1 /* 2 * Copyright (c) 2015 The WebM project authors. All Rights Reserved. 3 * 4 * Use of this source code is governed by a BSD-style license 5 * that can be found in the LICENSE file in the root of the source 6 * tree. An additional intellectual property rights grant can be found 7 * in the file PATENTS. All contributing project authors may 8 * be found in the AUTHORS file in the root of the source tree. 9 */ 10 #ifndef VPX_DSP_X86_CONVOLVE_H_ 11 #define VPX_DSP_X86_CONVOLVE_H_ 12 13 #include <assert.h> 14 15 #include "./vpx_config.h" 16 #include "vpx/vpx_integer.h" 17 #include "vpx_ports/mem.h" 18 19 typedef void filter8_1dfunction(const uint8_t *src_ptr, ptrdiff_t src_pitch, 20 uint8_t *output_ptr, ptrdiff_t out_pitch, 21 uint32_t output_height, const int16_t *filter); 22 23 #define FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \ 24 void vpx_convolve8_##name##_##opt( \ 25 const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \ 26 ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \ 27 const int16_t *filter_y, int y_step_q4, int w, int h) { \ 28 (void)filter_x; \ 29 (void)x_step_q4; \ 30 (void)filter_y; \ 31 (void)y_step_q4; \ 32 assert(filter[3] != 128); \ 33 assert(step_q4 == 16); \ 34 if (filter[0] | filter[1] | filter[2]) { \ 35 while (w >= 16) { \ 36 vpx_filter_block1d16_##dir##8_##avg##opt(src_start, src_stride, dst, \ 37 dst_stride, h, filter); \ 38 src += 16; \ 39 dst += 16; \ 40 w -= 16; \ 41 } \ 42 if (w == 8) { \ 43 vpx_filter_block1d8_##dir##8_##avg##opt(src_start, src_stride, dst, \ 44 dst_stride, h, filter); \ 45 } else if (w == 4) { \ 46 vpx_filter_block1d4_##dir##8_##avg##opt(src_start, src_stride, dst, \ 47 dst_stride, h, filter); \ 48 } \ 49 } else { \ 50 while (w >= 16) { \ 51 vpx_filter_block1d16_##dir##2_##avg##opt(src, src_stride, dst, \ 52 dst_stride, h, filter); \ 53 src += 16; \ 54 dst += 16; \ 55 w -= 16; \ 56 } \ 57 if (w == 8) { \ 58 vpx_filter_block1d8_##dir##2_##avg##opt(src, src_stride, dst, \ 59 dst_stride, h, filter); \ 60 } else if (w == 4) { \ 61 vpx_filter_block1d4_##dir##2_##avg##opt(src, src_stride, dst, \ 62 dst_stride, h, filter); \ 63 } \ 64 } \ 65 } 66 67 #define FUN_CONV_2D(avg, opt) \ 68 void vpx_convolve8_##avg##opt( \ 69 const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst, \ 70 ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \ 71 const int16_t *filter_y, int y_step_q4, int w, int h) { \ 72 assert(filter_x[3] != 128); \ 73 assert(filter_y[3] != 128); \ 74 assert(w <= 64); \ 75 assert(h <= 64); \ 76 assert(x_step_q4 == 16); \ 77 assert(y_step_q4 == 16); \ 78 if (filter_x[0] | filter_x[1] | filter_x[2]) { \ 79 DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 71]); \ 80 vpx_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, fdata2, 64, \ 81 filter_x, x_step_q4, filter_y, y_step_q4, w, \ 82 h + 7); \ 83 vpx_convolve8_##avg##vert_##opt(fdata2 + 3 * 64, 64, dst, dst_stride, \ 84 filter_x, x_step_q4, filter_y, \ 85 y_step_q4, w, h); \ 86 } else { \ 87 DECLARE_ALIGNED(16, uint8_t, fdata2[64 * 65]); \ 88 vpx_convolve8_horiz_##opt(src, src_stride, fdata2, 64, filter_x, \ 89 x_step_q4, filter_y, y_step_q4, w, h + 1); \ 90 vpx_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, filter_x, \ 91 x_step_q4, filter_y, y_step_q4, w, h); \ 92 } \ 93 } 94 95 #if CONFIG_VP9_HIGHBITDEPTH 96 97 typedef void highbd_filter8_1dfunction(const uint16_t *src_ptr, 98 const ptrdiff_t src_pitch, 99 uint16_t *output_ptr, 100 ptrdiff_t out_pitch, 101 unsigned int output_height, 102 const int16_t *filter, int bd); 103 104 #define HIGH_FUN_CONV_1D(name, step_q4, filter, dir, src_start, avg, opt) \ 105 void vpx_highbd_convolve8_##name##_##opt( \ 106 const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \ 107 ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \ 108 const int16_t *filter_y, int y_step_q4, int w, int h, int bd) { \ 109 if (step_q4 == 16 && filter[3] != 128) { \ 110 if (filter[0] | filter[1] | filter[2]) { \ 111 while (w >= 16) { \ 112 vpx_highbd_filter_block1d16_##dir##8_##avg##opt( \ 113 src_start, src_stride, dst, dst_stride, h, filter, bd); \ 114 src += 16; \ 115 dst += 16; \ 116 w -= 16; \ 117 } \ 118 while (w >= 8) { \ 119 vpx_highbd_filter_block1d8_##dir##8_##avg##opt( \ 120 src_start, src_stride, dst, dst_stride, h, filter, bd); \ 121 src += 8; \ 122 dst += 8; \ 123 w -= 8; \ 124 } \ 125 while (w >= 4) { \ 126 vpx_highbd_filter_block1d4_##dir##8_##avg##opt( \ 127 src_start, src_stride, dst, dst_stride, h, filter, bd); \ 128 src += 4; \ 129 dst += 4; \ 130 w -= 4; \ 131 } \ 132 } else { \ 133 while (w >= 16) { \ 134 vpx_highbd_filter_block1d16_##dir##2_##avg##opt( \ 135 src, src_stride, dst, dst_stride, h, filter, bd); \ 136 src += 16; \ 137 dst += 16; \ 138 w -= 16; \ 139 } \ 140 while (w >= 8) { \ 141 vpx_highbd_filter_block1d8_##dir##2_##avg##opt( \ 142 src, src_stride, dst, dst_stride, h, filter, bd); \ 143 src += 8; \ 144 dst += 8; \ 145 w -= 8; \ 146 } \ 147 while (w >= 4) { \ 148 vpx_highbd_filter_block1d4_##dir##2_##avg##opt( \ 149 src, src_stride, dst, dst_stride, h, filter, bd); \ 150 src += 4; \ 151 dst += 4; \ 152 w -= 4; \ 153 } \ 154 } \ 155 } \ 156 if (w) { \ 157 vpx_highbd_convolve8_##name##_c(src, src_stride, dst, dst_stride, \ 158 filter_x, x_step_q4, filter_y, \ 159 y_step_q4, w, h, bd); \ 160 } \ 161 } 162 163 #define HIGH_FUN_CONV_2D(avg, opt) \ 164 void vpx_highbd_convolve8_##avg##opt( \ 165 const uint16_t *src, ptrdiff_t src_stride, uint16_t *dst, \ 166 ptrdiff_t dst_stride, const int16_t *filter_x, int x_step_q4, \ 167 const int16_t *filter_y, int y_step_q4, int w, int h, int bd) { \ 168 assert(w <= 64); \ 169 assert(h <= 64); \ 170 if (x_step_q4 == 16 && y_step_q4 == 16) { \ 171 if ((filter_x[0] | filter_x[1] | filter_x[2]) || filter_x[3] == 128) { \ 172 DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 71]); \ 173 vpx_highbd_convolve8_horiz_##opt(src - 3 * src_stride, src_stride, \ 174 fdata2, 64, filter_x, x_step_q4, \ 175 filter_y, y_step_q4, w, h + 7, bd); \ 176 vpx_highbd_convolve8_##avg##vert_##opt( \ 177 fdata2 + 192, 64, dst, dst_stride, filter_x, x_step_q4, filter_y, \ 178 y_step_q4, w, h, bd); \ 179 } else { \ 180 DECLARE_ALIGNED(16, uint16_t, fdata2[64 * 65]); \ 181 vpx_highbd_convolve8_horiz_##opt(src, src_stride, fdata2, 64, \ 182 filter_x, x_step_q4, filter_y, \ 183 y_step_q4, w, h + 1, bd); \ 184 vpx_highbd_convolve8_##avg##vert_##opt(fdata2, 64, dst, dst_stride, \ 185 filter_x, x_step_q4, filter_y, \ 186 y_step_q4, w, h, bd); \ 187 } \ 188 } else { \ 189 vpx_highbd_convolve8_##avg##c(src, src_stride, dst, dst_stride, \ 190 filter_x, x_step_q4, filter_y, y_step_q4, \ 191 w, h, bd); \ 192 } \ 193 } 194 #endif // CONFIG_VP9_HIGHBITDEPTH 195 196 #endif // VPX_DSP_X86_CONVOLVE_H_ 197