static uint8x8_t
paeth(uint8x8_t a, uint8x8_t b, uint8x8_t c)
{
uint8x8_t d, e;
uint16x8_t p1, pa, pb, pc;
p1 = vaddl_u8(a, b); /* a + b */
pc = vaddl_u8(c, c); /* c * 2 */
pa = vabdl_u8(b, c); /* pa */
pb = vabdl_u8(a, c); /* pb */
pc = vabdq_u16(p1, pc); /* pc */
p1 = vcleq_u16(pa, pb); /* pa <= pb */
pa = vcleq_u16(pa, pc); /* pa <= pc */
pb = vcleq_u16(pb, pc); /* pb <= pc */
p1 = vandq_u16(p1, pa); /* pa <= pb && pa <= pc */
d = vmovn_u16(pb);
e = vmovn_u16(p1);
d = vbsl_u8(d, b, c);
e = vbsl_u8(e, a, d);
return e;
}
void test_vandQu16 (void)
{
uint16x8_t out_uint16x8_t;
uint16x8_t arg0_uint16x8_t;
uint16x8_t arg1_uint16x8_t;
out_uint16x8_t = vandq_u16 (arg0_uint16x8_t, arg1_uint16x8_t);
}
uint16x8_t test_vandq_u16(uint16x8_t a, uint16x8_t b) {
// CHECK-LABEL: test_vandq_u16
return vandq_u16(a, b);
// CHECK: and {{v[0-9]+}}.16b, {{v[0-9]+}}.16b, {{v[0-9]+}}.16b
}
inline uint16x8_t vandq(const uint16x8_t & v0, const uint16x8_t & v1) { return vandq_u16(v0, v1); }
int neon_new(DATA32* _p0, DATA32* _p1, DATA32* _p2, DATA32* _p3, DATA32* _ax, DATA32 _ay, DATA32* result, int len) {
int ay = _ay;
int i;
DATA32* pbuf = result;
uint16x4_t ay_16x4;
uint16x4_t p0_16x4;
uint16x4_t p2_16x4;
uint16x8_t ax_16x8;
uint16x8_t p0_p2_16x8;
uint16x8_t p1_p3_16x8;
uint16x8_t x255_16x8;
uint32x2_t p0_p2_32x2;
uint32x2_t p1_p3_32x2;
uint32x2_t res_32x2;
uint8x8_t p0_p2_8x8;
uint8x8_t p1_p3_8x8;
uint8x8_t p2_8x8;
uint16x4_t temp_16x4;
ay_16x4 = vdup_n_u16(ay);
x255_16x8 = vdupq_n_u16(0xff);
for(i = 0; i < len; i++) {
DATA32 p0 = *_p0++;
DATA32 p1 = *_p1++;
DATA32 p2 = *_p2++;
DATA32 p3 = *_p3++;
int ax = *_ax++;
if (p0 | p1 | p2 | p3)
{
ax_16x8 = vdupq_n_u16(ax);
p0_p2_32x2 = vset_lane_u32(p0, p0_p2_32x2, 0);
p0_p2_32x2 = vset_lane_u32(p2, p0_p2_32x2, 1);
p1_p3_32x2 = vset_lane_u32(p1, p1_p3_32x2, 0);
p1_p3_32x2 = vset_lane_u32(p3, p1_p3_32x2, 1);
p0_p2_8x8 = vreinterpret_u8_u32(p0_p2_32x2);
p1_p3_8x8 = vreinterpret_u8_u32(p1_p3_32x2);
p1_p3_16x8 = vmovl_u8(p1_p3_8x8);
p0_p2_16x8 = vmovl_u8(p0_p2_8x8);
p1_p3_16x8 = vsubq_u16(p1_p3_16x8, p0_p2_16x8);
p1_p3_16x8 = vmulq_u16(p1_p3_16x8, ax_16x8);
p1_p3_16x8 = vshrq_n_u16(p1_p3_16x8, 8);
p1_p3_16x8 = vaddq_u16(p1_p3_16x8, p0_p2_16x8);
p1_p3_16x8 = vandq_u16(p1_p3_16x8, x255_16x8);
p0_16x4 = vget_low_u16(p1_p3_16x8);
p2_16x4 = vget_high_u16(p1_p3_16x8);
p2_16x4 = vsub_u16(p2_16x4, p0_16x4);
p2_16x4 = vmul_u16(p2_16x4, ay_16x4);
p2_16x4 = vshr_n_u16(p2_16x4, 8);
p2_16x4 = vadd_u16(p2_16x4, p0_16x4);
p1_p3_16x8 = vcombine_u16(temp_16x4, p2_16x4);
p2_8x8 = vmovn_u16(p1_p3_16x8);
res_32x2 = vreinterpret_u32_u8(p2_8x8);
vst1_lane_u32(pbuf++, res_32x2, 1);
}
else
*pbuf++ = p0;
}
return 0;
}
