xref: /openbmc/linux/arch/sparc/kernel/visemul.c (revision 8fdff1dc)
1 /* visemul.c: Emulation of VIS instructions.
2  *
3  * Copyright (C) 2006 David S. Miller (davem@davemloft.net)
4  */
5 #include <linux/kernel.h>
6 #include <linux/errno.h>
7 #include <linux/thread_info.h>
8 #include <linux/perf_event.h>
9 
10 #include <asm/ptrace.h>
11 #include <asm/pstate.h>
12 #include <asm/fpumacro.h>
13 #include <asm/uaccess.h>
14 #include <asm/cacheflush.h>
15 
16 /* OPF field of various VIS instructions.  */
17 
18 /* 000111011 - four 16-bit packs  */
19 #define FPACK16_OPF	0x03b
20 
21 /* 000111010 - two 32-bit packs  */
22 #define FPACK32_OPF	0x03a
23 
24 /* 000111101 - four 16-bit packs  */
25 #define FPACKFIX_OPF	0x03d
26 
27 /* 001001101 - four 16-bit expands  */
28 #define FEXPAND_OPF	0x04d
29 
30 /* 001001011 - two 32-bit merges */
31 #define FPMERGE_OPF	0x04b
32 
33 /* 000110001 - 8-by-16-bit partitoned product  */
34 #define FMUL8x16_OPF	0x031
35 
36 /* 000110011 - 8-by-16-bit upper alpha partitioned product  */
37 #define FMUL8x16AU_OPF	0x033
38 
39 /* 000110101 - 8-by-16-bit lower alpha partitioned product  */
40 #define FMUL8x16AL_OPF	0x035
41 
42 /* 000110110 - upper 8-by-16-bit partitioned product  */
43 #define FMUL8SUx16_OPF	0x036
44 
45 /* 000110111 - lower 8-by-16-bit partitioned product  */
46 #define FMUL8ULx16_OPF	0x037
47 
48 /* 000111000 - upper 8-by-16-bit partitioned product  */
49 #define FMULD8SUx16_OPF	0x038
50 
51 /* 000111001 - lower unsigned 8-by-16-bit partitioned product  */
52 #define FMULD8ULx16_OPF	0x039
53 
54 /* 000101000 - four 16-bit compare; set rd if src1 > src2  */
55 #define FCMPGT16_OPF	0x028
56 
57 /* 000101100 - two 32-bit compare; set rd if src1 > src2  */
58 #define FCMPGT32_OPF	0x02c
59 
60 /* 000100000 - four 16-bit compare; set rd if src1 <= src2  */
61 #define FCMPLE16_OPF	0x020
62 
63 /* 000100100 - two 32-bit compare; set rd if src1 <= src2  */
64 #define FCMPLE32_OPF	0x024
65 
66 /* 000100010 - four 16-bit compare; set rd if src1 != src2  */
67 #define FCMPNE16_OPF	0x022
68 
69 /* 000100110 - two 32-bit compare; set rd if src1 != src2  */
70 #define FCMPNE32_OPF	0x026
71 
72 /* 000101010 - four 16-bit compare; set rd if src1 == src2  */
73 #define FCMPEQ16_OPF	0x02a
74 
75 /* 000101110 - two 32-bit compare; set rd if src1 == src2  */
76 #define FCMPEQ32_OPF	0x02e
77 
78 /* 000000000 - Eight 8-bit edge boundary processing  */
79 #define EDGE8_OPF	0x000
80 
81 /* 000000001 - Eight 8-bit edge boundary processing, no CC */
82 #define EDGE8N_OPF	0x001
83 
84 /* 000000010 - Eight 8-bit edge boundary processing, little-endian  */
85 #define EDGE8L_OPF	0x002
86 
87 /* 000000011 - Eight 8-bit edge boundary processing, little-endian, no CC  */
88 #define EDGE8LN_OPF	0x003
89 
90 /* 000000100 - Four 16-bit edge boundary processing  */
91 #define EDGE16_OPF	0x004
92 
93 /* 000000101 - Four 16-bit edge boundary processing, no CC  */
94 #define EDGE16N_OPF	0x005
95 
96 /* 000000110 - Four 16-bit edge boundary processing, little-endian  */
97 #define EDGE16L_OPF	0x006
98 
99 /* 000000111 - Four 16-bit edge boundary processing, little-endian, no CC  */
100 #define EDGE16LN_OPF	0x007
101 
102 /* 000001000 - Two 32-bit edge boundary processing  */
103 #define EDGE32_OPF	0x008
104 
105 /* 000001001 - Two 32-bit edge boundary processing, no CC  */
106 #define EDGE32N_OPF	0x009
107 
108 /* 000001010 - Two 32-bit edge boundary processing, little-endian  */
109 #define EDGE32L_OPF	0x00a
110 
111 /* 000001011 - Two 32-bit edge boundary processing, little-endian, no CC  */
112 #define EDGE32LN_OPF	0x00b
113 
114 /* 000111110 - distance between 8 8-bit components  */
115 #define PDIST_OPF	0x03e
116 
117 /* 000010000 - convert 8-bit 3-D address to blocked byte address  */
118 #define ARRAY8_OPF	0x010
119 
120 /* 000010010 - convert 16-bit 3-D address to blocked byte address  */
121 #define ARRAY16_OPF	0x012
122 
123 /* 000010100 - convert 32-bit 3-D address to blocked byte address  */
124 #define ARRAY32_OPF	0x014
125 
126 /* 000011001 - Set the GSR.MASK field in preparation for a BSHUFFLE  */
127 #define BMASK_OPF	0x019
128 
129 /* 001001100 - Permute bytes as specified by GSR.MASK  */
130 #define BSHUFFLE_OPF	0x04c
131 
132 #define VIS_OPF_SHIFT	5
133 #define VIS_OPF_MASK	(0x1ff << VIS_OPF_SHIFT)
134 
135 #define RS1(INSN)	(((INSN) >> 14) & 0x1f)
136 #define RS2(INSN)	(((INSN) >>  0) & 0x1f)
137 #define RD(INSN)	(((INSN) >> 25) & 0x1f)
138 
139 static inline void maybe_flush_windows(unsigned int rs1, unsigned int rs2,
140 				       unsigned int rd, int from_kernel)
141 {
142 	if (rs2 >= 16 || rs1 >= 16 || rd >= 16) {
143 		if (from_kernel != 0)
144 			__asm__ __volatile__("flushw");
145 		else
146 			flushw_user();
147 	}
148 }
149 
150 static unsigned long fetch_reg(unsigned int reg, struct pt_regs *regs)
151 {
152 	unsigned long value, fp;
153 
154 	if (reg < 16)
155 		return (!reg ? 0 : regs->u_regs[reg]);
156 
157 	fp = regs->u_regs[UREG_FP];
158 
159 	if (regs->tstate & TSTATE_PRIV) {
160 		struct reg_window *win;
161 		win = (struct reg_window *)(fp + STACK_BIAS);
162 		value = win->locals[reg - 16];
163 	} else if (!test_thread_64bit_stack(fp)) {
164 		struct reg_window32 __user *win32;
165 		win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
166 		get_user(value, &win32->locals[reg - 16]);
167 	} else {
168 		struct reg_window __user *win;
169 		win = (struct reg_window __user *)(fp + STACK_BIAS);
170 		get_user(value, &win->locals[reg - 16]);
171 	}
172 	return value;
173 }
174 
175 static inline unsigned long __user *__fetch_reg_addr_user(unsigned int reg,
176 							  struct pt_regs *regs)
177 {
178 	unsigned long fp = regs->u_regs[UREG_FP];
179 
180 	BUG_ON(reg < 16);
181 	BUG_ON(regs->tstate & TSTATE_PRIV);
182 
183 	if (!test_thread_64bit_stack(fp)) {
184 		struct reg_window32 __user *win32;
185 		win32 = (struct reg_window32 __user *)((unsigned long)((u32)fp));
186 		return (unsigned long __user *)&win32->locals[reg - 16];
187 	} else {
188 		struct reg_window __user *win;
189 		win = (struct reg_window __user *)(fp + STACK_BIAS);
190 		return &win->locals[reg - 16];
191 	}
192 }
193 
194 static inline unsigned long *__fetch_reg_addr_kern(unsigned int reg,
195 						   struct pt_regs *regs)
196 {
197 	BUG_ON(reg >= 16);
198 	BUG_ON(regs->tstate & TSTATE_PRIV);
199 
200 	return &regs->u_regs[reg];
201 }
202 
203 static void store_reg(struct pt_regs *regs, unsigned long val, unsigned long rd)
204 {
205 	if (rd < 16) {
206 		unsigned long *rd_kern = __fetch_reg_addr_kern(rd, regs);
207 
208 		*rd_kern = val;
209 	} else {
210 		unsigned long __user *rd_user = __fetch_reg_addr_user(rd, regs);
211 
212 		if (!test_thread_64bit_stack(regs->u_regs[UREG_FP]))
213 			__put_user((u32)val, (u32 __user *)rd_user);
214 		else
215 			__put_user(val, rd_user);
216 	}
217 }
218 
219 static inline unsigned long fpd_regval(struct fpustate *f,
220 				       unsigned int insn_regnum)
221 {
222 	insn_regnum = (((insn_regnum & 1) << 5) |
223 		       (insn_regnum & 0x1e));
224 
225 	return *(unsigned long *) &f->regs[insn_regnum];
226 }
227 
228 static inline unsigned long *fpd_regaddr(struct fpustate *f,
229 					 unsigned int insn_regnum)
230 {
231 	insn_regnum = (((insn_regnum & 1) << 5) |
232 		       (insn_regnum & 0x1e));
233 
234 	return (unsigned long *) &f->regs[insn_regnum];
235 }
236 
237 static inline unsigned int fps_regval(struct fpustate *f,
238 				      unsigned int insn_regnum)
239 {
240 	return f->regs[insn_regnum];
241 }
242 
243 static inline unsigned int *fps_regaddr(struct fpustate *f,
244 					unsigned int insn_regnum)
245 {
246 	return &f->regs[insn_regnum];
247 }
248 
249 struct edge_tab {
250 	u16 left, right;
251 };
252 static struct edge_tab edge8_tab[8] = {
253 	{ 0xff, 0x80 },
254 	{ 0x7f, 0xc0 },
255 	{ 0x3f, 0xe0 },
256 	{ 0x1f, 0xf0 },
257 	{ 0x0f, 0xf8 },
258 	{ 0x07, 0xfc },
259 	{ 0x03, 0xfe },
260 	{ 0x01, 0xff },
261 };
262 static struct edge_tab edge8_tab_l[8] = {
263 	{ 0xff, 0x01 },
264 	{ 0xfe, 0x03 },
265 	{ 0xfc, 0x07 },
266 	{ 0xf8, 0x0f },
267 	{ 0xf0, 0x1f },
268 	{ 0xe0, 0x3f },
269 	{ 0xc0, 0x7f },
270 	{ 0x80, 0xff },
271 };
272 static struct edge_tab edge16_tab[4] = {
273 	{ 0xf, 0x8 },
274 	{ 0x7, 0xc },
275 	{ 0x3, 0xe },
276 	{ 0x1, 0xf },
277 };
278 static struct edge_tab edge16_tab_l[4] = {
279 	{ 0xf, 0x1 },
280 	{ 0xe, 0x3 },
281 	{ 0xc, 0x7 },
282 	{ 0x8, 0xf },
283 };
284 static struct edge_tab edge32_tab[2] = {
285 	{ 0x3, 0x2 },
286 	{ 0x1, 0x3 },
287 };
288 static struct edge_tab edge32_tab_l[2] = {
289 	{ 0x3, 0x1 },
290 	{ 0x2, 0x3 },
291 };
292 
293 static void edge(struct pt_regs *regs, unsigned int insn, unsigned int opf)
294 {
295 	unsigned long orig_rs1, rs1, orig_rs2, rs2, rd_val;
296 	u16 left, right;
297 
298 	maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
299 	orig_rs1 = rs1 = fetch_reg(RS1(insn), regs);
300 	orig_rs2 = rs2 = fetch_reg(RS2(insn), regs);
301 
302 	if (test_thread_flag(TIF_32BIT)) {
303 		rs1 = rs1 & 0xffffffff;
304 		rs2 = rs2 & 0xffffffff;
305 	}
306 	switch (opf) {
307 	default:
308 	case EDGE8_OPF:
309 	case EDGE8N_OPF:
310 		left = edge8_tab[rs1 & 0x7].left;
311 		right = edge8_tab[rs2 & 0x7].right;
312 		break;
313 	case EDGE8L_OPF:
314 	case EDGE8LN_OPF:
315 		left = edge8_tab_l[rs1 & 0x7].left;
316 		right = edge8_tab_l[rs2 & 0x7].right;
317 		break;
318 
319 	case EDGE16_OPF:
320 	case EDGE16N_OPF:
321 		left = edge16_tab[(rs1 >> 1) & 0x3].left;
322 		right = edge16_tab[(rs2 >> 1) & 0x3].right;
323 		break;
324 
325 	case EDGE16L_OPF:
326 	case EDGE16LN_OPF:
327 		left = edge16_tab_l[(rs1 >> 1) & 0x3].left;
328 		right = edge16_tab_l[(rs2 >> 1) & 0x3].right;
329 		break;
330 
331 	case EDGE32_OPF:
332 	case EDGE32N_OPF:
333 		left = edge32_tab[(rs1 >> 2) & 0x1].left;
334 		right = edge32_tab[(rs2 >> 2) & 0x1].right;
335 		break;
336 
337 	case EDGE32L_OPF:
338 	case EDGE32LN_OPF:
339 		left = edge32_tab_l[(rs1 >> 2) & 0x1].left;
340 		right = edge32_tab_l[(rs2 >> 2) & 0x1].right;
341 		break;
342 	}
343 
344 	if ((rs1 & ~0x7UL) == (rs2 & ~0x7UL))
345 		rd_val = right & left;
346 	else
347 		rd_val = left;
348 
349 	store_reg(regs, rd_val, RD(insn));
350 
351 	switch (opf) {
352 	case EDGE8_OPF:
353 	case EDGE8L_OPF:
354 	case EDGE16_OPF:
355 	case EDGE16L_OPF:
356 	case EDGE32_OPF:
357 	case EDGE32L_OPF: {
358 		unsigned long ccr, tstate;
359 
360 		__asm__ __volatile__("subcc	%1, %2, %%g0\n\t"
361 				     "rd	%%ccr, %0"
362 				     : "=r" (ccr)
363 				     : "r" (orig_rs1), "r" (orig_rs2)
364 				     : "cc");
365 		tstate = regs->tstate & ~(TSTATE_XCC | TSTATE_ICC);
366 		regs->tstate = tstate | (ccr << 32UL);
367 	}
368 	}
369 }
370 
371 static void array(struct pt_regs *regs, unsigned int insn, unsigned int opf)
372 {
373 	unsigned long rs1, rs2, rd_val;
374 	unsigned int bits, bits_mask;
375 
376 	maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
377 	rs1 = fetch_reg(RS1(insn), regs);
378 	rs2 = fetch_reg(RS2(insn), regs);
379 
380 	bits = (rs2 > 5 ? 5 : rs2);
381 	bits_mask = (1UL << bits) - 1UL;
382 
383 	rd_val = ((((rs1 >> 11) & 0x3) <<  0) |
384 		  (((rs1 >> 33) & 0x3) <<  2) |
385 		  (((rs1 >> 55) & 0x1) <<  4) |
386 		  (((rs1 >> 13) & 0xf) <<  5) |
387 		  (((rs1 >> 35) & 0xf) <<  9) |
388 		  (((rs1 >> 56) & 0xf) << 13) |
389 		  (((rs1 >> 17) & bits_mask) << 17) |
390 		  (((rs1 >> 39) & bits_mask) << (17 + bits)) |
391 		  (((rs1 >> 60) & 0xf)       << (17 + (2*bits))));
392 
393 	switch (opf) {
394 	case ARRAY16_OPF:
395 		rd_val <<= 1;
396 		break;
397 
398 	case ARRAY32_OPF:
399 		rd_val <<= 2;
400 	}
401 
402 	store_reg(regs, rd_val, RD(insn));
403 }
404 
405 static void bmask(struct pt_regs *regs, unsigned int insn)
406 {
407 	unsigned long rs1, rs2, rd_val, gsr;
408 
409 	maybe_flush_windows(RS1(insn), RS2(insn), RD(insn), 0);
410 	rs1 = fetch_reg(RS1(insn), regs);
411 	rs2 = fetch_reg(RS2(insn), regs);
412 	rd_val = rs1 + rs2;
413 
414 	store_reg(regs, rd_val, RD(insn));
415 
416 	gsr = current_thread_info()->gsr[0] & 0xffffffff;
417 	gsr |= rd_val << 32UL;
418 	current_thread_info()->gsr[0] = gsr;
419 }
420 
421 static void bshuffle(struct pt_regs *regs, unsigned int insn)
422 {
423 	struct fpustate *f = FPUSTATE;
424 	unsigned long rs1, rs2, rd_val;
425 	unsigned long bmask, i;
426 
427 	bmask = current_thread_info()->gsr[0] >> 32UL;
428 
429 	rs1 = fpd_regval(f, RS1(insn));
430 	rs2 = fpd_regval(f, RS2(insn));
431 
432 	rd_val = 0UL;
433 	for (i = 0; i < 8; i++) {
434 		unsigned long which = (bmask >> (i * 4)) & 0xf;
435 		unsigned long byte;
436 
437 		if (which < 8)
438 			byte = (rs1 >> (which * 8)) & 0xff;
439 		else
440 			byte = (rs2 >> ((which-8)*8)) & 0xff;
441 		rd_val |= (byte << (i * 8));
442 	}
443 
444 	*fpd_regaddr(f, RD(insn)) = rd_val;
445 }
446 
447 static void pdist(struct pt_regs *regs, unsigned int insn)
448 {
449 	struct fpustate *f = FPUSTATE;
450 	unsigned long rs1, rs2, *rd, rd_val;
451 	unsigned long i;
452 
453 	rs1 = fpd_regval(f, RS1(insn));
454 	rs2 = fpd_regval(f, RS2(insn));
455 	rd = fpd_regaddr(f, RD(insn));
456 
457 	rd_val = *rd;
458 
459 	for (i = 0; i < 8; i++) {
460 		s16 s1, s2;
461 
462 		s1 = (rs1 >> (56 - (i * 8))) & 0xff;
463 		s2 = (rs2 >> (56 - (i * 8))) & 0xff;
464 
465 		/* Absolute value of difference. */
466 		s1 -= s2;
467 		if (s1 < 0)
468 			s1 = ~s1 + 1;
469 
470 		rd_val += s1;
471 	}
472 
473 	*rd = rd_val;
474 }
475 
476 static void pformat(struct pt_regs *regs, unsigned int insn, unsigned int opf)
477 {
478 	struct fpustate *f = FPUSTATE;
479 	unsigned long rs1, rs2, gsr, scale, rd_val;
480 
481 	gsr = current_thread_info()->gsr[0];
482 	scale = (gsr >> 3) & (opf == FPACK16_OPF ? 0xf : 0x1f);
483 	switch (opf) {
484 	case FPACK16_OPF: {
485 		unsigned long byte;
486 
487 		rs2 = fpd_regval(f, RS2(insn));
488 		rd_val = 0;
489 		for (byte = 0; byte < 4; byte++) {
490 			unsigned int val;
491 			s16 src = (rs2 >> (byte * 16UL)) & 0xffffUL;
492 			int scaled = src << scale;
493 			int from_fixed = scaled >> 7;
494 
495 			val = ((from_fixed < 0) ?
496 			       0 :
497 			       (from_fixed > 255) ?
498 			       255 : from_fixed);
499 
500 			rd_val |= (val << (8 * byte));
501 		}
502 		*fps_regaddr(f, RD(insn)) = rd_val;
503 		break;
504 	}
505 
506 	case FPACK32_OPF: {
507 		unsigned long word;
508 
509 		rs1 = fpd_regval(f, RS1(insn));
510 		rs2 = fpd_regval(f, RS2(insn));
511 		rd_val = (rs1 << 8) & ~(0x000000ff000000ffUL);
512 		for (word = 0; word < 2; word++) {
513 			unsigned long val;
514 			s32 src = (rs2 >> (word * 32UL));
515 			s64 scaled = src << scale;
516 			s64 from_fixed = scaled >> 23;
517 
518 			val = ((from_fixed < 0) ?
519 			       0 :
520 			       (from_fixed > 255) ?
521 			       255 : from_fixed);
522 
523 			rd_val |= (val << (32 * word));
524 		}
525 		*fpd_regaddr(f, RD(insn)) = rd_val;
526 		break;
527 	}
528 
529 	case FPACKFIX_OPF: {
530 		unsigned long word;
531 
532 		rs2 = fpd_regval(f, RS2(insn));
533 
534 		rd_val = 0;
535 		for (word = 0; word < 2; word++) {
536 			long val;
537 			s32 src = (rs2 >> (word * 32UL));
538 			s64 scaled = src << scale;
539 			s64 from_fixed = scaled >> 16;
540 
541 			val = ((from_fixed < -32768) ?
542 			       -32768 :
543 			       (from_fixed > 32767) ?
544 			       32767 : from_fixed);
545 
546 			rd_val |= ((val & 0xffff) << (word * 16));
547 		}
548 		*fps_regaddr(f, RD(insn)) = rd_val;
549 		break;
550 	}
551 
552 	case FEXPAND_OPF: {
553 		unsigned long byte;
554 
555 		rs2 = fps_regval(f, RS2(insn));
556 
557 		rd_val = 0;
558 		for (byte = 0; byte < 4; byte++) {
559 			unsigned long val;
560 			u8 src = (rs2 >> (byte * 8)) & 0xff;
561 
562 			val = src << 4;
563 
564 			rd_val |= (val << (byte * 16));
565 		}
566 		*fpd_regaddr(f, RD(insn)) = rd_val;
567 		break;
568 	}
569 
570 	case FPMERGE_OPF: {
571 		rs1 = fps_regval(f, RS1(insn));
572 		rs2 = fps_regval(f, RS2(insn));
573 
574 		rd_val = (((rs2 & 0x000000ff) <<  0) |
575 			  ((rs1 & 0x000000ff) <<  8) |
576 			  ((rs2 & 0x0000ff00) <<  8) |
577 			  ((rs1 & 0x0000ff00) << 16) |
578 			  ((rs2 & 0x00ff0000) << 16) |
579 			  ((rs1 & 0x00ff0000) << 24) |
580 			  ((rs2 & 0xff000000) << 24) |
581 			  ((rs1 & 0xff000000) << 32));
582 		*fpd_regaddr(f, RD(insn)) = rd_val;
583 		break;
584 	}
585 	}
586 }
587 
588 static void pmul(struct pt_regs *regs, unsigned int insn, unsigned int opf)
589 {
590 	struct fpustate *f = FPUSTATE;
591 	unsigned long rs1, rs2, rd_val;
592 
593 	switch (opf) {
594 	case FMUL8x16_OPF: {
595 		unsigned long byte;
596 
597 		rs1 = fps_regval(f, RS1(insn));
598 		rs2 = fpd_regval(f, RS2(insn));
599 
600 		rd_val = 0;
601 		for (byte = 0; byte < 4; byte++) {
602 			u16 src1 = (rs1 >> (byte *  8)) & 0x00ff;
603 			s16 src2 = (rs2 >> (byte * 16)) & 0xffff;
604 			u32 prod = src1 * src2;
605 			u16 scaled = ((prod & 0x00ffff00) >> 8);
606 
607 			/* Round up.  */
608 			if (prod & 0x80)
609 				scaled++;
610 			rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
611 		}
612 
613 		*fpd_regaddr(f, RD(insn)) = rd_val;
614 		break;
615 	}
616 
617 	case FMUL8x16AU_OPF:
618 	case FMUL8x16AL_OPF: {
619 		unsigned long byte;
620 		s16 src2;
621 
622 		rs1 = fps_regval(f, RS1(insn));
623 		rs2 = fps_regval(f, RS2(insn));
624 
625 		rd_val = 0;
626 		src2 = rs2 >> (opf == FMUL8x16AU_OPF ? 16 : 0);
627 		for (byte = 0; byte < 4; byte++) {
628 			u16 src1 = (rs1 >> (byte * 8)) & 0x00ff;
629 			u32 prod = src1 * src2;
630 			u16 scaled = ((prod & 0x00ffff00) >> 8);
631 
632 			/* Round up.  */
633 			if (prod & 0x80)
634 				scaled++;
635 			rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
636 		}
637 
638 		*fpd_regaddr(f, RD(insn)) = rd_val;
639 		break;
640 	}
641 
642 	case FMUL8SUx16_OPF:
643 	case FMUL8ULx16_OPF: {
644 		unsigned long byte, ushift;
645 
646 		rs1 = fpd_regval(f, RS1(insn));
647 		rs2 = fpd_regval(f, RS2(insn));
648 
649 		rd_val = 0;
650 		ushift = (opf == FMUL8SUx16_OPF) ? 8 : 0;
651 		for (byte = 0; byte < 4; byte++) {
652 			u16 src1;
653 			s16 src2;
654 			u32 prod;
655 			u16 scaled;
656 
657 			src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff);
658 			src2 = ((rs2 >> (16 * byte)) & 0xffff);
659 			prod = src1 * src2;
660 			scaled = ((prod & 0x00ffff00) >> 8);
661 
662 			/* Round up.  */
663 			if (prod & 0x80)
664 				scaled++;
665 			rd_val |= ((scaled & 0xffffUL) << (byte * 16UL));
666 		}
667 
668 		*fpd_regaddr(f, RD(insn)) = rd_val;
669 		break;
670 	}
671 
672 	case FMULD8SUx16_OPF:
673 	case FMULD8ULx16_OPF: {
674 		unsigned long byte, ushift;
675 
676 		rs1 = fps_regval(f, RS1(insn));
677 		rs2 = fps_regval(f, RS2(insn));
678 
679 		rd_val = 0;
680 		ushift = (opf == FMULD8SUx16_OPF) ? 8 : 0;
681 		for (byte = 0; byte < 2; byte++) {
682 			u16 src1;
683 			s16 src2;
684 			u32 prod;
685 			u16 scaled;
686 
687 			src1 = ((rs1 >> ((16 * byte) + ushift)) & 0x00ff);
688 			src2 = ((rs2 >> (16 * byte)) & 0xffff);
689 			prod = src1 * src2;
690 			scaled = ((prod & 0x00ffff00) >> 8);
691 
692 			/* Round up.  */
693 			if (prod & 0x80)
694 				scaled++;
695 			rd_val |= ((scaled & 0xffffUL) <<
696 				   ((byte * 32UL) + 7UL));
697 		}
698 		*fpd_regaddr(f, RD(insn)) = rd_val;
699 		break;
700 	}
701 	}
702 }
703 
704 static void pcmp(struct pt_regs *regs, unsigned int insn, unsigned int opf)
705 {
706 	struct fpustate *f = FPUSTATE;
707 	unsigned long rs1, rs2, rd_val, i;
708 
709 	rs1 = fpd_regval(f, RS1(insn));
710 	rs2 = fpd_regval(f, RS2(insn));
711 
712 	rd_val = 0;
713 
714 	switch (opf) {
715 	case FCMPGT16_OPF:
716 		for (i = 0; i < 4; i++) {
717 			s16 a = (rs1 >> (i * 16)) & 0xffff;
718 			s16 b = (rs2 >> (i * 16)) & 0xffff;
719 
720 			if (a > b)
721 				rd_val |= 8 >> i;
722 		}
723 		break;
724 
725 	case FCMPGT32_OPF:
726 		for (i = 0; i < 2; i++) {
727 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
728 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
729 
730 			if (a > b)
731 				rd_val |= 2 >> i;
732 		}
733 		break;
734 
735 	case FCMPLE16_OPF:
736 		for (i = 0; i < 4; i++) {
737 			s16 a = (rs1 >> (i * 16)) & 0xffff;
738 			s16 b = (rs2 >> (i * 16)) & 0xffff;
739 
740 			if (a <= b)
741 				rd_val |= 8 >> i;
742 		}
743 		break;
744 
745 	case FCMPLE32_OPF:
746 		for (i = 0; i < 2; i++) {
747 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
748 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
749 
750 			if (a <= b)
751 				rd_val |= 2 >> i;
752 		}
753 		break;
754 
755 	case FCMPNE16_OPF:
756 		for (i = 0; i < 4; i++) {
757 			s16 a = (rs1 >> (i * 16)) & 0xffff;
758 			s16 b = (rs2 >> (i * 16)) & 0xffff;
759 
760 			if (a != b)
761 				rd_val |= 8 >> i;
762 		}
763 		break;
764 
765 	case FCMPNE32_OPF:
766 		for (i = 0; i < 2; i++) {
767 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
768 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
769 
770 			if (a != b)
771 				rd_val |= 2 >> i;
772 		}
773 		break;
774 
775 	case FCMPEQ16_OPF:
776 		for (i = 0; i < 4; i++) {
777 			s16 a = (rs1 >> (i * 16)) & 0xffff;
778 			s16 b = (rs2 >> (i * 16)) & 0xffff;
779 
780 			if (a == b)
781 				rd_val |= 8 >> i;
782 		}
783 		break;
784 
785 	case FCMPEQ32_OPF:
786 		for (i = 0; i < 2; i++) {
787 			s32 a = (rs1 >> (i * 32)) & 0xffffffff;
788 			s32 b = (rs2 >> (i * 32)) & 0xffffffff;
789 
790 			if (a == b)
791 				rd_val |= 2 >> i;
792 		}
793 		break;
794 	}
795 
796 	maybe_flush_windows(0, 0, RD(insn), 0);
797 	store_reg(regs, rd_val, RD(insn));
798 }
799 
800 /* Emulate the VIS instructions which are not implemented in
801  * hardware on Niagara.
802  */
803 int vis_emul(struct pt_regs *regs, unsigned int insn)
804 {
805 	unsigned long pc = regs->tpc;
806 	unsigned int opf;
807 
808 	BUG_ON(regs->tstate & TSTATE_PRIV);
809 
810 	perf_sw_event(PERF_COUNT_SW_EMULATION_FAULTS, 1, regs, 0);
811 
812 	if (test_thread_flag(TIF_32BIT))
813 		pc = (u32)pc;
814 
815 	if (get_user(insn, (u32 __user *) pc))
816 		return -EFAULT;
817 
818 	save_and_clear_fpu();
819 
820 	opf = (insn & VIS_OPF_MASK) >> VIS_OPF_SHIFT;
821 	switch (opf) {
822 	default:
823 		return -EINVAL;
824 
825 	/* Pixel Formatting Instructions.  */
826 	case FPACK16_OPF:
827 	case FPACK32_OPF:
828 	case FPACKFIX_OPF:
829 	case FEXPAND_OPF:
830 	case FPMERGE_OPF:
831 		pformat(regs, insn, opf);
832 		break;
833 
834 	/* Partitioned Multiply Instructions  */
835 	case FMUL8x16_OPF:
836 	case FMUL8x16AU_OPF:
837 	case FMUL8x16AL_OPF:
838 	case FMUL8SUx16_OPF:
839 	case FMUL8ULx16_OPF:
840 	case FMULD8SUx16_OPF:
841 	case FMULD8ULx16_OPF:
842 		pmul(regs, insn, opf);
843 		break;
844 
845 	/* Pixel Compare Instructions  */
846 	case FCMPGT16_OPF:
847 	case FCMPGT32_OPF:
848 	case FCMPLE16_OPF:
849 	case FCMPLE32_OPF:
850 	case FCMPNE16_OPF:
851 	case FCMPNE32_OPF:
852 	case FCMPEQ16_OPF:
853 	case FCMPEQ32_OPF:
854 		pcmp(regs, insn, opf);
855 		break;
856 
857 	/* Edge Handling Instructions  */
858 	case EDGE8_OPF:
859 	case EDGE8N_OPF:
860 	case EDGE8L_OPF:
861 	case EDGE8LN_OPF:
862 	case EDGE16_OPF:
863 	case EDGE16N_OPF:
864 	case EDGE16L_OPF:
865 	case EDGE16LN_OPF:
866 	case EDGE32_OPF:
867 	case EDGE32N_OPF:
868 	case EDGE32L_OPF:
869 	case EDGE32LN_OPF:
870 		edge(regs, insn, opf);
871 		break;
872 
873 	/* Pixel Component Distance  */
874 	case PDIST_OPF:
875 		pdist(regs, insn);
876 		break;
877 
878 	/* Three-Dimensional Array Addressing Instructions  */
879 	case ARRAY8_OPF:
880 	case ARRAY16_OPF:
881 	case ARRAY32_OPF:
882 		array(regs, insn, opf);
883 		break;
884 
885 	/* Byte Mask and Shuffle Instructions  */
886 	case BMASK_OPF:
887 		bmask(regs, insn);
888 		break;
889 
890 	case BSHUFFLE_OPF:
891 		bshuffle(regs, insn);
892 		break;
893 	}
894 
895 	regs->tpc = regs->tnpc;
896 	regs->tnpc += 4;
897 	return 0;
898 }
899