dsputil_iwmmxt.c
Go to the documentation of this file.
1 /*
2  * iWMMXt optimized DSP utils
3  * Copyright (c) 2004 AGAWA Koji
4  *
5  * This file is part of Libav.
6  *
7  * Libav is free software; you can redistribute it and/or
8  * modify it under the terms of the GNU Lesser General Public
9  * License as published by the Free Software Foundation; either
10  * version 2.1 of the License, or (at your option) any later version.
11  *
12  * Libav is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15  * Lesser General Public License for more details.
16  *
17  * You should have received a copy of the GNU Lesser General Public
18  * License along with Libav; if not, write to the Free Software
19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
20  */
21 
22 #include "libavutil/cpu.h"
23 #include "libavcodec/dsputil.h"
24 
25 #define DEF(x, y) x ## _no_rnd_ ## y ##_iwmmxt
26 #define SET_RND(regd) __asm__ volatile ("mov r12, #1 \n\t tbcsth " #regd ", r12":::"r12");
27 #define WAVG2B "wavg2b"
29 #undef DEF
30 #undef SET_RND
31 #undef WAVG2B
32 
33 #define DEF(x, y) x ## _ ## y ##_iwmmxt
34 #define SET_RND(regd) __asm__ volatile ("mov r12, #2 \n\t tbcsth " #regd ", r12":::"r12");
35 #define WAVG2B "wavg2br"
37 #undef DEF
38 #undef SET_RND
39 #undef WAVG2BR
40 
41 // need scheduling
42 #define OP(AVG) \
43  __asm__ volatile ( \
44  /* alignment */ \
45  "and r12, %[pixels], #7 \n\t" \
46  "bic %[pixels], %[pixels], #7 \n\t" \
47  "tmcr wcgr1, r12 \n\t" \
48  \
49  "wldrd wr0, [%[pixels]] \n\t" \
50  "wldrd wr1, [%[pixels], #8] \n\t" \
51  "add %[pixels], %[pixels], %[line_size] \n\t" \
52  "walignr1 wr4, wr0, wr1 \n\t" \
53  \
54  "1: \n\t" \
55  \
56  "wldrd wr2, [%[pixels]] \n\t" \
57  "wldrd wr3, [%[pixels], #8] \n\t" \
58  "add %[pixels], %[pixels], %[line_size] \n\t" \
59  "pld [%[pixels]] \n\t" \
60  "walignr1 wr5, wr2, wr3 \n\t" \
61  AVG " wr6, wr4, wr5 \n\t" \
62  "wstrd wr6, [%[block]] \n\t" \
63  "add %[block], %[block], %[line_size] \n\t" \
64  \
65  "wldrd wr0, [%[pixels]] \n\t" \
66  "wldrd wr1, [%[pixels], #8] \n\t" \
67  "add %[pixels], %[pixels], %[line_size] \n\t" \
68  "walignr1 wr4, wr0, wr1 \n\t" \
69  "pld [%[pixels]] \n\t" \
70  AVG " wr6, wr4, wr5 \n\t" \
71  "wstrd wr6, [%[block]] \n\t" \
72  "add %[block], %[block], %[line_size] \n\t" \
73  \
74  "subs %[h], %[h], #2 \n\t" \
75  "bne 1b \n\t" \
76  : [block]"+r"(block), [pixels]"+r"(pixels), [h]"+r"(h) \
77  : [line_size]"r"(line_size) \
78  : "memory", "r12");
79 void put_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
80 {
81  OP("wavg2br");
82 }
83 void put_no_rnd_pixels8_y2_iwmmxt(uint8_t *block, const uint8_t *pixels, const int line_size, int h)
84 {
85  OP("wavg2b");
86 }
87 #undef OP
88 
89 void add_pixels_clamped_iwmmxt(const DCTELEM *block, uint8_t *pixels, int line_size)
90 {
91  uint8_t *pixels2 = pixels + line_size;
92 
93  __asm__ volatile (
94  "mov r12, #4 \n\t"
95  "1: \n\t"
96  "pld [%[pixels], %[line_size2]] \n\t"
97  "pld [%[pixels2], %[line_size2]] \n\t"
98  "wldrd wr4, [%[pixels]] \n\t"
99  "wldrd wr5, [%[pixels2]] \n\t"
100  "pld [%[block], #32] \n\t"
101  "wunpckelub wr6, wr4 \n\t"
102  "wldrd wr0, [%[block]] \n\t"
103  "wunpckehub wr7, wr4 \n\t"
104  "wldrd wr1, [%[block], #8] \n\t"
105  "wunpckelub wr8, wr5 \n\t"
106  "wldrd wr2, [%[block], #16] \n\t"
107  "wunpckehub wr9, wr5 \n\t"
108  "wldrd wr3, [%[block], #24] \n\t"
109  "add %[block], %[block], #32 \n\t"
110  "waddhss wr10, wr0, wr6 \n\t"
111  "waddhss wr11, wr1, wr7 \n\t"
112  "waddhss wr12, wr2, wr8 \n\t"
113  "waddhss wr13, wr3, wr9 \n\t"
114  "wpackhus wr14, wr10, wr11 \n\t"
115  "wpackhus wr15, wr12, wr13 \n\t"
116  "wstrd wr14, [%[pixels]] \n\t"
117  "add %[pixels], %[pixels], %[line_size2] \n\t"
118  "subs r12, r12, #1 \n\t"
119  "wstrd wr15, [%[pixels2]] \n\t"
120  "add %[pixels2], %[pixels2], %[line_size2] \n\t"
121  "bne 1b \n\t"
122  : [block]"+r"(block), [pixels]"+r"(pixels), [pixels2]"+r"(pixels2)
123  : [line_size2]"r"(line_size << 1)
124  : "cc", "memory", "r12");
125 }
126 
127 static void clear_blocks_iwmmxt(DCTELEM *blocks)
128 {
129  __asm__ volatile(
130  "wzero wr0 \n\t"
131  "mov r1, #(128 * 6 / 32) \n\t"
132  "1: \n\t"
133  "wstrd wr0, [%0] \n\t"
134  "wstrd wr0, [%0, #8] \n\t"
135  "wstrd wr0, [%0, #16] \n\t"
136  "wstrd wr0, [%0, #24] \n\t"
137  "subs r1, r1, #1 \n\t"
138  "add %0, %0, #32 \n\t"
139  "bne 1b \n\t"
140  : "+r"(blocks)
141  :
142  : "r1"
143  );
144 }
145 
146 static void nop(uint8_t *block, const uint8_t *pixels, int line_size, int h)
147 {
148  return;
149 }
150 
151 /* A run time test is not simple. If this file is compiled in
152  * then we should install the functions
153  */
154 
156 {
157  int mm_flags = AV_CPU_FLAG_IWMMXT; /* multimedia extension flags */
158  const int high_bit_depth = avctx->bits_per_raw_sample > 8;
159 
160  if (avctx->dsp_mask) {
161  if (avctx->dsp_mask & AV_CPU_FLAG_FORCE)
162  mm_flags |= (avctx->dsp_mask & 0xffff);
163  else
164  mm_flags &= ~(avctx->dsp_mask & 0xffff);
165  }
166 
167  if (!(mm_flags & AV_CPU_FLAG_IWMMXT)) return;
168 
170 
171  if (!high_bit_depth) {
173 
174  c->put_pixels_tab[0][0] = put_pixels16_iwmmxt;
175  c->put_pixels_tab[0][1] = put_pixels16_x2_iwmmxt;
176  c->put_pixels_tab[0][2] = put_pixels16_y2_iwmmxt;
177  c->put_pixels_tab[0][3] = put_pixels16_xy2_iwmmxt;
178  c->put_no_rnd_pixels_tab[0][0] = put_pixels16_iwmmxt;
179  c->put_no_rnd_pixels_tab[0][1] = put_no_rnd_pixels16_x2_iwmmxt;
180  c->put_no_rnd_pixels_tab[0][2] = put_no_rnd_pixels16_y2_iwmmxt;
181  c->put_no_rnd_pixels_tab[0][3] = put_no_rnd_pixels16_xy2_iwmmxt;
182 
183  c->put_pixels_tab[1][0] = put_pixels8_iwmmxt;
184  c->put_pixels_tab[1][1] = put_pixels8_x2_iwmmxt;
186  c->put_pixels_tab[1][3] = put_pixels8_xy2_iwmmxt;
187  c->put_no_rnd_pixels_tab[1][0] = put_pixels8_iwmmxt;
188  c->put_no_rnd_pixels_tab[1][1] = put_no_rnd_pixels8_x2_iwmmxt;
190  c->put_no_rnd_pixels_tab[1][3] = put_no_rnd_pixels8_xy2_iwmmxt;
191 
192  c->avg_pixels_tab[0][0] = avg_pixels16_iwmmxt;
193  c->avg_pixels_tab[0][1] = avg_pixels16_x2_iwmmxt;
194  c->avg_pixels_tab[0][2] = avg_pixels16_y2_iwmmxt;
195  c->avg_pixels_tab[0][3] = avg_pixels16_xy2_iwmmxt;
196  c->avg_no_rnd_pixels_tab[0][0] = avg_pixels16_iwmmxt;
197  c->avg_no_rnd_pixels_tab[0][1] = avg_no_rnd_pixels16_x2_iwmmxt;
198  c->avg_no_rnd_pixels_tab[0][2] = avg_no_rnd_pixels16_y2_iwmmxt;
199  c->avg_no_rnd_pixels_tab[0][3] = avg_no_rnd_pixels16_xy2_iwmmxt;
200 
201  c->avg_pixels_tab[1][0] = avg_pixels8_iwmmxt;
202  c->avg_pixels_tab[1][1] = avg_pixels8_x2_iwmmxt;
203  c->avg_pixels_tab[1][2] = avg_pixels8_y2_iwmmxt;
204  c->avg_pixels_tab[1][3] = avg_pixels8_xy2_iwmmxt;
205  c->avg_no_rnd_pixels_tab[1][0] = avg_no_rnd_pixels8_iwmmxt;
206  c->avg_no_rnd_pixels_tab[1][1] = avg_no_rnd_pixels8_x2_iwmmxt;
207  c->avg_no_rnd_pixels_tab[1][2] = avg_no_rnd_pixels8_y2_iwmmxt;
208  c->avg_no_rnd_pixels_tab[1][3] = avg_no_rnd_pixels8_xy2_iwmmxt;
209  }
210 }