]>
Commit | Line | Data |
---|---|---|
e9eb809d | 1 | /* Scalar evolution detector. |
66647d44 JJ |
2 | Copyright (C) 2003, 2004, 2005, 2006, 2007, 2008, 2009 |
3 | Free Software Foundation, Inc. | |
e9eb809d ZD |
4 | Contributed by Sebastian Pop <s.pop@laposte.net> |
5 | ||
6 | This file is part of GCC. | |
7 | ||
8 | GCC is free software; you can redistribute it and/or modify it under | |
9 | the terms of the GNU General Public License as published by the Free | |
9dcd6f09 | 10 | Software Foundation; either version 3, or (at your option) any later |
e9eb809d ZD |
11 | version. |
12 | ||
13 | GCC is distributed in the hope that it will be useful, but WITHOUT ANY | |
14 | 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. | |
17 | ||
18 | You should have received a copy of the GNU General Public License | |
9dcd6f09 NC |
19 | along with GCC; see the file COPYING3. If not see |
20 | <http://www.gnu.org/licenses/>. */ | |
e9eb809d | 21 | |
b8698a0f L |
22 | /* |
23 | Description: | |
24 | ||
9baba81b SP |
25 | This pass analyzes the evolution of scalar variables in loop |
26 | structures. The algorithm is based on the SSA representation, | |
27 | and on the loop hierarchy tree. This algorithm is not based on | |
28 | the notion of versions of a variable, as it was the case for the | |
29 | previous implementations of the scalar evolution algorithm, but | |
30 | it assumes that each defined name is unique. | |
31 | ||
32 | The notation used in this file is called "chains of recurrences", | |
33 | and has been proposed by Eugene Zima, Robert Van Engelen, and | |
34 | others for describing induction variables in programs. For example | |
35 | "b -> {0, +, 2}_1" means that the scalar variable "b" is equal to 0 | |
36 | when entering in the loop_1 and has a step 2 in this loop, in other | |
37 | words "for (b = 0; b < N; b+=2);". Note that the coefficients of | |
38 | this chain of recurrence (or chrec [shrek]) can contain the name of | |
39 | other variables, in which case they are called parametric chrecs. | |
40 | For example, "b -> {a, +, 2}_1" means that the initial value of "b" | |
41 | is the value of "a". In most of the cases these parametric chrecs | |
42 | are fully instantiated before their use because symbolic names can | |
43 | hide some difficult cases such as self-references described later | |
44 | (see the Fibonacci example). | |
b8698a0f | 45 | |
9baba81b | 46 | A short sketch of the algorithm is: |
b8698a0f | 47 | |
9baba81b SP |
48 | Given a scalar variable to be analyzed, follow the SSA edge to |
49 | its definition: | |
b8698a0f | 50 | |
726a989a | 51 | - When the definition is a GIMPLE_ASSIGN: if the right hand side |
9baba81b | 52 | (RHS) of the definition cannot be statically analyzed, the answer |
b8698a0f | 53 | of the analyzer is: "don't know". |
9baba81b SP |
54 | Otherwise, for all the variables that are not yet analyzed in the |
55 | RHS, try to determine their evolution, and finally try to | |
56 | evaluate the operation of the RHS that gives the evolution | |
57 | function of the analyzed variable. | |
58 | ||
59 | - When the definition is a condition-phi-node: determine the | |
60 | evolution function for all the branches of the phi node, and | |
61 | finally merge these evolutions (see chrec_merge). | |
62 | ||
63 | - When the definition is a loop-phi-node: determine its initial | |
64 | condition, that is the SSA edge defined in an outer loop, and | |
65 | keep it symbolic. Then determine the SSA edges that are defined | |
66 | in the body of the loop. Follow the inner edges until ending on | |
67 | another loop-phi-node of the same analyzed loop. If the reached | |
68 | loop-phi-node is not the starting loop-phi-node, then we keep | |
69 | this definition under a symbolic form. If the reached | |
70 | loop-phi-node is the same as the starting one, then we compute a | |
71 | symbolic stride on the return path. The result is then the | |
72 | symbolic chrec {initial_condition, +, symbolic_stride}_loop. | |
73 | ||
74 | Examples: | |
b8698a0f | 75 | |
9baba81b | 76 | Example 1: Illustration of the basic algorithm. |
b8698a0f | 77 | |
9baba81b SP |
78 | | a = 3 |
79 | | loop_1 | |
80 | | b = phi (a, c) | |
81 | | c = b + 1 | |
82 | | if (c > 10) exit_loop | |
83 | | endloop | |
b8698a0f | 84 | |
9baba81b SP |
85 | Suppose that we want to know the number of iterations of the |
86 | loop_1. The exit_loop is controlled by a COND_EXPR (c > 10). We | |
87 | ask the scalar evolution analyzer two questions: what's the | |
88 | scalar evolution (scev) of "c", and what's the scev of "10". For | |
89 | "10" the answer is "10" since it is a scalar constant. For the | |
90 | scalar variable "c", it follows the SSA edge to its definition, | |
91 | "c = b + 1", and then asks again what's the scev of "b". | |
92 | Following the SSA edge, we end on a loop-phi-node "b = phi (a, | |
93 | c)", where the initial condition is "a", and the inner loop edge | |
94 | is "c". The initial condition is kept under a symbolic form (it | |
95 | may be the case that the copy constant propagation has done its | |
96 | work and we end with the constant "3" as one of the edges of the | |
97 | loop-phi-node). The update edge is followed to the end of the | |
98 | loop, and until reaching again the starting loop-phi-node: b -> c | |
99 | -> b. At this point we have drawn a path from "b" to "b" from | |
100 | which we compute the stride in the loop: in this example it is | |
101 | "+1". The resulting scev for "b" is "b -> {a, +, 1}_1". Now | |
102 | that the scev for "b" is known, it is possible to compute the | |
103 | scev for "c", that is "c -> {a + 1, +, 1}_1". In order to | |
104 | determine the number of iterations in the loop_1, we have to | |
3f227a8c | 105 | instantiate_parameters (loop_1, {a + 1, +, 1}_1), that gives after some |
9baba81b SP |
106 | more analysis the scev {4, +, 1}_1, or in other words, this is |
107 | the function "f (x) = x + 4", where x is the iteration count of | |
108 | the loop_1. Now we have to solve the inequality "x + 4 > 10", | |
109 | and take the smallest iteration number for which the loop is | |
110 | exited: x = 7. This loop runs from x = 0 to x = 7, and in total | |
111 | there are 8 iterations. In terms of loop normalization, we have | |
112 | created a variable that is implicitly defined, "x" or just "_1", | |
113 | and all the other analyzed scalars of the loop are defined in | |
114 | function of this variable: | |
b8698a0f | 115 | |
9baba81b SP |
116 | a -> 3 |
117 | b -> {3, +, 1}_1 | |
118 | c -> {4, +, 1}_1 | |
b8698a0f L |
119 | |
120 | or in terms of a C program: | |
121 | ||
9baba81b SP |
122 | | a = 3 |
123 | | for (x = 0; x <= 7; x++) | |
124 | | { | |
125 | | b = x + 3 | |
126 | | c = x + 4 | |
127 | | } | |
b8698a0f | 128 | |
3f227a8c | 129 | Example 2a: Illustration of the algorithm on nested loops. |
b8698a0f | 130 | |
9baba81b SP |
131 | | loop_1 |
132 | | a = phi (1, b) | |
133 | | c = a + 2 | |
134 | | loop_2 10 times | |
135 | | b = phi (c, d) | |
136 | | d = b + 3 | |
137 | | endloop | |
138 | | endloop | |
b8698a0f | 139 | |
9baba81b SP |
140 | For analyzing the scalar evolution of "a", the algorithm follows |
141 | the SSA edge into the loop's body: "a -> b". "b" is an inner | |
b8698a0f L |
142 | loop-phi-node, and its analysis as in Example 1, gives: |
143 | ||
9baba81b SP |
144 | b -> {c, +, 3}_2 |
145 | d -> {c + 3, +, 3}_2 | |
b8698a0f | 146 | |
9baba81b SP |
147 | Following the SSA edge for the initial condition, we end on "c = a |
148 | + 2", and then on the starting loop-phi-node "a". From this point, | |
149 | the loop stride is computed: back on "c = a + 2" we get a "+2" in | |
150 | the loop_1, then on the loop-phi-node "b" we compute the overall | |
151 | effect of the inner loop that is "b = c + 30", and we get a "+30" | |
152 | in the loop_1. That means that the overall stride in loop_1 is | |
b8698a0f L |
153 | equal to "+32", and the result is: |
154 | ||
9baba81b SP |
155 | a -> {1, +, 32}_1 |
156 | c -> {3, +, 32}_1 | |
3f227a8c JS |
157 | |
158 | Example 2b: Multivariate chains of recurrences. | |
159 | ||
160 | | loop_1 | |
161 | | k = phi (0, k + 1) | |
162 | | loop_2 4 times | |
163 | | j = phi (0, j + 1) | |
164 | | loop_3 4 times | |
165 | | i = phi (0, i + 1) | |
166 | | A[j + k] = ... | |
167 | | endloop | |
168 | | endloop | |
169 | | endloop | |
170 | ||
171 | Analyzing the access function of array A with | |
172 | instantiate_parameters (loop_1, "j + k"), we obtain the | |
173 | instantiation and the analysis of the scalar variables "j" and "k" | |
174 | in loop_1. This leads to the scalar evolution {4, +, 1}_1: the end | |
175 | value of loop_2 for "j" is 4, and the evolution of "k" in loop_1 is | |
176 | {0, +, 1}_1. To obtain the evolution function in loop_3 and | |
177 | instantiate the scalar variables up to loop_1, one has to use: | |
a213b219 SP |
178 | instantiate_scev (block_before_loop (loop_1), loop_3, "j + k"). |
179 | The result of this call is {{0, +, 1}_1, +, 1}_2. | |
3f227a8c | 180 | |
9baba81b | 181 | Example 3: Higher degree polynomials. |
b8698a0f | 182 | |
9baba81b SP |
183 | | loop_1 |
184 | | a = phi (2, b) | |
185 | | c = phi (5, d) | |
186 | | b = a + 1 | |
187 | | d = c + a | |
188 | | endloop | |
b8698a0f | 189 | |
9baba81b SP |
190 | a -> {2, +, 1}_1 |
191 | b -> {3, +, 1}_1 | |
192 | c -> {5, +, a}_1 | |
193 | d -> {5 + a, +, a}_1 | |
b8698a0f | 194 | |
3f227a8c JS |
195 | instantiate_parameters (loop_1, {5, +, a}_1) -> {5, +, 2, +, 1}_1 |
196 | instantiate_parameters (loop_1, {5 + a, +, a}_1) -> {7, +, 3, +, 1}_1 | |
b8698a0f | 197 | |
9baba81b | 198 | Example 4: Lucas, Fibonacci, or mixers in general. |
b8698a0f | 199 | |
9baba81b SP |
200 | | loop_1 |
201 | | a = phi (1, b) | |
202 | | c = phi (3, d) | |
203 | | b = c | |
204 | | d = c + a | |
205 | | endloop | |
b8698a0f | 206 | |
9baba81b SP |
207 | a -> (1, c)_1 |
208 | c -> {3, +, a}_1 | |
b8698a0f | 209 | |
9baba81b SP |
210 | The syntax "(1, c)_1" stands for a PEELED_CHREC that has the |
211 | following semantics: during the first iteration of the loop_1, the | |
212 | variable contains the value 1, and then it contains the value "c". | |
213 | Note that this syntax is close to the syntax of the loop-phi-node: | |
214 | "a -> (1, c)_1" vs. "a = phi (1, c)". | |
b8698a0f | 215 | |
9baba81b SP |
216 | The symbolic chrec representation contains all the semantics of the |
217 | original code. What is more difficult is to use this information. | |
b8698a0f | 218 | |
9baba81b | 219 | Example 5: Flip-flops, or exchangers. |
b8698a0f | 220 | |
9baba81b SP |
221 | | loop_1 |
222 | | a = phi (1, b) | |
223 | | c = phi (3, d) | |
224 | | b = c | |
225 | | d = a | |
226 | | endloop | |
b8698a0f | 227 | |
9baba81b SP |
228 | a -> (1, c)_1 |
229 | c -> (3, a)_1 | |
b8698a0f | 230 | |
9baba81b | 231 | Based on these symbolic chrecs, it is possible to refine this |
b8698a0f L |
232 | information into the more precise PERIODIC_CHRECs: |
233 | ||
9baba81b SP |
234 | a -> |1, 3|_1 |
235 | c -> |3, 1|_1 | |
b8698a0f | 236 | |
9baba81b | 237 | This transformation is not yet implemented. |
b8698a0f | 238 | |
9baba81b | 239 | Further readings: |
b8698a0f | 240 | |
9baba81b SP |
241 | You can find a more detailed description of the algorithm in: |
242 | http://icps.u-strasbg.fr/~pop/DEA_03_Pop.pdf | |
243 | http://icps.u-strasbg.fr/~pop/DEA_03_Pop.ps.gz. But note that | |
244 | this is a preliminary report and some of the details of the | |
245 | algorithm have changed. I'm working on a research report that | |
246 | updates the description of the algorithms to reflect the design | |
247 | choices used in this implementation. | |
b8698a0f | 248 | |
9baba81b SP |
249 | A set of slides show a high level overview of the algorithm and run |
250 | an example through the scalar evolution analyzer: | |
251 | http://cri.ensmp.fr/~pop/gcc/mar04/slides.pdf | |
252 | ||
253 | The slides that I have presented at the GCC Summit'04 are available | |
254 | at: http://cri.ensmp.fr/~pop/gcc/20040604/gccsummit-lno-spop.pdf | |
255 | */ | |
256 | ||
e9eb809d ZD |
257 | #include "config.h" |
258 | #include "system.h" | |
259 | #include "coretypes.h" | |
260 | #include "tm.h" | |
e9eb809d ZD |
261 | #include "ggc.h" |
262 | #include "tree.h" | |
9d2b0e12 | 263 | #include "real.h" |
9baba81b SP |
264 | |
265 | /* These RTL headers are needed for basic-block.h. */ | |
e9eb809d ZD |
266 | #include "rtl.h" |
267 | #include "basic-block.h" | |
268 | #include "diagnostic.h" | |
269 | #include "tree-flow.h" | |
270 | #include "tree-dump.h" | |
271 | #include "timevar.h" | |
272 | #include "cfgloop.h" | |
273 | #include "tree-chrec.h" | |
274 | #include "tree-scalar-evolution.h" | |
9baba81b SP |
275 | #include "tree-pass.h" |
276 | #include "flags.h" | |
c59dabbe | 277 | #include "params.h" |
9baba81b SP |
278 | |
279 | static tree analyze_scalar_evolution_1 (struct loop *, tree, tree); | |
9baba81b | 280 | |
a213b219 SP |
281 | /* The cached information about an SSA name VAR, claiming that below |
282 | basic block INSTANTIATED_BELOW, the value of VAR can be expressed | |
283 | as CHREC. */ | |
9baba81b | 284 | |
d1b38208 | 285 | struct GTY(()) scev_info_str { |
a213b219 | 286 | basic_block instantiated_below; |
9baba81b SP |
287 | tree var; |
288 | tree chrec; | |
289 | }; | |
290 | ||
291 | /* Counters for the scev database. */ | |
292 | static unsigned nb_set_scev = 0; | |
293 | static unsigned nb_get_scev = 0; | |
294 | ||
295 | /* The following trees are unique elements. Thus the comparison of | |
296 | another element to these elements should be done on the pointer to | |
297 | these trees, and not on their value. */ | |
298 | ||
299 | /* The SSA_NAMEs that are not yet analyzed are qualified with NULL_TREE. */ | |
300 | tree chrec_not_analyzed_yet; | |
301 | ||
302 | /* Reserved to the cases where the analyzer has detected an | |
303 | undecidable property at compile time. */ | |
304 | tree chrec_dont_know; | |
305 | ||
306 | /* When the analyzer has detected that a property will never | |
307 | happen, then it qualifies it with chrec_known. */ | |
308 | tree chrec_known; | |
309 | ||
9e2f83a5 | 310 | static GTY ((param_is (struct scev_info_str))) htab_t scalar_evolution_info; |
9baba81b SP |
311 | |
312 | \f | |
a213b219 | 313 | /* Constructs a new SCEV_INFO_STR structure for VAR and INSTANTIATED_BELOW. */ |
9baba81b SP |
314 | |
315 | static inline struct scev_info_str * | |
a213b219 | 316 | new_scev_info_str (basic_block instantiated_below, tree var) |
9baba81b SP |
317 | { |
318 | struct scev_info_str *res; | |
b8698a0f | 319 | |
9e2f83a5 | 320 | res = GGC_NEW (struct scev_info_str); |
9baba81b SP |
321 | res->var = var; |
322 | res->chrec = chrec_not_analyzed_yet; | |
a213b219 SP |
323 | res->instantiated_below = instantiated_below; |
324 | ||
9baba81b SP |
325 | return res; |
326 | } | |
327 | ||
328 | /* Computes a hash function for database element ELT. */ | |
329 | ||
330 | static hashval_t | |
331 | hash_scev_info (const void *elt) | |
332 | { | |
741ac903 | 333 | return SSA_NAME_VERSION (((const struct scev_info_str *) elt)->var); |
9baba81b SP |
334 | } |
335 | ||
336 | /* Compares database elements E1 and E2. */ | |
337 | ||
338 | static int | |
339 | eq_scev_info (const void *e1, const void *e2) | |
340 | { | |
cceb1885 GDR |
341 | const struct scev_info_str *elt1 = (const struct scev_info_str *) e1; |
342 | const struct scev_info_str *elt2 = (const struct scev_info_str *) e2; | |
9baba81b | 343 | |
a213b219 SP |
344 | return (elt1->var == elt2->var |
345 | && elt1->instantiated_below == elt2->instantiated_below); | |
9baba81b SP |
346 | } |
347 | ||
348 | /* Deletes database element E. */ | |
349 | ||
350 | static void | |
351 | del_scev_info (void *e) | |
352 | { | |
9e2f83a5 | 353 | ggc_free (e); |
9baba81b SP |
354 | } |
355 | ||
a213b219 SP |
356 | /* Get the scalar evolution of VAR for INSTANTIATED_BELOW basic block. |
357 | A first query on VAR returns chrec_not_analyzed_yet. */ | |
9baba81b SP |
358 | |
359 | static tree * | |
a213b219 | 360 | find_var_scev_info (basic_block instantiated_below, tree var) |
9baba81b SP |
361 | { |
362 | struct scev_info_str *res; | |
363 | struct scev_info_str tmp; | |
364 | PTR *slot; | |
365 | ||
366 | tmp.var = var; | |
a213b219 | 367 | tmp.instantiated_below = instantiated_below; |
9baba81b SP |
368 | slot = htab_find_slot (scalar_evolution_info, &tmp, INSERT); |
369 | ||
370 | if (!*slot) | |
a213b219 | 371 | *slot = new_scev_info_str (instantiated_below, var); |
cceb1885 | 372 | res = (struct scev_info_str *) *slot; |
9baba81b SP |
373 | |
374 | return &res->chrec; | |
375 | } | |
376 | ||
9baba81b SP |
377 | /* Return true when CHREC contains symbolic names defined in |
378 | LOOP_NB. */ | |
379 | ||
b8698a0f | 380 | bool |
ed7a4b4b | 381 | chrec_contains_symbols_defined_in_loop (const_tree chrec, unsigned loop_nb) |
9baba81b | 382 | { |
5039610b SL |
383 | int i, n; |
384 | ||
9baba81b SP |
385 | if (chrec == NULL_TREE) |
386 | return false; | |
387 | ||
ad6003f2 | 388 | if (is_gimple_min_invariant (chrec)) |
9baba81b SP |
389 | return false; |
390 | ||
391 | if (TREE_CODE (chrec) == VAR_DECL | |
392 | || TREE_CODE (chrec) == PARM_DECL | |
393 | || TREE_CODE (chrec) == FUNCTION_DECL | |
394 | || TREE_CODE (chrec) == LABEL_DECL | |
395 | || TREE_CODE (chrec) == RESULT_DECL | |
396 | || TREE_CODE (chrec) == FIELD_DECL) | |
397 | return true; | |
398 | ||
399 | if (TREE_CODE (chrec) == SSA_NAME) | |
400 | { | |
726a989a | 401 | gimple def = SSA_NAME_DEF_STMT (chrec); |
9baba81b | 402 | struct loop *def_loop = loop_containing_stmt (def); |
42fd6772 | 403 | struct loop *loop = get_loop (loop_nb); |
9baba81b SP |
404 | |
405 | if (def_loop == NULL) | |
406 | return false; | |
407 | ||
408 | if (loop == def_loop || flow_loop_nested_p (loop, def_loop)) | |
409 | return true; | |
410 | ||
411 | return false; | |
412 | } | |
413 | ||
5039610b SL |
414 | n = TREE_OPERAND_LENGTH (chrec); |
415 | for (i = 0; i < n; i++) | |
b8698a0f | 416 | if (chrec_contains_symbols_defined_in_loop (TREE_OPERAND (chrec, i), |
5039610b SL |
417 | loop_nb)) |
418 | return true; | |
419 | return false; | |
9baba81b SP |
420 | } |
421 | ||
422 | /* Return true when PHI is a loop-phi-node. */ | |
423 | ||
424 | static bool | |
726a989a | 425 | loop_phi_node_p (gimple phi) |
9baba81b SP |
426 | { |
427 | /* The implementation of this function is based on the following | |
428 | property: "all the loop-phi-nodes of a loop are contained in the | |
429 | loop's header basic block". */ | |
430 | ||
726a989a | 431 | return loop_containing_stmt (phi)->header == gimple_bb (phi); |
9baba81b SP |
432 | } |
433 | ||
434 | /* Compute the scalar evolution for EVOLUTION_FN after crossing LOOP. | |
435 | In general, in the case of multivariate evolutions we want to get | |
436 | the evolution in different loops. LOOP specifies the level for | |
437 | which to get the evolution. | |
b8698a0f | 438 | |
9baba81b | 439 | Example: |
b8698a0f | 440 | |
9baba81b SP |
441 | | for (j = 0; j < 100; j++) |
442 | | { | |
443 | | for (k = 0; k < 100; k++) | |
444 | | { | |
b8698a0f | 445 | | i = k + j; - Here the value of i is a function of j, k. |
9baba81b | 446 | | } |
b8698a0f | 447 | | ... = i - Here the value of i is a function of j. |
9baba81b | 448 | | } |
b8698a0f L |
449 | | ... = i - Here the value of i is a scalar. |
450 | ||
451 | Example: | |
452 | ||
9baba81b SP |
453 | | i_0 = ... |
454 | | loop_1 10 times | |
455 | | i_1 = phi (i_0, i_2) | |
456 | | i_2 = i_1 + 2 | |
457 | | endloop | |
b8698a0f | 458 | |
9baba81b SP |
459 | This loop has the same effect as: |
460 | LOOP_1 has the same effect as: | |
b8698a0f | 461 | |
9baba81b | 462 | | i_1 = i_0 + 20 |
b8698a0f L |
463 | |
464 | The overall effect of the loop, "i_0 + 20" in the previous example, | |
465 | is obtained by passing in the parameters: LOOP = 1, | |
9baba81b SP |
466 | EVOLUTION_FN = {i_0, +, 2}_1. |
467 | */ | |
b8698a0f | 468 | |
42e6eec5 | 469 | tree |
9baba81b SP |
470 | compute_overall_effect_of_inner_loop (struct loop *loop, tree evolution_fn) |
471 | { | |
472 | bool val = false; | |
473 | ||
474 | if (evolution_fn == chrec_dont_know) | |
475 | return chrec_dont_know; | |
476 | ||
477 | else if (TREE_CODE (evolution_fn) == POLYNOMIAL_CHREC) | |
478 | { | |
677cc14d ZD |
479 | struct loop *inner_loop = get_chrec_loop (evolution_fn); |
480 | ||
481 | if (inner_loop == loop | |
482 | || flow_loop_nested_p (loop, inner_loop)) | |
9baba81b | 483 | { |
a14865db | 484 | tree nb_iter = number_of_latch_executions (inner_loop); |
9baba81b SP |
485 | |
486 | if (nb_iter == chrec_dont_know) | |
487 | return chrec_dont_know; | |
488 | else | |
489 | { | |
490 | tree res; | |
491 | ||
9baba81b SP |
492 | /* evolution_fn is the evolution function in LOOP. Get |
493 | its value in the nb_iter-th iteration. */ | |
494 | res = chrec_apply (inner_loop->num, evolution_fn, nb_iter); | |
42e6eec5 SP |
495 | |
496 | if (chrec_contains_symbols_defined_in_loop (res, loop->num)) | |
497 | res = instantiate_parameters (loop, res); | |
498 | ||
8c27b7d4 | 499 | /* Continue the computation until ending on a parent of LOOP. */ |
9baba81b SP |
500 | return compute_overall_effect_of_inner_loop (loop, res); |
501 | } | |
502 | } | |
503 | else | |
504 | return evolution_fn; | |
505 | } | |
b8698a0f | 506 | |
9baba81b SP |
507 | /* If the evolution function is an invariant, there is nothing to do. */ |
508 | else if (no_evolution_in_loop_p (evolution_fn, loop->num, &val) && val) | |
509 | return evolution_fn; | |
b8698a0f | 510 | |
9baba81b SP |
511 | else |
512 | return chrec_dont_know; | |
513 | } | |
514 | ||
515 | /* Determine whether the CHREC is always positive/negative. If the expression | |
516 | cannot be statically analyzed, return false, otherwise set the answer into | |
517 | VALUE. */ | |
518 | ||
519 | bool | |
520 | chrec_is_positive (tree chrec, bool *value) | |
521 | { | |
16a2acea | 522 | bool value0, value1, value2; |
a14865db | 523 | tree end_value, nb_iter; |
b8698a0f | 524 | |
9baba81b SP |
525 | switch (TREE_CODE (chrec)) |
526 | { | |
527 | case POLYNOMIAL_CHREC: | |
528 | if (!chrec_is_positive (CHREC_LEFT (chrec), &value0) | |
529 | || !chrec_is_positive (CHREC_RIGHT (chrec), &value1)) | |
530 | return false; | |
b8698a0f | 531 | |
9baba81b SP |
532 | /* FIXME -- overflows. */ |
533 | if (value0 == value1) | |
534 | { | |
535 | *value = value0; | |
536 | return true; | |
537 | } | |
538 | ||
539 | /* Otherwise the chrec is under the form: "{-197, +, 2}_1", | |
540 | and the proof consists in showing that the sign never | |
541 | changes during the execution of the loop, from 0 to | |
542 | loop->nb_iterations. */ | |
543 | if (!evolution_function_is_affine_p (chrec)) | |
544 | return false; | |
545 | ||
a14865db | 546 | nb_iter = number_of_latch_executions (get_chrec_loop (chrec)); |
9baba81b SP |
547 | if (chrec_contains_undetermined (nb_iter)) |
548 | return false; | |
549 | ||
9baba81b SP |
550 | #if 0 |
551 | /* TODO -- If the test is after the exit, we may decrease the number of | |
552 | iterations by one. */ | |
553 | if (after_exit) | |
16a2acea | 554 | nb_iter = chrec_fold_minus (type, nb_iter, build_int_cst (type, 1)); |
9baba81b SP |
555 | #endif |
556 | ||
557 | end_value = chrec_apply (CHREC_VARIABLE (chrec), chrec, nb_iter); | |
b8698a0f | 558 | |
9baba81b SP |
559 | if (!chrec_is_positive (end_value, &value2)) |
560 | return false; | |
b8698a0f | 561 | |
9baba81b SP |
562 | *value = value0; |
563 | return value0 == value1; | |
b8698a0f | 564 | |
9baba81b SP |
565 | case INTEGER_CST: |
566 | *value = (tree_int_cst_sgn (chrec) == 1); | |
567 | return true; | |
b8698a0f | 568 | |
9baba81b SP |
569 | default: |
570 | return false; | |
571 | } | |
572 | } | |
573 | ||
574 | /* Associate CHREC to SCALAR. */ | |
575 | ||
576 | static void | |
a213b219 | 577 | set_scalar_evolution (basic_block instantiated_below, tree scalar, tree chrec) |
9baba81b SP |
578 | { |
579 | tree *scalar_info; | |
b8698a0f | 580 | |
9baba81b SP |
581 | if (TREE_CODE (scalar) != SSA_NAME) |
582 | return; | |
583 | ||
a213b219 | 584 | scalar_info = find_var_scev_info (instantiated_below, scalar); |
b8698a0f | 585 | |
9baba81b SP |
586 | if (dump_file) |
587 | { | |
588 | if (dump_flags & TDF_DETAILS) | |
589 | { | |
590 | fprintf (dump_file, "(set_scalar_evolution \n"); | |
a213b219 SP |
591 | fprintf (dump_file, " instantiated_below = %d \n", |
592 | instantiated_below->index); | |
9baba81b SP |
593 | fprintf (dump_file, " (scalar = "); |
594 | print_generic_expr (dump_file, scalar, 0); | |
595 | fprintf (dump_file, ")\n (scalar_evolution = "); | |
596 | print_generic_expr (dump_file, chrec, 0); | |
597 | fprintf (dump_file, "))\n"); | |
598 | } | |
599 | if (dump_flags & TDF_STATS) | |
600 | nb_set_scev++; | |
601 | } | |
b8698a0f | 602 | |
9baba81b SP |
603 | *scalar_info = chrec; |
604 | } | |
605 | ||
a213b219 SP |
606 | /* Retrieve the chrec associated to SCALAR instantiated below |
607 | INSTANTIATED_BELOW block. */ | |
9baba81b SP |
608 | |
609 | static tree | |
a213b219 | 610 | get_scalar_evolution (basic_block instantiated_below, tree scalar) |
9baba81b SP |
611 | { |
612 | tree res; | |
b8698a0f | 613 | |
9baba81b SP |
614 | if (dump_file) |
615 | { | |
616 | if (dump_flags & TDF_DETAILS) | |
617 | { | |
618 | fprintf (dump_file, "(get_scalar_evolution \n"); | |
619 | fprintf (dump_file, " (scalar = "); | |
620 | print_generic_expr (dump_file, scalar, 0); | |
621 | fprintf (dump_file, ")\n"); | |
622 | } | |
623 | if (dump_flags & TDF_STATS) | |
624 | nb_get_scev++; | |
625 | } | |
b8698a0f | 626 | |
9baba81b SP |
627 | switch (TREE_CODE (scalar)) |
628 | { | |
629 | case SSA_NAME: | |
a213b219 | 630 | res = *find_var_scev_info (instantiated_below, scalar); |
9baba81b SP |
631 | break; |
632 | ||
633 | case REAL_CST: | |
325217ed | 634 | case FIXED_CST: |
9baba81b SP |
635 | case INTEGER_CST: |
636 | res = scalar; | |
637 | break; | |
638 | ||
639 | default: | |
640 | res = chrec_not_analyzed_yet; | |
641 | break; | |
642 | } | |
b8698a0f | 643 | |
9baba81b SP |
644 | if (dump_file && (dump_flags & TDF_DETAILS)) |
645 | { | |
646 | fprintf (dump_file, " (scalar_evolution = "); | |
647 | print_generic_expr (dump_file, res, 0); | |
648 | fprintf (dump_file, "))\n"); | |
649 | } | |
b8698a0f | 650 | |
9baba81b SP |
651 | return res; |
652 | } | |
653 | ||
654 | /* Helper function for add_to_evolution. Returns the evolution | |
655 | function for an assignment of the form "a = b + c", where "a" and | |
656 | "b" are on the strongly connected component. CHREC_BEFORE is the | |
657 | information that we already have collected up to this point. | |
b8698a0f L |
658 | TO_ADD is the evolution of "c". |
659 | ||
9baba81b SP |
660 | When CHREC_BEFORE has an evolution part in LOOP_NB, add to this |
661 | evolution the expression TO_ADD, otherwise construct an evolution | |
662 | part for this loop. */ | |
663 | ||
664 | static tree | |
e2157b49 | 665 | add_to_evolution_1 (unsigned loop_nb, tree chrec_before, tree to_add, |
726a989a | 666 | gimple at_stmt) |
9baba81b | 667 | { |
e2157b49 | 668 | tree type, left, right; |
677cc14d | 669 | struct loop *loop = get_loop (loop_nb), *chloop; |
e2157b49 | 670 | |
9baba81b SP |
671 | switch (TREE_CODE (chrec_before)) |
672 | { | |
673 | case POLYNOMIAL_CHREC: | |
677cc14d ZD |
674 | chloop = get_chrec_loop (chrec_before); |
675 | if (chloop == loop | |
676 | || flow_loop_nested_p (chloop, loop)) | |
9baba81b SP |
677 | { |
678 | unsigned var; | |
e2157b49 SP |
679 | |
680 | type = chrec_type (chrec_before); | |
b8698a0f | 681 | |
9baba81b | 682 | /* When there is no evolution part in this loop, build it. */ |
677cc14d | 683 | if (chloop != loop) |
9baba81b SP |
684 | { |
685 | var = loop_nb; | |
686 | left = chrec_before; | |
7e0923cd SP |
687 | right = SCALAR_FLOAT_TYPE_P (type) |
688 | ? build_real (type, dconst0) | |
689 | : build_int_cst (type, 0); | |
9baba81b SP |
690 | } |
691 | else | |
692 | { | |
693 | var = CHREC_VARIABLE (chrec_before); | |
694 | left = CHREC_LEFT (chrec_before); | |
695 | right = CHREC_RIGHT (chrec_before); | |
696 | } | |
697 | ||
e2157b49 | 698 | to_add = chrec_convert (type, to_add, at_stmt); |
5be014d5 AP |
699 | right = chrec_convert_rhs (type, right, at_stmt); |
700 | right = chrec_fold_plus (chrec_type (right), right, to_add); | |
e2157b49 | 701 | return build_polynomial_chrec (var, left, right); |
9baba81b SP |
702 | } |
703 | else | |
e2157b49 | 704 | { |
677cc14d ZD |
705 | gcc_assert (flow_loop_nested_p (loop, chloop)); |
706 | ||
e2157b49 SP |
707 | /* Search the evolution in LOOP_NB. */ |
708 | left = add_to_evolution_1 (loop_nb, CHREC_LEFT (chrec_before), | |
709 | to_add, at_stmt); | |
710 | right = CHREC_RIGHT (chrec_before); | |
5be014d5 | 711 | right = chrec_convert_rhs (chrec_type (left), right, at_stmt); |
e2157b49 SP |
712 | return build_polynomial_chrec (CHREC_VARIABLE (chrec_before), |
713 | left, right); | |
714 | } | |
b8698a0f | 715 | |
9baba81b SP |
716 | default: |
717 | /* These nodes do not depend on a loop. */ | |
718 | if (chrec_before == chrec_dont_know) | |
719 | return chrec_dont_know; | |
e2157b49 SP |
720 | |
721 | left = chrec_before; | |
5be014d5 | 722 | right = chrec_convert_rhs (chrec_type (left), to_add, at_stmt); |
e2157b49 | 723 | return build_polynomial_chrec (loop_nb, left, right); |
9baba81b SP |
724 | } |
725 | } | |
726 | ||
727 | /* Add TO_ADD to the evolution part of CHREC_BEFORE in the dimension | |
b8698a0f L |
728 | of LOOP_NB. |
729 | ||
9baba81b SP |
730 | Description (provided for completeness, for those who read code in |
731 | a plane, and for my poor 62 bytes brain that would have forgotten | |
732 | all this in the next two or three months): | |
b8698a0f | 733 | |
9baba81b SP |
734 | The algorithm of translation of programs from the SSA representation |
735 | into the chrecs syntax is based on a pattern matching. After having | |
736 | reconstructed the overall tree expression for a loop, there are only | |
737 | two cases that can arise: | |
b8698a0f | 738 | |
9baba81b SP |
739 | 1. a = loop-phi (init, a + expr) |
740 | 2. a = loop-phi (init, expr) | |
b8698a0f | 741 | |
9baba81b SP |
742 | where EXPR is either a scalar constant with respect to the analyzed |
743 | loop (this is a degree 0 polynomial), or an expression containing | |
744 | other loop-phi definitions (these are higher degree polynomials). | |
b8698a0f | 745 | |
9baba81b | 746 | Examples: |
b8698a0f L |
747 | |
748 | 1. | |
9baba81b SP |
749 | | init = ... |
750 | | loop_1 | |
751 | | a = phi (init, a + 5) | |
752 | | endloop | |
b8698a0f L |
753 | |
754 | 2. | |
9baba81b SP |
755 | | inita = ... |
756 | | initb = ... | |
757 | | loop_1 | |
758 | | a = phi (inita, 2 * b + 3) | |
759 | | b = phi (initb, b + 1) | |
760 | | endloop | |
b8698a0f L |
761 | |
762 | For the first case, the semantics of the SSA representation is: | |
763 | ||
9baba81b | 764 | | a (x) = init + \sum_{j = 0}^{x - 1} expr (j) |
b8698a0f | 765 | |
9baba81b SP |
766 | that is, there is a loop index "x" that determines the scalar value |
767 | of the variable during the loop execution. During the first | |
768 | iteration, the value is that of the initial condition INIT, while | |
769 | during the subsequent iterations, it is the sum of the initial | |
770 | condition with the sum of all the values of EXPR from the initial | |
b8698a0f L |
771 | iteration to the before last considered iteration. |
772 | ||
9baba81b | 773 | For the second case, the semantics of the SSA program is: |
b8698a0f | 774 | |
9baba81b SP |
775 | | a (x) = init, if x = 0; |
776 | | expr (x - 1), otherwise. | |
b8698a0f | 777 | |
9baba81b | 778 | The second case corresponds to the PEELED_CHREC, whose syntax is |
b8698a0f L |
779 | close to the syntax of a loop-phi-node: |
780 | ||
9baba81b | 781 | | phi (init, expr) vs. (init, expr)_x |
b8698a0f | 782 | |
9baba81b | 783 | The proof of the translation algorithm for the first case is a |
b8698a0f L |
784 | proof by structural induction based on the degree of EXPR. |
785 | ||
9baba81b SP |
786 | Degree 0: |
787 | When EXPR is a constant with respect to the analyzed loop, or in | |
788 | other words when EXPR is a polynomial of degree 0, the evolution of | |
789 | the variable A in the loop is an affine function with an initial | |
790 | condition INIT, and a step EXPR. In order to show this, we start | |
791 | from the semantics of the SSA representation: | |
b8698a0f | 792 | |
9baba81b | 793 | f (x) = init + \sum_{j = 0}^{x - 1} expr (j) |
b8698a0f | 794 | |
9baba81b | 795 | and since "expr (j)" is a constant with respect to "j", |
b8698a0f L |
796 | |
797 | f (x) = init + x * expr | |
798 | ||
9baba81b SP |
799 | Finally, based on the semantics of the pure sum chrecs, by |
800 | identification we get the corresponding chrecs syntax: | |
b8698a0f L |
801 | |
802 | f (x) = init * \binom{x}{0} + expr * \binom{x}{1} | |
9baba81b | 803 | f (x) -> {init, +, expr}_x |
b8698a0f | 804 | |
9baba81b SP |
805 | Higher degree: |
806 | Suppose that EXPR is a polynomial of degree N with respect to the | |
807 | analyzed loop_x for which we have already determined that it is | |
808 | written under the chrecs syntax: | |
b8698a0f | 809 | |
9baba81b | 810 | | expr (x) -> {b_0, +, b_1, +, ..., +, b_{n-1}} (x) |
b8698a0f | 811 | |
9baba81b | 812 | We start from the semantics of the SSA program: |
b8698a0f | 813 | |
9baba81b SP |
814 | | f (x) = init + \sum_{j = 0}^{x - 1} expr (j) |
815 | | | |
b8698a0f | 816 | | f (x) = init + \sum_{j = 0}^{x - 1} |
9baba81b SP |
817 | | (b_0 * \binom{j}{0} + ... + b_{n-1} * \binom{j}{n-1}) |
818 | | | |
b8698a0f L |
819 | | f (x) = init + \sum_{j = 0}^{x - 1} |
820 | | \sum_{k = 0}^{n - 1} (b_k * \binom{j}{k}) | |
9baba81b | 821 | | |
b8698a0f L |
822 | | f (x) = init + \sum_{k = 0}^{n - 1} |
823 | | (b_k * \sum_{j = 0}^{x - 1} \binom{j}{k}) | |
9baba81b | 824 | | |
b8698a0f L |
825 | | f (x) = init + \sum_{k = 0}^{n - 1} |
826 | | (b_k * \binom{x}{k + 1}) | |
9baba81b | 827 | | |
b8698a0f L |
828 | | f (x) = init + b_0 * \binom{x}{1} + ... |
829 | | + b_{n-1} * \binom{x}{n} | |
9baba81b | 830 | | |
b8698a0f L |
831 | | f (x) = init * \binom{x}{0} + b_0 * \binom{x}{1} + ... |
832 | | + b_{n-1} * \binom{x}{n} | |
9baba81b | 833 | | |
b8698a0f | 834 | |
9baba81b | 835 | And finally from the definition of the chrecs syntax, we identify: |
b8698a0f L |
836 | | f (x) -> {init, +, b_0, +, ..., +, b_{n-1}}_x |
837 | ||
9baba81b SP |
838 | This shows the mechanism that stands behind the add_to_evolution |
839 | function. An important point is that the use of symbolic | |
840 | parameters avoids the need of an analysis schedule. | |
b8698a0f | 841 | |
9baba81b | 842 | Example: |
b8698a0f | 843 | |
9baba81b SP |
844 | | inita = ... |
845 | | initb = ... | |
b8698a0f | 846 | | loop_1 |
9baba81b SP |
847 | | a = phi (inita, a + 2 + b) |
848 | | b = phi (initb, b + 1) | |
849 | | endloop | |
b8698a0f | 850 | |
9baba81b | 851 | When analyzing "a", the algorithm keeps "b" symbolically: |
b8698a0f | 852 | |
9baba81b | 853 | | a -> {inita, +, 2 + b}_1 |
b8698a0f | 854 | |
9baba81b | 855 | Then, after instantiation, the analyzer ends on the evolution: |
b8698a0f | 856 | |
9baba81b SP |
857 | | a -> {inita, +, 2 + initb, +, 1}_1 |
858 | ||
859 | */ | |
860 | ||
b8698a0f | 861 | static tree |
e2157b49 | 862 | add_to_evolution (unsigned loop_nb, tree chrec_before, enum tree_code code, |
726a989a | 863 | tree to_add, gimple at_stmt) |
9baba81b SP |
864 | { |
865 | tree type = chrec_type (to_add); | |
866 | tree res = NULL_TREE; | |
b8698a0f | 867 | |
9baba81b SP |
868 | if (to_add == NULL_TREE) |
869 | return chrec_before; | |
b8698a0f | 870 | |
9baba81b SP |
871 | /* TO_ADD is either a scalar, or a parameter. TO_ADD is not |
872 | instantiated at this point. */ | |
873 | if (TREE_CODE (to_add) == POLYNOMIAL_CHREC) | |
874 | /* This should not happen. */ | |
875 | return chrec_dont_know; | |
b8698a0f | 876 | |
9baba81b SP |
877 | if (dump_file && (dump_flags & TDF_DETAILS)) |
878 | { | |
879 | fprintf (dump_file, "(add_to_evolution \n"); | |
880 | fprintf (dump_file, " (loop_nb = %d)\n", loop_nb); | |
881 | fprintf (dump_file, " (chrec_before = "); | |
882 | print_generic_expr (dump_file, chrec_before, 0); | |
883 | fprintf (dump_file, ")\n (to_add = "); | |
884 | print_generic_expr (dump_file, to_add, 0); | |
885 | fprintf (dump_file, ")\n"); | |
886 | } | |
887 | ||
888 | if (code == MINUS_EXPR) | |
9d2b0e12 VR |
889 | to_add = chrec_fold_multiply (type, to_add, SCALAR_FLOAT_TYPE_P (type) |
890 | ? build_real (type, dconstm1) | |
891 | : build_int_cst_type (type, -1)); | |
9baba81b | 892 | |
e2157b49 | 893 | res = add_to_evolution_1 (loop_nb, chrec_before, to_add, at_stmt); |
9baba81b SP |
894 | |
895 | if (dump_file && (dump_flags & TDF_DETAILS)) | |
896 | { | |
897 | fprintf (dump_file, " (res = "); | |
898 | print_generic_expr (dump_file, res, 0); | |
899 | fprintf (dump_file, "))\n"); | |
900 | } | |
901 | ||
902 | return res; | |
903 | } | |
904 | ||
905 | /* Helper function. */ | |
906 | ||
907 | static inline tree | |
b8698a0f | 908 | set_nb_iterations_in_loop (struct loop *loop, |
9baba81b SP |
909 | tree res) |
910 | { | |
9baba81b SP |
911 | if (dump_file && (dump_flags & TDF_DETAILS)) |
912 | { | |
913 | fprintf (dump_file, " (set_nb_iterations_in_loop = "); | |
914 | print_generic_expr (dump_file, res, 0); | |
915 | fprintf (dump_file, "))\n"); | |
916 | } | |
b8698a0f | 917 | |
9baba81b SP |
918 | loop->nb_iterations = res; |
919 | return res; | |
920 | } | |
921 | ||
922 | \f | |
923 | ||
924 | /* This section selects the loops that will be good candidates for the | |
925 | scalar evolution analysis. For the moment, greedily select all the | |
926 | loop nests we could analyze. */ | |
927 | ||
9baba81b SP |
928 | /* For a loop with a single exit edge, return the COND_EXPR that |
929 | guards the exit edge. If the expression is too difficult to | |
930 | analyze, then give up. */ | |
931 | ||
b8698a0f | 932 | gimple |
22ea9ec0 | 933 | get_loop_exit_condition (const struct loop *loop) |
9baba81b | 934 | { |
726a989a | 935 | gimple res = NULL; |
ac8f6c69 | 936 | edge exit_edge = single_exit (loop); |
b8698a0f | 937 | |
9baba81b SP |
938 | if (dump_file && (dump_flags & TDF_DETAILS)) |
939 | fprintf (dump_file, "(get_loop_exit_condition \n "); | |
b8698a0f | 940 | |
82b85a85 | 941 | if (exit_edge) |
9baba81b | 942 | { |
726a989a | 943 | gimple stmt; |
b8698a0f | 944 | |
726a989a RB |
945 | stmt = last_stmt (exit_edge->src); |
946 | if (gimple_code (stmt) == GIMPLE_COND) | |
947 | res = stmt; | |
9baba81b | 948 | } |
b8698a0f | 949 | |
9baba81b SP |
950 | if (dump_file && (dump_flags & TDF_DETAILS)) |
951 | { | |
726a989a | 952 | print_gimple_stmt (dump_file, res, 0, 0); |
9baba81b SP |
953 | fprintf (dump_file, ")\n"); |
954 | } | |
b8698a0f | 955 | |
9baba81b SP |
956 | return res; |
957 | } | |
958 | ||
959 | /* Recursively determine and enqueue the exit conditions for a loop. */ | |
960 | ||
b8698a0f L |
961 | static void |
962 | get_exit_conditions_rec (struct loop *loop, | |
726a989a | 963 | VEC(gimple,heap) **exit_conditions) |
9baba81b SP |
964 | { |
965 | if (!loop) | |
966 | return; | |
b8698a0f | 967 | |
9baba81b SP |
968 | /* Recurse on the inner loops, then on the next (sibling) loops. */ |
969 | get_exit_conditions_rec (loop->inner, exit_conditions); | |
970 | get_exit_conditions_rec (loop->next, exit_conditions); | |
b8698a0f | 971 | |
ac8f6c69 | 972 | if (single_exit (loop)) |
9baba81b | 973 | { |
726a989a | 974 | gimple loop_condition = get_loop_exit_condition (loop); |
b8698a0f | 975 | |
9baba81b | 976 | if (loop_condition) |
726a989a | 977 | VEC_safe_push (gimple, heap, *exit_conditions, loop_condition); |
9baba81b SP |
978 | } |
979 | } | |
980 | ||
981 | /* Select the candidate loop nests for the analysis. This function | |
471854f8 | 982 | initializes the EXIT_CONDITIONS array. */ |
9baba81b SP |
983 | |
984 | static void | |
726a989a | 985 | select_loops_exit_conditions (VEC(gimple,heap) **exit_conditions) |
9baba81b | 986 | { |
d73be268 | 987 | struct loop *function_body = current_loops->tree_root; |
b8698a0f | 988 | |
9baba81b SP |
989 | get_exit_conditions_rec (function_body->inner, exit_conditions); |
990 | } | |
991 | ||
992 | \f | |
993 | /* Depth first search algorithm. */ | |
994 | ||
c59dabbe SP |
995 | typedef enum t_bool { |
996 | t_false, | |
997 | t_true, | |
998 | t_dont_know | |
999 | } t_bool; | |
1000 | ||
1001 | ||
726a989a | 1002 | static t_bool follow_ssa_edge (struct loop *loop, gimple, gimple, tree *, int); |
9baba81b | 1003 | |
726a989a | 1004 | /* Follow the ssa edge into the binary expression RHS0 CODE RHS1. |
9baba81b SP |
1005 | Return true if the strongly connected component has been found. */ |
1006 | ||
c59dabbe | 1007 | static t_bool |
726a989a RB |
1008 | follow_ssa_edge_binary (struct loop *loop, gimple at_stmt, |
1009 | tree type, tree rhs0, enum tree_code code, tree rhs1, | |
1010 | gimple halting_phi, tree *evolution_of_loop, int limit) | |
9baba81b | 1011 | { |
c59dabbe | 1012 | t_bool res = t_false; |
b2a93c0a | 1013 | tree evol; |
726a989a | 1014 | |
5be014d5 | 1015 | switch (code) |
9baba81b | 1016 | { |
5be014d5 | 1017 | case POINTER_PLUS_EXPR: |
9baba81b | 1018 | case PLUS_EXPR: |
9baba81b SP |
1019 | if (TREE_CODE (rhs0) == SSA_NAME) |
1020 | { | |
1021 | if (TREE_CODE (rhs1) == SSA_NAME) | |
1022 | { | |
b8698a0f | 1023 | /* Match an assignment under the form: |
9baba81b | 1024 | "a = b + c". */ |
b8698a0f | 1025 | |
9e824336 ZD |
1026 | /* We want only assignments of form "name + name" contribute to |
1027 | LIMIT, as the other cases do not necessarily contribute to | |
1028 | the complexity of the expression. */ | |
1029 | limit++; | |
1030 | ||
b2a93c0a | 1031 | evol = *evolution_of_loop; |
b8698a0f | 1032 | res = follow_ssa_edge |
726a989a | 1033 | (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi, &evol, limit); |
b8698a0f | 1034 | |
c59dabbe | 1035 | if (res == t_true) |
b8698a0f L |
1036 | *evolution_of_loop = add_to_evolution |
1037 | (loop->num, | |
1038 | chrec_convert (type, evol, at_stmt), | |
5be014d5 | 1039 | code, rhs1, at_stmt); |
b8698a0f | 1040 | |
c59dabbe | 1041 | else if (res == t_false) |
9baba81b | 1042 | { |
b8698a0f L |
1043 | res = follow_ssa_edge |
1044 | (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi, | |
c59dabbe | 1045 | evolution_of_loop, limit); |
b8698a0f | 1046 | |
c59dabbe | 1047 | if (res == t_true) |
b8698a0f L |
1048 | *evolution_of_loop = add_to_evolution |
1049 | (loop->num, | |
1050 | chrec_convert (type, *evolution_of_loop, at_stmt), | |
5be014d5 | 1051 | code, rhs0, at_stmt); |
c59dabbe SP |
1052 | |
1053 | else if (res == t_dont_know) | |
1054 | *evolution_of_loop = chrec_dont_know; | |
9baba81b | 1055 | } |
c59dabbe SP |
1056 | |
1057 | else if (res == t_dont_know) | |
1058 | *evolution_of_loop = chrec_dont_know; | |
9baba81b | 1059 | } |
b8698a0f | 1060 | |
9baba81b SP |
1061 | else |
1062 | { | |
b8698a0f | 1063 | /* Match an assignment under the form: |
9baba81b | 1064 | "a = b + ...". */ |
b8698a0f L |
1065 | res = follow_ssa_edge |
1066 | (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi, | |
c59dabbe SP |
1067 | evolution_of_loop, limit); |
1068 | if (res == t_true) | |
b8698a0f | 1069 | *evolution_of_loop = add_to_evolution |
726a989a | 1070 | (loop->num, chrec_convert (type, *evolution_of_loop, |
1e8552eb | 1071 | at_stmt), |
5be014d5 | 1072 | code, rhs1, at_stmt); |
c59dabbe SP |
1073 | |
1074 | else if (res == t_dont_know) | |
1075 | *evolution_of_loop = chrec_dont_know; | |
9baba81b SP |
1076 | } |
1077 | } | |
b8698a0f | 1078 | |
9baba81b SP |
1079 | else if (TREE_CODE (rhs1) == SSA_NAME) |
1080 | { | |
b8698a0f | 1081 | /* Match an assignment under the form: |
9baba81b | 1082 | "a = ... + c". */ |
b8698a0f L |
1083 | res = follow_ssa_edge |
1084 | (loop, SSA_NAME_DEF_STMT (rhs1), halting_phi, | |
c59dabbe SP |
1085 | evolution_of_loop, limit); |
1086 | if (res == t_true) | |
b8698a0f | 1087 | *evolution_of_loop = add_to_evolution |
726a989a | 1088 | (loop->num, chrec_convert (type, *evolution_of_loop, |
1e8552eb | 1089 | at_stmt), |
5be014d5 | 1090 | code, rhs0, at_stmt); |
c59dabbe SP |
1091 | |
1092 | else if (res == t_dont_know) | |
1093 | *evolution_of_loop = chrec_dont_know; | |
9baba81b SP |
1094 | } |
1095 | ||
1096 | else | |
b8698a0f | 1097 | /* Otherwise, match an assignment under the form: |
9baba81b SP |
1098 | "a = ... + ...". */ |
1099 | /* And there is nothing to do. */ | |
c59dabbe | 1100 | res = t_false; |
9baba81b | 1101 | break; |
b8698a0f | 1102 | |
9baba81b SP |
1103 | case MINUS_EXPR: |
1104 | /* This case is under the form "opnd0 = rhs0 - rhs1". */ | |
9baba81b | 1105 | if (TREE_CODE (rhs0) == SSA_NAME) |
9baba81b | 1106 | { |
b8698a0f | 1107 | /* Match an assignment under the form: |
f8e9d512 | 1108 | "a = b - ...". */ |
9e824336 ZD |
1109 | |
1110 | /* We want only assignments of form "name - name" contribute to | |
1111 | LIMIT, as the other cases do not necessarily contribute to | |
1112 | the complexity of the expression. */ | |
1113 | if (TREE_CODE (rhs1) == SSA_NAME) | |
1114 | limit++; | |
1115 | ||
b8698a0f | 1116 | res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0), halting_phi, |
c59dabbe SP |
1117 | evolution_of_loop, limit); |
1118 | if (res == t_true) | |
b8698a0f | 1119 | *evolution_of_loop = add_to_evolution |
726a989a | 1120 | (loop->num, chrec_convert (type, *evolution_of_loop, at_stmt), |
e2157b49 | 1121 | MINUS_EXPR, rhs1, at_stmt); |
c59dabbe SP |
1122 | |
1123 | else if (res == t_dont_know) | |
1124 | *evolution_of_loop = chrec_dont_know; | |
9baba81b | 1125 | } |
9baba81b | 1126 | else |
b8698a0f | 1127 | /* Otherwise, match an assignment under the form: |
9baba81b SP |
1128 | "a = ... - ...". */ |
1129 | /* And there is nothing to do. */ | |
c59dabbe | 1130 | res = t_false; |
9baba81b | 1131 | break; |
726a989a RB |
1132 | |
1133 | default: | |
1134 | res = t_false; | |
1135 | } | |
1136 | ||
1137 | return res; | |
1138 | } | |
b8698a0f | 1139 | |
726a989a RB |
1140 | /* Follow the ssa edge into the expression EXPR. |
1141 | Return true if the strongly connected component has been found. */ | |
1142 | ||
1143 | static t_bool | |
b8698a0f | 1144 | follow_ssa_edge_expr (struct loop *loop, gimple at_stmt, tree expr, |
726a989a RB |
1145 | gimple halting_phi, tree *evolution_of_loop, int limit) |
1146 | { | |
5aefc6a0 EB |
1147 | enum tree_code code = TREE_CODE (expr); |
1148 | tree type = TREE_TYPE (expr), rhs0, rhs1; | |
1149 | t_bool res; | |
1150 | ||
726a989a | 1151 | /* The EXPR is one of the following cases: |
b8698a0f | 1152 | - an SSA_NAME, |
726a989a | 1153 | - an INTEGER_CST, |
b8698a0f L |
1154 | - a PLUS_EXPR, |
1155 | - a POINTER_PLUS_EXPR, | |
726a989a RB |
1156 | - a MINUS_EXPR, |
1157 | - an ASSERT_EXPR, | |
1158 | - other cases are not yet handled. */ | |
5aefc6a0 | 1159 | |
726a989a RB |
1160 | switch (code) |
1161 | { | |
5aefc6a0 | 1162 | CASE_CONVERT: |
726a989a RB |
1163 | /* This assignment is under the form "a_1 = (cast) rhs. */ |
1164 | res = follow_ssa_edge_expr (loop, at_stmt, TREE_OPERAND (expr, 0), | |
1165 | halting_phi, evolution_of_loop, limit); | |
1166 | *evolution_of_loop = chrec_convert (type, *evolution_of_loop, at_stmt); | |
1167 | break; | |
1168 | ||
1169 | case INTEGER_CST: | |
1170 | /* This assignment is under the form "a_1 = 7". */ | |
1171 | res = t_false; | |
1172 | break; | |
5aefc6a0 | 1173 | |
726a989a RB |
1174 | case SSA_NAME: |
1175 | /* This assignment is under the form: "a_1 = b_2". */ | |
b8698a0f | 1176 | res = follow_ssa_edge |
726a989a RB |
1177 | (loop, SSA_NAME_DEF_STMT (expr), halting_phi, evolution_of_loop, limit); |
1178 | break; | |
5aefc6a0 | 1179 | |
726a989a RB |
1180 | case POINTER_PLUS_EXPR: |
1181 | case PLUS_EXPR: | |
1182 | case MINUS_EXPR: | |
1183 | /* This case is under the form "rhs0 +- rhs1". */ | |
1184 | rhs0 = TREE_OPERAND (expr, 0); | |
1185 | rhs1 = TREE_OPERAND (expr, 1); | |
5aefc6a0 EB |
1186 | type = TREE_TYPE (rhs0); |
1187 | STRIP_USELESS_TYPE_CONVERSION (rhs0); | |
1188 | STRIP_USELESS_TYPE_CONVERSION (rhs1); | |
1189 | res = follow_ssa_edge_binary (loop, at_stmt, type, rhs0, code, rhs1, | |
1190 | halting_phi, evolution_of_loop, limit); | |
1191 | break; | |
726a989a | 1192 | |
0bca51f0 | 1193 | case ASSERT_EXPR: |
5aefc6a0 EB |
1194 | /* This assignment is of the form: "a_1 = ASSERT_EXPR <a_2, ...>" |
1195 | It must be handled as a copy assignment of the form a_1 = a_2. */ | |
1196 | rhs0 = ASSERT_EXPR_VAR (expr); | |
1197 | if (TREE_CODE (rhs0) == SSA_NAME) | |
1198 | res = follow_ssa_edge (loop, SSA_NAME_DEF_STMT (rhs0), | |
1199 | halting_phi, evolution_of_loop, limit); | |
1200 | else | |
1201 | res = t_false; | |
1202 | break; | |
0bca51f0 | 1203 | |
9baba81b | 1204 | default: |
c59dabbe | 1205 | res = t_false; |
9baba81b SP |
1206 | break; |
1207 | } | |
5aefc6a0 | 1208 | |
9baba81b SP |
1209 | return res; |
1210 | } | |
1211 | ||
726a989a RB |
1212 | /* Follow the ssa edge into the right hand side of an assignment STMT. |
1213 | Return true if the strongly connected component has been found. */ | |
1214 | ||
1215 | static t_bool | |
1216 | follow_ssa_edge_in_rhs (struct loop *loop, gimple stmt, | |
1217 | gimple halting_phi, tree *evolution_of_loop, int limit) | |
1218 | { | |
726a989a | 1219 | enum tree_code code = gimple_assign_rhs_code (stmt); |
5aefc6a0 EB |
1220 | tree type = gimple_expr_type (stmt), rhs1, rhs2; |
1221 | t_bool res; | |
726a989a | 1222 | |
5aefc6a0 | 1223 | switch (code) |
726a989a | 1224 | { |
5aefc6a0 EB |
1225 | CASE_CONVERT: |
1226 | /* This assignment is under the form "a_1 = (cast) rhs. */ | |
1227 | res = follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt), | |
1228 | halting_phi, evolution_of_loop, limit); | |
1229 | *evolution_of_loop = chrec_convert (type, *evolution_of_loop, stmt); | |
1230 | break; | |
1231 | ||
1232 | case POINTER_PLUS_EXPR: | |
1233 | case PLUS_EXPR: | |
1234 | case MINUS_EXPR: | |
1235 | rhs1 = gimple_assign_rhs1 (stmt); | |
1236 | rhs2 = gimple_assign_rhs2 (stmt); | |
1237 | type = TREE_TYPE (rhs1); | |
1238 | res = follow_ssa_edge_binary (loop, stmt, type, rhs1, code, rhs2, | |
218d1c24 | 1239 | halting_phi, evolution_of_loop, limit); |
5aefc6a0 | 1240 | break; |
218d1c24 | 1241 | |
726a989a | 1242 | default: |
5aefc6a0 EB |
1243 | if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS) |
1244 | res = follow_ssa_edge_expr (loop, stmt, gimple_assign_rhs1 (stmt), | |
1245 | halting_phi, evolution_of_loop, limit); | |
1246 | else | |
1247 | res = t_false; | |
1248 | break; | |
726a989a | 1249 | } |
5aefc6a0 EB |
1250 | |
1251 | return res; | |
726a989a RB |
1252 | } |
1253 | ||
9baba81b SP |
1254 | /* Checks whether the I-th argument of a PHI comes from a backedge. */ |
1255 | ||
1256 | static bool | |
726a989a | 1257 | backedge_phi_arg_p (gimple phi, int i) |
9baba81b | 1258 | { |
726a989a | 1259 | const_edge e = gimple_phi_arg_edge (phi, i); |
9baba81b SP |
1260 | |
1261 | /* We would in fact like to test EDGE_DFS_BACK here, but we do not care | |
1262 | about updating it anywhere, and this should work as well most of the | |
1263 | time. */ | |
1264 | if (e->flags & EDGE_IRREDUCIBLE_LOOP) | |
1265 | return true; | |
1266 | ||
1267 | return false; | |
1268 | } | |
1269 | ||
1270 | /* Helper function for one branch of the condition-phi-node. Return | |
1271 | true if the strongly connected component has been found following | |
1272 | this path. */ | |
1273 | ||
c59dabbe | 1274 | static inline t_bool |
9baba81b | 1275 | follow_ssa_edge_in_condition_phi_branch (int i, |
b8698a0f L |
1276 | struct loop *loop, |
1277 | gimple condition_phi, | |
726a989a | 1278 | gimple halting_phi, |
9baba81b | 1279 | tree *evolution_of_branch, |
c59dabbe | 1280 | tree init_cond, int limit) |
9baba81b SP |
1281 | { |
1282 | tree branch = PHI_ARG_DEF (condition_phi, i); | |
1283 | *evolution_of_branch = chrec_dont_know; | |
1284 | ||
1285 | /* Do not follow back edges (they must belong to an irreducible loop, which | |
1286 | we really do not want to worry about). */ | |
1287 | if (backedge_phi_arg_p (condition_phi, i)) | |
c59dabbe | 1288 | return t_false; |
9baba81b SP |
1289 | |
1290 | if (TREE_CODE (branch) == SSA_NAME) | |
1291 | { | |
1292 | *evolution_of_branch = init_cond; | |
b8698a0f | 1293 | return follow_ssa_edge (loop, SSA_NAME_DEF_STMT (branch), halting_phi, |
c59dabbe | 1294 | evolution_of_branch, limit); |
9baba81b SP |
1295 | } |
1296 | ||
b8698a0f | 1297 | /* This case occurs when one of the condition branches sets |
89dbed81 | 1298 | the variable to a constant: i.e. a phi-node like |
b8698a0f L |
1299 | "a_2 = PHI <a_7(5), 2(6)>;". |
1300 | ||
1301 | FIXME: This case have to be refined correctly: | |
9baba81b SP |
1302 | in some cases it is possible to say something better than |
1303 | chrec_dont_know, for example using a wrap-around notation. */ | |
c59dabbe | 1304 | return t_false; |
9baba81b SP |
1305 | } |
1306 | ||
1307 | /* This function merges the branches of a condition-phi-node in a | |
1308 | loop. */ | |
1309 | ||
c59dabbe | 1310 | static t_bool |
9baba81b | 1311 | follow_ssa_edge_in_condition_phi (struct loop *loop, |
b8698a0f L |
1312 | gimple condition_phi, |
1313 | gimple halting_phi, | |
c59dabbe | 1314 | tree *evolution_of_loop, int limit) |
9baba81b | 1315 | { |
726a989a | 1316 | int i, n; |
9baba81b SP |
1317 | tree init = *evolution_of_loop; |
1318 | tree evolution_of_branch; | |
c59dabbe SP |
1319 | t_bool res = follow_ssa_edge_in_condition_phi_branch (0, loop, condition_phi, |
1320 | halting_phi, | |
1321 | &evolution_of_branch, | |
1322 | init, limit); | |
1323 | if (res == t_false || res == t_dont_know) | |
1324 | return res; | |
9baba81b | 1325 | |
9baba81b SP |
1326 | *evolution_of_loop = evolution_of_branch; |
1327 | ||
726a989a | 1328 | n = gimple_phi_num_args (condition_phi); |
726a989a | 1329 | for (i = 1; i < n; i++) |
9baba81b | 1330 | { |
e0afb98a SP |
1331 | /* Quickly give up when the evolution of one of the branches is |
1332 | not known. */ | |
1333 | if (*evolution_of_loop == chrec_dont_know) | |
c59dabbe | 1334 | return t_true; |
e0afb98a | 1335 | |
788d3075 RG |
1336 | /* Increase the limit by the PHI argument number to avoid exponential |
1337 | time and memory complexity. */ | |
c59dabbe SP |
1338 | res = follow_ssa_edge_in_condition_phi_branch (i, loop, condition_phi, |
1339 | halting_phi, | |
1340 | &evolution_of_branch, | |
788d3075 | 1341 | init, limit + i); |
c59dabbe SP |
1342 | if (res == t_false || res == t_dont_know) |
1343 | return res; | |
9baba81b SP |
1344 | |
1345 | *evolution_of_loop = chrec_merge (*evolution_of_loop, | |
1346 | evolution_of_branch); | |
1347 | } | |
b8698a0f | 1348 | |
c59dabbe | 1349 | return t_true; |
9baba81b SP |
1350 | } |
1351 | ||
1352 | /* Follow an SSA edge in an inner loop. It computes the overall | |
1353 | effect of the loop, and following the symbolic initial conditions, | |
1354 | it follows the edges in the parent loop. The inner loop is | |
1355 | considered as a single statement. */ | |
1356 | ||
c59dabbe | 1357 | static t_bool |
9baba81b | 1358 | follow_ssa_edge_inner_loop_phi (struct loop *outer_loop, |
b8698a0f | 1359 | gimple loop_phi_node, |
726a989a | 1360 | gimple halting_phi, |
c59dabbe | 1361 | tree *evolution_of_loop, int limit) |
9baba81b SP |
1362 | { |
1363 | struct loop *loop = loop_containing_stmt (loop_phi_node); | |
1364 | tree ev = analyze_scalar_evolution (loop, PHI_RESULT (loop_phi_node)); | |
1365 | ||
1366 | /* Sometimes, the inner loop is too difficult to analyze, and the | |
1367 | result of the analysis is a symbolic parameter. */ | |
1368 | if (ev == PHI_RESULT (loop_phi_node)) | |
1369 | { | |
c59dabbe | 1370 | t_bool res = t_false; |
726a989a | 1371 | int i, n = gimple_phi_num_args (loop_phi_node); |
9baba81b | 1372 | |
726a989a | 1373 | for (i = 0; i < n; i++) |
9baba81b SP |
1374 | { |
1375 | tree arg = PHI_ARG_DEF (loop_phi_node, i); | |
1376 | basic_block bb; | |
1377 | ||
1378 | /* Follow the edges that exit the inner loop. */ | |
726a989a | 1379 | bb = gimple_phi_arg_edge (loop_phi_node, i)->src; |
9baba81b | 1380 | if (!flow_bb_inside_loop_p (loop, bb)) |
726a989a RB |
1381 | res = follow_ssa_edge_expr (outer_loop, loop_phi_node, |
1382 | arg, halting_phi, | |
1383 | evolution_of_loop, limit); | |
c59dabbe SP |
1384 | if (res == t_true) |
1385 | break; | |
9baba81b SP |
1386 | } |
1387 | ||
1388 | /* If the path crosses this loop-phi, give up. */ | |
c59dabbe | 1389 | if (res == t_true) |
9baba81b SP |
1390 | *evolution_of_loop = chrec_dont_know; |
1391 | ||
1392 | return res; | |
1393 | } | |
1394 | ||
1395 | /* Otherwise, compute the overall effect of the inner loop. */ | |
1396 | ev = compute_overall_effect_of_inner_loop (loop, ev); | |
726a989a RB |
1397 | return follow_ssa_edge_expr (outer_loop, loop_phi_node, ev, halting_phi, |
1398 | evolution_of_loop, limit); | |
9baba81b SP |
1399 | } |
1400 | ||
1401 | /* Follow an SSA edge from a loop-phi-node to itself, constructing a | |
1402 | path that is analyzed on the return walk. */ | |
1403 | ||
c59dabbe | 1404 | static t_bool |
726a989a | 1405 | follow_ssa_edge (struct loop *loop, gimple def, gimple halting_phi, |
c59dabbe | 1406 | tree *evolution_of_loop, int limit) |
9baba81b SP |
1407 | { |
1408 | struct loop *def_loop; | |
b8698a0f | 1409 | |
726a989a | 1410 | if (gimple_nop_p (def)) |
c59dabbe | 1411 | return t_false; |
b8698a0f | 1412 | |
c59dabbe | 1413 | /* Give up if the path is longer than the MAX that we allow. */ |
9e824336 | 1414 | if (limit > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE)) |
c59dabbe | 1415 | return t_dont_know; |
b8698a0f | 1416 | |
9baba81b | 1417 | def_loop = loop_containing_stmt (def); |
b8698a0f | 1418 | |
726a989a | 1419 | switch (gimple_code (def)) |
9baba81b | 1420 | { |
726a989a | 1421 | case GIMPLE_PHI: |
9baba81b SP |
1422 | if (!loop_phi_node_p (def)) |
1423 | /* DEF is a condition-phi-node. Follow the branches, and | |
1424 | record their evolutions. Finally, merge the collected | |
1425 | information and set the approximation to the main | |
1426 | variable. */ | |
b8698a0f | 1427 | return follow_ssa_edge_in_condition_phi |
c59dabbe | 1428 | (loop, def, halting_phi, evolution_of_loop, limit); |
9baba81b SP |
1429 | |
1430 | /* When the analyzed phi is the halting_phi, the | |
1431 | depth-first search is over: we have found a path from | |
1432 | the halting_phi to itself in the loop. */ | |
1433 | if (def == halting_phi) | |
c59dabbe | 1434 | return t_true; |
b8698a0f | 1435 | |
9baba81b | 1436 | /* Otherwise, the evolution of the HALTING_PHI depends |
89dbed81 | 1437 | on the evolution of another loop-phi-node, i.e. the |
9baba81b SP |
1438 | evolution function is a higher degree polynomial. */ |
1439 | if (def_loop == loop) | |
c59dabbe | 1440 | return t_false; |
b8698a0f | 1441 | |
9baba81b SP |
1442 | /* Inner loop. */ |
1443 | if (flow_loop_nested_p (loop, def_loop)) | |
b8698a0f | 1444 | return follow_ssa_edge_inner_loop_phi |
9e824336 | 1445 | (loop, def, halting_phi, evolution_of_loop, limit + 1); |
9baba81b SP |
1446 | |
1447 | /* Outer loop. */ | |
c59dabbe | 1448 | return t_false; |
9baba81b | 1449 | |
726a989a | 1450 | case GIMPLE_ASSIGN: |
b8698a0f | 1451 | return follow_ssa_edge_in_rhs (loop, def, halting_phi, |
c59dabbe | 1452 | evolution_of_loop, limit); |
b8698a0f | 1453 | |
9baba81b SP |
1454 | default: |
1455 | /* At this level of abstraction, the program is just a set | |
726a989a | 1456 | of GIMPLE_ASSIGNs and PHI_NODEs. In principle there is no |
9baba81b | 1457 | other node to be handled. */ |
c59dabbe | 1458 | return t_false; |
9baba81b SP |
1459 | } |
1460 | } | |
1461 | ||
1462 | \f | |
1463 | ||
1464 | /* Given a LOOP_PHI_NODE, this function determines the evolution | |
1465 | function from LOOP_PHI_NODE to LOOP_PHI_NODE in the loop. */ | |
1466 | ||
1467 | static tree | |
b8698a0f | 1468 | analyze_evolution_in_loop (gimple loop_phi_node, |
9baba81b SP |
1469 | tree init_cond) |
1470 | { | |
726a989a | 1471 | int i, n = gimple_phi_num_args (loop_phi_node); |
9baba81b SP |
1472 | tree evolution_function = chrec_not_analyzed_yet; |
1473 | struct loop *loop = loop_containing_stmt (loop_phi_node); | |
1474 | basic_block bb; | |
b8698a0f | 1475 | |
9baba81b SP |
1476 | if (dump_file && (dump_flags & TDF_DETAILS)) |
1477 | { | |
1478 | fprintf (dump_file, "(analyze_evolution_in_loop \n"); | |
1479 | fprintf (dump_file, " (loop_phi_node = "); | |
726a989a | 1480 | print_gimple_stmt (dump_file, loop_phi_node, 0, 0); |
9baba81b SP |
1481 | fprintf (dump_file, ")\n"); |
1482 | } | |
b8698a0f | 1483 | |
726a989a | 1484 | for (i = 0; i < n; i++) |
9baba81b SP |
1485 | { |
1486 | tree arg = PHI_ARG_DEF (loop_phi_node, i); | |
726a989a RB |
1487 | gimple ssa_chain; |
1488 | tree ev_fn; | |
874caa00 | 1489 | t_bool res; |
9baba81b SP |
1490 | |
1491 | /* Select the edges that enter the loop body. */ | |
726a989a | 1492 | bb = gimple_phi_arg_edge (loop_phi_node, i)->src; |
9baba81b SP |
1493 | if (!flow_bb_inside_loop_p (loop, bb)) |
1494 | continue; | |
f29deac9 | 1495 | |
9baba81b SP |
1496 | if (TREE_CODE (arg) == SSA_NAME) |
1497 | { | |
f29deac9 SP |
1498 | bool val = false; |
1499 | ||
9baba81b SP |
1500 | ssa_chain = SSA_NAME_DEF_STMT (arg); |
1501 | ||
1502 | /* Pass in the initial condition to the follow edge function. */ | |
1503 | ev_fn = init_cond; | |
c59dabbe | 1504 | res = follow_ssa_edge (loop, ssa_chain, loop_phi_node, &ev_fn, 0); |
f29deac9 SP |
1505 | |
1506 | /* If ev_fn has no evolution in the inner loop, and the | |
1507 | init_cond is not equal to ev_fn, then we have an | |
1508 | ambiguity between two possible values, as we cannot know | |
1509 | the number of iterations at this point. */ | |
1510 | if (TREE_CODE (ev_fn) != POLYNOMIAL_CHREC | |
1511 | && no_evolution_in_loop_p (ev_fn, loop->num, &val) && val | |
1512 | && !operand_equal_p (init_cond, ev_fn, 0)) | |
1513 | ev_fn = chrec_dont_know; | |
9baba81b SP |
1514 | } |
1515 | else | |
874caa00 | 1516 | res = t_false; |
f29deac9 | 1517 | |
9baba81b SP |
1518 | /* When it is impossible to go back on the same |
1519 | loop_phi_node by following the ssa edges, the | |
89dbed81 | 1520 | evolution is represented by a peeled chrec, i.e. the |
9baba81b | 1521 | first iteration, EV_FN has the value INIT_COND, then |
b8698a0f | 1522 | all the other iterations it has the value of ARG. |
9baba81b | 1523 | For the moment, PEELED_CHREC nodes are not built. */ |
874caa00 | 1524 | if (res != t_true) |
9baba81b | 1525 | ev_fn = chrec_dont_know; |
b8698a0f | 1526 | |
9baba81b | 1527 | /* When there are multiple back edges of the loop (which in fact never |
8c27b7d4 | 1528 | happens currently, but nevertheless), merge their evolutions. */ |
9baba81b SP |
1529 | evolution_function = chrec_merge (evolution_function, ev_fn); |
1530 | } | |
b8698a0f | 1531 | |
9baba81b SP |
1532 | if (dump_file && (dump_flags & TDF_DETAILS)) |
1533 | { | |
1534 | fprintf (dump_file, " (evolution_function = "); | |
1535 | print_generic_expr (dump_file, evolution_function, 0); | |
1536 | fprintf (dump_file, "))\n"); | |
1537 | } | |
b8698a0f | 1538 | |
9baba81b SP |
1539 | return evolution_function; |
1540 | } | |
1541 | ||
1542 | /* Given a loop-phi-node, return the initial conditions of the | |
1543 | variable on entry of the loop. When the CCP has propagated | |
1544 | constants into the loop-phi-node, the initial condition is | |
1545 | instantiated, otherwise the initial condition is kept symbolic. | |
1546 | This analyzer does not analyze the evolution outside the current | |
1547 | loop, and leaves this task to the on-demand tree reconstructor. */ | |
1548 | ||
b8698a0f | 1549 | static tree |
726a989a | 1550 | analyze_initial_condition (gimple loop_phi_node) |
9baba81b | 1551 | { |
726a989a | 1552 | int i, n; |
9baba81b | 1553 | tree init_cond = chrec_not_analyzed_yet; |
726a989a | 1554 | struct loop *loop = loop_containing_stmt (loop_phi_node); |
b8698a0f | 1555 | |
9baba81b SP |
1556 | if (dump_file && (dump_flags & TDF_DETAILS)) |
1557 | { | |
1558 | fprintf (dump_file, "(analyze_initial_condition \n"); | |
1559 | fprintf (dump_file, " (loop_phi_node = \n"); | |
726a989a | 1560 | print_gimple_stmt (dump_file, loop_phi_node, 0, 0); |
9baba81b SP |
1561 | fprintf (dump_file, ")\n"); |
1562 | } | |
b8698a0f | 1563 | |
726a989a RB |
1564 | n = gimple_phi_num_args (loop_phi_node); |
1565 | for (i = 0; i < n; i++) | |
9baba81b SP |
1566 | { |
1567 | tree branch = PHI_ARG_DEF (loop_phi_node, i); | |
726a989a | 1568 | basic_block bb = gimple_phi_arg_edge (loop_phi_node, i)->src; |
b8698a0f | 1569 | |
9baba81b SP |
1570 | /* When the branch is oriented to the loop's body, it does |
1571 | not contribute to the initial condition. */ | |
1572 | if (flow_bb_inside_loop_p (loop, bb)) | |
1573 | continue; | |
1574 | ||
1575 | if (init_cond == chrec_not_analyzed_yet) | |
1576 | { | |
1577 | init_cond = branch; | |
1578 | continue; | |
1579 | } | |
1580 | ||
1581 | if (TREE_CODE (branch) == SSA_NAME) | |
1582 | { | |
1583 | init_cond = chrec_dont_know; | |
1584 | break; | |
1585 | } | |
1586 | ||
1587 | init_cond = chrec_merge (init_cond, branch); | |
1588 | } | |
1589 | ||
1590 | /* Ooops -- a loop without an entry??? */ | |
1591 | if (init_cond == chrec_not_analyzed_yet) | |
1592 | init_cond = chrec_dont_know; | |
1593 | ||
bf1cbdc6 RG |
1594 | /* During early loop unrolling we do not have fully constant propagated IL. |
1595 | Handle degenerate PHIs here to not miss important unrollings. */ | |
1596 | if (TREE_CODE (init_cond) == SSA_NAME) | |
1597 | { | |
1598 | gimple def = SSA_NAME_DEF_STMT (init_cond); | |
1599 | tree res; | |
1600 | if (gimple_code (def) == GIMPLE_PHI | |
1601 | && (res = degenerate_phi_result (def)) != NULL_TREE | |
1602 | /* Only allow invariants here, otherwise we may break | |
1603 | loop-closed SSA form. */ | |
1604 | && is_gimple_min_invariant (res)) | |
1605 | init_cond = res; | |
1606 | } | |
1607 | ||
9baba81b SP |
1608 | if (dump_file && (dump_flags & TDF_DETAILS)) |
1609 | { | |
1610 | fprintf (dump_file, " (init_cond = "); | |
1611 | print_generic_expr (dump_file, init_cond, 0); | |
1612 | fprintf (dump_file, "))\n"); | |
1613 | } | |
b8698a0f | 1614 | |
9baba81b SP |
1615 | return init_cond; |
1616 | } | |
1617 | ||
1618 | /* Analyze the scalar evolution for LOOP_PHI_NODE. */ | |
1619 | ||
b8698a0f | 1620 | static tree |
726a989a | 1621 | interpret_loop_phi (struct loop *loop, gimple loop_phi_node) |
9baba81b SP |
1622 | { |
1623 | tree res; | |
1624 | struct loop *phi_loop = loop_containing_stmt (loop_phi_node); | |
1625 | tree init_cond; | |
b8698a0f | 1626 | |
9baba81b SP |
1627 | if (phi_loop != loop) |
1628 | { | |
1629 | struct loop *subloop; | |
1630 | tree evolution_fn = analyze_scalar_evolution | |
1631 | (phi_loop, PHI_RESULT (loop_phi_node)); | |
1632 | ||
1633 | /* Dive one level deeper. */ | |
9ba025a2 | 1634 | subloop = superloop_at_depth (phi_loop, loop_depth (loop) + 1); |
9baba81b SP |
1635 | |
1636 | /* Interpret the subloop. */ | |
1637 | res = compute_overall_effect_of_inner_loop (subloop, evolution_fn); | |
1638 | return res; | |
1639 | } | |
1640 | ||
1641 | /* Otherwise really interpret the loop phi. */ | |
1642 | init_cond = analyze_initial_condition (loop_phi_node); | |
1643 | res = analyze_evolution_in_loop (loop_phi_node, init_cond); | |
1644 | ||
1645 | return res; | |
1646 | } | |
1647 | ||
1648 | /* This function merges the branches of a condition-phi-node, | |
1649 | contained in the outermost loop, and whose arguments are already | |
1650 | analyzed. */ | |
1651 | ||
1652 | static tree | |
726a989a | 1653 | interpret_condition_phi (struct loop *loop, gimple condition_phi) |
9baba81b | 1654 | { |
726a989a | 1655 | int i, n = gimple_phi_num_args (condition_phi); |
9baba81b | 1656 | tree res = chrec_not_analyzed_yet; |
b8698a0f | 1657 | |
726a989a | 1658 | for (i = 0; i < n; i++) |
9baba81b SP |
1659 | { |
1660 | tree branch_chrec; | |
b8698a0f | 1661 | |
9baba81b SP |
1662 | if (backedge_phi_arg_p (condition_phi, i)) |
1663 | { | |
1664 | res = chrec_dont_know; | |
1665 | break; | |
1666 | } | |
1667 | ||
1668 | branch_chrec = analyze_scalar_evolution | |
1669 | (loop, PHI_ARG_DEF (condition_phi, i)); | |
b8698a0f | 1670 | |
9baba81b SP |
1671 | res = chrec_merge (res, branch_chrec); |
1672 | } | |
1673 | ||
1674 | return res; | |
1675 | } | |
1676 | ||
726a989a | 1677 | /* Interpret the operation RHS1 OP RHS2. If we didn't |
29836d07 | 1678 | analyze this node before, follow the definitions until ending |
726a989a | 1679 | either on an analyzed GIMPLE_ASSIGN, or on a loop-phi-node. On the |
9baba81b SP |
1680 | return path, this function propagates evolutions (ala constant copy |
1681 | propagation). OPND1 is not a GIMPLE expression because we could | |
1682 | analyze the effect of an inner loop: see interpret_loop_phi. */ | |
1683 | ||
1684 | static tree | |
726a989a RB |
1685 | interpret_rhs_expr (struct loop *loop, gimple at_stmt, |
1686 | tree type, tree rhs1, enum tree_code code, tree rhs2) | |
9baba81b | 1687 | { |
726a989a RB |
1688 | tree res, chrec1, chrec2; |
1689 | ||
1690 | if (get_gimple_rhs_class (code) == GIMPLE_SINGLE_RHS) | |
1691 | { | |
1692 | if (is_gimple_min_invariant (rhs1)) | |
1693 | return chrec_convert (type, rhs1, at_stmt); | |
1694 | ||
1695 | if (code == SSA_NAME) | |
1696 | return chrec_convert (type, analyze_scalar_evolution (loop, rhs1), | |
1697 | at_stmt); | |
1e8552eb | 1698 | |
726a989a RB |
1699 | if (code == ASSERT_EXPR) |
1700 | { | |
1701 | rhs1 = ASSERT_EXPR_VAR (rhs1); | |
1702 | return chrec_convert (type, analyze_scalar_evolution (loop, rhs1), | |
1703 | at_stmt); | |
1704 | } | |
1705 | ||
1706 | return chrec_dont_know; | |
1707 | } | |
1e8552eb | 1708 | |
726a989a | 1709 | switch (code) |
9baba81b | 1710 | { |
5be014d5 | 1711 | case POINTER_PLUS_EXPR: |
726a989a RB |
1712 | chrec1 = analyze_scalar_evolution (loop, rhs1); |
1713 | chrec2 = analyze_scalar_evolution (loop, rhs2); | |
1714 | chrec1 = chrec_convert (type, chrec1, at_stmt); | |
1715 | chrec2 = chrec_convert (sizetype, chrec2, at_stmt); | |
1716 | res = chrec_fold_plus (type, chrec1, chrec2); | |
5be014d5 AP |
1717 | break; |
1718 | ||
9baba81b | 1719 | case PLUS_EXPR: |
726a989a RB |
1720 | chrec1 = analyze_scalar_evolution (loop, rhs1); |
1721 | chrec2 = analyze_scalar_evolution (loop, rhs2); | |
1722 | chrec1 = chrec_convert (type, chrec1, at_stmt); | |
1723 | chrec2 = chrec_convert (type, chrec2, at_stmt); | |
1724 | res = chrec_fold_plus (type, chrec1, chrec2); | |
9baba81b | 1725 | break; |
b8698a0f | 1726 | |
9baba81b | 1727 | case MINUS_EXPR: |
726a989a RB |
1728 | chrec1 = analyze_scalar_evolution (loop, rhs1); |
1729 | chrec2 = analyze_scalar_evolution (loop, rhs2); | |
1730 | chrec1 = chrec_convert (type, chrec1, at_stmt); | |
1731 | chrec2 = chrec_convert (type, chrec2, at_stmt); | |
1732 | res = chrec_fold_minus (type, chrec1, chrec2); | |
9baba81b SP |
1733 | break; |
1734 | ||
1735 | case NEGATE_EXPR: | |
726a989a RB |
1736 | chrec1 = analyze_scalar_evolution (loop, rhs1); |
1737 | chrec1 = chrec_convert (type, chrec1, at_stmt); | |
9a75ede0 | 1738 | /* TYPE may be integer, real or complex, so use fold_convert. */ |
726a989a | 1739 | res = chrec_fold_multiply (type, chrec1, |
9a75ede0 | 1740 | fold_convert (type, integer_minus_one_node)); |
9baba81b SP |
1741 | break; |
1742 | ||
418df9d7 JJ |
1743 | case BIT_NOT_EXPR: |
1744 | /* Handle ~X as -1 - X. */ | |
1745 | chrec1 = analyze_scalar_evolution (loop, rhs1); | |
1746 | chrec1 = chrec_convert (type, chrec1, at_stmt); | |
1747 | res = chrec_fold_minus (type, | |
1748 | fold_convert (type, integer_minus_one_node), | |
1749 | chrec1); | |
1750 | break; | |
1751 | ||
9baba81b | 1752 | case MULT_EXPR: |
726a989a RB |
1753 | chrec1 = analyze_scalar_evolution (loop, rhs1); |
1754 | chrec2 = analyze_scalar_evolution (loop, rhs2); | |
1755 | chrec1 = chrec_convert (type, chrec1, at_stmt); | |
1756 | chrec2 = chrec_convert (type, chrec2, at_stmt); | |
1757 | res = chrec_fold_multiply (type, chrec1, chrec2); | |
0bca51f0 | 1758 | break; |
b8698a0f | 1759 | |
1043771b | 1760 | CASE_CONVERT: |
726a989a RB |
1761 | chrec1 = analyze_scalar_evolution (loop, rhs1); |
1762 | res = chrec_convert (type, chrec1, at_stmt); | |
9baba81b | 1763 | break; |
b8698a0f | 1764 | |
9baba81b SP |
1765 | default: |
1766 | res = chrec_dont_know; | |
1767 | break; | |
1768 | } | |
b8698a0f | 1769 | |
9baba81b SP |
1770 | return res; |
1771 | } | |
1772 | ||
726a989a RB |
1773 | /* Interpret the expression EXPR. */ |
1774 | ||
1775 | static tree | |
1776 | interpret_expr (struct loop *loop, gimple at_stmt, tree expr) | |
1777 | { | |
1778 | enum tree_code code; | |
1779 | tree type = TREE_TYPE (expr), op0, op1; | |
1780 | ||
1781 | if (automatically_generated_chrec_p (expr)) | |
1782 | return expr; | |
1783 | ||
1784 | if (TREE_CODE (expr) == POLYNOMIAL_CHREC) | |
1785 | return chrec_dont_know; | |
1786 | ||
1787 | extract_ops_from_tree (expr, &code, &op0, &op1); | |
1788 | ||
1789 | return interpret_rhs_expr (loop, at_stmt, type, | |
1790 | op0, code, op1); | |
1791 | } | |
1792 | ||
1793 | /* Interpret the rhs of the assignment STMT. */ | |
1794 | ||
1795 | static tree | |
1796 | interpret_gimple_assign (struct loop *loop, gimple stmt) | |
1797 | { | |
1798 | tree type = TREE_TYPE (gimple_assign_lhs (stmt)); | |
1799 | enum tree_code code = gimple_assign_rhs_code (stmt); | |
1800 | ||
1801 | return interpret_rhs_expr (loop, stmt, type, | |
1802 | gimple_assign_rhs1 (stmt), code, | |
1803 | gimple_assign_rhs2 (stmt)); | |
1804 | } | |
1805 | ||
9baba81b SP |
1806 | \f |
1807 | ||
b8698a0f | 1808 | /* This section contains all the entry points: |
9baba81b SP |
1809 | - number_of_iterations_in_loop, |
1810 | - analyze_scalar_evolution, | |
1811 | - instantiate_parameters. | |
1812 | */ | |
1813 | ||
1814 | /* Compute and return the evolution function in WRTO_LOOP, the nearest | |
1815 | common ancestor of DEF_LOOP and USE_LOOP. */ | |
1816 | ||
b8698a0f L |
1817 | static tree |
1818 | compute_scalar_evolution_in_loop (struct loop *wrto_loop, | |
1819 | struct loop *def_loop, | |
9baba81b SP |
1820 | tree ev) |
1821 | { | |
1822 | tree res; | |
1823 | if (def_loop == wrto_loop) | |
1824 | return ev; | |
1825 | ||
9ba025a2 | 1826 | def_loop = superloop_at_depth (def_loop, loop_depth (wrto_loop) + 1); |
9baba81b SP |
1827 | res = compute_overall_effect_of_inner_loop (def_loop, ev); |
1828 | ||
1829 | return analyze_scalar_evolution_1 (wrto_loop, res, chrec_not_analyzed_yet); | |
1830 | } | |
1831 | ||
1832 | /* Helper recursive function. */ | |
1833 | ||
1834 | static tree | |
1835 | analyze_scalar_evolution_1 (struct loop *loop, tree var, tree res) | |
1836 | { | |
726a989a RB |
1837 | tree type = TREE_TYPE (var); |
1838 | gimple def; | |
9baba81b SP |
1839 | basic_block bb; |
1840 | struct loop *def_loop; | |
1841 | ||
42d375ed | 1842 | if (loop == NULL || TREE_CODE (type) == VECTOR_TYPE) |
9baba81b SP |
1843 | return chrec_dont_know; |
1844 | ||
1845 | if (TREE_CODE (var) != SSA_NAME) | |
726a989a | 1846 | return interpret_expr (loop, NULL, var); |
9baba81b SP |
1847 | |
1848 | def = SSA_NAME_DEF_STMT (var); | |
726a989a | 1849 | bb = gimple_bb (def); |
9baba81b SP |
1850 | def_loop = bb ? bb->loop_father : NULL; |
1851 | ||
1852 | if (bb == NULL | |
1853 | || !flow_bb_inside_loop_p (loop, bb)) | |
1854 | { | |
1855 | /* Keep the symbolic form. */ | |
1856 | res = var; | |
1857 | goto set_and_end; | |
1858 | } | |
1859 | ||
1860 | if (res != chrec_not_analyzed_yet) | |
1861 | { | |
1862 | if (loop != bb->loop_father) | |
b8698a0f | 1863 | res = compute_scalar_evolution_in_loop |
9baba81b SP |
1864 | (find_common_loop (loop, bb->loop_father), bb->loop_father, res); |
1865 | ||
1866 | goto set_and_end; | |
1867 | } | |
1868 | ||
1869 | if (loop != def_loop) | |
1870 | { | |
1871 | res = analyze_scalar_evolution_1 (def_loop, var, chrec_not_analyzed_yet); | |
1872 | res = compute_scalar_evolution_in_loop (loop, def_loop, res); | |
1873 | ||
1874 | goto set_and_end; | |
1875 | } | |
1876 | ||
726a989a | 1877 | switch (gimple_code (def)) |
9baba81b | 1878 | { |
726a989a RB |
1879 | case GIMPLE_ASSIGN: |
1880 | res = interpret_gimple_assign (loop, def); | |
9baba81b SP |
1881 | break; |
1882 | ||
726a989a | 1883 | case GIMPLE_PHI: |
9baba81b SP |
1884 | if (loop_phi_node_p (def)) |
1885 | res = interpret_loop_phi (loop, def); | |
1886 | else | |
1887 | res = interpret_condition_phi (loop, def); | |
1888 | break; | |
1889 | ||
1890 | default: | |
1891 | res = chrec_dont_know; | |
1892 | break; | |
1893 | } | |
1894 | ||
1895 | set_and_end: | |
1896 | ||
1897 | /* Keep the symbolic form. */ | |
1898 | if (res == chrec_dont_know) | |
1899 | res = var; | |
1900 | ||
1901 | if (loop == def_loop) | |
a213b219 | 1902 | set_scalar_evolution (block_before_loop (loop), var, res); |
9baba81b SP |
1903 | |
1904 | return res; | |
1905 | } | |
1906 | ||
52bdd655 SP |
1907 | /* Analyzes and returns the scalar evolution of the ssa_name VAR in |
1908 | LOOP. LOOP is the loop in which the variable is used. | |
b8698a0f | 1909 | |
9baba81b SP |
1910 | Example of use: having a pointer VAR to a SSA_NAME node, STMT a |
1911 | pointer to the statement that uses this variable, in order to | |
1912 | determine the evolution function of the variable, use the following | |
1913 | calls: | |
b8698a0f | 1914 | |
52bdd655 SP |
1915 | loop_p loop = loop_containing_stmt (stmt); |
1916 | tree chrec_with_symbols = analyze_scalar_evolution (loop, var); | |
3f227a8c | 1917 | tree chrec_instantiated = instantiate_parameters (loop, chrec_with_symbols); |
9baba81b SP |
1918 | */ |
1919 | ||
b8698a0f | 1920 | tree |
9baba81b SP |
1921 | analyze_scalar_evolution (struct loop *loop, tree var) |
1922 | { | |
1923 | tree res; | |
1924 | ||
1925 | if (dump_file && (dump_flags & TDF_DETAILS)) | |
1926 | { | |
1927 | fprintf (dump_file, "(analyze_scalar_evolution \n"); | |
1928 | fprintf (dump_file, " (loop_nb = %d)\n", loop->num); | |
1929 | fprintf (dump_file, " (scalar = "); | |
1930 | print_generic_expr (dump_file, var, 0); | |
1931 | fprintf (dump_file, ")\n"); | |
1932 | } | |
1933 | ||
a213b219 SP |
1934 | res = get_scalar_evolution (block_before_loop (loop), var); |
1935 | res = analyze_scalar_evolution_1 (loop, var, res); | |
9baba81b | 1936 | |
9baba81b SP |
1937 | if (dump_file && (dump_flags & TDF_DETAILS)) |
1938 | fprintf (dump_file, ")\n"); | |
1939 | ||
1940 | return res; | |
1941 | } | |
1942 | ||
1943 | /* Analyze scalar evolution of use of VERSION in USE_LOOP with respect to | |
f017bf5e | 1944 | WRTO_LOOP (which should be a superloop of USE_LOOP) |
a6f778b2 ZD |
1945 | |
1946 | FOLDED_CASTS is set to true if resolve_mixers used | |
1947 | chrec_convert_aggressive (TODO -- not really, we are way too conservative | |
b8698a0f L |
1948 | at the moment in order to keep things simple). |
1949 | ||
f017bf5e ZD |
1950 | To illustrate the meaning of USE_LOOP and WRTO_LOOP, consider the following |
1951 | example: | |
1952 | ||
1953 | for (i = 0; i < 100; i++) -- loop 1 | |
1954 | { | |
1955 | for (j = 0; j < 100; j++) -- loop 2 | |
1956 | { | |
1957 | k1 = i; | |
1958 | k2 = j; | |
1959 | ||
1960 | use2 (k1, k2); | |
1961 | ||
1962 | for (t = 0; t < 100; t++) -- loop 3 | |
1963 | use3 (k1, k2); | |
1964 | ||
1965 | } | |
1966 | use1 (k1, k2); | |
1967 | } | |
1968 | ||
1969 | Both k1 and k2 are invariants in loop3, thus | |
1970 | analyze_scalar_evolution_in_loop (loop3, loop3, k1) = k1 | |
1971 | analyze_scalar_evolution_in_loop (loop3, loop3, k2) = k2 | |
1972 | ||
1973 | As they are invariant, it does not matter whether we consider their | |
1974 | usage in loop 3 or loop 2, hence | |
1975 | analyze_scalar_evolution_in_loop (loop2, loop3, k1) = | |
1976 | analyze_scalar_evolution_in_loop (loop2, loop2, k1) = i | |
1977 | analyze_scalar_evolution_in_loop (loop2, loop3, k2) = | |
1978 | analyze_scalar_evolution_in_loop (loop2, loop2, k2) = [0,+,1]_2 | |
1979 | ||
1980 | Similarly for their evolutions with respect to loop 1. The values of K2 | |
1981 | in the use in loop 2 vary independently on loop 1, thus we cannot express | |
1982 | the evolution with respect to loop 1: | |
1983 | analyze_scalar_evolution_in_loop (loop1, loop3, k1) = | |
1984 | analyze_scalar_evolution_in_loop (loop1, loop2, k1) = [0,+,1]_1 | |
1985 | analyze_scalar_evolution_in_loop (loop1, loop3, k2) = | |
1986 | analyze_scalar_evolution_in_loop (loop1, loop2, k2) = dont_know | |
1987 | ||
1988 | The value of k2 in the use in loop 1 is known, though: | |
1989 | analyze_scalar_evolution_in_loop (loop1, loop1, k1) = [0,+,1]_1 | |
1990 | analyze_scalar_evolution_in_loop (loop1, loop1, k2) = 100 | |
1991 | */ | |
9baba81b SP |
1992 | |
1993 | static tree | |
1994 | analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop, | |
a6f778b2 | 1995 | tree version, bool *folded_casts) |
9baba81b SP |
1996 | { |
1997 | bool val = false; | |
a6f778b2 | 1998 | tree ev = version, tmp; |
9baba81b | 1999 | |
b8698a0f | 2000 | /* We cannot just do |
f017bf5e ZD |
2001 | |
2002 | tmp = analyze_scalar_evolution (use_loop, version); | |
2003 | ev = resolve_mixers (wrto_loop, tmp); | |
2004 | ||
2005 | as resolve_mixers would query the scalar evolution with respect to | |
2006 | wrto_loop. For example, in the situation described in the function | |
2007 | comment, suppose that wrto_loop = loop1, use_loop = loop3 and | |
2008 | version = k2. Then | |
2009 | ||
2010 | analyze_scalar_evolution (use_loop, version) = k2 | |
2011 | ||
2012 | and resolve_mixers (loop1, k2) finds that the value of k2 in loop 1 | |
2013 | is 100, which is a wrong result, since we are interested in the | |
2014 | value in loop 3. | |
2015 | ||
2016 | Instead, we need to proceed from use_loop to wrto_loop loop by loop, | |
2017 | each time checking that there is no evolution in the inner loop. */ | |
2018 | ||
a6f778b2 ZD |
2019 | if (folded_casts) |
2020 | *folded_casts = false; | |
9baba81b SP |
2021 | while (1) |
2022 | { | |
a6f778b2 ZD |
2023 | tmp = analyze_scalar_evolution (use_loop, ev); |
2024 | ev = resolve_mixers (use_loop, tmp); | |
2025 | ||
2026 | if (folded_casts && tmp != ev) | |
2027 | *folded_casts = true; | |
9baba81b SP |
2028 | |
2029 | if (use_loop == wrto_loop) | |
2030 | return ev; | |
2031 | ||
2032 | /* If the value of the use changes in the inner loop, we cannot express | |
2033 | its value in the outer loop (we might try to return interval chrec, | |
2034 | but we do not have a user for it anyway) */ | |
2035 | if (!no_evolution_in_loop_p (ev, use_loop->num, &val) | |
2036 | || !val) | |
2037 | return chrec_dont_know; | |
2038 | ||
9ba025a2 | 2039 | use_loop = loop_outer (use_loop); |
9baba81b SP |
2040 | } |
2041 | } | |
2042 | ||
a213b219 SP |
2043 | /* Returns from CACHE the value for VERSION instantiated below |
2044 | INSTANTIATED_BELOW block. */ | |
eb0bc7af ZD |
2045 | |
2046 | static tree | |
a213b219 SP |
2047 | get_instantiated_value (htab_t cache, basic_block instantiated_below, |
2048 | tree version) | |
eb0bc7af ZD |
2049 | { |
2050 | struct scev_info_str *info, pattern; | |
b8698a0f | 2051 | |
eb0bc7af | 2052 | pattern.var = version; |
a213b219 | 2053 | pattern.instantiated_below = instantiated_below; |
858904db | 2054 | info = (struct scev_info_str *) htab_find (cache, &pattern); |
eb0bc7af ZD |
2055 | |
2056 | if (info) | |
2057 | return info->chrec; | |
2058 | else | |
2059 | return NULL_TREE; | |
2060 | } | |
2061 | ||
a213b219 SP |
2062 | /* Sets in CACHE the value of VERSION instantiated below basic block |
2063 | INSTANTIATED_BELOW to VAL. */ | |
eb0bc7af ZD |
2064 | |
2065 | static void | |
a213b219 SP |
2066 | set_instantiated_value (htab_t cache, basic_block instantiated_below, |
2067 | tree version, tree val) | |
eb0bc7af ZD |
2068 | { |
2069 | struct scev_info_str *info, pattern; | |
2070 | PTR *slot; | |
b8698a0f | 2071 | |
eb0bc7af | 2072 | pattern.var = version; |
0bfdfbf6 | 2073 | pattern.instantiated_below = instantiated_below; |
eb0bc7af ZD |
2074 | slot = htab_find_slot (cache, &pattern, INSERT); |
2075 | ||
cceb1885 | 2076 | if (!*slot) |
a213b219 | 2077 | *slot = new_scev_info_str (instantiated_below, version); |
cceb1885 | 2078 | info = (struct scev_info_str *) *slot; |
eb0bc7af ZD |
2079 | info->chrec = val; |
2080 | } | |
2081 | ||
18aed06a SP |
2082 | /* Return the closed_loop_phi node for VAR. If there is none, return |
2083 | NULL_TREE. */ | |
2084 | ||
2085 | static tree | |
2086 | loop_closed_phi_def (tree var) | |
2087 | { | |
2088 | struct loop *loop; | |
2089 | edge exit; | |
726a989a RB |
2090 | gimple phi; |
2091 | gimple_stmt_iterator psi; | |
18aed06a SP |
2092 | |
2093 | if (var == NULL_TREE | |
2094 | || TREE_CODE (var) != SSA_NAME) | |
2095 | return NULL_TREE; | |
2096 | ||
2097 | loop = loop_containing_stmt (SSA_NAME_DEF_STMT (var)); | |
ac8f6c69 | 2098 | exit = single_exit (loop); |
18aed06a SP |
2099 | if (!exit) |
2100 | return NULL_TREE; | |
2101 | ||
726a989a RB |
2102 | for (psi = gsi_start_phis (exit->dest); !gsi_end_p (psi); gsi_next (&psi)) |
2103 | { | |
2104 | phi = gsi_stmt (psi); | |
2105 | if (PHI_ARG_DEF_FROM_EDGE (phi, exit) == var) | |
2106 | return PHI_RESULT (phi); | |
2107 | } | |
18aed06a SP |
2108 | |
2109 | return NULL_TREE; | |
2110 | } | |
2111 | ||
9e5dc77f | 2112 | static tree instantiate_scev_r (basic_block, struct loop *, tree, bool, |
320f5a78 SP |
2113 | htab_t, int); |
2114 | ||
2115 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW | |
2116 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
2117 | ||
2495a183 | 2118 | CHREC is an SSA_NAME to be instantiated. |
320f5a78 SP |
2119 | |
2120 | CACHE is the cache of already instantiated values. | |
2121 | ||
2122 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2123 | may wrap in signed/pointer type are folded, as long as the value of | |
2124 | the chrec is preserved. | |
2125 | ||
2126 | SIZE_EXPR is used for computing the size of the expression to be | |
2127 | instantiated, and to stop if it exceeds some limit. */ | |
2128 | ||
2129 | static tree | |
2495a183 SP |
2130 | instantiate_scev_name (basic_block instantiate_below, |
2131 | struct loop *evolution_loop, tree chrec, | |
2132 | bool fold_conversions, htab_t cache, int size_expr) | |
320f5a78 | 2133 | { |
2495a183 SP |
2134 | tree res; |
2135 | struct loop *def_loop; | |
2136 | basic_block def_bb = gimple_bb (SSA_NAME_DEF_STMT (chrec)); | |
20179b0d | 2137 | |
2495a183 SP |
2138 | /* A parameter (or loop invariant and we do not want to include |
2139 | evolutions in outer loops), nothing to do. */ | |
2140 | if (!def_bb | |
2141 | || loop_depth (def_bb->loop_father) == 0 | |
2142 | || dominated_by_p (CDI_DOMINATORS, instantiate_below, def_bb)) | |
2143 | return chrec; | |
20179b0d | 2144 | |
2495a183 SP |
2145 | /* We cache the value of instantiated variable to avoid exponential |
2146 | time complexity due to reevaluations. We also store the convenient | |
2147 | value in the cache in order to prevent infinite recursion -- we do | |
2148 | not want to instantiate the SSA_NAME if it is in a mixer | |
2149 | structure. This is used for avoiding the instantiation of | |
2150 | recursively defined functions, such as: | |
320f5a78 | 2151 | |
2495a183 | 2152 | | a_2 -> {0, +, 1, +, a_2}_1 */ |
20179b0d | 2153 | |
2495a183 SP |
2154 | res = get_instantiated_value (cache, instantiate_below, chrec); |
2155 | if (res) | |
2156 | return res; | |
20179b0d | 2157 | |
2495a183 | 2158 | res = chrec_dont_know; |
20179b0d | 2159 | set_instantiated_value (cache, instantiate_below, chrec, res); |
320f5a78 | 2160 | |
2495a183 SP |
2161 | def_loop = find_common_loop (evolution_loop, def_bb->loop_father); |
2162 | ||
320f5a78 SP |
2163 | /* If the analysis yields a parametric chrec, instantiate the |
2164 | result again. */ | |
2165 | res = analyze_scalar_evolution (def_loop, chrec); | |
2166 | ||
2167 | /* Don't instantiate loop-closed-ssa phi nodes. */ | |
2168 | if (TREE_CODE (res) == SSA_NAME | |
2169 | && (loop_containing_stmt (SSA_NAME_DEF_STMT (res)) == NULL | |
2170 | || (loop_depth (loop_containing_stmt (SSA_NAME_DEF_STMT (res))) | |
2171 | > loop_depth (def_loop)))) | |
2172 | { | |
2173 | if (res == chrec) | |
2174 | res = loop_closed_phi_def (chrec); | |
2175 | else | |
2176 | res = chrec; | |
2177 | ||
2178 | if (res == NULL_TREE | |
2179 | || !dominated_by_p (CDI_DOMINATORS, instantiate_below, | |
2180 | gimple_bb (SSA_NAME_DEF_STMT (res)))) | |
2181 | res = chrec_dont_know; | |
2182 | } | |
2183 | ||
2184 | else if (res != chrec_dont_know) | |
9e5dc77f | 2185 | res = instantiate_scev_r (instantiate_below, evolution_loop, res, |
320f5a78 SP |
2186 | fold_conversions, cache, size_expr); |
2187 | ||
2188 | /* Store the correct value to the cache. */ | |
2189 | set_instantiated_value (cache, instantiate_below, chrec, res); | |
2190 | return res; | |
20179b0d | 2191 | |
320f5a78 SP |
2192 | } |
2193 | ||
ec6636eb SP |
2194 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW |
2195 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
2196 | ||
2197 | CHREC is a polynomial chain of recurrence to be instantiated. | |
2198 | ||
2199 | CACHE is the cache of already instantiated values. | |
2200 | ||
2201 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2202 | may wrap in signed/pointer type are folded, as long as the value of | |
2203 | the chrec is preserved. | |
2204 | ||
2205 | SIZE_EXPR is used for computing the size of the expression to be | |
2206 | instantiated, and to stop if it exceeds some limit. */ | |
2207 | ||
2208 | static tree | |
2209 | instantiate_scev_poly (basic_block instantiate_below, | |
2210 | struct loop *evolution_loop, tree chrec, | |
2211 | bool fold_conversions, htab_t cache, int size_expr) | |
2212 | { | |
2213 | tree op1; | |
9e5dc77f | 2214 | tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, |
ec6636eb SP |
2215 | CHREC_LEFT (chrec), fold_conversions, cache, |
2216 | size_expr); | |
2217 | if (op0 == chrec_dont_know) | |
2218 | return chrec_dont_know; | |
2219 | ||
9e5dc77f | 2220 | op1 = instantiate_scev_r (instantiate_below, evolution_loop, |
ec6636eb SP |
2221 | CHREC_RIGHT (chrec), fold_conversions, cache, |
2222 | size_expr); | |
2223 | if (op1 == chrec_dont_know) | |
2224 | return chrec_dont_know; | |
2225 | ||
2226 | if (CHREC_LEFT (chrec) != op0 | |
2227 | || CHREC_RIGHT (chrec) != op1) | |
2228 | { | |
4bf4e169 SP |
2229 | unsigned var = CHREC_VARIABLE (chrec); |
2230 | ||
2231 | /* When the instantiated stride or base has an evolution in an | |
2232 | innermost loop, return chrec_dont_know, as this is not a | |
2233 | valid SCEV representation. In the reduced testcase for | |
2234 | PR40281 we would have {0, +, {1, +, 1}_2}_1 that has no | |
2235 | meaning. */ | |
2236 | if ((tree_is_chrec (op0) && CHREC_VARIABLE (op0) > var) | |
2237 | || (tree_is_chrec (op1) && CHREC_VARIABLE (op1) > var)) | |
2238 | return chrec_dont_know; | |
2239 | ||
ec6636eb | 2240 | op1 = chrec_convert_rhs (chrec_type (op0), op1, NULL); |
4bf4e169 | 2241 | chrec = build_polynomial_chrec (var, op0, op1); |
ec6636eb | 2242 | } |
4bf4e169 | 2243 | |
ec6636eb SP |
2244 | return chrec; |
2245 | } | |
2246 | ||
15fda317 SP |
2247 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW |
2248 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
2249 | ||
ffa34f4b | 2250 | "C0 CODE C1" is a binary expression of type TYPE to be instantiated. |
15fda317 SP |
2251 | |
2252 | CACHE is the cache of already instantiated values. | |
2253 | ||
2254 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2255 | may wrap in signed/pointer type are folded, as long as the value of | |
2256 | the chrec is preserved. | |
2257 | ||
2258 | SIZE_EXPR is used for computing the size of the expression to be | |
2259 | instantiated, and to stop if it exceeds some limit. */ | |
2260 | ||
2261 | static tree | |
2262 | instantiate_scev_binary (basic_block instantiate_below, | |
ffa34f4b SP |
2263 | struct loop *evolution_loop, tree chrec, enum tree_code code, |
2264 | tree type, tree c0, tree c1, | |
15fda317 SP |
2265 | bool fold_conversions, htab_t cache, int size_expr) |
2266 | { | |
2267 | tree op1; | |
9e5dc77f | 2268 | tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, |
ffa34f4b | 2269 | c0, fold_conversions, cache, |
15fda317 SP |
2270 | size_expr); |
2271 | if (op0 == chrec_dont_know) | |
2272 | return chrec_dont_know; | |
2273 | ||
9e5dc77f | 2274 | op1 = instantiate_scev_r (instantiate_below, evolution_loop, |
ffa34f4b | 2275 | c1, fold_conversions, cache, |
15fda317 SP |
2276 | size_expr); |
2277 | if (op1 == chrec_dont_know) | |
2278 | return chrec_dont_know; | |
2279 | ||
ffa34f4b SP |
2280 | if (c0 != op0 |
2281 | || c1 != op1) | |
15fda317 | 2282 | { |
15fda317 SP |
2283 | op0 = chrec_convert (type, op0, NULL); |
2284 | op1 = chrec_convert_rhs (type, op1, NULL); | |
2285 | ||
ffa34f4b | 2286 | switch (code) |
15fda317 SP |
2287 | { |
2288 | case POINTER_PLUS_EXPR: | |
2289 | case PLUS_EXPR: | |
2290 | return chrec_fold_plus (type, op0, op1); | |
2291 | ||
2292 | case MINUS_EXPR: | |
2293 | return chrec_fold_minus (type, op0, op1); | |
2294 | ||
2295 | case MULT_EXPR: | |
2296 | return chrec_fold_multiply (type, op0, op1); | |
2297 | ||
2298 | default: | |
2299 | gcc_unreachable (); | |
2300 | } | |
2301 | } | |
2302 | ||
ffa34f4b | 2303 | return chrec ? chrec : fold_build2 (code, type, c0, c1); |
15fda317 SP |
2304 | } |
2305 | ||
a213b219 | 2306 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW |
9c382ce9 SP |
2307 | and EVOLUTION_LOOP, that were left under a symbolic form. |
2308 | ||
2309 | "CHREC" that stands for a convert expression "(TYPE) OP" is to be | |
2310 | instantiated. | |
2311 | ||
2312 | CACHE is the cache of already instantiated values. | |
2313 | ||
2314 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2315 | may wrap in signed/pointer type are folded, as long as the value of | |
2316 | the chrec is preserved. | |
2317 | ||
2318 | SIZE_EXPR is used for computing the size of the expression to be | |
2319 | instantiated, and to stop if it exceeds some limit. */ | |
2320 | ||
2321 | static tree | |
2322 | instantiate_scev_convert (basic_block instantiate_below, | |
2323 | struct loop *evolution_loop, tree chrec, | |
2324 | tree type, tree op, | |
2325 | bool fold_conversions, htab_t cache, int size_expr) | |
2326 | { | |
9e5dc77f | 2327 | tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, op, |
9c382ce9 SP |
2328 | fold_conversions, cache, size_expr); |
2329 | ||
2330 | if (op0 == chrec_dont_know) | |
2331 | return chrec_dont_know; | |
2332 | ||
2333 | if (fold_conversions) | |
2334 | { | |
2335 | tree tmp = chrec_convert_aggressive (type, op0); | |
2336 | if (tmp) | |
2337 | return tmp; | |
2338 | } | |
2339 | ||
2340 | if (chrec && op0 == op) | |
2341 | return chrec; | |
2342 | ||
2343 | /* If we used chrec_convert_aggressive, we can no longer assume that | |
2344 | signed chrecs do not overflow, as chrec_convert does, so avoid | |
2345 | calling it in that case. */ | |
2346 | if (fold_conversions) | |
2347 | return fold_convert (type, op0); | |
2348 | ||
2349 | return chrec_convert (type, op0, NULL); | |
2350 | } | |
2351 | ||
7ec0665d SP |
2352 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW |
2353 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
2354 | ||
4b9d48a1 | 2355 | CHREC is a BIT_NOT_EXPR or a NEGATE_EXPR expression to be instantiated. |
7ec0665d | 2356 | Handle ~X as -1 - X. |
4b9d48a1 | 2357 | Handle -X as -1 * X. |
7ec0665d SP |
2358 | |
2359 | CACHE is the cache of already instantiated values. | |
2360 | ||
2361 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2362 | may wrap in signed/pointer type are folded, as long as the value of | |
2363 | the chrec is preserved. | |
2364 | ||
2365 | SIZE_EXPR is used for computing the size of the expression to be | |
2366 | instantiated, and to stop if it exceeds some limit. */ | |
2367 | ||
2368 | static tree | |
4b9d48a1 SP |
2369 | instantiate_scev_not (basic_block instantiate_below, |
2370 | struct loop *evolution_loop, tree chrec, | |
20179b0d | 2371 | enum tree_code code, tree type, tree op, |
4b9d48a1 | 2372 | bool fold_conversions, htab_t cache, int size_expr) |
7ec0665d | 2373 | { |
20179b0d | 2374 | tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, op, |
7ec0665d | 2375 | fold_conversions, cache, size_expr); |
20179b0d | 2376 | |
7ec0665d SP |
2377 | if (op0 == chrec_dont_know) |
2378 | return chrec_dont_know; | |
2379 | ||
20179b0d | 2380 | if (op != op0) |
7ec0665d SP |
2381 | { |
2382 | op0 = chrec_convert (type, op0, NULL); | |
4b9d48a1 | 2383 | |
20179b0d | 2384 | switch (code) |
4b9d48a1 SP |
2385 | { |
2386 | case BIT_NOT_EXPR: | |
2387 | return chrec_fold_minus | |
2388 | (type, fold_convert (type, integer_minus_one_node), op0); | |
2389 | ||
2390 | case NEGATE_EXPR: | |
2391 | return chrec_fold_multiply | |
2392 | (type, fold_convert (type, integer_minus_one_node), op0); | |
2393 | ||
2394 | default: | |
2395 | gcc_unreachable (); | |
2396 | } | |
7ec0665d | 2397 | } |
4b9d48a1 | 2398 | |
20179b0d | 2399 | return chrec ? chrec : fold_build1 (code, type, op0); |
7ec0665d SP |
2400 | } |
2401 | ||
d814176c SP |
2402 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW |
2403 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
2404 | ||
2405 | CHREC is an expression with 3 operands to be instantiated. | |
2406 | ||
2407 | CACHE is the cache of already instantiated values. | |
2408 | ||
2409 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2410 | may wrap in signed/pointer type are folded, as long as the value of | |
2411 | the chrec is preserved. | |
2412 | ||
2413 | SIZE_EXPR is used for computing the size of the expression to be | |
2414 | instantiated, and to stop if it exceeds some limit. */ | |
2415 | ||
2416 | static tree | |
2417 | instantiate_scev_3 (basic_block instantiate_below, | |
2418 | struct loop *evolution_loop, tree chrec, | |
2419 | bool fold_conversions, htab_t cache, int size_expr) | |
2420 | { | |
2421 | tree op1, op2; | |
9e5dc77f | 2422 | tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, |
d814176c SP |
2423 | TREE_OPERAND (chrec, 0), |
2424 | fold_conversions, cache, size_expr); | |
2425 | if (op0 == chrec_dont_know) | |
2426 | return chrec_dont_know; | |
2427 | ||
9e5dc77f | 2428 | op1 = instantiate_scev_r (instantiate_below, evolution_loop, |
d814176c SP |
2429 | TREE_OPERAND (chrec, 1), |
2430 | fold_conversions, cache, size_expr); | |
2431 | if (op1 == chrec_dont_know) | |
2432 | return chrec_dont_know; | |
2433 | ||
9e5dc77f | 2434 | op2 = instantiate_scev_r (instantiate_below, evolution_loop, |
d814176c SP |
2435 | TREE_OPERAND (chrec, 2), |
2436 | fold_conversions, cache, size_expr); | |
2437 | if (op2 == chrec_dont_know) | |
2438 | return chrec_dont_know; | |
2439 | ||
2440 | if (op0 == TREE_OPERAND (chrec, 0) | |
2441 | && op1 == TREE_OPERAND (chrec, 1) | |
2442 | && op2 == TREE_OPERAND (chrec, 2)) | |
2443 | return chrec; | |
2444 | ||
2445 | return fold_build3 (TREE_CODE (chrec), | |
2446 | TREE_TYPE (chrec), op0, op1, op2); | |
2447 | } | |
2448 | ||
9c382ce9 SP |
2449 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW |
2450 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
5b78fc3e | 2451 | |
9e5dc77f SP |
2452 | CHREC is an expression with 2 operands to be instantiated. |
2453 | ||
2454 | CACHE is the cache of already instantiated values. | |
2455 | ||
2456 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2457 | may wrap in signed/pointer type are folded, as long as the value of | |
2458 | the chrec is preserved. | |
2459 | ||
2460 | SIZE_EXPR is used for computing the size of the expression to be | |
2461 | instantiated, and to stop if it exceeds some limit. */ | |
2462 | ||
2463 | static tree | |
2464 | instantiate_scev_2 (basic_block instantiate_below, | |
2465 | struct loop *evolution_loop, tree chrec, | |
2466 | bool fold_conversions, htab_t cache, int size_expr) | |
2467 | { | |
2468 | tree op1; | |
2469 | tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, | |
2470 | TREE_OPERAND (chrec, 0), | |
2471 | fold_conversions, cache, size_expr); | |
2472 | if (op0 == chrec_dont_know) | |
2473 | return chrec_dont_know; | |
2474 | ||
2475 | op1 = instantiate_scev_r (instantiate_below, evolution_loop, | |
2476 | TREE_OPERAND (chrec, 1), | |
2477 | fold_conversions, cache, size_expr); | |
2478 | if (op1 == chrec_dont_know) | |
2479 | return chrec_dont_know; | |
2480 | ||
2481 | if (op0 == TREE_OPERAND (chrec, 0) | |
2482 | && op1 == TREE_OPERAND (chrec, 1)) | |
2483 | return chrec; | |
2484 | ||
2485 | return fold_build2 (TREE_CODE (chrec), TREE_TYPE (chrec), op0, op1); | |
2486 | } | |
2487 | ||
2488 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW | |
2489 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
2490 | ||
2491 | CHREC is an expression with 2 operands to be instantiated. | |
5b78fc3e JS |
2492 | |
2493 | CACHE is the cache of already instantiated values. | |
2494 | ||
2495 | FOLD_CONVERSIONS should be set to true when the conversions that | |
2496 | may wrap in signed/pointer type are folded, as long as the value of | |
2497 | the chrec is preserved. | |
2498 | ||
3f227a8c JS |
2499 | SIZE_EXPR is used for computing the size of the expression to be |
2500 | instantiated, and to stop if it exceeds some limit. */ | |
9c382ce9 | 2501 | |
9baba81b | 2502 | static tree |
a213b219 | 2503 | instantiate_scev_1 (basic_block instantiate_below, |
3f227a8c | 2504 | struct loop *evolution_loop, tree chrec, |
5b78fc3e | 2505 | bool fold_conversions, htab_t cache, int size_expr) |
9baba81b | 2506 | { |
9e5dc77f SP |
2507 | tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, |
2508 | TREE_OPERAND (chrec, 0), | |
2509 | fold_conversions, cache, size_expr); | |
2510 | ||
2511 | if (op0 == chrec_dont_know) | |
2512 | return chrec_dont_know; | |
2513 | ||
2514 | if (op0 == TREE_OPERAND (chrec, 0)) | |
2515 | return chrec; | |
2516 | ||
2517 | return fold_build1 (TREE_CODE (chrec), TREE_TYPE (chrec), op0); | |
2518 | } | |
2519 | ||
2520 | /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW | |
2521 | and EVOLUTION_LOOP, that were left under a symbolic form. | |
2522 | ||
2523 | CHREC is the scalar evolution to instantiate. | |
2524 | ||
2525 | CACHE is the cache of already instantiated values. | |
2282a0e6 | 2526 | |
9e5dc77f SP |
2527 | FOLD_CONVERSIONS should be set to true when the conversions that |
2528 | may wrap in signed/pointer type are folded, as long as the value of | |
2529 | the chrec is preserved. | |
2530 | ||
2531 | SIZE_EXPR is used for computing the size of the expression to be | |
2532 | instantiated, and to stop if it exceeds some limit. */ | |
2533 | ||
2534 | static tree | |
2535 | instantiate_scev_r (basic_block instantiate_below, | |
2536 | struct loop *evolution_loop, tree chrec, | |
2537 | bool fold_conversions, htab_t cache, int size_expr) | |
2538 | { | |
47ae9e4c SP |
2539 | /* Give up if the expression is larger than the MAX that we allow. */ |
2540 | if (size_expr++ > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE)) | |
2541 | return chrec_dont_know; | |
2542 | ||
d7770457 SP |
2543 | if (automatically_generated_chrec_p (chrec) |
2544 | || is_gimple_min_invariant (chrec)) | |
9baba81b SP |
2545 | return chrec; |
2546 | ||
2547 | switch (TREE_CODE (chrec)) | |
2548 | { | |
2549 | case SSA_NAME: | |
320f5a78 SP |
2550 | return instantiate_scev_name (instantiate_below, evolution_loop, chrec, |
2551 | fold_conversions, cache, size_expr); | |
9baba81b SP |
2552 | |
2553 | case POLYNOMIAL_CHREC: | |
ec6636eb SP |
2554 | return instantiate_scev_poly (instantiate_below, evolution_loop, chrec, |
2555 | fold_conversions, cache, size_expr); | |
9baba81b | 2556 | |
5be014d5 | 2557 | case POINTER_PLUS_EXPR: |
9baba81b | 2558 | case PLUS_EXPR: |
9baba81b | 2559 | case MINUS_EXPR: |
9baba81b | 2560 | case MULT_EXPR: |
15fda317 | 2561 | return instantiate_scev_binary (instantiate_below, evolution_loop, chrec, |
ffa34f4b SP |
2562 | TREE_CODE (chrec), chrec_type (chrec), |
2563 | TREE_OPERAND (chrec, 0), | |
2564 | TREE_OPERAND (chrec, 1), | |
15fda317 | 2565 | fold_conversions, cache, size_expr); |
9baba81b | 2566 | |
1043771b | 2567 | CASE_CONVERT: |
9c382ce9 SP |
2568 | return instantiate_scev_convert (instantiate_below, evolution_loop, chrec, |
2569 | TREE_TYPE (chrec), TREE_OPERAND (chrec, 0), | |
2570 | fold_conversions, cache, size_expr); | |
9baba81b | 2571 | |
4b9d48a1 | 2572 | case NEGATE_EXPR: |
418df9d7 | 2573 | case BIT_NOT_EXPR: |
4b9d48a1 | 2574 | return instantiate_scev_not (instantiate_below, evolution_loop, chrec, |
20179b0d SP |
2575 | TREE_CODE (chrec), TREE_TYPE (chrec), |
2576 | TREE_OPERAND (chrec, 0), | |
4b9d48a1 | 2577 | fold_conversions, cache, size_expr); |
418df9d7 | 2578 | |
9baba81b SP |
2579 | case SCEV_NOT_KNOWN: |
2580 | return chrec_dont_know; | |
2581 | ||
2582 | case SCEV_KNOWN: | |
2583 | return chrec_known; | |
15fda317 | 2584 | |
9baba81b SP |
2585 | default: |
2586 | break; | |
2587 | } | |
2588 | ||
0dfb0dc6 SP |
2589 | if (VL_EXP_CLASS_P (chrec)) |
2590 | return chrec_dont_know; | |
2591 | ||
9baba81b SP |
2592 | switch (TREE_CODE_LENGTH (TREE_CODE (chrec))) |
2593 | { | |
2594 | case 3: | |
d814176c SP |
2595 | return instantiate_scev_3 (instantiate_below, evolution_loop, chrec, |
2596 | fold_conversions, cache, size_expr); | |
9baba81b SP |
2597 | |
2598 | case 2: | |
9e5dc77f SP |
2599 | return instantiate_scev_2 (instantiate_below, evolution_loop, chrec, |
2600 | fold_conversions, cache, size_expr); | |
7ec0665d | 2601 | |
9baba81b | 2602 | case 1: |
9e5dc77f SP |
2603 | return instantiate_scev_1 (instantiate_below, evolution_loop, chrec, |
2604 | fold_conversions, cache, size_expr); | |
9baba81b SP |
2605 | |
2606 | case 0: | |
2607 | return chrec; | |
2608 | ||
2609 | default: | |
2610 | break; | |
2611 | } | |
2612 | ||
2613 | /* Too complicated to handle. */ | |
2614 | return chrec_dont_know; | |
2615 | } | |
e9eb809d ZD |
2616 | |
2617 | /* Analyze all the parameters of the chrec that were left under a | |
a213b219 SP |
2618 | symbolic form. INSTANTIATE_BELOW is the basic block that stops the |
2619 | recursive instantiation of parameters: a parameter is a variable | |
2620 | that is defined in a basic block that dominates INSTANTIATE_BELOW or | |
2621 | a function parameter. */ | |
e9eb809d ZD |
2622 | |
2623 | tree | |
a213b219 | 2624 | instantiate_scev (basic_block instantiate_below, struct loop *evolution_loop, |
3f227a8c | 2625 | tree chrec) |
e9eb809d | 2626 | { |
9baba81b | 2627 | tree res; |
eb0bc7af | 2628 | htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info); |
9baba81b SP |
2629 | |
2630 | if (dump_file && (dump_flags & TDF_DETAILS)) | |
2631 | { | |
3f227a8c | 2632 | fprintf (dump_file, "(instantiate_scev \n"); |
a213b219 | 2633 | fprintf (dump_file, " (instantiate_below = %d)\n", instantiate_below->index); |
3f227a8c | 2634 | fprintf (dump_file, " (evolution_loop = %d)\n", evolution_loop->num); |
9baba81b SP |
2635 | fprintf (dump_file, " (chrec = "); |
2636 | print_generic_expr (dump_file, chrec, 0); | |
2637 | fprintf (dump_file, ")\n"); | |
2638 | } | |
b8698a0f | 2639 | |
9e5dc77f | 2640 | res = instantiate_scev_r (instantiate_below, evolution_loop, chrec, false, |
5b78fc3e | 2641 | cache, 0); |
9baba81b SP |
2642 | |
2643 | if (dump_file && (dump_flags & TDF_DETAILS)) | |
2644 | { | |
2645 | fprintf (dump_file, " (res = "); | |
2646 | print_generic_expr (dump_file, res, 0); | |
2647 | fprintf (dump_file, "))\n"); | |
2648 | } | |
eb0bc7af ZD |
2649 | |
2650 | htab_delete (cache); | |
b8698a0f | 2651 | |
9baba81b SP |
2652 | return res; |
2653 | } | |
2654 | ||
2655 | /* Similar to instantiate_parameters, but does not introduce the | |
2282a0e6 ZD |
2656 | evolutions in outer loops for LOOP invariants in CHREC, and does not |
2657 | care about causing overflows, as long as they do not affect value | |
2658 | of an expression. */ | |
9baba81b | 2659 | |
3cb960c7 | 2660 | tree |
9baba81b SP |
2661 | resolve_mixers (struct loop *loop, tree chrec) |
2662 | { | |
eb0bc7af | 2663 | htab_t cache = htab_create (10, hash_scev_info, eq_scev_info, del_scev_info); |
9e5dc77f | 2664 | tree ret = instantiate_scev_r (block_before_loop (loop), loop, chrec, true, |
a213b219 | 2665 | cache, 0); |
eb0bc7af ZD |
2666 | htab_delete (cache); |
2667 | return ret; | |
9baba81b SP |
2668 | } |
2669 | ||
b8698a0f | 2670 | /* Entry point for the analysis of the number of iterations pass. |
9baba81b SP |
2671 | This function tries to safely approximate the number of iterations |
2672 | the loop will run. When this property is not decidable at compile | |
2673 | time, the result is chrec_dont_know. Otherwise the result is | |
2674 | a scalar or a symbolic parameter. | |
b8698a0f | 2675 | |
9baba81b | 2676 | Example of analysis: suppose that the loop has an exit condition: |
b8698a0f | 2677 | |
9baba81b | 2678 | "if (b > 49) goto end_loop;" |
b8698a0f | 2679 | |
9baba81b SP |
2680 | and that in a previous analysis we have determined that the |
2681 | variable 'b' has an evolution function: | |
b8698a0f L |
2682 | |
2683 | "EF = {23, +, 5}_2". | |
2684 | ||
9baba81b SP |
2685 | When we evaluate the function at the point 5, i.e. the value of the |
2686 | variable 'b' after 5 iterations in the loop, we have EF (5) = 48, | |
2687 | and EF (6) = 53. In this case the value of 'b' on exit is '53' and | |
2688 | the loop body has been executed 6 times. */ | |
2689 | ||
b8698a0f | 2690 | tree |
a14865db | 2691 | number_of_latch_executions (struct loop *loop) |
9baba81b SP |
2692 | { |
2693 | tree res, type; | |
2694 | edge exit; | |
2695 | struct tree_niter_desc niter_desc; | |
2696 | ||
2697 | /* Determine whether the number_of_iterations_in_loop has already | |
2698 | been computed. */ | |
2699 | res = loop->nb_iterations; | |
2700 | if (res) | |
2701 | return res; | |
2702 | res = chrec_dont_know; | |
2703 | ||
2704 | if (dump_file && (dump_flags & TDF_DETAILS)) | |
2705 | fprintf (dump_file, "(number_of_iterations_in_loop\n"); | |
b8698a0f | 2706 | |
ac8f6c69 | 2707 | exit = single_exit (loop); |
82b85a85 | 2708 | if (!exit) |
9baba81b | 2709 | goto end; |
9baba81b | 2710 | |
f9cc1a70 | 2711 | if (!number_of_iterations_exit (loop, exit, &niter_desc, false)) |
9baba81b SP |
2712 | goto end; |
2713 | ||
2714 | type = TREE_TYPE (niter_desc.niter); | |
2715 | if (integer_nonzerop (niter_desc.may_be_zero)) | |
5212068f | 2716 | res = build_int_cst (type, 0); |
9baba81b SP |
2717 | else if (integer_zerop (niter_desc.may_be_zero)) |
2718 | res = niter_desc.niter; | |
2719 | else | |
2720 | res = chrec_dont_know; | |
2721 | ||
2722 | end: | |
2723 | return set_nb_iterations_in_loop (loop, res); | |
2724 | } | |
2725 | ||
a14865db ZD |
2726 | /* Returns the number of executions of the exit condition of LOOP, |
2727 | i.e., the number by one higher than number_of_latch_executions. | |
fa10beec | 2728 | Note that unlike number_of_latch_executions, this number does |
a14865db ZD |
2729 | not necessarily fit in the unsigned variant of the type of |
2730 | the control variable -- if the number of iterations is a constant, | |
2731 | we return chrec_dont_know if adding one to number_of_latch_executions | |
2732 | overflows; however, in case the number of iterations is symbolic | |
2733 | expression, the caller is responsible for dealing with this | |
2734 | the possible overflow. */ | |
2735 | ||
b8698a0f | 2736 | tree |
a14865db ZD |
2737 | number_of_exit_cond_executions (struct loop *loop) |
2738 | { | |
2739 | tree ret = number_of_latch_executions (loop); | |
2740 | tree type = chrec_type (ret); | |
2741 | ||
2742 | if (chrec_contains_undetermined (ret)) | |
2743 | return ret; | |
2744 | ||
2745 | ret = chrec_fold_plus (type, ret, build_int_cst (type, 1)); | |
2746 | if (TREE_CODE (ret) == INTEGER_CST | |
2747 | && TREE_OVERFLOW (ret)) | |
2748 | return chrec_dont_know; | |
2749 | ||
2750 | return ret; | |
2751 | } | |
2752 | ||
9baba81b SP |
2753 | /* One of the drivers for testing the scalar evolutions analysis. |
2754 | This function computes the number of iterations for all the loops | |
2755 | from the EXIT_CONDITIONS array. */ | |
2756 | ||
b8698a0f | 2757 | static void |
726a989a | 2758 | number_of_iterations_for_all_loops (VEC(gimple,heap) **exit_conditions) |
9baba81b SP |
2759 | { |
2760 | unsigned int i; | |
2761 | unsigned nb_chrec_dont_know_loops = 0; | |
2762 | unsigned nb_static_loops = 0; | |
726a989a | 2763 | gimple cond; |
b8698a0f | 2764 | |
726a989a | 2765 | for (i = 0; VEC_iterate (gimple, *exit_conditions, i, cond); i++) |
9baba81b | 2766 | { |
a14865db | 2767 | tree res = number_of_latch_executions (loop_containing_stmt (cond)); |
9baba81b SP |
2768 | if (chrec_contains_undetermined (res)) |
2769 | nb_chrec_dont_know_loops++; | |
2770 | else | |
2771 | nb_static_loops++; | |
2772 | } | |
b8698a0f | 2773 | |
9baba81b SP |
2774 | if (dump_file) |
2775 | { | |
2776 | fprintf (dump_file, "\n(\n"); | |
2777 | fprintf (dump_file, "-----------------------------------------\n"); | |
2778 | fprintf (dump_file, "%d\tnb_chrec_dont_know_loops\n", nb_chrec_dont_know_loops); | |
2779 | fprintf (dump_file, "%d\tnb_static_loops\n", nb_static_loops); | |
42fd6772 | 2780 | fprintf (dump_file, "%d\tnb_total_loops\n", number_of_loops ()); |
9baba81b SP |
2781 | fprintf (dump_file, "-----------------------------------------\n"); |
2782 | fprintf (dump_file, ")\n\n"); | |
b8698a0f | 2783 | |
0c8efed8 | 2784 | print_loops (dump_file, 3); |
9baba81b SP |
2785 | } |
2786 | } | |
2787 | ||
2788 | \f | |
2789 | ||
2790 | /* Counters for the stats. */ | |
2791 | ||
b8698a0f | 2792 | struct chrec_stats |
9baba81b SP |
2793 | { |
2794 | unsigned nb_chrecs; | |
2795 | unsigned nb_affine; | |
2796 | unsigned nb_affine_multivar; | |
2797 | unsigned nb_higher_poly; | |
2798 | unsigned nb_chrec_dont_know; | |
2799 | unsigned nb_undetermined; | |
2800 | }; | |
2801 | ||
2802 | /* Reset the counters. */ | |
2803 | ||
2804 | static inline void | |
2805 | reset_chrecs_counters (struct chrec_stats *stats) | |
2806 | { | |
2807 | stats->nb_chrecs = 0; | |
2808 | stats->nb_affine = 0; | |
2809 | stats->nb_affine_multivar = 0; | |
2810 | stats->nb_higher_poly = 0; | |
2811 | stats->nb_chrec_dont_know = 0; | |
2812 | stats->nb_undetermined = 0; | |
2813 | } | |
2814 | ||
2815 | /* Dump the contents of a CHREC_STATS structure. */ | |
2816 | ||
2817 | static void | |
2818 | dump_chrecs_stats (FILE *file, struct chrec_stats *stats) | |
2819 | { | |
2820 | fprintf (file, "\n(\n"); | |
2821 | fprintf (file, "-----------------------------------------\n"); | |
2822 | fprintf (file, "%d\taffine univariate chrecs\n", stats->nb_affine); | |
2823 | fprintf (file, "%d\taffine multivariate chrecs\n", stats->nb_affine_multivar); | |
b8698a0f | 2824 | fprintf (file, "%d\tdegree greater than 2 polynomials\n", |
9baba81b SP |
2825 | stats->nb_higher_poly); |
2826 | fprintf (file, "%d\tchrec_dont_know chrecs\n", stats->nb_chrec_dont_know); | |
2827 | fprintf (file, "-----------------------------------------\n"); | |
2828 | fprintf (file, "%d\ttotal chrecs\n", stats->nb_chrecs); | |
b8698a0f | 2829 | fprintf (file, "%d\twith undetermined coefficients\n", |
9baba81b SP |
2830 | stats->nb_undetermined); |
2831 | fprintf (file, "-----------------------------------------\n"); | |
b8698a0f | 2832 | fprintf (file, "%d\tchrecs in the scev database\n", |
9baba81b SP |
2833 | (int) htab_elements (scalar_evolution_info)); |
2834 | fprintf (file, "%d\tsets in the scev database\n", nb_set_scev); | |
2835 | fprintf (file, "%d\tgets in the scev database\n", nb_get_scev); | |
2836 | fprintf (file, "-----------------------------------------\n"); | |
2837 | fprintf (file, ")\n\n"); | |
2838 | } | |
2839 | ||
2840 | /* Gather statistics about CHREC. */ | |
2841 | ||
2842 | static void | |
2843 | gather_chrec_stats (tree chrec, struct chrec_stats *stats) | |
2844 | { | |
2845 | if (dump_file && (dump_flags & TDF_STATS)) | |
2846 | { | |
2847 | fprintf (dump_file, "(classify_chrec "); | |
2848 | print_generic_expr (dump_file, chrec, 0); | |
2849 | fprintf (dump_file, "\n"); | |
2850 | } | |
b8698a0f | 2851 | |
9baba81b | 2852 | stats->nb_chrecs++; |
b8698a0f | 2853 | |
9baba81b SP |
2854 | if (chrec == NULL_TREE) |
2855 | { | |
2856 | stats->nb_undetermined++; | |
2857 | return; | |
2858 | } | |
b8698a0f | 2859 | |
9baba81b SP |
2860 | switch (TREE_CODE (chrec)) |
2861 | { | |
2862 | case POLYNOMIAL_CHREC: | |
2863 | if (evolution_function_is_affine_p (chrec)) | |
2864 | { | |
2865 | if (dump_file && (dump_flags & TDF_STATS)) | |
2866 | fprintf (dump_file, " affine_univariate\n"); | |
2867 | stats->nb_affine++; | |
2868 | } | |
a50411de | 2869 | else if (evolution_function_is_affine_multivariate_p (chrec, 0)) |
9baba81b SP |
2870 | { |
2871 | if (dump_file && (dump_flags & TDF_STATS)) | |
2872 | fprintf (dump_file, " affine_multivariate\n"); | |
2873 | stats->nb_affine_multivar++; | |
2874 | } | |
2875 | else | |
2876 | { | |
2877 | if (dump_file && (dump_flags & TDF_STATS)) | |
2878 | fprintf (dump_file, " higher_degree_polynomial\n"); | |
2879 | stats->nb_higher_poly++; | |
2880 | } | |
b8698a0f | 2881 | |
9baba81b SP |
2882 | break; |
2883 | ||
2884 | default: | |
2885 | break; | |
2886 | } | |
b8698a0f | 2887 | |
9baba81b SP |
2888 | if (chrec_contains_undetermined (chrec)) |
2889 | { | |
2890 | if (dump_file && (dump_flags & TDF_STATS)) | |
2891 | fprintf (dump_file, " undetermined\n"); | |
2892 | stats->nb_undetermined++; | |
2893 | } | |
b8698a0f | 2894 | |
9baba81b SP |
2895 | if (dump_file && (dump_flags & TDF_STATS)) |
2896 | fprintf (dump_file, ")\n"); | |
2897 | } | |
2898 | ||
2899 | /* One of the drivers for testing the scalar evolutions analysis. | |
2900 | This function analyzes the scalar evolution of all the scalars | |
2901 | defined as loop phi nodes in one of the loops from the | |
b8698a0f L |
2902 | EXIT_CONDITIONS array. |
2903 | ||
9baba81b SP |
2904 | TODO Optimization: A loop is in canonical form if it contains only |
2905 | a single scalar loop phi node. All the other scalars that have an | |
2906 | evolution in the loop are rewritten in function of this single | |
2907 | index. This allows the parallelization of the loop. */ | |
2908 | ||
b8698a0f | 2909 | static void |
726a989a | 2910 | analyze_scalar_evolution_for_all_loop_phi_nodes (VEC(gimple,heap) **exit_conditions) |
9baba81b SP |
2911 | { |
2912 | unsigned int i; | |
2913 | struct chrec_stats stats; | |
726a989a RB |
2914 | gimple cond, phi; |
2915 | gimple_stmt_iterator psi; | |
b8698a0f | 2916 | |
9baba81b | 2917 | reset_chrecs_counters (&stats); |
b8698a0f | 2918 | |
726a989a | 2919 | for (i = 0; VEC_iterate (gimple, *exit_conditions, i, cond); i++) |
9baba81b SP |
2920 | { |
2921 | struct loop *loop; | |
2922 | basic_block bb; | |
726a989a | 2923 | tree chrec; |
b8698a0f | 2924 | |
5310bac6 | 2925 | loop = loop_containing_stmt (cond); |
9baba81b | 2926 | bb = loop->header; |
b8698a0f | 2927 | |
726a989a RB |
2928 | for (psi = gsi_start_phis (bb); !gsi_end_p (psi); gsi_next (&psi)) |
2929 | { | |
2930 | phi = gsi_stmt (psi); | |
2931 | if (is_gimple_reg (PHI_RESULT (phi))) | |
2932 | { | |
b8698a0f L |
2933 | chrec = instantiate_parameters |
2934 | (loop, | |
726a989a | 2935 | analyze_scalar_evolution (loop, PHI_RESULT (phi))); |
b8698a0f | 2936 | |
726a989a RB |
2937 | if (dump_file && (dump_flags & TDF_STATS)) |
2938 | gather_chrec_stats (chrec, &stats); | |
2939 | } | |
2940 | } | |
9baba81b | 2941 | } |
b8698a0f | 2942 | |
9baba81b SP |
2943 | if (dump_file && (dump_flags & TDF_STATS)) |
2944 | dump_chrecs_stats (dump_file, &stats); | |
2945 | } | |
2946 | ||
2947 | /* Callback for htab_traverse, gathers information on chrecs in the | |
2948 | hashtable. */ | |
2949 | ||
2950 | static int | |
2951 | gather_stats_on_scev_database_1 (void **slot, void *stats) | |
2952 | { | |
cceb1885 | 2953 | struct scev_info_str *entry = (struct scev_info_str *) *slot; |
9baba81b | 2954 | |
cceb1885 | 2955 | gather_chrec_stats (entry->chrec, (struct chrec_stats *) stats); |
9baba81b SP |
2956 | |
2957 | return 1; | |
2958 | } | |
2959 | ||
2960 | /* Classify the chrecs of the whole database. */ | |
2961 | ||
b8698a0f | 2962 | void |
9baba81b SP |
2963 | gather_stats_on_scev_database (void) |
2964 | { | |
2965 | struct chrec_stats stats; | |
b8698a0f | 2966 | |
9baba81b SP |
2967 | if (!dump_file) |
2968 | return; | |
b8698a0f | 2969 | |
9baba81b | 2970 | reset_chrecs_counters (&stats); |
b8698a0f | 2971 | |
9baba81b SP |
2972 | htab_traverse (scalar_evolution_info, gather_stats_on_scev_database_1, |
2973 | &stats); | |
2974 | ||
2975 | dump_chrecs_stats (dump_file, &stats); | |
2976 | } | |
2977 | ||
2978 | \f | |
2979 | ||
2980 | /* Initializer. */ | |
2981 | ||
2982 | static void | |
2983 | initialize_scalar_evolutions_analyzer (void) | |
2984 | { | |
2985 | /* The elements below are unique. */ | |
2986 | if (chrec_dont_know == NULL_TREE) | |
2987 | { | |
2988 | chrec_not_analyzed_yet = NULL_TREE; | |
2989 | chrec_dont_know = make_node (SCEV_NOT_KNOWN); | |
2990 | chrec_known = make_node (SCEV_KNOWN); | |
d5ab5675 ZD |
2991 | TREE_TYPE (chrec_dont_know) = void_type_node; |
2992 | TREE_TYPE (chrec_known) = void_type_node; | |
9baba81b SP |
2993 | } |
2994 | } | |
2995 | ||
2996 | /* Initialize the analysis of scalar evolutions for LOOPS. */ | |
2997 | ||
2998 | void | |
d73be268 | 2999 | scev_initialize (void) |
9baba81b | 3000 | { |
42fd6772 ZD |
3001 | loop_iterator li; |
3002 | struct loop *loop; | |
9baba81b | 3003 | |
9e2f83a5 ZD |
3004 | scalar_evolution_info = htab_create_alloc (100, |
3005 | hash_scev_info, | |
3006 | eq_scev_info, | |
3007 | del_scev_info, | |
3008 | ggc_calloc, | |
3009 | ggc_free); | |
b8698a0f | 3010 | |
9baba81b SP |
3011 | initialize_scalar_evolutions_analyzer (); |
3012 | ||
42fd6772 ZD |
3013 | FOR_EACH_LOOP (li, loop, 0) |
3014 | { | |
3015 | loop->nb_iterations = NULL_TREE; | |
3016 | } | |
9baba81b SP |
3017 | } |
3018 | ||
3019 | /* Cleans up the information cached by the scalar evolutions analysis. */ | |
3020 | ||
3021 | void | |
3022 | scev_reset (void) | |
3023 | { | |
42fd6772 | 3024 | loop_iterator li; |
9baba81b SP |
3025 | struct loop *loop; |
3026 | ||
3027 | if (!scalar_evolution_info || !current_loops) | |
3028 | return; | |
3029 | ||
39f8a3b0 | 3030 | htab_empty (scalar_evolution_info); |
42fd6772 | 3031 | FOR_EACH_LOOP (li, loop, 0) |
9baba81b | 3032 | { |
42fd6772 | 3033 | loop->nb_iterations = NULL_TREE; |
9baba81b | 3034 | } |
e9eb809d ZD |
3035 | } |
3036 | ||
f017bf5e ZD |
3037 | /* Checks whether use of OP in USE_LOOP behaves as a simple affine iv with |
3038 | respect to WRTO_LOOP and returns its base and step in IV if possible | |
3039 | (see analyze_scalar_evolution_in_loop for more details on USE_LOOP | |
3040 | and WRTO_LOOP). If ALLOW_NONCONSTANT_STEP is true, we want step to be | |
3041 | invariant in LOOP. Otherwise we require it to be an integer constant. | |
b8698a0f | 3042 | |
f017bf5e ZD |
3043 | IV->no_overflow is set to true if we are sure the iv cannot overflow (e.g. |
3044 | because it is computed in signed arithmetics). Consequently, adding an | |
3045 | induction variable | |
b8698a0f | 3046 | |
f017bf5e ZD |
3047 | for (i = IV->base; ; i += IV->step) |
3048 | ||
3049 | is only safe if IV->no_overflow is false, or TYPE_OVERFLOW_UNDEFINED is | |
3050 | false for the type of the induction variable, or you can prove that i does | |
3051 | not wrap by some other argument. Otherwise, this might introduce undefined | |
3052 | behavior, and | |
b8698a0f | 3053 | |
f017bf5e ZD |
3054 | for (i = iv->base; ; i = (type) ((unsigned type) i + (unsigned type) iv->step)) |
3055 | ||
3056 | must be used instead. */ | |
e9eb809d ZD |
3057 | |
3058 | bool | |
f017bf5e ZD |
3059 | simple_iv (struct loop *wrto_loop, struct loop *use_loop, tree op, |
3060 | affine_iv *iv, bool allow_nonconstant_step) | |
e9eb809d | 3061 | { |
9baba81b | 3062 | tree type, ev; |
a6f778b2 | 3063 | bool folded_casts; |
9baba81b | 3064 | |
a6f778b2 ZD |
3065 | iv->base = NULL_TREE; |
3066 | iv->step = NULL_TREE; | |
3067 | iv->no_overflow = false; | |
9baba81b SP |
3068 | |
3069 | type = TREE_TYPE (op); | |
3070 | if (TREE_CODE (type) != INTEGER_TYPE | |
3071 | && TREE_CODE (type) != POINTER_TYPE) | |
3072 | return false; | |
3073 | ||
f017bf5e | 3074 | ev = analyze_scalar_evolution_in_loop (wrto_loop, use_loop, op, |
a6f778b2 | 3075 | &folded_casts); |
f017bf5e ZD |
3076 | if (chrec_contains_undetermined (ev) |
3077 | || chrec_contains_symbols_defined_in_loop (ev, wrto_loop->num)) | |
9baba81b SP |
3078 | return false; |
3079 | ||
f017bf5e | 3080 | if (tree_does_not_contain_chrecs (ev)) |
9baba81b | 3081 | { |
a6f778b2 | 3082 | iv->base = ev; |
6e42ce54 | 3083 | iv->step = build_int_cst (TREE_TYPE (ev), 0); |
a6f778b2 | 3084 | iv->no_overflow = true; |
9baba81b SP |
3085 | return true; |
3086 | } | |
3087 | ||
3088 | if (TREE_CODE (ev) != POLYNOMIAL_CHREC | |
f017bf5e | 3089 | || CHREC_VARIABLE (ev) != (unsigned) wrto_loop->num) |
9baba81b SP |
3090 | return false; |
3091 | ||
a6f778b2 | 3092 | iv->step = CHREC_RIGHT (ev); |
f017bf5e ZD |
3093 | if ((!allow_nonconstant_step && TREE_CODE (iv->step) != INTEGER_CST) |
3094 | || tree_contains_chrecs (iv->step, NULL)) | |
9baba81b | 3095 | return false; |
9be872b7 | 3096 | |
a6f778b2 | 3097 | iv->base = CHREC_LEFT (ev); |
f017bf5e | 3098 | if (tree_contains_chrecs (iv->base, NULL)) |
9baba81b SP |
3099 | return false; |
3100 | ||
eeef0e45 ILT |
3101 | iv->no_overflow = !folded_casts && TYPE_OVERFLOW_UNDEFINED (type); |
3102 | ||
9baba81b SP |
3103 | return true; |
3104 | } | |
3105 | ||
3106 | /* Runs the analysis of scalar evolutions. */ | |
3107 | ||
3108 | void | |
3109 | scev_analysis (void) | |
3110 | { | |
726a989a | 3111 | VEC(gimple,heap) *exit_conditions; |
b8698a0f | 3112 | |
726a989a | 3113 | exit_conditions = VEC_alloc (gimple, heap, 37); |
d73be268 | 3114 | select_loops_exit_conditions (&exit_conditions); |
9baba81b SP |
3115 | |
3116 | if (dump_file && (dump_flags & TDF_STATS)) | |
5310bac6 | 3117 | analyze_scalar_evolution_for_all_loop_phi_nodes (&exit_conditions); |
b8698a0f | 3118 | |
5310bac6 | 3119 | number_of_iterations_for_all_loops (&exit_conditions); |
726a989a | 3120 | VEC_free (gimple, heap, exit_conditions); |
e9eb809d | 3121 | } |
9baba81b SP |
3122 | |
3123 | /* Finalize the scalar evolution analysis. */ | |
3124 | ||
3125 | void | |
3126 | scev_finalize (void) | |
3127 | { | |
d51157de ZD |
3128 | if (!scalar_evolution_info) |
3129 | return; | |
9baba81b | 3130 | htab_delete (scalar_evolution_info); |
c7b852c8 | 3131 | scalar_evolution_info = NULL; |
9baba81b SP |
3132 | } |
3133 | ||
771f882e ZD |
3134 | /* Returns true if the expression EXPR is considered to be too expensive |
3135 | for scev_const_prop. */ | |
3136 | ||
3137 | bool | |
3138 | expression_expensive_p (tree expr) | |
3139 | { | |
3140 | enum tree_code code; | |
3141 | ||
3142 | if (is_gimple_val (expr)) | |
3143 | return false; | |
3144 | ||
3145 | code = TREE_CODE (expr); | |
3146 | if (code == TRUNC_DIV_EXPR | |
3147 | || code == CEIL_DIV_EXPR | |
3148 | || code == FLOOR_DIV_EXPR | |
3149 | || code == ROUND_DIV_EXPR | |
3150 | || code == TRUNC_MOD_EXPR | |
3151 | || code == CEIL_MOD_EXPR | |
3152 | || code == FLOOR_MOD_EXPR | |
3153 | || code == ROUND_MOD_EXPR | |
3154 | || code == EXACT_DIV_EXPR) | |
3155 | { | |
3156 | /* Division by power of two is usually cheap, so we allow it. | |
3157 | Forbid anything else. */ | |
3158 | if (!integer_pow2p (TREE_OPERAND (expr, 1))) | |
3159 | return true; | |
3160 | } | |
3161 | ||
3162 | switch (TREE_CODE_CLASS (code)) | |
3163 | { | |
3164 | case tcc_binary: | |
3165 | case tcc_comparison: | |
3166 | if (expression_expensive_p (TREE_OPERAND (expr, 1))) | |
3167 | return true; | |
3168 | ||
3169 | /* Fallthru. */ | |
3170 | case tcc_unary: | |
3171 | return expression_expensive_p (TREE_OPERAND (expr, 0)); | |
3172 | ||
3173 | default: | |
3174 | return true; | |
3175 | } | |
3176 | } | |
3177 | ||
684aaf29 | 3178 | /* Replace ssa names for that scev can prove they are constant by the |
3ac01fde ZD |
3179 | appropriate constants. Also perform final value replacement in loops, |
3180 | in case the replacement expressions are cheap. | |
b8698a0f | 3181 | |
684aaf29 ZD |
3182 | We only consider SSA names defined by phi nodes; rest is left to the |
3183 | ordinary constant propagation pass. */ | |
3184 | ||
c2924966 | 3185 | unsigned int |
684aaf29 ZD |
3186 | scev_const_prop (void) |
3187 | { | |
3188 | basic_block bb; | |
726a989a RB |
3189 | tree name, type, ev; |
3190 | gimple phi, ass; | |
3ac01fde | 3191 | struct loop *loop, *ex_loop; |
684aaf29 | 3192 | bitmap ssa_names_to_remove = NULL; |
3ac01fde | 3193 | unsigned i; |
42fd6772 | 3194 | loop_iterator li; |
726a989a | 3195 | gimple_stmt_iterator psi; |
684aaf29 | 3196 | |
d51157de | 3197 | if (number_of_loops () <= 1) |
c2924966 | 3198 | return 0; |
684aaf29 ZD |
3199 | |
3200 | FOR_EACH_BB (bb) | |
3201 | { | |
3202 | loop = bb->loop_father; | |
3203 | ||
726a989a | 3204 | for (psi = gsi_start_phis (bb); !gsi_end_p (psi); gsi_next (&psi)) |
684aaf29 | 3205 | { |
726a989a | 3206 | phi = gsi_stmt (psi); |
684aaf29 ZD |
3207 | name = PHI_RESULT (phi); |
3208 | ||
3209 | if (!is_gimple_reg (name)) | |
3210 | continue; | |
3211 | ||
3212 | type = TREE_TYPE (name); | |
3213 | ||
3214 | if (!POINTER_TYPE_P (type) | |
3215 | && !INTEGRAL_TYPE_P (type)) | |
3216 | continue; | |
3217 | ||
3218 | ev = resolve_mixers (loop, analyze_scalar_evolution (loop, name)); | |
3219 | if (!is_gimple_min_invariant (ev) | |
3220 | || !may_propagate_copy (name, ev)) | |
3221 | continue; | |
3222 | ||
3223 | /* Replace the uses of the name. */ | |
18aed06a SP |
3224 | if (name != ev) |
3225 | replace_uses_by (name, ev); | |
684aaf29 ZD |
3226 | |
3227 | if (!ssa_names_to_remove) | |
3228 | ssa_names_to_remove = BITMAP_ALLOC (NULL); | |
3229 | bitmap_set_bit (ssa_names_to_remove, SSA_NAME_VERSION (name)); | |
3230 | } | |
3231 | } | |
3232 | ||
9b3b55a1 DN |
3233 | /* Remove the ssa names that were replaced by constants. We do not |
3234 | remove them directly in the previous cycle, since this | |
3235 | invalidates scev cache. */ | |
684aaf29 ZD |
3236 | if (ssa_names_to_remove) |
3237 | { | |
3238 | bitmap_iterator bi; | |
684aaf29 ZD |
3239 | |
3240 | EXECUTE_IF_SET_IN_BITMAP (ssa_names_to_remove, 0, i, bi) | |
3241 | { | |
726a989a | 3242 | gimple_stmt_iterator psi; |
684aaf29 ZD |
3243 | name = ssa_name (i); |
3244 | phi = SSA_NAME_DEF_STMT (name); | |
3245 | ||
726a989a RB |
3246 | gcc_assert (gimple_code (phi) == GIMPLE_PHI); |
3247 | psi = gsi_for_stmt (phi); | |
3248 | remove_phi_node (&psi, true); | |
684aaf29 ZD |
3249 | } |
3250 | ||
3251 | BITMAP_FREE (ssa_names_to_remove); | |
3252 | scev_reset (); | |
3253 | } | |
3ac01fde ZD |
3254 | |
3255 | /* Now the regular final value replacement. */ | |
42fd6772 | 3256 | FOR_EACH_LOOP (li, loop, LI_FROM_INNERMOST) |
3ac01fde ZD |
3257 | { |
3258 | edge exit; | |
726a989a RB |
3259 | tree def, rslt, niter; |
3260 | gimple_stmt_iterator bsi; | |
3ac01fde | 3261 | |
3ac01fde ZD |
3262 | /* If we do not know exact number of iterations of the loop, we cannot |
3263 | replace the final value. */ | |
ac8f6c69 | 3264 | exit = single_exit (loop); |
a6f778b2 ZD |
3265 | if (!exit) |
3266 | continue; | |
3267 | ||
a14865db | 3268 | niter = number_of_latch_executions (loop); |
b3ce5b6e | 3269 | if (niter == chrec_dont_know) |
3ac01fde | 3270 | continue; |
925196ed ZD |
3271 | |
3272 | /* Ensure that it is possible to insert new statements somewhere. */ | |
3273 | if (!single_pred_p (exit->dest)) | |
3274 | split_loop_exit_edge (exit); | |
726a989a | 3275 | bsi = gsi_after_labels (exit->dest); |
925196ed | 3276 | |
9ba025a2 ZD |
3277 | ex_loop = superloop_at_depth (loop, |
3278 | loop_depth (exit->dest->loop_father) + 1); | |
3ac01fde | 3279 | |
726a989a | 3280 | for (psi = gsi_start_phis (exit->dest); !gsi_end_p (psi); ) |
3ac01fde | 3281 | { |
726a989a | 3282 | phi = gsi_stmt (psi); |
925196ed | 3283 | rslt = PHI_RESULT (phi); |
3ac01fde | 3284 | def = PHI_ARG_DEF_FROM_EDGE (phi, exit); |
925196ed | 3285 | if (!is_gimple_reg (def)) |
726a989a RB |
3286 | { |
3287 | gsi_next (&psi); | |
3288 | continue; | |
3289 | } | |
3ac01fde ZD |
3290 | |
3291 | if (!POINTER_TYPE_P (TREE_TYPE (def)) | |
3292 | && !INTEGRAL_TYPE_P (TREE_TYPE (def))) | |
726a989a RB |
3293 | { |
3294 | gsi_next (&psi); | |
3295 | continue; | |
3296 | } | |
3ac01fde | 3297 | |
a6f778b2 | 3298 | def = analyze_scalar_evolution_in_loop (ex_loop, loop, def, NULL); |
925196ed | 3299 | def = compute_overall_effect_of_inner_loop (ex_loop, def); |
3ac01fde | 3300 | if (!tree_does_not_contain_chrecs (def) |
e5db3515 ZD |
3301 | || chrec_contains_symbols_defined_in_loop (def, ex_loop->num) |
3302 | /* Moving the computation from the loop may prolong life range | |
3303 | of some ssa names, which may cause problems if they appear | |
3304 | on abnormal edges. */ | |
771f882e ZD |
3305 | || contains_abnormal_ssa_name_p (def) |
3306 | /* Do not emit expensive expressions. The rationale is that | |
3307 | when someone writes a code like | |
3308 | ||
3309 | while (n > 45) n -= 45; | |
3310 | ||
3311 | he probably knows that n is not large, and does not want it | |
3312 | to be turned into n %= 45. */ | |
3313 | || expression_expensive_p (def)) | |
726a989a RB |
3314 | { |
3315 | gsi_next (&psi); | |
3316 | continue; | |
3317 | } | |
3ac01fde | 3318 | |
9b3b55a1 | 3319 | /* Eliminate the PHI node and replace it by a computation outside |
925196ed ZD |
3320 | the loop. */ |
3321 | def = unshare_expr (def); | |
726a989a RB |
3322 | remove_phi_node (&psi, false); |
3323 | ||
3324 | def = force_gimple_operand_gsi (&bsi, def, false, NULL_TREE, | |
3325 | true, GSI_SAME_STMT); | |
3326 | ass = gimple_build_assign (rslt, def); | |
3327 | gsi_insert_before (&bsi, ass, GSI_SAME_STMT); | |
3ac01fde ZD |
3328 | } |
3329 | } | |
c2924966 | 3330 | return 0; |
684aaf29 | 3331 | } |
9e2f83a5 ZD |
3332 | |
3333 | #include "gt-tree-scalar-evolution.h" |