Mercurial > hg > CbC > CbC_gcc
annotate gcc/tree-call-cdce.c @ 63:b7f97abdc517 gcc-4.6-20100522
update gcc from gcc-4.5.0 to gcc-4.6
| author | ryoma <e075725@ie.u-ryukyu.ac.jp> |
|---|---|
| date | Mon, 24 May 2010 12:47:05 +0900 |
| parents | 77e2b8dfacca |
| children | f6334be47118 |
| rev | line source |
|---|---|
| 0 | 1 /* Conditional Dead Call Elimination pass for the GNU compiler. |
|
63
b7f97abdc517
update gcc from gcc-4.5.0 to gcc-4.6
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
55
diff
changeset
|
2 Copyright (C) 2008, 2009, 2010 |
| 0 | 3 Free Software Foundation, Inc. |
| 4 Contributed by Xinliang David Li <davidxl@google.com> | |
| 5 | |
| 6 This file is part of GCC. | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
7 |
| 0 | 8 GCC is free software; you can redistribute it and/or modify it |
| 9 under the terms of the GNU General Public License as published by the | |
| 10 Free Software Foundation; either version 3, or (at your option) any | |
| 11 later version. | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
12 |
| 0 | 13 GCC is distributed in the hope that it will be useful, but WITHOUT |
| 14 ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
| 15 FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
| 16 for more details. | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
17 |
| 0 | 18 You should have received a copy of the GNU General Public License |
| 19 along with GCC; see the file COPYING3. If not see | |
| 20 <http://www.gnu.org/licenses/>. */ | |
| 21 | |
| 22 #include "config.h" | |
| 23 #include "system.h" | |
| 24 #include "coretypes.h" | |
| 25 #include "tm.h" | |
| 26 #include "basic-block.h" | |
| 27 #include "tree.h" | |
| 28 #include "diagnostic.h" | |
|
63
b7f97abdc517
update gcc from gcc-4.5.0 to gcc-4.6
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
55
diff
changeset
|
29 #include "gimple-pretty-print.h" |
| 0 | 30 #include "tree-flow.h" |
| 31 #include "gimple.h" | |
| 32 #include "tree-dump.h" | |
| 33 #include "tree-pass.h" | |
| 34 #include "timevar.h" | |
| 35 #include "flags.h" | |
| 36 | |
| 37 | |
| 38 /* Conditional dead call elimination | |
| 39 | |
| 40 Some builtin functions can set errno on error conditions, but they | |
| 41 are otherwise pure. If the result of a call to such a function is | |
| 42 not used, the compiler can still not eliminate the call without | |
| 43 powerful interprocedural analysis to prove that the errno is not | |
| 44 checked. However, if the conditions under which the error occurs | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
45 are known, the compiler can conditionally dead code eliminate the |
| 0 | 46 calls by shrink-wrapping the semi-dead calls into the error condition: |
| 47 | |
| 48 built_in_call (args) | |
| 49 ==> | |
| 50 if (error_cond (args)) | |
| 51 built_in_call (args) | |
| 52 | |
| 53 An actual simple example is : | |
| 54 log (x); // Mostly dead call | |
| 55 ==> | |
| 56 if (x < 0) | |
| 57 log (x); | |
| 58 With this change, call to log (x) is effectively eliminated, as | |
| 59 in majority of the cases, log won't be called with x out of | |
| 60 range. The branch is totally predictable, so the branch cost | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
61 is low. |
| 0 | 62 |
| 63 Note that library functions are not supposed to clear errno to zero without | |
| 64 error. See IEEE Std 1003.1, section 2.3 Error Numbers, and section 7.5:3 of | |
| 65 ISO/IEC 9899 (C99). | |
| 66 | |
| 67 The condition wrapping the builtin call is conservatively set to avoid too | |
| 68 aggressive (wrong) shrink wrapping. The optimization is called conditional | |
| 69 dead call elimination because the call is eliminated under the condition | |
| 70 that the input arguments would not lead to domain or range error (for | |
| 71 instance when x <= 0 for a log (x) call), however the chances that the error | |
| 72 condition is hit is very low (those builtin calls which are conditionally | |
| 73 dead are usually part of the C++ abstraction penalty exposed after | |
| 74 inlining). */ | |
| 75 | |
| 76 | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
77 /* A structure for representing input domain of |
| 0 | 78 a function argument in integer. If the lower |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
79 bound is -inf, has_lb is set to false. If the |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
80 upper bound is +inf, has_ub is false. |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
81 is_lb_inclusive and is_ub_inclusive are flags |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
82 to indicate if lb and ub value are inclusive |
| 0 | 83 respectively. */ |
| 84 | |
| 85 typedef struct input_domain | |
| 86 { | |
| 87 int lb; | |
| 88 int ub; | |
| 89 bool has_lb; | |
| 90 bool has_ub; | |
| 91 bool is_lb_inclusive; | |
| 92 bool is_ub_inclusive; | |
| 93 } inp_domain; | |
| 94 | |
| 95 /* A helper function to construct and return an input | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
96 domain object. LB is the lower bound, HAS_LB is |
| 0 | 97 a boolean flag indicating if the lower bound exists, |
| 98 and LB_INCLUSIVE is a boolean flag indicating if the | |
| 99 lower bound is inclusive or not. UB, HAS_UB, and | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
100 UB_INCLUSIVE have the same meaning, but for upper |
| 0 | 101 bound of the domain. */ |
| 102 | |
| 103 static inp_domain | |
| 104 get_domain (int lb, bool has_lb, bool lb_inclusive, | |
| 105 int ub, bool has_ub, bool ub_inclusive) | |
| 106 { | |
| 107 inp_domain domain; | |
| 108 domain.lb = lb; | |
| 109 domain.has_lb = has_lb; | |
| 110 domain.is_lb_inclusive = lb_inclusive; | |
| 111 domain.ub = ub; | |
| 112 domain.has_ub = has_ub; | |
| 113 domain.is_ub_inclusive = ub_inclusive; | |
| 114 return domain; | |
| 115 } | |
| 116 | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
117 /* A helper function to check the target format for the |
| 0 | 118 argument type. In this implementation, only IEEE formats |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
119 are supported. ARG is the call argument to be checked. |
| 0 | 120 Returns true if the format is supported. To support other |
| 121 target formats, function get_no_error_domain needs to be | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
122 enhanced to have range bounds properly computed. Since |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
123 the check is cheap (very small number of candidates |
| 0 | 124 to be checked), the result is not cached for each float type. */ |
| 125 | |
| 126 static bool | |
| 127 check_target_format (tree arg) | |
| 128 { | |
| 129 tree type; | |
| 130 enum machine_mode mode; | |
| 131 const struct real_format *rfmt; | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
132 |
| 0 | 133 type = TREE_TYPE (arg); |
| 134 mode = TYPE_MODE (type); | |
| 135 rfmt = REAL_MODE_FORMAT (mode); | |
| 136 if ((mode == SFmode | |
| 137 && (rfmt == &ieee_single_format || rfmt == &mips_single_format | |
| 138 || rfmt == &motorola_single_format)) | |
| 139 || (mode == DFmode | |
| 140 && (rfmt == &ieee_double_format || rfmt == &mips_double_format | |
| 141 || rfmt == &motorola_double_format)) | |
| 142 /* For long double, we can not really check XFmode | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
143 which is only defined on intel platforms. |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
144 Candidate pre-selection using builtin function |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
145 code guarantees that we are checking formats |
| 0 | 146 for long double modes: double, quad, and extended. */ |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
147 || (mode != SFmode && mode != DFmode |
| 0 | 148 && (rfmt == &ieee_quad_format |
| 149 || rfmt == &mips_quad_format | |
| 150 || rfmt == &ieee_extended_motorola_format | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
151 || rfmt == &ieee_extended_intel_96_format |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
152 || rfmt == &ieee_extended_intel_128_format |
| 0 | 153 || rfmt == &ieee_extended_intel_96_round_53_format))) |
| 154 return true; | |
| 155 | |
| 156 return false; | |
| 157 } | |
| 158 | |
| 159 | |
| 160 /* A helper function to help select calls to pow that are suitable for | |
| 161 conditional DCE transformation. It looks for pow calls that can be | |
| 162 guided with simple conditions. Such calls either have constant base | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
163 values or base values converted from integers. Returns true if |
| 0 | 164 the pow call POW_CALL is a candidate. */ |
| 165 | |
| 166 /* The maximum integer bit size for base argument of a pow call | |
| 167 that is suitable for shrink-wrapping transformation. */ | |
| 168 #define MAX_BASE_INT_BIT_SIZE 32 | |
| 169 | |
| 170 static bool | |
| 171 check_pow (gimple pow_call) | |
| 172 { | |
| 173 tree base, expn; | |
| 174 enum tree_code bc, ec; | |
| 175 | |
| 176 if (gimple_call_num_args (pow_call) != 2) | |
| 177 return false; | |
| 178 | |
| 179 base = gimple_call_arg (pow_call, 0); | |
| 180 expn = gimple_call_arg (pow_call, 1); | |
| 181 | |
| 182 if (!check_target_format (expn)) | |
| 183 return false; | |
| 184 | |
| 185 bc = TREE_CODE (base); | |
| 186 ec = TREE_CODE (expn); | |
| 187 | |
| 188 /* Folding candidates are not interesting. | |
| 189 Can actually assert that it is already folded. */ | |
| 190 if (ec == REAL_CST && bc == REAL_CST) | |
| 191 return false; | |
| 192 | |
| 193 if (bc == REAL_CST) | |
| 194 { | |
| 195 /* Only handle a fixed range of constant. */ | |
| 196 REAL_VALUE_TYPE mv; | |
| 197 REAL_VALUE_TYPE bcv = TREE_REAL_CST (base); | |
| 198 if (REAL_VALUES_EQUAL (bcv, dconst1)) | |
| 199 return false; | |
| 200 if (REAL_VALUES_LESS (bcv, dconst1)) | |
| 201 return false; | |
| 202 real_from_integer (&mv, TYPE_MODE (TREE_TYPE (base)), 256, 0, 1); | |
| 203 if (REAL_VALUES_LESS (mv, bcv)) | |
| 204 return false; | |
| 205 return true; | |
| 206 } | |
| 207 else if (bc == SSA_NAME) | |
| 208 { | |
| 209 tree base_val0, base_var, type; | |
| 210 gimple base_def; | |
| 211 int bit_sz; | |
| 212 | |
| 213 /* Only handles cases where base value is converted | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
214 from integer values. */ |
| 0 | 215 base_def = SSA_NAME_DEF_STMT (base); |
| 216 if (gimple_code (base_def) != GIMPLE_ASSIGN) | |
| 217 return false; | |
| 218 | |
| 219 if (gimple_assign_rhs_code (base_def) != FLOAT_EXPR) | |
| 220 return false; | |
| 221 base_val0 = gimple_assign_rhs1 (base_def); | |
| 222 | |
| 223 base_var = SSA_NAME_VAR (base_val0); | |
| 224 if (!DECL_P (base_var)) | |
| 225 return false; | |
| 226 | |
| 227 type = TREE_TYPE (base_var); | |
| 228 if (TREE_CODE (type) != INTEGER_TYPE) | |
| 229 return false; | |
| 230 bit_sz = TYPE_PRECISION (type); | |
| 231 /* If the type of the base is too wide, | |
| 232 the resulting shrink wrapping condition | |
| 233 will be too conservative. */ | |
| 234 if (bit_sz > MAX_BASE_INT_BIT_SIZE) | |
| 235 return false; | |
| 236 | |
| 237 return true; | |
| 238 } | |
| 239 else | |
| 240 return false; | |
| 241 } | |
| 242 | |
| 243 /* A helper function to help select candidate function calls that are | |
| 244 suitable for conditional DCE. Candidate functions must have single | |
| 245 valid input domain in this implementation except for pow (see check_pow). | |
| 246 Returns true if the function call is a candidate. */ | |
| 247 | |
| 248 static bool | |
| 249 check_builtin_call (gimple bcall) | |
| 250 { | |
| 251 tree arg; | |
| 252 | |
| 253 arg = gimple_call_arg (bcall, 0); | |
| 254 return check_target_format (arg); | |
| 255 } | |
| 256 | |
| 257 /* A helper function to determine if a builtin function call is a | |
| 258 candidate for conditional DCE. Returns true if the builtin call | |
| 259 is a candidate. */ | |
| 260 | |
| 261 static bool | |
| 262 is_call_dce_candidate (gimple call) | |
| 263 { | |
| 264 tree fn; | |
| 265 enum built_in_function fnc; | |
| 266 | |
| 267 /* Only potentially dead calls are considered. */ | |
| 268 if (gimple_call_lhs (call)) | |
| 269 return false; | |
| 270 | |
| 271 fn = gimple_call_fndecl (call); | |
| 272 if (!fn | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
273 || !DECL_BUILT_IN (fn) |
| 0 | 274 || (DECL_BUILT_IN_CLASS (fn) != BUILT_IN_NORMAL)) |
| 275 return false; | |
| 276 | |
| 277 fnc = DECL_FUNCTION_CODE (fn); | |
| 278 switch (fnc) | |
| 279 { | |
| 280 /* Trig functions. */ | |
| 281 CASE_FLT_FN (BUILT_IN_ACOS): | |
| 282 CASE_FLT_FN (BUILT_IN_ASIN): | |
| 283 /* Hyperbolic functions. */ | |
| 284 CASE_FLT_FN (BUILT_IN_ACOSH): | |
| 285 CASE_FLT_FN (BUILT_IN_ATANH): | |
| 286 CASE_FLT_FN (BUILT_IN_COSH): | |
| 287 CASE_FLT_FN (BUILT_IN_SINH): | |
| 288 /* Log functions. */ | |
| 289 CASE_FLT_FN (BUILT_IN_LOG): | |
| 290 CASE_FLT_FN (BUILT_IN_LOG2): | |
| 291 CASE_FLT_FN (BUILT_IN_LOG10): | |
| 292 CASE_FLT_FN (BUILT_IN_LOG1P): | |
| 293 /* Exp functions. */ | |
| 294 CASE_FLT_FN (BUILT_IN_EXP): | |
| 295 CASE_FLT_FN (BUILT_IN_EXP2): | |
| 296 CASE_FLT_FN (BUILT_IN_EXP10): | |
| 297 CASE_FLT_FN (BUILT_IN_EXPM1): | |
| 298 CASE_FLT_FN (BUILT_IN_POW10): | |
| 299 /* Sqrt. */ | |
| 300 CASE_FLT_FN (BUILT_IN_SQRT): | |
| 301 return check_builtin_call (call); | |
| 302 /* Special one: two argument pow. */ | |
| 303 case BUILT_IN_POW: | |
| 304 return check_pow (call); | |
| 305 default: | |
| 306 break; | |
| 307 } | |
| 308 | |
| 309 return false; | |
| 310 } | |
| 311 | |
| 312 | |
| 313 /* A helper function to generate gimple statements for | |
| 314 one bound comparison. ARG is the call argument to | |
| 315 be compared with the bound, LBUB is the bound value | |
| 316 in integer, TCODE is the tree_code of the comparison, | |
| 317 TEMP_NAME1/TEMP_NAME2 are names of the temporaries, | |
| 318 CONDS is a vector holding the produced GIMPLE statements, | |
| 319 and NCONDS points to the variable holding the number | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
320 of logical comparisons. CONDS is either empty or |
| 0 | 321 a list ended with a null tree. */ |
| 322 | |
| 323 static void | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
324 gen_one_condition (tree arg, int lbub, |
| 0 | 325 enum tree_code tcode, |
| 326 const char *temp_name1, | |
| 327 const char *temp_name2, | |
| 328 VEC (gimple, heap) *conds, | |
| 329 unsigned *nconds) | |
| 330 { | |
| 331 tree lbub_real_cst, lbub_cst, float_type; | |
| 332 tree temp, tempn, tempc, tempcn; | |
| 333 gimple stmt1, stmt2, stmt3; | |
| 334 | |
| 335 float_type = TREE_TYPE (arg); | |
| 336 lbub_cst = build_int_cst (integer_type_node, lbub); | |
| 337 lbub_real_cst = build_real_from_int_cst (float_type, lbub_cst); | |
| 338 | |
| 339 temp = create_tmp_var (float_type, temp_name1); | |
| 340 stmt1 = gimple_build_assign (temp, arg); | |
| 341 tempn = make_ssa_name (temp, stmt1); | |
| 342 gimple_assign_set_lhs (stmt1, tempn); | |
| 343 | |
| 344 tempc = create_tmp_var (boolean_type_node, temp_name2); | |
| 345 stmt2 = gimple_build_assign (tempc, | |
| 346 fold_build2 (tcode, | |
| 347 boolean_type_node, | |
| 348 tempn, lbub_real_cst)); | |
| 349 tempcn = make_ssa_name (tempc, stmt2); | |
| 350 gimple_assign_set_lhs (stmt2, tempcn); | |
| 351 | |
| 352 stmt3 = gimple_build_cond_from_tree (tempcn, NULL_TREE, NULL_TREE); | |
| 353 VEC_quick_push (gimple, conds, stmt1); | |
| 354 VEC_quick_push (gimple, conds, stmt2); | |
| 355 VEC_quick_push (gimple, conds, stmt3); | |
| 356 (*nconds)++; | |
| 357 } | |
| 358 | |
| 359 /* A helper function to generate GIMPLE statements for | |
| 360 out of input domain check. ARG is the call argument | |
| 361 to be runtime checked, DOMAIN holds the valid domain | |
| 362 for the given function, CONDS points to the vector | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
363 holding the result GIMPLE statements. *NCONDS is |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
364 the number of logical comparisons. This function |
| 0 | 365 produces no more than two logical comparisons, one |
| 366 for lower bound check, one for upper bound check. */ | |
| 367 | |
| 368 static void | |
| 369 gen_conditions_for_domain (tree arg, inp_domain domain, | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
370 VEC (gimple, heap) *conds, |
| 0 | 371 unsigned *nconds) |
| 372 { | |
| 373 if (domain.has_lb) | |
| 374 gen_one_condition (arg, domain.lb, | |
| 375 (domain.is_lb_inclusive | |
| 376 ? LT_EXPR : LE_EXPR), | |
| 377 "DCE_COND_LB", "DCE_COND_LB_TEST", | |
| 378 conds, nconds); | |
| 379 | |
| 380 if (domain.has_ub) | |
| 381 { | |
| 382 /* Now push a separator. */ | |
| 383 if (domain.has_lb) | |
| 384 VEC_quick_push (gimple, conds, NULL); | |
| 385 | |
| 386 gen_one_condition (arg, domain.ub, | |
| 387 (domain.is_ub_inclusive | |
| 388 ? GT_EXPR : GE_EXPR), | |
| 389 "DCE_COND_UB", "DCE_COND_UB_TEST", | |
| 390 conds, nconds); | |
| 391 } | |
| 392 } | |
| 393 | |
| 394 | |
| 395 /* A helper function to generate condition | |
| 396 code for the y argument in call pow (some_const, y). | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
397 See candidate selection in check_pow. Since the |
| 0 | 398 candidates' base values have a limited range, |
| 399 the guarded code generated for y are simple: | |
| 400 if (y > max_y) | |
| 401 pow (const, y); | |
| 402 Note max_y can be computed separately for each | |
| 403 const base, but in this implementation, we | |
| 404 choose to compute it using the max base | |
| 405 in the allowed range for the purpose of | |
| 406 simplicity. BASE is the constant base value, | |
| 407 EXPN is the expression for the exponent argument, | |
| 408 *CONDS is the vector to hold resulting statements, | |
| 409 and *NCONDS is the number of logical conditions. */ | |
| 410 | |
| 411 static void | |
| 412 gen_conditions_for_pow_cst_base (tree base, tree expn, | |
| 413 VEC (gimple, heap) *conds, | |
| 414 unsigned *nconds) | |
| 415 { | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
416 inp_domain exp_domain; |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
417 /* Validate the range of the base constant to make |
| 0 | 418 sure it is consistent with check_pow. */ |
| 419 REAL_VALUE_TYPE mv; | |
| 420 REAL_VALUE_TYPE bcv = TREE_REAL_CST (base); | |
| 421 gcc_assert (!REAL_VALUES_EQUAL (bcv, dconst1) | |
| 422 && !REAL_VALUES_LESS (bcv, dconst1)); | |
| 423 real_from_integer (&mv, TYPE_MODE (TREE_TYPE (base)), 256, 0, 1); | |
| 424 gcc_assert (!REAL_VALUES_LESS (mv, bcv)); | |
| 425 | |
| 426 exp_domain = get_domain (0, false, false, | |
| 427 127, true, false); | |
| 428 | |
| 429 gen_conditions_for_domain (expn, exp_domain, | |
| 430 conds, nconds); | |
| 431 } | |
| 432 | |
| 433 /* Generate error condition code for pow calls with | |
| 434 non constant base values. The candidates selected | |
| 435 have their base argument value converted from | |
| 436 integer (see check_pow) value (1, 2, 4 bytes), and | |
| 437 the max exp value is computed based on the size | |
| 438 of the integer type (i.e. max possible base value). | |
| 439 The resulting input domain for exp argument is thus | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
440 conservative (smaller than the max value allowed by |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
441 the runtime value of the base). BASE is the integer |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
442 base value, EXPN is the expression for the exponent |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
443 argument, *CONDS is the vector to hold resulting |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
444 statements, and *NCONDS is the number of logical |
| 0 | 445 conditions. */ |
| 446 | |
| 447 static void | |
| 448 gen_conditions_for_pow_int_base (tree base, tree expn, | |
| 449 VEC (gimple, heap) *conds, | |
| 450 unsigned *nconds) | |
| 451 { | |
| 452 gimple base_def; | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
453 tree base_val0; |
| 0 | 454 tree base_var, int_type; |
| 455 tree temp, tempn; | |
| 456 tree cst0; | |
| 457 gimple stmt1, stmt2; | |
| 458 int bit_sz, max_exp; | |
| 459 inp_domain exp_domain; | |
| 460 | |
| 461 base_def = SSA_NAME_DEF_STMT (base); | |
| 462 base_val0 = gimple_assign_rhs1 (base_def); | |
| 463 base_var = SSA_NAME_VAR (base_val0); | |
| 464 int_type = TREE_TYPE (base_var); | |
| 465 bit_sz = TYPE_PRECISION (int_type); | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
466 gcc_assert (bit_sz > 0 |
| 0 | 467 && bit_sz <= MAX_BASE_INT_BIT_SIZE); |
| 468 | |
| 469 /* Determine the max exp argument value according to | |
| 470 the size of the base integer. The max exp value | |
| 471 is conservatively estimated assuming IEEE754 double | |
| 472 precision format. */ | |
| 473 if (bit_sz == 8) | |
| 474 max_exp = 128; | |
| 475 else if (bit_sz == 16) | |
| 476 max_exp = 64; | |
| 477 else | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
478 { |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
479 gcc_assert (bit_sz == MAX_BASE_INT_BIT_SIZE); |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
480 max_exp = 32; |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
481 } |
| 0 | 482 |
| 483 /* For pow ((double)x, y), generate the following conditions: | |
| 484 cond 1: | |
| 485 temp1 = x; | |
| 486 if (temp1 <= 0) | |
| 487 | |
| 488 cond 2: | |
| 489 temp2 = y; | |
| 490 if (temp2 > max_exp_real_cst) */ | |
| 491 | |
| 492 /* Generate condition in reverse order -- first | |
| 493 the condition for the exp argument. */ | |
| 494 | |
| 495 exp_domain = get_domain (0, false, false, | |
| 496 max_exp, true, true); | |
| 497 | |
| 498 gen_conditions_for_domain (expn, exp_domain, | |
| 499 conds, nconds); | |
| 500 | |
| 501 /* Now generate condition for the base argument. | |
| 502 Note it does not use the helper function | |
| 503 gen_conditions_for_domain because the base | |
| 504 type is integer. */ | |
| 505 | |
| 506 /* Push a separator. */ | |
| 507 VEC_quick_push (gimple, conds, NULL); | |
| 508 | |
| 509 temp = create_tmp_var (int_type, "DCE_COND1"); | |
| 510 cst0 = build_int_cst (int_type, 0); | |
| 511 stmt1 = gimple_build_assign (temp, base_val0); | |
| 512 tempn = make_ssa_name (temp, stmt1); | |
| 513 gimple_assign_set_lhs (stmt1, tempn); | |
| 514 stmt2 = gimple_build_cond (LE_EXPR, tempn, cst0, NULL_TREE, NULL_TREE); | |
| 515 | |
| 516 VEC_quick_push (gimple, conds, stmt1); | |
| 517 VEC_quick_push (gimple, conds, stmt2); | |
| 518 (*nconds)++; | |
| 519 } | |
| 520 | |
| 521 /* Method to generate conditional statements for guarding conditionally | |
| 522 dead calls to pow. One or more statements can be generated for | |
| 523 each logical condition. Statement groups of different conditions | |
| 524 are separated by a NULL tree and they are stored in the VEC | |
| 525 conds. The number of logical conditions are stored in *nconds. | |
| 526 | |
| 527 See C99 standard, 7.12.7.4:2, for description of pow (x, y). | |
| 528 The precise condition for domain errors are complex. In this | |
| 529 implementation, a simplified (but conservative) valid domain | |
| 530 for x and y are used: x is positive to avoid dom errors, while | |
| 531 y is smaller than a upper bound (depending on x) to avoid range | |
| 532 errors. Runtime code is generated to check x (if not constant) | |
| 533 and y against the valid domain. If it is out, jump to the call, | |
| 534 otherwise the call is bypassed. POW_CALL is the call statement, | |
| 535 *CONDS is a vector holding the resulting condition statements, | |
| 536 and *NCONDS is the number of logical conditions. */ | |
| 537 | |
| 538 static void | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
539 gen_conditions_for_pow (gimple pow_call, VEC (gimple, heap) *conds, |
| 0 | 540 unsigned *nconds) |
| 541 { | |
| 542 tree base, expn; | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
543 enum tree_code bc; |
| 0 | 544 |
| 545 #ifdef ENABLE_CHECKING | |
| 546 gcc_assert (check_pow (pow_call)); | |
| 547 #endif | |
| 548 | |
| 549 *nconds = 0; | |
| 550 | |
| 551 base = gimple_call_arg (pow_call, 0); | |
| 552 expn = gimple_call_arg (pow_call, 1); | |
| 553 | |
| 554 bc = TREE_CODE (base); | |
| 555 | |
| 556 if (bc == REAL_CST) | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
557 gen_conditions_for_pow_cst_base (base, expn, conds, nconds); |
| 0 | 558 else if (bc == SSA_NAME) |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
559 gen_conditions_for_pow_int_base (base, expn, conds, nconds); |
| 0 | 560 else |
| 561 gcc_unreachable (); | |
| 562 } | |
| 563 | |
| 564 /* A helper routine to help computing the valid input domain | |
| 565 for a builtin function. See C99 7.12.7 for details. In this | |
| 566 implementation, we only handle single region domain. The | |
| 567 resulting region can be conservative (smaller) than the actual | |
| 568 one and rounded to integers. Some of the bounds are documented | |
| 569 in the standard, while other limit constants are computed | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
570 assuming IEEE floating point format (for SF and DF modes). |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
571 Since IEEE only sets minimum requirements for long double format, |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
572 different long double formats exist under different implementations |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
573 (e.g, 64 bit double precision (DF), 80 bit double-extended |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
574 precision (XF), and 128 bit quad precision (QF) ). For simplicity, |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
575 in this implementation, the computed bounds for long double assume |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
576 64 bit format (DF), and are therefore conservative. Another |
| 0 | 577 assumption is that single precision float type is always SF mode, |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
578 and double type is DF mode. This function is quite |
| 0 | 579 implementation specific, so it may not be suitable to be part of |
| 580 builtins.c. This needs to be revisited later to see if it can | |
| 581 be leveraged in x87 assembly expansion. */ | |
| 582 | |
| 583 static inp_domain | |
| 584 get_no_error_domain (enum built_in_function fnc) | |
| 585 { | |
| 586 switch (fnc) | |
| 587 { | |
| 588 /* Trig functions: return [-1, +1] */ | |
| 589 CASE_FLT_FN (BUILT_IN_ACOS): | |
| 590 CASE_FLT_FN (BUILT_IN_ASIN): | |
| 591 return get_domain (-1, true, true, | |
| 592 1, true, true); | |
| 593 /* Hyperbolic functions. */ | |
| 594 CASE_FLT_FN (BUILT_IN_ACOSH): | |
| 595 /* acosh: [1, +inf) */ | |
| 596 return get_domain (1, true, true, | |
| 597 1, false, false); | |
| 598 CASE_FLT_FN (BUILT_IN_ATANH): | |
| 599 /* atanh: (-1, +1) */ | |
| 600 return get_domain (-1, true, false, | |
| 601 1, true, false); | |
| 602 case BUILT_IN_COSHF: | |
| 603 case BUILT_IN_SINHF: | |
| 604 /* coshf: (-89, +89) */ | |
| 605 return get_domain (-89, true, false, | |
| 606 89, true, false); | |
| 607 case BUILT_IN_COSH: | |
| 608 case BUILT_IN_SINH: | |
| 609 case BUILT_IN_COSHL: | |
| 610 case BUILT_IN_SINHL: | |
| 611 /* cosh: (-710, +710) */ | |
| 612 return get_domain (-710, true, false, | |
| 613 710, true, false); | |
| 614 /* Log functions: (0, +inf) */ | |
| 615 CASE_FLT_FN (BUILT_IN_LOG): | |
| 616 CASE_FLT_FN (BUILT_IN_LOG2): | |
| 617 CASE_FLT_FN (BUILT_IN_LOG10): | |
| 618 return get_domain (0, true, false, | |
| 619 0, false, false); | |
| 620 CASE_FLT_FN (BUILT_IN_LOG1P): | |
| 621 return get_domain (-1, true, false, | |
| 622 0, false, false); | |
| 623 /* Exp functions. */ | |
| 624 case BUILT_IN_EXPF: | |
| 625 case BUILT_IN_EXPM1F: | |
| 626 /* expf: (-inf, 88) */ | |
| 627 return get_domain (-1, false, false, | |
| 628 88, true, false); | |
| 629 case BUILT_IN_EXP: | |
| 630 case BUILT_IN_EXPM1: | |
| 631 case BUILT_IN_EXPL: | |
| 632 case BUILT_IN_EXPM1L: | |
| 633 /* exp: (-inf, 709) */ | |
| 634 return get_domain (-1, false, false, | |
| 635 709, true, false); | |
| 636 case BUILT_IN_EXP2F: | |
| 637 /* exp2f: (-inf, 128) */ | |
| 638 return get_domain (-1, false, false, | |
| 639 128, true, false); | |
| 640 case BUILT_IN_EXP2: | |
| 641 case BUILT_IN_EXP2L: | |
| 642 /* exp2: (-inf, 1024) */ | |
| 643 return get_domain (-1, false, false, | |
| 644 1024, true, false); | |
| 645 case BUILT_IN_EXP10F: | |
| 646 case BUILT_IN_POW10F: | |
| 647 /* exp10f: (-inf, 38) */ | |
| 648 return get_domain (-1, false, false, | |
| 649 38, true, false); | |
| 650 case BUILT_IN_EXP10: | |
| 651 case BUILT_IN_POW10: | |
| 652 case BUILT_IN_EXP10L: | |
| 653 case BUILT_IN_POW10L: | |
| 654 /* exp10: (-inf, 308) */ | |
| 655 return get_domain (-1, false, false, | |
| 656 308, true, false); | |
| 657 /* sqrt: [0, +inf) */ | |
| 658 CASE_FLT_FN (BUILT_IN_SQRT): | |
| 659 return get_domain (0, true, true, | |
| 660 0, false, false); | |
| 661 default: | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
662 gcc_unreachable (); |
| 0 | 663 } |
| 664 | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
665 gcc_unreachable (); |
| 0 | 666 } |
| 667 | |
| 668 /* The function to generate shrink wrap conditions for a partially | |
| 669 dead builtin call whose return value is not used anywhere, | |
| 670 but has to be kept live due to potential error condition. | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
671 BI_CALL is the builtin call, CONDS is the vector of statements |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
672 for condition code, NCODES is the pointer to the number of |
| 0 | 673 logical conditions. Statements belonging to different logical |
| 674 condition are separated by NULL tree in the vector. */ | |
| 675 | |
| 676 static void | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
677 gen_shrink_wrap_conditions (gimple bi_call, VEC (gimple, heap) *conds, |
| 0 | 678 unsigned int *nconds) |
| 679 { | |
| 680 gimple call; | |
| 681 tree fn; | |
| 682 enum built_in_function fnc; | |
| 683 | |
| 684 gcc_assert (nconds && conds); | |
| 685 gcc_assert (VEC_length (gimple, conds) == 0); | |
| 686 gcc_assert (is_gimple_call (bi_call)); | |
| 687 | |
| 688 call = bi_call; | |
| 689 fn = gimple_call_fndecl (call); | |
| 690 gcc_assert (fn && DECL_BUILT_IN (fn)); | |
| 691 fnc = DECL_FUNCTION_CODE (fn); | |
| 692 *nconds = 0; | |
| 693 | |
| 694 if (fnc == BUILT_IN_POW) | |
| 695 gen_conditions_for_pow (call, conds, nconds); | |
| 696 else | |
| 697 { | |
| 698 tree arg; | |
| 699 inp_domain domain = get_no_error_domain (fnc); | |
| 700 *nconds = 0; | |
| 701 arg = gimple_call_arg (bi_call, 0); | |
| 702 gen_conditions_for_domain (arg, domain, conds, nconds); | |
| 703 } | |
| 704 | |
| 705 return; | |
| 706 } | |
| 707 | |
| 708 | |
| 709 /* Probability of the branch (to the call) is taken. */ | |
| 710 #define ERR_PROB 0.01 | |
| 711 | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
712 /* The function to shrink wrap a partially dead builtin call |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
713 whose return value is not used anywhere, but has to be kept |
| 0 | 714 live due to potential error condition. Returns true if the |
| 715 transformation actually happens. */ | |
| 716 | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
717 static bool |
| 0 | 718 shrink_wrap_one_built_in_call (gimple bi_call) |
| 719 { | |
| 720 gimple_stmt_iterator bi_call_bsi; | |
| 721 basic_block bi_call_bb, join_tgt_bb, guard_bb, guard_bb0; | |
| 722 edge join_tgt_in_edge_from_call, join_tgt_in_edge_fall_thru; | |
| 723 edge bi_call_in_edge0, guard_bb_in_edge; | |
| 724 VEC (gimple, heap) *conds; | |
| 725 unsigned tn_cond_stmts, nconds; | |
| 726 unsigned ci; | |
| 727 gimple cond_expr = NULL; | |
| 728 gimple cond_expr_start; | |
| 729 tree bi_call_label_decl; | |
| 730 gimple bi_call_label; | |
| 731 | |
| 732 conds = VEC_alloc (gimple, heap, 12); | |
| 733 gen_shrink_wrap_conditions (bi_call, conds, &nconds); | |
| 734 | |
| 735 /* This can happen if the condition generator decides | |
| 736 it is not beneficial to do the transformation. Just | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
737 return false and do not do any transformation for |
| 0 | 738 the call. */ |
| 739 if (nconds == 0) | |
| 740 return false; | |
| 741 | |
| 742 bi_call_bb = gimple_bb (bi_call); | |
| 743 | |
| 744 /* Now find the join target bb -- split | |
| 745 bi_call_bb if needed. */ | |
| 746 bi_call_bsi = gsi_for_stmt (bi_call); | |
| 747 | |
| 748 join_tgt_in_edge_from_call = split_block (bi_call_bb, bi_call); | |
| 749 bi_call_bsi = gsi_for_stmt (bi_call); | |
| 750 | |
| 751 join_tgt_bb = join_tgt_in_edge_from_call->dest; | |
| 752 | |
| 753 /* Now it is time to insert the first conditional expression | |
| 754 into bi_call_bb and split this bb so that bi_call is | |
| 755 shrink-wrapped. */ | |
| 756 tn_cond_stmts = VEC_length (gimple, conds); | |
| 757 cond_expr = NULL; | |
| 758 cond_expr_start = VEC_index (gimple, conds, 0); | |
| 759 for (ci = 0; ci < tn_cond_stmts; ci++) | |
| 760 { | |
| 761 gimple c = VEC_index (gimple, conds, ci); | |
| 762 gcc_assert (c || ci != 0); | |
| 763 if (!c) | |
| 764 break; | |
| 765 gsi_insert_before (&bi_call_bsi, c, GSI_SAME_STMT); | |
| 766 cond_expr = c; | |
| 767 } | |
| 768 nconds--; | |
| 769 ci++; | |
| 770 gcc_assert (cond_expr && gimple_code (cond_expr) == GIMPLE_COND); | |
| 771 | |
| 772 /* Now the label. */ | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
773 bi_call_label_decl = create_artificial_label (gimple_location (bi_call)); |
| 0 | 774 bi_call_label = gimple_build_label (bi_call_label_decl); |
| 775 gsi_insert_before (&bi_call_bsi, bi_call_label, GSI_SAME_STMT); | |
| 776 | |
| 777 bi_call_in_edge0 = split_block (bi_call_bb, cond_expr); | |
| 778 bi_call_in_edge0->flags &= ~EDGE_FALLTHRU; | |
| 779 bi_call_in_edge0->flags |= EDGE_TRUE_VALUE; | |
| 780 guard_bb0 = bi_call_bb; | |
| 781 bi_call_bb = bi_call_in_edge0->dest; | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
782 join_tgt_in_edge_fall_thru = make_edge (guard_bb0, join_tgt_bb, |
| 0 | 783 EDGE_FALSE_VALUE); |
| 784 | |
| 785 bi_call_in_edge0->probability = REG_BR_PROB_BASE * ERR_PROB; | |
| 786 join_tgt_in_edge_fall_thru->probability = | |
| 787 REG_BR_PROB_BASE - bi_call_in_edge0->probability; | |
| 788 | |
| 789 /* Code generation for the rest of the conditions */ | |
| 790 guard_bb = guard_bb0; | |
| 791 while (nconds > 0) | |
| 792 { | |
| 793 unsigned ci0; | |
| 794 edge bi_call_in_edge; | |
| 795 gimple_stmt_iterator guard_bsi = gsi_for_stmt (cond_expr_start); | |
| 796 ci0 = ci; | |
| 797 cond_expr_start = VEC_index (gimple, conds, ci0); | |
| 798 for (; ci < tn_cond_stmts; ci++) | |
| 799 { | |
| 800 gimple c = VEC_index (gimple, conds, ci); | |
| 801 gcc_assert (c || ci != ci0); | |
| 802 if (!c) | |
| 803 break; | |
| 804 gsi_insert_before (&guard_bsi, c, GSI_SAME_STMT); | |
| 805 cond_expr = c; | |
| 806 } | |
| 807 nconds--; | |
| 808 ci++; | |
| 809 gcc_assert (cond_expr && gimple_code (cond_expr) == GIMPLE_COND); | |
| 810 guard_bb_in_edge = split_block (guard_bb, cond_expr); | |
| 811 guard_bb_in_edge->flags &= ~EDGE_FALLTHRU; | |
| 812 guard_bb_in_edge->flags |= EDGE_FALSE_VALUE; | |
| 813 | |
| 814 bi_call_in_edge = make_edge (guard_bb, bi_call_bb, EDGE_TRUE_VALUE); | |
| 815 | |
| 816 bi_call_in_edge->probability = REG_BR_PROB_BASE * ERR_PROB; | |
| 817 guard_bb_in_edge->probability = | |
| 818 REG_BR_PROB_BASE - bi_call_in_edge->probability; | |
| 819 } | |
| 820 | |
| 821 VEC_free (gimple, heap, conds); | |
| 822 if (dump_file && (dump_flags & TDF_DETAILS)) | |
| 823 { | |
| 824 location_t loc; | |
| 825 loc = gimple_location (bi_call); | |
| 826 fprintf (dump_file, | |
| 827 "%s:%d: note: function call is shrink-wrapped" | |
| 828 " into error conditions.\n", | |
| 829 LOCATION_FILE (loc), LOCATION_LINE (loc)); | |
| 830 } | |
| 831 | |
| 832 return true; | |
| 833 } | |
| 834 | |
| 835 /* The top level function for conditional dead code shrink | |
| 836 wrapping transformation. */ | |
| 837 | |
| 838 static bool | |
| 839 shrink_wrap_conditional_dead_built_in_calls (VEC (gimple, heap) *calls) | |
| 840 { | |
| 841 bool changed = false; | |
| 842 unsigned i = 0; | |
| 843 | |
| 844 unsigned n = VEC_length (gimple, calls); | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
845 if (n == 0) |
| 0 | 846 return false; |
| 847 | |
| 848 for (; i < n ; i++) | |
| 849 { | |
| 850 gimple bi_call = VEC_index (gimple, calls, i); | |
| 851 changed |= shrink_wrap_one_built_in_call (bi_call); | |
| 852 } | |
| 853 | |
| 854 return changed; | |
| 855 } | |
| 856 | |
| 857 /* Pass entry points. */ | |
| 858 | |
| 859 static unsigned int | |
| 860 tree_call_cdce (void) | |
| 861 { | |
| 862 basic_block bb; | |
| 863 gimple_stmt_iterator i; | |
| 864 bool something_changed = false; | |
| 865 VEC (gimple, heap) *cond_dead_built_in_calls = NULL; | |
| 866 FOR_EACH_BB (bb) | |
| 867 { | |
| 868 /* Collect dead call candidates. */ | |
| 869 for (i = gsi_start_bb (bb); !gsi_end_p (i); gsi_next (&i)) | |
| 870 { | |
| 871 gimple stmt = gsi_stmt (i); | |
| 872 if (is_gimple_call (stmt) | |
| 873 && is_call_dce_candidate (stmt)) | |
| 874 { | |
| 875 if (dump_file && (dump_flags & TDF_DETAILS)) | |
| 876 { | |
| 877 fprintf (dump_file, "Found conditional dead call: "); | |
| 878 print_gimple_stmt (dump_file, stmt, 0, TDF_SLIM); | |
| 879 fprintf (dump_file, "\n"); | |
| 880 } | |
| 881 if (cond_dead_built_in_calls == NULL) | |
| 882 cond_dead_built_in_calls = VEC_alloc (gimple, heap, 64); | |
| 883 VEC_safe_push (gimple, heap, cond_dead_built_in_calls, stmt); | |
| 884 } | |
| 885 } | |
| 886 } | |
| 887 | |
| 888 if (cond_dead_built_in_calls == NULL) | |
| 889 return 0; | |
| 890 | |
| 891 something_changed | |
| 892 = shrink_wrap_conditional_dead_built_in_calls (cond_dead_built_in_calls); | |
| 893 | |
| 894 VEC_free (gimple, heap, cond_dead_built_in_calls); | |
| 895 | |
| 896 if (something_changed) | |
| 897 { | |
| 898 free_dominance_info (CDI_DOMINATORS); | |
| 899 free_dominance_info (CDI_POST_DOMINATORS); | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
900 /* As we introduced new control-flow we need to insert PHI-nodes |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
901 for the call-clobbers of the remaining call. */ |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
902 mark_sym_for_renaming (gimple_vop (cfun)); |
|
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
903 return (TODO_update_ssa | TODO_cleanup_cfg | TODO_ggc_collect |
| 0 | 904 | TODO_remove_unused_locals); |
| 905 } | |
| 906 else | |
| 907 return 0; | |
| 908 } | |
| 909 | |
| 910 static bool | |
| 911 gate_call_cdce (void) | |
| 912 { | |
| 913 /* The limit constants used in the implementation | |
| 914 assume IEEE floating point format. Other formats | |
| 915 can be supported in the future if needed. */ | |
|
55
77e2b8dfacca
update it from 4.4.3 to 4.5.0
ryoma <e075725@ie.u-ryukyu.ac.jp>
parents:
0
diff
changeset
|
916 return flag_tree_builtin_call_dce != 0 && optimize_function_for_speed_p (cfun); |
| 0 | 917 } |
| 918 | |
| 919 struct gimple_opt_pass pass_call_cdce = | |
| 920 { | |
| 921 { | |
| 922 GIMPLE_PASS, | |
| 923 "cdce", /* name */ | |
| 924 gate_call_cdce, /* gate */ | |
| 925 tree_call_cdce, /* execute */ | |
| 926 NULL, /* sub */ | |
| 927 NULL, /* next */ | |
| 928 0, /* static_pass_number */ | |
| 929 TV_TREE_CALL_CDCE, /* tv_id */ | |
| 930 PROP_cfg | PROP_ssa, /* properties_required */ | |
| 931 0, /* properties_provided */ | |
| 932 0, /* properties_destroyed */ | |
| 933 0, /* todo_flags_start */ | |
| 934 TODO_dump_func | TODO_verify_ssa /* todo_flags_finish */ | |
| 935 } | |
| 936 }; |