Mercurial > libavcodec.hg
annotate bfin/vp3_idct_bfin.S @ 12512:58bd8cb86f5d libavcodec
Remove unused variable.
author | rbultje |
---|---|
date | Fri, 24 Sep 2010 15:31:46 +0000 |
parents | 8327c5b4df9b |
children |
rev | line source |
---|---|
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
1 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
2 * vp3_idct BlackFin |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
3 * |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
4 * Copyright (C) 2007 Marc Hoffman <marc.hoffman@analog.com> |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
5 * |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
6 * This file is part of FFmpeg. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
7 * |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
8 * FFmpeg is free software; you can redistribute it and/or |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
9 * modify it under the terms of the GNU Lesser General Public |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
10 * License as published by the Free Software Foundation; either |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
11 * version 2.1 of the License, or (at your option) any later version. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
12 * |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
13 * FFmpeg is distributed in the hope that it will be useful, |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
16 * Lesser General Public License for more details. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
17 * |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
18 * You should have received a copy of the GNU Lesser General Public |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
19 * License along with FFmpeg; if not, write to the Free Software |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
21 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
22 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
23 This blackfin DSP code implements an 8x8 inverse type II DCT. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
24 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
25 Prototype : void ff_bfin_vp3_idct(DCTELEM *in) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
26 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
27 Registers Used : A0, A1, R0-R7, I0-I3, B0, B2, B3, M0-M2, L0-L3, P0-P5, LC0. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
28 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
29 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
30 |
11063 | 31 #include "config.h" |
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
32 #include "config_bfin.h" |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
33 |
11063 | 34 #if defined(__FDPIC__) && CONFIG_SRAM |
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
35 .section .l1.data.B,"aw",@progbits |
6362
78aa57eba353
FLAT objects cannot have multiple sections, so using the L1 attributes breaks
diego
parents:
5776
diff
changeset
|
36 #else |
78aa57eba353
FLAT objects cannot have multiple sections, so using the L1 attributes breaks
diego
parents:
5776
diff
changeset
|
37 .data |
78aa57eba353
FLAT objects cannot have multiple sections, so using the L1 attributes breaks
diego
parents:
5776
diff
changeset
|
38 #endif |
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
39 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
40 .align 4; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
41 coefs: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
42 .short 0x5a82; // C4 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
43 .short 0x5a82; // C4 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
44 .short 0x30FC; //cos(3pi/8) C6 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
45 .short 0x7642; //cos(pi/8) C2 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
46 .short 0x18F9; //cos(7pi/16) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
47 .short 0x7D8A; //cos(pi/16) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
48 .short 0x471D; //cos(5pi/16) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
49 .short 0x6A6E; //cos(3pi/16) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
50 .short 0x18F9; //cos(7pi/16) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
51 .short 0x7D8A; //cos(pi/16) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
52 |
11063 | 53 #if defined(__FDPIC__) && CONFIG_SRAM |
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
54 .section .l1.data.A |
6362
78aa57eba353
FLAT objects cannot have multiple sections, so using the L1 attributes breaks
diego
parents:
5776
diff
changeset
|
55 #endif |
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
56 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
57 vtmp: .space 256 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
58 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
59 #define TMP0 FP-8 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
60 #define TMP1 FP-12 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
61 #define TMP2 FP-16 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
62 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
63 |
6362
78aa57eba353
FLAT objects cannot have multiple sections, so using the L1 attributes breaks
diego
parents:
5776
diff
changeset
|
64 .text |
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
65 DEFUN(vp3_idct,mL1, |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
66 (DCTELEM *block)): |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
67 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
68 /********************** Function Prologue *********************************/ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
69 link 16; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
70 [--SP] = (R7:4, P5:3); // Push the registers onto the stack. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
71 B0 = R0; // Pointer to Input matrix |
6362
78aa57eba353
FLAT objects cannot have multiple sections, so using the L1 attributes breaks
diego
parents:
5776
diff
changeset
|
72 RELOC(R1, P3, coefs); // Pointer to Coefficients |
78aa57eba353
FLAT objects cannot have multiple sections, so using the L1 attributes breaks
diego
parents:
5776
diff
changeset
|
73 RELOC(R2, P3, vtmp); // Pointer to Temporary matrix |
5776
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
74 B3 = R1; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
75 B2 = R2; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
76 L3 = 20; // L3 is used for making the coefficient array |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
77 // circular. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
78 // MUST BE RESTORED TO ZERO at function exit. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
79 M1 = 16 (X); // All these registers are initialized for |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
80 M3 = 8(X); // modifying address offsets. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
81 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
82 I0 = B0; // I0 points to Input Element (0, 0). |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
83 I2 = B0; // I2 points to Input Element (0, 0). |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
84 I2 += M3 || R0.H = W[I0]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
85 // Element 0 is read into R0.H |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
86 I1 = I2; // I1 points to input Element (0, 6). |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
87 I1 += 4 || R0.L = W[I2++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
88 // I2 points to input Element (0, 4). |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
89 // Element 4 is read into R0.L. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
90 P2 = 8 (X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
91 P3 = 32 (X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
92 P4 = -32 (X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
93 P5 = 98 (X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
94 R7 = 0x8000(Z); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
95 I3 = B3; // I3 points to Coefficients |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
96 P0 = B2; // P0 points to array Element (0, 0) of temp |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
97 P1 = B2; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
98 R7 = [I3++] || [TMP2]=R7; // Coefficient C4 is read into R7.H and R7.L. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
99 MNOP; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
100 NOP; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
101 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
102 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
103 * A1 = Y0 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
104 * A0 = Y0 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
105 * A1 = A1 + Y4 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
106 * A0 = A0 - Y4 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
107 * load: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
108 * R1=(Y2,Y6) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
109 * R7=(C2,C6) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
110 * res: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
111 * R3=Y0, R2=Y4 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
112 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
113 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+= 4 || R1.L=W[I1++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
114 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
115 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
116 LSETUP (.0, .1) LC0 = P2; // perform 8 1d idcts |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
117 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
118 P2 = 112 (X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
119 P1 = P1 + P2; // P1 points to element (7, 0) of temp buffer. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
120 P2 = -94(X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
121 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
122 .0: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
123 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
124 * A1 = Y2 * cos(3pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
125 * A0 = Y2 * cos(pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
126 * A1 = A1 - Y6 * cos(pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
127 * A0 = A0 + Y6 * cos(3pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
128 * R5 = (Y1,Y7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
129 * R7 = (C1,C7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
130 * res: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
131 * R1=Y2, R0=Y6 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
132 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
133 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
134 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
135 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
136 * Y0 = Y0 + Y6. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
137 * Y4 = Y4 + Y2. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
138 * Y2 = Y4 - Y2. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
139 * Y6 = Y0 - Y6. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
140 * R3 is saved |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
141 * R6.l=Y3 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
142 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
143 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
144 R3=R3+R0, R0=R3-R0; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
145 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
146 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
147 * Compute the odd portion (1,3,5,7) even is done. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
148 * |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
149 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
150 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
151 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
152 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
153 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
154 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
155 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
156 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
157 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
158 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
159 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
160 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
161 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
162 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
163 // R3=Y1, R2=Y7, R7=Y5, R6=Y3 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
164 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
165 /* Transpose write column. */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
166 R5.H=R4+R2 (RND12); // Y0=Y0+Y7 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
167 R5.L=R4-R2 (RND12) || R4 = [TMP1]; // Y7=Y7-Y0 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
168 R2.H=R1+R7 (RND12) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
169 R2.L=R1-R7 (RND12) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
170 R5.H=R0-R3 (RND12) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
171 R5.L=R0+R3 (RND12) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
172 R3.H=R4-R6 (RND12) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
173 R3.L=R4+R6 (RND12) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
174 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
175 /* pipeline loop start, + drain Y3, Y4 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
176 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
177 .1: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
178 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
179 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
180 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
181 I0 = B2; // I0 points to Input Element (0, 0) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
182 I2 = B2; // I2 points to Input Element (0, 0) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
183 I2 += M3 || R0.H = W[I0]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
184 // Y0 is read in R0.H |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
185 I1 = I2; // I1 points to input Element (0, 6) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
186 I1 += 4 || R0.L = W[I2++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
187 // I2 points to input Element (0, 4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
188 // Y4 is read in R0.L |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
189 P2 = 8 (X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
190 I3 = B3; // I3 points to Coefficients |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
191 P0 = B0; // P0 points to array Element (0, 0) for writing |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
192 // output |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
193 P1 = B0; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
194 R7 = [I3++]; // R7.H = C4 and R7.L = C4 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
195 NOP; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
196 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
197 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
198 * A1 = Y0 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
199 * A0 = Y0 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
200 * A1 = A1 + Y4 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
201 * A0 = A0 - Y4 * cos(pi/4) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
202 * load: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
203 * R1=(Y2,Y6) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
204 * R7=(C2,C6) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
205 * res: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
206 * R3=Y0, R2=Y4 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
207 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
208 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || I0+=4 || R1.L=W[I1++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
209 R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || R1.H=W[I0--] || R7=[I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
210 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
211 LSETUP (.2, .3) LC0 = P2; // peform 8 1d idcts |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
212 P2 = 112 (X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
213 P1 = P1 + P2; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
214 P2 = -94(X); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
215 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
216 .2: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
217 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
218 * A1 = Y2 * cos(3pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
219 * A0 = Y2 * cos(pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
220 * A1 = A1 - Y6 * cos(pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
221 * A0 = A0 + Y6 * cos(3pi/8) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
222 * R5 = (Y1,Y7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
223 * R7 = (C1,C7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
224 * res: |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
225 * R1=Y2, R0=Y6 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
226 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
227 A1=R7.L*R1.H, A0=R7.H*R1.H (IS) || I0+=4 || R5.H=W[I0]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
228 R1=(A1-=R7.H*R1.L), R0=(A0+=R7.L*R1.L) (IS) || R5.L=W[I1--] || R7=[I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
229 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
230 * Y0 = Y0 + Y6. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
231 * Y4 = Y4 + Y2. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
232 * Y2 = Y4 - Y2. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
233 * Y6 = Y0 - Y6. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
234 * R3 is saved |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
235 * R6.l=Y3 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
236 * note: R3: Y0, R2: Y4, R1: Y2, R0: Y6 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
237 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
238 R3=R3+R0, R0=R3-R0; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
239 R2=R2+R1, R1=R2-R1 || [TMP0]=R3 || R6.L=W[I0--]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
240 /* |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
241 * Compute the odd portion (1,3,5,7) even is done. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
242 * |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
243 * Y1 = C7 * Y1 - C1 * Y7 + C3 * Y5 - C5 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
244 * Y7 = C1 * Y1 + C7 * Y7 + C5 * Y5 + C3 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
245 * Y5 = C5 * Y1 + C3 * Y7 + C7 * Y5 - C1 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
246 * Y3 = C3 * Y1 - C5 * Y7 - C1 * Y5 - C7 * Y3. |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
247 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
248 // R5=(Y1,Y7) R6=(Y5,Y3) // R7=(C1,C7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
249 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || [TMP1]=R2 || R6.H=W[I2--]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
250 A1-=R7.H*R5.L, A0+=R7.L*R5.L (IS) || I0-=4 || R7=[I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
251 A1+=R7.H*R6.H, A0+=R7.L*R6.H (IS) || I0+=M1; // R7=(C3,C5) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
252 R3 =(A1-=R7.L*R6.L), R2 =(A0+=R7.H*R6.L) (IS); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
253 A1 =R7.L*R5.H, A0 =R7.H*R5.H (IS) || R4=[TMP0]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
254 A1+=R7.H*R5.L, A0-=R7.L*R5.L (IS) || I1+=M1 || R7=[I3++]; // R7=(C1,C7) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
255 A1+=R7.L*R6.H, A0-=R7.H*R6.H (IS); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
256 R7 =(A1-=R7.H*R6.L), R6 =(A0-=R7.L*R6.L) (IS) || I2+=M1; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
257 // R3=Y1, R2=Y7, R7=Y5, R6=Y3 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
258 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
259 /* Transpose write column. */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
260 R5.H=R4+R2 (RND20); // Y0=Y0+Y7 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
261 R5.L=R4-R2 (RND20) || R4 = [TMP1]; // Y7=Y7-Y0 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
262 R5=R5>>>2(v); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
263 R2.H=R1+R7 (RND20) || W[P0++P3]=R5.H; // Y2=Y2+Y5 st Y0 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
264 R2.L=R1-R7 (RND20) || W[P1++P4]=R5.L || R7=[I3++]; // Y5=Y2-Y5 st Y7 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
265 R2=R2>>>2(v); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
266 R5.H=R0-R3 (RND20) || W[P0++P3]=R2.H || R1.L=W[I1++]; // Y1=Y6-Y1 st Y2 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
267 R5.L=R0+R3 (RND20) || W[P1++P4]=R2.L || R0.H=W[I0++]; // Y6=Y6+Y1 st Y5 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
268 R5=R5>>>2(v); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
269 R3.H=R4-R6 (RND20) || W[P0++P3]=R5.H || R0.L=W[I2++]; // Y3=Y3-Y4 st Y1 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
270 R3.L=R4+R6 (RND20) || W[P1++P4]=R5.L || R1.H=W[I0++]; // Y4=Y3+Y4 st Y6 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
271 R3=R3>>>2(v); |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
272 /* pipeline loop start, + drain Y3, Y4 */ |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
273 A1=R7.H*R0.H, A0=R7.H*R0.H (IS) || W[P0++P2]= R3.H || R1.H = W[I0--]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
274 .3: R3=(A1+=R7.H*R0.L), R2=(A0-=R7.H*R0.L) (IS) || W[P1++P5]= R3.L || R7 = [I3++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
275 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
276 L3 = 0; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
277 (R7:4,P5:3)=[SP++]; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
278 unlink; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
279 RTS; |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
280 DEFUN_END(vp3_idct) |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
281 |
73ec16dbcbc1
blackfin optimized vp3 transform and infastructure for idct
mhoffman
parents:
diff
changeset
|
282 |