void test_vgetQ_lanes16 (void)
{
int16_t out_int16_t;
int16x8_t arg0_int16x8_t;
out_int16_t = vgetq_lane_s16 (arg0_int16x8_t, 1);
}
/* s16x8 mv mul */
void mw_neon_mv_mul_s16x8(short * A, int Row, int T, short * B, short * C)
{
int i = 0;
int k = 0;
int16x8_t neon_b, neon_c;
int16x8_t neon_a0, neon_a1, neon_a2, neon_a3, neon_a4, neon_a5, neon_a6, neon_a7;
int16x8_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_s16(0);
for (k = 0; k < T; k+=8)
{
int j = k * T + i;
neon_a0 = vld1q_s16(A + j);
j+=Row;
neon_a1 = vld1q_s16(A + j);
j+=Row;
neon_a2 = vld1q_s16(A + j);
j+=Row;
neon_a3 = vld1q_s16(A + j);
j+=Row;
neon_a4 = vld1q_s16(A + j);
j+=Row;
neon_a5 = vld1q_s16(A + j);
j+=Row;
neon_a6 = vld1q_s16(A + j);
j+=Row;
neon_a7 = vld1q_s16(A + j);
neon_b = vld1q_s16(B + k);
neon_b0 = vdupq_n_s16(vgetq_lane_s16(neon_b, 0));
neon_b1 = vdupq_n_s16(vgetq_lane_s16(neon_b, 1));
neon_b2 = vdupq_n_s16(vgetq_lane_s16(neon_b, 2));
neon_b3 = vdupq_n_s16(vgetq_lane_s16(neon_b, 3));
neon_b4 = vdupq_n_s16(vgetq_lane_s16(neon_b, 4));
neon_b5 = vdupq_n_s16(vgetq_lane_s16(neon_b, 5));
neon_b6 = vdupq_n_s16(vgetq_lane_s16(neon_b, 6));
neon_b7 = vdupq_n_s16(vgetq_lane_s16(neon_b, 7));
neon_c = vaddq_s16(vmulq_s16(neon_a0, neon_b0), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a1, neon_b1), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a2, neon_b2), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a3, neon_b3), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a4, neon_b4), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a5, neon_b5), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a6, neon_b6), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a7, neon_b7), neon_c);
}
vst1q_s16(C + i, neon_c);
}
}
int32_t test_vgetq_lane_s16(int16x8_t v1) {
// CHECK: test_vgetq_lane_s16
return vgetq_lane_s16(v1, 6)+1;
// CHECK: smov {{w[0-9]+}}, {{v[0-9]+}}.h[6]
}
rfx_dwt_2d_decode_block_horiz_NEON(INT16 * l, INT16 * h, INT16 * dst, int subband_width)
{
int y, n;
INT16 * l_ptr = l;
INT16 * h_ptr = h;
INT16 * dst_ptr = dst;
for (y = 0; y < subband_width; y++)
{
/* Even coefficients */
for (n = 0; n < subband_width; n+=8)
{
// dst[2n] = l[n] - ((h[n-1] + h[n] + 1) >> 1);
int16x8_t l_n = vld1q_s16(l_ptr);
int16x8_t h_n = vld1q_s16(h_ptr);
int16x8_t h_n_m = vld1q_s16(h_ptr - 1);
if (n == 0)
{
int16_t first = vgetq_lane_s16(h_n_m, 1);
h_n_m = vsetq_lane_s16(first, h_n_m, 0);
}
int16x8_t tmp_n = vaddq_s16(h_n, h_n_m);
tmp_n = vaddq_s16(tmp_n, vdupq_n_s16(1));
tmp_n = vshrq_n_s16(tmp_n, 1);
int16x8_t dst_n = vsubq_s16(l_n, tmp_n);
vst1q_s16(l_ptr, dst_n);
l_ptr+=8;
h_ptr+=8;
}
l_ptr -= subband_width;
h_ptr -= subband_width;
/* Odd coefficients */
for (n = 0; n < subband_width; n+=8)
{
// dst[2n + 1] = (h[n] << 1) + ((dst[2n] + dst[2n + 2]) >> 1);
int16x8_t h_n = vld1q_s16(h_ptr);
h_n = vshlq_n_s16(h_n, 1);
int16x8x2_t dst_n;
dst_n.val[0] = vld1q_s16(l_ptr);
int16x8_t dst_n_p = vld1q_s16(l_ptr + 1);
if (n == subband_width - 8)
{
int16_t last = vgetq_lane_s16(dst_n_p, 6);
dst_n_p = vsetq_lane_s16(last, dst_n_p, 7);
}
dst_n.val[1] = vaddq_s16(dst_n_p, dst_n.val[0]);
dst_n.val[1] = vshrq_n_s16(dst_n.val[1], 1);
dst_n.val[1] = vaddq_s16(dst_n.val[1], h_n);
vst2q_s16(dst_ptr, dst_n);
l_ptr+=8;
h_ptr+=8;
dst_ptr+=16;
}
}
}
int16_t test_vgetq_lane_s16(int16x8_t a) {
// CHECK-LABEL: test_vgetq_lane_s16:
// CHECK-NEXT: umov.h w0, v0[7]
// CHECK-NEXT: ret
return vgetq_lane_s16(a, 7);
}
/* s16x8 mm mul */
void mw_neon_mm_mul_s16x8(short * A, int Row, int T, short * B, int Col, short * C)
{
int i, k, j;
int16x8_t neon_b, neon_c;
int16x8_t neon_a0, neon_a1, neon_a2, neon_a3, neon_a4, neon_a5, neon_a6, neon_a7;
int16x8_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_s16(0);
for (j = 0; j < T; j+=8)
{
int j_T = j * T + i;
int k_Row = k * Row;
neon_a0 = vld1q_s16(A + j_T);
j_T+=Row;
neon_a1 = vld1q_s16(A + j_T);
j_T+=Row;
neon_a2 = vld1q_s16(A + j_T);
j_T+=Row;
neon_a3 = vld1q_s16(A + j_T);
j_T+=Row;
neon_a4 = vld1q_s16(A + j_T);
j_T+=Row;
neon_a5 = vld1q_s16(A + j_T);
j_T+=Row;
neon_a6 = vld1q_s16(A + j_T);
j_T+=Row;
neon_a7 = vld1q_s16(A + j_T);
neon_b = vld1q_s16(B + k_Row + j);
neon_b0 = vdupq_n_s16(vgetq_lane_s16(neon_b, 0));
neon_b1 = vdupq_n_s16(vgetq_lane_s16(neon_b, 1));
neon_b2 = vdupq_n_s16(vgetq_lane_s16(neon_b, 2));
neon_b3 = vdupq_n_s16(vgetq_lane_s16(neon_b, 3));
neon_b4 = vdupq_n_s16(vgetq_lane_s16(neon_b, 4));
neon_b5 = vdupq_n_s16(vgetq_lane_s16(neon_b, 5));
neon_b6 = vdupq_n_s16(vgetq_lane_s16(neon_b, 6));
neon_b7 = vdupq_n_s16(vgetq_lane_s16(neon_b, 7));
neon_c = vaddq_s16(vmulq_s16(neon_a0, neon_b0), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a1, neon_b1), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a2, neon_b2), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a3, neon_b3), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a4, neon_b4), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a5, neon_b5), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a6, neon_b6), neon_c);
neon_c = vaddq_s16(vmulq_s16(neon_a7, neon_b7), neon_c);
vst1q_lane_s16(C + k_Row + i, neon_c, 0);
vst1q_lane_s16(C + k_Row + i + 1, neon_c, 1);
vst1q_lane_s16(C + k_Row + i + 2, neon_c, 2);
vst1q_lane_s16(C + k_Row + i + 3, neon_c, 3);
vst1q_lane_s16(C + k_Row + i + 4, neon_c, 4);
vst1q_lane_s16(C + k_Row + i + 5, neon_c, 5);
vst1q_lane_s16(C + k_Row + i + 6, neon_c, 6);
vst1q_lane_s16(C + k_Row + i + 7, neon_c, 7);
}
}
}
}
// CHECK-LABEL: define i16 @test_vgetq_lane_s16(<8 x i16> %a) #0 {
// CHECK: [[TMP0:%.*]] = bitcast <8 x i16> %a to <16 x i8>
// CHECK: [[TMP1:%.*]] = bitcast <16 x i8> [[TMP0]] to <8 x i16>
// CHECK: [[VGETQ_LANE:%.*]] = extractelement <8 x i16> [[TMP1]], i32 7
// CHECK: ret i16 [[VGETQ_LANE]]
int16_t test_vgetq_lane_s16(int16x8_t a) {
return vgetq_lane_s16(a, 7);
}
