C++ CONVERT_TO_SHORTPTR函数代码示例

static void highbd_convolve_avg_horiz(const uint8_t *src8, ptrdiff_t src_stride,
                                      uint8_t *dst8, ptrdiff_t dst_stride,
                                      const InterpKernel *x_filters,
                                      int x0_q4, int x_step_q4,
                                      int w, int h, int bd) {
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= SUBPEL_TAPS / 2 - 1;
  for (y = 0; y < h; ++y) {
    int x_q4 = x0_q4;
    for (x = 0; x < w; ++x) {
      const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
      const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_x[k] * x_filter[k];
      dst[x] = ROUND_POWER_OF_TWO(dst[x] +
          clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
      x_q4 += x_step_q4;
    }
    src += src_stride;
    dst += dst_stride;
  }
}

void aom_yv12_partial_copy_v_c(const YV12_BUFFER_CONFIG *src_bc,
                               YV12_BUFFER_CONFIG *dst_bc, int hstart, int hend,
                               int vstart, int vend) {
  int row;
  const uint8_t *src = src_bc->v_buffer;
  uint8_t *dst = dst_bc->v_buffer;

  if (src_bc->flags & YV12_FLAG_HIGHBITDEPTH) {
    const uint16_t *src16 =
        CONVERT_TO_SHORTPTR(src + vstart * src_bc->uv_stride + hstart);
    uint16_t *dst16 =
        CONVERT_TO_SHORTPTR(dst + vstart * dst_bc->uv_stride + hstart);
    for (row = vstart; row < vend; ++row) {
      memcpy(dst16, src16, (hend - hstart) * sizeof(uint16_t));
      src16 += src_bc->uv_stride;
      dst16 += dst_bc->uv_stride;
    }
    return;
  }

  src = (src + vstart * src_bc->uv_stride + hstart);
  dst = (dst + vstart * dst_bc->uv_stride + hstart);

  for (row = vstart; row < vend; ++row) {
    memcpy(dst, src, (hend - hstart));
    src += src_bc->uv_stride;
    dst += dst_bc->uv_stride;
  }
}

    static INLINE
    unsigned int highbd_masked_sad(const uint8_t *src8, int src_stride,
                                   const uint8_t *a8, int a_stride,
                                   const uint8_t *b8, int b_stride,
                                   const uint8_t *m, int m_stride, int width,
                                   int height) {
  int y, x;
  unsigned int sad = 0;
  const uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);

  for (y = 0; y < height; y++) {
    for (x = 0; x < width; x++) {
      const uint16_t pred = AOM_BLEND_A64(m[x], a[x], b[x]);
      sad += abs(pred - src[x]);
    }

    src += src_stride;
    a += a_stride;
    b += b_stride;
    m += m_stride;
  }
  sad = (sad + 31) >> 6;

  return sad;
}

void aom_highbd_blend_a64_vmask_c(uint8_t *dst_8, uint32_t dst_stride,
                                  const uint8_t *src0_8, uint32_t src0_stride,
                                  const uint8_t *src1_8, uint32_t src1_stride,
                                  const uint8_t *mask, int h, int w, int bd) {
  int i, j;
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst_8);
  const uint16_t *src0 = CONVERT_TO_SHORTPTR(src0_8);
  const uint16_t *src1 = CONVERT_TO_SHORTPTR(src1_8);
  (void)bd;

  assert(IMPLIES(src0 == dst, src0_stride == dst_stride));
  assert(IMPLIES(src1 == dst, src1_stride == dst_stride));

  assert(h >= 1);
  assert(w >= 1);
  assert(IS_POWER_OF_TWO(h));
  assert(IS_POWER_OF_TWO(w));

  assert(bd == 8 || bd == 10 || bd == 12);

  for (i = 0; i < h; ++i) {
    const int m = mask[i];
    for (j = 0; j < w; ++j) {
      dst[i * dst_stride + j] = AOM_BLEND_A64(m, src0[i * src0_stride + j],
                                              src1[i * src1_stride + j]);
    }
  }
}

void vpx_yv12_copy_y_c(const YV12_BUFFER_CONFIG *src_ybc,
                       YV12_BUFFER_CONFIG *dst_ybc) {
  int row;
  const uint8_t *src = src_ybc->y_buffer;
  uint8_t *dst = dst_ybc->y_buffer;

#if CONFIG_VP9_HIGHBITDEPTH
  if (src_ybc->flags & YV12_FLAG_HIGHBITDEPTH) {
    const uint16_t *src16 = CONVERT_TO_SHORTPTR(src);
    uint16_t *dst16 = CONVERT_TO_SHORTPTR(dst);
    for (row = 0; row < src_ybc->y_height; ++row) {
      memcpy(dst16, src16, src_ybc->y_width * sizeof(uint16_t));
      src16 += src_ybc->y_stride;
      dst16 += dst_ybc->y_stride;
    }
    return;
  }
#endif

  for (row = 0; row < src_ybc->y_height; ++row) {
    memcpy(dst, src, src_ybc->y_width);
    src += src_ybc->y_stride;
    dst += dst_ybc->y_stride;
  }
}

static void highbd_convolve_avg_vert(const uint8_t *src8, ptrdiff_t src_stride,
                                     uint8_t *dst8, ptrdiff_t dst_stride,
                                     const InterpKernel *y_filters,
                                     int y0_q4, int y_step_q4, int w, int h,
                                     int bd) {
  int x, y;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  src -= src_stride * (SUBPEL_TAPS / 2 - 1);
  for (x = 0; x < w; ++x) {
    int y_q4 = y0_q4;
    for (y = 0; y < h; ++y) {
      const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
      const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
      int k, sum = 0;
      for (k = 0; k < SUBPEL_TAPS; ++k)
        sum += src_y[k * src_stride] * y_filter[k];
      dst[y * dst_stride] = ROUND_POWER_OF_TWO(dst[y * dst_stride] +
          clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd), 1);
      y_q4 += y_step_q4;
    }
    ++src;
    ++dst;
  }
}

static INLINE unsigned int highbd_masked_sad4xh_ssse3(
    const uint8_t *src8, int src_stride, const uint8_t *a8, int a_stride,
    const uint8_t *b8, int b_stride, const uint8_t *m_ptr, int m_stride,
    int height) {
  const uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *a_ptr = CONVERT_TO_SHORTPTR(a8);
  const uint16_t *b_ptr = CONVERT_TO_SHORTPTR(b8);
  int y;
  __m128i res = _mm_setzero_si128();
  const __m128i mask_max = _mm_set1_epi16((1 << AOM_BLEND_A64_ROUND_BITS));
  const __m128i round_const =
      _mm_set1_epi32((1 << AOM_BLEND_A64_ROUND_BITS) >> 1);
  const __m128i one = _mm_set1_epi16(1);

  for (y = 0; y < height; y += 2) {
    const __m128i src = _mm_unpacklo_epi64(
        _mm_loadl_epi64((const __m128i *)src_ptr),
        _mm_loadl_epi64((const __m128i *)&src_ptr[src_stride]));
    const __m128i a =
        _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)a_ptr),
                           _mm_loadl_epi64((const __m128i *)&a_ptr[a_stride]));
    const __m128i b =
        _mm_unpacklo_epi64(_mm_loadl_epi64((const __m128i *)b_ptr),
                           _mm_loadl_epi64((const __m128i *)&b_ptr[b_stride]));
    // Zero-extend mask to 16 bits
    const __m128i m = _mm_unpacklo_epi8(
        _mm_unpacklo_epi32(
            _mm_cvtsi32_si128(*(const uint32_t *)m_ptr),
            _mm_cvtsi32_si128(*(const uint32_t *)&m_ptr[m_stride])),
        _mm_setzero_si128());
    const __m128i m_inv = _mm_sub_epi16(mask_max, m);

    const __m128i data_l = _mm_unpacklo_epi16(a, b);
    const __m128i mask_l = _mm_unpacklo_epi16(m, m_inv);
    __m128i pred_l = _mm_madd_epi16(data_l, mask_l);
    pred_l = _mm_srai_epi32(_mm_add_epi32(pred_l, round_const),
                            AOM_BLEND_A64_ROUND_BITS);

    const __m128i data_r = _mm_unpackhi_epi16(a, b);
    const __m128i mask_r = _mm_unpackhi_epi16(m, m_inv);
    __m128i pred_r = _mm_madd_epi16(data_r, mask_r);
    pred_r = _mm_srai_epi32(_mm_add_epi32(pred_r, round_const),
                            AOM_BLEND_A64_ROUND_BITS);

    const __m128i pred = _mm_packs_epi32(pred_l, pred_r);
    const __m128i diff = _mm_abs_epi16(_mm_sub_epi16(pred, src));
    res = _mm_add_epi32(res, _mm_madd_epi16(diff, one));

    src_ptr += src_stride * 2;
    a_ptr += a_stride * 2;
    b_ptr += b_stride * 2;
    m_ptr += m_stride * 2;
  }
  res = _mm_hadd_epi32(res, res);
  res = _mm_hadd_epi32(res, res);
  int sad = _mm_cvtsi128_si32(res);
  return (sad + 31) >> 6;
}

static INLINE unsigned int highbd_masked_sad_ssse3(
    const uint8_t *src8, int src_stride, const uint8_t *a8, int a_stride,
    const uint8_t *b8, int b_stride, const uint8_t *m_ptr, int m_stride,
    int width, int height) {
  const uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *a_ptr = CONVERT_TO_SHORTPTR(a8);
  const uint16_t *b_ptr = CONVERT_TO_SHORTPTR(b8);
  int x, y;
  __m128i res = _mm_setzero_si128();
  const __m128i mask_max = _mm_set1_epi16((1 << AOM_BLEND_A64_ROUND_BITS));
  const __m128i round_const =
      _mm_set1_epi32((1 << AOM_BLEND_A64_ROUND_BITS) >> 1);
  const __m128i one = _mm_set1_epi16(1);

  for (y = 0; y < height; y++) {
    for (x = 0; x < width; x += 8) {
      const __m128i src = _mm_loadu_si128((const __m128i *)&src_ptr[x]);
      const __m128i a = _mm_loadu_si128((const __m128i *)&a_ptr[x]);
      const __m128i b = _mm_loadu_si128((const __m128i *)&b_ptr[x]);
      // Zero-extend mask to 16 bits
      const __m128i m = _mm_unpacklo_epi8(
          _mm_loadl_epi64((const __m128i *)&m_ptr[x]), _mm_setzero_si128());
      const __m128i m_inv = _mm_sub_epi16(mask_max, m);

      const __m128i data_l = _mm_unpacklo_epi16(a, b);
      const __m128i mask_l = _mm_unpacklo_epi16(m, m_inv);
      __m128i pred_l = _mm_madd_epi16(data_l, mask_l);
      pred_l = _mm_srai_epi32(_mm_add_epi32(pred_l, round_const),
                              AOM_BLEND_A64_ROUND_BITS);

      const __m128i data_r = _mm_unpackhi_epi16(a, b);
      const __m128i mask_r = _mm_unpackhi_epi16(m, m_inv);
      __m128i pred_r = _mm_madd_epi16(data_r, mask_r);
      pred_r = _mm_srai_epi32(_mm_add_epi32(pred_r, round_const),
                              AOM_BLEND_A64_ROUND_BITS);

      // Note: the maximum value in pred_l/r is (2^bd)-1 < 2^15,
      // so it is safe to do signed saturation here.
      const __m128i pred = _mm_packs_epi32(pred_l, pred_r);
      // There is no 16-bit SAD instruction, so we have to synthesize
      // an 8-element SAD. We do this by storing 4 32-bit partial SADs,
      // and accumulating them at the end
      const __m128i diff = _mm_abs_epi16(_mm_sub_epi16(pred, src));
      res = _mm_add_epi32(res, _mm_madd_epi16(diff, one));
    }

    src_ptr += src_stride;
    a_ptr += a_stride;
    b_ptr += b_stride;
    m_ptr += m_stride;
  }
  // At this point, we have four 32-bit partial SADs stored in 'res'.
  res = _mm_hadd_epi32(res, res);
  res = _mm_hadd_epi32(res, res);
  int sad = _mm_cvtsi128_si32(res);
  return (sad + 31) >> 6;
}

static INLINE unsigned int highbd_masked_sad8xh_avx2(
    const uint8_t *src8, int src_stride, const uint8_t *a8, int a_stride,
    const uint8_t *b8, int b_stride, const uint8_t *m_ptr, int m_stride,
    int height) {
  const uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src8);
  const uint16_t *a_ptr = CONVERT_TO_SHORTPTR(a8);
  const uint16_t *b_ptr = CONVERT_TO_SHORTPTR(b8);
  int y;
  __m256i res = _mm256_setzero_si256();
  const __m256i mask_max = _mm256_set1_epi16((1 << AOM_BLEND_A64_ROUND_BITS));
  const __m256i round_const =
      _mm256_set1_epi32((1 << AOM_BLEND_A64_ROUND_BITS) >> 1);
  const __m256i one = _mm256_set1_epi16(1);

  for (y = 0; y < height; y += 2) {
    const __m256i src = xx_loadu2_m128i(src_ptr + src_stride, src_ptr);
    const __m256i a = xx_loadu2_m128i(a_ptr + a_stride, a_ptr);
    const __m256i b = xx_loadu2_m128i(b_ptr + b_stride, b_ptr);
    // Zero-extend mask to 16 bits
    const __m256i m = _mm256_cvtepu8_epi16(_mm_unpacklo_epi64(
        _mm_loadl_epi64((const __m128i *)(m_ptr)),
        _mm_loadl_epi64((const __m128i *)(m_ptr + m_stride))));
    const __m256i m_inv = _mm256_sub_epi16(mask_max, m);

    const __m256i data_l = _mm256_unpacklo_epi16(a, b);
    const __m256i mask_l = _mm256_unpacklo_epi16(m, m_inv);
    __m256i pred_l = _mm256_madd_epi16(data_l, mask_l);
    pred_l = _mm256_srai_epi32(_mm256_add_epi32(pred_l, round_const),
                               AOM_BLEND_A64_ROUND_BITS);

    const __m256i data_r = _mm256_unpackhi_epi16(a, b);
    const __m256i mask_r = _mm256_unpackhi_epi16(m, m_inv);
    __m256i pred_r = _mm256_madd_epi16(data_r, mask_r);
    pred_r = _mm256_srai_epi32(_mm256_add_epi32(pred_r, round_const),
                               AOM_BLEND_A64_ROUND_BITS);

    // Note: the maximum value in pred_l/r is (2^bd)-1 < 2^15,
    // so it is safe to do signed saturation here.
    const __m256i pred = _mm256_packs_epi32(pred_l, pred_r);
    // There is no 16-bit SAD instruction, so we have to synthesize
    // an 8-element SAD. We do this by storing 4 32-bit partial SADs,
    // and accumulating them at the end
    const __m256i diff = _mm256_abs_epi16(_mm256_sub_epi16(pred, src));
    res = _mm256_add_epi32(res, _mm256_madd_epi16(diff, one));

    src_ptr += src_stride << 1;
    a_ptr += a_stride << 1;
    b_ptr += b_stride << 1;
    m_ptr += m_stride << 1;
  }
  // At this point, we have four 32-bit partial SADs stored in 'res'.
  res = _mm256_hadd_epi32(res, res);
  res = _mm256_hadd_epi32(res, res);
  int sad = _mm256_extract_epi32(res, 0) + _mm256_extract_epi32(res, 4);
  return (sad + 31) >> 6;
}

unsigned int vpx_highbd_10_mse16x16_sse2(const uint8_t *src8, int src_stride,
                                         const uint8_t *ref8, int ref_stride,
                                         unsigned int *sse) {
  int sum;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
  highbd_10_variance_sse2(src, src_stride, ref, ref_stride, 16, 16,
                          sse, &sum, vpx_highbd_calc16x16var_sse2, 16);
  return *sse;
}

unsigned int vpx_highbd_12_mse8x8_sse2(const uint8_t *src8, int src_stride,
                                       const uint8_t *ref8, int ref_stride,
                                       unsigned int *sse) {
  int sum;
  uint16_t *src = CONVERT_TO_SHORTPTR(src8);
  uint16_t *ref = CONVERT_TO_SHORTPTR(ref8);
  highbd_12_variance_sse2(src, src_stride, ref, ref_stride, 8, 8,
                          sse, &sum, vpx_highbd_calc8x8var_sse2, 8);
  return *sse;
}

        static INLINE
    unsigned int highbd_sad(const uint8_t *a8, int a_stride, const uint8_t *b8,
                            int b_stride, int width, int height) {
  int y, x;
  unsigned int sad = 0;
  const uint16_t *a = CONVERT_TO_SHORTPTR(a8);
  const uint16_t *b = CONVERT_TO_SHORTPTR(b8);
  for (y = 0; y < height; y++) {
    for (x = 0; x < width; x++) sad += abs(a[x] - b[x]);

    a += a_stride;
    b += b_stride;
  }
  return sad;
}

static void highbd_var_filter_block2d_bil_first_pass(
    const uint8_t *src_ptr8,
    uint16_t *output_ptr,
    unsigned int src_pixels_per_line,
    int pixel_step,
    unsigned int output_height,
    unsigned int output_width,
    const int16_t *vp9_filter) {
  unsigned int i, j;
  uint16_t *src_ptr = CONVERT_TO_SHORTPTR(src_ptr8);
  for (i = 0; i < output_height; i++) {
    for (j = 0; j < output_width; j++) {
      output_ptr[j] =
          ROUND_POWER_OF_TWO((int)src_ptr[0] * vp9_filter[0] +
                             (int)src_ptr[pixel_step] * vp9_filter[1],
                             FILTER_BITS);

      src_ptr++;
    }

    // Next row...
    src_ptr += src_pixels_per_line - output_width;
    output_ptr += output_width;
  }
}

void vp10_highbd_iht16x16_256_add_c(const tran_low_t *input, uint8_t *dest8,
                                    int stride, int tx_type, int bd) {
  int i, j;
  tran_low_t out[16 * 16];
  tran_low_t *outptr = out;
  tran_low_t temp_in[16], temp_out[16];
  const highbd_transform_2d ht = HIGH_IHT_16[tx_type];
  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);

  // Rows
  for (i = 0; i < 16; ++i) {
    ht.rows(input, outptr, bd);
    input += 16;
    outptr += 16;
  }

  // Columns
  for (i = 0; i < 16; ++i) {
    for (j = 0; j < 16; ++j) temp_in[j] = out[j * 16 + i];
    ht.cols(temp_in, temp_out, bd);
    for (j = 0; j < 16; ++j) {
      dest[j * stride + i] = highbd_clip_pixel_add(
          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 6), bd);
    }
  }
}

void vp10_highbd_iht8x8_64_add_c(const tran_low_t *input, uint8_t *dest8,
                                 int stride, int tx_type, int bd) {
  int i, j;
  tran_low_t out[8 * 8];
  tran_low_t *outptr = out;
  tran_low_t temp_in[8], temp_out[8];
  const highbd_transform_2d ht = HIGH_IHT_8[tx_type];
  uint16_t *dest = CONVERT_TO_SHORTPTR(dest8);

  // Inverse transform row vectors.
  for (i = 0; i < 8; ++i) {
    ht.rows(input, outptr, bd);
    input += 8;
    outptr += 8;
  }

  // Inverse transform column vectors.
  for (i = 0; i < 8; ++i) {
    for (j = 0; j < 8; ++j) temp_in[j] = out[j * 8 + i];
    ht.cols(temp_in, temp_out, bd);
    for (j = 0; j < 8; ++j) {
      dest[j * stride + i] = highbd_clip_pixel_add(
          dest[j * stride + i], ROUND_POWER_OF_TWO(temp_out[j], 5), bd);
    }
  }
}