1 /*
2 * Copyright (c) 2020, 2023, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "asm/assembler.hpp"
27 #include "asm/assembler.inline.hpp"
28 #include "opto/c2_MacroAssembler.hpp"
29 #include "opto/intrinsicnode.hpp"
30 #include "runtime/vm_version.hpp"
31
32 #ifdef PRODUCT
33 #define BLOCK_COMMENT(str) // nothing
34 #else
35 #define BLOCK_COMMENT(str) block_comment(str)
36 #endif
37 #define BIND(label) bind(label); BLOCK_COMMENT(#label ":")
38
39 // Intrinsics for CompactStrings
40
41 // Compress char[] to byte[] by compressing 16 bytes at once.
42 void C2_MacroAssembler::string_compress_16(Register src, Register dst, Register cnt,
43 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
44 Label& Lfailure, bool ascii) {
45
46 const Register tmp0 = R0;
47 const int byte_mask = ascii ? 0x7F : 0xFF;
48 assert_different_registers(src, dst, cnt, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
49 Label Lloop, Lslow;
50
51 // Check if cnt >= 8 (= 16 bytes)
52 lis(tmp1, byte_mask); // tmp1 = 0x00FF00FF00FF00FF (non ascii case)
53 srwi_(tmp2, cnt, 3);
54 beq(CCR0, Lslow);
55 ori(tmp1, tmp1, byte_mask);
56 rldimi(tmp1, tmp1, 32, 0);
57 mtctr(tmp2);
58
59 // 2x unrolled loop
60 bind(Lloop);
61 ld(tmp2, 0, src); // _0_1_2_3 (Big Endian)
62 ld(tmp4, 8, src); // _4_5_6_7
63
64 orr(tmp0, tmp2, tmp4);
65 rldicl(tmp3, tmp2, 6*8, 64-24); // _____1_2
66 rldimi(tmp2, tmp2, 2*8, 2*8); // _0_2_3_3
67 rldicl(tmp5, tmp4, 6*8, 64-24); // _____5_6
68 rldimi(tmp4, tmp4, 2*8, 2*8); // _4_6_7_7
69
70 andc_(tmp0, tmp0, tmp1);
71 bne(CCR0, Lfailure); // Not latin1/ascii.
72 addi(src, src, 16);
73
74 rlwimi(tmp3, tmp2, 0*8, 24, 31);// _____1_3
75 srdi(tmp2, tmp2, 3*8); // ____0_2_
76 rlwimi(tmp5, tmp4, 0*8, 24, 31);// _____5_7
77 srdi(tmp4, tmp4, 3*8); // ____4_6_
78
79 orr(tmp2, tmp2, tmp3); // ____0123
80 orr(tmp4, tmp4, tmp5); // ____4567
81
82 stw(tmp2, 0, dst);
83 stw(tmp4, 4, dst);
84 addi(dst, dst, 8);
85 bdnz(Lloop);
86
87 bind(Lslow); // Fallback to slow version
88 }
89
90 // Compress char[] to byte[]. cnt must be positive int.
91 void C2_MacroAssembler::string_compress(Register src, Register dst, Register cnt, Register tmp,
92 Label& Lfailure, bool ascii) {
93 const int byte_mask = ascii ? 0x7F : 0xFF;
94 Label Lloop;
95 mtctr(cnt);
96
97 bind(Lloop);
98 lhz(tmp, 0, src);
99 cmplwi(CCR0, tmp, byte_mask);
100 bgt(CCR0, Lfailure); // Not latin1/ascii.
101 addi(src, src, 2);
102 stb(tmp, 0, dst);
103 addi(dst, dst, 1);
104 bdnz(Lloop);
105 }
106
107 void C2_MacroAssembler::encode_iso_array(Register src, Register dst, Register len,
108 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5,
109 Register result, bool ascii) {
110 Label Lslow, Lfailure1, Lfailure2, Ldone;
111
112 string_compress_16(src, dst, len, tmp1, tmp2, tmp3, tmp4, tmp5, Lfailure1, ascii);
113 rldicl_(result, len, 0, 64-3); // Remaining characters.
114 beq(CCR0, Ldone);
115 bind(Lslow);
116 string_compress(src, dst, result, tmp2, Lfailure2, ascii);
117 li(result, 0);
118 b(Ldone);
119
120 bind(Lfailure1);
121 mr(result, len);
122 mfctr(tmp1);
123 rldimi_(result, tmp1, 3, 0); // Remaining characters.
124 beq(CCR0, Ldone);
125 b(Lslow);
126
127 bind(Lfailure2);
128 mfctr(result); // Remaining characters.
129
130 bind(Ldone);
131 subf(result, result, len);
132 }
133
134 // Inflate byte[] to char[] by inflating 16 bytes at once.
135 void C2_MacroAssembler::string_inflate_16(Register src, Register dst, Register cnt,
136 Register tmp1, Register tmp2, Register tmp3, Register tmp4, Register tmp5) {
137 const Register tmp0 = R0;
138 assert_different_registers(src, dst, cnt, tmp0, tmp1, tmp2, tmp3, tmp4, tmp5);
139 Label Lloop, Lslow;
140
141 // Check if cnt >= 8
142 srwi_(tmp2, cnt, 3);
143 beq(CCR0, Lslow);
144 lis(tmp1, 0xFF); // tmp1 = 0x00FF00FF
145 ori(tmp1, tmp1, 0xFF);
146 mtctr(tmp2);
147
148 // 2x unrolled loop
149 bind(Lloop);
150 lwz(tmp2, 0, src); // ____0123 (Big Endian)
151 lwz(tmp4, 4, src); // ____4567
152 addi(src, src, 8);
153
154 rldicl(tmp3, tmp2, 7*8, 64-8); // _______2
155 rlwimi(tmp2, tmp2, 3*8, 16, 23);// ____0113
156 rldicl(tmp5, tmp4, 7*8, 64-8); // _______6
157 rlwimi(tmp4, tmp4, 3*8, 16, 23);// ____4557
158
159 andc(tmp0, tmp2, tmp1); // ____0_1_
160 rlwimi(tmp2, tmp3, 2*8, 0, 23); // _____2_3
161 andc(tmp3, tmp4, tmp1); // ____4_5_
162 rlwimi(tmp4, tmp5, 2*8, 0, 23); // _____6_7
163
164 rldimi(tmp2, tmp0, 3*8, 0*8); // _0_1_2_3
165 rldimi(tmp4, tmp3, 3*8, 0*8); // _4_5_6_7
166
167 std(tmp2, 0, dst);
168 std(tmp4, 8, dst);
169 addi(dst, dst, 16);
170 bdnz(Lloop);
171
172 bind(Lslow); // Fallback to slow version
173 }
174
175 // Inflate byte[] to char[]. cnt must be positive int.
176 void C2_MacroAssembler::string_inflate(Register src, Register dst, Register cnt, Register tmp) {
177 Label Lloop;
178 mtctr(cnt);
179
180 bind(Lloop);
181 lbz(tmp, 0, src);
182 addi(src, src, 1);
183 sth(tmp, 0, dst);
184 addi(dst, dst, 2);
185 bdnz(Lloop);
186 }
187
188 void C2_MacroAssembler::string_compare(Register str1, Register str2,
189 Register cnt1, Register cnt2,
190 Register tmp1, Register result, int ae) {
191 const Register tmp0 = R0,
192 diff = tmp1;
193
194 assert_different_registers(str1, str2, cnt1, cnt2, tmp0, tmp1, result);
195 Label Ldone, Lslow, Lloop, Lreturn_diff;
196
197 // Note: Making use of the fact that compareTo(a, b) == -compareTo(b, a)
198 // we interchange str1 and str2 in the UL case and negate the result.
199 // Like this, str1 is always latin1 encoded, except for the UU case.
200 // In addition, we need 0 (or sign which is 0) extend.
201
202 if (ae == StrIntrinsicNode::UU) {
203 srwi(cnt1, cnt1, 1);
204 } else {
205 clrldi(cnt1, cnt1, 32);
206 }
207
208 if (ae != StrIntrinsicNode::LL) {
209 srwi(cnt2, cnt2, 1);
210 } else {
211 clrldi(cnt2, cnt2, 32);
212 }
213
214 // See if the lengths are different, and calculate min in cnt1.
215 // Save diff in case we need it for a tie-breaker.
216 subf_(diff, cnt2, cnt1); // diff = cnt1 - cnt2
217 // if (diff > 0) { cnt1 = cnt2; }
218 if (VM_Version::has_isel()) {
219 isel(cnt1, CCR0, Assembler::greater, /*invert*/ false, cnt2);
220 } else {
221 Label Lskip;
222 blt(CCR0, Lskip);
223 mr(cnt1, cnt2);
224 bind(Lskip);
225 }
226
227 // Rename registers
228 Register chr1 = result;
229 Register chr2 = tmp0;
230
231 // Compare multiple characters in fast loop (only implemented for same encoding).
232 int stride1 = 8, stride2 = 8;
233 if (ae == StrIntrinsicNode::LL || ae == StrIntrinsicNode::UU) {
234 int log2_chars_per_iter = (ae == StrIntrinsicNode::LL) ? 3 : 2;
235 Label Lfastloop, Lskipfast;
236
237 srwi_(tmp0, cnt1, log2_chars_per_iter);
238 beq(CCR0, Lskipfast);
239 rldicl(cnt2, cnt1, 0, 64 - log2_chars_per_iter); // Remaining characters.
240 li(cnt1, 1 << log2_chars_per_iter); // Initialize for failure case: Rescan characters from current iteration.
241 mtctr(tmp0);
242
243 bind(Lfastloop);
244 ld(chr1, 0, str1);
245 ld(chr2, 0, str2);
246 cmpd(CCR0, chr1, chr2);
247 bne(CCR0, Lslow);
248 addi(str1, str1, stride1);
249 addi(str2, str2, stride2);
250 bdnz(Lfastloop);
251 mr(cnt1, cnt2); // Remaining characters.
252 bind(Lskipfast);
253 }
254
255 // Loop which searches the first difference character by character.
256 cmpwi(CCR0, cnt1, 0);
257 beq(CCR0, Lreturn_diff);
258 bind(Lslow);
259 mtctr(cnt1);
260
261 switch (ae) {
262 case StrIntrinsicNode::LL: stride1 = 1; stride2 = 1; break;
263 case StrIntrinsicNode::UL: // fallthru (see comment above)
264 case StrIntrinsicNode::LU: stride1 = 1; stride2 = 2; break;
265 case StrIntrinsicNode::UU: stride1 = 2; stride2 = 2; break;
266 default: ShouldNotReachHere(); break;
267 }
268
269 bind(Lloop);
270 if (stride1 == 1) { lbz(chr1, 0, str1); } else { lhz(chr1, 0, str1); }
271 if (stride2 == 1) { lbz(chr2, 0, str2); } else { lhz(chr2, 0, str2); }
272 subf_(result, chr2, chr1); // result = chr1 - chr2
273 bne(CCR0, Ldone);
274 addi(str1, str1, stride1);
275 addi(str2, str2, stride2);
276 bdnz(Lloop);
277
278 // If strings are equal up to min length, return the length difference.
279 bind(Lreturn_diff);
280 mr(result, diff);
281
282 // Otherwise, return the difference between the first mismatched chars.
283 bind(Ldone);
284 if (ae == StrIntrinsicNode::UL) {
285 neg(result, result); // Negate result (see note above).
286 }
287 }
288
289 void C2_MacroAssembler::array_equals(bool is_array_equ, Register ary1, Register ary2,
290 Register limit, Register tmp1, Register result, bool is_byte) {
291 const Register tmp0 = R0;
292 assert_different_registers(ary1, ary2, limit, tmp0, tmp1, result);
293 Label Ldone, Lskiploop, Lloop, Lfastloop, Lskipfast;
294 bool limit_needs_shift = false;
295
296 if (is_array_equ) {
297 const int length_offset = arrayOopDesc::length_offset_in_bytes();
298 const int base_offset = arrayOopDesc::base_offset_in_bytes(is_byte ? T_BYTE : T_CHAR);
299
300 // Return true if the same array.
301 cmpd(CCR0, ary1, ary2);
302 beq(CCR0, Lskiploop);
303
304 // Return false if one of them is null.
305 cmpdi(CCR0, ary1, 0);
306 cmpdi(CCR1, ary2, 0);
307 li(result, 0);
308 cror(CCR0, Assembler::equal, CCR1, Assembler::equal);
309 beq(CCR0, Ldone);
310
311 // Load the lengths of arrays.
312 lwz(limit, length_offset, ary1);
313 lwz(tmp0, length_offset, ary2);
314
315 // Return false if the two arrays are not equal length.
316 cmpw(CCR0, limit, tmp0);
317 bne(CCR0, Ldone);
318
319 // Load array addresses.
320 addi(ary1, ary1, base_offset);
321 addi(ary2, ary2, base_offset);
322 } else {
323 limit_needs_shift = !is_byte;
324 li(result, 0); // Assume not equal.
325 }
326
327 // Rename registers
328 Register chr1 = tmp0;
329 Register chr2 = tmp1;
330
331 // Compare 8 bytes per iteration in fast loop.
332 const int log2_chars_per_iter = is_byte ? 3 : 2;
333
334 srwi_(tmp0, limit, log2_chars_per_iter + (limit_needs_shift ? 1 : 0));
335 beq(CCR0, Lskipfast);
336 mtctr(tmp0);
337
338 bind(Lfastloop);
339 ld(chr1, 0, ary1);
340 ld(chr2, 0, ary2);
341 addi(ary1, ary1, 8);
342 addi(ary2, ary2, 8);
343 cmpd(CCR0, chr1, chr2);
344 bne(CCR0, Ldone);
345 bdnz(Lfastloop);
346
347 bind(Lskipfast);
348 rldicl_(limit, limit, limit_needs_shift ? 64 - 1 : 0, 64 - log2_chars_per_iter); // Remaining characters.
349 beq(CCR0, Lskiploop);
350 mtctr(limit);
351
352 // Character by character.
353 bind(Lloop);
354 if (is_byte) {
355 lbz(chr1, 0, ary1);
356 lbz(chr2, 0, ary2);
357 addi(ary1, ary1, 1);
358 addi(ary2, ary2, 1);
359 } else {
360 lhz(chr1, 0, ary1);
361 lhz(chr2, 0, ary2);
362 addi(ary1, ary1, 2);
363 addi(ary2, ary2, 2);
364 }
365 cmpw(CCR0, chr1, chr2);
366 bne(CCR0, Ldone);
367 bdnz(Lloop);
368
369 bind(Lskiploop);
370 li(result, 1); // All characters are equal.
371 bind(Ldone);
372 }
373
374 void C2_MacroAssembler::string_indexof(Register result, Register haystack, Register haycnt,
375 Register needle, ciTypeArray* needle_values, Register needlecnt, int needlecntval,
376 Register tmp1, Register tmp2, Register tmp3, Register tmp4, int ae) {
377
378 // Ensure 0<needlecnt<=haycnt in ideal graph as prerequisite!
379 Label L_TooShort, L_Found, L_NotFound, L_End;
380 Register last_addr = haycnt, // Kill haycnt at the beginning.
381 addr = tmp1,
382 n_start = tmp2,
383 ch1 = tmp3,
384 ch2 = R0;
385
386 assert(ae != StrIntrinsicNode::LU, "Invalid encoding");
387 const int h_csize = (ae == StrIntrinsicNode::LL) ? 1 : 2;
388 const int n_csize = (ae == StrIntrinsicNode::UU) ? 2 : 1;
389
390 // **************************************************************************************************
391 // Prepare for main loop: optimized for needle count >=2, bail out otherwise.
392 // **************************************************************************************************
393
394 // Compute last haystack addr to use if no match gets found.
395 clrldi(haycnt, haycnt, 32); // Ensure positive int is valid as 64 bit value.
396 addi(addr, haystack, -h_csize); // Accesses use pre-increment.
397 if (needlecntval == 0) { // variable needlecnt
398 cmpwi(CCR6, needlecnt, 2);
399 clrldi(needlecnt, needlecnt, 32); // Ensure positive int is valid as 64 bit value.
400 blt(CCR6, L_TooShort); // Variable needlecnt: handle short needle separately.
401 }
402
403 if (n_csize == 2) { lwz(n_start, 0, needle); } else { lhz(n_start, 0, needle); } // Load first 2 characters of needle.
404
405 if (needlecntval == 0) { // variable needlecnt
406 subf(ch1, needlecnt, haycnt); // Last character index to compare is haycnt-needlecnt.
407 addi(needlecnt, needlecnt, -2); // Rest of needle.
408 } else { // constant needlecnt
409 guarantee(needlecntval != 1, "IndexOf with single-character needle must be handled separately");
410 assert((needlecntval & 0x7fff) == needlecntval, "wrong immediate");
411 addi(ch1, haycnt, -needlecntval); // Last character index to compare is haycnt-needlecnt.
412 if (needlecntval > 3) { li(needlecnt, needlecntval - 2); } // Rest of needle.
413 }
414
415 if (h_csize == 2) { slwi(ch1, ch1, 1); } // Scale to number of bytes.
416
417 if (ae ==StrIntrinsicNode::UL) {
418 srwi(tmp4, n_start, 1*8); // ___0
419 rlwimi(n_start, tmp4, 2*8, 0, 23); // _0_1
420 }
421
422 add(last_addr, haystack, ch1); // Point to last address to compare (haystack+2*(haycnt-needlecnt)).
423
424 // Main Loop (now we have at least 2 characters).
425 Label L_OuterLoop, L_InnerLoop, L_FinalCheck, L_Comp1, L_Comp2;
426 bind(L_OuterLoop); // Search for 1st 2 characters.
427 Register addr_diff = tmp4;
428 subf(addr_diff, addr, last_addr); // Difference between already checked address and last address to check.
429 addi(addr, addr, h_csize); // This is the new address we want to use for comparing.
430 srdi_(ch2, addr_diff, h_csize);
431 beq(CCR0, L_FinalCheck); // 2 characters left?
432 mtctr(ch2); // num of characters / 2
433 bind(L_InnerLoop); // Main work horse (2x unrolled search loop)
434 if (h_csize == 2) { // Load 2 characters of haystack (ignore alignment).
435 lwz(ch1, 0, addr);
436 lwz(ch2, 2, addr);
437 } else {
438 lhz(ch1, 0, addr);
439 lhz(ch2, 1, addr);
440 }
441 cmpw(CCR0, ch1, n_start); // Compare 2 characters (1 would be sufficient but try to reduce branches to CompLoop).
442 cmpw(CCR1, ch2, n_start);
443 beq(CCR0, L_Comp1); // Did we find the needle start?
444 beq(CCR1, L_Comp2);
445 addi(addr, addr, 2 * h_csize);
446 bdnz(L_InnerLoop);
447 bind(L_FinalCheck);
448 andi_(addr_diff, addr_diff, h_csize); // Remaining characters not covered by InnerLoop: (num of characters) & 1.
449 beq(CCR0, L_NotFound);
450 if (h_csize == 2) { lwz(ch1, 0, addr); } else { lhz(ch1, 0, addr); } // One position left at which we have to compare.
451 cmpw(CCR1, ch1, n_start);
452 beq(CCR1, L_Comp1);
453 bind(L_NotFound);
454 li(result, -1); // not found
455 b(L_End);
456
457 // **************************************************************************************************
458 // Special Case: unfortunately, the variable needle case can be called with needlecnt<2
459 // **************************************************************************************************
460 if (needlecntval == 0) { // We have to handle these cases separately.
461 Label L_OneCharLoop;
462 bind(L_TooShort);
463 mtctr(haycnt);
464 if (n_csize == 2) { lhz(n_start, 0, needle); } else { lbz(n_start, 0, needle); } // First character of needle
465 bind(L_OneCharLoop);
466 if (h_csize == 2) { lhzu(ch1, 2, addr); } else { lbzu(ch1, 1, addr); }
467 cmpw(CCR1, ch1, n_start);
468 beq(CCR1, L_Found); // Did we find the one character needle?
469 bdnz(L_OneCharLoop);
470 li(result, -1); // Not found.
471 b(L_End);
472 }
473
474 // **************************************************************************************************
475 // Regular Case Part II: compare rest of needle (first 2 characters have been compared already)
476 // **************************************************************************************************
477
478 // Compare the rest
479 bind(L_Comp2);
480 addi(addr, addr, h_csize); // First comparison has failed, 2nd one hit.
481 bind(L_Comp1); // Addr points to possible needle start.
482 if (needlecntval != 2) { // Const needlecnt==2?
483 if (needlecntval != 3) {
484 if (needlecntval == 0) { beq(CCR6, L_Found); } // Variable needlecnt==2?
485 Register n_ind = tmp4,
486 h_ind = n_ind;
487 li(n_ind, 2 * n_csize); // First 2 characters are already compared, use index 2.
488 mtctr(needlecnt); // Decremented by 2, still > 0.
489 Label L_CompLoop;
490 bind(L_CompLoop);
491 if (ae ==StrIntrinsicNode::UL) {
492 h_ind = ch1;
493 sldi(h_ind, n_ind, 1);
494 }
495 if (n_csize == 2) { lhzx(ch2, needle, n_ind); } else { lbzx(ch2, needle, n_ind); }
496 if (h_csize == 2) { lhzx(ch1, addr, h_ind); } else { lbzx(ch1, addr, h_ind); }
497 cmpw(CCR1, ch1, ch2);
498 bne(CCR1, L_OuterLoop);
499 addi(n_ind, n_ind, n_csize);
500 bdnz(L_CompLoop);
501 } else { // No loop required if there's only one needle character left.
502 if (n_csize == 2) { lhz(ch2, 2 * 2, needle); } else { lbz(ch2, 2 * 1, needle); }
503 if (h_csize == 2) { lhz(ch1, 2 * 2, addr); } else { lbz(ch1, 2 * 1, addr); }
504 cmpw(CCR1, ch1, ch2);
505 bne(CCR1, L_OuterLoop);
506 }
507 }
508 // Return index ...
509 bind(L_Found);
510 subf(result, haystack, addr); // relative to haystack, ...
511 if (h_csize == 2) { srdi(result, result, 1); } // in characters.
512 bind(L_End);
513 } // string_indexof
514
515 void C2_MacroAssembler::string_indexof_char(Register result, Register haystack, Register haycnt,
516 Register needle, jchar needleChar, Register tmp1, Register tmp2, bool is_byte) {
517 assert_different_registers(haystack, haycnt, needle, tmp1, tmp2);
518
519 Label L_InnerLoop, L_FinalCheck, L_Found1, L_Found2, L_NotFound, L_End;
520 Register addr = tmp1,
521 ch1 = tmp2,
522 ch2 = R0;
523
524 const int h_csize = is_byte ? 1 : 2;
525
526 //4:
527 srwi_(tmp2, haycnt, 1); // Shift right by exact_log2(UNROLL_FACTOR).
528 mr(addr, haystack);
529 beq(CCR0, L_FinalCheck);
530 mtctr(tmp2); // Move to count register.
531 //8:
532 bind(L_InnerLoop); // Main work horse (2x unrolled search loop).
533 if (!is_byte) {
534 lhz(ch1, 0, addr);
535 lhz(ch2, 2, addr);
536 } else {
537 lbz(ch1, 0, addr);
538 lbz(ch2, 1, addr);
539 }
540 (needle != R0) ? cmpw(CCR0, ch1, needle) : cmplwi(CCR0, ch1, (unsigned int)needleChar);
541 (needle != R0) ? cmpw(CCR1, ch2, needle) : cmplwi(CCR1, ch2, (unsigned int)needleChar);
542 beq(CCR0, L_Found1); // Did we find the needle?
543 beq(CCR1, L_Found2);
544 addi(addr, addr, 2 * h_csize);
545 bdnz(L_InnerLoop);
546 //16:
547 bind(L_FinalCheck);
548 andi_(R0, haycnt, 1);
549 beq(CCR0, L_NotFound);
550 if (!is_byte) { lhz(ch1, 0, addr); } else { lbz(ch1, 0, addr); } // One position left at which we have to compare.
551 (needle != R0) ? cmpw(CCR1, ch1, needle) : cmplwi(CCR1, ch1, (unsigned int)needleChar);
552 beq(CCR1, L_Found1);
553 //21:
554 bind(L_NotFound);
555 li(result, -1); // Not found.
556 b(L_End);
557
558 bind(L_Found2);
559 addi(addr, addr, h_csize);
560 //24:
561 bind(L_Found1); // Return index ...
562 subf(result, haystack, addr); // relative to haystack, ...
563 if (!is_byte) { srdi(result, result, 1); } // in characters.
564 bind(L_End);
565 } // string_indexof_char
566
567
568 void C2_MacroAssembler::count_positives(Register src, Register cnt, Register result,
569 Register tmp1, Register tmp2) {
570 const Register tmp0 = R0;
571 assert_different_registers(src, result, cnt, tmp0, tmp1, tmp2);
572 Label Lfastloop, Lslow, Lloop, Ldone;
573
574 // Check if cnt >= 8 (= 16 bytes)
575 lis(tmp1, (int)(short)0x8080); // tmp1 = 0x8080808080808080
576 srwi_(tmp2, cnt, 4);
577 mr(result, src); // Use result reg to point to the current position.
578 beq(CCR0, Lslow);
579 ori(tmp1, tmp1, 0x8080);
580 rldimi(tmp1, tmp1, 32, 0);
581 mtctr(tmp2);
582
583 // 2x unrolled loop
584 bind(Lfastloop);
585 ld(tmp2, 0, result);
586 ld(tmp0, 8, result);
587
588 orr(tmp0, tmp2, tmp0);
589
590 and_(tmp0, tmp0, tmp1);
591 bne(CCR0, Lslow); // Found negative byte.
592 addi(result, result, 16);
593 bdnz(Lfastloop);
594
595 bind(Lslow); // Fallback to slow version.
596 subf(tmp0, src, result); // Bytes known positive.
597 subf_(tmp0, tmp0, cnt); // Remaining Bytes.
598 beq(CCR0, Ldone);
599 mtctr(tmp0);
600 bind(Lloop);
601 lbz(tmp0, 0, result);
602 andi_(tmp0, tmp0, 0x80);
603 bne(CCR0, Ldone); // Found negative byte.
604 addi(result, result, 1);
605 bdnz(Lloop);
606
607 bind(Ldone);
608 subf(result, src, result); // Result is offset from src.
609 }