| 3960 |
Serge |
1 |
/* |
|
|
2 |
decode_3dnow.s - 3DNow! optimized synth_1to1() |
|
|
3 |
|
|
|
4 |
copyright ?-2007 by the mpg123 project - free software under the terms of the LGPL 2.1 |
|
|
5 |
see COPYING and AUTHORS files in distribution or http://mpg123.org |
|
|
6 |
initially written by Syuuhei Kashiyama |
|
|
7 |
|
|
|
8 |
This code based 'decode_3dnow.s' by Syuuhei Kashiyama |
|
|
9 |
,only two types of changes have been made: |
|
|
10 |
|
|
|
11 |
- remove PREFETCH instruction for speedup |
|
|
12 |
- change function name for support 3DNow! automatic detect |
|
|
13 |
- femms moved to before 'call dct64_3dnow' |
|
|
14 |
|
|
|
15 |
You can find Kashiyama's original 3dnow! support patch |
|
|
16 |
(for mpg123-0.59o) at |
|
|
17 |
http://user.ecc.u-tokyo.ac.jp/~g810370/linux-simd/ (Japanese). |
|
|
18 |
|
|
|
19 |
by KIMURA Takuhiro - until 31.Mar.1999 |
|
|
20 |
- after 1.Apr.1999 |
|
|
21 |
|
|
|
22 |
|
|
|
23 |
|
|
|
24 |
Replacement of synth_1to1() with AMD's 3DNow! SIMD operations support |
|
|
25 |
|
|
|
26 |
Syuuhei Kashiyama |
|
|
27 |
|
|
|
28 |
The author of this program disclaim whole expressed or implied |
|
|
29 |
warranties with regard to this program, and in no event shall the |
|
|
30 |
author of this program liable to whatever resulted from the use of |
|
|
31 |
this program. Use it at your own risk. |
|
|
32 |
*/ |
|
|
33 |
|
|
|
34 |
#include "mangle.h" |
|
|
35 |
|
|
|
36 |
#ifdef ACCURATE_ROUNDING |
|
|
37 |
#ifndef __APPLE__ |
|
|
38 |
.section .rodata |
|
|
39 |
#else |
|
|
40 |
.data |
|
|
41 |
#endif |
|
|
42 |
ALIGN8 |
|
|
43 |
max_s16: |
|
|
44 |
.long 1191181824 /* 32767.0 */ |
|
|
45 |
.long 1191181824 |
|
|
46 |
min_s16: |
|
|
47 |
.long -956301312 /* -32768.0 */ |
|
|
48 |
.long -956301312 |
|
|
49 |
ftoi_magic: |
|
|
50 |
.long 1262485504 /* 2^23 + 2^22 */ |
|
|
51 |
.long 1262485504 |
|
|
52 |
#endif |
|
|
53 |
.text |
|
|
54 |
ALIGN16 |
|
|
55 |
.globl ASM_NAME(synth_1to1_3dnow_asm) |
|
|
56 |
/* int synth_1to1_3dnow_asm(real *bandPtr, int channel, unsigned char *out, unsigned char *buffs, int *bo, real *decwin); */ |
|
|
57 |
ASM_NAME(synth_1to1_3dnow_asm): |
|
|
58 |
subl $24,%esp |
|
|
59 |
pushl %ebp |
|
|
60 |
pushl %edi |
|
|
61 |
xorl %ebp,%ebp |
|
|
62 |
pushl %esi |
|
|
63 |
pushl %ebx |
|
|
64 |
/* stack old: 0=ebx 4=esi 8=edi 12=ebp 16,20,24,28,32,36=local 40=back 44=bandptr 48=channel 52=out 56=pnt */ |
|
|
65 |
/* stack new: 0=ebx 4=esi 8=edi 12=ebp 16,20,24,28,32,36=local 40=back 44=bandptr 48=channel 52=out 56=buffs 60=bo 64=decwin */ |
|
|
66 |
#define OUT 52(%esp) |
|
|
67 |
#define CHANNEL 48(%esp) |
|
|
68 |
#define BANDPTR 44(%esp) |
|
|
69 |
#define BUFFS 56(%esp) |
|
|
70 |
#define BO 60(%esp) |
|
|
71 |
#define DECWIN 64(%esp) |
|
|
72 |
#define LOCAL0 16(%esp) |
|
|
73 |
#define LOCAL1 20(%esp) |
|
|
74 |
#define LOCAL5 36(%esp) |
|
|
75 |
movl OUT,%esi |
|
|
76 |
movl %esi,LOCAL0 /* save buffer start (samples pointer) to another local var */ |
|
|
77 |
movl CHANNEL,%ebx |
|
|
78 |
movl BO,%esi /* bo address */ |
|
|
79 |
movl (%esi),%edx /* bo value */ |
|
|
80 |
|
|
|
81 |
femms |
|
|
82 |
testl %ebx,%ebx |
|
|
83 |
jne .L26 |
|
|
84 |
/* if(!channel) */ |
|
|
85 |
decl %edx /* --bo */ |
|
|
86 |
andl $15,%edx |
|
|
87 |
movl %edx,(%esi) /* save bo */ |
|
|
88 |
movl BUFFS,%ecx |
|
|
89 |
jmp .L27 |
|
|
90 |
.L26: /* if(channel) */ |
|
|
91 |
addl $2,LOCAL0 /* samples++ */ |
|
|
92 |
movl BUFFS,%ecx |
|
|
93 |
addl $2176,%ecx |
|
|
94 |
.L27: |
|
|
95 |
/* edx (and it's lower end) still holds bo value */ |
|
|
96 |
testb $1,%dl /* bo & 0x1 */ |
|
|
97 |
je .L28 |
|
|
98 |
movl %edx,LOCAL5 |
|
|
99 |
movl %ecx,%ebx |
|
|
100 |
movl BANDPTR,%esi |
|
|
101 |
movl %edx,%edi |
|
|
102 |
pushl %esi |
|
|
103 |
sall $2,%edi |
|
|
104 |
movl %ebx,%eax |
|
|
105 |
movl %edi,24(%esp) /* LOCAL1, actually */ |
|
|
106 |
addl %edi,%eax |
|
|
107 |
pushl %eax |
|
|
108 |
movl %edx,%eax |
|
|
109 |
incl %eax |
|
|
110 |
andl $15,%eax |
|
|
111 |
leal 1088(,%eax,4),%eax |
|
|
112 |
addl %ebx,%eax |
|
|
113 |
pushl %eax |
|
|
114 |
call ASM_NAME(dct64_3dnow) |
|
|
115 |
addl $12,%esp |
|
|
116 |
jmp .L29 |
|
|
117 |
.L28: |
|
|
118 |
leal 1(%edx),%esi |
|
|
119 |
movl BANDPTR,%edi |
|
|
120 |
movl %esi,LOCAL5 |
|
|
121 |
leal 1092(%ecx,%edx,4),%eax |
|
|
122 |
pushl %edi |
|
|
123 |
leal 1088(%ecx),%ebx |
|
|
124 |
pushl %eax |
|
|
125 |
sall $2,%esi |
|
|
126 |
leal (%ecx,%edx,4),%eax |
|
|
127 |
pushl %eax |
|
|
128 |
call ASM_NAME(dct64_3dnow) |
|
|
129 |
addl $12,%esp |
|
|
130 |
movl %esi,LOCAL1 |
|
|
131 |
.L29: |
|
|
132 |
movl DECWIN,%edx |
|
|
133 |
addl $64,%edx |
|
|
134 |
movl $16,%ecx |
|
|
135 |
subl LOCAL1,%edx |
|
|
136 |
movl LOCAL0,%edi |
|
|
137 |
|
|
|
138 |
pcmpeqb %mm7,%mm7 |
|
|
139 |
pslld $31,%mm7 |
|
|
140 |
movq (%edx),%mm0 |
|
|
141 |
movq (%ebx),%mm1 |
|
|
142 |
ALIGN32 |
|
|
143 |
.L33: |
|
|
144 |
movq 8(%edx),%mm3 |
|
|
145 |
pfmul %mm1,%mm0 |
|
|
146 |
movq 8(%ebx),%mm4 |
|
|
147 |
movq 16(%edx),%mm5 |
|
|
148 |
pfmul %mm4,%mm3 |
|
|
149 |
movq 16(%ebx),%mm6 |
|
|
150 |
pfadd %mm3,%mm0 |
|
|
151 |
movq 24(%edx),%mm1 |
|
|
152 |
pfmul %mm6,%mm5 |
|
|
153 |
movq 24(%ebx),%mm2 |
|
|
154 |
pfadd %mm5,%mm0 |
|
|
155 |
movq 32(%edx),%mm3 |
|
|
156 |
pfmul %mm2,%mm1 |
|
|
157 |
movq 32(%ebx),%mm4 |
|
|
158 |
pfadd %mm1,%mm0 |
|
|
159 |
movq 40(%edx),%mm5 |
|
|
160 |
pfmul %mm4,%mm3 |
|
|
161 |
movq 40(%ebx),%mm6 |
|
|
162 |
pfadd %mm3,%mm0 |
|
|
163 |
movq 48(%edx),%mm1 |
|
|
164 |
pfmul %mm6,%mm5 |
|
|
165 |
movq 48(%ebx),%mm2 |
|
|
166 |
pfadd %mm0,%mm5 |
|
|
167 |
movq 56(%edx),%mm3 |
|
|
168 |
pfmul %mm1,%mm2 |
|
|
169 |
movq 56(%ebx),%mm4 |
|
|
170 |
pfadd %mm5,%mm2 |
|
|
171 |
addl $64,%ebx |
|
|
172 |
subl $-128,%edx |
|
|
173 |
movq (%edx),%mm0 |
|
|
174 |
pfmul %mm4,%mm3 |
|
|
175 |
movq (%ebx),%mm1 |
|
|
176 |
pfadd %mm3,%mm2 |
|
|
177 |
movq %mm2,%mm3 |
|
|
178 |
psrlq $32,%mm3 |
|
|
179 |
pfsub %mm3,%mm2 |
|
|
180 |
incl %ebp |
|
|
181 |
#ifdef ACCURATE_ROUNDING |
|
|
182 |
pfmin (max_s16),%mm2 |
|
|
183 |
pfmax (min_s16),%mm2 |
|
|
184 |
pfadd (ftoi_magic),%mm2 |
|
|
185 |
#else |
|
|
186 |
pf2id %mm2,%mm2 |
|
|
187 |
packssdw %mm2,%mm2 |
|
|
188 |
#endif |
|
|
189 |
movd %mm2,%eax |
|
|
190 |
movw %ax,0(%edi) |
|
|
191 |
addl $4,%edi |
|
|
192 |
decl %ecx |
|
|
193 |
jnz .L33 |
|
|
194 |
|
|
|
195 |
movd (%ebx),%mm0 |
|
|
196 |
movd (%edx),%mm1 |
|
|
197 |
punpckldq 8(%ebx),%mm0 |
|
|
198 |
punpckldq 8(%edx),%mm1 |
|
|
199 |
movd 16(%ebx),%mm3 |
|
|
200 |
movd 16(%edx),%mm4 |
|
|
201 |
pfmul %mm1,%mm0 |
|
|
202 |
punpckldq 24(%ebx),%mm3 |
|
|
203 |
punpckldq 24(%edx),%mm4 |
|
|
204 |
movd 32(%ebx),%mm5 |
|
|
205 |
movd 32(%edx),%mm6 |
|
|
206 |
pfmul %mm4,%mm3 |
|
|
207 |
punpckldq 40(%ebx),%mm5 |
|
|
208 |
punpckldq 40(%edx),%mm6 |
|
|
209 |
pfadd %mm3,%mm0 |
|
|
210 |
movd 48(%ebx),%mm1 |
|
|
211 |
movd 48(%edx),%mm2 |
|
|
212 |
pfmul %mm6,%mm5 |
|
|
213 |
punpckldq 56(%ebx),%mm1 |
|
|
214 |
punpckldq 56(%edx),%mm2 |
|
|
215 |
pfadd %mm5,%mm0 |
|
|
216 |
pfmul %mm2,%mm1 |
|
|
217 |
pfadd %mm1,%mm0 |
|
|
218 |
pfacc %mm1,%mm0 |
|
|
219 |
#ifdef ACCURATE_ROUNDING |
|
|
220 |
pfmin (max_s16),%mm0 |
|
|
221 |
pfmax (min_s16),%mm0 |
|
|
222 |
pfadd (ftoi_magic),%mm0 |
|
|
223 |
#else |
|
|
224 |
pf2id %mm0,%mm0 |
|
|
225 |
packssdw %mm0,%mm0 |
|
|
226 |
#endif |
|
|
227 |
movd %mm0,%eax |
|
|
228 |
movw %ax,0(%edi) |
|
|
229 |
incl %ebp |
|
|
230 |
movl LOCAL5,%esi |
|
|
231 |
addl $-64,%ebx |
|
|
232 |
movl $15,%ebp |
|
|
233 |
addl $4,%edi |
|
|
234 |
leal -128(%edx,%esi,8),%edx |
|
|
235 |
|
|
|
236 |
movl $15,%ecx |
|
|
237 |
movd (%ebx),%mm0 |
|
|
238 |
movd -4(%edx),%mm1 |
|
|
239 |
punpckldq 4(%ebx),%mm0 |
|
|
240 |
punpckldq -8(%edx),%mm1 |
|
|
241 |
ALIGN32 |
|
|
242 |
.L46: |
|
|
243 |
movd 8(%ebx),%mm3 |
|
|
244 |
movd -12(%edx),%mm4 |
|
|
245 |
pfmul %mm1,%mm0 |
|
|
246 |
punpckldq 12(%ebx),%mm3 |
|
|
247 |
punpckldq -16(%edx),%mm4 |
|
|
248 |
movd 16(%ebx),%mm5 |
|
|
249 |
movd -20(%edx),%mm6 |
|
|
250 |
pfmul %mm4,%mm3 |
|
|
251 |
punpckldq 20(%ebx),%mm5 |
|
|
252 |
punpckldq -24(%edx),%mm6 |
|
|
253 |
pfadd %mm3,%mm0 |
|
|
254 |
movd 24(%ebx),%mm1 |
|
|
255 |
movd -28(%edx),%mm2 |
|
|
256 |
pfmul %mm6,%mm5 |
|
|
257 |
punpckldq 28(%ebx),%mm1 |
|
|
258 |
punpckldq -32(%edx),%mm2 |
|
|
259 |
pfadd %mm5,%mm0 |
|
|
260 |
movd 32(%ebx),%mm3 |
|
|
261 |
movd -36(%edx),%mm4 |
|
|
262 |
pfmul %mm2,%mm1 |
|
|
263 |
punpckldq 36(%ebx),%mm3 |
|
|
264 |
punpckldq -40(%edx),%mm4 |
|
|
265 |
pfadd %mm1,%mm0 |
|
|
266 |
movd 40(%ebx),%mm5 |
|
|
267 |
movd -44(%edx),%mm6 |
|
|
268 |
pfmul %mm4,%mm3 |
|
|
269 |
punpckldq 44(%ebx),%mm5 |
|
|
270 |
punpckldq -48(%edx),%mm6 |
|
|
271 |
pfadd %mm3,%mm0 |
|
|
272 |
movd 48(%ebx),%mm1 |
|
|
273 |
movd -52(%edx),%mm2 |
|
|
274 |
pfmul %mm6,%mm5 |
|
|
275 |
punpckldq 52(%ebx),%mm1 |
|
|
276 |
punpckldq -56(%edx),%mm2 |
|
|
277 |
pfadd %mm0,%mm5 |
|
|
278 |
movd 56(%ebx),%mm3 |
|
|
279 |
movd -60(%edx),%mm4 |
|
|
280 |
pfmul %mm2,%mm1 |
|
|
281 |
punpckldq 60(%ebx),%mm3 |
|
|
282 |
punpckldq (%edx),%mm4 |
|
|
283 |
pfadd %mm1,%mm5 |
|
|
284 |
addl $-128,%edx |
|
|
285 |
addl $-64,%ebx |
|
|
286 |
movd (%ebx),%mm0 |
|
|
287 |
movd -4(%edx),%mm1 |
|
|
288 |
pfmul %mm4,%mm3 |
|
|
289 |
punpckldq 4(%ebx),%mm0 |
|
|
290 |
punpckldq -8(%edx),%mm1 |
|
|
291 |
pfadd %mm5,%mm3 |
|
|
292 |
pfacc %mm3,%mm3 |
|
|
293 |
incl %ebp |
|
|
294 |
pxor %mm7,%mm3 |
|
|
295 |
#ifdef ACCURATE_ROUNDING |
|
|
296 |
pfmin (max_s16),%mm3 |
|
|
297 |
pfmax (min_s16),%mm3 |
|
|
298 |
pfadd (ftoi_magic),%mm3 |
|
|
299 |
#else |
|
|
300 |
pf2id %mm3,%mm3 |
|
|
301 |
packssdw %mm3,%mm3 |
|
|
302 |
#endif |
|
|
303 |
movd %mm3,%eax |
|
|
304 |
movw %ax,(%edi) |
|
|
305 |
addl $4,%edi |
|
|
306 |
decl %ecx |
|
|
307 |
jnz .L46 |
|
|
308 |
|
|
|
309 |
femms |
|
|
310 |
movl %ebp,%eax |
|
|
311 |
popl %ebx |
|
|
312 |
popl %esi |
|
|
313 |
popl %edi |
|
|
314 |
popl %ebp |
|
|
315 |
addl $24,%esp |
|
|
316 |
ret |
|
|
317 |
|
|
|
318 |
NONEXEC_STACK |