Commit | Line | Data |
---|---|---|
90c4eee3 MW |
1 | /// -*- mode: asm; asm-comment-char: 0 -*- |
2 | ||
3 | ///-------------------------------------------------------------------------- | |
4 | /// Preliminaries. | |
5 | ||
6 | #include <sys/syscall.h> | |
7 | ||
8 | #if defined(__i386__) || defined(__x86_64__) | |
06297a93 MW |
9 | |
10 | .intel_syntax noprefix | |
11 | ||
90c4eee3 MW |
12 | #elif defined(__arm__) |
13 | ||
14 | .macro ret | |
15 | bx r14 | |
16 | .endm | |
17 | ||
18 | .arch armv7-a | |
19 | ||
20 | #elif defined(__aarch64__) | |
21 | ||
22 | .macro cmov rd, rn, cc | |
23 | csel \rd, \rn, \rd, \cc | |
24 | .endm | |
25 | #define _COND(_) \ | |
26 | _(eq) _(ne) _(cs) _(cc) _(vs) _(vc) _(mi) _(pl) \ | |
27 | _(ge) _(lt) _(gt) _(le) _(hi) _(ls) _(al) _(nv) \ | |
28 | _(hs) _(lo) | |
29 | #define _INST(_) \ | |
30 | _(ccmp) _(ccmn) \ | |
31 | _(csel) _(cmov) \ | |
32 | _(csinc) _(cinc) _(cset) \ | |
33 | _(csneg) _(cneg) \ | |
34 | _(csinv) _(cinv) _(csetm) | |
35 | #define _CONDVAR(cc) _definstvar cc; | |
36 | #define _INSTVARS(inst) \ | |
37 | .macro _definstvar cc; \ | |
38 | .macro inst.\cc args:vararg; inst \args, \cc; .endm; \ | |
39 | .endm; \ | |
40 | _COND(_CONDVAR); \ | |
41 | .purgem _definstvar; | |
42 | _INST(_INSTVARS) | |
43 | #undef _COND | |
44 | #undef _INST | |
45 | #undef _CONDVAR | |
46 | #undef _INSTVARS | |
47 | ||
48 | #define CCMP_N 8 | |
49 | #define CCMP_Z 4 | |
50 | #define CCMP_C 2 | |
51 | #define CCMP_V 1 | |
52 | ||
53 | #define CCMP_MI CCMP_N | |
54 | #define CCMP_PL 0 | |
55 | #define CCMP_EQ CCMP_Z | |
56 | #define CCMP_NE 0 | |
57 | #define CCMP_CS CCMP_C | |
58 | #define CCMP_HS CCMP_C | |
59 | #define CCMP_CC 0 | |
60 | #define CCMP_LO 0 | |
61 | #define CCMP_VS CCMP_V | |
62 | #define CCMP_VC 0 | |
63 | #define CCMP_HI CCMP_C | |
64 | #define CCMP_LS 0 | |
65 | #define CCMP_LT CCMP_N | |
66 | #define CCMP_GE 0 | |
67 | #define CCMP_LE CCMP_N | |
68 | #define CCMP_GT 0 | |
69 | ||
70 | #else | |
71 | # error "not supported" | |
72 | #endif | |
06297a93 MW |
73 | |
74 | .macro proc name | |
75 | .globl \name | |
76 | .type \name, STT_FUNC | |
77 | .p2align 4 | |
78 | \name\(): | |
79 | .macro endproc | |
80 | .size \name, . - \name | |
81 | .purgem endproc | |
82 | .endm | |
83 | .endm | |
84 | ||
85 | .macro ch c | |
90c4eee3 MW |
86 | #if defined(__i386__) |
87 | ||
88 | pushf | |
89 | push eax | |
90 | push ebx | |
91 | push ecx | |
92 | push edx | |
93 | push ebp | |
94 | mov ebp, esp | |
95 | and esp, -16 | |
96 | ||
97 | push \c | |
98 | call putchar@plt | |
99 | ||
100 | call get_pc_ebx | |
101 | add ebx, offset _GLOBAL_OFFSET_TABLE | |
102 | mov eax, [ebx + stdout@GOT] | |
103 | mov eax, [eax] | |
104 | call fflush@plt | |
105 | ||
106 | mov esp, ebp | |
107 | pop ebp | |
108 | pop edx | |
109 | pop ecx | |
110 | pop ebx | |
111 | pop eax | |
112 | popf | |
113 | ||
114 | #elif defined(__x86_64__) | |
115 | ||
06297a93 MW |
116 | pushf |
117 | push rax | |
118 | push rcx | |
119 | push rdx | |
120 | push rsi | |
121 | push rdi | |
122 | push r8 | |
123 | push r9 | |
124 | push rbp | |
125 | mov rbp, rsp | |
126 | and rsp, -16 | |
127 | ||
128 | mov rdi, \c | |
129 | call putchar@plt | |
130 | ||
131 | mov rdi, [rip + stdout] | |
132 | call fflush@plt | |
133 | ||
134 | mov rsp, rbp | |
135 | pop rbp | |
136 | pop r9 | |
137 | pop r8 | |
138 | pop rdi | |
139 | pop rsi | |
140 | pop rdx | |
141 | pop rcx | |
142 | pop rax | |
143 | popf | |
90c4eee3 MW |
144 | |
145 | #elif defined(__arm__) | |
146 | ||
147 | stmfd r13!, {r0-r4, r12, r14} | |
148 | ||
149 | mov r4, r13 | |
150 | bic r14, r4, #15 | |
151 | mov r13, r14 | |
152 | ||
153 | mov r0, #\c | |
154 | bl putchar@plt | |
155 | ||
156 | ldr r14, .L$_c$gotoff$\@ | |
157 | .L$_c$gotpc$\@: | |
158 | add r14, pc, r14 | |
159 | b .L$_c$cont$\@ | |
160 | .L$_c$gotoff$\@: | |
161 | .word _GLOBAL_OFFSET_TABLE - .L$_c$gotpc$\@ - 8 | |
162 | .L$_c$cont$\@: | |
163 | bl fflush@plt | |
164 | ||
165 | mov r13, r4 | |
166 | ldmfd r13!, {r0-r4, r12, r14} | |
167 | ||
168 | #elif defined(__aarch64__) | |
169 | ||
170 | sub sp, sp, #20*8 | |
171 | stp x0, x1, [sp, #0] | |
172 | stp x2, x3, [sp, #16] | |
173 | stp x4, x5, [sp, #32] | |
174 | stp x6, x7, [sp, #48] | |
175 | stp x8, x9, [sp, #64] | |
176 | stp x10, x11, [sp, #80] | |
177 | stp x12, x13, [sp, #96] | |
178 | stp x14, x15, [sp, #112] | |
179 | stp x16, x17, [sp, #128] | |
180 | mrs x16, nzcv | |
181 | stp x16, x30, [sp, #144] | |
182 | ||
183 | mov w0, #\c | |
184 | bl putchar | |
185 | adrp x0, :got:stdout | |
186 | ldr x0, [x0, #:got_lo12:stdout] | |
187 | ldr x0, [x0] | |
188 | bl fflush | |
189 | ||
190 | ldp x16, x30, [sp, #144] | |
191 | msr nzcv, x16 | |
192 | ldp x16, x17, [sp, #128] | |
193 | ldp x14, x15, [sp, #112] | |
194 | ldp x12, x13, [sp, #96] | |
195 | ldp x10, x11, [sp, #80] | |
196 | ldp x8, x9, [sp, #64] | |
197 | ldp x6, x7, [sp, #48] | |
198 | ldp x4, x5, [sp, #32] | |
199 | ldp x2, x3, [sp, #16] | |
200 | ldp x0, x1, [sp, #0] | |
201 | add sp, sp, #20*8 | |
202 | ||
203 | #else | |
204 | # error "not supported" | |
205 | #endif | |
06297a93 MW |
206 | .endm |
207 | ||
90c4eee3 MW |
208 | .macro notimpl |
209 | #if defined(__i386__) || defined(__x86_64__) | |
210 | ud2 | |
211 | #elif defined(__arm__) | |
212 | udf | |
213 | #elif defined(__aarch64__) | |
214 | hlt #0 | |
215 | #else | |
216 | # error "not supported" | |
217 | #endif | |
218 | .endm | |
219 | ||
220 | .section .note.GNU-stack, "", %progbits | |
221 | ||
06297a93 MW |
222 | .text |
223 | ||
90c4eee3 MW |
224 | #if defined(__i386__) |
225 | get_pc_ebx: | |
226 | mov ebx, [esp] | |
227 | ret | |
228 | #endif | |
229 | ||
230 | ||
06297a93 MW |
231 | proc call_example |
232 | ||
90c4eee3 MW |
233 | #if defined(__i386__) |
234 | ||
235 | push ebx // ebx | |
236 | push esi // esi, ebx | |
237 | push edi // edi, esi, ebx | |
238 | push ebp // flags, ebp, ..., ebx | |
239 | pushf | |
240 | ||
241 | mov edi, [esp + 4*6] | |
242 | mov esi, [esp + 4*7] | |
243 | push esi // regs, flags, ebp, ..., ebx | |
244 | ||
245 | call get_pc_ebx | |
246 | lea eax, [ebx + 9f - .] | |
247 | push eax // cont, regs, flags, ebp, ..., ebx | |
248 | push edi // func, cont, regs, flags, ebp, ..., ebx | |
249 | ||
250 | mov eax, [esi + 28] | |
251 | pushf | |
252 | pop ecx | |
253 | and eax, 0x0cd5 | |
254 | and ecx, ~0x0cd5 | |
255 | or eax, ecx | |
256 | push eax | |
257 | popf | |
258 | mov eax, [esi + 0] | |
259 | mov ebx, [esi + 4] | |
260 | mov ecx, [esi + 8] | |
261 | mov edx, [esi + 12] | |
262 | mov edi, [esi + 20] | |
263 | mov ebp, [esi + 24] | |
264 | mov esi, [esi + 16] | |
265 | ||
266 | ret // -> func; regs, flags, ebp, ..., ebx | |
267 | ||
268 | 9: pushf // eflags, regs, flags, ebp, ..., ebx | |
269 | push esi // esi, eflags, regs, flags, ebp, ..., ebx | |
270 | mov esi, [esp + 8] | |
271 | mov [esi + 0], eax | |
272 | mov [esi + 4], ebx | |
273 | mov [esi + 8], ecx | |
274 | mov [esi + 12], edx | |
275 | mov [esi + 20], edi | |
276 | mov [esi + 24], ebp | |
277 | pop eax // rflags, regs, flags, ebp, ..., ebx | |
278 | mov [esi + 16], eax | |
279 | pop eax // regs, flags, ebp, ..., ebx | |
280 | mov [esi + 28], eax | |
281 | ||
282 | add esp, 4 // flags, ebp, ..., ebx | |
283 | popf // ebp, ..., ebx | |
284 | pop ebp // ..., ebx | |
285 | pop edi | |
286 | pop esi | |
287 | pop ebx // | |
288 | ret | |
289 | ||
290 | #elif defined(__x86_64__) | |
291 | ||
06297a93 MW |
292 | push rbx // rbx |
293 | push r10 | |
294 | push r11 | |
295 | push r12 | |
296 | push r13 | |
297 | push r14 | |
298 | push r15 | |
299 | push rbp // flags, rbp, ..., rbx | |
300 | pushf | |
301 | ||
302 | push rsi // regs, flags, rbp, ..., rbx | |
303 | ||
304 | lea rax, [rip + 9f] | |
305 | push rax // cont, regs, flags, rbp, ..., rbx | |
306 | push rdi // func, cont, regs, flags, rbp, ..., rbx | |
307 | ||
90c4eee3 | 308 | mov rax, [rsi + 8*15] |
06297a93 MW |
309 | pushf |
310 | pop rcx | |
311 | and rax, 0x0cd5 | |
312 | and rcx, ~0x0cd5 | |
313 | or rax, rcx | |
314 | push rax | |
315 | popf | |
90c4eee3 MW |
316 | mov rax, [rsi + 0] |
317 | mov rbx, [rsi + 8] | |
318 | mov rcx, [rsi + 16] | |
319 | mov rdx, [rsi + 24] | |
320 | mov rdi, [rsi + 40] | |
321 | mov rbp, [rsi + 48] | |
322 | mov r8, [rsi + 56] | |
323 | mov r9, [rsi + 64] | |
324 | mov r10, [rsi + 72] | |
325 | mov r11, [rsi + 80] | |
326 | mov r12, [rsi + 88] | |
327 | mov r13, [rsi + 96] | |
328 | mov r14, [rsi + 104] | |
329 | mov r15, [rsi + 112] | |
330 | mov rsi, [rsi + 32] | |
06297a93 MW |
331 | |
332 | ret // -> func; regs, flags, rbp, ..., rbx | |
333 | ||
334 | 9: pushf // rflags, regs, flags, rbp, ..., rbx | |
335 | push rsi // rsi, rflags, regs, flags, rbp, ..., rbx | |
336 | mov rsi, [rsp + 16] | |
90c4eee3 MW |
337 | mov [rsi + 0], rax |
338 | mov [rsi + 8], rbx | |
339 | mov [rsi + 16], rcx | |
340 | mov [rsi + 24], rdx | |
341 | mov [rsi + 40], rdi | |
342 | mov [rsi + 48], rbp | |
343 | mov [rsi + 56], r8 | |
344 | mov [rsi + 64], r9 | |
345 | mov [rsi + 72], r10 | |
346 | mov [rsi + 80], r11 | |
347 | mov [rsi + 88], r12 | |
348 | mov [rsi + 96], r13 | |
349 | mov [rsi + 104], r14 | |
350 | mov [rsi + 112], r15 | |
06297a93 | 351 | pop rax // rflags, regs, flags, rbp, ..., rbx |
90c4eee3 | 352 | mov [rsi + 32], rax |
06297a93 | 353 | pop rax // regs, flags, rbp, ..., rbx |
90c4eee3 | 354 | mov [rsi + 120], rax |
06297a93 MW |
355 | |
356 | add rsp, 8 // flags, rbp, ..., rbx | |
357 | popf // rbp, ..., rbx | |
358 | pop rbp // ..., rbx | |
359 | pop r15 | |
360 | pop r14 | |
361 | pop r13 | |
362 | pop r12 | |
363 | pop r11 | |
364 | pop r10 | |
365 | pop rbx // | |
366 | ret | |
367 | ||
90c4eee3 MW |
368 | #elif defined(__arm__) |
369 | ||
370 | stmfd r13!, {r0, r1, r4-r11, r14} | |
371 | ldmia r1, {r0-r12, r14} | |
372 | msr cpsr, r14 | |
373 | mov r14, pc | |
374 | ldr pc, [r13], #4 | |
375 | ldr r14, [r13], #4 | |
376 | stmia r14!, {r0-r12} | |
377 | mrs r0, cpsr | |
378 | str r0, [r14] | |
379 | ldmfd r13!, {r4-r11, pc} | |
380 | ||
381 | #elif defined(__aarch64__) | |
382 | ||
383 | stp x29, x30, [sp, #-13*8]! | |
384 | mov x29, sp | |
385 | stp x19, x20, [sp, #16] | |
386 | stp x21, x22, [sp, #32] | |
387 | stp x23, x24, [sp, #48] | |
388 | stp x25, x26, [sp, #64] | |
389 | stp x27, x28, [sp, #80] | |
390 | str x1, [sp, #96] | |
391 | ||
392 | mov x16, x0 | |
393 | ||
394 | ldr x17, [x1, #128] | |
395 | ldp x14, x15, [x1, #112] | |
396 | ldp x12, x13, [x1, #96] | |
397 | ldp x10, x11, [x1, #80] | |
398 | ldp x8, x9, [x1, #64] | |
399 | ldp x6, x7, [x1, #48] | |
400 | ldp x4, x5, [x1, #32] | |
401 | ldp x2, x3, [x1, #16] | |
402 | ldp x0, x1, [x1, #0] | |
403 | msr nzcv, x17 | |
404 | ||
405 | blr x16 | |
406 | ||
407 | ldr x16, [sp, #96] | |
408 | mrs x17, nzcv | |
409 | str x17, [x16, #128] | |
410 | stp x14, x15, [x16, #112] | |
411 | stp x12, x13, [x16, #96] | |
412 | stp x10, x11, [x16, #80] | |
413 | stp x8, x9, [x16, #64] | |
414 | stp x6, x7, [x16, #48] | |
415 | stp x4, x5, [x16, #32] | |
416 | stp x2, x3, [x16, #16] | |
417 | stp x0, x1, [x16, #0] | |
418 | ||
419 | ldp x19, x20, [sp, #16] | |
420 | ldp x21, x22, [sp, #32] | |
421 | ldp x23, x24, [sp, #48] | |
422 | ldp x25, x26, [sp, #64] | |
423 | ldp x27, x28, [sp, #80] | |
424 | ldp x29, x30, [sp], #13*8 | |
425 | ||
426 | #else | |
427 | # error "not supported" | |
428 | #endif | |
429 | ||
06297a93 MW |
430 | endproc |
431 | ||
432 | proc nop | |
433 | ||
434 | ret | |
435 | ||
436 | endproc | |
437 | ||
438 | ///-------------------------------------------------------------------------- | |
90c4eee3 | 439 | /// 0x00--0x0f |
06297a93 MW |
440 | |
441 | proc x00 | |
442 | ||
443 | // clear all 64 bits of extended traditional registers | |
90c4eee3 MW |
444 | |
445 | #if defined(__x86_64__) | |
446 | ||
447 | xor eax, eax // clear rax | |
448 | lea rbx, [0] // rbx -> _|_ | |
06297a93 | 449 | loop . // iterate, decrement rcx until zero |
90c4eee3 MW |
450 | mov rdx, 0 // set rdx = 0 |
451 | and esi, 0 // clear all bits of rsi | |
452 | sub edi, edi // set rdi = edi - edi = 0 | |
06297a93 MW |
453 | push 0 |
454 | pop rbp // pop 0 into rbp | |
455 | ||
90c4eee3 MW |
456 | #elif defined(__i386__) |
457 | ||
458 | xor eax, eax | |
459 | lea ebx, [0] | |
460 | loop . | |
461 | mov edx, 0 | |
462 | and esi, 0 | |
463 | sub edi, edi | |
464 | push 0 | |
465 | pop ebp | |
466 | ||
467 | #elif defined(__arm__) | |
468 | ||
469 | eor r0, r0, r0 | |
470 | rsb r1, r1, r1 | |
471 | 0: subs r2, r2, #1 | |
472 | bne 0b | |
473 | mov r3, #0 | |
474 | and r4, r4, #0 | |
475 | sub r5, r5, r5 | |
476 | ||
477 | #elif defined(__aarch64__) | |
478 | ||
479 | eor w0, w0, w0 | |
480 | mov w1, wzr | |
481 | 0: sub w2, w2, #1 | |
482 | cbnz w2, 0b | |
483 | mov w3, #0 | |
484 | and w4, w4, wzr | |
485 | sub w5, w5, w5 | |
486 | ||
487 | #else | |
488 | notimpl | |
489 | #endif | |
490 | ||
06297a93 MW |
491 | ret |
492 | ||
493 | endproc | |
494 | ||
495 | proc x01 | |
496 | ||
497 | // advance a fibonacci pair by c steps | |
498 | // | |
499 | // on entry, a and d are f_{i+1} and f_i; on exit, they are f_{i+c+1} | |
500 | // and f_{i+c}, where f_{i+1} = f_i + f_{i-1} | |
90c4eee3 MW |
501 | |
502 | #if defined(__x86_64__) | |
503 | ||
06297a93 MW |
504 | 0: xadd rax, rdx // a, d = a + d, a |
505 | // = f_{i+1} + f_i, f_{i+1} | |
506 | // = f_{i+2}, f_{i+1} | |
507 | loop 0b // advance i, decrement c, iterate | |
508 | ||
90c4eee3 MW |
509 | #elif defined(__i386__) |
510 | ||
511 | 0: xadd eax, edx | |
512 | loop 0b | |
513 | ||
514 | #elif defined(__arm__) | |
515 | ||
516 | 0: subs r2, r2, #2 | |
517 | add r3, r3, r0 | |
518 | blo 8f | |
519 | add r0, r0, r3 | |
520 | bhi 0b | |
521 | ||
522 | 8: movne r0, r3 | |
523 | ||
524 | #elif defined(__aarch64__) | |
525 | ||
526 | 0: subs x2, x2, #2 | |
527 | add x3, x3, x0 | |
528 | b.lo 8f | |
529 | add x0, x0, x3 | |
530 | b.hi 0b | |
531 | ||
532 | 8: cmov.ne x0, x3 | |
533 | ||
534 | #else | |
535 | notimpl | |
536 | #endif | |
537 | ||
06297a93 MW |
538 | ret |
539 | ||
540 | endproc | |
541 | ||
542 | proc x02 | |
543 | ||
544 | // boolean canonify a: if a = 0 on entry, leave it zero; otherwise | |
545 | // set a = 1 | |
90c4eee3 MW |
546 | |
547 | #if defined(__x86_64__) | |
548 | ||
06297a93 MW |
549 | neg rax // set cf iff a /= 0 |
550 | sbb rax, rax // a = a - a - cf = -cf | |
551 | neg rax // a = cf | |
552 | ||
90c4eee3 MW |
553 | #elif defined(__i386__) |
554 | ||
555 | neg eax | |
556 | sbb eax, eax | |
557 | neg eax | |
558 | ||
559 | #elif defined(__arm__) | |
560 | ||
561 | movs r1, r0 // the easy way | |
562 | movne r1, #1 // mvnne r1, #1 for mask | |
563 | ||
564 | cmp r0, #1 // clear cf iff a == 0 | |
565 | sbc r2, r0, r0 // c' = a - a - 1 + cf = cf - 1 | |
566 | add r2, r2, #1 // c' = cf | |
567 | ||
568 | sub r3, r0, r0, lsr #1 // d' top bit clear; d' = 0 iff a = 0 | |
569 | rsb r3, r3, #0 // d' top bit set iff a /= 0 | |
570 | mov r3, r3, lsr #31 // asr for mask | |
571 | ||
572 | rsbs r0, r0, #0 | |
573 | sbc r0, r0, r0 | |
574 | rsb r0, r0, #0 | |
575 | ||
576 | #elif defined(__aarch64__) | |
577 | ||
578 | cmp x0, #0 // trivial | |
579 | cset.ne x1 // csetm for mask | |
580 | ||
581 | cmp xzr, x0 // set cf iff a == 0 | |
582 | sbc x2, x0, x0 // c' = a - a - 1 + cf = cf - 1 | |
583 | neg x2, x2 // c' = 1 - cf | |
584 | ||
585 | sub x3, x0, x0, lsr #1 // if a < 2^63 then a' = ceil(d/2) < | |
586 | // 2^63 | |
587 | // if a >= 2^63, write a = 2^63 + t | |
588 | // with t < 2^63; d' = 2^63 - 2^62 + | |
589 | // ceil(t/2) = 2^62 + ceil(t/2), and | |
590 | // ceil(t/2) < 2^62 | |
591 | // anyway d' < 2^63 and d' = 0 iff | |
592 | // a = 0 | |
593 | neg x3, x3 // d' top bit set iff a /= 0 | |
594 | lsr x3, x3, #63 // asr for mask | |
595 | ||
596 | cmp x0, #1 // set cf iff a /= 0 | |
597 | adc x0, xzr, xzr // a' = 0 + 0 + cf = cf | |
598 | ||
599 | #else | |
600 | notimpl | |
601 | #endif | |
602 | ||
06297a93 MW |
603 | ret |
604 | ||
605 | endproc | |
606 | ||
607 | proc x03 | |
608 | ||
609 | // set a = min(a, d) (unsigned); clobber c, d | |
90c4eee3 MW |
610 | |
611 | #if defined(__x86_64__) | |
612 | ||
06297a93 MW |
613 | sub rdx, rax // d' = d - a; set cf if a > d |
614 | sbb rcx, rcx // c = -cf = -[a > d] | |
615 | and rcx, rdx // c = a > d ? d - a : 0 | |
616 | add rax, rcx // a' = a > d ? d : a | |
617 | ||
90c4eee3 MW |
618 | #elif defined(__i386__) |
619 | ||
620 | sub edx, eax | |
621 | sbb ecx, ecx | |
622 | and ecx, edx | |
623 | add eax, ecx | |
624 | ||
625 | #elif defined(__arm__) | |
626 | ||
627 | cmp r0, r3 // the easy way | |
628 | movlo r1, r0 // only needed for out-of-place | |
629 | movhs r1, r3 | |
630 | ||
631 | subs r3, r3, r0 | |
632 | sbc r12, r12, r12 | |
633 | and r12, r12, r3 | |
634 | add r0, r0, r12 | |
635 | ||
636 | #elif defined(__aarch64__) | |
637 | ||
638 | cmp x0, x3 // the easy way | |
639 | csel.lo x1, x0, x3 | |
640 | ||
641 | subs x3, x3, x0 // d' = d - a; set cf if d >= a | |
642 | sbc x16, xzr, xzr // t = -1 + cf = -[a > d] | |
643 | and x16, x16, x3 // t = a > d ? d - a : 0 | |
644 | add x0, x0, x16 // a' = a > d ? d : a | |
645 | ||
646 | #else | |
647 | notimpl | |
648 | #endif | |
649 | ||
06297a93 MW |
650 | ret |
651 | ||
652 | endproc | |
653 | ||
654 | proc x04 | |
655 | ||
656 | // switch case? | |
90c4eee3 MW |
657 | |
658 | #if defined(__x86_64__) | |
659 | ||
660 | // unrelated playing | |
661 | mov ecx, eax | |
662 | mov rbx, -1 | |
663 | mov edx, ecx | |
664 | sub edx, '0' | |
665 | cmp edx, 10 | |
666 | cmovb rbx, rdx | |
667 | or ecx, 0x20 | |
668 | mov edx, ecx | |
669 | sub edx, 'a' | |
670 | sub ecx, 'a' - 10 | |
671 | cmp edx, 6 | |
672 | cmovb rbx, rcx | |
673 | ||
674 | xor al, 0x20 | |
675 | ||
676 | #elif defined(__i386__) | |
677 | ||
678 | // unrelated playing | |
679 | mov ecx, eax | |
680 | mov ebx, -1 | |
681 | mov edx, ecx | |
682 | sub edx, '0' | |
683 | cmp edx, 10 | |
684 | cmovb ebx, edx | |
685 | or ecx, 0x20 | |
686 | mov edx, ecx | |
687 | sub edx, 'a' | |
688 | sub ecx, 'a' - 10 | |
689 | cmp edx, 6 | |
690 | cmovb ebx, ecx | |
691 | ||
06297a93 MW |
692 | xor al, 0x20 |
693 | ||
90c4eee3 MW |
694 | #elif defined(__arm__) |
695 | ||
696 | // unrelated playing | |
697 | mvn r1, #0 | |
698 | sub r12, r0, #'0' | |
699 | cmp r12, #10 | |
700 | movlo r1, r12 | |
701 | orr r12, r0, #0x20 | |
702 | sub r12, r12, #'a' | |
703 | cmp r12, #6 | |
704 | addlo r1, r12, #10 | |
705 | ||
706 | eor r0, r0, #0x20 | |
707 | ||
708 | #elif defined(__aarch64__) | |
709 | ||
710 | // unrelated playing | |
711 | mov x1, #-1 | |
712 | sub w16, w0, #'0' | |
713 | cmp w16, #10 | |
714 | cmov.lo x1, x16 | |
715 | orr w16, w0, #0x20 | |
716 | sub w16, w16, #'a' - 10 | |
717 | cmp w16, #10 | |
718 | ccmp.hs w16, #16, #CCMP_HS | |
719 | cmov.lo x1, x16 | |
720 | ||
721 | eor w0, w0, #0x20 | |
722 | ||
723 | #else | |
724 | notimpl | |
725 | #endif | |
726 | ||
06297a93 MW |
727 | ret |
728 | ||
729 | endproc | |
730 | ||
731 | proc x05 | |
732 | ||
733 | // answer whether 5 <= a </<= 9. | |
90c4eee3 MW |
734 | |
735 | #if defined(__x86_64__) | |
736 | ||
06297a93 MW |
737 | sub rax, 5 // a' = a - 5 |
738 | cmp rax, 4 // is a' - 5 </<= 4? | |
739 | ||
740 | // cc a' a | |
741 | // | |
742 | // z/e a' = 4 a = 9 | |
743 | // nz/ne a' /= 4 a /= 9 | |
744 | // | |
745 | // a/nbe a' > 4 a > 9 or a < 5 | |
746 | // nc/ae/nb a' >= 4 a >= 9 or a < 5 | |
747 | // c/b/nae a' < 4 5 <= a < 9 | |
748 | // be/na a' <= 4 5 <= a <= 9 | |
749 | // | |
750 | // o a' < -2^63 + 4 -2^63 + 5 <= a < -2^63 + 9 | |
751 | // no a' >= -2^63 + 4 a >= -2^63 + 9 or | |
752 | // a < -2^63 + 5 | |
753 | // s -2^63 + 4 <= a' < 4 -2^63 + 9 <= a < 9 | |
754 | // ns a' < -2^63 + 4 or a < -2^63 + 9 or a >= 9 | |
755 | // a' >= 4 | |
756 | // ge/nl a' >= 4 a >= 9 or a < -2^63 + 5 | |
757 | // l/nge a' < 4 -2^63 + 5 <= a < 9 | |
758 | // g/nle a' > 4 a > 9 or a < -2^63 + 5 | |
759 | // le/ng a' <= 4 -2^63 + 5 <= a <= 9 | |
760 | ||
90c4eee3 MW |
761 | #elif defined(__i386__) |
762 | ||
763 | sub eax, 5 | |
764 | cmp eax, 4 | |
765 | ||
766 | #elif defined(__arm__) | |
767 | ||
768 | // i dimly remember having a slick way to do this way back in the | |
769 | // day, but i can't figure it out any more. | |
770 | sub r0, #5 | |
771 | cmp r0, #4 | |
772 | ||
773 | #elif defined(__aarch64__) | |
774 | ||
775 | // literal translation is too obvious | |
776 | cmp x0, #5 | |
777 | ccmp.hs x0, #9, #CCMP_HS | |
778 | ||
779 | #else | |
780 | notimpl | |
781 | #endif | |
782 | ||
06297a93 MW |
783 | ret |
784 | ||
785 | endproc | |
786 | ||
787 | proc x06 | |
788 | ||
789 | // leave a unchanged, but set zf if a = 0, cf if a /= 0, clear of, | |
790 | // set sf to msb(a) | |
90c4eee3 MW |
791 | |
792 | #if defined(__x86_64__) | |
793 | ||
06297a93 MW |
794 | not rax // a' = -a - 1 |
795 | inc rax // a' = -a | |
796 | neg rax // a' = a | |
797 | ||
90c4eee3 MW |
798 | #elif defined(__i386__) |
799 | ||
800 | not eax | |
801 | inc eax | |
802 | neg eax | |
803 | ||
804 | #elif defined(__arm__) | |
805 | ||
806 | mvn r0, r0 | |
807 | add r0, r0, #1 | |
808 | rsbs r0, r0, #0 // cf has opposite sense | |
809 | ||
810 | #elif defined(__aarch64__) | |
811 | ||
812 | mvn x0, x0 | |
813 | add x0, x0, #1 | |
814 | negs x0, x0 // cf has opposite sense | |
815 | ||
816 | #else | |
817 | notimpl | |
818 | #endif | |
819 | ||
06297a93 MW |
820 | ret |
821 | ||
822 | endproc | |
823 | ||
824 | proc x07 | |
825 | ||
826 | // same as before (?) | |
90c4eee3 MW |
827 | |
828 | #if defined(__x86_64__) | |
829 | ||
06297a93 MW |
830 | inc rax // a' = a + 1 |
831 | neg rax // a' = -a - 1 | |
832 | inc rax // a' = -a | |
833 | neg rax // a' = a | |
834 | ||
90c4eee3 MW |
835 | #elif defined(__i386__) |
836 | ||
837 | inc eax | |
838 | neg eax | |
839 | inc eax | |
840 | neg eax | |
841 | ||
842 | #elif defined(__arm__) | |
843 | ||
844 | add r0, r0, #1 | |
845 | rsb r0, r0, #0 | |
846 | add r0, r0, #1 | |
847 | rsbs r0, r0, #0 | |
848 | ||
849 | #elif defined(__aarch64__) | |
850 | ||
851 | add x0, x0, #1 | |
852 | neg x0, x0 | |
853 | add x0, x0, #1 | |
854 | negs x0, x0 // cf has opposite sense | |
855 | ||
856 | #else | |
857 | notimpl | |
858 | #endif | |
859 | ||
06297a93 MW |
860 | ret |
861 | ||
862 | endproc | |
863 | ||
864 | proc x08 | |
865 | ||
866 | // floor((a + d)/2), correctly handling overflow conditions; final cf | |
867 | // is lsb(a + d), probably uninteresting | |
90c4eee3 MW |
868 | |
869 | #if defined(__x86_64__) | |
870 | ||
06297a93 MW |
871 | add rax, rdx // cf || a' = a + d |
872 | rcr rax, 1 // shift 65-bit result right by one | |
873 | // place; lsb moves into carry | |
874 | ||
90c4eee3 MW |
875 | #elif defined(__i386__) |
876 | ||
877 | add eax, edx | |
878 | rcr eax, 1 | |
879 | ||
880 | #elif defined(__arm__) | |
881 | ||
882 | // like the two-instruction a64 version | |
883 | sub r1, r3, r0 | |
884 | add r1, r0, r1, lsr #1 | |
885 | ||
886 | // the slick version, similar to the above | |
887 | adds r0, r0, r3 | |
888 | mov r0, r0, rrx | |
889 | ||
890 | #elif defined(__aarch64__) | |
891 | ||
892 | // a64 lacks a32's rrx. literal translation. | |
893 | adds x1, x0, x3 // cf || a' = a + d | |
894 | adc x16, xzr, xzr // realize cf in extra register | |
895 | extr x1, x16, x1, #1 // shift down one place | |
896 | ||
897 | // two instruction version: clobbers additional register. (if you | |
898 | // wanted the answer in any other register, even overwriting d, then | |
899 | // this is unnecessary.) also depends on d >= a. | |
900 | sub x16, x3, x0 // compute difference | |
901 | add x0, x0, x16, lsr #1 // add half of it (rounded down) | |
902 | ||
903 | #else | |
904 | notimpl | |
905 | #endif | |
906 | ||
06297a93 MW |
907 | ret |
908 | ||
909 | endproc | |
910 | ||
911 | proc x09 | |
912 | ||
913 | // a = a/8, rounded to nearest; i.e., floor(a/8) if a == 0, 1, 2, 3 | |
914 | // (mod 8), or ceil(a/8) if a == 4, 5, 6, 7 (mod 8). | |
90c4eee3 MW |
915 | |
916 | #if defined(__x86_64__) | |
917 | ||
06297a93 MW |
918 | shr rax, 3 // a' = floor(a/8); cf = 1 if a == |
919 | // 4, 5, 6, 7 (mod 8) | |
920 | adc rax, 0 // a' = floor(a/8) + cf | |
921 | ||
90c4eee3 MW |
922 | #elif defined(__i386__) |
923 | ||
924 | shr eax, 3 | |
925 | adc eax, 0 | |
926 | ||
927 | #elif defined(__arm__) | |
928 | ||
929 | movs r0, r0, lsr #3 | |
930 | adc r0, r0, #0 | |
931 | ||
932 | #elif defined(__aarch64__) | |
933 | ||
934 | tst x0, #4 | |
935 | orr x0, xzr, x0, lsr #3 | |
936 | cinc.ne x0, x0 | |
937 | ||
938 | #else | |
939 | notimpl | |
940 | #endif | |
941 | ||
06297a93 MW |
942 | ret |
943 | ||
944 | endproc | |
945 | ||
946 | proc x0a | |
947 | ||
948 | // increment c-byte little-endian bignum at rdi | |
90c4eee3 MW |
949 | |
950 | #if defined(__x86_64__) | |
951 | ||
06297a93 MW |
952 | add byte ptr [rdi], 1 |
953 | 0: inc rdi | |
954 | adc byte ptr [rdi], 0 | |
955 | loop 0b | |
956 | ||
90c4eee3 MW |
957 | #elif defined(__i386__) |
958 | ||
959 | add byte ptr [edi], 1 | |
960 | 0: inc edi | |
961 | adc byte ptr [edi], 0 | |
962 | loop 0b | |
963 | ||
964 | #elif defined(__arm__) | |
965 | ||
966 | mov r12, #256 // set initial carry | |
967 | 0: ldrb r0, [r5] | |
968 | subs r2, r2, #1 | |
969 | add r12, r0, r12, lsr #8 | |
970 | strb r12, [r5], #1 | |
971 | bne 0b | |
972 | ||
973 | #elif defined(__aarch64__) | |
974 | ||
975 | mov w17, #256 // set initial carry | |
976 | 0: ldrb w16, [x5] | |
977 | sub x2, x2, #1 | |
978 | add w17, w16, w17, lsr #8 | |
979 | strb w17, [x5], #1 | |
980 | cbnz x2, 0b | |
981 | ||
982 | #else | |
983 | notimpl | |
984 | #endif | |
985 | ||
06297a93 MW |
986 | ret |
987 | ||
988 | endproc | |
989 | ||
990 | proc x0b | |
991 | ||
992 | // negate double-precision d:a | |
90c4eee3 MW |
993 | |
994 | #if defined(__x86_64__) | |
995 | ||
06297a93 MW |
996 | not rdx // d' = -d - 1 |
997 | neg rax // a' = -a; | |
998 | // cf = 1 iff a /= 0 | |
999 | sbb rdx, -1 // d' = -d - cf | |
1000 | ||
90c4eee3 MW |
1001 | #elif defined(__i386__) |
1002 | ||
1003 | not edx | |
1004 | neg eax | |
1005 | sbb edx, -1 | |
1006 | ||
1007 | #elif defined(__arm__) | |
1008 | ||
1009 | // reverse subtract is awesome | |
1010 | rsbs r0, r0, #0 | |
1011 | rsc r3, r3, #0 | |
1012 | ||
1013 | #elif defined(__aarch64__) | |
1014 | ||
1015 | // easy way: everything is better with zero registers. | |
1016 | negs x0, x0 | |
1017 | ngc x3, x3 | |
1018 | ||
1019 | #else | |
1020 | notimpl | |
1021 | #endif | |
1022 | ||
06297a93 MW |
1023 | ret |
1024 | ||
1025 | endproc | |
1026 | ||
1027 | proc x0c | |
1028 | ||
1029 | // rotate is distributive over xor. | |
1030 | ||
90c4eee3 MW |
1031 | #if defined(__x86_64__) |
1032 | ||
06297a93 MW |
1033 | // rax // = a_1 || a_0 |
1034 | // rbx // = b_1 || b_0 | |
1035 | mov rcx, rax // = a_1 || a_0 | |
1036 | ||
1037 | xor rcx, rbx // = (a_1 XOR b_1) || (a_0 XOR b_0) | |
1038 | ror rcx, 0xd // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1039 | ||
1040 | ror rax, 0xd // = a_0 || a_1 | |
1041 | ror rbx, 0xd // = b_0 || b_1 | |
1042 | xor rax, rbx // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1043 | ||
1044 | cmp rax, rcx // always equal | |
1045 | ||
90c4eee3 MW |
1046 | #elif defined(__i386__) |
1047 | ||
1048 | mov ecx, eax // = a_1 || a_0 | |
1049 | ||
1050 | xor ecx, ebx // = (a_1 XOR b_1) || (a_0 XOR b_0) | |
1051 | ror ecx, 0xd // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1052 | ||
1053 | ror eax, 0xd // = a_0 || a_1 | |
1054 | ror ebx, 0xd // = b_0 || b_1 | |
1055 | xor eax, ebx // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1056 | ||
1057 | cmp eax, ecx // always equal | |
1058 | ||
1059 | #elif defined(__arm__) | |
1060 | ||
1061 | ||
1062 | // r0 // = a_1 || a_0 | |
1063 | // r1 // = b_1 || b_0 | |
1064 | eor r2, r0, r1 // = (a_1 XOR b_1) || (a_0 XOR b_0) | |
1065 | mov r2, r2, ror #13 // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1066 | ||
1067 | mov r1, r1, ror #13 // = b_0 || b_1 | |
1068 | eor r0, r1, r0, ror #13 // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1069 | ||
1070 | cmp r0, r2 // always equal | |
1071 | ||
1072 | #elif defined(__aarch64__) | |
1073 | ||
1074 | // x0 // = a_1 || a_0 | |
1075 | // x1 // = b_1 || b_0 | |
1076 | eor x2, x0, x1 // = (a_1 XOR b_1) || (a_0 XOR b_0) | |
1077 | ror x2, x2, #13 // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1078 | ||
1079 | ror x1, x1, #13 // = b_0 || b_1 | |
1080 | eor x0, x1, x0, ror #13 // = (a_0 XOR b_0) || (a_1 XOR b_1) | |
1081 | ||
1082 | cmp x0, x2 // always equal | |
1083 | ||
1084 | #else | |
1085 | notimpl | |
1086 | #endif | |
1087 | ||
06297a93 MW |
1088 | ret |
1089 | ||
1090 | endproc | |
1091 | ||
1092 | proc x0d | |
1093 | ||
1094 | // and is distributive over xor. | |
1095 | ||
90c4eee3 MW |
1096 | #if defined(__x86_64__) |
1097 | ||
06297a93 MW |
1098 | mov rdx, rbx // = b |
1099 | ||
1100 | xor rbx, rcx // = b XOR c | |
1101 | and rbx, rax // = a AND (b XOR c) | |
1102 | ||
1103 | and rdx, rax // = a AND b | |
1104 | and rax, rcx // = a AND c | |
1105 | xor rax, rdx // = (a AND b) XOR (a AND c) | |
1106 | // = a AND (b XOR c) | |
1107 | ||
1108 | cmp rax, rbx // always equal | |
1109 | ||
90c4eee3 MW |
1110 | #elif defined(__i386__) |
1111 | ||
1112 | mov edx, ebx // = b | |
1113 | ||
1114 | xor ebx, ecx // = b XOR c | |
1115 | and ebx, eax // = a AND (b XOR c) | |
1116 | ||
1117 | and edx, eax // = a AND b | |
1118 | and eax, ecx // = a AND c | |
1119 | xor eax, edx // = (a AND b) XOR (a AND c) | |
1120 | // = a AND (b XOR c) | |
1121 | ||
1122 | cmp eax, ebx // always equal | |
1123 | ||
1124 | #elif defined(__arm__) | |
1125 | ||
1126 | and r3, r0, r1 // = a AND b | |
1127 | ||
1128 | eor r1, r1, r2 // = b XOR c | |
1129 | and r1, r1, r0 // = a AND (b XOR c) | |
1130 | ||
1131 | and r0, r0, r2 // = a AND c | |
1132 | eor r0, r0, r3 // = (a AND b) XOR (a AND c) | |
1133 | // = a AND (b XOR c) | |
1134 | ||
1135 | cmp r0, r1 // always equal | |
1136 | ||
1137 | #elif defined(__aarch64__) | |
1138 | ||
1139 | and x3, x0, x1 // = a AND b | |
1140 | ||
1141 | eor x1, x1, x2 // = b XOR c | |
1142 | and x1, x1, x0 // = a AND (b XOR c) | |
1143 | ||
1144 | and x0, x0, x2 // = a AND c | |
1145 | eor x0, x0, x3 // = (a AND b) XOR (a AND c) | |
1146 | // = a AND (b XOR c) | |
1147 | ||
1148 | cmp x0, x1 // always equal | |
1149 | ||
1150 | #else | |
1151 | notimpl | |
1152 | #endif | |
1153 | ||
06297a93 MW |
1154 | ret |
1155 | ||
1156 | endproc | |
1157 | ||
1158 | proc x0e | |
1159 | ||
1160 | // de morgan's law | |
1161 | ||
90c4eee3 MW |
1162 | #if defined(__x86_64__) |
1163 | ||
06297a93 MW |
1164 | mov rcx, rax // = a |
1165 | ||
1166 | and rcx, rbx // = a AND b | |
1167 | not rcx // = NOT (a AND b) | |
1168 | ||
1169 | not rax // = NOT a | |
1170 | not rbx // = NOT b | |
1171 | or rax, rbx // = (NOT a) OR (NOT b) | |
1172 | // = NOT (a AND b) | |
1173 | ||
90c4eee3 MW |
1174 | cmp rax, rcx // always equal |
1175 | ||
1176 | #elif defined(__i386__) | |
1177 | ||
1178 | mov ecx, eax // = a | |
1179 | ||
1180 | and ecx, ebx // = a AND b | |
1181 | not ecx // = NOT (a AND b) | |
1182 | ||
1183 | not eax // = NOT a | |
1184 | not ebx // = NOT b | |
1185 | or eax, ebx // = (NOT a) OR (NOT b) | |
1186 | // = NOT (a AND b) | |
1187 | ||
1188 | cmp eax, ecx // always equal | |
1189 | ||
1190 | #elif defined(__arm__) | |
1191 | ||
1192 | and r2, r0, r1 // = a AND b | |
1193 | mvn r2, r2 // = NOT (a AND b) | |
1194 | ||
1195 | mvn r0, r0 // = NOT a | |
1196 | mvn r1, r1 // = NOT b | |
1197 | orr r0, r0, r1 // = (NOT a) OR (NOT b) | |
1198 | ||
1199 | cmp r0, r2 // always equal | |
1200 | ||
1201 | #elif defined(__aarch64__) | |
1202 | ||
1203 | and x2, x0, x1 // = a AND b | |
1204 | mvn x2, x2 // = NOT (a AND b) | |
1205 | ||
1206 | mvn x0, x0 // = NOT a | |
1207 | orn x0, x0, x1 // = (NOT a) OR (NOT b) | |
1208 | ||
1209 | cmp x0, x2 // always equal | |
1210 | ||
1211 | #else | |
1212 | notimpl | |
1213 | #endif | |
06297a93 MW |
1214 | |
1215 | ret | |
1216 | ||
1217 | endproc | |
1218 | ||
1219 | proc x0f | |
1220 | ||
1221 | // replace input buffer bytes with cumulative XORs with initial a; | |
1222 | // final a is XOR of all buffer bytes and initial a. | |
1223 | // | |
1224 | // not sure why you'd do this. | |
1225 | ||
90c4eee3 | 1226 | #if defined(__x86_64__) |
06297a93 MW |
1227 | |
1228 | 0: xor [rsi], al | |
1229 | lodsb | |
1230 | loop 0b | |
1231 | ||
90c4eee3 MW |
1232 | #elif defined(__i386__) |
1233 | ||
1234 | 0: xor [esi], al | |
1235 | lodsb | |
1236 | loop 0b | |
1237 | ||
1238 | #elif defined(__arm__) | |
1239 | ||
1240 | 0: ldrb r12, [r4] | |
1241 | subs r2, r2, #1 | |
1242 | eor r0, r0, r12 | |
1243 | strb r0, [r4], #1 | |
1244 | bne 0b | |
1245 | ||
1246 | #elif defined(__aarch64__) | |
1247 | ||
1248 | 0: ldrb w16, [x4] | |
1249 | sub x2, x2, #1 | |
1250 | eor w0, w0, w16 | |
1251 | strb w0, [x4], #1 | |
1252 | cbnz x2, 0b | |
1253 | ||
1254 | #else | |
1255 | notimpl | |
1256 | #endif | |
1257 | ||
06297a93 MW |
1258 | ret |
1259 | ||
1260 | endproc | |
1261 | ||
90c4eee3 MW |
1262 | ///-------------------------------------------------------------------------- |
1263 | /// 0x10--0x1f | |
1264 | ||
06297a93 MW |
1265 | proc x10 |
1266 | ||
1267 | // four different ways to swap a pair of registers. | |
1268 | ||
90c4eee3 MW |
1269 | #if defined(__x86_64__) |
1270 | ||
06297a93 MW |
1271 | push rax |
1272 | push rcx | |
1273 | pop rax | |
1274 | pop rcx | |
1275 | ||
1276 | xor rax, rcx | |
1277 | xor rcx, rax | |
1278 | xor rax, rcx | |
1279 | ||
1280 | add rax, rcx | |
1281 | sub rcx, rax | |
1282 | add rax, rcx | |
1283 | neg rcx | |
1284 | ||
1285 | xchg rax, rcx | |
1286 | ||
90c4eee3 MW |
1287 | #elif defined(__i386__) |
1288 | ||
1289 | push eax | |
1290 | push ecx | |
1291 | pop eax | |
1292 | pop ecx | |
1293 | ||
1294 | xor eax, ecx | |
1295 | xor ecx, eax | |
1296 | xor eax, ecx | |
1297 | ||
1298 | add eax, ecx | |
1299 | sub ecx, eax | |
1300 | add eax, ecx | |
1301 | neg ecx | |
1302 | ||
1303 | xchg eax, ecx | |
1304 | ||
1305 | #elif defined(__arm__) | |
1306 | ||
1307 | stmfd r13!, {r0, r2} | |
1308 | ldr r0, [r13, #4] | |
1309 | ldr r2, [r13], #8 | |
1310 | ||
1311 | eor r0, r0, r2 | |
1312 | eor r2, r2, r0 | |
1313 | eor r0, r0, r2 | |
1314 | ||
1315 | sub r0, r0, r2 | |
1316 | add r2, r2, r0 | |
1317 | rsb r0, r0, r2 // don't need 3-addr with reverse-sub | |
1318 | ||
1319 | mov r12, r0 | |
1320 | mov r0, r2 | |
1321 | mov r2, r0 | |
1322 | ||
1323 | #elif defined(__aarch64__) | |
1324 | ||
1325 | // anything you can do | |
1326 | stp x0, x2, [sp, #-16]! | |
1327 | ldp x2, x0, [sp], #16 | |
1328 | ||
1329 | eor x0, x0, x2 | |
1330 | eor x2, x2, x0 | |
1331 | eor x0, x0, x2 | |
1332 | ||
1333 | // the add/sub/add thing was daft. you can do it in three if you're | |
1334 | // clever -- and have three-address operations. | |
1335 | sub x0, x0, x2 | |
1336 | add x2, x2, x0 | |
1337 | sub x0, x2, x0 | |
1338 | ||
1339 | // but we lack a fourth. we can't do this in fewer than three | |
1340 | // instructions without hitting memory. only `ldp' will modify two | |
1341 | // registers at a time, so we need at least two instructions -- but | |
1342 | // if the first one sets one of our two registers to its final value | |
1343 | // then we lose the other input value with no way to recover it, so | |
1344 | // we must either write a fresh third register, or write something | |
1345 | // other than the final value, and in both cases we need a third | |
1346 | // instruction to fix everything up. we've done the wrong-something- | |
1347 | // other trick twice, so here's the captain-obvious use-a-third- | |
1348 | // register version. | |
1349 | mov x16, x0 | |
1350 | mov x0, x2 | |
1351 | mov x2, x16 | |
1352 | ||
1353 | #else | |
1354 | notimpl | |
1355 | #endif | |
1356 | ||
06297a93 MW |
1357 | ret |
1358 | ||
1359 | endproc | |
1360 | ||
1361 | proc x11 | |
1362 | ||
1363 | // assuming a is initialized to zero, set a to the inclusive or of | |
1364 | // the xor-differences of corresponding bytes in the c-byte strings | |
1365 | // at si and di. | |
1366 | // | |
1367 | // in particular, a will be zero (and zf set) if and only if the two | |
1368 | // strings are equal. | |
1369 | ||
90c4eee3 MW |
1370 | #if defined(__x86_64__) |
1371 | ||
06297a93 MW |
1372 | 0: mov dl, [rsi] |
1373 | xor dl, [rdi] | |
1374 | inc rsi | |
1375 | inc rdi | |
1376 | or al, dl | |
1377 | loop 0b | |
1378 | ||
90c4eee3 MW |
1379 | #elif defined(__i386__) |
1380 | ||
1381 | 0: mov dl, [esi] | |
1382 | xor dl, [edi] | |
1383 | inc esi | |
1384 | inc edi | |
1385 | or al, dl | |
1386 | loop 0b | |
1387 | ||
1388 | #elif defined(__arm__) | |
1389 | ||
1390 | 0: ldrb r1, [r4], #1 | |
1391 | ldrb r12, [r5], #1 | |
1392 | subs r2, r2, #1 | |
1393 | eor r12, r12, r1 | |
1394 | orr r0, r0, r12 | |
1395 | bne 0b | |
1396 | ||
1397 | #elif defined(__aarch64__) | |
1398 | ||
1399 | 0: ldrb w16, [x4], #1 | |
1400 | ldrb w17, [x5], #1 | |
1401 | sub x2, x2, #1 | |
1402 | eor w16, w16, w17 | |
1403 | orr w0, w0, w16 | |
1404 | cbnz x2, 0b | |
1405 | ||
1406 | #else | |
1407 | notimpl | |
1408 | #endif | |
1409 | ||
06297a93 MW |
1410 | ret |
1411 | ||
1412 | endproc | |
1413 | ||
1414 | proc x12 | |
1415 | ||
1416 | // an obtuse way of adding two registers. for any bit position, a | |
1417 | // OR d is set if and only if at least one of a and d has a bit set | |
1418 | // in that position, and a AND d is set if and only if both have a | |
1419 | // bit set in that position. essentially, then, what we've done is | |
1420 | // move all of the set bits in d to a, unless there's already a bit | |
1421 | // there. this clearly doesn't change the sum. | |
1422 | ||
90c4eee3 MW |
1423 | #if defined(__x86_64__) |
1424 | ||
06297a93 MW |
1425 | mov rcx, rdx // c' = d |
1426 | and rdx, rax // d' = a AND d | |
1427 | or rax, rcx // a' = a OR d | |
1428 | add rax, rdx | |
1429 | ||
90c4eee3 MW |
1430 | #elif defined(__i386__) |
1431 | ||
1432 | mov ecx, edx // c' = d | |
1433 | and edx, eax // d' = a AND d | |
1434 | or eax, ecx // a' = a OR d | |
1435 | add eax, edx | |
1436 | ||
1437 | #elif defined(__arm__) | |
1438 | ||
1439 | and r2, r0, r3 // c' = a AND d | |
1440 | orr r0, r0, r3 // a' = a OR d | |
1441 | add r0, r0, r2 | |
1442 | ||
1443 | #elif defined(__aarch64__) | |
1444 | ||
1445 | and x2, x0, x3 // c' = a AND d | |
1446 | orr x0, x0, x3 // a' = a OR d | |
1447 | add x0, x0, x2 | |
1448 | ||
1449 | #else | |
1450 | notimpl | |
1451 | #endif | |
1452 | ||
06297a93 MW |
1453 | ret |
1454 | ||
1455 | endproc | |
1456 | ||
1457 | proc x13 | |
1458 | ||
1459 | // ok, so this is a really obtuse way of adding a and b; the result | |
1460 | // is in a and d. but why does it work? | |
1461 | ||
90c4eee3 MW |
1462 | #if defined(__x86_64__) |
1463 | ||
06297a93 MW |
1464 | mov rcx, 0x40 // carry chains at most 64 long |
1465 | 0: mov rdx, rax // copy a' | |
1466 | xor rax, rbx // low bits of each bitwise sum | |
1467 | and rbx, rdx // carry bits from each bitwise sum | |
90c4eee3 MW |
1468 | shl rbx, 1 // carry them into next position |
1469 | loop 0b | |
1470 | ||
1471 | #elif defined(__i386__) | |
1472 | ||
1473 | mov ecx, 0x40 // carry chains at most 64 long | |
1474 | 0: mov edx, eax // copy a' | |
1475 | xor eax, ebx // low bits of each bitwise sum | |
1476 | and ebx, edx // carry bits from each bitwise sum | |
1477 | shl ebx, 1 // carry them into next position | |
06297a93 MW |
1478 | loop 0b |
1479 | ||
90c4eee3 MW |
1480 | #elif defined(__arm__) |
1481 | ||
1482 | mov r2, #0x40 | |
1483 | 0: and r3, r0, r1 | |
1484 | subs r2, r2, #1 | |
1485 | eor r0, r0, r1 | |
1486 | lsl r1, r3, #1 | |
1487 | bne 0b | |
1488 | ||
1489 | #elif defined(__aarch64__) | |
1490 | ||
1491 | mov x2, #0x40 | |
1492 | 0: and x3, x0, x1 | |
1493 | sub x2, x2, #1 | |
1494 | eor x0, x0, x1 | |
1495 | lsl x1, x3, #1 | |
1496 | cbnz x2, 0b | |
1497 | ||
1498 | #else | |
1499 | notimpl | |
1500 | #endif | |
1501 | ||
06297a93 MW |
1502 | ret |
1503 | ||
1504 | endproc | |
1505 | ||
1506 | proc x14 | |
1507 | ||
1508 | // floor((a + d)/2), like x08. | |
1509 | ||
90c4eee3 MW |
1510 | #if defined(__x86_64__) |
1511 | ||
06297a93 MW |
1512 | mov rcx, rax // copy a for later |
1513 | and rcx, rdx // carry bits | |
1514 | ||
1515 | xor rax, rdx // low bits of each bitwise sum | |
1516 | shr rax, 1 // divide by 2; carries now in place | |
1517 | ||
1518 | add rax, rcx // add the carries; done | |
1519 | ||
90c4eee3 MW |
1520 | #elif defined(__i386__) |
1521 | ||
1522 | mov ecx, eax // copy a for later | |
1523 | and ecx, edx // carry bits | |
1524 | ||
1525 | xor eax, edx // low bits of each bitwise sum | |
1526 | shr eax, 1 // divide by 2; carries now in place | |
1527 | ||
1528 | add eax, ecx // add the carries; done | |
1529 | ||
1530 | #elif defined(__arm__) | |
1531 | ||
1532 | and r2, r0, r3 | |
1533 | eor r0, r0, r3 | |
1534 | add r0, r2, r0, lsr #1 | |
1535 | ||
1536 | #elif defined(__aarch64__) | |
1537 | ||
1538 | and x2, x0, x3 | |
1539 | eor x0, x0, x3 | |
1540 | add x0, x2, x0, lsr #1 | |
1541 | ||
1542 | #else | |
1543 | notimpl | |
1544 | #endif | |
1545 | ||
06297a93 MW |
1546 | ret |
1547 | ||
1548 | endproc | |
1549 | ||
1550 | proc x15 | |
1551 | ||
1552 | // sign extension 32 -> 64 bits. | |
1553 | ||
90c4eee3 MW |
1554 | #if defined(__x86_64__) |
1555 | ||
1556 | movsx rbx, eax // like this? | |
06297a93 MW |
1557 | |
1558 | mov rdx, 0xffffffff80000000 | |
1559 | add rax, rdx // if bit 31 of a is set then bits | |
1560 | // 31--63 of a' are clear; otherwise, | |
1561 | // these bits are all set -- which is | |
1562 | // exactly backwards | |
1563 | xor rax, rdx // so fix it | |
1564 | ||
90c4eee3 MW |
1565 | #elif defined(__i386__) |
1566 | ||
1567 | movsx ebx, ax // like this? | |
1568 | ||
1569 | mov edx, 0xffff8000 | |
1570 | add eax, edx // if bit 31 of a is set then bits | |
1571 | // 31--63 of a' are clear; otherwise, | |
1572 | // these bits are all set -- which is | |
1573 | // exactly backwards | |
1574 | xor eax, edx // so fix it | |
1575 | ||
1576 | #elif defined(__arm__) | |
1577 | ||
1578 | sxth r1, r0 // like this | |
1579 | ||
1580 | mov r12, #0x80000000 | |
1581 | add r0, r0, r12, asr #16 | |
1582 | eor r0, r0, r12, asr #16 | |
1583 | ||
1584 | #elif defined(__aarch64__) | |
1585 | ||
1586 | sxtw x1, w0 // like this | |
1587 | ||
1588 | mov x16, #0xffffffff80000000 | |
1589 | add x0, x0, x16 | |
1590 | eor x0, x0, x16 | |
1591 | ||
1592 | #else | |
1593 | notimpl | |
1594 | #endif | |
1595 | ||
06297a93 MW |
1596 | ret |
1597 | ||
1598 | endproc | |
1599 | ||
1600 | proc x16 | |
1601 | ||
90c4eee3 MW |
1602 | // ??? i don't know why you'd want to calculate this. |
1603 | ||
1604 | #if defined(__x86_64__) | |
06297a93 MW |
1605 | |
1606 | xor rax, rbx // a' = a XOR b | |
1607 | xor rbx, rcx // b' = b XOR c | |
1608 | mov rsi, rax // t = a XOR b | |
1609 | add rsi, rbx // t = (a XOR b) + (b XOR c) | |
1610 | cmovc rax, rbx // a' = cf ? b XOR c : a XOR b | |
1611 | xor rax, rbx // a' = cf ? 0 : a XOR c | |
1612 | cmp rax, rsi | |
1613 | ||
90c4eee3 MW |
1614 | #elif defined(__i386__) |
1615 | ||
1616 | xor eax, ebx // a' = a XOR b | |
1617 | xor ebx, ecx // b' = b XOR c | |
1618 | mov esi, eax // t = a XOR b | |
1619 | add esi, ebx // t = (a XOR b) + (b XOR c) | |
1620 | cmovc eax, ebx // a' = cf ? b XOR c : a XOR b | |
1621 | xor eax, ebx // a' = cf ? 0 : a XOR c | |
1622 | cmp eax, esi | |
1623 | ||
1624 | #elif defined(__arm__) | |
1625 | ||
1626 | eor r0, r0, r1 | |
1627 | eor r1, r1, r2 | |
1628 | adds r4, r0, r1 | |
1629 | movcs r0, r1 | |
1630 | eor r0, r0, r1 | |
1631 | cmp r0, r4 | |
1632 | ||
1633 | #elif defined(__aarch64__) | |
1634 | ||
1635 | eor x0, x0, x1 | |
1636 | eor x1, x1, x2 | |
1637 | adds x4, x0, x1 | |
1638 | cmov.cs x0, x1 | |
1639 | eor x0, x0, x1 | |
1640 | cmp x0, x4 | |
1641 | ||
1642 | #else | |
1643 | notimpl | |
1644 | #endif | |
1645 | ||
06297a93 MW |
1646 | ret |
1647 | ||
1648 | endproc | |
1649 | ||
1650 | proc x17 | |
1651 | ||
90c4eee3 MW |
1652 | // absolute value |
1653 | ||
1654 | #if defined(__x86_64__) | |
1655 | ||
1656 | cqo // d = a < 0 ? -1 : 0 | |
1657 | xor rax, rdx // a' = a < 0 ? -a - 1 : a | |
1658 | sub rax, rdx // a' = a < 0 ? -a : a | |
1659 | ||
1660 | #elif defined(__i386__) | |
1661 | ||
1662 | cdq // d = a < 0 ? -1 : 0 | |
1663 | xor eax, edx // a' = a < 0 ? -a - 1 : a | |
1664 | sub eax, edx // a' = a < 0 ? -a : a | |
1665 | ||
1666 | #elif defined(__arm__) | |
1667 | ||
1668 | // direct approach | |
1669 | movs r1, r0 | |
1670 | rsbmi r1, r0, #0 | |
1671 | ||
1672 | // faithful-ish conversion | |
1673 | eor r3, r0, r0, asr #31 | |
1674 | sub r0, r3, r0, asr #31 | |
1675 | ||
1676 | #elif defined(__aarch64__) | |
1677 | ||
1678 | // direct approach | |
1679 | tst x0, #1 << 63 | |
1680 | cneg.ne x1, x0 | |
1681 | ||
1682 | // faithful-ish conversion | |
1683 | eor x3, x0, x0, asr #63 | |
1684 | sub x0, x3, x0, asr #63 | |
1685 | ||
1686 | #else | |
1687 | notimpl | |
1688 | #endif | |
1689 | ||
1690 | ret | |
06297a93 MW |
1691 | |
1692 | endproc | |
1693 | ||
1694 | proc x18 | |
1695 | ||
90c4eee3 MW |
1696 | // should always set sf, clear zf, unless we get rescheduled to a |
1697 | // different core. | |
1698 | ||
1699 | #if defined(__x86_64__) | |
1700 | ||
1701 | rdtsc // d || a = cycles | |
1702 | shl rdx, 0x20 | |
1703 | or rax, rdx // a = cycles | |
1704 | mov rcx, rax // c = cycles | |
1705 | ||
1706 | rdtsc // d || a = cycles' | |
1707 | shl rdx, 0x20 | |
1708 | or rax, rdx // a = cycles' | |
1709 | ||
1710 | cmp rcx, rax | |
1711 | ||
1712 | #elif defined(__i386__) | |
1713 | ||
1714 | rdtsc // d || a = cycles | |
1715 | mov ebx, eax | |
1716 | mov ecx, edx // c || b = cycles | |
1717 | ||
1718 | rdtsc // d || a = cycles' | |
1719 | ||
1720 | sub ebx, eax | |
1721 | sbb ecx, edx | |
1722 | ||
1723 | #elif defined(__arm__) | |
1724 | ||
1725 | // cycle clock not available in user mode | |
1726 | mrrc p15, 0, r0, r1, c9 | |
1727 | mrrc p15, 0, r2, r3, c9 | |
1728 | subs r0, r0, r2 | |
1729 | sbcs r1, r1, r3 | |
1730 | ||
1731 | #elif defined(__aarch64__) | |
1732 | ||
1733 | // cycle clock not available in user mode | |
1734 | mrs x0, pmccntr_el0 | |
1735 | mrs x1, pmccntr_el0 | |
1736 | cmp x0, x1 | |
1737 | ||
1738 | #else | |
1739 | notimpl | |
1740 | #endif | |
1741 | ||
1742 | ret | |
06297a93 MW |
1743 | |
1744 | endproc | |
1745 | ||
1746 | proc x19 | |
1747 | ||
90c4eee3 MW |
1748 | // stupid way to capture a pointer to inline data and jump past it. |
1749 | // confuses the return-address predictor something chronic. worse | |
1750 | // because amd64 calling convention doesn't usually pass arguments on | |
1751 | // the stack. | |
1752 | ||
1753 | #if defined(__x86_64__) | |
1754 | ||
1755 | call 8f | |
1756 | .string "hello world!\n\0" | |
1757 | 8: call print_str | |
1758 | add rsp, 8 | |
1759 | ret | |
1760 | ||
1761 | print_str: | |
1762 | // actually implement this ridiculous thing | |
1763 | mov rsi, [rsp + 8] | |
1764 | xor edx, edx | |
1765 | 0: mov al, [rsi + rdx] | |
1766 | inc rdx | |
1767 | cmp al, 0 | |
1768 | jnz 0b | |
1769 | mov eax, SYS_write | |
1770 | mov edi, 1 | |
1771 | dec rdx | |
1772 | syscall // clobbers r11 :-( | |
1773 | ret | |
1774 | ||
1775 | #elif defined(__i386__) | |
1776 | ||
1777 | call 8f | |
1778 | .string "hello world!\n\0" | |
1779 | 8: call print_str | |
1780 | add esp, 4 | |
1781 | ret | |
1782 | ||
1783 | print_str: | |
1784 | // actually implement this ridiculous thing | |
1785 | mov ecx, [esp + 4] | |
1786 | xor edx, edx | |
1787 | 0: mov al, [ecx + edx] | |
1788 | inc edx | |
1789 | cmp al, 0 | |
1790 | jnz 0b | |
1791 | mov eax, SYS_write | |
1792 | mov ebx, 1 | |
1793 | dec edx | |
1794 | int 0x80 | |
1795 | ret | |
1796 | ||
1797 | #elif defined(__arm__) | |
1798 | ||
1799 | // why am i doing this? | |
1800 | stmfd r13!, {r14} | |
1801 | bl 8f | |
1802 | .string "hello world!\n\0" | |
1803 | .balign 4 | |
1804 | 8: mov r1, r14 // might as well make it easy on myself | |
1805 | bl print_str | |
1806 | ldmfd r13!, {pc} | |
1807 | ||
1808 | print_str: | |
1809 | mov r2, #0 | |
1810 | 0: ldrb r0, [r1, r2] | |
1811 | cmp r0, #0 | |
1812 | addne r2, r2, #1 | |
1813 | bne 0b | |
1814 | mov r0, #1 | |
1815 | mov r7, #SYS_write | |
1816 | swi 0 | |
1817 | bx r14 | |
1818 | ||
1819 | #elif defined(__aarch64__) | |
1820 | ||
1821 | // why am i doing this? | |
1822 | str x30, [sp, #-16]! | |
1823 | bl 8f | |
1824 | .string "hello world!\n\0" | |
1825 | .balign 4 | |
1826 | 8: mov x1, x30 // might as well make it easy on myself | |
1827 | bl print_str | |
1828 | ldr x30, [sp], #16 | |
1829 | ret | |
1830 | ||
1831 | print_str: | |
1832 | mov x2, #0 | |
1833 | 0: ldrb w0, [x1, x2] | |
1834 | cmp w0, #0 | |
1835 | cinc.ne x2, x2 | |
1836 | b.ne 0b | |
1837 | mov x0, #1 | |
1838 | mov x8, #SYS_write | |
1839 | svc #0 | |
1840 | ret | |
1841 | ||
1842 | #else | |
1843 | notimpl | |
1844 | #endif | |
06297a93 MW |
1845 | |
1846 | endproc | |
1847 | ||
1848 | proc x1a | |
1849 | ||
90c4eee3 MW |
1850 | // collect the current instruction-pointer address. this was an old |
1851 | // 32-bit i386 trick for position-independent code, but (a) it | |
1852 | // confuses the return predictor, and (b) amd64 has true pc-relative | |
1853 | // addressing. | |
1854 | ||
1855 | #if defined(__x86_64__) | |
1856 | ||
1857 | // the actual example | |
1858 | call 0f | |
1859 | 0: pop rax | |
1860 | ||
1861 | // the modern i386 trick doesn't confuse the return-address | |
1862 | // predictor. | |
1863 | call calladdr_rbx | |
1864 | sub rbx, . - 0b | |
1865 | ||
1866 | // but rip-relative addressing is even better | |
1867 | lea rcx, [rip + 0b] | |
1868 | ||
1869 | ret | |
1870 | ||
1871 | calladdr_rbx: | |
1872 | mov rbx, [rsp] | |
1873 | ret | |
1874 | ||
1875 | #elif defined(__i386__) | |
1876 | ||
1877 | // the actual example | |
1878 | call 0f | |
1879 | 0: pop eax | |
1880 | ||
1881 | // the modern i386 trick doesn't confuse the return-address | |
1882 | // predictor. | |
1883 | call get_pc_ebx | |
1884 | sub ebx, . - 0b | |
1885 | ||
1886 | ret | |
1887 | ||
1888 | #elif defined(__arm__) | |
1889 | ||
1890 | stmfd r13!, {r14} | |
1891 | ||
1892 | bl 0f | |
1893 | 0: mov r0, r14 | |
1894 | ||
1895 | bl return | |
1896 | sub r1, r14, #. - 0b | |
1897 | ||
1898 | adr r2, 0b | |
1899 | ||
1900 | ldmfd r13!, {pc} | |
1901 | ||
1902 | return: bx r14 | |
1903 | ||
1904 | #elif defined(__aarch64__) | |
1905 | ||
1906 | str x30, [sp, #-16]! | |
1907 | ||
1908 | // we can do all of the above using a64 | |
1909 | bl 0f | |
1910 | 0: mov x0, x30 | |
1911 | ||
1912 | bl return | |
1913 | sub x1, x30, #. - 0b | |
1914 | ||
1915 | adr x2, 0b | |
1916 | ||
1917 | ldr x30, [sp], #16 | |
1918 | return: ret | |
1919 | ||
1920 | #else | |
1921 | notimpl | |
1922 | #endif | |
06297a93 MW |
1923 | |
1924 | endproc | |
1925 | ||
1926 | proc x1b | |
1927 | ||
90c4eee3 MW |
1928 | #if defined(__x86_64__) |
1929 | ||
1930 | // retpolines: an mitigation against adversarially influenced | |
1931 | // speculative execution at indirect branches. if an adversary can | |
1932 | // prepare a branch-target buffer entry matching an indirect branch | |
1933 | // in the victim's address space then they can cause the victim to | |
1934 | // /speculatively/ (but not architecturally) execute any code in | |
1935 | // their address space, possibly leading to leaking secrets through | |
1936 | // the cache. retpolines aren't susceptible to this because the | |
1937 | // predicted destination address is from the return-prediction stack | |
1938 | // which the adversary can't prime. the performance penalty is still | |
1939 | // essentially a branch misprediction -- for this return, and | |
1940 | // possibly all others already stacked. | |
1941 | ||
1942 | // (try not to crash) | |
1943 | lea rax, [rip + 9f] | |
1944 | ||
1945 | push rax | |
1946 | 9: ret | |
1947 | ||
1948 | #elif defined(__i386__) | |
1949 | ||
1950 | call get_pc_ebx | |
1951 | lea eax, [ebx + 9f - .] | |
1952 | ||
1953 | push eax | |
1954 | 9: ret | |
1955 | ||
1956 | #elif defined(__arm__) | |
1957 | ||
1958 | stmfd r13!, {r14} | |
1959 | ||
1960 | adr r14, 8f | |
1961 | bx r14 | |
1962 | ||
1963 | 8: ldmfd r13!, {pc} | |
1964 | ||
1965 | #elif defined(__aarch64__) | |
1966 | ||
1967 | str x30, [sp, #-16]! | |
1968 | ||
1969 | adr x30, 8f | |
1970 | ret | |
1971 | ||
1972 | 8: ldr x30, [sp], #16 | |
1973 | ret | |
1974 | ||
1975 | #else | |
1976 | notimpl | |
1977 | #endif | |
06297a93 MW |
1978 | |
1979 | endproc | |
1980 | ||
1981 | proc x1c | |
1982 | ||
90c4eee3 MW |
1983 | // ok, having a hard time seeing a use for this. the most important |
1984 | // thing to note is that sp is set from `pop' /after/ it's | |
1985 | // incremented. | |
1986 | ||
1987 | #if defined(__x86_64__) | |
1988 | ||
1989 | // try not to crash | |
1990 | mov rax, rsp | |
1991 | and rsp, -16 | |
1992 | push rax | |
1993 | ||
1994 | pop rsp | |
1995 | ||
1996 | // check it worked | |
1997 | mov rbx, rsp | |
1998 | ret | |
1999 | ||
2000 | #elif defined(__i386__) | |
2001 | ||
2002 | // try not to crash | |
2003 | mov eax, esp | |
2004 | and esp, -16 | |
2005 | push eax | |
2006 | ||
2007 | pop esp | |
2008 | ||
2009 | // check it worked | |
2010 | mov ebx, esp | |
2011 | ret | |
2012 | ||
2013 | #elif defined(__arm__) | |
2014 | ||
2015 | // not even going to dignify this | |
2016 | notimpl | |
2017 | ||
2018 | #elif defined(__aarch64__) | |
2019 | ||
2020 | // not even going to dignify this | |
2021 | notimpl | |
2022 | ||
2023 | #else | |
2024 | notimpl | |
2025 | #endif | |
06297a93 MW |
2026 | |
2027 | endproc | |
2028 | ||
2029 | proc x1d | |
2030 | ||
90c4eee3 MW |
2031 | // monumentally cheesy way to copy 8 n bytes from buff1 to buff2. |
2032 | // also clobbers words at buff2 + 8 n and buff2 - 8 for good measure. | |
2033 | ||
2034 | n = 4 | |
2035 | ||
2036 | #if defined(__x86_64__) | |
2037 | ||
2038 | mov rax, rsp // safekeeping | |
2039 | ||
2040 | // we're toast if we get hit by a signal now. fingers crossed... | |
2041 | .if 0 | |
2042 | mov rsp, buff2 + 8*n + 8 | |
2043 | mov rbp, buff1 + 8*n | |
2044 | .else | |
2045 | lea rsp, [rdi + 8*n + 16] | |
2046 | lea rbp, [rsi + 8*n] | |
2047 | .endif | |
2048 | enter 0, n + 1 | |
2049 | ||
2050 | // precise action: | |
2051 | // | |
2052 | // +---------+ +---------+ | |
2053 | // rbp -> | ??? | rsp -> | ??? | | |
2054 | // +---------+ +---------+ | |
2055 | // | w_{n-1} | | rbp | <- rbp' | |
2056 | // +---------+ +---------+ | |
2057 | // | ... | | w_{n-1} | | |
2058 | // +---------+ +---------+ | |
2059 | // | w_1 | | ... | | |
2060 | // +---------+ +---------+ | |
2061 | // | w_0 | | w_1 | | |
2062 | // +---------+ +---------+ | |
2063 | // | w_0 | | |
2064 | // +---------+ | |
2065 | // | rbp' | <- rsp' | |
2066 | // +---------+ | |
2067 | ||
2068 | mov rdx, rsp | |
2069 | mov rsp, rax | |
2070 | ||
2071 | #elif defined(__i386__) | |
2072 | ||
2073 | mov eax, esp // safekeeping | |
2074 | ||
2075 | // we're toast if we get hit by a signal now. fingers crossed... | |
2076 | .if 0 | |
2077 | mov esp, buff2 + 4*n + 4 | |
2078 | mov ebp, buff1 + 4*n | |
2079 | .else | |
2080 | lea esp, [edi + 4*n + 8] | |
2081 | lea ebp, [esi + 4*n] | |
2082 | .endif | |
2083 | enter 0, n + 1 | |
2084 | ||
2085 | mov edx, esp | |
2086 | mov esp, eax | |
2087 | ||
2088 | #elif defined(__arm__) | |
2089 | ||
2090 | add r4, r4, #4*n | |
2091 | add r5, r5, #4*n + 8 | |
2092 | ||
2093 | str r4, [r5, #-4]! | |
2094 | .rept n/2 | |
2095 | ldrd r0, r1, [r4, #-8]! | |
2096 | strd r0, r1, [r5, #-8]! | |
2097 | .endr | |
2098 | add r4, r5, #4*n | |
2099 | str r4, [r5, #-4]! | |
2100 | ||
2101 | #elif defined(__aarch64__) | |
2102 | ||
2103 | // omgwtf. let's not actually screw with the stack pointer. | |
2104 | ||
2105 | add x4, x4, #8*n | |
2106 | add x5, x5, #8*n + 16 | |
2107 | ||
2108 | str x4, [x5, #-8]! | |
2109 | .rept n/2 | |
2110 | ldp x16, x17, [x4, #-16]! | |
2111 | stp x16, x17, [x5, #-16]! | |
2112 | .endr | |
2113 | add x4, x5, #8*n | |
2114 | str x4, [x5, #-8]! | |
2115 | ||
2116 | #else | |
2117 | notimpl | |
2118 | #endif | |
2119 | ||
2120 | ret | |
06297a93 MW |
2121 | |
2122 | endproc | |
2123 | ||
2124 | proc x1e | |
2125 | ||
90c4eee3 MW |
2126 | // convert nibble value to (uppercase) hex; other input values yield |
2127 | // nonsense. | |
2128 | ||
2129 | #if defined(__x86_64__) | |
2130 | ||
2131 | // das doesn't work in 64-bit mode; best i can come up with | |
2132 | mov edx, eax | |
2133 | add al, '0' | |
2134 | add dl, 'A' - 10 | |
2135 | cmp al, '9' + 1 | |
2136 | cmovae eax, edx | |
2137 | ||
2138 | #elif defined(__i386__) | |
2139 | ||
2140 | cmp al, 0x0a // cf = 1 iff a < 10 | |
2141 | sbb al, 0x69 // if 0 <= a < 10, a' = a - 0x6a, so | |
2142 | // 0x96 <= a' < 0x70, setting af, cf | |
2143 | // if 10 <= a < 16, a' = a - 0x69, so | |
2144 | // 0x71 <= a' < 0x77, setting cf but | |
2145 | // clearing af | |
2146 | das // if 0 <= a < 10, then af and cf are | |
2147 | // both set, so set subtract 0x66 | |
2148 | // from a' leaving 0x30 <= a' < 0x3a; | |
2149 | // if 10 <= a < 16 then af clear but | |
2150 | // cf set, so subtract 0x60 from a' | |
2151 | // leaving 0x41 <= a' < 0x47 | |
2152 | ||
2153 | #elif defined(__arm__) | |
2154 | ||
2155 | // significantly less tricksy | |
2156 | cmp r0, #10 | |
2157 | addlo r0, r0, #'0' | |
2158 | addhs r0, r0, #'A' - 10 | |
2159 | ||
2160 | #elif defined(__aarch64__) | |
2161 | ||
2162 | // with less versatile conditional execution this is the best we can | |
2163 | // do | |
2164 | cmp w0, #10 | |
2165 | add w16, w0, #'A' - 10 | |
2166 | add w0, w0, #'0' | |
2167 | cmov.hs w0, w16 | |
2168 | ||
2169 | #else | |
2170 | notimpl | |
2171 | #endif | |
2172 | ||
2173 | ret | |
06297a93 MW |
2174 | |
2175 | endproc | |
2176 | ||
2177 | proc x1f | |
2178 | ||
90c4eee3 MW |
2179 | // verify collatz conjecture starting at a; assume a /= 0! |
2180 | ||
2181 | #if defined(__x86_64__) | |
2182 | ||
2183 | 0: bsf rcx, rax // clobber c if a = 0 | |
2184 | shr rax, cl // a = 2^c a' | |
2185 | cmp rdx, 0 | |
2186 | je 1f | |
2187 | stosq | |
2188 | dec rdx | |
2189 | 1: | |
2190 | cmp rax, 1 // done? | |
2191 | je 9f | |
2192 | lea rax, [2*rax + rax + 1] // a' = 3 a' + 1 | |
2193 | jmp 0b // again | |
2194 | ||
2195 | 9: ret | |
2196 | ||
2197 | #elif defined(__i386__) | |
2198 | ||
2199 | 0: bsf ecx, eax // clobber c if a = 0 | |
2200 | shr eax, cl // a = 2^c a' | |
2201 | cmp edx, 0 | |
2202 | je 1f | |
2203 | stosd | |
2204 | dec edx | |
2205 | 1: | |
2206 | cmp eax, 1 // done? | |
2207 | je 9f | |
2208 | lea eax, [2*eax + eax + 1] // a' = 3 a' + 1 | |
2209 | jmp 0b // again | |
2210 | ||
2211 | 9: ret | |
2212 | ||
2213 | #elif defined(__arm__) | |
2214 | ||
2215 | // rbit introduced in armv7 | |
2216 | 0: rbit r2, r0 | |
2217 | clz r2, r2 | |
2218 | mov r0, r0, lsr r2 // a = 2^c a' | |
2219 | cmp r3, #0 | |
2220 | strne r0, [r5], #4 | |
2221 | subne r3, r3, #1 | |
2222 | cmp r0, #1 | |
2223 | adcne r0, r0, r0, lsl #1 // a' = 3 a' + 1 (because c set) | |
2224 | bne 0b | |
2225 | ||
2226 | ret | |
2227 | ||
2228 | #elif defined(__aarch64__) | |
2229 | ||
2230 | 0: rbit w2, w0 | |
2231 | clz w2, w2 | |
2232 | lsr w0, w0, w2 // a = 2^c a' | |
2233 | cmp x3, #0 | |
2234 | beq 1f | |
2235 | str x0, [x5], #8 | |
2236 | sub x3, x3, #1 | |
2237 | 1: | |
2238 | cmp w0, #1 | |
2239 | add w16, w0, w0, lsl #1 // t = 3 a' + 1 (because c set) | |
2240 | csinc.eq w0, w0, w16 | |
2241 | b.ne 0b | |
2242 | ||
2243 | ret | |
2244 | ||
2245 | #else | |
2246 | notimpl | |
2247 | #endif | |
06297a93 MW |
2248 | |
2249 | endproc | |
2250 | ||
90c4eee3 MW |
2251 | ///-------------------------------------------------------------------------- |
2252 | /// 0x20--0x2f | |
2253 | ||
06297a93 MW |
2254 | proc x20 |
2255 | ||
90c4eee3 MW |
2256 | // calculate 1337 a slowly |
2257 | ||
2258 | #if defined(__x86_64__) | |
2259 | ||
2260 | // original version | |
2261 | mov rcx, rax // c = a | |
2262 | shl rcx, 2 // c = 4 a | |
2263 | add rcx, rax // c = 5 a | |
2264 | shl rcx, 3 // c = 40 a | |
2265 | add rcx, rax // c = 41 a | |
2266 | shl rcx, 1 // c = 82 a | |
2267 | add rcx, rax // c = 83 a | |
2268 | shl rcx, 1 // c = 166 a | |
2269 | add rcx, rax // c = 167 a | |
2270 | shl rcx, 3 // c = 1336 a | |
2271 | add rcx, rax // c = 1337 a | |
2272 | ||
2273 | // a quick way | |
2274 | lea rdx, [2*rax + rax] // t = 3 a | |
2275 | shl rdx, 6 // t = 192 a | |
2276 | sub rdx, rax // t = 191 a | |
2277 | lea rbx, [8*rdx] // b = 1528 a | |
2278 | sub rbx, rdx // b = 1337 a | |
2279 | ||
2280 | #elif defined(__i386__) | |
2281 | ||
2282 | // original version | |
2283 | mov ecx, eax // c = a | |
2284 | shl ecx, 2 // c = 4 a | |
2285 | add ecx, eax // c = 5 a | |
2286 | shl ecx, 3 // c = 40 a | |
2287 | add ecx, eax // c = 41 a | |
2288 | shl ecx, 1 // c = 82 a | |
2289 | add ecx, eax // c = 83 a | |
2290 | shl ecx, 1 // c = 166 a | |
2291 | add ecx, eax // c = 167 a | |
2292 | shl ecx, 3 // c = 1336 a | |
2293 | add ecx, eax // c = 1337 a | |
2294 | ||
2295 | // a quick way | |
2296 | lea edx, [2*eax + eax] // t = 3 a | |
2297 | shl edx, 6 // t = 192 a | |
2298 | sub edx, eax // t = 191 a | |
2299 | lea ebx, [8*edx] // b = 1528 a | |
2300 | sub ebx, edx // b = 1337 a | |
2301 | ||
2302 | #elif defined(__arm__) | |
2303 | ||
2304 | // original version, ish | |
2305 | add r2, r0, r0, lsl #2 // c = 5 a | |
2306 | add r2, r0, r2, lsl #3 // c = 41 a | |
2307 | add r2, r0, r2, lsl #1 // c = 83 a | |
2308 | add r2, r0, r2, lsl #1 // c = 167 a | |
2309 | add r2, r0, r2, lsl #3 // c = 1337 a | |
2310 | ||
2311 | // quicker way | |
2312 | add r1, r0, r0, lsl #1 // b = 3 a | |
2313 | rsb r1, r0, r1, lsl #6 // b = 191 a | |
2314 | rsb r1, r1, r1, lsl #3 // b = 1337 a | |
2315 | ||
2316 | #elif defined(__aarch64__) | |
2317 | ||
2318 | // original version, ish | |
2319 | add x2, x0, x0, lsl #2 // c = 5 a | |
2320 | add x2, x0, x2, lsl #3 // c = 41 a | |
2321 | add x2, x0, x2, lsl #1 // c = 83 a | |
2322 | add x2, x0, x2, lsl #1 // c = 167 a | |
2323 | add x2, x0, x2, lsl #3 // c = 1337 a | |
2324 | ||
2325 | // sleazy because no rsb | |
2326 | add x1, x0, x0, lsl #1 // b = 3 a | |
2327 | sub x1, x0, x1, lsl #6 // b = -191 a | |
2328 | sub x1, x1, x1, lsl #3 // b = 1337 a | |
2329 | ||
2330 | #else | |
2331 | notimpl | |
2332 | #endif | |
06297a93 MW |
2333 | |
2334 | ret | |
2335 | ||
2336 | endproc | |
2337 | ||
2338 | proc x21 | |
2339 | ||
90c4eee3 MW |
2340 | // multiply complex numbers a + b i and c + d i |
2341 | // | |
2342 | // (a + b i) (c + d i) = (a c - b d) + (a d + b c) i | |
2343 | // | |
2344 | // somewhat slick approach uses only three multiplications | |
2345 | ||
2346 | #if defined(__x86_64__) | |
2347 | ||
2348 | mov rsi, rax // t = a | |
2349 | add rax, rbx // a' = a + b | |
2350 | mov rdi, rdx // u = d | |
2351 | sub rdx, rcx // d' = d - c | |
2352 | add rdi, rcx // u = c + d | |
2353 | ||
2354 | imul rax, rcx // a' = c (a + b) | |
2355 | imul rsi, rdx // t = a (d - c) | |
2356 | imul rdi, rbx // u = b (c + d) | |
2357 | ||
2358 | add rsi, rax // t = a (d - c) + c (a + b) | |
2359 | mov rbx, rsi // b' = a (d - c) + c (a + b) | |
2360 | // = a d + b c | |
2361 | sub rax, rdi // a' = c (a + b) - b (c + d) | |
2362 | // = a c - b d | |
2363 | ||
2364 | #elif defined(__i386__) | |
2365 | ||
2366 | mov esi, eax // t = a | |
2367 | add eax, ebx // a' = a + b | |
2368 | mov edi, edx // u = d | |
2369 | sub edx, ecx // d' = d - c | |
2370 | add edi, ecx // u = c + d | |
2371 | ||
2372 | imul eax, ecx // a' = c (a + b) | |
2373 | imul esi, edx // t = a (d - c) | |
2374 | imul edi, ebx // u = b (c + d) | |
2375 | ||
2376 | add esi, eax // t = a (d - c) + c (a + b) | |
2377 | mov ebx, esi // b' = a (d - c) + c (a + b) | |
2378 | // = a d + b c | |
2379 | sub eax, edi // a' = c (a + b) - b (c + d) | |
2380 | // = a c - b d | |
2381 | ||
2382 | #elif defined(__arm__) | |
2383 | ||
2384 | add r4, r0, r1 // t = a + b | |
2385 | add r5, r2, r3 // u = c + d | |
2386 | sub r3, r3, r2 // d' = d - c | |
2387 | ||
2388 | // mls introduced in armv7 | |
2389 | mul r4, r4, r2 // t = c (a + b) | |
2390 | mov r2, r1 // c' = a (bah!) | |
2391 | mla r1, r0, r3, r4 // b' = a (d - c) + c (a + b) | |
2392 | // = a d + b c | |
2393 | mls r0, r2, r5, r4 // a' = c (a + b) - b (c + d) | |
2394 | // = a c - b d | |
2395 | ||
2396 | #elif defined(__aarch64__) | |
2397 | ||
2398 | add x4, x0, x1 // t = a + b | |
2399 | add x5, x2, x3 // u = c + d | |
2400 | sub x3, x3, x2 // d' = d - c | |
2401 | ||
2402 | // mls intxoduced in axmv7 | |
2403 | mul x4, x4, x2 // t = c (a + b) | |
2404 | mov x2, x1 // c' = a (bah!) | |
2405 | madd x1, x0, x3, x4 // b' = a (d - c) + c (a + b) | |
2406 | // = a d + b c | |
2407 | msub x0, x2, x5, x4 // a' = c (a + b) - b (c + d) | |
2408 | // = a c - b d | |
2409 | ||
2410 | #else | |
2411 | notimpl | |
2412 | #endif | |
2413 | ||
2414 | ret | |
06297a93 MW |
2415 | |
2416 | endproc | |
2417 | ||
2418 | proc x22 | |
2419 | ||
90c4eee3 MW |
2420 | // divide by 3 |
2421 | ||
2422 | #if defined(__x86_64__) | |
2423 | ||
2424 | mov rdx, 0xaaaaaaaaaaaaaaab // = ceil(2/3 2^64) | |
2425 | mul rdx // d' || a' =~ 2/3 a 2^64 | |
2426 | shr rdx, 1 // d' = floor(a/3) | |
2427 | mov rax, rdx // a' = floor(a/3) | |
2428 | ||
2429 | // we start with 0 <= a < 2^64. write f = ceil(2/3 2^64), so that | |
2430 | // 2/3 < f/2^64 < 2/3 + 1/2^64. then floor(2/3 a) <= floor(a f/2^64) | |
2431 | // <= floor(2/3 a + a/2^64), but a < 2^64 so a/2^64 < 1 and | |
2432 | // floor(a f/2^64) = floor(2/3 a). | |
2433 | ||
2434 | #elif defined(__i386__) | |
2435 | ||
2436 | mov edx, 0xaaaaaaab // = ceil(2/3 2^32) | |
2437 | mul edx // d' || a' =~ 2/3 a 2^32 | |
2438 | shr edx, 1 // d' = floor(a/3) | |
2439 | mov eax, edx // a' = floor(a/3) | |
2440 | ||
2441 | #elif defined(__arm__) | |
2442 | ||
2443 | ldr r12, =0xaaaaaaab | |
2444 | umull r12, r0, r0, r12 | |
2445 | mov r0, r0, lsr #1 | |
2446 | ||
2447 | #elif defined(__aarch64__) | |
2448 | ||
2449 | ldr x16, =0xaaaaaaaaaaaaaaab | |
2450 | umulh x0, x0, x16 | |
2451 | lsr x0, x0, #1 | |
2452 | ||
2453 | #else | |
2454 | notimpl | |
2455 | #endif | |
2456 | ||
2457 | ret | |
06297a93 MW |
2458 | |
2459 | endproc | |
2460 | ||
2461 | proc x23 | |
2462 | ||
90c4eee3 MW |
2463 | #if defined(__x86_64__) |
2464 | ||
2465 | // main loop: shorten a preserving residue class mod 3 | |
2466 | 0: cmp rax, 5 | |
2467 | jbe 8f | |
2468 | // a > 5 | |
2469 | mov rdx, rax // d' = a | |
2470 | shr rdx, 2 // d' = floor(a/4) | |
2471 | and rax, 3 // a = 4 d' + a' (0 <= a' < 4) | |
2472 | add rax, rdx // a' == a (mod 3) but a' < a/4 + 4 | |
2473 | jmp 0b | |
2474 | ||
2475 | // fix up final value 0 <= a < 6: want 0 <= a < 3 | |
2476 | // | |
2477 | // the tricky part is actually a = 3; but the other final cases take | |
2478 | // additional iterations which we can avoid. | |
2479 | 8: cmp rax, 3 // set cf iff a < 3 | |
2480 | cmc // set cf iff a >= 3 | |
2481 | sbb rdx, rdx // d' = a >= 3 ? -1 : 0 | |
2482 | and rdx, 3 // d' = a >= 3 ? 3 : 0 | |
2483 | sub rax, rdx // a' = a - (a >= 3 ? 3 : 0) | |
2484 | // = a (mod 3) | |
2485 | ||
2486 | #elif defined(__i386__) | |
2487 | ||
2488 | // main loop: shorten a preserving residue class mod 3 | |
2489 | 0: cmp eax, 5 | |
2490 | jbe 8f | |
2491 | // a > 5 | |
2492 | mov edx, eax // d' = a | |
2493 | shr edx, 2 // d' = floor(a/4) | |
2494 | and eax, 3 // a = 4 d' + a' (0 <= a' < 4) | |
2495 | add eax, edx // a' == a (mod 3) but a' < a/4 + 4 | |
2496 | jmp 0b | |
2497 | ||
2498 | // fix up final value 0 <= a < 6: want 0 <= a < 3 | |
2499 | // | |
2500 | // the tricky part is actually a = 3; but the other final cases take | |
2501 | // additional iterations which we can avoid. | |
2502 | 8: cmp eax, 3 // set cf iff a < 3 | |
2503 | cmc // set cf iff a >= 3 | |
2504 | sbb edx, edx // d' = a >= 3 ? -1 : 0 | |
2505 | and edx, 3 // d' = a >= 3 ? 3 : 0 | |
2506 | sub eax, edx // a' = a - (a >= 3 ? 3 : 0) | |
2507 | // = a (mod 3) | |
2508 | ||
2509 | #elif defined(__arm__) | |
2510 | ||
2511 | 0: cmp r0, #6 | |
2512 | andhs r12, r0, #3 | |
2513 | addhs r0, r12, r0, lsr #2 | |
2514 | bhs 0b | |
2515 | ||
2516 | cmp r0, #3 | |
2517 | subhs r0, r0, #3 | |
2518 | ||
2519 | #elif defined(__aarch64__) | |
2520 | ||
2521 | 0: cmp x0, #6 | |
2522 | // blunder on through regardless since this doesn't affect the result | |
2523 | and x16, x0, #3 | |
2524 | add x0, x16, x0, lsr #2 | |
2525 | b.hs 0b | |
2526 | ||
2527 | subs x16, x0, #3 | |
2528 | cmov.hs x0, x16 | |
2529 | ||
2530 | #else | |
2531 | notimpl | |
2532 | #endif | |
2533 | ||
2534 | ret | |
06297a93 MW |
2535 | |
2536 | endproc | |
2537 | ||
2538 | proc x24 | |
2539 | ||
90c4eee3 MW |
2540 | // invert (odd) a mod 2^64 |
2541 | // | |
2542 | // suppose a a_i == 1 (mod 2^{2^i}) | |
2543 | // | |
2544 | // clearly good for i = 0, since 2^i = 1 and 2^{2^i} = 2, and a_0 = | |
2545 | // a == 1 (mod 2) by assumption | |
2546 | // | |
2547 | // write a a_i == b_i 2^{2^i} + 1 (mod 2^{2^{i+1}}) | |
2548 | // then b_i == (a a_i - 1)/2^{2^i} (mod 2^{2^i}) | |
2549 | // to lift inverse, we want x such that a x == -b_i (mod 2^{2^i}); | |
2550 | // clearly x = -a_i b_i will do, since a a_i == 1 (mod 2^{2^i}) | |
2551 | // then: | |
2552 | // a_{i+1} = a_i - a_i b_i 2^{2^i} = a_i (1 - (a a_i - 1)) | |
2553 | // = 2 a_i - a a_i^2 | |
2554 | // | |
2555 | // check: | |
2556 | // a a_{i+1} = 2 a a_i - a^2 a_i^2 | |
2557 | // == 2 a a_i - (b_i 2^{2^i} + 1)^2 | |
2558 | // == 2 (b_i 2^{2^i} + 1) - | |
2559 | // (b_i^2 2^{2^{i+1}} + 2 b_i 2^{2^i} + 1) | |
2560 | // == 1 (mod 2^{2^{i+1}}) | |
2561 | ||
2562 | #if defined(__x86_64__) | |
2563 | ||
2564 | // rax // a_0 = a | |
2565 | mov rbx, rax // b' = a | |
2566 | mov rsi, rax // t = a_0 | |
2567 | ||
2568 | 0: | |
2569 | cmp rbp, 0 | |
2570 | je 1f | |
2571 | stosq | |
2572 | dec rbp | |
2573 | 1: | |
2574 | mul rbx // a' = a a_i | |
2575 | mov rcx, rax // c = a a_i | |
2576 | ||
2577 | sub rax, 2 // a' = a a_i - 2 | |
2578 | neg rax // a' = 2 - a a_i | |
2579 | mul rsi // a_{i+1} = a_i (2 - a a_i) | |
2580 | // = 2 a_i - a a_i^2 | |
2581 | mov rsi, rax // t = a_{i+1} | |
2582 | ||
2583 | cmp rcx, 1 // done? | |
2584 | ja 0b // no -- iterate | |
2585 | ||
2586 | #elif defined(__i386__) | |
2587 | ||
2588 | // eax // a_0 = a | |
2589 | mov ebx, eax // b' = a | |
2590 | mov esi, eax // t = a_0 | |
2591 | ||
2592 | 0: | |
2593 | cmp ebp, 0 | |
2594 | je 1f | |
2595 | stosd | |
2596 | dec ebp | |
2597 | 1: | |
2598 | mul ebx // a' = a a_i | |
2599 | mov ecx, eax // c = a a_i | |
2600 | ||
2601 | sub eax, 2 // a' = a a_i - 2 | |
2602 | jb 9f // done if < 2 | |
2603 | neg eax // a' = 2 - a a_i | |
2604 | mul esi // a_{i+1} = a_i (2 - a a_i) | |
2605 | // = 2 a_i - a a_i^2 | |
2606 | mov esi, eax // t = a_{i+1} | |
2607 | ||
2608 | jmp 0b // and iterate | |
2609 | 9: mov eax, esi // restore | |
2610 | ||
2611 | #elif defined(__arm__) | |
2612 | ||
2613 | // r0 // a_0 = a | |
2614 | mov r1, r0 // b' = a | |
2615 | ||
2616 | 0: | |
2617 | cmp r6, #0 | |
2618 | strne r0, [r5], #4 | |
2619 | subne r6, r6, #1 | |
2620 | mul r2, r0, r1 // c = a a_i | |
2621 | rsbs r2, r2, #2 // c = 2 - a a_i | |
2622 | mul r0, r0, r2 // a_{i+1} = a_i (2 - a a_i) | |
2623 | // = 2 a_i - a a_i^2 | |
2624 | blo 0b | |
2625 | ||
2626 | #elif defined(__aarch64__) | |
2627 | ||
2628 | // x0 // a_0 = a | |
2629 | mov x1, x0 // b' = a | |
2630 | mov x16, #2 // because we have no rsb | |
2631 | ||
2632 | 0: | |
2633 | cmp x6, #0 | |
2634 | b.eq 1f | |
2635 | str x0, [x5], #8 | |
2636 | sub x6, x6, #1 | |
2637 | 1: | |
2638 | mul x2, x0, x1 // c = a a_i | |
2639 | subs x2, x16, x2 // c = 2 - a a_i | |
2640 | mul x0, x0, x2 // a_{i+1} = a_i (2 - a a_i) | |
2641 | // = 2 a_i - a a_i^2 | |
2642 | b.lo 0b | |
2643 | ||
2644 | #else | |
2645 | notimpl | |
2646 | #endif | |
2647 | ||
2648 | ret | |
06297a93 MW |
2649 | |
2650 | endproc | |
2651 | ||
2652 | proc x25 | |
2653 | ||
90c4eee3 MW |
2654 | // a poor approximation to pi/4 |
2655 | // | |
2656 | // think of x and y as being in 16.16 fixed-point format. we sample | |
2657 | // points in the unit square, and determine how many of them are | |
2658 | // within a unit quarter-circle centred at the origin. the area of | |
2659 | // the quarter-circle is pi/4. | |
2660 | ||
2661 | #if defined(__x86_64__) | |
2662 | ||
2663 | xor eax, eax // a = 0 | |
2664 | mov rcx, 1 | |
2665 | shl rcx, 0x20 // c =~ 4 billion | |
2666 | ||
2667 | 0: movzx rbx, cx // x = low 16 bits of c | |
2668 | imul rbx, rbx // b = x^2 | |
2669 | ||
2670 | ror rcx, 0x10 // switch halves of c | |
2671 | movzx rdx, cx // y = high 16 bits of c | |
2672 | imul rdx, rdx // d = y^2 | |
2673 | rol rcx, 0x10 // switch back | |
2674 | ||
2675 | add rbx, rdx // r^2 = x^2 + y^2 | |
2676 | shr rbx, 0x20 // r^2 >= 1? | |
2677 | cmp rbx, 1 // set cf iff r^2 >= 1 | |
2678 | adc rax, 0 // and add onto accumulator | |
2679 | loop 0b | |
2680 | ||
2681 | #elif defined(__i386__) | |
2682 | ||
2683 | // this is actually better done in 32 bits. the carry has the wrong | |
2684 | // sense here, so instead deduct one for each point outside the | |
2685 | // quarter-circle rather than adding one for each point inside it. | |
2686 | xor eax, eax | |
2687 | xor ecx, ecx | |
2688 | ||
2689 | 0: movzx ebx, cx | |
2690 | imul ebx, ebx | |
2691 | ||
2692 | ror ecx, 0x10 | |
2693 | movzx edx, cx | |
2694 | imul edx, edx | |
2695 | rol ecx, 0x10 | |
2696 | ||
2697 | add ebx, edx // see? | |
2698 | sbb eax, 0 | |
2699 | loop 0b | |
2700 | ||
2701 | #elif defined(__arm__) | |
2702 | ||
2703 | mov r0, #0 | |
2704 | mov r2, #0 | |
2705 | ||
2706 | 0: uxth r1, r2, ror #0 | |
2707 | uxth r3, r2, ror #16 | |
2708 | mul r1, r1, r1 | |
2709 | mul r3, r3, r3 | |
2710 | cmn r1, r3 // mlas doesn't set cf usefully | |
2711 | addcc r0, r0, #1 | |
2712 | adds r2, r2, #1 | |
2713 | bne 0b | |
2714 | ||
2715 | #elif defined(__aarch64__) | |
2716 | ||
2717 | mov w0, #0 | |
2718 | mov w2, #0 | |
2719 | ||
2720 | 0: ubfx w1, w2, #0, #16 | |
2721 | ubfx w3, w2, #16, #16 | |
2722 | sub w2, w2, #1 | |
2723 | mul w1, w1, w1 | |
2724 | mul w3, w3, w3 | |
2725 | cmn w1, w3 | |
2726 | cinc.cc w0, w0 | |
2727 | cbnz w2, 0b | |
2728 | ||
2729 | #else | |
2730 | notimpl | |
2731 | #endif | |
2732 | ||
2733 | ret | |
06297a93 MW |
2734 | |
2735 | endproc | |
2736 | ||
2737 | proc x26 | |
2738 | ||
90c4eee3 MW |
2739 | #if defined(__x86_64__) |
2740 | ||
2741 | notimpl | |
2742 | ||
2743 | #elif defined(__i386__) | |
2744 | ||
2745 | notimpl | |
2746 | ||
2747 | #elif defined(__arm__) | |
2748 | ||
2749 | notimpl | |
2750 | ||
2751 | #elif defined(__aarch64__) | |
2752 | ||
2753 | notimpl | |
2754 | ||
2755 | #else | |
2756 | notimpl | |
2757 | #endif | |
06297a93 MW |
2758 | |
2759 | endproc | |
2760 | ||
2761 | proc x27 | |
2762 | ||
90c4eee3 MW |
2763 | #if defined(__x86_64__) |
2764 | ||
2765 | notimpl | |
2766 | ||
2767 | #elif defined(__i386__) | |
2768 | ||
2769 | notimpl | |
2770 | ||
2771 | #elif defined(__arm__) | |
2772 | ||
2773 | notimpl | |
2774 | ||
2775 | #elif defined(__aarch64__) | |
2776 | ||
2777 | notimpl | |
2778 | ||
2779 | #else | |
2780 | notimpl | |
2781 | #endif | |
06297a93 MW |
2782 | |
2783 | endproc | |
2784 | ||
2785 | proc x28 | |
2786 | ||
90c4eee3 MW |
2787 | #if defined(__x86_64__) |
2788 | ||
2789 | notimpl | |
2790 | ||
2791 | #elif defined(__i386__) | |
2792 | ||
2793 | notimpl | |
2794 | ||
2795 | #elif defined(__arm__) | |
2796 | ||
2797 | notimpl | |
2798 | ||
2799 | #elif defined(__aarch64__) | |
2800 | ||
2801 | notimpl | |
2802 | ||
2803 | #else | |
2804 | notimpl | |
2805 | #endif | |
06297a93 MW |
2806 | |
2807 | endproc | |
2808 | ||
2809 | proc x29 | |
2810 | ||
90c4eee3 MW |
2811 | #if defined(__x86_64__) |
2812 | ||
2813 | notimpl | |
2814 | ||
2815 | #elif defined(__i386__) | |
2816 | ||
2817 | notimpl | |
2818 | ||
2819 | #elif defined(__arm__) | |
2820 | ||
2821 | notimpl | |
2822 | ||
2823 | #elif defined(__aarch64__) | |
2824 | ||
2825 | notimpl | |
2826 | ||
2827 | #else | |
2828 | notimpl | |
2829 | #endif | |
06297a93 MW |
2830 | |
2831 | endproc | |
2832 | ||
2833 | proc x2a | |
2834 | ||
90c4eee3 MW |
2835 | #if defined(__x86_64__) |
2836 | ||
2837 | notimpl | |
2838 | ||
2839 | #elif defined(__i386__) | |
2840 | ||
2841 | notimpl | |
2842 | ||
2843 | #elif defined(__arm__) | |
2844 | ||
2845 | notimpl | |
2846 | ||
2847 | #elif defined(__aarch64__) | |
2848 | ||
2849 | notimpl | |
2850 | ||
2851 | #else | |
2852 | notimpl | |
2853 | #endif | |
06297a93 MW |
2854 | |
2855 | endproc | |
2856 | ||
2857 | proc x2b | |
2858 | ||
90c4eee3 MW |
2859 | #if defined(__x86_64__) |
2860 | ||
2861 | notimpl | |
2862 | ||
2863 | #elif defined(__i386__) | |
2864 | ||
2865 | notimpl | |
2866 | ||
2867 | #elif defined(__arm__) | |
2868 | ||
2869 | notimpl | |
2870 | ||
2871 | #elif defined(__aarch64__) | |
2872 | ||
2873 | notimpl | |
2874 | ||
2875 | #else | |
2876 | notimpl | |
2877 | #endif | |
06297a93 MW |
2878 | |
2879 | endproc | |
2880 | ||
2881 | proc x2c | |
2882 | ||
90c4eee3 MW |
2883 | #if defined(__x86_64__) |
2884 | ||
2885 | notimpl | |
2886 | ||
2887 | #elif defined(__i386__) | |
2888 | ||
2889 | notimpl | |
2890 | ||
2891 | #elif defined(__arm__) | |
2892 | ||
2893 | notimpl | |
2894 | ||
2895 | #elif defined(__aarch64__) | |
2896 | ||
2897 | notimpl | |
2898 | ||
2899 | #else | |
2900 | notimpl | |
2901 | #endif | |
06297a93 MW |
2902 | |
2903 | endproc | |
2904 | ||
2905 | proc x2d | |
2906 | ||
90c4eee3 MW |
2907 | #if defined(__x86_64__) |
2908 | ||
2909 | notimpl | |
2910 | ||
2911 | #elif defined(__i386__) | |
2912 | ||
2913 | notimpl | |
2914 | ||
2915 | #elif defined(__arm__) | |
2916 | ||
2917 | notimpl | |
2918 | ||
2919 | #elif defined(__aarch64__) | |
2920 | ||
2921 | notimpl | |
2922 | ||
2923 | #else | |
2924 | notimpl | |
2925 | #endif | |
06297a93 MW |
2926 | |
2927 | endproc | |
2928 | ||
2929 | proc x2e | |
2930 | ||
90c4eee3 MW |
2931 | #if defined(__x86_64__) |
2932 | ||
2933 | notimpl | |
2934 | ||
2935 | #elif defined(__i386__) | |
2936 | ||
2937 | notimpl | |
2938 | ||
2939 | #elif defined(__arm__) | |
2940 | ||
2941 | notimpl | |
2942 | ||
2943 | #elif defined(__aarch64__) | |
2944 | ||
2945 | notimpl | |
2946 | ||
2947 | #else | |
2948 | notimpl | |
2949 | #endif | |
06297a93 MW |
2950 | |
2951 | endproc | |
2952 | ||
2953 | proc x2f | |
2954 | ||
90c4eee3 MW |
2955 | #if defined(__x86_64__) |
2956 | ||
2957 | notimpl | |
2958 | ||
2959 | #elif defined(__i386__) | |
2960 | ||
2961 | notimpl | |
2962 | ||
2963 | #elif defined(__arm__) | |
2964 | ||
2965 | notimpl | |
2966 | ||
2967 | #elif defined(__aarch64__) | |
2968 | ||
2969 | notimpl | |
2970 | ||
2971 | #else | |
2972 | notimpl | |
2973 | #endif | |
06297a93 MW |
2974 | |
2975 | endproc | |
2976 | ||
90c4eee3 MW |
2977 | ///-------------------------------------------------------------------------- |
2978 | /// 0x30--0x3f | |
2979 | ||
06297a93 MW |
2980 | proc x30 |
2981 | ||
90c4eee3 MW |
2982 | #if defined(__x86_64__) |
2983 | ||
2984 | notimpl | |
2985 | ||
2986 | #elif defined(__i386__) | |
2987 | ||
2988 | notimpl | |
2989 | ||
2990 | #elif defined(__arm__) | |
2991 | ||
2992 | notimpl | |
2993 | ||
2994 | #elif defined(__aarch64__) | |
2995 | ||
2996 | notimpl | |
2997 | ||
2998 | #else | |
2999 | notimpl | |
3000 | #endif | |
06297a93 MW |
3001 | |
3002 | ret | |
3003 | ||
3004 | endproc | |
3005 | ||
3006 | proc x31 | |
3007 | ||
90c4eee3 MW |
3008 | #if defined(__x86_64__) |
3009 | ||
3010 | notimpl | |
3011 | ||
3012 | #elif defined(__i386__) | |
3013 | ||
3014 | notimpl | |
3015 | ||
3016 | #elif defined(__arm__) | |
3017 | ||
3018 | notimpl | |
3019 | ||
3020 | #elif defined(__aarch64__) | |
3021 | ||
3022 | notimpl | |
3023 | ||
3024 | #else | |
3025 | notimpl | |
3026 | #endif | |
06297a93 MW |
3027 | |
3028 | endproc | |
3029 | ||
3030 | proc x32 | |
3031 | ||
90c4eee3 MW |
3032 | #if defined(__x86_64__) |
3033 | ||
3034 | notimpl | |
3035 | ||
3036 | #elif defined(__i386__) | |
3037 | ||
3038 | notimpl | |
3039 | ||
3040 | #elif defined(__arm__) | |
3041 | ||
3042 | notimpl | |
3043 | ||
3044 | #elif defined(__aarch64__) | |
3045 | ||
3046 | notimpl | |
3047 | ||
3048 | #else | |
3049 | notimpl | |
3050 | #endif | |
06297a93 MW |
3051 | |
3052 | endproc | |
3053 | ||
3054 | proc x33 | |
3055 | ||
90c4eee3 MW |
3056 | #if defined(__x86_64__) |
3057 | ||
3058 | notimpl | |
3059 | ||
3060 | #elif defined(__i386__) | |
3061 | ||
3062 | notimpl | |
3063 | ||
3064 | #elif defined(__arm__) | |
3065 | ||
3066 | notimpl | |
3067 | ||
3068 | #elif defined(__aarch64__) | |
3069 | ||
3070 | notimpl | |
3071 | ||
3072 | #else | |
3073 | notimpl | |
3074 | #endif | |
06297a93 MW |
3075 | |
3076 | endproc | |
3077 | ||
3078 | proc x34 | |
3079 | ||
90c4eee3 MW |
3080 | #if defined(__x86_64__) |
3081 | ||
3082 | notimpl | |
3083 | ||
3084 | #elif defined(__i386__) | |
3085 | ||
3086 | notimpl | |
3087 | ||
3088 | #elif defined(__arm__) | |
3089 | ||
3090 | notimpl | |
3091 | ||
3092 | #elif defined(__aarch64__) | |
3093 | ||
3094 | notimpl | |
3095 | ||
3096 | #else | |
3097 | notimpl | |
3098 | #endif | |
06297a93 MW |
3099 | |
3100 | endproc | |
3101 | ||
3102 | proc x35 | |
3103 | ||
90c4eee3 MW |
3104 | #if defined(__x86_64__) |
3105 | ||
3106 | notimpl | |
3107 | ||
3108 | #elif defined(__i386__) | |
3109 | ||
3110 | notimpl | |
3111 | ||
3112 | #elif defined(__arm__) | |
3113 | ||
3114 | notimpl | |
3115 | ||
3116 | #elif defined(__aarch64__) | |
3117 | ||
3118 | notimpl | |
3119 | ||
3120 | #else | |
3121 | notimpl | |
3122 | #endif | |
06297a93 MW |
3123 | |
3124 | endproc | |
3125 | ||
3126 | proc x36 | |
3127 | ||
90c4eee3 MW |
3128 | #if defined(__x86_64__) |
3129 | ||
3130 | notimpl | |
3131 | ||
3132 | #elif defined(__i386__) | |
3133 | ||
3134 | notimpl | |
3135 | ||
3136 | #elif defined(__arm__) | |
3137 | ||
3138 | notimpl | |
3139 | ||
3140 | #elif defined(__aarch64__) | |
3141 | ||
3142 | notimpl | |
3143 | ||
3144 | #else | |
3145 | notimpl | |
3146 | #endif | |
06297a93 MW |
3147 | |
3148 | endproc | |
3149 | ||
3150 | proc x37 | |
3151 | ||
90c4eee3 MW |
3152 | #if defined(__x86_64__) |
3153 | ||
3154 | notimpl | |
3155 | ||
3156 | #elif defined(__i386__) | |
3157 | ||
3158 | notimpl | |
3159 | ||
3160 | #elif defined(__arm__) | |
3161 | ||
3162 | notimpl | |
3163 | ||
3164 | #elif defined(__aarch64__) | |
3165 | ||
3166 | notimpl | |
3167 | ||
3168 | #else | |
3169 | notimpl | |
3170 | #endif | |
06297a93 MW |
3171 | |
3172 | endproc | |
3173 | ||
3174 | proc x38 | |
3175 | ||
90c4eee3 MW |
3176 | #if defined(__x86_64__) |
3177 | ||
3178 | notimpl | |
3179 | ||
3180 | #elif defined(__i386__) | |
3181 | ||
3182 | notimpl | |
3183 | ||
3184 | #elif defined(__arm__) | |
3185 | ||
3186 | notimpl | |
3187 | ||
3188 | #elif defined(__aarch64__) | |
3189 | ||
3190 | notimpl | |
3191 | ||
3192 | #else | |
3193 | notimpl | |
3194 | #endif | |
06297a93 MW |
3195 | |
3196 | endproc | |
3197 | ||
3198 | proc x39 | |
3199 | ||
90c4eee3 MW |
3200 | #if defined(__x86_64__) |
3201 | ||
3202 | notimpl | |
3203 | ||
3204 | #elif defined(__i386__) | |
3205 | ||
3206 | notimpl | |
3207 | ||
3208 | #elif defined(__arm__) | |
3209 | ||
3210 | notimpl | |
3211 | ||
3212 | #elif defined(__aarch64__) | |
3213 | ||
3214 | notimpl | |
3215 | ||
3216 | #else | |
3217 | notimpl | |
3218 | #endif | |
06297a93 MW |
3219 | |
3220 | endproc | |
3221 | ||
3222 | proc x3a | |
3223 | ||
90c4eee3 MW |
3224 | #if defined(__x86_64__) |
3225 | ||
3226 | notimpl | |
3227 | ||
3228 | #elif defined(__i386__) | |
3229 | ||
3230 | notimpl | |
3231 | ||
3232 | #elif defined(__arm__) | |
3233 | ||
3234 | notimpl | |
3235 | ||
3236 | #elif defined(__aarch64__) | |
3237 | ||
3238 | notimpl | |
3239 | ||
3240 | #else | |
3241 | notimpl | |
3242 | #endif | |
06297a93 MW |
3243 | |
3244 | endproc | |
3245 | ||
3246 | proc x3b | |
3247 | ||
90c4eee3 MW |
3248 | #if defined(__x86_64__) |
3249 | ||
3250 | notimpl | |
3251 | ||
3252 | #elif defined(__i386__) | |
3253 | ||
3254 | notimpl | |
3255 | ||
3256 | #elif defined(__arm__) | |
3257 | ||
3258 | notimpl | |
3259 | ||
3260 | #elif defined(__aarch64__) | |
3261 | ||
3262 | notimpl | |
3263 | ||
3264 | #else | |
3265 | notimpl | |
3266 | #endif | |
06297a93 MW |
3267 | |
3268 | endproc | |
3269 | ||
3270 | proc x3c | |
3271 | ||
90c4eee3 MW |
3272 | #if defined(__x86_64__) |
3273 | ||
3274 | notimpl | |
3275 | ||
3276 | #elif defined(__i386__) | |
3277 | ||
3278 | notimpl | |
3279 | ||
3280 | #elif defined(__arm__) | |
3281 | ||
3282 | notimpl | |
3283 | ||
3284 | #elif defined(__aarch64__) | |
3285 | ||
3286 | notimpl | |
3287 | ||
3288 | #else | |
3289 | notimpl | |
3290 | #endif | |
06297a93 MW |
3291 | |
3292 | endproc | |
3293 | ||
3294 | proc x3d | |
3295 | ||
90c4eee3 MW |
3296 | #if defined(__x86_64__) |
3297 | ||
3298 | notimpl | |
3299 | ||
3300 | #elif defined(__i386__) | |
3301 | ||
3302 | notimpl | |
3303 | ||
3304 | #elif defined(__arm__) | |
3305 | ||
3306 | notimpl | |
3307 | ||
3308 | #elif defined(__aarch64__) | |
3309 | ||
3310 | notimpl | |
3311 | ||
3312 | #else | |
3313 | notimpl | |
3314 | #endif | |
06297a93 MW |
3315 | |
3316 | endproc | |
3317 | ||
3318 | proc x3e | |
3319 | ||
90c4eee3 MW |
3320 | #if defined(__x86_64__) |
3321 | ||
3322 | notimpl | |
3323 | ||
3324 | #elif defined(__i386__) | |
3325 | ||
3326 | notimpl | |
3327 | ||
3328 | #elif defined(__arm__) | |
3329 | ||
3330 | notimpl | |
3331 | ||
3332 | #elif defined(__aarch64__) | |
3333 | ||
3334 | notimpl | |
3335 | ||
3336 | #else | |
3337 | notimpl | |
3338 | #endif | |
06297a93 MW |
3339 | |
3340 | endproc | |
3341 | ||
3342 | proc x3f | |
3343 | ||
90c4eee3 MW |
3344 | #if defined(__x86_64__) |
3345 | ||
3346 | notimpl | |
3347 | ||
3348 | #elif defined(__i386__) | |
3349 | ||
3350 | notimpl | |
3351 | ||
3352 | #elif defined(__arm__) | |
3353 | ||
3354 | notimpl | |
3355 | ||
3356 | #elif defined(__aarch64__) | |
3357 | ||
3358 | notimpl | |
3359 | ||
3360 | #else | |
3361 | notimpl | |
3362 | #endif | |
06297a93 MW |
3363 | |
3364 | endproc | |
90c4eee3 MW |
3365 | |
3366 | ///----- That's all, folks -------------------------------------------------- |