/* s32x4 mm mul */
void mw_neon_mm_mul_s32x4(int * A, int Row, int T, int * B, int Col, int * C)
{
int i, k, j;
int32x4_t neon_b, neon_c;
int32x4_t neon_a0, neon_a1, neon_a2, neon_a3;
int32x4_t neon_b0, neon_b1, neon_b2, neon_b3;
for (i = 0; i < Row; i+=4)
{
for (k = 0; k < Col; k+=1)
{
neon_c = vmovq_n_s32(0);
for (j = 0; j < T; j+=4)
{
int j_T = j * T + i;
int k_Row = k * Row;
neon_a0 = vld1q_s32(A + j_T);
j_T+=Row;
neon_a1 = vld1q_s32(A + j_T);
j_T+=Row;
neon_a2 = vld1q_s32(A + j_T);
j_T+=Row;
neon_a3 = vld1q_s32(A + j_T);
neon_b = vld1q_s32(B + k_Row + j);
neon_b0 = vdupq_n_s32(vgetq_lane_s32(neon_b, 0));
neon_b1 = vdupq_n_s32(vgetq_lane_s32(neon_b, 1));
neon_b2 = vdupq_n_s32(vgetq_lane_s32(neon_b, 2));
neon_b3 = vdupq_n_s32(vgetq_lane_s32(neon_b, 3));
neon_c = vaddq_s32(vmulq_s32(neon_a0, neon_b0), neon_c);
neon_c = vaddq_s32(vmulq_s32(neon_a1, neon_b1), neon_c);
neon_c = vaddq_s32(vmulq_s32(neon_a2, neon_b2), neon_c);
neon_c = vaddq_s32(vmulq_s32(neon_a3, neon_b3), neon_c);
vst1q_lane_s32(C + k_Row + i, neon_c, 0);
vst1q_lane_s32(C + k_Row + i + 1, neon_c, 1);
vst1q_lane_s32(C + k_Row + i + 2, neon_c, 2);
vst1q_lane_s32(C + k_Row + i + 3, neon_c, 3);
}
}
}
}
void test_vmulQs32 (void)
{
int32x4_t out_int32x4_t;
int32x4_t arg0_int32x4_t;
int32x4_t arg1_int32x4_t;
out_int32x4_t = vmulq_s32 (arg0_int32x4_t, arg1_int32x4_t);
}
/* s32x4 mv mul */
void mw_neon_mv_mul_s32x4(int * A, int Row, int T, int * B, int * C)
{
int i = 0;
int k = 0;
int32x4_t neon_b, neon_c;
int32x4_t neon_a0, neon_a1, neon_a2, neon_a3;
int32x4_t neon_b0, neon_b1, neon_b2, neon_b3;
for (i = 0; i < Row; i+=4)
{
neon_c = vmovq_n_s32(0);
for (k = 0; k < T; k+=4)
{
int j = k * T + i;
neon_a0 = vld1q_s32(A + j);
j+=Row;
neon_a1 = vld1q_s32(A + j);
j+=Row;
neon_a2 = vld1q_s32(A + j);
j+=Row;
neon_a3 = vld1q_s32(A + j);
neon_b = vld1q_s32(B + k);
neon_b0 = vdupq_n_s32(vgetq_lane_s32(neon_b, 0));
neon_b1 = vdupq_n_s32(vgetq_lane_s32(neon_b, 1));
neon_b2 = vdupq_n_s32(vgetq_lane_s32(neon_b, 2));
neon_b3 = vdupq_n_s32(vgetq_lane_s32(neon_b, 3));
neon_c = vaddq_s32(vmulq_s32(neon_a0, neon_b0), neon_c);
neon_c = vaddq_s32(vmulq_s32(neon_a1, neon_b1), neon_c);
neon_c = vaddq_s32(vmulq_s32(neon_a2, neon_b2), neon_c);
neon_c = vaddq_s32(vmulq_s32(neon_a3, neon_b3), neon_c);
}
vst1q_s32(C + i, neon_c);
}
}
static INLINE void IADST4x4_1D(int16x4_t *d3s16, int16x4_t *d4s16,
int16x4_t *d5s16, int16x8_t *q3s16,
int16x8_t *q8s16, int16x8_t *q9s16) {
int16x4_t d6s16, d16s16, d17s16, d18s16, d19s16;
int32x4_t q8s32, q9s32, q10s32, q11s32, q12s32, q13s32, q14s32, q15s32;
d6s16 = vget_low_s16(*q3s16);
d16s16 = vget_low_s16(*q8s16);
d17s16 = vget_high_s16(*q8s16);
d18s16 = vget_low_s16(*q9s16);
d19s16 = vget_high_s16(*q9s16);
q10s32 = vmull_s16(*d3s16, d16s16);
q11s32 = vmull_s16(*d4s16, d16s16);
q12s32 = vmull_s16(d6s16, d17s16);
q13s32 = vmull_s16(*d5s16, d18s16);
q14s32 = vmull_s16(*d3s16, d18s16);
q15s32 = vmovl_s16(d16s16);
q15s32 = vaddw_s16(q15s32, d19s16);
q8s32 = vmull_s16(*d4s16, d19s16);
q15s32 = vsubw_s16(q15s32, d18s16);
q9s32 = vmull_s16(*d5s16, d19s16);
q10s32 = vaddq_s32(q10s32, q13s32);
q10s32 = vaddq_s32(q10s32, q8s32);
q11s32 = vsubq_s32(q11s32, q14s32);
q8s32 = vdupq_n_s32(sinpi_3_9);
q11s32 = vsubq_s32(q11s32, q9s32);
q15s32 = vmulq_s32(q15s32, q8s32);
q13s32 = vaddq_s32(q10s32, q12s32);
q10s32 = vaddq_s32(q10s32, q11s32);
q14s32 = vaddq_s32(q11s32, q12s32);
q10s32 = vsubq_s32(q10s32, q12s32);
d16s16 = vqrshrn_n_s32(q13s32, 14);
d17s16 = vqrshrn_n_s32(q14s32, 14);
d18s16 = vqrshrn_n_s32(q15s32, 14);
d19s16 = vqrshrn_n_s32(q10s32, 14);
*q8s16 = vcombine_s16(d16s16, d17s16);
*q9s16 = vcombine_s16(d18s16, d19s16);
return;
}
