bfin: Refactor duplicated assembly-related macros
[libav.git] / libavcodec / bfin / vp3_idct_bfin.S
1 /*
2 * vp3_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_vp3_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 */
30
31 #include "libavutil/bfin/asm.h"
32
33 SECTION_L1_DATA_B
34
35 .align 4;
36 coefs:
37 .short 0x5a82; // C4
38 .short 0x5a82; // C4
39 .short 0x30FC; //cos(3pi/8) C6
40 .short 0x7642; //cos(pi/8) C2
41 .short 0x18F9; //cos(7pi/16)
42 .short 0x7D8A; //cos(pi/16)
43 .short 0x471D; //cos(5pi/16)
44 .short 0x6A6E; //cos(3pi/16)
45 .short 0x18F9; //cos(7pi/16)
46 .short 0x7D8A; //cos(pi/16)
47
48 SECTION_L1_DATA_A
49
50 vtmp: .space 256
51
52 #define TMP0 FP-8
53 #define TMP1 FP-12
54 #define TMP2 FP-16
55
56
57 .text
58 DEFUN(vp3_idct,mL1,
59 (int16_t *block)):
60
61 /********************** Function Prologue *********************************/
62 link 16;
63 [--SP] = (R7:4, P5:3); // Push the registers onto the stack.
64 B0 = R0; // Pointer to Input matrix
65 RELOC(R1, P3, coefs); // Pointer to Coefficients
66 RELOC(R2, P3, vtmp); // Pointer to Temporary matrix
67 B3 = R1;
68 B2 = R2;
69 L3 = 20; // L3 is used for making the coefficient array
70 // circular.
71 // MUST BE RESTORED TO ZERO at function exit.
72 M1 = 16 (X); // All these registers are initialized for
73 M3 = 8(X); // modifying address offsets.
74
75 I0 = B0; // I0 points to Input Element (0, 0).
76 I2 = B0; // I2 points to Input Element (0, 0).
77 I2 += M3 || R0.H = W[I0];
78 // Element 0 is read into R0.H
79 I1 = I2; // I1 points to input Element (0, 6).
80 I1 += 4 || R0.L = W[I2++];
81 // I2 points to input Element (0, 4).
82 // Element 4 is read into R0.L.
83 P2 = 8 (X);
84 P3 = 32 (X);
85 P4 = -32 (X);
86 P5 = 98 (X);
87 R7 = 0x8000(Z);
88 I3 = B3; // I3 points to Coefficients
89 P0 = B2; // P0 points to array Element (0, 0) of temp
90 P1 = B2;
91 R7 = [I3++] || [TMP2]=R7; // Coefficient C4 is read into R7.H and R7.L.
92 MNOP;
93 NOP;
94
95 /*
96 * A1 = Y0 * cos(pi/4)
97 * A0 = Y0 * cos(pi/4)
98 * A1 = A1 + Y4 * cos(pi/4)
99 * A0 = A0 - Y4 * cos(pi/4)
100 * load:
101 * R1=(Y2,Y6)
102 * R7=(C2,C6)
103 * res:
104 * R3=Y0, R2=Y4
105 */
106 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+= 4 || R1.L=W[I1++];
107 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
108
109 LSETUP (.0, .1) LC0 = P2; // perform 8 1d idcts
110
111 P2 = 112 (X);
112 P1 = P1 + P2; // P1 points to element (7, 0) of temp buffer.
113 P2 = -94(X);
114
115 .0:
116 /*
117 * A1 = Y2 * cos(3pi/8)
118 * A0 = Y2 * cos(pi/8)
119 * A1 = A1 - Y6 * cos(pi/8)
120 * A0 = A0 + Y6 * cos(3pi/8)
121 * R5 = (Y1,Y7)
122 * R7 = (C1,C7)
123 * res:
124 * R1=Y2, R0=Y6
125 */
126 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0];
127 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++];
128 /*
129 * Y0 = Y0 + Y6.
130 * Y4 = Y4 + Y2.
131 * Y2 = Y4 - Y2.
132 * Y6 = Y0 - Y6.
133 * R3 is saved
134 * R6.l=Y3
135 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
136 */
137 R3=R3+R0, R0=R3-R0;
138 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
139 /*
140 * Compute the odd portion (1,3,5,7) even is done.
141 *
142 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
143 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
144 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
145 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
146 */
147 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7)
148 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--];
149 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++];
150 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5)
151 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
152 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0];
153 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7)
154 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS);
155 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
156 // R3=Y1, R2=Y7, R7=Y5, R6=Y3
157
158 /* Transpose write column. */
159 R5.H=R4+R2 (RND12); // Y0=Y0+Y7
160 R5.L=R4-R2 (RND12) || R4 = [TMP1]; // Y7=Y7-Y0
161 R2.H=R1+R7 (RND12) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0
162 R2.L=R1-R7 (RND12) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7
163 R5.H=R0-R3 (RND12) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
164 R5.L=R0+R3 (RND12) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
165 R3.H=R4-R6 (RND12) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
166 R3.L=R4+R6 (RND12) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
167
168 /* pipeline loop start, + drain Y3, Y4 */
169 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--];
170 .1: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
171
172
173
174 I0 = B2; // I0 points to Input Element (0, 0)
175 I2 = B2; // I2 points to Input Element (0, 0)
176 I2 += M3 || R0.H = W[I0];
177 // Y0 is read in R0.H
178 I1 = I2; // I1 points to input Element (0, 6)
179 I1 += 4 || R0.L = W[I2++];
180 // I2 points to input Element (0, 4)
181 // Y4 is read in R0.L
182 P2 = 8 (X);
183 I3 = B3; // I3 points to Coefficients
184 P0 = B0; // P0 points to array Element (0, 0) for writing
185 // output
186 P1 = B0;
187 R7 = [I3++]; // R7.H = C4 and R7.L = C4
188 NOP;
189
190 /*
191 * A1 = Y0 * cos(pi/4)
192 * A0 = Y0 * cos(pi/4)
193 * A1 = A1 + Y4 * cos(pi/4)
194 * A0 = A0 - Y4 * cos(pi/4)
195 * load:
196 * R1=(Y2,Y6)
197 * R7=(C2,C6)
198 * res:
199 * R3=Y0, R2=Y4
200 */
201 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+=4 || R1.L=W[I1++];
202 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++];
203
204 LSETUP (.2, .3) LC0 = P2; // peform 8 1d idcts
205 P2 = 112 (X);
206 P1 = P1 + P2;
207 P2 = -94(X);
208
209 .2:
210 /*
211 * A1 = Y2 * cos(3pi/8)
212 * A0 = Y2 * cos(pi/8)
213 * A1 = A1 - Y6 * cos(pi/8)
214 * A0 = A0 + Y6 * cos(3pi/8)
215 * R5 = (Y1,Y7)
216 * R7 = (C1,C7)
217 * res:
218 * R1=Y2, R0=Y6
219 */
220 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0];
221 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++];
222 /*
223 * Y0 = Y0 + Y6.
224 * Y4 = Y4 + Y2.
225 * Y2 = Y4 - Y2.
226 * Y6 = Y0 - Y6.
227 * R3 is saved
228 * R6.l=Y3
229 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6
230 */
231 R3=R3+R0, R0=R3-R0;
232 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--];
233 /*
234 * Compute the odd portion (1,3,5,7) even is done.
235 *
236 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3.
237 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3.
238 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3.
239 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3.
240 */
241 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7)
242 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--];
243 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++];
244 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5)
245 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS);
246 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0];
247 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7)
248 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS);
249 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1;
250 // R3=Y1, R2=Y7, R7=Y5, R6=Y3
251
252 /* Transpose write column. */
253 R5.H=R4+R2 (RND20); // Y0=Y0+Y7
254 R5.L=R4-R2 (RND20) || R4 = [TMP1]; // Y7=Y7-Y0
255 R5=R5>>>2(v);
256 R2.H=R1+R7 (RND20) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0
257 R2.L=R1-R7 (RND20) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7
258 R2=R2>>>2(v);
259 R5.H=R0-R3 (RND20) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2
260 R5.L=R0+R3 (RND20) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5
261 R5=R5>>>2(v);
262 R3.H=R4-R6 (RND20) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1
263 R3.L=R4+R6 (RND20) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6
264 R3=R3>>>2(v);
265 /* pipeline loop start, + drain Y3, Y4 */
266 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--];
267 .3: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++];
268
269 L3 = 0;
270 (R7:4,P5:3)=[SP++];
271 unlink;
272 RTS;
273 DEFUN_END(vp3_idct)