]>
Commit | Line | Data |
---|---|---|
ccd57e8a | 1 | /* Helper routines for memory move and comparison insns. |
d353bf18 | 2 | Copyright (C) 2013-2015 Free Software Foundation, Inc. |
ccd57e8a | 3 | |
4 | This file is part of GCC. | |
5 | ||
6 | GCC is free software; you can redistribute it and/or modify | |
7 | it under the terms of the GNU General Public License as published by | |
8 | the Free Software Foundation; either version 3, or (at your option) | |
9 | any later version. | |
10 | ||
11 | GCC is distributed in the hope that it will be useful, | |
12 | but WITHOUT ANY WARRANTY; without even the implied warranty of | |
13 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
14 | GNU General Public License for more details. | |
15 | ||
16 | You should have received a copy of the GNU General Public License | |
17 | along with GCC; see the file COPYING3. If not see | |
18 | <http://www.gnu.org/licenses/>. */ | |
19 | ||
20 | #include "config.h" | |
21 | #include "system.h" | |
22 | #include "coretypes.h" | |
23 | #include "tm.h" | |
24 | #include "machmode.h" | |
25 | #include "rtl.h" | |
b20a8bb4 | 26 | #include "hash-set.h" |
27 | #include "vec.h" | |
28 | #include "double-int.h" | |
29 | #include "input.h" | |
30 | #include "alias.h" | |
31 | #include "symtab.h" | |
32 | #include "wide-int.h" | |
33 | #include "inchash.h" | |
ccd57e8a | 34 | #include "tree.h" |
35 | #include "expr.h" | |
36 | #include "tm_p.h" | |
94ea8568 | 37 | #include "predict.h" |
94ea8568 | 38 | #include "hard-reg-set.h" |
39 | #include "input.h" | |
40 | #include "function.h" | |
41 | #include "dominance.h" | |
42 | #include "cfg.h" | |
43 | #include "cfgrtl.h" | |
44 | #include "cfganal.h" | |
45 | #include "lcm.h" | |
46 | #include "cfgbuild.h" | |
47 | #include "cfgcleanup.h" | |
ccd57e8a | 48 | #include "basic-block.h" |
49 | ||
50 | /* Like force_operand, but guarantees that VALUE ends up in TARGET. */ | |
51 | static void | |
52 | force_into (rtx value, rtx target) | |
53 | { | |
54 | value = force_operand (value, target); | |
55 | if (! rtx_equal_p (value, target)) | |
56 | emit_insn (gen_move_insn (target, value)); | |
57 | } | |
58 | ||
59 | /* Emit code to perform a block move. Choose the best method. | |
60 | ||
61 | OPERANDS[0] is the destination. | |
62 | OPERANDS[1] is the source. | |
63 | OPERANDS[2] is the size. | |
64 | OPERANDS[3] is the alignment safe to use. */ | |
65 | bool | |
66 | expand_block_move (rtx *operands) | |
67 | { | |
68 | int align = INTVAL (operands[3]); | |
69 | int constp = (CONST_INT_P (operands[2])); | |
70 | int bytes = (constp ? INTVAL (operands[2]) : 0); | |
71 | ||
72 | if (! constp) | |
73 | return false; | |
74 | ||
75 | /* If we could use mov.l to move words and dest is word-aligned, we | |
76 | can use movua.l for loads and still generate a relatively short | |
77 | and efficient sequence. */ | |
5b271aff | 78 | if (TARGET_SH4A && align < 4 |
ccd57e8a | 79 | && MEM_ALIGN (operands[0]) >= 32 |
80 | && can_move_by_pieces (bytes, 32)) | |
81 | { | |
82 | rtx dest = copy_rtx (operands[0]); | |
83 | rtx src = copy_rtx (operands[1]); | |
84 | /* We could use different pseudos for each copied word, but | |
85 | since movua can only load into r0, it's kind of | |
86 | pointless. */ | |
87 | rtx temp = gen_reg_rtx (SImode); | |
88 | rtx src_addr = copy_addr_to_reg (XEXP (src, 0)); | |
89 | int copied = 0; | |
90 | ||
91 | while (copied + 4 <= bytes) | |
92 | { | |
93 | rtx to = adjust_address (dest, SImode, copied); | |
94 | rtx from = adjust_automodify_address (src, BLKmode, | |
95 | src_addr, copied); | |
96 | ||
97 | set_mem_size (from, 4); | |
98 | emit_insn (gen_movua (temp, from)); | |
99 | emit_move_insn (src_addr, plus_constant (Pmode, src_addr, 4)); | |
100 | emit_move_insn (to, temp); | |
101 | copied += 4; | |
102 | } | |
103 | ||
104 | if (copied < bytes) | |
105 | move_by_pieces (adjust_address (dest, BLKmode, copied), | |
106 | adjust_automodify_address (src, BLKmode, | |
107 | src_addr, copied), | |
108 | bytes - copied, align, 0); | |
109 | ||
110 | return true; | |
111 | } | |
112 | ||
113 | /* If it isn't a constant number of bytes, or if it doesn't have 4 byte | |
114 | alignment, or if it isn't a multiple of 4 bytes, then fail. */ | |
115 | if (align < 4 || (bytes % 4 != 0)) | |
116 | return false; | |
117 | ||
118 | if (TARGET_HARD_SH4) | |
119 | { | |
120 | if (bytes < 12) | |
121 | return false; | |
122 | else if (bytes == 12) | |
123 | { | |
124 | rtx func_addr_rtx = gen_reg_rtx (Pmode); | |
125 | rtx r4 = gen_rtx_REG (SImode, 4); | |
126 | rtx r5 = gen_rtx_REG (SImode, 5); | |
127 | ||
128 | function_symbol (func_addr_rtx, "__movmemSI12_i4", SFUNC_STATIC); | |
129 | force_into (XEXP (operands[0], 0), r4); | |
130 | force_into (XEXP (operands[1], 0), r5); | |
131 | emit_insn (gen_block_move_real_i4 (func_addr_rtx)); | |
132 | return true; | |
133 | } | |
134 | else if (! optimize_size) | |
135 | { | |
136 | const char *entry_name; | |
137 | rtx func_addr_rtx = gen_reg_rtx (Pmode); | |
138 | int dwords; | |
139 | rtx r4 = gen_rtx_REG (SImode, 4); | |
140 | rtx r5 = gen_rtx_REG (SImode, 5); | |
141 | rtx r6 = gen_rtx_REG (SImode, 6); | |
142 | ||
143 | entry_name = (bytes & 4 ? "__movmem_i4_odd" : "__movmem_i4_even"); | |
144 | function_symbol (func_addr_rtx, entry_name, SFUNC_STATIC); | |
145 | force_into (XEXP (operands[0], 0), r4); | |
146 | force_into (XEXP (operands[1], 0), r5); | |
147 | ||
148 | dwords = bytes >> 3; | |
149 | emit_insn (gen_move_insn (r6, GEN_INT (dwords - 1))); | |
150 | emit_insn (gen_block_lump_real_i4 (func_addr_rtx)); | |
151 | return true; | |
152 | } | |
153 | else | |
154 | return false; | |
155 | } | |
156 | if (bytes < 64) | |
157 | { | |
158 | char entry[30]; | |
159 | rtx func_addr_rtx = gen_reg_rtx (Pmode); | |
160 | rtx r4 = gen_rtx_REG (SImode, 4); | |
161 | rtx r5 = gen_rtx_REG (SImode, 5); | |
162 | ||
163 | sprintf (entry, "__movmemSI%d", bytes); | |
164 | function_symbol (func_addr_rtx, entry, SFUNC_STATIC); | |
165 | force_into (XEXP (operands[0], 0), r4); | |
166 | force_into (XEXP (operands[1], 0), r5); | |
167 | emit_insn (gen_block_move_real (func_addr_rtx)); | |
168 | return true; | |
169 | } | |
170 | ||
171 | /* This is the same number of bytes as a memcpy call, but to a different | |
172 | less common function name, so this will occasionally use more space. */ | |
173 | if (! optimize_size) | |
174 | { | |
175 | rtx func_addr_rtx = gen_reg_rtx (Pmode); | |
176 | int final_switch, while_loop; | |
177 | rtx r4 = gen_rtx_REG (SImode, 4); | |
178 | rtx r5 = gen_rtx_REG (SImode, 5); | |
179 | rtx r6 = gen_rtx_REG (SImode, 6); | |
180 | ||
181 | function_symbol (func_addr_rtx, "__movmem", SFUNC_STATIC); | |
182 | force_into (XEXP (operands[0], 0), r4); | |
183 | force_into (XEXP (operands[1], 0), r5); | |
184 | ||
185 | /* r6 controls the size of the move. 16 is decremented from it | |
186 | for each 64 bytes moved. Then the negative bit left over is used | |
187 | as an index into a list of move instructions. e.g., a 72 byte move | |
188 | would be set up with size(r6) = 14, for one iteration through the | |
189 | big while loop, and a switch of -2 for the last part. */ | |
190 | ||
191 | final_switch = 16 - ((bytes / 4) % 16); | |
192 | while_loop = ((bytes / 4) / 16 - 1) * 16; | |
193 | emit_insn (gen_move_insn (r6, GEN_INT (while_loop + final_switch))); | |
194 | emit_insn (gen_block_lump_real (func_addr_rtx)); | |
195 | return true; | |
196 | } | |
197 | ||
198 | return false; | |
199 | } | |
200 | ||
f4a24c51 | 201 | static const int prob_unlikely = REG_BR_PROB_BASE / 10; |
202 | static const int prob_likely = REG_BR_PROB_BASE / 4; | |
c2daae6e | 203 | |
ccd57e8a | 204 | /* Emit code to perform a strcmp. |
205 | ||
206 | OPERANDS[0] is the destination. | |
207 | OPERANDS[1] is the first string. | |
208 | OPERANDS[2] is the second string. | |
c2daae6e | 209 | OPERANDS[3] is the known alignment. */ |
ccd57e8a | 210 | bool |
211 | sh_expand_cmpstr (rtx *operands) | |
212 | { | |
c2daae6e | 213 | rtx addr1 = operands[1]; |
214 | rtx addr2 = operands[2]; | |
215 | rtx s1_addr = copy_addr_to_reg (XEXP (addr1, 0)); | |
216 | rtx s2_addr = copy_addr_to_reg (XEXP (addr2, 0)); | |
ccd57e8a | 217 | rtx tmp0 = gen_reg_rtx (SImode); |
218 | rtx tmp1 = gen_reg_rtx (SImode); | |
219 | rtx tmp2 = gen_reg_rtx (SImode); | |
220 | rtx tmp3 = gen_reg_rtx (SImode); | |
221 | ||
c2daae6e | 222 | rtx jump; |
79f6a8ed | 223 | rtx_code_label *L_return = gen_label_rtx (); |
224 | rtx_code_label *L_loop_byte = gen_label_rtx (); | |
225 | rtx_code_label *L_end_loop_byte = gen_label_rtx (); | |
226 | rtx_code_label *L_loop_long = gen_label_rtx (); | |
227 | rtx_code_label *L_end_loop_long = gen_label_rtx (); | |
ccd57e8a | 228 | |
b421555d | 229 | int align = INTVAL (operands[3]); |
ccd57e8a | 230 | |
231 | emit_move_insn (tmp0, const0_rtx); | |
232 | ||
b421555d | 233 | if (align < 4) |
234 | { | |
235 | emit_insn (gen_iorsi3 (tmp1, s1_addr, s2_addr)); | |
f07efc97 | 236 | emit_insn (gen_tstsi_t (tmp1, GEN_INT (3))); |
b421555d | 237 | jump = emit_jump_insn (gen_branch_false (L_loop_byte)); |
238 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
239 | } | |
ccd57e8a | 240 | |
c2daae6e | 241 | addr1 = adjust_automodify_address (addr1, SImode, s1_addr, 0); |
242 | addr2 = adjust_automodify_address (addr2, SImode, s2_addr, 0); | |
ccd57e8a | 243 | |
244 | /* tmp2 is aligned, OK to load. */ | |
245 | emit_move_insn (tmp3, addr2); | |
246 | emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 4)); | |
247 | ||
f4a24c51 | 248 | /* start long loop. */ |
ccd57e8a | 249 | emit_label (L_loop_long); |
250 | ||
251 | emit_move_insn (tmp2, tmp3); | |
252 | ||
253 | /* tmp1 is aligned, OK to load. */ | |
254 | emit_move_insn (tmp1, addr1); | |
255 | emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, 4)); | |
256 | ||
257 | /* Is there a 0 byte ? */ | |
258 | emit_insn (gen_andsi3 (tmp3, tmp3, tmp1)); | |
259 | ||
260 | emit_insn (gen_cmpstr_t (tmp0, tmp3)); | |
261 | jump = emit_jump_insn (gen_branch_true (L_end_loop_long)); | |
262 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
263 | ||
264 | emit_insn (gen_cmpeqsi_t (tmp1, tmp2)); | |
265 | ||
266 | /* tmp2 is aligned, OK to load. */ | |
267 | emit_move_insn (tmp3, addr2); | |
268 | emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 4)); | |
269 | ||
270 | jump = emit_jump_insn (gen_branch_true (L_loop_long)); | |
271 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
272 | /* end loop. */ | |
273 | ||
b421555d | 274 | /* Fallthu, substract words. */ |
ccd57e8a | 275 | if (TARGET_LITTLE_ENDIAN) |
276 | { | |
277 | rtx low_1 = gen_lowpart (HImode, tmp1); | |
278 | rtx low_2 = gen_lowpart (HImode, tmp2); | |
279 | ||
280 | emit_insn (gen_rotlhi3_8 (low_1, low_1)); | |
281 | emit_insn (gen_rotlhi3_8 (low_2, low_2)); | |
282 | emit_insn (gen_rotlsi3_16 (tmp1, tmp1)); | |
283 | emit_insn (gen_rotlsi3_16 (tmp2, tmp2)); | |
284 | emit_insn (gen_rotlhi3_8 (low_1, low_1)); | |
285 | emit_insn (gen_rotlhi3_8 (low_2, low_2)); | |
286 | } | |
287 | ||
288 | jump = emit_jump_insn (gen_jump_compact (L_return)); | |
289 | emit_barrier_after (jump); | |
290 | ||
ccd57e8a | 291 | emit_label (L_end_loop_long); |
292 | ||
293 | emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, -4)); | |
294 | emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, -4)); | |
295 | ||
b421555d | 296 | /* start byte loop. */ |
c2daae6e | 297 | addr1 = adjust_address (addr1, QImode, 0); |
298 | addr2 = adjust_address (addr2, QImode, 0); | |
b421555d | 299 | |
ccd57e8a | 300 | emit_label (L_loop_byte); |
301 | ||
302 | emit_insn (gen_extendqisi2 (tmp2, addr2)); | |
303 | emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 1)); | |
304 | ||
305 | emit_insn (gen_extendqisi2 (tmp1, addr1)); | |
306 | emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, 1)); | |
307 | ||
308 | emit_insn (gen_cmpeqsi_t (tmp2, const0_rtx)); | |
309 | jump = emit_jump_insn (gen_branch_true (L_end_loop_byte)); | |
310 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
311 | ||
312 | emit_insn (gen_cmpeqsi_t (tmp1, tmp2)); | |
b421555d | 313 | if (flag_delayed_branch) |
314 | emit_insn (gen_zero_extendqisi2 (tmp2, gen_lowpart (QImode, tmp2))); | |
315 | jump = emit_jump_insn (gen_branch_true (L_loop_byte)); | |
ccd57e8a | 316 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); |
317 | /* end loop. */ | |
318 | ||
319 | emit_label (L_end_loop_byte); | |
320 | ||
b421555d | 321 | if (! flag_delayed_branch) |
322 | emit_insn (gen_zero_extendqisi2 (tmp2, gen_lowpart (QImode, tmp2))); | |
ccd57e8a | 323 | emit_insn (gen_zero_extendqisi2 (tmp1, gen_lowpart (QImode, tmp1))); |
324 | ||
325 | emit_label (L_return); | |
326 | ||
327 | emit_insn (gen_subsi3 (operands[0], tmp1, tmp2)); | |
328 | ||
329 | return true; | |
330 | } | |
331 | ||
b421555d | 332 | /* Emit code to perform a strncmp. |
333 | ||
334 | OPERANDS[0] is the destination. | |
335 | OPERANDS[1] is the first string. | |
336 | OPERANDS[2] is the second string. | |
337 | OPERANDS[3] is the length. | |
c2daae6e | 338 | OPERANDS[4] is the known alignment. */ |
b421555d | 339 | bool |
340 | sh_expand_cmpnstr (rtx *operands) | |
341 | { | |
c2daae6e | 342 | rtx addr1 = operands[1]; |
343 | rtx addr2 = operands[2]; | |
344 | rtx s1_addr = copy_addr_to_reg (XEXP (addr1, 0)); | |
345 | rtx s2_addr = copy_addr_to_reg (XEXP (addr2, 0)); | |
b421555d | 346 | rtx tmp1 = gen_reg_rtx (SImode); |
347 | rtx tmp2 = gen_reg_rtx (SImode); | |
348 | ||
c2daae6e | 349 | rtx jump; |
79f6a8ed | 350 | rtx_code_label *L_return = gen_label_rtx (); |
351 | rtx_code_label *L_loop_byte = gen_label_rtx (); | |
352 | rtx_code_label *L_end_loop_byte = gen_label_rtx (); | |
b421555d | 353 | |
b421555d | 354 | rtx len = force_reg (SImode, operands[3]); |
75c9129c | 355 | int constp = CONST_INT_P (operands[3]); |
b421555d | 356 | |
f4a24c51 | 357 | /* Loop on a register count. */ |
75c9129c | 358 | if (constp) |
b421555d | 359 | { |
75c9129c | 360 | rtx tmp0 = gen_reg_rtx (SImode); |
b421555d | 361 | rtx tmp3 = gen_reg_rtx (SImode); |
362 | rtx lenw = gen_reg_rtx (SImode); | |
b421555d | 363 | |
79f6a8ed | 364 | rtx_code_label *L_loop_long = gen_label_rtx (); |
365 | rtx_code_label *L_end_loop_long = gen_label_rtx (); | |
b421555d | 366 | |
75c9129c | 367 | int align = INTVAL (operands[4]); |
368 | int bytes = INTVAL (operands[3]); | |
369 | int witers = bytes / 4; | |
370 | ||
371 | if (witers > 1) | |
f4a24c51 | 372 | { |
373 | addr1 = adjust_automodify_address (addr1, SImode, s1_addr, 0); | |
374 | addr2 = adjust_automodify_address (addr2, SImode, s2_addr, 0); | |
375 | ||
376 | emit_move_insn (tmp0, const0_rtx); | |
377 | ||
378 | if (align < 4) | |
379 | { | |
380 | emit_insn (gen_iorsi3 (tmp1, s1_addr, s2_addr)); | |
f07efc97 | 381 | emit_insn (gen_tstsi_t (tmp1, GEN_INT (3))); |
f4a24c51 | 382 | jump = emit_jump_insn (gen_branch_false (L_loop_byte)); |
383 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
384 | } | |
385 | ||
386 | /* word count. Do we have iterations ? */ | |
387 | emit_insn (gen_lshrsi3 (lenw, len, GEN_INT (2))); | |
388 | ||
389 | /* start long loop. */ | |
390 | emit_label (L_loop_long); | |
391 | ||
392 | /* tmp2 is aligned, OK to load. */ | |
393 | emit_move_insn (tmp2, addr2); | |
394 | emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, | |
395 | GET_MODE_SIZE (SImode))); | |
396 | ||
397 | /* tmp1 is aligned, OK to load. */ | |
398 | emit_move_insn (tmp1, addr1); | |
399 | emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, | |
400 | GET_MODE_SIZE (SImode))); | |
401 | ||
402 | /* Is there a 0 byte ? */ | |
403 | emit_insn (gen_andsi3 (tmp3, tmp2, tmp1)); | |
404 | ||
405 | emit_insn (gen_cmpstr_t (tmp0, tmp3)); | |
406 | jump = emit_jump_insn (gen_branch_true (L_end_loop_long)); | |
407 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
408 | ||
409 | emit_insn (gen_cmpeqsi_t (tmp1, tmp2)); | |
410 | jump = emit_jump_insn (gen_branch_false (L_end_loop_long)); | |
411 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
412 | ||
413 | if (TARGET_SH2) | |
414 | emit_insn (gen_dect (lenw, lenw)); | |
415 | else | |
416 | { | |
417 | emit_insn (gen_addsi3 (lenw, lenw, GEN_INT (-1))); | |
418 | emit_insn (gen_tstsi_t (lenw, lenw)); | |
419 | } | |
420 | ||
421 | jump = emit_jump_insn (gen_branch_false (L_loop_long)); | |
422 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
423 | ||
424 | int sbytes = bytes % 4; | |
425 | ||
426 | /* end loop. Reached max iterations. */ | |
427 | if (sbytes == 0) | |
428 | { | |
7756601d | 429 | emit_insn (gen_subsi3 (operands[0], tmp1, tmp2)); |
f4a24c51 | 430 | jump = emit_jump_insn (gen_jump_compact (L_return)); |
431 | emit_barrier_after (jump); | |
432 | } | |
433 | else | |
434 | { | |
435 | /* Remaining bytes to check. */ | |
436 | ||
437 | addr1 = adjust_automodify_address (addr1, QImode, s1_addr, 0); | |
438 | addr2 = adjust_automodify_address (addr2, QImode, s2_addr, 0); | |
439 | ||
440 | while (sbytes--) | |
441 | { | |
442 | emit_insn (gen_extendqisi2 (tmp1, addr1)); | |
443 | emit_insn (gen_extendqisi2 (tmp2, addr2)); | |
444 | ||
445 | emit_insn (gen_cmpeqsi_t (tmp2, const0_rtx)); | |
446 | jump = emit_jump_insn (gen_branch_true (L_end_loop_byte)); | |
447 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
448 | ||
449 | emit_insn (gen_cmpeqsi_t (tmp1, tmp2)); | |
450 | if (flag_delayed_branch) | |
451 | emit_insn (gen_zero_extendqisi2 (tmp2, | |
452 | gen_lowpart (QImode, | |
453 | tmp2))); | |
454 | jump = emit_jump_insn (gen_branch_false (L_end_loop_byte)); | |
455 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
456 | ||
457 | addr1 = adjust_address (addr1, QImode, | |
458 | GET_MODE_SIZE (QImode)); | |
459 | addr2 = adjust_address (addr2, QImode, | |
460 | GET_MODE_SIZE (QImode)); | |
461 | } | |
462 | ||
463 | jump = emit_jump_insn (gen_jump_compact( L_end_loop_byte)); | |
464 | emit_barrier_after (jump); | |
465 | } | |
466 | ||
467 | emit_label (L_end_loop_long); | |
468 | ||
469 | /* Found last word. Restart it byte per byte. */ | |
470 | ||
471 | emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, | |
472 | -GET_MODE_SIZE (SImode))); | |
473 | emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, | |
474 | -GET_MODE_SIZE (SImode))); | |
475 | ||
476 | /* fall thru. */ | |
477 | } | |
75c9129c | 478 | |
479 | addr1 = adjust_automodify_address (addr1, QImode, s1_addr, 0); | |
480 | addr2 = adjust_automodify_address (addr2, QImode, s2_addr, 0); | |
481 | ||
482 | while (bytes--) | |
f4a24c51 | 483 | { |
484 | emit_insn (gen_extendqisi2 (tmp1, addr1)); | |
485 | emit_insn (gen_extendqisi2 (tmp2, addr2)); | |
486 | ||
487 | emit_insn (gen_cmpeqsi_t (tmp2, const0_rtx)); | |
488 | jump = emit_jump_insn (gen_branch_true (L_end_loop_byte)); | |
489 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
490 | ||
491 | emit_insn (gen_cmpeqsi_t (tmp1, tmp2)); | |
492 | if (flag_delayed_branch) | |
493 | emit_insn (gen_zero_extendqisi2 (tmp2, | |
494 | gen_lowpart (QImode, tmp2))); | |
495 | jump = emit_jump_insn (gen_branch_false (L_end_loop_byte)); | |
496 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
497 | ||
498 | addr1 = adjust_address (addr1, QImode, GET_MODE_SIZE (QImode)); | |
499 | addr2 = adjust_address (addr2, QImode, GET_MODE_SIZE (QImode)); | |
500 | } | |
75c9129c | 501 | |
502 | jump = emit_jump_insn (gen_jump_compact( L_end_loop_byte)); | |
503 | emit_barrier_after (jump); | |
b421555d | 504 | } |
7756601d | 505 | else |
506 | { | |
507 | emit_insn (gen_cmpeqsi_t (len, const0_rtx)); | |
508 | emit_move_insn (operands[0], const0_rtx); | |
509 | jump = emit_jump_insn (gen_branch_true (L_return)); | |
510 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
511 | } | |
b421555d | 512 | |
75c9129c | 513 | addr1 = adjust_automodify_address (addr1, QImode, s1_addr, 0); |
514 | addr2 = adjust_automodify_address (addr2, QImode, s2_addr, 0); | |
c2daae6e | 515 | |
516 | emit_label (L_loop_byte); | |
517 | ||
518 | emit_insn (gen_extendqisi2 (tmp2, addr2)); | |
519 | emit_move_insn (s2_addr, plus_constant (Pmode, s2_addr, 1)); | |
520 | ||
521 | emit_insn (gen_extendqisi2 (tmp1, addr1)); | |
522 | emit_move_insn (s1_addr, plus_constant (Pmode, s1_addr, 1)); | |
523 | ||
524 | emit_insn (gen_cmpeqsi_t (tmp2, const0_rtx)); | |
525 | jump = emit_jump_insn (gen_branch_true (L_end_loop_byte)); | |
526 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
527 | ||
528 | emit_insn (gen_cmpeqsi_t (tmp1, tmp2)); | |
529 | if (flag_delayed_branch) | |
530 | emit_insn (gen_zero_extendqisi2 (tmp2, gen_lowpart (QImode, tmp2))); | |
531 | jump = emit_jump_insn (gen_branch_false (L_end_loop_byte)); | |
532 | add_int_reg_note (jump, REG_BR_PROB, prob_unlikely); | |
533 | ||
534 | if (TARGET_SH2) | |
535 | emit_insn (gen_dect (len, len)); | |
536 | else | |
537 | { | |
538 | emit_insn (gen_addsi3 (len, len, GEN_INT (-1))); | |
539 | emit_insn (gen_tstsi_t (len, len)); | |
540 | } | |
541 | ||
542 | jump = emit_jump_insn (gen_branch_false (L_loop_byte)); | |
543 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
544 | /* end byte loop. */ | |
545 | ||
546 | emit_label (L_end_loop_byte); | |
547 | ||
548 | if (! flag_delayed_branch) | |
549 | emit_insn (gen_zero_extendqisi2 (tmp2, gen_lowpart (QImode, tmp2))); | |
550 | emit_insn (gen_zero_extendqisi2 (tmp1, gen_lowpart (QImode, tmp1))); | |
551 | ||
c2daae6e | 552 | emit_insn (gen_subsi3 (operands[0], tmp1, tmp2)); |
553 | ||
7756601d | 554 | emit_label (L_return); |
555 | ||
c2daae6e | 556 | return true; |
557 | } | |
558 | ||
f4a24c51 | 559 | /* Emit code to perform a strlen. |
c2daae6e | 560 | |
561 | OPERANDS[0] is the destination. | |
562 | OPERANDS[1] is the string. | |
563 | OPERANDS[2] is the char to search. | |
564 | OPERANDS[3] is the alignment. */ | |
565 | bool | |
566 | sh_expand_strlen (rtx *operands) | |
567 | { | |
568 | rtx addr1 = operands[1]; | |
569 | rtx current_addr = copy_addr_to_reg (XEXP (addr1, 0)); | |
570 | rtx start_addr = gen_reg_rtx (Pmode); | |
571 | rtx tmp0 = gen_reg_rtx (SImode); | |
572 | rtx tmp1 = gen_reg_rtx (SImode); | |
79f6a8ed | 573 | rtx_code_label *L_return = gen_label_rtx (); |
574 | rtx_code_label *L_loop_byte = gen_label_rtx (); | |
c2daae6e | 575 | |
576 | rtx jump; | |
79f6a8ed | 577 | rtx_code_label *L_loop_long = gen_label_rtx (); |
578 | rtx_code_label *L_end_loop_long = gen_label_rtx (); | |
c2daae6e | 579 | |
580 | int align = INTVAL (operands[3]); | |
581 | ||
582 | emit_move_insn (operands[0], GEN_INT (-1)); | |
583 | ||
584 | /* remember start of string. */ | |
585 | emit_move_insn (start_addr, current_addr); | |
586 | ||
587 | if (align < 4) | |
588 | { | |
f07efc97 | 589 | emit_insn (gen_tstsi_t (current_addr, GEN_INT (3))); |
c2daae6e | 590 | jump = emit_jump_insn (gen_branch_false (L_loop_byte)); |
591 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
592 | } | |
593 | ||
594 | emit_move_insn (tmp0, operands[2]); | |
595 | ||
596 | addr1 = adjust_automodify_address (addr1, SImode, current_addr, 0); | |
597 | ||
9c00010f | 598 | /* start long loop. */ |
c2daae6e | 599 | emit_label (L_loop_long); |
600 | ||
601 | /* tmp1 is aligned, OK to load. */ | |
602 | emit_move_insn (tmp1, addr1); | |
603 | emit_move_insn (current_addr, plus_constant (Pmode, current_addr, 4)); | |
604 | ||
605 | /* Is there a 0 byte ? */ | |
606 | emit_insn (gen_cmpstr_t (tmp0, tmp1)); | |
b421555d | 607 | |
c2daae6e | 608 | jump = emit_jump_insn (gen_branch_false (L_loop_long)); |
609 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
610 | /* end loop. */ | |
b421555d | 611 | |
c2daae6e | 612 | emit_label (L_end_loop_long); |
b421555d | 613 | |
c2daae6e | 614 | emit_move_insn (current_addr, plus_constant (Pmode, current_addr, -4)); |
b421555d | 615 | |
c2daae6e | 616 | addr1 = adjust_address (addr1, QImode, 0); |
b421555d | 617 | |
751d4d6f | 618 | /* unroll remaining bytes. */ |
9c00010f | 619 | for (int i = 0; i < 4; ++i) |
620 | { | |
621 | emit_insn (gen_extendqisi2 (tmp1, addr1)); | |
622 | emit_move_insn (current_addr, plus_constant (Pmode, current_addr, 1)); | |
623 | emit_insn (gen_cmpeqsi_t (tmp1, const0_rtx)); | |
624 | jump = emit_jump_insn (gen_branch_true (L_return)); | |
625 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
626 | } | |
751d4d6f | 627 | |
751d4d6f | 628 | emit_barrier_after (jump); |
629 | ||
630 | /* start byte loop. */ | |
c2daae6e | 631 | emit_label (L_loop_byte); |
b421555d | 632 | |
c2daae6e | 633 | emit_insn (gen_extendqisi2 (tmp1, addr1)); |
634 | emit_move_insn (current_addr, plus_constant (Pmode, current_addr, 1)); | |
b421555d | 635 | |
c2daae6e | 636 | emit_insn (gen_cmpeqsi_t (tmp1, const0_rtx)); |
637 | jump = emit_jump_insn (gen_branch_false (L_loop_byte)); | |
638 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
b421555d | 639 | |
c2daae6e | 640 | /* end loop. */ |
b421555d | 641 | |
751d4d6f | 642 | emit_label (L_return); |
643 | ||
9c00010f | 644 | emit_insn (gen_addsi3 (start_addr, start_addr, GEN_INT (1))); |
c2daae6e | 645 | emit_insn (gen_subsi3 (operands[0], current_addr, start_addr)); |
b421555d | 646 | |
c2daae6e | 647 | return true; |
b421555d | 648 | } |
1878fb5b | 649 | |
f4a24c51 | 650 | /* Emit code to perform a memset. |
1878fb5b | 651 | |
652 | OPERANDS[0] is the destination. | |
653 | OPERANDS[1] is the size; | |
654 | OPERANDS[2] is the char to search. | |
655 | OPERANDS[3] is the alignment. */ | |
656 | void | |
657 | sh_expand_setmem (rtx *operands) | |
658 | { | |
79f6a8ed | 659 | rtx_code_label *L_loop_byte = gen_label_rtx (); |
660 | rtx_code_label *L_loop_word = gen_label_rtx (); | |
661 | rtx_code_label *L_return = gen_label_rtx (); | |
1878fb5b | 662 | rtx jump; |
663 | rtx dest = copy_rtx (operands[0]); | |
664 | rtx dest_addr = copy_addr_to_reg (XEXP (dest, 0)); | |
665 | rtx val = force_reg (SImode, operands[2]); | |
666 | int align = INTVAL (operands[3]); | |
1878fb5b | 667 | rtx len = force_reg (SImode, operands[1]); |
668 | ||
669 | if (! CONST_INT_P (operands[1])) | |
670 | return; | |
671 | ||
f4a24c51 | 672 | int count = INTVAL (operands[1]); |
1878fb5b | 673 | |
674 | if (CONST_INT_P (operands[2]) | |
675 | && (INTVAL (operands[2]) == 0 || INTVAL (operands[2]) == -1) && count > 8) | |
676 | { | |
677 | rtx lenw = gen_reg_rtx (SImode); | |
678 | ||
679 | if (align < 4) | |
f4a24c51 | 680 | { |
f07efc97 | 681 | emit_insn (gen_tstsi_t (dest_addr, GEN_INT (3))); |
f4a24c51 | 682 | jump = emit_jump_insn (gen_branch_false (L_loop_byte)); |
683 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
684 | } | |
1878fb5b | 685 | |
f4a24c51 | 686 | /* word count. Do we have iterations ? */ |
1878fb5b | 687 | emit_insn (gen_lshrsi3 (lenw, len, GEN_INT (2))); |
688 | ||
689 | dest = adjust_automodify_address (dest, SImode, dest_addr, 0); | |
690 | ||
691 | /* start loop. */ | |
692 | emit_label (L_loop_word); | |
693 | ||
694 | if (TARGET_SH2) | |
695 | emit_insn (gen_dect (lenw, lenw)); | |
696 | else | |
f4a24c51 | 697 | { |
698 | emit_insn (gen_addsi3 (lenw, lenw, GEN_INT (-1))); | |
699 | emit_insn (gen_tstsi_t (lenw, lenw)); | |
700 | } | |
1878fb5b | 701 | |
702 | emit_move_insn (dest, val); | |
703 | emit_move_insn (dest_addr, plus_constant (Pmode, dest_addr, | |
f4a24c51 | 704 | GET_MODE_SIZE (SImode))); |
1878fb5b | 705 | |
706 | ||
707 | jump = emit_jump_insn (gen_branch_false (L_loop_word)); | |
708 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
709 | count = count % 4; | |
710 | ||
711 | dest = adjust_address (dest, QImode, 0); | |
712 | ||
713 | val = gen_lowpart (QImode, val); | |
714 | ||
715 | while (count--) | |
f4a24c51 | 716 | { |
717 | emit_move_insn (dest, val); | |
718 | emit_move_insn (dest_addr, plus_constant (Pmode, dest_addr, | |
719 | GET_MODE_SIZE (QImode))); | |
720 | } | |
1878fb5b | 721 | |
722 | jump = emit_jump_insn (gen_jump_compact (L_return)); | |
723 | emit_barrier_after (jump); | |
724 | } | |
725 | ||
726 | dest = adjust_automodify_address (dest, QImode, dest_addr, 0); | |
727 | ||
728 | /* start loop. */ | |
729 | emit_label (L_loop_byte); | |
730 | ||
731 | if (TARGET_SH2) | |
732 | emit_insn (gen_dect (len, len)); | |
733 | else | |
734 | { | |
735 | emit_insn (gen_addsi3 (len, len, GEN_INT (-1))); | |
736 | emit_insn (gen_tstsi_t (len, len)); | |
737 | } | |
738 | ||
739 | val = gen_lowpart (QImode, val); | |
740 | emit_move_insn (dest, val); | |
741 | emit_move_insn (dest_addr, plus_constant (Pmode, dest_addr, | |
742 | GET_MODE_SIZE (QImode))); | |
743 | ||
744 | jump = emit_jump_insn (gen_branch_false (L_loop_byte)); | |
745 | add_int_reg_note (jump, REG_BR_PROB, prob_likely); | |
746 | ||
747 | emit_label (L_return); | |
1878fb5b | 748 | } |