void test_vzipQs16 (void)
{
int16x8x2_t out_int16x8x2_t;
int16x8_t arg0_int16x8_t;
int16x8_t arg1_int16x8_t;
out_int16x8x2_t = vzipq_s16 (arg0_int16x8_t, arg1_int16x8_t);
}
void multadd_real_vector_complex_scalar(int16_t *x,
int16_t *alpha,
int16_t *y,
uint32_t N)
{
uint32_t i;
// do 8 multiplications at a time
simd_q15_t alpha_r_128,alpha_i_128,yr,yi,*x_128=(simd_q15_t*)x,*y_128=(simd_q15_t*)y;
int j;
// printf("alpha = %d,%d\n",alpha[0],alpha[1]);
alpha_r_128 = set1_int16(alpha[0]);
alpha_i_128 = set1_int16(alpha[1]);
j=0;
for (i=0; i<N>>3; i++) {
yr = mulhi_s1_int16(alpha_r_128,x_128[i]);
yi = mulhi_s1_int16(alpha_i_128,x_128[i]);
#if defined(__x86_64__) || defined(__i386__)
y_128[j] = _mm_adds_epi16(y_128[j],_mm_unpacklo_epi16(yr,yi));
j++;
y_128[j] = _mm_adds_epi16(y_128[j],_mm_unpackhi_epi16(yr,yi));
j++;
#elif defined(__arm__)
int16x8x2_t yint;
yint = vzipq_s16(yr,yi);
y_128[j] = adds_int16(y_128[j],yint.val[0]);
j++;
y_128[j] = adds_int16(y_128[j],yint.val[1]);
j++;
#endif
}
_mm_empty();
_m_empty();
}
static void SharpYUVFilterRow_NEON(const int16_t* A, const int16_t* B, int len,
const uint16_t* best_y, uint16_t* out) {
int i;
const int16x8_t max = vdupq_n_s16(MAX_Y);
const int16x8_t zero = vdupq_n_s16(0);
for (i = 0; i + 8 <= len; i += 8) {
const int16x8_t a0 = vld1q_s16(A + i + 0);
const int16x8_t a1 = vld1q_s16(A + i + 1);
const int16x8_t b0 = vld1q_s16(B + i + 0);
const int16x8_t b1 = vld1q_s16(B + i + 1);
const int16x8_t a0b1 = vaddq_s16(a0, b1);
const int16x8_t a1b0 = vaddq_s16(a1, b0);
const int16x8_t a0a1b0b1 = vaddq_s16(a0b1, a1b0); // A0+A1+B0+B1
const int16x8_t a0b1_2 = vaddq_s16(a0b1, a0b1); // 2*(A0+B1)
const int16x8_t a1b0_2 = vaddq_s16(a1b0, a1b0); // 2*(A1+B0)
const int16x8_t c0 = vshrq_n_s16(vaddq_s16(a0b1_2, a0a1b0b1), 3);
const int16x8_t c1 = vshrq_n_s16(vaddq_s16(a1b0_2, a0a1b0b1), 3);
const int16x8_t d0 = vaddq_s16(c1, a0);
const int16x8_t d1 = vaddq_s16(c0, a1);
const int16x8_t e0 = vrshrq_n_s16(d0, 1);
const int16x8_t e1 = vrshrq_n_s16(d1, 1);
const int16x8x2_t f = vzipq_s16(e0, e1);
const int16x8_t g0 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 0));
const int16x8_t g1 = vreinterpretq_s16_u16(vld1q_u16(best_y + 2 * i + 8));
const int16x8_t h0 = vaddq_s16(g0, f.val[0]);
const int16x8_t h1 = vaddq_s16(g1, f.val[1]);
const int16x8_t i0 = vmaxq_s16(vminq_s16(h0, max), zero);
const int16x8_t i1 = vmaxq_s16(vminq_s16(h1, max), zero);
vst1q_u16(out + 2 * i + 0, vreinterpretq_u16_s16(i0));
vst1q_u16(out + 2 * i + 8, vreinterpretq_u16_s16(i1));
}
for (; i < len; ++i) {
const int a0b1 = A[i + 0] + B[i + 1];
const int a1b0 = A[i + 1] + B[i + 0];
const int a0a1b0b1 = a0b1 + a1b0 + 8;
const int v0 = (8 * A[i + 0] + 2 * a1b0 + a0a1b0b1) >> 4;
const int v1 = (8 * A[i + 1] + 2 * a0b1 + a0a1b0b1) >> 4;
out[2 * i + 0] = clip_y(best_y[2 * i + 0] + v0);
out[2 * i + 1] = clip_y(best_y[2 * i + 1] + v1);
}
}
int16x8x2_t test_vzipq_s16(int16x8_t a, int16x8_t b) {
// CHECK-LABEL: test_vzipq_s16
return vzipq_s16(a, b);
// CHECK: zip1 {{v[0-9]+}}.8h, {{v[0-9]+}}.8h
// CHECK: zip2 {{v[0-9]+}}.8h, {{v[0-9]+}}.8h
}
