yading@10: /*
yading@10:  * Copyright (C) 2013 Wei Gao <weigao@multicorewareinc.com>
yading@10:  *
yading@10:  *
yading@10:  * This file is part of FFmpeg.
yading@10:  *
yading@10:  * FFmpeg is free software; you can redistribute it and/or
yading@10:  * modify it under the terms of the GNU Lesser General Public
yading@10:  * License as published by the Free Software Foundation; either
yading@10:  * version 2.1 of the License, or (at your option) any later version.
yading@10:  *
yading@10:  * FFmpeg is distributed in the hope that it will be useful,
yading@10:  * but WITHOUT ANY WARRANTY; without even the implied warranty of
yading@10:  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
yading@10:  * Lesser General Public License for more details.
yading@10:  *
yading@10:  * You should have received a copy of the GNU Lesser General Public
yading@10:  * License along with FFmpeg; if not, write to the Free Software
yading@10:  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
yading@10:  */
yading@10: 
yading@10: #ifndef AVFILTER_DESHAKE_KERNEL_H
yading@10: #define AVFILTER_DESHAKE_KERNEL_H
yading@10: 
yading@10: #include "libavutil/opencl.h"
yading@10: 
yading@10: const char *ff_kernel_deshake_opencl = AV_OPENCL_KERNEL(
yading@10: 
yading@10: inline unsigned char pixel(global const unsigned char *src, float x, float y,
yading@10:                            int w, int h,int stride, unsigned char def)
yading@10: {
yading@10:     return (x < 0 || y < 0 || x >= w || y >= h) ? def : src[(int)x + (int)y * stride];
yading@10: }
yading@10: unsigned char interpolate_nearest(float x, float y, global const unsigned char *src,
yading@10:                                   int width, int height, int stride, unsigned char def)
yading@10: {
yading@10:     return pixel(src, (int)(x + 0.5), (int)(y + 0.5), width, height, stride, def);
yading@10: }
yading@10: 
yading@10: unsigned char interpolate_bilinear(float x, float y, global const unsigned char *src,
yading@10:                                    int width, int height, int stride, unsigned char def)
yading@10: {
yading@10:     int x_c, x_f, y_c, y_f;
yading@10:     int v1, v2, v3, v4;
yading@10: 
yading@10:     if (x < -1 || x > width || y < -1 || y > height) {
yading@10:         return def;
yading@10:     } else {
yading@10:         x_f = (int)x;
yading@10:         x_c = x_f + 1;
yading@10: 
yading@10:         y_f = (int)y;
yading@10:         y_c = y_f + 1;
yading@10: 
yading@10:         v1 = pixel(src, x_c, y_c, width, height, stride, def);
yading@10:         v2 = pixel(src, x_c, y_f, width, height, stride, def);
yading@10:         v3 = pixel(src, x_f, y_c, width, height, stride, def);
yading@10:         v4 = pixel(src, x_f, y_f, width, height, stride, def);
yading@10: 
yading@10:         return (v1*(x - x_f)*(y - y_f) + v2*((x - x_f)*(y_c - y)) +
yading@10:                 v3*(x_c - x)*(y - y_f) + v4*((x_c - x)*(y_c - y)));
yading@10:     }
yading@10: }
yading@10: 
yading@10: unsigned char interpolate_biquadratic(float x, float y, global const unsigned char *src,
yading@10:                                       int width, int height, int stride, unsigned char def)
yading@10: {
yading@10:     int     x_c, x_f, y_c, y_f;
yading@10:     unsigned char v1,  v2,  v3,  v4;
yading@10:     float   f1,  f2,  f3,  f4;
yading@10: 
yading@10:     if (x < - 1 || x > width || y < -1 || y > height)
yading@10:         return def;
yading@10:     else {
yading@10:         x_f = (int)x;
yading@10:         x_c = x_f + 1;
yading@10:         y_f = (int)y;
yading@10:         y_c = y_f + 1;
yading@10: 
yading@10:         v1 = pixel(src, x_c, y_c, width, height, stride, def);
yading@10:         v2 = pixel(src, x_c, y_f, width, height, stride, def);
yading@10:         v3 = pixel(src, x_f, y_c, width, height, stride, def);
yading@10:         v4 = pixel(src, x_f, y_f, width, height, stride, def);
yading@10: 
yading@10:         f1 = 1 - sqrt((x_c - x) * (y_c - y));
yading@10:         f2 = 1 - sqrt((x_c - x) * (y - y_f));
yading@10:         f3 = 1 - sqrt((x - x_f) * (y_c - y));
yading@10:         f4 = 1 - sqrt((x - x_f) * (y - y_f));
yading@10:         return (v1 * f1 + v2 * f2 + v3 * f3 + v4 * f4) / (f1 + f2 + f3 + f4);
yading@10:     }
yading@10: }
yading@10: 
yading@10: inline const float clipf(float a, float amin, float amax)
yading@10: {
yading@10:     if      (a < amin) return amin;
yading@10:     else if (a > amax) return amax;
yading@10:     else               return a;
yading@10: }
yading@10: 
yading@10: inline int mirror(int v, int m)
yading@10: {
yading@10:     while ((unsigned)v > (unsigned)m) {
yading@10:         v = -v;
yading@10:         if (v < 0)
yading@10:             v += 2 * m;
yading@10:     }
yading@10:     return v;
yading@10: }
yading@10: 
yading@10: kernel void avfilter_transform(global  unsigned char *src,
yading@10:                                global  unsigned char *dst,
yading@10:                                global          float *matrix,
yading@10:                                global          float *matrix2,
yading@10:                                                  int interpolate,
yading@10:                                                  int fillmethod,
yading@10:                                                  int src_stride_lu,
yading@10:                                                  int dst_stride_lu,
yading@10:                                                  int src_stride_ch,
yading@10:                                                  int dst_stride_ch,
yading@10:                                                  int height,
yading@10:                                                  int width,
yading@10:                                                  int ch,
yading@10:                                                  int cw)
yading@10: {
yading@10:      int global_id = get_global_id(0);
yading@10: 
yading@10:      global unsigned char *dst_y = dst;
yading@10:      global unsigned char *dst_u = dst_y + height * dst_stride_lu;
yading@10:      global unsigned char *dst_v = dst_u + ch * dst_stride_ch;
yading@10: 
yading@10:      global unsigned char *src_y = src;
yading@10:      global unsigned char *src_u = src_y + height * src_stride_lu;
yading@10:      global unsigned char *src_v = src_u + ch * src_stride_ch;
yading@10: 
yading@10:      global unsigned char *tempdst;
yading@10:      global unsigned char *tempsrc;
yading@10: 
yading@10:      int x;
yading@10:      int y;
yading@10:      float x_s;
yading@10:      float y_s;
yading@10:      int tempsrc_stride;
yading@10:      int tempdst_stride;
yading@10:      int temp_height;
yading@10:      int temp_width;
yading@10:      int curpos;
yading@10:      unsigned char def = 0;
yading@10:      if (global_id < width*height) {
yading@10:         y = global_id/width;
yading@10:         x = global_id%width;
yading@10:         x_s = x * matrix[0] + y * matrix[1] + matrix[2];
yading@10:         y_s = x * matrix[3] + y * matrix[4] + matrix[5];
yading@10:         tempdst = dst_y;
yading@10:         tempsrc = src_y;
yading@10:         tempsrc_stride = src_stride_lu;
yading@10:         tempdst_stride = dst_stride_lu;
yading@10:         temp_height = height;
yading@10:         temp_width = width;
yading@10:      } else if ((global_id >= width*height)&&(global_id < width*height + ch*cw)) {
yading@10:         y = (global_id - width*height)/cw;
yading@10:         x = (global_id - width*height)%cw;
yading@10:         x_s = x * matrix2[0] + y * matrix2[1] + matrix2[2];
yading@10:         y_s = x * matrix2[3] + y * matrix2[4] + matrix2[5];
yading@10:         tempdst = dst_u;
yading@10:         tempsrc = src_u;
yading@10:         tempsrc_stride = src_stride_ch;
yading@10:         tempdst_stride = dst_stride_ch;
yading@10:         temp_height = ch;
yading@10:         temp_width = cw;
yading@10:      } else {
yading@10:         y = (global_id - width*height - ch*cw)/cw;
yading@10:         x = (global_id - width*height - ch*cw)%cw;
yading@10:         x_s = x * matrix2[0] + y * matrix2[1] + matrix2[2];
yading@10:         y_s = x * matrix2[3] + y * matrix2[4] + matrix2[5];
yading@10:         tempdst = dst_v;
yading@10:         tempsrc = src_v;
yading@10:         tempsrc_stride = src_stride_ch;
yading@10:         tempdst_stride = dst_stride_ch;
yading@10:         temp_height = ch;
yading@10:         temp_width = cw;
yading@10:      }
yading@10:      curpos = y * tempdst_stride + x;
yading@10:      switch (fillmethod) {
yading@10:         case 0: //FILL_BLANK
yading@10:             def = 0;
yading@10:             break;
yading@10:         case 1: //FILL_ORIGINAL
yading@10:             def = tempsrc[y*tempsrc_stride+x];
yading@10:             break;
yading@10:         case 2: //FILL_CLAMP
yading@10:             y_s = clipf(y_s, 0, temp_height - 1);
yading@10:             x_s = clipf(x_s, 0, temp_width - 1);
yading@10:             def = tempsrc[(int)y_s * tempsrc_stride + (int)x_s];
yading@10:             break;
yading@10:         case 3: //FILL_MIRROR
yading@10:             y_s = mirror(y_s,temp_height - 1);
yading@10:             x_s = mirror(x_s,temp_width - 1);
yading@10:             def = tempsrc[(int)y_s * tempsrc_stride + (int)x_s];
yading@10:             break;
yading@10:     }
yading@10:     switch (interpolate) {
yading@10:         case 0: //INTERPOLATE_NEAREST
yading@10:             tempdst[curpos] = interpolate_nearest(x_s, y_s, tempsrc, temp_width, temp_height, tempsrc_stride, def);
yading@10:             break;
yading@10:         case 1: //INTERPOLATE_BILINEAR
yading@10:             tempdst[curpos] = interpolate_bilinear(x_s, y_s, tempsrc, temp_width, temp_height, tempsrc_stride, def);
yading@10:             break;
yading@10:         case 2: //INTERPOLATE_BIQUADRATIC
yading@10:             tempdst[curpos] = interpolate_biquadratic(x_s, y_s, tempsrc, temp_width, temp_height, tempsrc_stride, def);
yading@10:             break;
yading@10:         default:
yading@10:             return;
yading@10:     }
yading@10: }
yading@10: );
yading@10: 
yading@10: #endif /* AVFILTER_DESHAKE_KERNEL_H */