Statistics
| Branch: | Revision:

ffmpeg / libavcodec / bfin / idct_bfin.S @ 67fd620c

History | View | Annotate | Download (11.1 KB)

1
/*
2
 * idct BlackFin
3
 *
4
 * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
5
 *
6
 * This file is part of FFmpeg.
7
 *
8
 * FFmpeg is free software; you can redistribute it and/or
9
 * modify it under the terms of the GNU Lesser General Public
10
 * License as published by the Free Software Foundation; either
11
 * version 2.1 of the License, or (at your option) any later version.
12
 *
13
 * FFmpeg is distributed in the hope that it will be useful,
14
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16
 * Lesser General Public License for more details.
17
 *
18
 * You should have received a copy of the GNU Lesser General Public
19
 * License along with FFmpeg; if not, write to the Free Software
20
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21
 */
22
/*
23
   This blackfin DSP code implements an 8x8 inverse type II DCT.
24

    
25
Prototype       : void ff_bfin_idct(DCTELEM *in)
26

    
27
Registers Used  : A0, A1, R0-R7, I0-I3, B0, B2, B3, M0-M2, L0-L3, P0-P5, LC0.
28

    
29
Performance     :
30
                    Code Size   : 498 Bytes.
31
                    Cycle Count : 417 Cycles
32

    
33

    
34
-----------------------------------------------------------
35
FFMPEG conformance testing results
36
-----------------------------------------------------------
37

    
38
dct-test: modified with the following
39
            dct_error("BFINidct", 1, ff_bfin_idct, idct, test);
40
produces the following output
41

    
42
root:/u/ffmpeg/bhead/libavcodec> ./dct-test -i
43
ffmpeg DCT/IDCT test
44

    
45
    8    15    -2    21    24    17     0    10
46
    2   -10    -5    -5    -3     7   -14    -3
47
    2   -13   -10   -19    18    -6     6    -2
48
    9     4    16    -3     9    12    10    15
49
   15    -9    -2    10     1    16     0   -15
50
  -15     5     7     3    13     0    13    20
51
   -6   -15    24     9   -18     1     9   -22
52
   -8    25    23     2    -7     0    30    13
53
IDCT BFINidct: err_inf=1 err2=0.01002344 syserr=0.00150000 maxout=266 blockSumErr=64
54
IDCT BFINidct: 88.3 kdct/s
55

    
56
*/
57

    
58
#include "config_bfin.h"
59

    
60
.section .l1.data.B,"aw",@progbits
61

    
62
.align 4;
63
coefs:
64
.short 0x5a82;           //  C4
65
.short 0x5a82;           //  C4
66
.short 0x30FC;           //cos(3pi/8)  C6
67
.short 0x7642;           //cos(pi/8)   C2
68
.short 0x18F9;           //cos(7pi/16)
69
.short 0x7D8A;           //cos(pi/16)
70
.short 0x471D;           //cos(5pi/16)
71
.short 0x6A6E;           //cos(3pi/16)
72
.short 0x18F9;           //cos(7pi/16)
73
.short 0x7D8A;           //cos(pi/16)
74

    
75
.section .l1.data.A
76

    
77
vtmp: .space 256
78

    
79
#define TMP0 FP-8
80
#define TMP1 FP-12
81
#define TMP2 FP-16
82

    
83

    
84
DEFUN(idct,mL1,
85
        (DCTELEM *block)):
86

    
87
/********************** Function Prologue *********************************/
88
    link 16;
89
    [--SP] = (R7:4, P5:3);   // Push the registers onto the stack.
90
    B0 = R0;                 // Pointer to Input matrix
91
    R1 = [P3+coefs@GOT17M4]; // Pointer to Coefficients
92
    R2 = [P3+vtmp@GOT17M4];  // Pointer to Temporary matrix
93
    B3 = R1;
94
    B2 = R2;
95
    L3 = 20;                // L3 is used for making the coefficient array
96
                            // circular.
97
                            // MUST BE RESTORED TO ZERO at function exit.
98
    M1 = 16 (X);            // All these registers are initialized for
99
    M3 = 8(X);              // modifying address offsets.
100

    
101
    I0 = B0;                // I0 points to Input Element (0, 0).
102
    I2 = B0;                // I2 points to Input Element (0, 0).
103
    I2 += M3 || R0.H = W[I0];
104
                            // Element 0 is read into R0.H
105
    I1 = I2;                // I1 points to input Element (0, 6).
106
    I1 += 4  || R0.L = W[I2++];
107
                            // I2 points to input Element (0, 4).
108
                            // Element 4 is read into R0.L.
109
    P2 = 8 (X);
110
    P3 = 32 (X);
111
    P4 = -32 (X);
112
    P5 = 98 (X);
113
    R7 = 0x8000(Z);
114
    I3 = B3;                // I3 points to Coefficients
115
    P0 = B2;                // P0 points to array Element (0, 0) of temp
116
    P1 = B2;
117
    R7 = [I3++] || [TMP2]=R7;            // Coefficient C4 is read into R7.H and R7.L.
118
    MNOP;
119
    NOP;
120

    
121
    /*
122
     *   A1 =      Y0 * cos(pi/4)
123
     *   A0 =      Y0 * cos(pi/4)
124
     *   A1 = A1 + Y4 * cos(pi/4)
125
     *   A0 = A0 - Y4 * cos(pi/4)
126
     *   load:
127
     *     R1=(Y2,Y6)
128
     *     R7=(C2,C6)
129
     *   res:
130
     *     R3=Y0, R2=Y4
131
     */
132
    A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || I0+= 4       || R1.L=W[I1++];
133
    R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
134

    
135
    LSETUP (.0, .1) LC0 = P2; // perform 8 1d idcts
136

    
137
    P2 = 112 (X);
138
    P1 = P1 + P2;           // P1 points to element (7, 0) of temp buffer.
139
    P2 = -94(X);
140

    
141
.0:
142
       /*
143
        *   A1 =      Y2 * cos(3pi/8)
144
        *   A0 =      Y2 * cos(pi/8)
145
        *   A1 = A1 - Y6 * cos(pi/8)
146
        *   A0 = A0 + Y6 * cos(3pi/8)
147
        *      R5 = (Y1,Y7)
148
        *      R7 = (C1,C7)
149
        *   res:
150
        *      R1=Y2, R0=Y6
151
        */
152
        A1=R7.L*R1.H,       A0=R7.H*R1.H (IS)        || I0+=4        || R5.H=W[I0];
153
        R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS)  || R5.L=W[I1--] || R7=[I3++];
154
        /*
155
        *   Y0 = Y0 + Y6.
156
        *   Y4 = Y4 + Y2.
157
        *   Y2 = Y4 - Y2.
158
        *   Y6 = Y0 - Y6.
159
        *     R3 is saved
160
        *     R6.l=Y3
161
        * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
162
        */
163
        R3=R3+R0, R0=R3-R0;
164
        R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
165
        /*
166
         *  Compute the odd portion (1,3,5,7) even is done.
167
         *
168
         *  Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
169
         *  Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
170
         *  Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
171
         *  Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
172
         */
173
        //  R5=(Y1,Y7)  R6=(Y5,Y3)                                                   // R7=(C1,C7)
174
        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || [TMP1]=R2 || R6.H=W[I2--];
175
        A1-=R7.H*R5.L,       A0+=R7.L*R5.L (IS)       || I0-=4     || R7=[I3++];
176
        A1+=R7.H*R6.H,       A0+=R7.L*R6.H (IS)       || I0+=M1;                     // R7=(C3,C5)
177
        R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
178
        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || R4=[TMP0];
179
        A1+=R7.H*R5.L,       A0-=R7.L*R5.L (IS)       || I1+=M1    || R7=[I3++];     // R7=(C1,C7)
180
        A1+=R7.L*R6.H,       A0-=R7.H*R6.H (IS);
181
        R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
182
        // R3=Y1, R2=Y7, R7=Y5, R6=Y3
183

    
184
        /* Transpose write column. */
185
        R5.H=R4+R2 (RND12);                                   // Y0=Y0+Y7
186
        R5.L=R4-R2 (RND12) || R4 = [TMP1];                    // Y7=Y7-Y0
187
        R2.H=R1+R7 (RND12) || W[P0++P3]=R5.H;                 // Y2=Y2+Y5 st Y0
188
        R2.L=R1-R7 (RND12) || W[P1++P4]=R5.L || R7=[I3++];    // Y5=Y2-Y5 st Y7
189
        R5.H=R0-R3 (RND12) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
190
        R5.L=R0+R3 (RND12) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
191
        R3.H=R4-R6 (RND12) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
192
        R3.L=R4+R6 (RND12) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
193

    
194
        /* pipeline loop start, + drain Y3, Y4 */
195
        A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || W[P0++P2]= R3.H || R1.H = W[I0--];
196
.1:     R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
197

    
198

    
199

    
200
    I0 = B2;                // I0 points to Input Element (0, 0)
201
    I2 = B2;                // I2 points to Input Element (0, 0)
202
    I2 += M3 || R0.H = W[I0];
203
                            // Y0 is read in R0.H
204
    I1 = I2;                // I1 points to input Element (0, 6)
205
    I1 += 4  || R0.L = W[I2++];
206
                            // I2 points to input Element (0, 4)
207
                            // Y4 is read in R0.L
208
    P2 = 8 (X);
209
    I3 = B3;                // I3 points to Coefficients
210
    P0 = B0;                // P0 points to array Element (0, 0) for writing
211
                            // output
212
    P1 = B0;
213
    R7 = [I3++];            // R7.H = C4 and R7.L = C4
214
    NOP;
215

    
216
    /*
217
     *   A1 =      Y0 * cos(pi/4)
218
     *   A0 =      Y0 * cos(pi/4)
219
     *   A1 = A1 + Y4 * cos(pi/4)
220
     *   A0 = A0 - Y4 * cos(pi/4)
221
     *   load:
222
     *     R1=(Y2,Y6)
223
     *     R7=(C2,C6)
224
     *   res:
225
     *     R3=Y0, R2=Y4
226
     */
227
    A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || I0+=4        || R1.L=W[I1++];
228
    R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
229

    
230
    LSETUP (.2, .3) LC0 = P2; // peform 8 1d idcts
231
    P2 = 112 (X);
232
    P1 = P1 + P2;
233
    P2 = -94(X);
234

    
235
.2:
236
        /*
237
         *   A1 =      Y2 * cos(3pi/8)
238
         *   A0 =      Y2 * cos(pi/8)
239
         *   A1 = A1 - Y6 * cos(pi/8)
240
         *   A0 = A0 + Y6 * cos(3pi/8)
241
         *      R5 = (Y1,Y7)
242
         *      R7 = (C1,C7)
243
         *   res:
244
         *      R1=Y2, R0=Y6
245
         */
246
        A1=R7.L*R1.H,       A0=R7.H*R1.H (IS)        || I0+=4        || R5.H=W[I0];
247
        R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS)  || R5.L=W[I1--] || R7=[I3++];
248
        /*
249
        *   Y0 = Y0 + Y6.
250
        *   Y4 = Y4 + Y2.
251
        *   Y2 = Y4 - Y2.
252
        *   Y6 = Y0 - Y6.
253
        *     R3 is saved
254
        *     R6.l=Y3
255
        * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
256
        */
257
        R3=R3+R0, R0=R3-R0;
258
        R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
259
        /*
260
         *  Compute the odd portion (1,3,5,7) even is done.
261
         *
262
         *  Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
263
         *  Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
264
         *  Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
265
         *  Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
266
         */
267
        //  R5=(Y1,Y7)  R6=(Y5,Y3)                                                   // R7=(C1,C7)
268
        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || [TMP1]=R2 || R6.H=W[I2--];
269
        A1-=R7.H*R5.L,       A0+=R7.L*R5.L (IS)       || I0-=4     || R7=[I3++];
270
        A1+=R7.H*R6.H,       A0+=R7.L*R6.H (IS)       || I0+=M1;                     // R7=(C3,C5)
271
        R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
272
        A1 =R7.L*R5.H,       A0 =R7.H*R5.H (IS)       || R4=[TMP0];
273
        A1+=R7.H*R5.L,       A0-=R7.L*R5.L (IS)       || I1+=M1    || R7=[I3++];     // R7=(C1,C7)
274
        A1+=R7.L*R6.H,       A0-=R7.H*R6.H (IS);
275
        R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
276
        // R3=Y1, R2=Y7, R7=Y5, R6=Y3
277

    
278
        /* Transpose write column. */
279
        R5.H=R4+R2 (RND20);                                   // Y0=Y0+Y7
280
        R5.L=R4-R2 (RND20) || R4 = [TMP1];                    // Y7=Y7-Y0
281
        R2.H=R1+R7 (RND20) || W[P0++P3]=R5.H;                 // Y2=Y2+Y5 st Y0
282
        R2.L=R1-R7 (RND20) || W[P1++P4]=R5.L || R7=[I3++];    // Y5=Y2-Y5 st Y7
283
        R5.H=R0-R3 (RND20) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
284
        R5.L=R0+R3 (RND20) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
285
        R3.H=R4-R6 (RND20) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
286
        R3.L=R4+R6 (RND20) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
287

    
288
        /* pipeline loop start, + drain Y3, Y4 */
289
        A1=R7.H*R0.H,       A0=R7.H*R0.H (IS)       || W[P0++P2]= R3.H || R1.H = W[I0--];
290
.3:     R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
291

    
292
    L3 = 0;
293
    (R7:4,P5:3)=[SP++];
294
    unlink;
295
    RTS;
296

    
297