source: mainline/arch/ia64/src/fpu_context.c@ a84af84

lfn serial ticket/834-toolchain-update topic/msim-upgrade topic/simplify-dev-export
Last change on this file since a84af84 was 41fa6f2, checked in by Jakub Vana <jakub.vana@…>, 19 years ago

Itanium FPU Lazy context switching… but not so much tested

  • Property mode set to 100644
File size: 10.8 KB
Line 
1/*
2 * Copyright (C) 2005 Jakub Vana
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 *
9 * - Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * - Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * - The name of the author may not be used to endorse or promote products
15 * derived from this software without specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 *
28 */
29
30#include <fpu_context.h>
31#include <arch/register.h>
32#include <print.h>
33
34
35void fpu_context_save(fpu_context_t *fctx){
36
37 asm volatile(
38
39 "stf.spill [%0]=f32,0x80\n"
40 "stf.spill [%1]=f33,0x80\n"
41 "stf.spill [%2]=f34,0x80\n"
42 "stf.spill [%3]=f35,0x80\n"
43 "stf.spill [%4]=f36,0x80\n"
44 "stf.spill [%5]=f37,0x80\n"
45 "stf.spill [%6]=f38,0x80\n"
46 "stf.spill [%7]=f39,0x80\n;;"
47
48 "stf.spill [%0]=f40,0x80\n"
49 "stf.spill [%1]=f41,0x80\n"
50 "stf.spill [%2]=f42,0x80\n"
51 "stf.spill [%3]=f43,0x80\n"
52 "stf.spill [%4]=f44,0x80\n"
53 "stf.spill [%5]=f45,0x80\n"
54 "stf.spill [%6]=f46,0x80\n"
55 "stf.spill [%7]=f47,0x80\n;;"
56
57 "stf.spill [%0]=f48,0x80\n"
58 "stf.spill [%1]=f49,0x80\n"
59 "stf.spill [%2]=f50,0x80\n"
60 "stf.spill [%3]=f51,0x80\n"
61 "stf.spill [%4]=f52,0x80\n"
62 "stf.spill [%5]=f53,0x80\n"
63 "stf.spill [%6]=f54,0x80\n"
64 "stf.spill [%7]=f55,0x80\n;;"
65
66 "stf.spill [%0]=f56,0x80\n"
67 "stf.spill [%1]=f57,0x80\n"
68 "stf.spill [%2]=f58,0x80\n"
69 "stf.spill [%3]=f59,0x80\n"
70 "stf.spill [%4]=f60,0x80\n"
71 "stf.spill [%5]=f61,0x80\n"
72 "stf.spill [%6]=f62,0x80\n"
73 "stf.spill [%7]=f63,0x80\n;;"
74
75 "stf.spill [%0]=f64,0x80\n"
76 "stf.spill [%1]=f65,0x80\n"
77 "stf.spill [%2]=f66,0x80\n"
78 "stf.spill [%3]=f67,0x80\n"
79 "stf.spill [%4]=f68,0x80\n"
80 "stf.spill [%5]=f69,0x80\n"
81 "stf.spill [%6]=f70,0x80\n"
82 "stf.spill [%7]=f71,0x80\n;;"
83
84 "stf.spill [%0]=f72,0x80\n"
85 "stf.spill [%1]=f73,0x80\n"
86 "stf.spill [%2]=f74,0x80\n"
87 "stf.spill [%3]=f75,0x80\n"
88 "stf.spill [%4]=f76,0x80\n"
89 "stf.spill [%5]=f77,0x80\n"
90 "stf.spill [%6]=f78,0x80\n"
91 "stf.spill [%7]=f79,0x80\n;;"
92
93 "stf.spill [%0]=f80,0x80\n"
94 "stf.spill [%1]=f81,0x80\n"
95 "stf.spill [%2]=f82,0x80\n"
96 "stf.spill [%3]=f83,0x80\n"
97 "stf.spill [%4]=f84,0x80\n"
98 "stf.spill [%5]=f85,0x80\n"
99 "stf.spill [%6]=f86,0x80\n"
100 "stf.spill [%7]=f87,0x80\n;;"
101
102 "stf.spill [%0]=f88,0x80\n"
103 "stf.spill [%1]=f89,0x80\n"
104 "stf.spill [%2]=f90,0x80\n"
105 "stf.spill [%3]=f91,0x80\n"
106 "stf.spill [%4]=f92,0x80\n"
107 "stf.spill [%5]=f93,0x80\n"
108 "stf.spill [%6]=f94,0x80\n"
109 "stf.spill [%7]=f95,0x80\n;;"
110
111
112 "stf.spill [%0]=f96,0x80\n"
113 "stf.spill [%1]=f97,0x80\n"
114 "stf.spill [%2]=f98,0x80\n"
115 "stf.spill [%3]=f99,0x80\n"
116 "stf.spill [%4]=f100,0x80\n"
117 "stf.spill [%5]=f101,0x80\n"
118 "stf.spill [%6]=f102,0x80\n"
119 "stf.spill [%7]=f103,0x80\n;;"
120
121 "stf.spill [%0]=f104,0x80\n"
122 "stf.spill [%1]=f105,0x80\n"
123 "stf.spill [%2]=f106,0x80\n"
124 "stf.spill [%3]=f107,0x80\n"
125 "stf.spill [%4]=f108,0x80\n"
126 "stf.spill [%5]=f109,0x80\n"
127 "stf.spill [%6]=f110,0x80\n"
128 "stf.spill [%7]=f111,0x80\n;;"
129
130 "stf.spill [%0]=f112,0x80\n"
131 "stf.spill [%1]=f113,0x80\n"
132 "stf.spill [%2]=f114,0x80\n"
133 "stf.spill [%3]=f115,0x80\n"
134 "stf.spill [%4]=f116,0x80\n"
135 "stf.spill [%5]=f117,0x80\n"
136 "stf.spill [%6]=f118,0x80\n"
137 "stf.spill [%7]=f119,0x80\n;;"
138
139 "stf.spill [%0]=f120,0x80\n"
140 "stf.spill [%1]=f121,0x80\n"
141 "stf.spill [%2]=f122,0x80\n"
142 "stf.spill [%3]=f123,0x80\n"
143 "stf.spill [%4]=f124,0x80\n"
144 "stf.spill [%5]=f125,0x80\n"
145 "stf.spill [%6]=f126,0x80\n"
146 "stf.spill [%7]=f127,0x80\n;;"
147
148
149 :
150 :"r" (&((fctx->fr)[0])),"r" (&((fctx->fr)[1])),"r" (&((fctx->fr)[2])),"r" (&((fctx->fr)[3])),
151 "r" (&((fctx->fr)[4])),"r" (&((fctx->fr)[5])),"r" (&((fctx->fr)[6])),"r" (&((fctx->fr)[7]))
152 );
153
154}
155
156
157void fpu_context_restore(fpu_context_t *fctx)
158{
159
160 asm volatile(
161 "ldf.fill f32=[%0],0x80\n"
162 "ldf.fill f33=[%1],0x80\n"
163 "ldf.fill f34=[%2],0x80\n"
164 "ldf.fill f35=[%3],0x80\n"
165 "ldf.fill f36=[%4],0x80\n"
166 "ldf.fill f37=[%5],0x80\n"
167 "ldf.fill f38=[%6],0x80\n"
168 "ldf.fill f39=[%7],0x80\n;;"
169
170 "ldf.fill f40=[%0],0x80\n"
171 "ldf.fill f41=[%1],0x80\n"
172 "ldf.fill f42=[%2],0x80\n"
173 "ldf.fill f43=[%3],0x80\n"
174 "ldf.fill f44=[%4],0x80\n"
175 "ldf.fill f45=[%5],0x80\n"
176 "ldf.fill f46=[%6],0x80\n"
177 "ldf.fill f47=[%7],0x80\n;;"
178
179 "ldf.fill f48=[%0],0x80\n"
180 "ldf.fill f49=[%1],0x80\n"
181 "ldf.fill f50=[%2],0x80\n"
182 "ldf.fill f51=[%3],0x80\n"
183 "ldf.fill f52=[%4],0x80\n"
184 "ldf.fill f53=[%5],0x80\n"
185 "ldf.fill f54=[%6],0x80\n"
186 "ldf.fill f55=[%7],0x80\n;;"
187
188 "ldf.fill f56=[%0],0x80\n"
189 "ldf.fill f57=[%1],0x80\n"
190 "ldf.fill f58=[%2],0x80\n"
191 "ldf.fill f59=[%3],0x80\n"
192 "ldf.fill f60=[%4],0x80\n"
193 "ldf.fill f61=[%5],0x80\n"
194 "ldf.fill f62=[%6],0x80\n"
195 "ldf.fill f63=[%7],0x80\n;;"
196
197 "ldf.fill f64=[%0],0x80\n"
198 "ldf.fill f65=[%1],0x80\n"
199 "ldf.fill f66=[%2],0x80\n"
200 "ldf.fill f67=[%3],0x80\n"
201 "ldf.fill f68=[%4],0x80\n"
202 "ldf.fill f69=[%5],0x80\n"
203 "ldf.fill f70=[%6],0x80\n"
204 "ldf.fill f71=[%7],0x80\n;;"
205
206 "ldf.fill f72=[%0],0x80\n"
207 "ldf.fill f73=[%1],0x80\n"
208 "ldf.fill f74=[%2],0x80\n"
209 "ldf.fill f75=[%3],0x80\n"
210 "ldf.fill f76=[%4],0x80\n"
211 "ldf.fill f77=[%5],0x80\n"
212 "ldf.fill f78=[%6],0x80\n"
213 "ldf.fill f79=[%7],0x80\n;;"
214
215 "ldf.fill f80=[%0],0x80\n"
216 "ldf.fill f81=[%1],0x80\n"
217 "ldf.fill f82=[%2],0x80\n"
218 "ldf.fill f83=[%3],0x80\n"
219 "ldf.fill f84=[%4],0x80\n"
220 "ldf.fill f85=[%5],0x80\n"
221 "ldf.fill f86=[%6],0x80\n"
222 "ldf.fill f87=[%7],0x80\n;;"
223
224 "ldf.fill f88=[%0],0x80\n"
225 "ldf.fill f89=[%1],0x80\n"
226 "ldf.fill f90=[%2],0x80\n"
227 "ldf.fill f91=[%3],0x80\n"
228 "ldf.fill f92=[%4],0x80\n"
229 "ldf.fill f93=[%5],0x80\n"
230 "ldf.fill f94=[%6],0x80\n"
231 "ldf.fill f95=[%7],0x80\n;;"
232
233
234 "ldf.fill f96=[%0],0x80\n"
235 "ldf.fill f97=[%1],0x80\n"
236 "ldf.fill f98=[%2],0x80\n"
237 "ldf.fill f99=[%3],0x80\n"
238 "ldf.fill f100=[%4],0x80\n"
239 "ldf.fill f101=[%5],0x80\n"
240 "ldf.fill f102=[%6],0x80\n"
241 "ldf.fill f103=[%7],0x80\n;;"
242
243 "ldf.fill f104=[%0],0x80\n"
244 "ldf.fill f105=[%1],0x80\n"
245 "ldf.fill f106=[%2],0x80\n"
246 "ldf.fill f107=[%3],0x80\n"
247 "ldf.fill f108=[%4],0x80\n"
248 "ldf.fill f109=[%5],0x80\n"
249 "ldf.fill f110=[%6],0x80\n"
250 "ldf.fill f111=[%7],0x80\n;;"
251
252 "ldf.fill f112=[%0],0x80\n"
253 "ldf.fill f113=[%1],0x80\n"
254 "ldf.fill f114=[%2],0x80\n"
255 "ldf.fill f115=[%3],0x80\n"
256 "ldf.fill f116=[%4],0x80\n"
257 "ldf.fill f117=[%5],0x80\n"
258 "ldf.fill f118=[%6],0x80\n"
259 "ldf.fill f119=[%7],0x80\n;;"
260
261 "ldf.fill f120=[%0],0x80\n"
262 "ldf.fill f121=[%1],0x80\n"
263 "ldf.fill f122=[%2],0x80\n"
264 "ldf.fill f123=[%3],0x80\n"
265 "ldf.fill f124=[%4],0x80\n"
266 "ldf.fill f125=[%5],0x80\n"
267 "ldf.fill f126=[%6],0x80\n"
268 "ldf.fill f127=[%7],0x80\n;;"
269
270
271 :
272 :"r" (&((fctx->fr)[0])),"r" (&((fctx->fr)[1])),"r" (&((fctx->fr)[2])),"r" (&((fctx->fr)[3])),
273 "r" (&((fctx->fr)[4])),"r" (&((fctx->fr)[5])),"r" (&((fctx->fr)[6])),"r" (&((fctx->fr)[7]))
274 );
275}
276
277void fpu_enable(void)
278{
279 __u64 a = 0 ;
280 asm volatile(
281 "rsm %0;;"
282 "srlz.i\n"
283 "srlz.d;;\n"
284 :
285 :"i" (PSR_DFH_MASK)
286 );
287 asm volatile
288 (
289 "mov %0=ar.fpsr;;\n"
290 "or %0=%0,%1;;\n"
291 "mov ar.fpsr=%0;;\n"
292 : "+r" (a)
293 : "r" (0x38)
294 );
295
296}
297
298void fpu_disable(void)
299{
300
301 __u64 a = 0 ;
302 asm volatile(
303 "ssm %0;;\n"
304 "srlz.i\n"
305 "srlz.d;;\n"
306 :
307 :"i" (PSR_DFH_MASK)
308 );
309 asm volatile
310 (
311 "mov %0=ar.fpsr;;\n"
312 "or %0=%0,%1;;\n"
313 "mov ar.fpsr=%0;;\n"
314 : "+r" (a)
315 : "r" (0x38)
316 );
317
318}
319
320void fpu_init(void)
321{
322 __u64 a = 0 ;
323 asm volatile
324 (
325 "mov %0=ar.fpsr;;\n"
326 "or %0=%0,%1;;\n"
327 "mov ar.fpsr=%0;;\n"
328 : "+r" (a)
329 : "r" (0x38)
330 );
331
332 asm volatile(
333 "mov f2=f0\n"
334 "mov f3=f0\n"
335 "mov f4=f0\n"
336 "mov f5=f0\n"
337 "mov f6=f0\n"
338 "mov f7=f0\n"
339 "mov f8=f0\n"
340 "mov f9=f0\n"
341
342 "mov f10=f0\n"
343 "mov f11=f0\n"
344 "mov f12=f0\n"
345 "mov f13=f0\n"
346 "mov f14=f0\n"
347 "mov f15=f0\n"
348 "mov f16=f0\n"
349 "mov f17=f0\n"
350 "mov f18=f0\n"
351 "mov f19=f0\n"
352
353 "mov f20=f0\n"
354 "mov f21=f0\n"
355 "mov f22=f0\n"
356 "mov f23=f0\n"
357 "mov f24=f0\n"
358 "mov f25=f0\n"
359 "mov f26=f0\n"
360 "mov f27=f0\n"
361 "mov f28=f0\n"
362 "mov f29=f0\n"
363
364 "mov f30=f0\n"
365 "mov f31=f0\n"
366 "mov f32=f0\n"
367 "mov f33=f0\n"
368 "mov f34=f0\n"
369 "mov f35=f0\n"
370 "mov f36=f0\n"
371 "mov f37=f0\n"
372 "mov f38=f0\n"
373 "mov f39=f0\n"
374
375 "mov f40=f0\n"
376 "mov f41=f0\n"
377 "mov f42=f0\n"
378 "mov f43=f0\n"
379 "mov f44=f0\n"
380 "mov f45=f0\n"
381 "mov f46=f0\n"
382 "mov f47=f0\n"
383 "mov f48=f0\n"
384 "mov f49=f0\n"
385
386 "mov f50=f0\n"
387 "mov f51=f0\n"
388 "mov f52=f0\n"
389 "mov f53=f0\n"
390 "mov f54=f0\n"
391 "mov f55=f0\n"
392 "mov f56=f0\n"
393 "mov f57=f0\n"
394 "mov f58=f0\n"
395 "mov f59=f0\n"
396
397 "mov f60=f0\n"
398 "mov f61=f0\n"
399 "mov f62=f0\n"
400 "mov f63=f0\n"
401 "mov f64=f0\n"
402 "mov f65=f0\n"
403 "mov f66=f0\n"
404 "mov f67=f0\n"
405 "mov f68=f0\n"
406 "mov f69=f0\n"
407
408 "mov f70=f0\n"
409 "mov f71=f0\n"
410 "mov f72=f0\n"
411 "mov f73=f0\n"
412 "mov f74=f0\n"
413 "mov f75=f0\n"
414 "mov f76=f0\n"
415 "mov f77=f0\n"
416 "mov f78=f0\n"
417 "mov f79=f0\n"
418
419 "mov f80=f0\n"
420 "mov f81=f0\n"
421 "mov f82=f0\n"
422 "mov f83=f0\n"
423 "mov f84=f0\n"
424 "mov f85=f0\n"
425 "mov f86=f0\n"
426 "mov f87=f0\n"
427 "mov f88=f0\n"
428 "mov f89=f0\n"
429
430 "mov f90=f0\n"
431 "mov f91=f0\n"
432 "mov f92=f0\n"
433 "mov f93=f0\n"
434 "mov f94=f0\n"
435 "mov f95=f0\n"
436 "mov f96=f0\n"
437 "mov f97=f0\n"
438 "mov f98=f0\n"
439 "mov f99=f0\n"
440
441 "mov f100=f0\n"
442 "mov f101=f0\n"
443 "mov f102=f0\n"
444 "mov f103=f0\n"
445 "mov f104=f0\n"
446 "mov f105=f0\n"
447 "mov f106=f0\n"
448 "mov f107=f0\n"
449 "mov f108=f0\n"
450 "mov f109=f0\n"
451
452 "mov f110=f0\n"
453 "mov f111=f0\n"
454 "mov f112=f0\n"
455 "mov f113=f0\n"
456 "mov f114=f0\n"
457 "mov f115=f0\n"
458 "mov f116=f0\n"
459 "mov f117=f0\n"
460 "mov f118=f0\n"
461 "mov f119=f0\n"
462
463 "mov f120=f0\n"
464 "mov f121=f0\n"
465 "mov f122=f0\n"
466 "mov f123=f0\n"
467 "mov f124=f0\n"
468 "mov f125=f0\n"
469 "mov f126=f0\n"
470 "mov f127=f0\n"
471
472 );
473
474}
475
Note: See TracBrowser for help on using the repository browser.