
uint8x8_t m1_8x8;
|
||||
uint8x8_t m_8x8;
|
||||
uint8x8_t x1_8x8;
|
||||
uint32x4_t x0_32x4;
|
||||
uint32x4_t cond_32x4;
|
||||
|
||||
c_32x2 = vdup_n_u32(c);
|
||||
c_8x8 = vreinterpret_u8_u32(c_32x2);
|
||||
c_16x8 = vmovl_u8(c_8x8);
|
||||
c_i16x8 = vreinterpretq_s16_u16(c_16x8);
|
||||
x1_8x16 = vdupq_n_u8(0x1);
|
||||
x1_8x8 = vget_low_u8(x1_8x16);
|
||||
x1_32x4 = vreinterpretq_u32_u8(x1_8x16);
|
||||
x0_32x4 = vdupq_n_u32(0x0);
|
||||
|
||||
DATA32 *start = d;
|
||||
int size = l;
|
||||
DATA32 *end = start + (size & ~3);
|
||||
while (start < end) {
|
||||
int k = *((int *)m);
|
||||
if (k == 0)
|
||||
{
|
||||
m+=4;
|
||||
start+=4;
|
||||
continue;
|
||||
}
|
||||
|
||||
m_32x2 = vld1_lane_u32((DATA32*)m, m_32x2, 0);
|
||||
d_32x4 = vld1q_u32(start);
|
||||
d_8x16 = vreinterpretq_u8_u32(d_32x4);
|
||||
d0_8x8 = vget_low_u8(d_8x16);
|
||||
d1_8x8 = vget_high_u8(d_8x16);
|
||||
|
||||
m_8x8 = vreinterpret_u8_u32(m_32x2);
|
||||
m_16x8 = vmovl_u8(m_8x8);
|
||||
m_8x16 = vreinterpretq_u8_u16(m_16x8);
|
||||
m_8x8 = vget_low_u8(m_8x16);
|
||||
m_16x8 = vmovl_u8(m_8x8);
|
||||
m_32x4 = vreinterpretq_u32_u16(m_16x8);
|
||||
|
||||
m_32x4 = vmulq_u32(m_32x4, x1_32x4);
|
||||
m_8x16 = vreinterpretq_u8_u32(m_32x4);
|
||||
m0_8x8 = vget_low_u8(m_8x16);
|
||||
m1_8x8 = vget_high_u8(m_8x16);
|
||||
m0_16x8 = vaddl_u8(m0_8x8, x1_8x8);
|
||||
m1_16x8 = vaddl_u8(m1_8x8, x1_8x8);
|
||||
|
||||
m0_i16x8 = vreinterpretq_s16_u16(m0_16x8);
|
||||
m1_i16x8 = vreinterpretq_s16_u16(m1_16x8);
|
||||
|
||||
d0_16x8 = vmovl_u8(d0_8x8);
|
||||
d1_16x8 = vmovl_u8(d1_8x8);
|
||||
|
||||
d0_i16x8 = vreinterpretq_s16_u16(d0_16x8);
|
||||
d1_i16x8 = vreinterpretq_s16_u16(d1_16x8);
|
||||
|
||||
dc0_i16x8 = vsubq_s16(c_i16x8, d0_i16x8);
|
||||
dc1_i16x8 = vsubq_s16(c_i16x8, d1_i16x8);
|
||||
|
||||
dc0_i16x8 = vmulq_s16(dc0_i16x8, m0_i16x8);
|
||||
dc1_i16x8 = vmulq_s16(dc1_i16x8, m1_i16x8);
|
||||
|
||||
dc0_i16x8 = vshrq_n_s16(dc0_i16x8, 8);
|
||||
dc1_i16x8 = vshrq_n_s16(dc1_i16x8, 8);
|
||||
|
||||
dc0_i16x8 = vaddq_s16(dc0_i16x8, d0_i16x8);
|
||||
dc1_i16x8 = vaddq_s16(dc1_i16x8, d1_i16x8);
|
||||
|
||||
dc0_i8x8 = vmovn_s16(dc0_i16x8);
|
||||
dc1_i8x8 = vmovn_s16(dc1_i16x8);
|
||||
|
||||
dc_i8x16 = vcombine_s8(dc0_i8x8, dc1_i8x8);
|
||||
dc_32x4 = vreinterpretq_u32_s8(dc_i8x16);
|
||||
|
||||
cond_32x4 = vceqq_u32(m_32x4, x0_32x4);
|
||||
dc_32x4 = vbslq_u32(cond_32x4, d_32x4, dc_32x4);
|
||||
|
||||
vst1q_u32(start, dc_32x4);
|
||||
m+=4;
|
||||
start+=4;
|
||||
}
|
||||
end += (size & 3);
|
||||
while (start < end) {
|
||||
DATA32 alpha = *m;
|
||||
alpha++;
|
||||
*start = INTERP_256(alpha, c, *start);
|
||||
m++; start++;
|
||||
}
|
||||
#else
|
||||
DATA32 *e,*tmp;
|
||||
int alpha;
|
||||
|
|
Loading…
Reference in New Issue