• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (c) 2016, Alliance for Open Media. All rights reserved
3  *
4  * This source code is subject to the terms of the BSD 2 Clause License and
5  * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
6  * was not distributed with this source code in the LICENSE file, you can
7  * obtain it at www.aomedia.org/license/software. If the Alliance for Open
8  * Media Patent License 1.0 was not distributed with this source code in the
9  * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
10  */
11 
12 #include <assert.h>
13 #include <string.h>
14 
15 #include "config/aom_config.h"
16 #include "config/aom_dsp_rtcd.h"
17 
18 #include "aom/aom_integer.h"
19 #include "aom_dsp/aom_dsp_common.h"
20 #include "aom_dsp/aom_filter.h"
21 #include "aom_ports/mem.h"
22 
horz_scalar_product(const uint8_t * a,const int16_t * b)23 static INLINE int horz_scalar_product(const uint8_t *a, const int16_t *b) {
24   int sum = 0;
25   for (int k = 0; k < SUBPEL_TAPS; ++k) sum += a[k] * b[k];
26   return sum;
27 }
28 
vert_scalar_product(const uint8_t * a,ptrdiff_t a_stride,const int16_t * b)29 static INLINE int vert_scalar_product(const uint8_t *a, ptrdiff_t a_stride,
30                                       const int16_t *b) {
31   int sum = 0;
32   for (int k = 0; k < SUBPEL_TAPS; ++k) sum += a[k * a_stride] * b[k];
33   return sum;
34 }
35 
convolve_horiz(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const InterpKernel * x_filters,int x0_q4,int x_step_q4,int w,int h)36 static void convolve_horiz(const uint8_t *src, ptrdiff_t src_stride,
37                            uint8_t *dst, ptrdiff_t dst_stride,
38                            const InterpKernel *x_filters, int x0_q4,
39                            int x_step_q4, int w, int h) {
40   src -= SUBPEL_TAPS / 2 - 1;
41   for (int y = 0; y < h; ++y) {
42     int x_q4 = x0_q4;
43     for (int x = 0; x < w; ++x) {
44       const uint8_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
45       const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
46       const int sum = horz_scalar_product(src_x, x_filter);
47       dst[x] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
48       x_q4 += x_step_q4;
49     }
50     src += src_stride;
51     dst += dst_stride;
52   }
53 }
54 
convolve_vert(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const InterpKernel * y_filters,int y0_q4,int y_step_q4,int w,int h)55 static void convolve_vert(const uint8_t *src, ptrdiff_t src_stride,
56                           uint8_t *dst, ptrdiff_t dst_stride,
57                           const InterpKernel *y_filters, int y0_q4,
58                           int y_step_q4, int w, int h) {
59   src -= src_stride * (SUBPEL_TAPS / 2 - 1);
60 
61   for (int x = 0; x < w; ++x) {
62     int y_q4 = y0_q4;
63     for (int y = 0; y < h; ++y) {
64       const unsigned char *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
65       const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
66       const int sum = vert_scalar_product(src_y, src_stride, y_filter);
67       dst[y * dst_stride] = clip_pixel(ROUND_POWER_OF_TWO(sum, FILTER_BITS));
68       y_q4 += y_step_q4;
69     }
70     ++src;
71     ++dst;
72   }
73 }
74 
get_filter_base(const int16_t * filter)75 static const InterpKernel *get_filter_base(const int16_t *filter) {
76   // NOTE: This assumes that the filter table is 256-byte aligned.
77   // TODO(agrange) Modify to make independent of table alignment.
78   return (const InterpKernel *)(((intptr_t)filter) & ~((intptr_t)0xFF));
79 }
80 
get_filter_offset(const int16_t * f,const InterpKernel * base)81 static int get_filter_offset(const int16_t *f, const InterpKernel *base) {
82   return (int)((const InterpKernel *)(intptr_t)f - base);
83 }
84 
aom_convolve8_horiz_c(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const int16_t * filter_x,int x_step_q4,const int16_t * filter_y,int y_step_q4,int w,int h)85 void aom_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
86                            uint8_t *dst, ptrdiff_t dst_stride,
87                            const int16_t *filter_x, int x_step_q4,
88                            const int16_t *filter_y, int y_step_q4, int w,
89                            int h) {
90   const InterpKernel *const filters_x = get_filter_base(filter_x);
91   const int x0_q4 = get_filter_offset(filter_x, filters_x);
92 
93   (void)filter_y;
94   (void)y_step_q4;
95 
96   convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4, x_step_q4,
97                  w, h);
98 }
99 
aom_convolve8_vert_c(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const int16_t * filter_x,int x_step_q4,const int16_t * filter_y,int y_step_q4,int w,int h)100 void aom_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
101                           uint8_t *dst, ptrdiff_t dst_stride,
102                           const int16_t *filter_x, int x_step_q4,
103                           const int16_t *filter_y, int y_step_q4, int w,
104                           int h) {
105   const InterpKernel *const filters_y = get_filter_base(filter_y);
106   const int y0_q4 = get_filter_offset(filter_y, filters_y);
107 
108   (void)filter_x;
109   (void)x_step_q4;
110 
111   convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4, y_step_q4,
112                 w, h);
113 }
114 
aom_convolve_copy_c(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const int16_t * filter_x,int filter_x_stride,const int16_t * filter_y,int filter_y_stride,int w,int h)115 void aom_convolve_copy_c(const uint8_t *src, ptrdiff_t src_stride, uint8_t *dst,
116                          ptrdiff_t dst_stride, const int16_t *filter_x,
117                          int filter_x_stride, const int16_t *filter_y,
118                          int filter_y_stride, int w, int h) {
119   int r;
120 
121   (void)filter_x;
122   (void)filter_x_stride;
123   (void)filter_y;
124   (void)filter_y_stride;
125 
126   for (r = h; r > 0; --r) {
127     memcpy(dst, src, w);
128     src += src_stride;
129     dst += dst_stride;
130   }
131 }
132 
highbd_vert_scalar_product(const uint16_t * a,ptrdiff_t a_stride,const int16_t * b)133 static INLINE int highbd_vert_scalar_product(const uint16_t *a,
134                                              ptrdiff_t a_stride,
135                                              const int16_t *b) {
136   int sum = 0;
137   for (int k = 0; k < SUBPEL_TAPS; ++k) sum += a[k * a_stride] * b[k];
138   return sum;
139 }
140 
highbd_horz_scalar_product(const uint16_t * a,const int16_t * b)141 static INLINE int highbd_horz_scalar_product(const uint16_t *a,
142                                              const int16_t *b) {
143   int sum = 0;
144   for (int k = 0; k < SUBPEL_TAPS; ++k) sum += a[k] * b[k];
145   return sum;
146 }
147 
highbd_convolve_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)148 static void highbd_convolve_horiz(const uint8_t *src8, ptrdiff_t src_stride,
149                                   uint8_t *dst8, ptrdiff_t dst_stride,
150                                   const InterpKernel *x_filters, int x0_q4,
151                                   int x_step_q4, int w, int h, int bd) {
152   uint16_t *src = CONVERT_TO_SHORTPTR(src8);
153   uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
154   src -= SUBPEL_TAPS / 2 - 1;
155   for (int y = 0; y < h; ++y) {
156     int x_q4 = x0_q4;
157     for (int x = 0; x < w; ++x) {
158       const uint16_t *const src_x = &src[x_q4 >> SUBPEL_BITS];
159       const int16_t *const x_filter = x_filters[x_q4 & SUBPEL_MASK];
160       const int sum = highbd_horz_scalar_product(src_x, x_filter);
161       dst[x] = clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
162       x_q4 += x_step_q4;
163     }
164     src += src_stride;
165     dst += dst_stride;
166   }
167 }
168 
highbd_convolve_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)169 static void highbd_convolve_vert(const uint8_t *src8, ptrdiff_t src_stride,
170                                  uint8_t *dst8, ptrdiff_t dst_stride,
171                                  const InterpKernel *y_filters, int y0_q4,
172                                  int y_step_q4, int w, int h, int bd) {
173   uint16_t *src = CONVERT_TO_SHORTPTR(src8);
174   uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
175   src -= src_stride * (SUBPEL_TAPS / 2 - 1);
176   for (int x = 0; x < w; ++x) {
177     int y_q4 = y0_q4;
178     for (int y = 0; y < h; ++y) {
179       const uint16_t *src_y = &src[(y_q4 >> SUBPEL_BITS) * src_stride];
180       const int16_t *const y_filter = y_filters[y_q4 & SUBPEL_MASK];
181       const int sum = highbd_vert_scalar_product(src_y, src_stride, y_filter);
182       dst[y * dst_stride] =
183           clip_pixel_highbd(ROUND_POWER_OF_TWO(sum, FILTER_BITS), bd);
184       y_q4 += y_step_q4;
185     }
186     ++src;
187     ++dst;
188   }
189 }
190 
aom_highbd_convolve8_horiz_c(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const int16_t * filter_x,int x_step_q4,const int16_t * filter_y,int y_step_q4,int w,int h,int bd)191 void aom_highbd_convolve8_horiz_c(const uint8_t *src, ptrdiff_t src_stride,
192                                   uint8_t *dst, ptrdiff_t dst_stride,
193                                   const int16_t *filter_x, int x_step_q4,
194                                   const int16_t *filter_y, int y_step_q4, int w,
195                                   int h, int bd) {
196   const InterpKernel *const filters_x = get_filter_base(filter_x);
197   const int x0_q4 = get_filter_offset(filter_x, filters_x);
198   (void)filter_y;
199   (void)y_step_q4;
200 
201   highbd_convolve_horiz(src, src_stride, dst, dst_stride, filters_x, x0_q4,
202                         x_step_q4, w, h, bd);
203 }
204 
aom_highbd_convolve8_vert_c(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const int16_t * filter_x,int x_step_q4,const int16_t * filter_y,int y_step_q4,int w,int h,int bd)205 void aom_highbd_convolve8_vert_c(const uint8_t *src, ptrdiff_t src_stride,
206                                  uint8_t *dst, ptrdiff_t dst_stride,
207                                  const int16_t *filter_x, int x_step_q4,
208                                  const int16_t *filter_y, int y_step_q4, int w,
209                                  int h, int bd) {
210   const InterpKernel *const filters_y = get_filter_base(filter_y);
211   const int y0_q4 = get_filter_offset(filter_y, filters_y);
212   (void)filter_x;
213   (void)x_step_q4;
214 
215   highbd_convolve_vert(src, src_stride, dst, dst_stride, filters_y, y0_q4,
216                        y_step_q4, w, h, bd);
217 }
218 
aom_highbd_convolve_copy_c(const uint8_t * src8,ptrdiff_t src_stride,uint8_t * dst8,ptrdiff_t dst_stride,const int16_t * filter_x,int filter_x_stride,const int16_t * filter_y,int filter_y_stride,int w,int h,int bd)219 void aom_highbd_convolve_copy_c(const uint8_t *src8, ptrdiff_t src_stride,
220                                 uint8_t *dst8, ptrdiff_t dst_stride,
221                                 const int16_t *filter_x, int filter_x_stride,
222                                 const int16_t *filter_y, int filter_y_stride,
223                                 int w, int h, int bd) {
224   int r;
225   uint16_t *src = CONVERT_TO_SHORTPTR(src8);
226   uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
227   (void)filter_x;
228   (void)filter_y;
229   (void)filter_x_stride;
230   (void)filter_y_stride;
231   (void)bd;
232 
233   for (r = h; r > 0; --r) {
234     memcpy(dst, src, w * sizeof(uint16_t));
235     src += src_stride;
236     dst += dst_stride;
237   }
238 }
239