diff options
author | Sam Ravnborg <sam@ravnborg.org> | 2008-11-16 23:07:11 -0500 |
---|---|---|
committer | David S. Miller <davem@davemloft.net> | 2008-12-04 12:16:56 -0500 |
commit | 5115f39c2034cb80a050d996a2a6343bce189628 (patch) | |
tree | f106dcf237f9f7de49d4ad5d5f3b81968f663e46 /arch/sparc/math-emu | |
parent | 774434bf33bdc876c2818eba34e416fe1fc7a746 (diff) |
sparc64: unify math-emu
Move relavent files to sparc/math-emu and
adjust path/include accordingly.
Signed-off-by: Sam Ravnborg <sam@ravnborg.org>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'arch/sparc/math-emu')
-rw-r--r-- | arch/sparc/math-emu/math_64.c | 513 | ||||
-rw-r--r-- | arch/sparc/math-emu/sfp-util_64.h | 120 |
2 files changed, 633 insertions, 0 deletions
diff --git a/arch/sparc/math-emu/math_64.c b/arch/sparc/math-emu/math_64.c new file mode 100644 index 000000000000..6863c9bde25c --- /dev/null +++ b/arch/sparc/math-emu/math_64.c | |||
@@ -0,0 +1,513 @@ | |||
1 | /* | ||
2 | * arch/sparc64/math-emu/math.c | ||
3 | * | ||
4 | * Copyright (C) 1997,1999 Jakub Jelinek (jj@ultra.linux.cz) | ||
5 | * Copyright (C) 1999 David S. Miller (davem@redhat.com) | ||
6 | * | ||
7 | * Emulation routines originate from soft-fp package, which is part | ||
8 | * of glibc and has appropriate copyrights in it. | ||
9 | */ | ||
10 | |||
11 | #include <linux/types.h> | ||
12 | #include <linux/sched.h> | ||
13 | #include <linux/errno.h> | ||
14 | |||
15 | #include <asm/fpumacro.h> | ||
16 | #include <asm/ptrace.h> | ||
17 | #include <asm/uaccess.h> | ||
18 | |||
19 | #include "sfp-util_64.h" | ||
20 | #include <math-emu/soft-fp.h> | ||
21 | #include <math-emu/single.h> | ||
22 | #include <math-emu/double.h> | ||
23 | #include <math-emu/quad.h> | ||
24 | |||
25 | /* QUAD - ftt == 3 */ | ||
26 | #define FMOVQ 0x003 | ||
27 | #define FNEGQ 0x007 | ||
28 | #define FABSQ 0x00b | ||
29 | #define FSQRTQ 0x02b | ||
30 | #define FADDQ 0x043 | ||
31 | #define FSUBQ 0x047 | ||
32 | #define FMULQ 0x04b | ||
33 | #define FDIVQ 0x04f | ||
34 | #define FDMULQ 0x06e | ||
35 | #define FQTOX 0x083 | ||
36 | #define FXTOQ 0x08c | ||
37 | #define FQTOS 0x0c7 | ||
38 | #define FQTOD 0x0cb | ||
39 | #define FITOQ 0x0cc | ||
40 | #define FSTOQ 0x0cd | ||
41 | #define FDTOQ 0x0ce | ||
42 | #define FQTOI 0x0d3 | ||
43 | /* SUBNORMAL - ftt == 2 */ | ||
44 | #define FSQRTS 0x029 | ||
45 | #define FSQRTD 0x02a | ||
46 | #define FADDS 0x041 | ||
47 | #define FADDD 0x042 | ||
48 | #define FSUBS 0x045 | ||
49 | #define FSUBD 0x046 | ||
50 | #define FMULS 0x049 | ||
51 | #define FMULD 0x04a | ||
52 | #define FDIVS 0x04d | ||
53 | #define FDIVD 0x04e | ||
54 | #define FSMULD 0x069 | ||
55 | #define FSTOX 0x081 | ||
56 | #define FDTOX 0x082 | ||
57 | #define FDTOS 0x0c6 | ||
58 | #define FSTOD 0x0c9 | ||
59 | #define FSTOI 0x0d1 | ||
60 | #define FDTOI 0x0d2 | ||
61 | #define FXTOS 0x084 /* Only Ultra-III generates this. */ | ||
62 | #define FXTOD 0x088 /* Only Ultra-III generates this. */ | ||
63 | #if 0 /* Optimized inline in sparc64/kernel/entry.S */ | ||
64 | #define FITOS 0x0c4 /* Only Ultra-III generates this. */ | ||
65 | #endif | ||
66 | #define FITOD 0x0c8 /* Only Ultra-III generates this. */ | ||
67 | /* FPOP2 */ | ||
68 | #define FCMPQ 0x053 | ||
69 | #define FCMPEQ 0x057 | ||
70 | #define FMOVQ0 0x003 | ||
71 | #define FMOVQ1 0x043 | ||
72 | #define FMOVQ2 0x083 | ||
73 | #define FMOVQ3 0x0c3 | ||
74 | #define FMOVQI 0x103 | ||
75 | #define FMOVQX 0x183 | ||
76 | #define FMOVQZ 0x027 | ||
77 | #define FMOVQLE 0x047 | ||
78 | #define FMOVQLZ 0x067 | ||
79 | #define FMOVQNZ 0x0a7 | ||
80 | #define FMOVQGZ 0x0c7 | ||
81 | #define FMOVQGE 0x0e7 | ||
82 | |||
83 | #define FSR_TEM_SHIFT 23UL | ||
84 | #define FSR_TEM_MASK (0x1fUL << FSR_TEM_SHIFT) | ||
85 | #define FSR_AEXC_SHIFT 5UL | ||
86 | #define FSR_AEXC_MASK (0x1fUL << FSR_AEXC_SHIFT) | ||
87 | #define FSR_CEXC_SHIFT 0UL | ||
88 | #define FSR_CEXC_MASK (0x1fUL << FSR_CEXC_SHIFT) | ||
89 | |||
90 | /* All routines returning an exception to raise should detect | ||
91 | * such exceptions _before_ rounding to be consistent with | ||
92 | * the behavior of the hardware in the implemented cases | ||
93 | * (and thus with the recommendations in the V9 architecture | ||
94 | * manual). | ||
95 | * | ||
96 | * We return 0 if a SIGFPE should be sent, 1 otherwise. | ||
97 | */ | ||
98 | static inline int record_exception(struct pt_regs *regs, int eflag) | ||
99 | { | ||
100 | u64 fsr = current_thread_info()->xfsr[0]; | ||
101 | int would_trap; | ||
102 | |||
103 | /* Determine if this exception would have generated a trap. */ | ||
104 | would_trap = (fsr & ((long)eflag << FSR_TEM_SHIFT)) != 0UL; | ||
105 | |||
106 | /* If trapping, we only want to signal one bit. */ | ||
107 | if(would_trap != 0) { | ||
108 | eflag &= ((fsr & FSR_TEM_MASK) >> FSR_TEM_SHIFT); | ||
109 | if((eflag & (eflag - 1)) != 0) { | ||
110 | if(eflag & FP_EX_INVALID) | ||
111 | eflag = FP_EX_INVALID; | ||
112 | else if(eflag & FP_EX_OVERFLOW) | ||
113 | eflag = FP_EX_OVERFLOW; | ||
114 | else if(eflag & FP_EX_UNDERFLOW) | ||
115 | eflag = FP_EX_UNDERFLOW; | ||
116 | else if(eflag & FP_EX_DIVZERO) | ||
117 | eflag = FP_EX_DIVZERO; | ||
118 | else if(eflag & FP_EX_INEXACT) | ||
119 | eflag = FP_EX_INEXACT; | ||
120 | } | ||
121 | } | ||
122 | |||
123 | /* Set CEXC, here is the rule: | ||
124 | * | ||
125 | * In general all FPU ops will set one and only one | ||
126 | * bit in the CEXC field, this is always the case | ||
127 | * when the IEEE exception trap is enabled in TEM. | ||
128 | */ | ||
129 | fsr &= ~(FSR_CEXC_MASK); | ||
130 | fsr |= ((long)eflag << FSR_CEXC_SHIFT); | ||
131 | |||
132 | /* Set the AEXC field, rule is: | ||
133 | * | ||
134 | * If a trap would not be generated, the | ||
135 | * CEXC just generated is OR'd into the | ||
136 | * existing value of AEXC. | ||
137 | */ | ||
138 | if(would_trap == 0) | ||
139 | fsr |= ((long)eflag << FSR_AEXC_SHIFT); | ||
140 | |||
141 | /* If trapping, indicate fault trap type IEEE. */ | ||
142 | if(would_trap != 0) | ||
143 | fsr |= (1UL << 14); | ||
144 | |||
145 | current_thread_info()->xfsr[0] = fsr; | ||
146 | |||
147 | /* If we will not trap, advance the program counter over | ||
148 | * the instruction being handled. | ||
149 | */ | ||
150 | if(would_trap == 0) { | ||
151 | regs->tpc = regs->tnpc; | ||
152 | regs->tnpc += 4; | ||
153 | } | ||
154 | |||
155 | return (would_trap ? 0 : 1); | ||
156 | } | ||
157 | |||
158 | typedef union { | ||
159 | u32 s; | ||
160 | u64 d; | ||
161 | u64 q[2]; | ||
162 | } *argp; | ||
163 | |||
164 | int do_mathemu(struct pt_regs *regs, struct fpustate *f) | ||
165 | { | ||
166 | unsigned long pc = regs->tpc; | ||
167 | unsigned long tstate = regs->tstate; | ||
168 | u32 insn = 0; | ||
169 | int type = 0; | ||
170 | /* ftt tells which ftt it may happen in, r is rd, b is rs2 and a is rs1. The *u arg tells | ||
171 | whether the argument should be packed/unpacked (0 - do not unpack/pack, 1 - unpack/pack) | ||
172 | non-u args tells the size of the argument (0 - no argument, 1 - single, 2 - double, 3 - quad */ | ||
173 | #define TYPE(ftt, r, ru, b, bu, a, au) type = (au << 2) | (a << 0) | (bu << 5) | (b << 3) | (ru << 8) | (r << 6) | (ftt << 9) | ||
174 | int freg; | ||
175 | static u64 zero[2] = { 0L, 0L }; | ||
176 | int flags; | ||
177 | FP_DECL_EX; | ||
178 | FP_DECL_S(SA); FP_DECL_S(SB); FP_DECL_S(SR); | ||
179 | FP_DECL_D(DA); FP_DECL_D(DB); FP_DECL_D(DR); | ||
180 | FP_DECL_Q(QA); FP_DECL_Q(QB); FP_DECL_Q(QR); | ||
181 | int IR; | ||
182 | long XR, xfsr; | ||
183 | |||
184 | if (tstate & TSTATE_PRIV) | ||
185 | die_if_kernel("unfinished/unimplemented FPop from kernel", regs); | ||
186 | if (test_thread_flag(TIF_32BIT)) | ||
187 | pc = (u32)pc; | ||
188 | if (get_user(insn, (u32 __user *) pc) != -EFAULT) { | ||
189 | if ((insn & 0xc1f80000) == 0x81a00000) /* FPOP1 */ { | ||
190 | switch ((insn >> 5) & 0x1ff) { | ||
191 | /* QUAD - ftt == 3 */ | ||
192 | case FMOVQ: | ||
193 | case FNEGQ: | ||
194 | case FABSQ: TYPE(3,3,0,3,0,0,0); break; | ||
195 | case FSQRTQ: TYPE(3,3,1,3,1,0,0); break; | ||
196 | case FADDQ: | ||
197 | case FSUBQ: | ||
198 | case FMULQ: | ||
199 | case FDIVQ: TYPE(3,3,1,3,1,3,1); break; | ||
200 | case FDMULQ: TYPE(3,3,1,2,1,2,1); break; | ||
201 | case FQTOX: TYPE(3,2,0,3,1,0,0); break; | ||
202 | case FXTOQ: TYPE(3,3,1,2,0,0,0); break; | ||
203 | case FQTOS: TYPE(3,1,1,3,1,0,0); break; | ||
204 | case FQTOD: TYPE(3,2,1,3,1,0,0); break; | ||
205 | case FITOQ: TYPE(3,3,1,1,0,0,0); break; | ||
206 | case FSTOQ: TYPE(3,3,1,1,1,0,0); break; | ||
207 | case FDTOQ: TYPE(3,3,1,2,1,0,0); break; | ||
208 | case FQTOI: TYPE(3,1,0,3,1,0,0); break; | ||
209 | |||
210 | /* We can get either unimplemented or unfinished | ||
211 | * for these cases. Pre-Niagara systems generate | ||
212 | * unfinished fpop for SUBNORMAL cases, and Niagara | ||
213 | * always gives unimplemented fpop for fsqrt{s,d}. | ||
214 | */ | ||
215 | case FSQRTS: { | ||
216 | unsigned long x = current_thread_info()->xfsr[0]; | ||
217 | |||
218 | x = (x >> 14) & 0xf; | ||
219 | TYPE(x,1,1,1,1,0,0); | ||
220 | break; | ||
221 | } | ||
222 | |||
223 | case FSQRTD: { | ||
224 | unsigned long x = current_thread_info()->xfsr[0]; | ||
225 | |||
226 | x = (x >> 14) & 0xf; | ||
227 | TYPE(x,2,1,2,1,0,0); | ||
228 | break; | ||
229 | } | ||
230 | |||
231 | /* SUBNORMAL - ftt == 2 */ | ||
232 | case FADDD: | ||
233 | case FSUBD: | ||
234 | case FMULD: | ||
235 | case FDIVD: TYPE(2,2,1,2,1,2,1); break; | ||
236 | case FADDS: | ||
237 | case FSUBS: | ||
238 | case FMULS: | ||
239 | case FDIVS: TYPE(2,1,1,1,1,1,1); break; | ||
240 | case FSMULD: TYPE(2,2,1,1,1,1,1); break; | ||
241 | case FSTOX: TYPE(2,2,0,1,1,0,0); break; | ||
242 | case FDTOX: TYPE(2,2,0,2,1,0,0); break; | ||
243 | case FDTOS: TYPE(2,1,1,2,1,0,0); break; | ||
244 | case FSTOD: TYPE(2,2,1,1,1,0,0); break; | ||
245 | case FSTOI: TYPE(2,1,0,1,1,0,0); break; | ||
246 | case FDTOI: TYPE(2,1,0,2,1,0,0); break; | ||
247 | |||
248 | /* Only Ultra-III generates these */ | ||
249 | case FXTOS: TYPE(2,1,1,2,0,0,0); break; | ||
250 | case FXTOD: TYPE(2,2,1,2,0,0,0); break; | ||
251 | #if 0 /* Optimized inline in sparc64/kernel/entry.S */ | ||
252 | case FITOS: TYPE(2,1,1,1,0,0,0); break; | ||
253 | #endif | ||
254 | case FITOD: TYPE(2,2,1,1,0,0,0); break; | ||
255 | } | ||
256 | } | ||
257 | else if ((insn & 0xc1f80000) == 0x81a80000) /* FPOP2 */ { | ||
258 | IR = 2; | ||
259 | switch ((insn >> 5) & 0x1ff) { | ||
260 | case FCMPQ: TYPE(3,0,0,3,1,3,1); break; | ||
261 | case FCMPEQ: TYPE(3,0,0,3,1,3,1); break; | ||
262 | /* Now the conditional fmovq support */ | ||
263 | case FMOVQ0: | ||
264 | case FMOVQ1: | ||
265 | case FMOVQ2: | ||
266 | case FMOVQ3: | ||
267 | /* fmovq %fccX, %fY, %fZ */ | ||
268 | if (!((insn >> 11) & 3)) | ||
269 | XR = current_thread_info()->xfsr[0] >> 10; | ||
270 | else | ||
271 | XR = current_thread_info()->xfsr[0] >> (30 + ((insn >> 10) & 0x6)); | ||
272 | XR &= 3; | ||
273 | IR = 0; | ||
274 | switch ((insn >> 14) & 0x7) { | ||
275 | /* case 0: IR = 0; break; */ /* Never */ | ||
276 | case 1: if (XR) IR = 1; break; /* Not Equal */ | ||
277 | case 2: if (XR == 1 || XR == 2) IR = 1; break; /* Less or Greater */ | ||
278 | case 3: if (XR & 1) IR = 1; break; /* Unordered or Less */ | ||
279 | case 4: if (XR == 1) IR = 1; break; /* Less */ | ||
280 | case 5: if (XR & 2) IR = 1; break; /* Unordered or Greater */ | ||
281 | case 6: if (XR == 2) IR = 1; break; /* Greater */ | ||
282 | case 7: if (XR == 3) IR = 1; break; /* Unordered */ | ||
283 | } | ||
284 | if ((insn >> 14) & 8) | ||
285 | IR ^= 1; | ||
286 | break; | ||
287 | case FMOVQI: | ||
288 | case FMOVQX: | ||
289 | /* fmovq %[ix]cc, %fY, %fZ */ | ||
290 | XR = regs->tstate >> 32; | ||
291 | if ((insn >> 5) & 0x80) | ||
292 | XR >>= 4; | ||
293 | XR &= 0xf; | ||
294 | IR = 0; | ||
295 | freg = ((XR >> 2) ^ XR) & 2; | ||
296 | switch ((insn >> 14) & 0x7) { | ||
297 | /* case 0: IR = 0; break; */ /* Never */ | ||
298 | case 1: if (XR & 4) IR = 1; break; /* Equal */ | ||
299 | case 2: if ((XR & 4) || freg) IR = 1; break; /* Less or Equal */ | ||
300 | case 3: if (freg) IR = 1; break; /* Less */ | ||
301 | case 4: if (XR & 5) IR = 1; break; /* Less or Equal Unsigned */ | ||
302 | case 5: if (XR & 1) IR = 1; break; /* Carry Set */ | ||
303 | case 6: if (XR & 8) IR = 1; break; /* Negative */ | ||
304 | case 7: if (XR & 2) IR = 1; break; /* Overflow Set */ | ||
305 | } | ||
306 | if ((insn >> 14) & 8) | ||
307 | IR ^= 1; | ||
308 | break; | ||
309 | case FMOVQZ: | ||
310 | case FMOVQLE: | ||
311 | case FMOVQLZ: | ||
312 | case FMOVQNZ: | ||
313 | case FMOVQGZ: | ||
314 | case FMOVQGE: | ||
315 | freg = (insn >> 14) & 0x1f; | ||
316 | if (!freg) | ||
317 | XR = 0; | ||
318 | else if (freg < 16) | ||
319 | XR = regs->u_regs[freg]; | ||
320 | else if (test_thread_flag(TIF_32BIT)) { | ||
321 | struct reg_window32 __user *win32; | ||
322 | flushw_user (); | ||
323 | win32 = (struct reg_window32 __user *)((unsigned long)((u32)regs->u_regs[UREG_FP])); | ||
324 | get_user(XR, &win32->locals[freg - 16]); | ||
325 | } else { | ||
326 | struct reg_window __user *win; | ||
327 | flushw_user (); | ||
328 | win = (struct reg_window __user *)(regs->u_regs[UREG_FP] + STACK_BIAS); | ||
329 | get_user(XR, &win->locals[freg - 16]); | ||
330 | } | ||
331 | IR = 0; | ||
332 | switch ((insn >> 10) & 3) { | ||
333 | case 1: if (!XR) IR = 1; break; /* Register Zero */ | ||
334 | case 2: if (XR <= 0) IR = 1; break; /* Register Less Than or Equal to Zero */ | ||
335 | case 3: if (XR < 0) IR = 1; break; /* Register Less Than Zero */ | ||
336 | } | ||
337 | if ((insn >> 10) & 4) | ||
338 | IR ^= 1; | ||
339 | break; | ||
340 | } | ||
341 | if (IR == 0) { | ||
342 | /* The fmov test was false. Do a nop instead */ | ||
343 | current_thread_info()->xfsr[0] &= ~(FSR_CEXC_MASK); | ||
344 | regs->tpc = regs->tnpc; | ||
345 | regs->tnpc += 4; | ||
346 | return 1; | ||
347 | } else if (IR == 1) { | ||
348 | /* Change the instruction into plain fmovq */ | ||
349 | insn = (insn & 0x3e00001f) | 0x81a00060; | ||
350 | TYPE(3,3,0,3,0,0,0); | ||
351 | } | ||
352 | } | ||
353 | } | ||
354 | if (type) { | ||
355 | argp rs1 = NULL, rs2 = NULL, rd = NULL; | ||
356 | |||
357 | freg = (current_thread_info()->xfsr[0] >> 14) & 0xf; | ||
358 | if (freg != (type >> 9)) | ||
359 | goto err; | ||
360 | current_thread_info()->xfsr[0] &= ~0x1c000; | ||
361 | freg = ((insn >> 14) & 0x1f); | ||
362 | switch (type & 0x3) { | ||
363 | case 3: if (freg & 2) { | ||
364 | current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */; | ||
365 | goto err; | ||
366 | } | ||
367 | case 2: freg = ((freg & 1) << 5) | (freg & 0x1e); | ||
368 | case 1: rs1 = (argp)&f->regs[freg]; | ||
369 | flags = (freg < 32) ? FPRS_DL : FPRS_DU; | ||
370 | if (!(current_thread_info()->fpsaved[0] & flags)) | ||
371 | rs1 = (argp)&zero; | ||
372 | break; | ||
373 | } | ||
374 | switch (type & 0x7) { | ||
375 | case 7: FP_UNPACK_QP (QA, rs1); break; | ||
376 | case 6: FP_UNPACK_DP (DA, rs1); break; | ||
377 | case 5: FP_UNPACK_SP (SA, rs1); break; | ||
378 | } | ||
379 | freg = (insn & 0x1f); | ||
380 | switch ((type >> 3) & 0x3) { | ||
381 | case 3: if (freg & 2) { | ||
382 | current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */; | ||
383 | goto err; | ||
384 | } | ||
385 | case 2: freg = ((freg & 1) << 5) | (freg & 0x1e); | ||
386 | case 1: rs2 = (argp)&f->regs[freg]; | ||
387 | flags = (freg < 32) ? FPRS_DL : FPRS_DU; | ||
388 | if (!(current_thread_info()->fpsaved[0] & flags)) | ||
389 | rs2 = (argp)&zero; | ||
390 | break; | ||
391 | } | ||
392 | switch ((type >> 3) & 0x7) { | ||
393 | case 7: FP_UNPACK_QP (QB, rs2); break; | ||
394 | case 6: FP_UNPACK_DP (DB, rs2); break; | ||
395 | case 5: FP_UNPACK_SP (SB, rs2); break; | ||
396 | } | ||
397 | freg = ((insn >> 25) & 0x1f); | ||
398 | switch ((type >> 6) & 0x3) { | ||
399 | case 3: if (freg & 2) { | ||
400 | current_thread_info()->xfsr[0] |= (6 << 14) /* invalid_fp_register */; | ||
401 | goto err; | ||
402 | } | ||
403 | case 2: freg = ((freg & 1) << 5) | (freg & 0x1e); | ||
404 | case 1: rd = (argp)&f->regs[freg]; | ||
405 | flags = (freg < 32) ? FPRS_DL : FPRS_DU; | ||
406 | if (!(current_thread_info()->fpsaved[0] & FPRS_FEF)) { | ||
407 | current_thread_info()->fpsaved[0] = FPRS_FEF; | ||
408 | current_thread_info()->gsr[0] = 0; | ||
409 | } | ||
410 | if (!(current_thread_info()->fpsaved[0] & flags)) { | ||
411 | if (freg < 32) | ||
412 | memset(f->regs, 0, 32*sizeof(u32)); | ||
413 | else | ||
414 | memset(f->regs+32, 0, 32*sizeof(u32)); | ||
415 | } | ||
416 | current_thread_info()->fpsaved[0] |= flags; | ||
417 | break; | ||
418 | } | ||
419 | switch ((insn >> 5) & 0x1ff) { | ||
420 | /* + */ | ||
421 | case FADDS: FP_ADD_S (SR, SA, SB); break; | ||
422 | case FADDD: FP_ADD_D (DR, DA, DB); break; | ||
423 | case FADDQ: FP_ADD_Q (QR, QA, QB); break; | ||
424 | /* - */ | ||
425 | case FSUBS: FP_SUB_S (SR, SA, SB); break; | ||
426 | case FSUBD: FP_SUB_D (DR, DA, DB); break; | ||
427 | case FSUBQ: FP_SUB_Q (QR, QA, QB); break; | ||
428 | /* * */ | ||
429 | case FMULS: FP_MUL_S (SR, SA, SB); break; | ||
430 | case FSMULD: FP_CONV (D, S, 1, 1, DA, SA); | ||
431 | FP_CONV (D, S, 1, 1, DB, SB); | ||
432 | case FMULD: FP_MUL_D (DR, DA, DB); break; | ||
433 | case FDMULQ: FP_CONV (Q, D, 2, 1, QA, DA); | ||
434 | FP_CONV (Q, D, 2, 1, QB, DB); | ||
435 | case FMULQ: FP_MUL_Q (QR, QA, QB); break; | ||
436 | /* / */ | ||
437 | case FDIVS: FP_DIV_S (SR, SA, SB); break; | ||
438 | case FDIVD: FP_DIV_D (DR, DA, DB); break; | ||
439 | case FDIVQ: FP_DIV_Q (QR, QA, QB); break; | ||
440 | /* sqrt */ | ||
441 | case FSQRTS: FP_SQRT_S (SR, SB); break; | ||
442 | case FSQRTD: FP_SQRT_D (DR, DB); break; | ||
443 | case FSQRTQ: FP_SQRT_Q (QR, QB); break; | ||
444 | /* mov */ | ||
445 | case FMOVQ: rd->q[0] = rs2->q[0]; rd->q[1] = rs2->q[1]; break; | ||
446 | case FABSQ: rd->q[0] = rs2->q[0] & 0x7fffffffffffffffUL; rd->q[1] = rs2->q[1]; break; | ||
447 | case FNEGQ: rd->q[0] = rs2->q[0] ^ 0x8000000000000000UL; rd->q[1] = rs2->q[1]; break; | ||
448 | /* float to int */ | ||
449 | case FSTOI: FP_TO_INT_S (IR, SB, 32, 1); break; | ||
450 | case FDTOI: FP_TO_INT_D (IR, DB, 32, 1); break; | ||
451 | case FQTOI: FP_TO_INT_Q (IR, QB, 32, 1); break; | ||
452 | case FSTOX: FP_TO_INT_S (XR, SB, 64, 1); break; | ||
453 | case FDTOX: FP_TO_INT_D (XR, DB, 64, 1); break; | ||
454 | case FQTOX: FP_TO_INT_Q (XR, QB, 64, 1); break; | ||
455 | /* int to float */ | ||
456 | case FITOQ: IR = rs2->s; FP_FROM_INT_Q (QR, IR, 32, int); break; | ||
457 | case FXTOQ: XR = rs2->d; FP_FROM_INT_Q (QR, XR, 64, long); break; | ||
458 | /* Only Ultra-III generates these */ | ||
459 | case FXTOS: XR = rs2->d; FP_FROM_INT_S (SR, XR, 64, long); break; | ||
460 | case FXTOD: XR = rs2->d; FP_FROM_INT_D (DR, XR, 64, long); break; | ||
461 | #if 0 /* Optimized inline in sparc64/kernel/entry.S */ | ||
462 | case FITOS: IR = rs2->s; FP_FROM_INT_S (SR, IR, 32, int); break; | ||
463 | #endif | ||
464 | case FITOD: IR = rs2->s; FP_FROM_INT_D (DR, IR, 32, int); break; | ||
465 | /* float to float */ | ||
466 | case FSTOD: FP_CONV (D, S, 1, 1, DR, SB); break; | ||
467 | case FSTOQ: FP_CONV (Q, S, 2, 1, QR, SB); break; | ||
468 | case FDTOQ: FP_CONV (Q, D, 2, 1, QR, DB); break; | ||
469 | case FDTOS: FP_CONV (S, D, 1, 1, SR, DB); break; | ||
470 | case FQTOS: FP_CONV (S, Q, 1, 2, SR, QB); break; | ||
471 | case FQTOD: FP_CONV (D, Q, 1, 2, DR, QB); break; | ||
472 | /* comparison */ | ||
473 | case FCMPQ: | ||
474 | case FCMPEQ: | ||
475 | FP_CMP_Q(XR, QB, QA, 3); | ||
476 | if (XR == 3 && | ||
477 | (((insn >> 5) & 0x1ff) == FCMPEQ || | ||
478 | FP_ISSIGNAN_Q(QA) || | ||
479 | FP_ISSIGNAN_Q(QB))) | ||
480 | FP_SET_EXCEPTION (FP_EX_INVALID); | ||
481 | } | ||
482 | if (!FP_INHIBIT_RESULTS) { | ||
483 | switch ((type >> 6) & 0x7) { | ||
484 | case 0: xfsr = current_thread_info()->xfsr[0]; | ||
485 | if (XR == -1) XR = 2; | ||
486 | switch (freg & 3) { | ||
487 | /* fcc0, 1, 2, 3 */ | ||
488 | case 0: xfsr &= ~0xc00; xfsr |= (XR << 10); break; | ||
489 | case 1: xfsr &= ~0x300000000UL; xfsr |= (XR << 32); break; | ||
490 | case 2: xfsr &= ~0xc00000000UL; xfsr |= (XR << 34); break; | ||
491 | case 3: xfsr &= ~0x3000000000UL; xfsr |= (XR << 36); break; | ||
492 | } | ||
493 | current_thread_info()->xfsr[0] = xfsr; | ||
494 | break; | ||
495 | case 1: rd->s = IR; break; | ||
496 | case 2: rd->d = XR; break; | ||
497 | case 5: FP_PACK_SP (rd, SR); break; | ||
498 | case 6: FP_PACK_DP (rd, DR); break; | ||
499 | case 7: FP_PACK_QP (rd, QR); break; | ||
500 | } | ||
501 | } | ||
502 | |||
503 | if(_fex != 0) | ||
504 | return record_exception(regs, _fex); | ||
505 | |||
506 | /* Success and no exceptions detected. */ | ||
507 | current_thread_info()->xfsr[0] &= ~(FSR_CEXC_MASK); | ||
508 | regs->tpc = regs->tnpc; | ||
509 | regs->tnpc += 4; | ||
510 | return 1; | ||
511 | } | ||
512 | err: return 0; | ||
513 | } | ||
diff --git a/arch/sparc/math-emu/sfp-util_64.h b/arch/sparc/math-emu/sfp-util_64.h new file mode 100644 index 000000000000..425d3cf01af4 --- /dev/null +++ b/arch/sparc/math-emu/sfp-util_64.h | |||
@@ -0,0 +1,120 @@ | |||
1 | /* | ||
2 | * arch/sparc64/math-emu/sfp-util.h | ||
3 | * | ||
4 | * Copyright (C) 1999 Jakub Jelinek (jj@ultra.linux.cz) | ||
5 | * Copyright (C) 1999 David S. Miller (davem@redhat.com) | ||
6 | * | ||
7 | */ | ||
8 | |||
9 | #include <linux/kernel.h> | ||
10 | #include <linux/sched.h> | ||
11 | #include <linux/types.h> | ||
12 | #include <asm/byteorder.h> | ||
13 | |||
14 | #define add_ssaaaa(sh, sl, ah, al, bh, bl) \ | ||
15 | __asm__ ("addcc %4,%5,%1\n\t" \ | ||
16 | "add %2,%3,%0\n\t" \ | ||
17 | "bcs,a,pn %%xcc, 1f\n\t" \ | ||
18 | "add %0, 1, %0\n" \ | ||
19 | "1:" \ | ||
20 | : "=r" ((UDItype)(sh)), \ | ||
21 | "=&r" ((UDItype)(sl)) \ | ||
22 | : "r" ((UDItype)(ah)), \ | ||
23 | "r" ((UDItype)(bh)), \ | ||
24 | "r" ((UDItype)(al)), \ | ||
25 | "r" ((UDItype)(bl)) \ | ||
26 | : "cc") | ||
27 | |||
28 | #define sub_ddmmss(sh, sl, ah, al, bh, bl) \ | ||
29 | __asm__ ("subcc %4,%5,%1\n\t" \ | ||
30 | "sub %2,%3,%0\n\t" \ | ||
31 | "bcs,a,pn %%xcc, 1f\n\t" \ | ||
32 | "sub %0, 1, %0\n" \ | ||
33 | "1:" \ | ||
34 | : "=r" ((UDItype)(sh)), \ | ||
35 | "=&r" ((UDItype)(sl)) \ | ||
36 | : "r" ((UDItype)(ah)), \ | ||
37 | "r" ((UDItype)(bh)), \ | ||
38 | "r" ((UDItype)(al)), \ | ||
39 | "r" ((UDItype)(bl)) \ | ||
40 | : "cc") | ||
41 | |||
42 | #define umul_ppmm(wh, wl, u, v) \ | ||
43 | do { \ | ||
44 | UDItype tmp1, tmp2, tmp3, tmp4; \ | ||
45 | __asm__ __volatile__ ( \ | ||
46 | "srl %7,0,%3\n\t" \ | ||
47 | "mulx %3,%6,%1\n\t" \ | ||
48 | "srlx %6,32,%2\n\t" \ | ||
49 | "mulx %2,%3,%4\n\t" \ | ||
50 | "sllx %4,32,%5\n\t" \ | ||
51 | "srl %6,0,%3\n\t" \ | ||
52 | "sub %1,%5,%5\n\t" \ | ||
53 | "srlx %5,32,%5\n\t" \ | ||
54 | "addcc %4,%5,%4\n\t" \ | ||
55 | "srlx %7,32,%5\n\t" \ | ||
56 | "mulx %3,%5,%3\n\t" \ | ||
57 | "mulx %2,%5,%5\n\t" \ | ||
58 | "sethi %%hi(0x80000000),%2\n\t" \ | ||
59 | "addcc %4,%3,%4\n\t" \ | ||
60 | "srlx %4,32,%4\n\t" \ | ||
61 | "add %2,%2,%2\n\t" \ | ||
62 | "movcc %%xcc,%%g0,%2\n\t" \ | ||
63 | "addcc %5,%4,%5\n\t" \ | ||
64 | "sllx %3,32,%3\n\t" \ | ||
65 | "add %1,%3,%1\n\t" \ | ||
66 | "add %5,%2,%0" \ | ||
67 | : "=r" ((UDItype)(wh)), \ | ||
68 | "=&r" ((UDItype)(wl)), \ | ||
69 | "=&r" (tmp1), "=&r" (tmp2), "=&r" (tmp3), "=&r" (tmp4) \ | ||
70 | : "r" ((UDItype)(u)), \ | ||
71 | "r" ((UDItype)(v)) \ | ||
72 | : "cc"); \ | ||
73 | } while (0) | ||
74 | |||
75 | #define udiv_qrnnd(q, r, n1, n0, d) \ | ||
76 | do { \ | ||
77 | UWtype __d1, __d0, __q1, __q0, __r1, __r0, __m; \ | ||
78 | __d1 = (d >> 32); \ | ||
79 | __d0 = (USItype)d; \ | ||
80 | \ | ||
81 | __r1 = (n1) % __d1; \ | ||
82 | __q1 = (n1) / __d1; \ | ||
83 | __m = (UWtype) __q1 * __d0; \ | ||
84 | __r1 = (__r1 << 32) | (n0 >> 32); \ | ||
85 | if (__r1 < __m) \ | ||
86 | { \ | ||
87 | __q1--, __r1 += (d); \ | ||
88 | if (__r1 >= (d)) /* i.e. we didn't get carry when adding to __r1 */ \ | ||
89 | if (__r1 < __m) \ | ||
90 | __q1--, __r1 += (d); \ | ||
91 | } \ | ||
92 | __r1 -= __m; \ | ||
93 | \ | ||
94 | __r0 = __r1 % __d1; \ | ||
95 | __q0 = __r1 / __d1; \ | ||
96 | __m = (UWtype) __q0 * __d0; \ | ||
97 | __r0 = (__r0 << 32) | ((USItype)n0); \ | ||
98 | if (__r0 < __m) \ | ||
99 | { \ | ||
100 | __q0--, __r0 += (d); \ | ||
101 | if (__r0 >= (d)) \ | ||
102 | if (__r0 < __m) \ | ||
103 | __q0--, __r0 += (d); \ | ||
104 | } \ | ||
105 | __r0 -= __m; \ | ||
106 | \ | ||
107 | (q) = (UWtype) (__q1 << 32) | __q0; \ | ||
108 | (r) = __r0; \ | ||
109 | } while (0) | ||
110 | |||
111 | #define UDIV_NEEDS_NORMALIZATION 1 | ||
112 | |||
113 | #define abort() \ | ||
114 | return 0 | ||
115 | |||
116 | #ifdef __BIG_ENDIAN | ||
117 | #define __BYTE_ORDER __BIG_ENDIAN | ||
118 | #else | ||
119 | #define __BYTE_ORDER __LITTLE_ENDIAN | ||
120 | #endif | ||