From 56c1a4fd4bb668f7981497688c4feae55d524aac Mon Sep 17 00:00:00 2001 From: Aditya Kumar Date: Tue, 18 Aug 2015 13:27:54 -0500 Subject: PR tree-optimization/52563, tree-optimization/62173 commit 5fe66b3cf99994fd9c8c68cea43aa1cf42eaa76d Author: amker Date: Tue Jun 2 03:33:35 2015 +0000 PR tree-optimization/52563 PR tree-optimization/62173 * tree-ssa-loop-ivopts.c (struct iv): New field. Reorder fields. (alloc_iv, set_iv): New parameter. (determine_biv_step): Delete. (find_bivs): Inline original determine_biv_step. Pass new argument to set_iv. (idx_find_step): Use no_overflow information for conversion. * tree-scalar-evolution.c (analyze_scalar_evolution_in_loop): Let resolve_mixers handle folded_casts. (instantiate_scev_name): Change bool parameter to bool pointer. (instantiate_scev_poly, instantiate_scev_binary): Ditto. (instantiate_array_ref, instantiate_scev_not): Ditto. (instantiate_scev_3, instantiate_scev_2): Ditto. (instantiate_scev_1, instantiate_scev_r): Ditto. (instantiate_scev_convert, ): Change parameter. Pass argument to chrec_convert_aggressive. (instantiate_scev): Change argument. (resolve_mixers): New parameter and set it. (scev_const_prop): New argument. * tree-scalar-evolution.h (resolve_mixers): New parameter. * tree-chrec.c (convert_affine_scev): Call chrec_convert instead of chrec_conert_1. (chrec_convert): New parameter. Move definition below. (chrec_convert_aggressive): New parameter and set it. Call convert_affine_scev. * tree-chrec.h (chrec_convert): New parameter. (chrec_convert_aggressive): Ditto. gcc/testsuite/ChangeLog PR tree-optimization/52563 PR tree-optimization/62173 * gcc.dg/tree-ssa/scev-3.c: Remove xfail. * gcc.dg/tree-ssa/scev-4.c: Ditto. git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@224009 138bc75d-0d04-0410-961f-82ee72b054a4 --- gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-3.c | 2 +- gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-4.c | 2 +- gcc-4.9/gcc/tree-chrec.c | 104 +++++++++++-------- gcc-4.9/gcc/tree-chrec.h | 4 +- gcc-4.9/gcc/tree-scalar-evolution.c | 138 ++++++++++++++----------- gcc-4.9/gcc/tree-scalar-evolution.h | 2 +- gcc-4.9/gcc/tree-ssa-loop-ivopts.c | 58 +++++------ 7 files changed, 171 insertions(+), 139 deletions(-) (limited to 'gcc-4.9') diff --git a/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-3.c b/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-3.c index 5cac1cefb..28d5c9327 100644 --- a/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-3.c +++ b/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-3.c @@ -14,5 +14,5 @@ f(int k) } } -/* { dg-final { scan-tree-dump-times "&a" 1 "optimized" { xfail { lp64 || llp64 } } } } */ +/* { dg-final { scan-tree-dump-times "&a" 1 "optimized" } } */ /* { dg-final { cleanup-tree-dump "optimized" } } */ diff --git a/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-4.c b/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-4.c index 5f15d622d..6c1e530a9 100644 --- a/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-4.c +++ b/gcc-4.9/gcc/testsuite/gcc.dg/tree-ssa/scev-4.c @@ -19,5 +19,5 @@ f(int k) } } -/* { dg-final { scan-tree-dump-times "&a" 1 "optimized" { xfail { lp64 || llp64 } } } } */ +/* { dg-final { scan-tree-dump-times "&a" 1 "optimized" } } */ /* { dg-final { cleanup-tree-dump "optimized" } } */ diff --git a/gcc-4.9/gcc/tree-chrec.c b/gcc-4.9/gcc/tree-chrec.c index b9350f015..3e34c65dc 100644 --- a/gcc-4.9/gcc/tree-chrec.c +++ b/gcc-4.9/gcc/tree-chrec.c @@ -1162,8 +1162,6 @@ nb_vars_in_chrec (tree chrec) } } -static tree chrec_convert_1 (tree, tree, gimple, bool); - /* Converts BASE and STEP of affine scev to TYPE. LOOP is the loop whose iv the scev corresponds to. AT_STMT is the statement at that the scev is evaluated. USE_OVERFLOW_SEMANTICS is true if this function should assume that @@ -1238,8 +1236,7 @@ convert_affine_scev (struct loop *loop, tree type, use_overflow_semantics)) return false; - new_base = chrec_convert_1 (type, *base, at_stmt, - use_overflow_semantics); + new_base = chrec_convert (type, *base, at_stmt, use_overflow_semantics); /* The step must be sign extended, regardless of the signedness of CT and TYPE. This only needs to be handled specially when CT is unsigned -- to avoid e.g. unsigned char [100, +, 255] @@ -1250,10 +1247,11 @@ convert_affine_scev (struct loop *loop, tree type, if (TYPE_PRECISION (step_type) > TYPE_PRECISION (ct) && TYPE_UNSIGNED (ct)) { tree signed_ct = build_nonstandard_integer_type (TYPE_PRECISION (ct), 0); - new_step = chrec_convert_1 (signed_ct, new_step, at_stmt, - use_overflow_semantics); + new_step = chrec_convert (signed_ct, new_step, at_stmt, + use_overflow_semantics); } - new_step = chrec_convert_1 (step_type, new_step, at_stmt, use_overflow_semantics); + new_step = chrec_convert (step_type, new_step, at_stmt, + use_overflow_semantics); if (automatically_generated_chrec_p (new_base) || automatically_generated_chrec_p (new_step)) @@ -1283,36 +1281,6 @@ chrec_convert_rhs (tree type, tree chrec, gimple at_stmt) return chrec_convert (type, chrec, at_stmt); } -/* Convert CHREC to TYPE. When the analyzer knows the context in - which the CHREC is built, it sets AT_STMT to the statement that - contains the definition of the analyzed variable, otherwise the - conversion is less accurate: the information is used for - determining a more accurate estimation of the number of iterations. - By default AT_STMT could be safely set to NULL_TREE. - - The following rule is always true: TREE_TYPE (chrec) == - TREE_TYPE (CHREC_LEFT (chrec)) == TREE_TYPE (CHREC_RIGHT (chrec)). - An example of what could happen when adding two chrecs and the type - of the CHREC_RIGHT is different than CHREC_LEFT is: - - {(uint) 0, +, (uchar) 10} + - {(uint) 0, +, (uchar) 250} - - that would produce a wrong result if CHREC_RIGHT is not (uint): - - {(uint) 0, +, (uchar) 4} - - instead of - - {(uint) 0, +, (uint) 260} -*/ - -tree -chrec_convert (tree type, tree chrec, gimple at_stmt) -{ - return chrec_convert_1 (type, chrec, at_stmt, true); -} - /* Convert CHREC to TYPE. When the analyzer knows the context in which the CHREC is built, it sets AT_STMT to the statement that contains the definition of the analyzed variable, otherwise the @@ -1404,15 +1372,53 @@ keep_cast: return res; } +/* Convert CHREC to TYPE. When the analyzer knows the context in + which the CHREC is built, it sets AT_STMT to the statement that + contains the definition of the analyzed variable, otherwise the + conversion is less accurate: the information is used for + determining a more accurate estimation of the number of iterations. + By default AT_STMT could be safely set to NULL_TREE. + + The following rule is always true: TREE_TYPE (chrec) == + TREE_TYPE (CHREC_LEFT (chrec)) == TREE_TYPE (CHREC_RIGHT (chrec)). + An example of what could happen when adding two chrecs and the type + of the CHREC_RIGHT is different than CHREC_LEFT is: + + {(uint) 0, +, (uchar) 10} + + {(uint) 0, +, (uchar) 250} + + that would produce a wrong result if CHREC_RIGHT is not (uint): + + {(uint) 0, +, (uchar) 4} + + instead of + + {(uint) 0, +, (uint) 260} + + USE_OVERFLOW_SEMANTICS is true if this function should assume that + the rules for overflow of the given language apply (e.g., that signed + arithmetics in C does not overflow) -- i.e., to use them to avoid unnecessary + tests, but also to enforce that the result follows them. */ + +tree +chrec_convert (tree type, tree chrec, gimple at_stmt, + bool use_overflow_semantics) +{ + return chrec_convert_1 (type, chrec, at_stmt, use_overflow_semantics); +} + /* Convert CHREC to TYPE, without regard to signed overflows. Returns the new chrec if something else than what chrec_convert would do happens, NULL_TREE - otherwise. */ + otherwise. This function set TRUE to variable pointed by FOLD_CONVERSIONS + if the result chrec may overflow. */ tree -chrec_convert_aggressive (tree type, tree chrec) +chrec_convert_aggressive (tree type, tree chrec, bool *fold_conversions) { tree inner_type, left, right, lc, rc, rtype; + gcc_assert (fold_conversions != NULL); + if (automatically_generated_chrec_p (chrec) || TREE_CODE (chrec) != POLYNOMIAL_CHREC) return NULL_TREE; @@ -1421,17 +1427,33 @@ chrec_convert_aggressive (tree type, tree chrec) if (TYPE_PRECISION (type) > TYPE_PRECISION (inner_type)) return NULL_TREE; + if (useless_type_conversion_p (type, inner_type)) + return NULL_TREE; + + if (!*fold_conversions && evolution_function_is_affine_p (chrec)) + { + tree base, step; + struct loop *loop; + + loop = get_chrec_loop (chrec); + base = CHREC_LEFT (chrec); + step = CHREC_RIGHT (chrec); + if (convert_affine_scev (loop, type, &base, &step, NULL, true)) + return build_polynomial_chrec (loop->num, base, step); + } rtype = POINTER_TYPE_P (type) ? sizetype : type; left = CHREC_LEFT (chrec); right = CHREC_RIGHT (chrec); - lc = chrec_convert_aggressive (type, left); + lc = chrec_convert_aggressive (type, left, fold_conversions); if (!lc) lc = chrec_convert (type, left, NULL); - rc = chrec_convert_aggressive (rtype, right); + rc = chrec_convert_aggressive (rtype, right, fold_conversions); if (!rc) rc = chrec_convert (rtype, right, NULL); + *fold_conversions = true; + return build_polynomial_chrec (CHREC_VARIABLE (chrec), lc, rc); } diff --git a/gcc-4.9/gcc/tree-chrec.h b/gcc-4.9/gcc/tree-chrec.h index 90cc7a74b..ea4627751 100644 --- a/gcc-4.9/gcc/tree-chrec.h +++ b/gcc-4.9/gcc/tree-chrec.h @@ -59,9 +59,9 @@ enum ev_direction scev_direction (const_tree); extern tree chrec_fold_plus (tree, tree, tree); extern tree chrec_fold_minus (tree, tree, tree); extern tree chrec_fold_multiply (tree, tree, tree); -extern tree chrec_convert (tree, tree, gimple); +extern tree chrec_convert (tree, tree, gimple, bool = true); extern tree chrec_convert_rhs (tree, tree, gimple); -extern tree chrec_convert_aggressive (tree, tree); +extern tree chrec_convert_aggressive (tree, tree, bool *); /* Operations. */ extern tree chrec_apply (unsigned, tree, tree); diff --git a/gcc-4.9/gcc/tree-scalar-evolution.c b/gcc-4.9/gcc/tree-scalar-evolution.c index f1ddc24b1..218baa406 100644 --- a/gcc-4.9/gcc/tree-scalar-evolution.c +++ b/gcc-4.9/gcc/tree-scalar-evolution.c @@ -2116,7 +2116,7 @@ analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop, /* We cannot just do tmp = analyze_scalar_evolution (use_loop, version); - ev = resolve_mixers (wrto_loop, tmp); + ev = resolve_mixers (wrto_loop, tmp, folded_casts); as resolve_mixers would query the scalar evolution with respect to wrto_loop. For example, in the situation described in the function @@ -2125,9 +2125,9 @@ analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop, analyze_scalar_evolution (use_loop, version) = k2 - and resolve_mixers (loop1, k2) finds that the value of k2 in loop 1 - is 100, which is a wrong result, since we are interested in the - value in loop 3. + and resolve_mixers (loop1, k2, folded_casts) finds that the value of + k2 in loop 1 is 100, which is a wrong result, since we are interested + in the value in loop 3. Instead, we need to proceed from use_loop to wrto_loop loop by loop, each time checking that there is no evolution in the inner loop. */ @@ -2137,10 +2137,7 @@ analyze_scalar_evolution_in_loop (struct loop *wrto_loop, struct loop *use_loop, while (1) { tmp = analyze_scalar_evolution (use_loop, ev); - ev = resolve_mixers (use_loop, tmp); - - if (folded_casts && tmp != ev) - *folded_casts = true; + ev = resolve_mixers (use_loop, tmp, folded_casts); if (use_loop == wrto_loop) return ev; @@ -2262,7 +2259,7 @@ loop_closed_phi_def (tree var) } static tree instantiate_scev_r (basic_block, struct loop *, struct loop *, - tree, bool, int); + tree, bool *, int); /* Analyze all the parameters of the chrec, between INSTANTIATE_BELOW and EVOLUTION_LOOP, that were left under a symbolic form. @@ -2271,9 +2268,10 @@ static tree instantiate_scev_r (basic_block, struct loop *, struct loop *, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2282,7 +2280,7 @@ static tree instantiate_scev_name (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, - bool fold_conversions, + bool *fold_conversions, int size_expr) { tree res; @@ -2376,9 +2374,10 @@ instantiate_scev_name (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2386,7 +2385,7 @@ instantiate_scev_name (basic_block instantiate_below, static tree instantiate_scev_poly (basic_block instantiate_below, struct loop *evolution_loop, struct loop *, - tree chrec, bool fold_conversions, int size_expr) + tree chrec, bool *fold_conversions, int size_expr) { tree op1; tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, @@ -2420,9 +2419,10 @@ instantiate_scev_poly (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2432,7 +2432,7 @@ instantiate_scev_binary (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, enum tree_code code, tree type, tree c0, tree c1, - bool fold_conversions, int size_expr) + bool *fold_conversions, int size_expr) { tree op1; tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, inner_loop, @@ -2478,9 +2478,10 @@ instantiate_scev_binary (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2488,7 +2489,7 @@ instantiate_scev_binary (basic_block instantiate_below, static tree instantiate_array_ref (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, - tree chrec, bool fold_conversions, int size_expr) + tree chrec, bool *fold_conversions, int size_expr) { tree res; tree index = TREE_OPERAND (chrec, 1); @@ -2515,9 +2516,10 @@ instantiate_array_ref (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2526,7 +2528,7 @@ static tree instantiate_scev_convert (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, tree type, tree op, - bool fold_conversions, int size_expr) + bool *fold_conversions, int size_expr) { tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, inner_loop, op, @@ -2537,19 +2539,21 @@ instantiate_scev_convert (basic_block instantiate_below, if (fold_conversions) { - tree tmp = chrec_convert_aggressive (type, op0); + tree tmp = chrec_convert_aggressive (type, op0, fold_conversions); if (tmp) return tmp; - } - if (chrec && op0 == op) - return chrec; + /* If we used chrec_convert_aggressive, we can no longer assume that + signed chrecs do not overflow, as chrec_convert does, so avoid + calling it in that case. */ + if (*fold_conversions) + { + if (chrec && op0 == op) + return chrec; - /* If we used chrec_convert_aggressive, we can no longer assume that - signed chrecs do not overflow, as chrec_convert does, so avoid - calling it in that case. */ - if (fold_conversions) - return fold_convert (type, op0); + return fold_convert (type, op0); + } + } return chrec_convert (type, op0, NULL); } @@ -2563,9 +2567,10 @@ instantiate_scev_convert (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2575,7 +2580,7 @@ instantiate_scev_not (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, enum tree_code code, tree type, tree op, - bool fold_conversions, int size_expr) + bool *fold_conversions, int size_expr) { tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, inner_loop, op, @@ -2613,9 +2618,10 @@ instantiate_scev_not (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2624,7 +2630,7 @@ static tree instantiate_scev_3 (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, - bool fold_conversions, int size_expr) + bool *fold_conversions, int size_expr) { tree op1, op2; tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, @@ -2661,9 +2667,10 @@ instantiate_scev_3 (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2672,7 +2679,7 @@ static tree instantiate_scev_2 (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, - bool fold_conversions, int size_expr) + bool *fold_conversions, int size_expr) { tree op1; tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, @@ -2701,9 +2708,10 @@ instantiate_scev_2 (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2712,7 +2720,7 @@ static tree instantiate_scev_1 (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, - bool fold_conversions, int size_expr) + bool *fold_conversions, int size_expr) { tree op0 = instantiate_scev_r (instantiate_below, evolution_loop, inner_loop, TREE_OPERAND (chrec, 0), @@ -2734,9 +2742,10 @@ instantiate_scev_1 (basic_block instantiate_below, CACHE is the cache of already instantiated values. - FOLD_CONVERSIONS should be set to true when the conversions that - may wrap in signed/pointer type are folded, as long as the value of - the chrec is preserved. + Variable pointed by FOLD_CONVERSIONS is set to TRUE when the + conversions that may wrap in signed/pointer type are folded, as long + as the value of the chrec is preserved. If FOLD_CONVERSIONS is NULL + then we don't do such fold. SIZE_EXPR is used for computing the size of the expression to be instantiated, and to stop if it exceeds some limit. */ @@ -2745,7 +2754,7 @@ static tree instantiate_scev_r (basic_block instantiate_below, struct loop *evolution_loop, struct loop *inner_loop, tree chrec, - bool fold_conversions, int size_expr) + bool *fold_conversions, int size_expr) { /* Give up if the expression is larger than the MAX that we allow. */ if (size_expr++ > PARAM_VALUE (PARAM_SCEV_MAX_EXPR_SIZE)) @@ -2870,7 +2879,7 @@ instantiate_scev (basic_block instantiate_below, struct loop *evolution_loop, } res = instantiate_scev_r (instantiate_below, evolution_loop, - NULL, chrec, false, 0); + NULL, chrec, NULL, 0); if (destr) { @@ -2894,9 +2903,10 @@ instantiate_scev (basic_block instantiate_below, struct loop *evolution_loop, of an expression. */ tree -resolve_mixers (struct loop *loop, tree chrec) +resolve_mixers (struct loop *loop, tree chrec, bool *folded_casts) { bool destr = false; + bool fold_conversions = false; if (!global_cache) { global_cache = new instantiate_cache_type; @@ -2904,7 +2914,10 @@ resolve_mixers (struct loop *loop, tree chrec) } tree ret = instantiate_scev_r (block_before_loop (loop), loop, NULL, - chrec, true, 0); + chrec, &fold_conversions, 0); + + if (folded_casts && !*folded_casts) + *folded_casts = fold_conversions; if (destr) { @@ -3369,7 +3382,8 @@ scev_const_prop (void) && !INTEGRAL_TYPE_P (type)) continue; - ev = resolve_mixers (loop, analyze_scalar_evolution (loop, name)); + ev = resolve_mixers (loop, analyze_scalar_evolution (loop, name), + NULL); if (!is_gimple_min_invariant (ev) || !may_propagate_copy (name, ev)) continue; diff --git a/gcc-4.9/gcc/tree-scalar-evolution.h b/gcc-4.9/gcc/tree-scalar-evolution.h index 556997684..cb9af5122 100644 --- a/gcc-4.9/gcc/tree-scalar-evolution.h +++ b/gcc-4.9/gcc/tree-scalar-evolution.h @@ -31,7 +31,7 @@ extern void scev_reset_htab (void); extern void scev_finalize (void); extern tree analyze_scalar_evolution (struct loop *, tree); extern tree instantiate_scev (basic_block, struct loop *, tree); -extern tree resolve_mixers (struct loop *, tree); +extern tree resolve_mixers (struct loop *, tree, bool *); extern void gather_stats_on_scev_database (void); extern unsigned int scev_const_prop (void); extern bool expression_expensive_p (tree); diff --git a/gcc-4.9/gcc/tree-ssa-loop-ivopts.c b/gcc-4.9/gcc/tree-ssa-loop-ivopts.c index 400187964..0bf8e2aff 100644 --- a/gcc-4.9/gcc/tree-ssa-loop-ivopts.c +++ b/gcc-4.9/gcc/tree-ssa-loop-ivopts.c @@ -142,9 +142,10 @@ struct iv tree base_object; /* A memory object to that the induction variable points. */ tree step; /* Step of the iv (constant only). */ tree ssa_name; /* The ssa name with the value. */ + unsigned use_id; /* The identifier in the use if it is the case. */ bool biv_p; /* Is it a biv? */ bool have_use_for; /* Do we already have a use for it? */ - unsigned use_id; /* The identifier in the use if it is the case. */ + bool no_overflow; /* True if the iv doesn't overflow. */ }; /* Per-ssa version information (induction variable descriptions, etc.). */ @@ -943,10 +944,10 @@ determine_base_object (tree expr) } /* Allocates an induction variable with given initial value BASE and step STEP - for loop LOOP. */ + for loop LOOP. NO_OVERFLOW implies the iv doesn't overflow. */ static struct iv * -alloc_iv (tree base, tree step) +alloc_iv (tree base, tree step, bool no_overflow = false) { tree base_object = base; struct iv *iv = XCNEW (struct iv); @@ -977,21 +978,24 @@ alloc_iv (tree base, tree step) iv->have_use_for = false; iv->use_id = 0; iv->ssa_name = NULL_TREE; + iv->no_overflow = no_overflow; return iv; } -/* Sets STEP and BASE for induction variable IV. */ +/* Sets STEP and BASE for induction variable IV. NO_OVERFLOW implies the IV + doesn't overflow. */ static void -set_iv (struct ivopts_data *data, tree iv, tree base, tree step) +set_iv (struct ivopts_data *data, tree iv, tree base, tree step, + bool no_overflow) { struct version_info *info = name_info (data, iv); gcc_assert (!info->iv); bitmap_set_bit (data->relevant, SSA_NAME_VERSION (iv)); - info->iv = alloc_iv (base, step); + info->iv = alloc_iv (base, step, no_overflow); info->iv->ssa_name = iv; } @@ -1013,37 +1017,19 @@ get_iv (struct ivopts_data *data, tree var) if (!bb || !flow_bb_inside_loop_p (data->current_loop, bb)) - set_iv (data, var, var, build_int_cst (type, 0)); + set_iv (data, var, var, build_int_cst (type, 0), true); } return name_info (data, var)->iv; } -/* Determines the step of a biv defined in PHI. Returns NULL if PHI does - not define a simple affine biv with nonzero step. */ - -static tree -determine_biv_step (gimple phi) -{ - struct loop *loop = gimple_bb (phi)->loop_father; - tree name = PHI_RESULT (phi); - affine_iv iv; - - if (virtual_operand_p (name)) - return NULL_TREE; - - if (!simple_iv (loop, loop, name, &iv, true)) - return NULL_TREE; - - return integer_zerop (iv.step) ? NULL_TREE : iv.step; -} - /* Finds basic ivs. */ static bool find_bivs (struct ivopts_data *data) { gimple phi; + affine_iv iv; tree step, type, base; bool found = false; struct loop *loop = data->current_loop; @@ -1056,10 +1042,16 @@ find_bivs (struct ivopts_data *data) if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (PHI_RESULT (phi))) continue; - step = determine_biv_step (phi); - if (!step) + if (virtual_operand_p (PHI_RESULT (phi))) continue; + if (!simple_iv (loop, loop, PHI_RESULT (phi), &iv, true)) + continue; + + if (integer_zerop (iv.step)) + continue; + + step = iv.step; base = PHI_ARG_DEF_FROM_EDGE (phi, loop_preheader_edge (loop)); base = expand_simple_operations (base); if (contains_abnormal_ssa_name_p (base) @@ -1076,7 +1068,7 @@ find_bivs (struct ivopts_data *data) step = fold_convert (type, step); } - set_iv (data, PHI_RESULT (phi), base, step); + set_iv (data, PHI_RESULT (phi), base, step, iv.no_overflow); found = true; } @@ -1172,7 +1164,7 @@ find_givs_in_stmt (struct ivopts_data *data, gimple stmt) if (!find_givs_in_stmt_scev (data, stmt, &iv)) return; - set_iv (data, gimple_assign_lhs (stmt), iv.base, iv.step); + set_iv (data, gimple_assign_lhs (stmt), iv.base, iv.step, iv.no_overflow); } /* Finds general ivs in basic block BB. */ @@ -1584,6 +1576,7 @@ idx_find_step (tree base, tree *idx, void *data) { struct ifs_ivopts_data *dta = (struct ifs_ivopts_data *) data; struct iv *iv; + bool use_overflow_semantics = false; tree step, iv_base, iv_step, lbound, off; struct loop *loop = dta->ivopts_data->current_loop; @@ -1643,9 +1636,12 @@ idx_find_step (tree base, tree *idx, void *data) iv_base = iv->base; iv_step = iv->step; + if (iv->no_overflow && nowrap_type_p (TREE_TYPE (iv_step))) + use_overflow_semantics = true; + if (!convert_affine_scev (dta->ivopts_data->current_loop, sizetype, &iv_base, &iv_step, dta->stmt, - false)) + use_overflow_semantics)) { /* The index might wrap. */ return false; -- cgit v1.2.3