2 * Copyright (C) 1995-2008 University of Karlsruhe. All right reserved.
4 * This file is part of libFirm.
6 * This file may be distributed and/or modified under the terms of the
7 * GNU General Public License version 2 as published by the Free Software
8 * Foundation and appearing in the file LICENSE.GPL included in the
9 * packaging of this file.
11 * Licensees holding valid libFirm Professional Edition licenses may use
12 * this file in accordance with the libFirm Commercial License.
13 * Agreement provided with the Software.
15 * This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
16 * WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22 * @brief Available Optimisations of libFirm.
25 #ifndef FIRM_IROPTIMIZE_H
26 #define FIRM_IROPTIMIZE_H
28 #include "firm_types.h"
31 * Control flow optimization.
33 * Removes empty blocks doing if simplifications and loop simplifications.
34 * A block is empty if it contains only a Jmp node and Phi nodes.
35 * Merges single entry single exit blocks with their predecessor
36 * and propagates dead control flow by calling equivalent_node().
37 * Independent of compiler flag it removes Tuples from cf edges,
38 * Bad predecessors from Blocks and Phis, and unnecessary predecessors of End.
40 * @bug So far destroys backedge information.
41 * @bug Chokes on Id nodes if called in a certain order with other
42 * optimizations. Call local_optimize_graph() before to remove
45 void optimize_cf(ir_graph *irg);
48 * Perform path-sensitive jump threading on the given graph.
50 * @param irg the graph
52 void opt_jumpthreading(ir_graph* irg);
55 * Try to simplify boolean expression in the given ir graph.
56 * eg. x < 5 && x < 6 becomes x < 5
58 * @param irg the graph
60 void opt_bool(ir_graph *irg);
63 * Try to reduce the number of conv nodes in the given ir graph.
65 * @param irg the graph
67 * @return non-zero if the optimization could be applied, 0 else
69 int conv_opt(ir_graph *irg);
72 * Do the scalar replacement optimization.
73 * Make a date flow analyze and split the
76 * @param irg the graph which should be optimized
78 void data_flow_scalar_replacement_opt(ir_graph *irg);
81 * A callback that checks whether a entity is an allocation
84 typedef int (*check_alloc_entity_func)(ir_entity *ent);
87 * Do simple and fast escape analysis for one graph.
89 * @param irg the graph
90 * @param callback a callback function to check whether a
91 * given entity is a allocation call
93 void escape_enalysis_irg(ir_graph *irg, check_alloc_entity_func callback);
96 * Do simple and fast escape analysis for all graphs.
98 * This optimization implements a simple and fast but inexact
99 * escape analysis. Some addresses might be marked as 'escaped' even
101 * The advantage is a low memory footprint and fast speed.
103 * @param run_scalar_replace if this flag in non-zero, scalar
104 * replacement optimization is run on graphs with removed
106 * @param callback a callback function to check whether a
107 * given entity is a allocation call
109 * This optimization removes allocation which are not used (rare) and replace
110 * allocation that can be proved dead at the end of the graph which stack variables.
112 * The creation of stack variable allows scalar replacement to be run only
113 * on those graphs that have been changed.
115 * This is most effective on Java where no other stack variables exists.
117 void escape_analysis(int run_scalar_replace, check_alloc_entity_func callback);
120 * Optimize function calls by handling const functions.
122 * This optimization first detects all "const functions", i.e.,
123 * IR graphs that neither read nor write memory (and hence did
124 * not create exceptions, as these use memory in Firm).
126 * The result of calls to such functions depends only on its
127 * arguments, hence those calls are no more pinned.
129 * This is a rather strong criteria, so do not expect that a
130 * lot of functions will be found. Moreover, all of them might
131 * already be inlined if inlining is activated.
132 * Anyway, it might be good for handling builtin's or pseudo-graphs,
133 * even if the later read/write memory (but we know how).
135 * This optimizations read the irg_const_function property of
136 * entities and and sets the irg_const_function property of
139 * If callee information is valid, we also optimize polymorphic Calls.
141 * @param force_run if non-zero, an optimization run is started even
142 * if no const function graph was detected.
143 * Else calls are only optimized if at least one
144 * const function graph was detected.
145 * @param callback a callback function to check whether a
146 * given entity is a allocation call
148 * If the frontend created external entities with the irg_const_function
149 * property set, the force_run parameter should be set, else
152 * @note This optimization destroys the link fields of nodes.
154 void optimize_funccalls(int force_run, check_alloc_entity_func callback);
157 * Does Partial Redundancy Elimination combined with
158 * Global Value Numbering.
159 * Can be used to replace place_code() completely.
161 * Based on VanDrunen and Hosking 2004.
163 * @param irg the graph
165 void do_gvn_pre(ir_graph *irg);
168 * This function is called to evaluate, if a mux can build
169 * of the current architecture.
170 * If it returns non-zero, a mux is created, else the code
172 * @param sel A selector of a Cond.
173 * @param phi_list List of Phi nodes about to be converted (linked via get_Phi_next() field)
174 * @param i First data predecessor involved in if conversion
175 * @param j Second data predecessor involved in if conversion
177 typedef int (*arch_allow_ifconv_func)(ir_node *sel, ir_node* phi_list, int i, int j);
180 * The parameters structure.
182 struct ir_settings_if_conv_t {
183 int max_depth; /**< The maximum depth up to which expressions
184 are examined when it has to be decided if they
185 can be placed into another block. */
186 arch_allow_ifconv_func allow_ifconv; /**< Evaluator function, if not set all possible Psi
187 nodes will be created. */
191 * Perform If conversion on a graph.
193 * @param irg The graph.
194 * @param params The parameters for the if conversion.
196 * Cannot handle blocks with Bad control predecessors, so call it after control
199 void opt_if_conv(ir_graph *irg, const ir_settings_if_conv_t *params);
201 void opt_sync(ir_graph *irg);
204 * Check if we can replace the load by a given const from
205 * the const code irg.
207 * @param load the load to replace
208 * @param c the constant
210 * @return in the modes match or can be transformed using a reinterpret cast
211 * returns a copy of the constant (possibly Conv'ed) on the
214 ir_node *can_replace_load_by_const(const ir_node *load, ir_node *c);
217 * Load/Store optimization.
219 * Removes redundant non-volatile Loads and Stores.
220 * May introduce Bad nodes if exceptional control flow
221 * is removed. The following cases are optimized:
223 * Load without result: A Load which has only a memory use
226 * Load after Store: A Load after a Store is removed, if
227 * the Load doesn't have an exception handler OR is in
228 * the same block as the Store.
230 * Load after Load: A Load after a Load is removed, if the
231 * Load doesn't have an exception handler OR is in the
232 * same block as the previous Load.
234 * Store before Store: A Store immediately before another
235 * Store in the same block is removed, if the Store doesn't
236 * have an exception handler.
238 * Store after Load: A Store after a Load is removed, if the
239 * Store doesn't have an exception handler.
241 * @return non-zero if the optimization could be applied, 0 else
243 int optimize_load_store(ir_graph *irg);
246 * New experimental alternative to optimize_load_store.
247 * Based on a dataflow analysis, so load/stores are moved out of loops
250 int opt_ldst(ir_graph *irg);
253 * Do Loop unrolling in the given graph.
255 void optimize_loop_unrolling(ir_graph *irg);
258 * Optimize the frame type of an irg by removing
259 * never touched entities.
261 * @param irg The graph whose frame type will be optimized
263 * This function did not change the graph, only it's frame type.
264 * The layout state of the frame type will be set to layout_undefined
265 * if entities were removed.
267 void opt_frame_irg(ir_graph *irg);
269 /** Possible flags for the Operator Scalar Replacement. */
270 typedef enum osr_flags {
271 osr_flag_none = 0, /**< no additional flags */
272 osr_flag_lftr_with_ov_check = 1, /**< do linear function test replacement
273 only if no overflow can occur. */
274 osr_flag_ignore_x86_shift = 2, /**< ignore Multiplications by 2, 4, 8 */
275 osr_flag_keep_reg_pressure = 4 /**< do NOT increase register pressure by introducing new
276 induction variables. */
279 /* FirmJNI cannot handle identical enum values... */
281 /** default setting */
282 #define osr_flag_default osr_flag_lftr_with_ov_check
285 * Do the Operator Scalar Replacement optimization and linear
286 * function test replacement for loop control.
287 * Can be switched off using the set_opt_strength_red() flag.
288 * In that case, only remove_phi_cycles() is executed.
290 * @param irg the graph which should be optimized
291 * @param flags set of osr_flags
293 * The linear function replacement test is controlled by the flags.
294 * If the osr_flag_lftr_with_ov_check is set, the replacement is only
295 * done if do overflow can occur.
296 * Otherwise it is ALWAYS done which might be insecure.
300 * for (i = 0; i < 100; ++i)
302 * might be replaced by
304 * for (i = 0; i < 400; i += 4)
308 * for (i = 0; i < 0x7FFFFFFF; ++i)
310 * will not be replaced by
312 * for (i = 0; i < 0xFFFFFFFC; i += 4)
314 * because of overflow.
318 * for (i = 0; i <= 0xF; ++i)
320 * will NOT be transformed into
322 * for (i = 0xFFFFFFF0; i <= 0xFFFFFFFF; ++i)
324 * although here is no direct overflow. The OV occurs when the ++i
325 * is executed (and would created an endless loop here!).
327 * For the same reason, a loop
329 * for (i = 0; i <= 9; i += x)
331 * will NOT be transformed because we cannot estimate whether an overflow
332 * might happen adding x.
334 * Note that i < a + 400 is also not possible with the current implementation
335 * although this might be allowed by other compilers...
337 * Note further that tests for equality can be handled some simpler (but are not
340 * This algorithm destroys the link field of nodes.
342 void opt_osr(ir_graph *irg, unsigned flags);
345 * Removes useless Phi cycles, i.e cycles of Phi nodes with only one
347 * This is automatically done in opt_osr(), so there is no need to call it
350 * @param irg the graph which should be optimized
352 * This algorithm destroys the link field of nodes.
354 void remove_phi_cycles(ir_graph *irg);
356 /** A default threshold. */
357 #define DEFAULT_CLONE_THRESHOLD 300
360 * Do procedure cloning. Evaluate a heuristic weight for every
361 * Call(..., Const, ...). If the weight is bigger than threshold,
362 * clone the entity and fix the calls.
364 * @param threshold the threshold for cloning
366 * The threshold is an estimation of how many instructions are saved
367 * when executing a cloned method. If threshold is 0.0, every possible
370 void proc_cloning(float threshold);
375 * Applies Reassociation rules to integer expressions.
376 * Beware: Works only if integer overflow might be ignored, as for C, Java
377 * and for address expression.
378 * Works only if Constant folding is activated.
380 * Uses loop information to detect loop-invariant (ie contant
381 * inside the loop) values.
383 * See Muchnik 12.3.1 Algebraic Simplification and Reassociation of
384 * Addressing Expressions.
386 * @return non-zero if the optimization could be applied, 0 else
388 int optimize_reassociation(ir_graph *irg);
391 * Normalize the Returns of a graph by creating a new End block
392 * with One Return(Phi).
393 * This is the preferred input for the if-conversion.
395 * In pseudocode, it means:
402 * is transformed into
410 void normalize_one_return(ir_graph *irg);
413 * Normalize the Returns of a graph by moving
414 * the Returns upwards as much as possible.
415 * This might be preferred for code generation.
417 * In pseudocode, it means:
425 * is transformed into
432 void normalize_n_returns(ir_graph *irg);
435 * Do the scalar replacement optimization.
436 * Replace local compound entities (like structures and arrays)
437 * with atomic values if possible. Does not handle classes yet.
439 * @param irg the graph which should be optimized
441 * @return non-zero, if at least one entity was replaced
443 int scalar_replacement_opt(ir_graph *irg);
445 /** Performs strength reduction for the passed graph. */
446 void reduce_strength(ir_graph *irg);
449 * Optimizes tail-recursion calls by converting them into loops.
450 * Depends on the flag opt_tail_recursion.
451 * Currently supports the following forms:
453 * - return x + func();
454 * - return func() - x;
455 * - return x * func();
458 * Does not work for Calls that use the exception stuff.
460 * @param irg the graph to be optimized
462 * @return non-zero if the optimization could be applied, 0 else
464 int opt_tail_rec_irg(ir_graph *irg);
467 * Optimize tail-recursion calls for all IR-Graphs.
468 * Can currently handle:
469 * - direct return value, i.e. return func().
470 * - additive return value, i.e. return x +/- func()
471 * - multiplicative return value, i.e. return x * func() or return -func()
473 * The current implementation must be run before optimize_funccalls(),
474 * because it expects the memory edges pointing to calls, which might be
475 * removed by optimize_funccalls().
477 void opt_tail_recursion(void);
479 /** This is the type for a method, that returns a pointer type to
480 * tp. This is needed in the normalization. */
481 typedef ir_type *(*gen_pointer_type_to_func)(ir_type *tp);
483 /** Insert Casts so that class type casts conform exactly with the type hierarchy.
485 * Formulated in Java, this achieves the following:
487 * For a class hierarchy
489 * class B extends A {}
490 * class C extends B {}
491 * we transforms a cast
496 * The algorithm works for Casts with class types, but also for Casts
497 * with all pointer types that point (over several indirections,
498 * i.e. ***A) to a class type. Normalizes all graphs. Computes type
499 * information (@see irtypeinfo.h) if not available.
500 * Invalidates trout information as new casts are generated.
502 * @param gppt_fct A function that returns a pointer type that points
503 * to the type given as argument. If this parameter is NULL, a default
504 * function is used that either uses trout information or performs a O(n)
505 * search to find an existing pointer type. If it can not find a type,
506 * generates a pointer type with mode_P_mach and suffix "cc_ptr_tp".
508 void normalize_irp_class_casts(gen_pointer_type_to_func gppt_fct);
511 /** Insert Casts so that class type casts conform exactly with the type hierarchy
514 * For more details see normalize_irp_class_casts().
516 * This transformation requires that type information is computed. @see irtypeinfo.h.
518 void normalize_irg_class_casts(ir_graph *irg, gen_pointer_type_to_func gppt_fct);
521 /** Optimize casting between class types.
524 * class B extends A { }
525 * class C extends B {}
526 * Performs the following transformations:
527 * C c = (C)(B)(A)(B)new C() --> C c = (C)(B)newC() --> C c = new C()
528 * (Optimizing downcasts as A a = (A)(B)(new A()) --> A a = new A() can
529 * be suppressed by setting the flag opt_suppress_downcast_optimization.
530 * Downcasting A to B might cause an exception. It is not clear
531 * whether this is modeled by the Firm Cast node, as it has no exception
533 * If there is inh_m() that overwrites m() in B:
534 * ((A) new B()).m() --> (new B()).inh_m()
535 * Phi((A)x, (A)y) --> (A) Phi (x, y) if (A) is an upcast.
537 * Computes type information if not available. @see irtypeinfo.h.
538 * Typeinformation is valid after optimization.
539 * Invalidates trout information.
541 void optimize_class_casts(void);
544 * CLiff Click's combo algorithm from "Combining Analyses, combining Optimizations".
546 * Does conditional constant propagation, unreachable code elimination and optimistic
547 * global value numbering at once.
549 * @param irg the graph to run on
551 void combo(ir_graph *irg);
553 /** Inlines all small methods at call sites where the called address comes
554 * from a SymConst node that references the entity representing the called
557 * The size argument is a rough measure for the code size of the method:
558 * Methods where the obstack containing the firm graph is smaller than
559 * size are inlined. Further only a limited number of calls are inlined.
560 * If the method contains more than 1024 inlineable calls none will be
562 * Inlining is only performed if flags `optimize' and `inlineing' are set.
563 * The graph may not be in state phase_building.
564 * It is recommended to call local_optimize_graph() after inlining as this
565 * function leaves a set of obscure Tuple nodes, e.g. a Proj-Tuple-Jmp
566 * combination as control flow operation.
568 void inline_small_irgs(ir_graph *irg, int size);
571 /** Inlineing with a different heuristic than inline_small_irgs().
573 * Inlines leave functions. If inlinening creates new leave
574 * function inlines these, too. (If g calls f, and f calls leave h,
575 * h is first inlined in f and then f in g.)
577 * Then inlines all small functions (this is not recursive).
579 * For a heuristic this inlineing uses firm node counts. It does
580 * not count auxiliary nodes as Proj, Tuple, End, Start, Id, Sync.
581 * If the ignore_runtime flag is set, calls to functions marked with the
582 * mtp_property_runtime property are ignored.
584 * @param maxsize Do not inline any calls if a method has more than
585 * maxsize firm nodes. It may reach this limit by
587 * @param leavesize Inline leave functions if they have less than leavesize
589 * @param size Inline all function smaller than size.
590 * @param ignore_runtime count a function only calling runtime functions as
593 void inline_leave_functions(unsigned maxsize, unsigned leavesize,
594 unsigned size, int ignore_runtime);
597 * Heuristic inliner. Calculates a benefice value for every call and inlines
598 * those calls with a value higher than the threshold.
600 * @param maxsize Do not inline any calls if a method has more than
601 * maxsize firm nodes. It may reach this limit by
603 * @param threshold inlining threshold
605 void inline_functions(unsigned maxsize, int inline_threshold);
608 * Combines congruent blocks into one.
610 * @param irg The IR-graph to optimize.
612 * @return non-zero if the graph was transformed
614 int shape_blocks(ir_graph *irg);