void test_vmovQ_nu16 (void)
{
uint16x8_t out_uint16x8_t;
uint16_t arg0_uint16_t;
out_uint16x8_t = vmovq_n_u16 (arg0_uint16_t);
}
/* u16x8 mv mul */
void mw_neon_mv_mul_u16x8(unsigned short * A, int Row, int T, unsigned short * B, unsigned short * C)
{
int i = 0;
int k = 0;
uint16x8_t neon_b, neon_c;
uint16x8_t neon_a0, neon_a1, neon_a2, neon_a3, neon_a4, neon_a5, neon_a6, neon_a7;
uint16x8_t neon_b0, neon_b1, neon_b2, neon_b3, neon_b4, neon_b5, neon_b6, neon_b7;
for (i = 0; i < Row; i+=8)
{
neon_c = vmovq_n_u16(0);
for (k = 0; k < T; k+=8)
{
int j = k * T + i;
neon_a0 = vld1q_u16(A + j);
j+=Row;
neon_a1 = vld1q_u16(A + j);
j+=Row;
neon_a2 = vld1q_u16(A + j);
j+=Row;
neon_a3 = vld1q_u16(A + j);
j+=Row;
neon_a4 = vld1q_u16(A + j);
j+=Row;
neon_a5 = vld1q_u16(A + j);
j+=Row;
neon_a6 = vld1q_u16(A + j);
j+=Row;
neon_a7 = vld1q_u16(A + j);
neon_b = vld1q_u16(B + k);
neon_b0 = vdupq_n_u16(vgetq_lane_u16(neon_b, 0));
neon_b1 = vdupq_n_u16(vgetq_lane_u16(neon_b, 1));
neon_b2 = vdupq_n_u16(vgetq_lane_u16(neon_b, 2));
neon_b3 = vdupq_n_u16(vgetq_lane_u16(neon_b, 3));
neon_b4 = vdupq_n_u16(vgetq_lane_u16(neon_b, 4));
neon_b5 = vdupq_n_u16(vgetq_lane_u16(neon_b, 5));
neon_b6 = vdupq_n_u16(vgetq_lane_u16(neon_b, 6));
neon_b7 = vdupq_n_u16(vgetq_lane_u16(neon_b, 7));
neon_c = vaddq_u16(vmulq_u16(neon_a0, neon_b0), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a1, neon_b1), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a2, neon_b2), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a3, neon_b3), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a4, neon_b4), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a5, neon_b5), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a6, neon_b6), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a7, neon_b7), neon_c);
}
vst1q_u16(C + i, neon_c);
}
}
uint16x8_t test_vmovq_n_u16(uint16_t v1) {
// CHECK: test_vmovq_n_u16
return vmovq_n_u16(v1);
// CHECK: dup {{v[0-9]+}}.8h, {{w[0-9]+}}
}
/* u16x8 mm mul */
void mw_neon_mm_mul_u16x8(unsigned short * A, int Row, int T, unsigned short * B, int Col, unsigned short * C)
{
int i, k, j;
uint16x8_t neon_b, neon_c;
uint16x8_t neon_a0, neon_a1, neon_a2, neon_a3, neon_a4, neon_a5, neon_a6, neon_a7;
uint16x8_t neon_b0, neon_b1, neon_b2, neon_b3, neon_b4, neon_b5, neon_b6, neon_b7;
for (i = 0; i < Row; i+=8)
{
for (k = 0; k < Col; k+=1)
{
neon_c = vmovq_n_u16(0);
for (j = 0; j < T; j+=8)
{
int j_T = j * T + i;
int k_Row = k * Row;
neon_a0 = vld1q_u16(A + j_T);
j_T+=Row;
neon_a1 = vld1q_u16(A + j_T);
j_T+=Row;
neon_a2 = vld1q_u16(A + j_T);
j_T+=Row;
neon_a3 = vld1q_u16(A + j_T);
j_T+=Row;
neon_a4 = vld1q_u16(A + j_T);
j_T+=Row;
neon_a5 = vld1q_u16(A + j_T);
j_T+=Row;
neon_a6 = vld1q_u16(A + j_T);
j_T+=Row;
neon_a7 = vld1q_u16(A + j_T);
neon_b = vld1q_u16(B + k_Row + j);
neon_b0 = vdupq_n_u16(vgetq_lane_u16(neon_b, 0));
neon_b1 = vdupq_n_u16(vgetq_lane_u16(neon_b, 1));
neon_b2 = vdupq_n_u16(vgetq_lane_u16(neon_b, 2));
neon_b3 = vdupq_n_u16(vgetq_lane_u16(neon_b, 3));
neon_b4 = vdupq_n_u16(vgetq_lane_u16(neon_b, 4));
neon_b5 = vdupq_n_u16(vgetq_lane_u16(neon_b, 5));
neon_b6 = vdupq_n_u16(vgetq_lane_u16(neon_b, 6));
neon_b7 = vdupq_n_u16(vgetq_lane_u16(neon_b, 7));
neon_c = vaddq_u16(vmulq_u16(neon_a0, neon_b0), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a1, neon_b1), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a2, neon_b2), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a3, neon_b3), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a4, neon_b4), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a5, neon_b5), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a6, neon_b6), neon_c);
neon_c = vaddq_u16(vmulq_u16(neon_a7, neon_b7), neon_c);
vst1q_lane_u16(C + k_Row + i, neon_c, 0);
vst1q_lane_u16(C + k_Row + i + 1, neon_c, 1);
vst1q_lane_u16(C + k_Row + i + 2, neon_c, 2);
vst1q_lane_u16(C + k_Row + i + 3, neon_c, 3);
vst1q_lane_u16(C + k_Row + i + 4, neon_c, 4);
vst1q_lane_u16(C + k_Row + i + 5, neon_c, 5);
vst1q_lane_u16(C + k_Row + i + 6, neon_c, 6);
vst1q_lane_u16(C + k_Row + i + 7, neon_c, 7);
}
}
}
}
