bfin: Refactor duplicated assembly-related macros
[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 "libavutil/bfin/asm.h"
59
60 SECTION_L1_DATA_B
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 .text
85 DEFUN(idct,mL1,
86 (int16_t *block)):
87
88 /********************** Function Prologue *********************************/
89 link 16;
90 [--SP] = (R7:4, P5:3); // Push the registers onto the stack.
91 B0 = R0; // Pointer to Input matrix
92 RELOC(R1, P3, coefs); // Pointer to Coefficients
93 RELOC(R2, P3, vtmp); // Pointer to Temporary matrix
94 B3 = R1;
95 B2 = R2;
96 L3 = 20; // L3 is used for making the coefficient array
97 // circular.
98 // MUST BE RESTORED TO ZERO at function exit.
99 M1 = 16 (X); // All these registers are initialized for
100 M3 = 8(X); // modifying address offsets.
101
102 I0 = B0; // I0 points to Input Element (0, 0).
103 I2 = B0; // I2 points to Input Element (0, 0).
104 I2 += M3 || R0.H = W[I0];
105 // Element 0 is read into R0.H
106 I1 = I2; // I1 points to input Element (0, 6).
107 I1 += 4 || R0.L = W[I2++];
108 // I2 points to input Element (0, 4).
109 // Element 4 is read into R0.L.
110 P2 = 8 (X);
111 P3 = 32 (X);
112 P4 = -32 (X);
113 P5 = 98 (X);
114 R7 = 0x8000(Z);
115 I3 = B3; // I3 points to Coefficients
116 P0 = B2; // P0 points to array Element (0, 0) of temp
117 P1 = B2;
118 R7 = [I3++] || [TMP2]=R7; // Coefficient C4 is read into R7.H and R7.L.
119 MNOP;
120 NOP;
121
122 /*
123 * A1 = Y0 * cos(pi/4)
124 * A0 = Y0 * cos(pi/4)
125 * A1 = A1 + Y4 * cos(pi/4)
126 * A0 = A0 - Y4 * cos(pi/4)
127 * load:
128 * R1=(Y2,Y6)
129 * R7=(C2,C6)
130 * res:
131 * R3=Y0, R2=Y4
132 */
133 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+= 4 || R1.L=W[I1++];
134 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
135
136 LSETUP (.0, .1) LC0 = P2; // perform 8 1d idcts
137
138 P2 = 112 (X);
139 P1 = P1 + P2; // P1 points to element (7, 0) of temp buffer.
140 P2 = -94(X);
141
142 .0:
143 /*
144 * A1 = Y2 * cos(3pi/8)
145 * A0 = Y2 * cos(pi/8)
146 * A1 = A1 - Y6 * cos(pi/8)
147 * A0 = A0 + Y6 * cos(3pi/8)
148 * R5 = (Y1,Y7)
149 * R7 = (C1,C7)
150 * res:
151 * R1=Y2, R0=Y6
152 */
153 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0];
154 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++];
155 /*
156 * Y0 = Y0 + Y6.
157 * Y4 = Y4 + Y2.
158 * Y2 = Y4 - Y2.
159 * Y6 = Y0 - Y6.
160 * R3 is saved
161 * R6.l=Y3
162 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
163 */
164 R3=R3+R0, R0=R3-R0;
165 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
166 /*
167 * Compute the odd portion (1,3,5,7) even is done.
168 *
169 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
170 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
171 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
172 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
173 */
174 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7)
175 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--];
176 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++];
177 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5)
178 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
179 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0];
180 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7)
181 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS);
182 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
183 // R3=Y1, R2=Y7, R7=Y5, R6=Y3
184
185 /* Transpose write column. */
186 R5.H=R4+R2 (RND12); // Y0=Y0+Y7
187 R5.L=R4-R2 (RND12) || R4 = [TMP1]; // Y7=Y7-Y0
188 R2.H=R1+R7 (RND12) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0
189 R2.L=R1-R7 (RND12) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7
190 R5.H=R0-R3 (RND12) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
191 R5.L=R0+R3 (RND12) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
192 R3.H=R4-R6 (RND12) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
193 R3.L=R4+R6 (RND12) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
194
195 /* pipeline loop start, + drain Y3, Y4 */
196 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--];
197 .1: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
198
199
200
201 I0 = B2; // I0 points to Input Element (0, 0)
202 I2 = B2; // I2 points to Input Element (0, 0)
203 I2 += M3 || R0.H = W[I0];
204 // Y0 is read in R0.H
205 I1 = I2; // I1 points to input Element (0, 6)
206 I1 += 4 || R0.L = W[I2++];
207 // I2 points to input Element (0, 4)
208 // Y4 is read in R0.L
209 P2 = 8 (X);
210 I3 = B3; // I3 points to Coefficients
211 P0 = B0; // P0 points to array Element (0, 0) for writing
212 // output
213 P1 = B0;
214 R7 = [I3++]; // R7.H = C4 and R7.L = C4
215 NOP;
216
217 /*
218 * A1 = Y0 * cos(pi/4)
219 * A0 = Y0 * cos(pi/4)
220 * A1 = A1 + Y4 * cos(pi/4)
221 * A0 = A0 - Y4 * cos(pi/4)
222 * load:
223 * R1=(Y2,Y6)
224 * R7=(C2,C6)
225 * res:
226 * R3=Y0, R2=Y4
227 */
228 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+=4 || R1.L=W[I1++];
229 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
230
231 LSETUP (.2, .3) LC0 = P2; // peform 8 1d idcts
232 P2 = 112 (X);
233 P1 = P1 + P2;
234 P2 = -94(X);
235
236 .2:
237 /*
238 * A1 = Y2 * cos(3pi/8)
239 * A0 = Y2 * cos(pi/8)
240 * A1 = A1 - Y6 * cos(pi/8)
241 * A0 = A0 + Y6 * cos(3pi/8)
242 * R5 = (Y1,Y7)
243 * R7 = (C1,C7)
244 * res:
245 * R1=Y2, R0=Y6
246 */
247 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0];
248 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++];
249 /*
250 * Y0 = Y0 + Y6.
251 * Y4 = Y4 + Y2.
252 * Y2 = Y4 - Y2.
253 * Y6 = Y0 - Y6.
254 * R3 is saved
255 * R6.l=Y3
256 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
257 */
258 R3=R3+R0, R0=R3-R0;
259 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
260 /*
261 * Compute the odd portion (1,3,5,7) even is done.
262 *
263 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
264 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
265 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
266 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
267 */
268 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7)
269 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--];
270 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++];
271 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5)
272 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
273 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0];
274 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7)
275 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS);
276 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
277 // R3=Y1, R2=Y7, R7=Y5, R6=Y3
278
279 /* Transpose write column. */
280 R5.H=R4+R2 (RND20); // Y0=Y0+Y7
281 R5.L=R4-R2 (RND20) || R4 = [TMP1]; // Y7=Y7-Y0
282 R2.H=R1+R7 (RND20) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0
283 R2.L=R1-R7 (RND20) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7
284 R5.H=R0-R3 (RND20) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
285 R5.L=R0+R3 (RND20) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
286 R3.H=R4-R6 (RND20) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
287 R3.L=R4+R6 (RND20) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
288
289 /* pipeline loop start, + drain Y3, Y4 */
290 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--];
291 .3: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
292
293 L3 = 0;
294 (R7:4,P5:3)=[SP++];
295 unlink;
296 RTS;
297 DEFUN_END(idct)