예제 #1
0
void vp9_tm_predictor_4x4_neon(uint8_t *dst, ptrdiff_t stride,
                               const uint8_t *above, const uint8_t *left) {
  int i;
  uint16x8_t q1u16, q3u16;
  int16x8_t q1s16;
  uint8x8_t d0u8 = vdup_n_u8(0);
  uint32x2_t d2u32 = vdup_n_u32(0);

  d0u8 = vld1_dup_u8(above - 1);
  d2u32 = vld1_lane_u32((const uint32_t *)above, d2u32, 0);
  q3u16 = vsubl_u8(vreinterpret_u8_u32(d2u32), d0u8);
  for (i = 0; i < 4; i++, dst += stride) {
    q1u16 = vdupq_n_u16((uint16_t)left[i]);
    q1s16 = vaddq_s16(vreinterpretq_s16_u16(q1u16),
                      vreinterpretq_s16_u16(q3u16));
    d0u8 = vqmovun_s16(q1s16);
    vst1_lane_u32((uint32_t *)dst, vreinterpret_u32_u8(d0u8), 0);
  }
}
예제 #2
0
void vp9_tm_predictor_8x8_neon(uint8_t *dst, ptrdiff_t stride,
                               const uint8_t *above, const uint8_t *left) {
  int j;
  uint16x8_t q0u16, q3u16, q10u16;
  int16x8_t q0s16;
  uint16x4_t d20u16;
  uint8x8_t d0u8, d2u8, d30u8;

  d0u8 = vld1_dup_u8(above - 1);
  d30u8 = vld1_u8(left);
  d2u8 = vld1_u8(above);
  q10u16 = vmovl_u8(d30u8);
  q3u16 = vsubl_u8(d2u8, d0u8);
  d20u16 = vget_low_u16(q10u16);
  for (j = 0; j < 2; j++, d20u16 = vget_high_u16(q10u16)) {
    q0u16 = vdupq_lane_u16(d20u16, 0);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += stride;
    q0u16 = vdupq_lane_u16(d20u16, 1);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += stride;
    q0u16 = vdupq_lane_u16(d20u16, 2);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += stride;
    q0u16 = vdupq_lane_u16(d20u16, 3);
    q0s16 = vaddq_s16(vreinterpretq_s16_u16(q3u16),
                      vreinterpretq_s16_u16(q0u16));
    d0u8 = vqmovun_s16(q0s16);
    vst1_u64((uint64_t *)dst, vreinterpret_u64_u8(d0u8));
    dst += stride;
  }
}
예제 #3
0
파일: vld1_dupu8.c 프로젝트: 0day-ci/gcc
void test_vld1_dupu8 (void)
{
  uint8x8_t out_uint8x8_t;

  out_uint8x8_t = vld1_dup_u8 (0);
}