0eeffc201628f5bc978db39fd0e73d024eb64b7a
[libav.git] / libavcodec / bfin / idct_bfin.S
1 /*
2 * idct BlackFin
3 *
4 * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com>
5 *
6 * This file is part of Libav.
7 *
8 * Libav 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 * Libav 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 Libav; 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(int16_t *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 Libav 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 libavcodec> ./dct-test -i
43 Libav 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.h"
59 #include "config_bfin.h"
60
61 #if defined(__FDPIC__) && CONFIG_SRAM
62 .section .l1.data.B,"aw",@progbits
63 #else
64 .data
65 #endif
66
67 .align 4;
68 coefs:
69 .short 0x5a82; // C4
70 .short 0x5a82; // C4
71 .short 0x30FC; //cos(3pi/8) C6
72 .short 0x7642; //cos(pi/8) C2
73 .short 0x18F9; //cos(7pi/16)
74 .short 0x7D8A; //cos(pi/16)
75 .short 0x471D; //cos(5pi/16)
76 .short 0x6A6E; //cos(3pi/16)
77 .short 0x18F9; //cos(7pi/16)
78 .short 0x7D8A; //cos(pi/16)
79
80 #if defined(__FDPIC__) && CONFIG_SRAM
81 .section .l1.data.A,"aw",@progbits
82 #endif
83
84 vtmp: .space 256
85
86 #define TMP0 FP-8
87 #define TMP1 FP-12
88 #define TMP2 FP-16
89
90
91 .text
92 DEFUN(idct,mL1,
93 (int16_t *block)):
94
95 /********************** Function Prologue *********************************/
96 link 16;
97 [--SP] = (R7:4, P5:3); // Push the registers onto the stack.
98 B0 = R0; // Pointer to Input matrix
99 RELOC(R1, P3, coefs); // Pointer to Coefficients
100 RELOC(R2, P3, vtmp); // Pointer to Temporary matrix
101 B3 = R1;
102 B2 = R2;
103 L3 = 20; // L3 is used for making the coefficient array
104 // circular.
105 // MUST BE RESTORED TO ZERO at function exit.
106 M1 = 16 (X); // All these registers are initialized for
107 M3 = 8(X); // modifying address offsets.
108
109 I0 = B0; // I0 points to Input Element (0, 0).
110 I2 = B0; // I2 points to Input Element (0, 0).
111 I2 += M3 || R0.H = W[I0];
112 // Element 0 is read into R0.H
113 I1 = I2; // I1 points to input Element (0, 6).
114 I1 += 4 || R0.L = W[I2++];
115 // I2 points to input Element (0, 4).
116 // Element 4 is read into R0.L.
117 P2 = 8 (X);
118 P3 = 32 (X);
119 P4 = -32 (X);
120 P5 = 98 (X);
121 R7 = 0x8000(Z);
122 I3 = B3; // I3 points to Coefficients
123 P0 = B2; // P0 points to array Element (0, 0) of temp
124 P1 = B2;
125 R7 = [I3++] || [TMP2]=R7; // Coefficient C4 is read into R7.H and R7.L.
126 MNOP;
127 NOP;
128
129 /*
130 * A1 = Y0 * cos(pi/4)
131 * A0 = Y0 * cos(pi/4)
132 * A1 = A1 + Y4 * cos(pi/4)
133 * A0 = A0 - Y4 * cos(pi/4)
134 * load:
135 * R1=(Y2,Y6)
136 * R7=(C2,C6)
137 * res:
138 * R3=Y0, R2=Y4
139 */
140 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+= 4 || R1.L=W[I1++];
141 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
142
143 LSETUP (.0, .1) LC0 = P2; // perform 8 1d idcts
144
145 P2 = 112 (X);
146 P1 = P1 + P2; // P1 points to element (7, 0) of temp buffer.
147 P2 = -94(X);
148
149 .0:
150 /*
151 * A1 = Y2 * cos(3pi/8)
152 * A0 = Y2 * cos(pi/8)
153 * A1 = A1 - Y6 * cos(pi/8)
154 * A0 = A0 + Y6 * cos(3pi/8)
155 * R5 = (Y1,Y7)
156 * R7 = (C1,C7)
157 * res:
158 * R1=Y2, R0=Y6
159 */
160 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0];
161 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++];
162 /*
163 * Y0 = Y0 + Y6.
164 * Y4 = Y4 + Y2.
165 * Y2 = Y4 - Y2.
166 * Y6 = Y0 - Y6.
167 * R3 is saved
168 * R6.l=Y3
169 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
170 */
171 R3=R3+R0, R0=R3-R0;
172 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
173 /*
174 * Compute the odd portion (1,3,5,7) even is done.
175 *
176 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
177 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
178 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
179 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
180 */
181 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7)
182 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--];
183 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++];
184 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5)
185 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
186 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0];
187 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7)
188 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS);
189 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
190 // R3=Y1, R2=Y7, R7=Y5, R6=Y3
191
192 /* Transpose write column. */
193 R5.H=R4+R2 (RND12); // Y0=Y0+Y7
194 R5.L=R4-R2 (RND12) || R4 = [TMP1]; // Y7=Y7-Y0
195 R2.H=R1+R7 (RND12) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0
196 R2.L=R1-R7 (RND12) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7
197 R5.H=R0-R3 (RND12) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
198 R5.L=R0+R3 (RND12) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
199 R3.H=R4-R6 (RND12) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
200 R3.L=R4+R6 (RND12) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
201
202 /* pipeline loop start, + drain Y3, Y4 */
203 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--];
204 .1: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
205
206
207
208 I0 = B2; // I0 points to Input Element (0, 0)
209 I2 = B2; // I2 points to Input Element (0, 0)
210 I2 += M3 || R0.H = W[I0];
211 // Y0 is read in R0.H
212 I1 = I2; // I1 points to input Element (0, 6)
213 I1 += 4 || R0.L = W[I2++];
214 // I2 points to input Element (0, 4)
215 // Y4 is read in R0.L
216 P2 = 8 (X);
217 I3 = B3; // I3 points to Coefficients
218 P0 = B0; // P0 points to array Element (0, 0) for writing
219 // output
220 P1 = B0;
221 R7 = [I3++]; // R7.H = C4 and R7.L = C4
222 NOP;
223
224 /*
225 * A1 = Y0 * cos(pi/4)
226 * A0 = Y0 * cos(pi/4)
227 * A1 = A1 + Y4 * cos(pi/4)
228 * A0 = A0 - Y4 * cos(pi/4)
229 * load:
230 * R1=(Y2,Y6)
231 * R7=(C2,C6)
232 * res:
233 * R3=Y0, R2=Y4
234 */
235 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+=4 || R1.L=W[I1++];
236 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
237
238 LSETUP (.2, .3) LC0 = P2; // peform 8 1d idcts
239 P2 = 112 (X);
240 P1 = P1 + P2;
241 P2 = -94(X);
242
243 .2:
244 /*
245 * A1 = Y2 * cos(3pi/8)
246 * A0 = Y2 * cos(pi/8)
247 * A1 = A1 - Y6 * cos(pi/8)
248 * A0 = A0 + Y6 * cos(3pi/8)
249 * R5 = (Y1,Y7)
250 * R7 = (C1,C7)
251 * res:
252 * R1=Y2, R0=Y6
253 */
254 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0];
255 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++];
256 /*
257 * Y0 = Y0 + Y6.
258 * Y4 = Y4 + Y2.
259 * Y2 = Y4 - Y2.
260 * Y6 = Y0 - Y6.
261 * R3 is saved
262 * R6.l=Y3
263 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
264 */
265 R3=R3+R0, R0=R3-R0;
266 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
267 /*
268 * Compute the odd portion (1,3,5,7) even is done.
269 *
270 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
271 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
272 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
273 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
274 */
275 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7)
276 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--];
277 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++];
278 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5)
279 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
280 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0];
281 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7)
282 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS);
283 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
284 // R3=Y1, R2=Y7, R7=Y5, R6=Y3
285
286 /* Transpose write column. */
287 R5.H=R4+R2 (RND20); // Y0=Y0+Y7
288 R5.L=R4-R2 (RND20) || R4 = [TMP1]; // Y7=Y7-Y0
289 R2.H=R1+R7 (RND20) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0
290 R2.L=R1-R7 (RND20) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7
291 R5.H=R0-R3 (RND20) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
292 R5.L=R0+R3 (RND20) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
293 R3.H=R4-R6 (RND20) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
294 R3.L=R4+R6 (RND20) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
295
296 /* pipeline loop start, + drain Y3, Y4 */
297 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--];
298 .3: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
299
300 L3 = 0;
301 (R7:4,P5:3)=[SP++];
302 unlink;
303 RTS;
304 DEFUN_END(idct)