From eea12c729f81ba778b6b37e0ddc3311b80e867af Mon Sep 17 00:00:00 2001 From: dnovillo Date: Thu, 2 Jun 2005 02:57:15 +0000 Subject: [PATCH] 2005-06-01 Diego Novillo PR 14341, PR 21332, PR 20701, PR 21029, PR 21086, PR 21090 PR 21289, PR 21348, PR 21367, PR 21368, PR 21458. * fold-const.c (invert_tree_comparison): Make extern. * tree-flow.h (enum value_range_type): Move to tree-ssa-propagate. (struct value_range_def): Limewise. (get_value_range): Remove. (dump_value_range): Remove. (dump_all_value_ranges): Remove. (debug_all_value_ranges): Remove. (vrp_evaluate_conditional): Declare. * tree-ssa-propagate.c (struct prop_stats_d): Add field num_pred_folded. (substitute_and_fold): Add argument use_ranges_p. Update all callers. If use_ranges_p is true, call fold_predicate_in to fold predicates using range information. Ignore ASSERT_EXPRs. Change debugging output to only show statements that have been folded. (replace_phi_args_in): Move debugging output code from substitute and fold. (fold_predicate_in): New local function. * tree-ssa-propagate.h (enum value_range_type): Move from tree-flow.h. (struct value_range_d): Likewise. Add field 'equiv'. (value_range_t): Rename from value_range. * tree-vrp.c (found_in_subgraph): Rename from found. (get_opposite_operand): Remove. (struct assert_locus_d): Declare. (assert_locus_t): Declare. (need_assert_for): Declare. (asserts_for): Declare. (blocks_visited): Declare. (vr_value): Declare. (set_value_range): Add argument 'equiv'. Don't drop to VARYING ranges that cover all values in the type. Make deep copy of equivalence set 'equiv'. (copy_value_range): New local function. (set_value_range_to_undefined): New local function. (compare_values): Return -2 if either value has overflowed. (range_includes_zero_p): New local function. (extract_range_from_assert): Flip the predicate code if the name being asserted is on the RHS of the predicate. Avoid creating unnecessary symbolic ranges if the comparison includes another name with a known numeric range. Update the equivalnce set of the new range when asserting EQ_EXPR predicates. (extract_range_from_ssa_name): Update the equivalence set of the new range with VAR. (extract_range_from_binary_expr): Also handle TRUTH_*_EXPR. If -fwrapv is used, set the resulting range to VARYING if the operation overflows. Otherwise, use TYPE_MIN_VALUE and TYPE_MAX_VALUE to represent -INF and +INF. Fix handling of *_DIV_EXPR. (extract_range_from_unary_expr): Handle MINUS_EXPR and ABS_EXPR properly by switching the range around if necessary. (extract_range_from_comparison): New local function. (extract_range_from_expr): Call it. (adjust_range_with_scev): Do not adjust the range if using wrapping arithmetic (-fwrapv). (dump_value_range): Also show equivalence set. Show -INF and +INF for TYPE_MIN_VALUE and TYPE_MAX_VALUE. (build_assert_expr_for): Also build ASSERT_EXPR for EQ_EXPR. (infer_value_range): Change return value to bool. Add arguments 'comp_code_p' and 'val_p'. Do not attempt to infer ranges from statements that may throw. Store the comparison code in comp_code_p. Store the other operand to be used in the predicate in val_p. (dump_asserts_for): New. (debug_asserts_for): New. (dump_all_asserts): New. (debug_all_asserts): New. (register_new_assert_for): New. (register_edge_assert_for): New. (find_conditional_asserts): New. (find_assert_locations): New. (process_assert_insertions_for): New. (process_assert_insertions): New. (insert_range_assertions): Initialize found_in_subgraph, blocks_visited, need_assert_for and asserts_for. Call find_assert_locations and process_assert_insertions. (remove_range_assertions): Add more documentation. (vrp_initialize): Change return type to void. Do not try to guess if running VRP is worth it. (compare_name_with_value): New. (compare_names): New. (vrp_evaluate_conditional): Add argument 'use_equiv_p'. If use_equiv_p is true, call compare_names and compare_name_with_value to compare all the ranges for every name in the equivalence set of the predicate operands. Update all callers. (vrp_meet): Try harder not to derive a VARYING range. If two values meet, the resulting equivalence set is the intersection of the two equivalence sets. (vrp_visit_phi_node): Call copy_value_range to get the current range information of the LHS. (vrp_finalize): Create a value vector representing all the names that ended up with exactly one value in their range. Call substitute_and_fold. (execute_vrp): Document equivalence sets in ranges. * tree.h (SSA_NAME_VALUE_RANGE): Remove. (struct tree_ssa_name): Remove field value_range. (invert_tree_comparison): Declare. testsuite/ChangeLog 2005-06-01 Diego Novillo PR 14341, PR 21332, PR 20701, PR 21086, PR 21090 PR 21289, PR 21348, PR 21367, PR 21368, PR 21458. * gcc.dg/tree-ssa/pr14341.c: New test. * gcc.dg/tree-ssa/pr14841.c: New test. * gcc.dg/tree-ssa/pr20701.c: New test. * gcc.dg/tree-ssa/pr21086.c: New test. * gcc.dg/tree-ssa/pr21090.c: New test. * gcc.dg/tree-ssa/pr21332.c: New test. * gcc.dg/tree-ssa/pr21458.c: New test. * gcc.dg/tree-ssa/pr21658.c: New test. * gcc.dg/tree-ssa/vrp01.c: New test. * gcc.dg/tree-ssa/vrp02.c: New test. * gcc.dg/tree-ssa/vrp03.c: New test. * gcc.dg/tree-ssa/vrp04.c: New test. * gcc.dg/tree-ssa/vrp05.c: New test. * gcc.dg/tree-ssa/vrp06.c: New test. * gcc.dg/tree-ssa/vrp07.c: New test. * gcc.dg/tree-ssa/vrp08.c: New test. * gcc.dg/tree-ssa/vrp09.c: New test. * gcc.dg/tree-ssa/vrp10.c: New test. * gcc.dg/tree-ssa/vrp11.c: New test. * gcc.dg/tree-ssa/vrp12.c: New test. * gcc.dg/tree-ssa/vrp13.c: New test. 2005-06-01 Alexandre Oliva PR 21029 * gcc.dg/tree-ssa/pr21029.c: New test. git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@100478 138bc75d-0d04-0410-961f-82ee72b054a4 --- gcc/ChangeLog | 108 ++ gcc/fold-const.c | 3 +- gcc/testsuite/ChangeLog | 31 + gcc/testsuite/gcc.dg/tree-ssa/pr14341.c | 16 + gcc/testsuite/gcc.dg/tree-ssa/pr14841.c | 2 +- gcc/testsuite/gcc.dg/tree-ssa/pr20701.c | 30 + gcc/testsuite/gcc.dg/tree-ssa/pr21029.c | 118 ++ gcc/testsuite/gcc.dg/tree-ssa/pr21086.c | 19 + gcc/testsuite/gcc.dg/tree-ssa/pr21090.c | 23 + gcc/testsuite/gcc.dg/tree-ssa/pr21332.c | 28 + gcc/testsuite/gcc.dg/tree-ssa/pr21458.c | 20 + gcc/testsuite/gcc.dg/tree-ssa/pr21658.c | 2 +- gcc/testsuite/gcc.dg/tree-ssa/vrp01.c | 28 + gcc/testsuite/gcc.dg/tree-ssa/vrp02.c | 24 + gcc/testsuite/gcc.dg/tree-ssa/vrp03.c | 35 + gcc/testsuite/gcc.dg/tree-ssa/vrp04.c | 13 + gcc/testsuite/gcc.dg/tree-ssa/vrp05.c | 20 + gcc/testsuite/gcc.dg/tree-ssa/vrp06.c | 31 + gcc/testsuite/gcc.dg/tree-ssa/vrp07.c | 36 + gcc/testsuite/gcc.dg/tree-ssa/vrp08.c | 23 + gcc/testsuite/gcc.dg/tree-ssa/vrp09.c | 31 + gcc/testsuite/gcc.dg/tree-ssa/vrp10.c | 25 + gcc/testsuite/gcc.dg/tree-ssa/vrp11.c | 20 + gcc/testsuite/gcc.dg/tree-ssa/vrp12.c | 23 + gcc/testsuite/gcc.dg/tree-ssa/vrp13.c | 139 ++ gcc/tree-flow.h | 34 +- gcc/tree-ssa-ccp.c | 2 +- gcc/tree-ssa-copy.c | 2 +- gcc/tree-ssa-propagate.c | 146 +- gcc/tree-ssa-propagate.h | 35 +- gcc/tree-vrp.c | 2384 ++++++++++++++++++++++++------- gcc/tree.h | 9 +- 32 files changed, 2873 insertions(+), 587 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr14341.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr20701.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr21029.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr21086.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr21090.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr21332.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/pr21458.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp01.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp02.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp03.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp04.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp05.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp06.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp07.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp08.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp09.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp10.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp11.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp12.c create mode 100644 gcc/testsuite/gcc.dg/tree-ssa/vrp13.c diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 8b879e5..c344fe0 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,111 @@ +2005-06-01 Diego Novillo + + PR 14341, PR 21332, PR 20701, PR 21029, PR 21086, PR 21090 + PR 21289, PR 21348, PR 21367, PR 21368, PR 21458. + * fold-const.c (invert_tree_comparison): Make extern. + * tree-flow.h (enum value_range_type): Move to tree-ssa-propagate. + (struct value_range_def): Limewise. + (get_value_range): Remove. + (dump_value_range): Remove. + (dump_all_value_ranges): Remove. + (debug_all_value_ranges): Remove. + (vrp_evaluate_conditional): Declare. + * tree-ssa-propagate.c (struct prop_stats_d): Add field + num_pred_folded. + (substitute_and_fold): Add argument use_ranges_p. + Update all callers. + If use_ranges_p is true, call fold_predicate_in to fold + predicates using range information. + Ignore ASSERT_EXPRs. + Change debugging output to only show statements that have been + folded. + (replace_phi_args_in): Move debugging output code from + substitute and fold. + (fold_predicate_in): New local function. + * tree-ssa-propagate.h (enum value_range_type): Move from + tree-flow.h. + (struct value_range_d): Likewise. + Add field 'equiv'. + (value_range_t): Rename from value_range. + * tree-vrp.c (found_in_subgraph): Rename from found. + (get_opposite_operand): Remove. + (struct assert_locus_d): Declare. + (assert_locus_t): Declare. + (need_assert_for): Declare. + (asserts_for): Declare. + (blocks_visited): Declare. + (vr_value): Declare. + (set_value_range): Add argument 'equiv'. + Don't drop to VARYING ranges that cover all values in the + type. + Make deep copy of equivalence set 'equiv'. + (copy_value_range): New local function. + (set_value_range_to_undefined): New local function. + (compare_values): Return -2 if either value has overflowed. + (range_includes_zero_p): New local function. + (extract_range_from_assert): Flip the predicate code if the + name being asserted is on the RHS of the predicate. + Avoid creating unnecessary symbolic ranges if the comparison + includes another name with a known numeric range. + Update the equivalnce set of the new range when asserting + EQ_EXPR predicates. + (extract_range_from_ssa_name): Update the equivalence set of + the new range with VAR. + (extract_range_from_binary_expr): Also handle TRUTH_*_EXPR. + If -fwrapv is used, set the resulting range to VARYING if the + operation overflows. Otherwise, use TYPE_MIN_VALUE and + TYPE_MAX_VALUE to represent -INF and +INF. + Fix handling of *_DIV_EXPR. + (extract_range_from_unary_expr): Handle MINUS_EXPR and + ABS_EXPR properly by switching the range around if necessary. + (extract_range_from_comparison): New local function. + (extract_range_from_expr): Call it. + (adjust_range_with_scev): Do not adjust the range if using + wrapping arithmetic (-fwrapv). + (dump_value_range): Also show equivalence set. + Show -INF and +INF for TYPE_MIN_VALUE and TYPE_MAX_VALUE. + (build_assert_expr_for): Also build ASSERT_EXPR for EQ_EXPR. + (infer_value_range): Change return value to bool. + Add arguments 'comp_code_p' and 'val_p'. + Do not attempt to infer ranges from statements that may throw. + Store the comparison code in comp_code_p. + Store the other operand to be used in the predicate in val_p. + (dump_asserts_for): New. + (debug_asserts_for): New. + (dump_all_asserts): New. + (debug_all_asserts): New. + (register_new_assert_for): New. + (register_edge_assert_for): New. + (find_conditional_asserts): New. + (find_assert_locations): New. + (process_assert_insertions_for): New. + (process_assert_insertions): New. + (insert_range_assertions): Initialize found_in_subgraph, + blocks_visited, need_assert_for and asserts_for. + Call find_assert_locations and process_assert_insertions. + (remove_range_assertions): Add more documentation. + (vrp_initialize): Change return type to void. + Do not try to guess if running VRP is worth it. + (compare_name_with_value): New. + (compare_names): New. + (vrp_evaluate_conditional): Add argument 'use_equiv_p'. If + use_equiv_p is true, call compare_names and + compare_name_with_value to compare all the ranges for every + name in the equivalence set of the predicate operands. + Update all callers. + (vrp_meet): Try harder not to derive a VARYING range. + If two values meet, the resulting equivalence set is the + intersection of the two equivalence sets. + (vrp_visit_phi_node): Call copy_value_range to get the current + range information of the LHS. + (vrp_finalize): Create a value vector representing all the + names that ended up with exactly one value in their range. + Call substitute_and_fold. + (execute_vrp): Document equivalence sets in ranges. + * tree.h (SSA_NAME_VALUE_RANGE): Remove. + (struct tree_ssa_name): Remove field value_range. + (invert_tree_comparison): Declare. + 2005-06-01 Daniel Berlin Fix PR tree-optimization/21839 diff --git a/gcc/fold-const.c b/gcc/fold-const.c index 05e016a..721701b 100644 --- a/gcc/fold-const.c +++ b/gcc/fold-const.c @@ -89,7 +89,6 @@ static tree negate_expr (tree); static tree split_tree (tree, enum tree_code, tree *, tree *, tree *, int); static tree associate_trees (tree, tree, enum tree_code, tree); static tree const_binop (enum tree_code, tree, tree, int); -static enum tree_code invert_tree_comparison (enum tree_code, bool); static enum comparison_code comparison_to_compcode (enum tree_code); static enum tree_code compcode_to_comparison (enum comparison_code); static tree combine_comparisons (enum tree_code, enum tree_code, @@ -2119,7 +2118,7 @@ pedantic_non_lvalue (tree x) comparisons, except for NE_EXPR and EQ_EXPR, so we receive a machine mode as well: if reversing the comparison is unsafe, return ERROR_MARK. */ -static enum tree_code +enum tree_code invert_tree_comparison (enum tree_code code, bool honor_nans) { if (honor_nans && flag_trapping_math) diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index 9f2246c..288242e 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,34 @@ +2005-06-01 Diego Novillo + + PR 14341, PR 21332, PR 20701, PR 21086, PR 21090 + PR 21289, PR 21348, PR 21367, PR 21368, PR 21458. + * gcc.dg/tree-ssa/pr14341.c: New test. + * gcc.dg/tree-ssa/pr14841.c: New test. + * gcc.dg/tree-ssa/pr20701.c: New test. + * gcc.dg/tree-ssa/pr21086.c: New test. + * gcc.dg/tree-ssa/pr21090.c: New test. + * gcc.dg/tree-ssa/pr21332.c: New test. + * gcc.dg/tree-ssa/pr21458.c: New test. + * gcc.dg/tree-ssa/pr21658.c: New test. + * gcc.dg/tree-ssa/vrp01.c: New test. + * gcc.dg/tree-ssa/vrp02.c: New test. + * gcc.dg/tree-ssa/vrp03.c: New test. + * gcc.dg/tree-ssa/vrp04.c: New test. + * gcc.dg/tree-ssa/vrp05.c: New test. + * gcc.dg/tree-ssa/vrp06.c: New test. + * gcc.dg/tree-ssa/vrp07.c: New test. + * gcc.dg/tree-ssa/vrp08.c: New test. + * gcc.dg/tree-ssa/vrp09.c: New test. + * gcc.dg/tree-ssa/vrp10.c: New test. + * gcc.dg/tree-ssa/vrp11.c: New test. + * gcc.dg/tree-ssa/vrp12.c: New test. + * gcc.dg/tree-ssa/vrp13.c: New test. + +2005-06-01 Alexandre Oliva + + PR 21029 + * gcc.dg/tree-ssa/pr21029.c: New test. + 2005-06-01 Roger Sayle * gfortran.dg/logint-1.f: New test case. diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr14341.c b/gcc/testsuite/gcc.dg/tree-ssa/pr14341.c new file mode 100644 index 0000000..07f51b7 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr14341.c @@ -0,0 +1,16 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +void fn_call (int); +int h(int, int); +void t() +{ + int i; + int x; + for( i = 0; i < 100000000; i++ ){ + fn_call (i < 100000000); + } +} + +/* { dg-final { scan-tree-dump-times "fn_call \\(1\\)" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr14841.c b/gcc/testsuite/gcc.dg/tree-ssa/pr14841.c index 4acc6ce..4f9c759 100644 --- a/gcc/testsuite/gcc.dg/tree-ssa/pr14841.c +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr14841.c @@ -25,5 +25,5 @@ foo (void) link_error (); } -/* { dg-final { scan-tree-dump-times "with if \\(0\\)" 1 "store_ccp"} } */ +/* { dg-final { scan-tree-dump-times "Folded statement: if " 1 "store_ccp"} } */ /* { dg-final { cleanup-tree-dump "store_ccp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr20701.c b/gcc/testsuite/gcc.dg/tree-ssa/pr20701.c new file mode 100644 index 0000000..a02c317 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr20701.c @@ -0,0 +1,30 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +typedef struct { + int code; +} *rtx; + +int +can_combine_p (rtx insn, rtx elt) +{ + rtx set; + + set = 0; + if (insn->code == 3) + set = insn; + else + { + set = elt; + if (set == 0) + return 0; + } + + if (set == 0) + return 1; + + return 0; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate.*to 0" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr21029.c b/gcc/testsuite/gcc.dg/tree-ssa/pr21029.c new file mode 100644 index 0000000..de2595a --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr21029.c @@ -0,0 +1,118 @@ +/* { dg-do run } */ +/* { dg-options "-O2 -fwrapv" } */ + +/* PR tree-optimization/21029 + + f() used to get optimized to an infinite loop by tree-vrp, because + j is assumed to be non-negative. Even though the conversion from + unsigned to signed has unspecified results if the expression value + is not representable in the signed type, the compiler itself (e.g., + the Ada front end) depends on wrap-around behavior. */ + +unsigned int f(void) { + unsigned char i = 123; + signed char j; + + do + if ((j = (signed char) i) < 0) + break; + else + i++; + while (1); + + return i; +} + +/* Now let's torture it a bit further. Narrowing conversions need + similar treatment. */ + +unsigned int f1 (void) { + unsigned short i = 123; + signed char j; + + do + if ((j = (signed char) i) < 0) + break; + else + i++; + while (1); + + return i; +} + +/* And so do widening conversions. */ + +unsigned int f2 (void) { + unsigned char i = 123; + signed short j; + + do + if ((j = (signed short) (signed char) i) < 0) + break; + else + i++; + while (1); + + return i; +} + +/* Check same-sign truncations with an increment that turns into + decrements. */ + +unsigned int f3 (void) { + signed short i = 5; + signed char j; + + do + if ((j = (signed char) i) < 0) + break; + else + i += 255; + while (1); + + return i; +} + +/* Check that the truncation above doesn't confuse the result of the + test after a widening conversion. */ + +unsigned int f4 (void) { + signed short i = -123; + signed int j; + + do + if ((j = (signed int) (signed char) i) > 0) + break; + else + i += 255; + while (1); + + return i; +} + +/* Even if we omit the widening truncation, the narrowing truncation + is implementation-defined. */ + +unsigned int f5 (void) { + signed long i = -123; + signed char j; + + do + if ((j = (signed char) i) > 0) + break; + else + i += 255; + while (1); + + return i; +} + +int main (void) { + f (); + f1 (); + f2 (); + f3 (); + f4 (); + f5 (); + return 0; +} diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr21086.c b/gcc/testsuite/gcc.dg/tree-ssa/pr21086.c new file mode 100644 index 0000000..547d1c8 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr21086.c @@ -0,0 +1,19 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +int +foo (int *p) +{ + int a = *p; + int b = p != 0; + + *p = b; + + if (b) + return a; + else + return 0; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate " 2 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr21090.c b/gcc/testsuite/gcc.dg/tree-ssa/pr21090.c new file mode 100644 index 0000000..e6b3c88 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr21090.c @@ -0,0 +1,23 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +int g, h; + +int +foo (int a) +{ + int *p; + + if (a) + p = &g; + else + p = &h; + + if (p != 0) + return 1; + else + return 0; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr21332.c b/gcc/testsuite/gcc.dg/tree-ssa/pr21332.c new file mode 100644 index 0000000..8dd75cc --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr21332.c @@ -0,0 +1,28 @@ +/* { dg-do run } */ +/* { dg-options "-O2" } */ + +// this testcase fails also on amd64: + +extern void abort (void); + +int f () +{ + return -1; +} + +int main () +{ + int b, c, i; + + b = 0; + c = f (); + if (c <= 0) + { + c = -c; + for (i = 0; i < c; i++) + b = 1; + if (!b) + abort (); + } + return 0; +} diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr21458.c b/gcc/testsuite/gcc.dg/tree-ssa/pr21458.c new file mode 100644 index 0000000..0222e7a --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr21458.c @@ -0,0 +1,20 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +extern void g (void); +extern void bar (int); + +int +foo (int a) +{ + int i; + + for (i = 1; i < 100; i++) + { + if (i) + g (); + } +} + +/* { dg-final { scan-tree-dump-times "Folding predicate.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/pr21658.c b/gcc/testsuite/gcc.dg/tree-ssa/pr21658.c index 993b493..2fa45b4 100644 --- a/gcc/testsuite/gcc.dg/tree-ssa/pr21658.c +++ b/gcc/testsuite/gcc.dg/tree-ssa/pr21658.c @@ -17,5 +17,5 @@ f (void) link_error (); } -/* { dg-final { scan-tree-dump-times "with if \\(0\\)" 1 "ccp"} } */ +/* { dg-final { scan-tree-dump-times "Folded statement: if " 1 "ccp"} } */ /* { dg-final { cleanup-tree-dump "ccp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp01.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp01.c new file mode 100644 index 0000000..1edda29 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp01.c @@ -0,0 +1,28 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +foo (int *p, int i) +{ + int x; + + if (i > 10) + { + if (p) + { + x = *p; + p = 0; + } + } + else + p = 0; + + /* This should be folded to if (1), but only if we insert an + assertion on the ELSE edge from the inner 'if (p)'. */ + if (p == 0) + return x + 1; + + return i; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate p_.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp02.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp02.c new file mode 100644 index 0000000..61cb308 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp02.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +struct A +{ + int a; + int b; +}; + +foo (struct A *p, struct A *q) +{ + int x = p->a; + if (p == q) + return q->a; + + /* We should fold this to 'if (1)' but the assertion for 'p == q' + was overwriting the assertion 'p != 0' from the first dereference + of 'p'. */ + if (p) + return x + p->b; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate p_.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp03.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp03.c new file mode 100644 index 0000000..20fcb91 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp03.c @@ -0,0 +1,35 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +struct A +{ + int a; + int b; +}; + +foo (struct A *p, struct A *q) +{ + int *r = 0; + + if (p) + { + if (p == q) + { + /* This should be folded to 'if (1)' because q is [p, p] + and p is ~[0, 0]. */ + if (q) + r = &q->a; + + /* This should be folded to 'if (1)' because q should be + ~[0, 0] and thus &q->a should be ~[0, 0]. */ + if (r) + return 5; + } + } + + return q->a; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate q_.*to 1" 1 "vrp" } } */ +/* { dg-final { scan-tree-dump-times "Folding predicate r_.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp04.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp04.c new file mode 100644 index 0000000..2632248 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp04.c @@ -0,0 +1,13 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +foo (int a, int b) +{ + if (a == b) + /* This should be folded to if (1) */ + if (a == b) + return a + b; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate a_.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp05.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp05.c new file mode 100644 index 0000000..767d3e1 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp05.c @@ -0,0 +1,20 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +foo (int k, int j) +{ + if (k >= 10) + { + if (j > k) + { + /* We should fold this to if (1). */ + if (j > 0) + return j; + } + } + + return j; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate j_.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp06.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp06.c new file mode 100644 index 0000000..a92db95 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp06.c @@ -0,0 +1,31 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +foo (int i, int j, int a) +{ + if (i >= 10) + if (i <= 30) + if (i == j) + { + a--; + + /* This should fold to 'if (0)'. */ + if (i < 0) + i = baz (); + + /* This should fold to 'if (1)'. */ + if (j > 0) + a--; + + /* This should fold to 'if (0)'. */ + if (i != j) + return 0; + } + + return i + a + j; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate i_.*to 0" 1 "vrp" } } */ +/* { dg-final { scan-tree-dump-times "Folding predicate j_.*to 1" 1 "vrp" } } */ +/* { dg-final { scan-tree-dump-times "Folding predicate i_.*to 0" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp07.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp07.c new file mode 100644 index 0000000..49df596 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp07.c @@ -0,0 +1,36 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp-details" } */ + +foo (int i, int *p) +{ + int j; + + if (i > 10) + { + if (p) + { + j = *p; + /* This should be folded to if (1) because of the parent 'if + (p)'. But the dereference of p above does not need an + assertion. */ + if (p) + return j + 1; + } + } + else + { + j = *p - 3; + /* This should be folded to if (0), because p has just been + dereferenced. But we were not inserting enough ASSERT_EXPRs + to figure it out. */ + if (!p) + return j - 4; + } + + return i; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate p_.*to 1" 1 "vrp" } } */ +/* { dg-final { scan-tree-dump-times "Folding predicate p_.*to 0" 1 "vrp" } } */ +/* { dg-final { scan-tree-dump-times "PREDICATE: p_\[0-9\] ne_expr 0B" 2 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp08.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp08.c new file mode 100644 index 0000000..5bf43fe --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp08.c @@ -0,0 +1,23 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fno-tree-fre -fdump-tree-vrp-details" } */ + +/* Compile with -fno-tree-fre -O2 to prevent CSEing *p. */ +foo (int a, int *p) +{ + int x = *p + 2; + int y = *p - 1; + int z = *p + 9; + + /* This should be folded to if (1), but only one ASSERT_EXPR should + be inserted. */ + if (p) + a = x + y + z; + else + a = x - y; + + return a; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate p_.*to 1" 1 "vrp" } } */ +/* { dg-final { scan-tree-dump-times "PREDICATE: p_. ne_expr 0" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp09.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp09.c new file mode 100644 index 0000000..6a2c983 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp09.c @@ -0,0 +1,31 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +foo (int *p) +{ + int x = baz (); + + if (p == 0) + goto L78; + else + { + x = *p; + /* This should be folded to if (1). */ + if (p) + x = x + 1; +L78: + /* This should not be folded to if (1). */ + if (p) + { + x = baz (*p); + /* This should be folded to if (1). */ + if (p) + return x + 3; + } + + return x - 3; + } +} + +/* { dg-final { scan-tree-dump-times "Folding predicate p_.. != 0B to 1" 2 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp10.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp10.c new file mode 100644 index 0000000..5a74584 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp10.c @@ -0,0 +1,25 @@ +/* { dg-do run } */ +/* { dg-options "-O2" } */ + +extern void abort (void); + +foo (int k, int j) +{ + if (k >= 10) + { + if (j > k) + { + /* We should fold this to if (0). */ + if (j < 10) + abort (); + } + } + + return j; +} + +main() +{ + foo (10, 3); + return 0; +} diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp11.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp11.c new file mode 100644 index 0000000..b9e5fca --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp11.c @@ -0,0 +1,20 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-vrp" } */ + +foo (int k, int j, int z) +{ + if (k > z) + { + if (j > k) + { + /* We should fold this to if (1). */ + if (j > z) + return j; + } + } + + return j; +} + +/* { dg-final { scan-tree-dump-times "Folding predicate.*to 1" 1 "vrp" } } */ +/* { dg-final { cleanup-tree-dump "vrp" } } */ diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp12.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp12.c new file mode 100644 index 0000000..1714a56 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp12.c @@ -0,0 +1,23 @@ +/* { dg-do link } */ +/* { dg-options -O2 } */ + +foo (int i) +{ + int x; + + x = i; + if (i < -10) + { + x = __builtin_abs (i); + /* VRP was incorrectly folding this to if (1). */ + if (x < 0) + link_error (); + } + + return x; +} + +main() +{ + foo (-30); +} diff --git a/gcc/testsuite/gcc.dg/tree-ssa/vrp13.c b/gcc/testsuite/gcc.dg/tree-ssa/vrp13.c new file mode 100644 index 0000000..cfc55d8 --- /dev/null +++ b/gcc/testsuite/gcc.dg/tree-ssa/vrp13.c @@ -0,0 +1,139 @@ +/* { dg-do run } */ +/* { dg-options -O2 } */ + +extern void abort (void); + +foo (int i, int j) +{ + int k; + + /* [-INF, -1] / [1, +INF] should not give [-1, -1]. */ + if (i <= -1) + if (j >= 1) + { + k = i / j; + if (k == -1) + abort (); + + return k; + } + + /* [-20, -10] / [2, 10] should give [-10, -1]. */ + if (i >= -20) + if (i <= -10) + if (j >= 2) + if (j <= 10) + { + k = i / j; + if (k < -10) + link_error (); + if (k > -1) + link_error (); + + return k; + } + + /* [-20, -10] / [-10, -2] should give [1, 10]. */ + if (i >= -20) + if (i <= -10) + if (j >= -10) + if (j <= -2) + { + k = i / j; + if (k < 1) + link_error (); + if (k > 10) + link_error (); + + return k; + } + + /* [-20, 10] / [2, 10] should give [-10, 5]. */ + if (i >= -20) + if (i <= 10) + if (j >= 2) + if (j <= 10) + { + k = i / j; + if (k < -10) + link_error (); + if (k > 5) + link_error (); + + return k; + } + + /* [-20, 10] / [-10, -2] should give [-5, 10]. */ + if (i >= -20) + if (i <= 10) + if (j >= -10) + if (j <= -2) + { + k = i / j; + if (k < -5) + link_error (); + if (k > 10) + link_error (); + + return k; + } + + /* [10, 20] / [2, 10] should give [1, 10]. */ + if (i >= 10) + if (i <= 20) + if (j >= 2) + if (j <= 10) + { + k = i / j; + if (k < 1) + link_error (); + if (k > 10) + link_error (); + + return k; + } + + /* [10, 20] / [-10, -2] should give [-10, -1]. */ + if (i >= 10) + if (i <= 20) + if (j >= -10) + if (j <= -2) + { + k = i / j; + if (k > -1) + link_error (); + if (k < -10) + link_error (); + + return k; + } + + abort (); +} + + +main() +{ + if (foo (-10, 5) != -2) + abort (); + + if (foo (-16, 4) != -4) + abort (); + + if (foo (-15, -5) != 3) + abort (); + + if (foo (8, 2) != 4) + abort (); + + if (foo (10, -2) != -5) + abort (); + + if (foo (20, 5) != 4) + abort (); + + if (foo (15, -3) != -5) + abort (); + + return 0; +} diff --git a/gcc/tree-flow.h b/gcc/tree-flow.h index fda3e57..2b044b8 100644 --- a/gcc/tree-flow.h +++ b/gcc/tree-flow.h @@ -83,34 +83,6 @@ struct ptr_info_def GTY(()) }; -/* Types of value ranges. */ -enum value_range_type { VR_UNDEFINED, VR_RANGE, VR_ANTI_RANGE, VR_VARYING }; - - -/* Ranges of values that can be associated with an SSA_NAME after VRP - has executed. */ -struct value_range_def GTY(()) -{ - /* Lattice value represented by this range. */ - enum value_range_type type; - - /* Minimum and maximum values represented by this range. These - values are _CST nodes that should be interpreted as follows: - - - If TYPE == VR_UNDEFINED then MIN and MAX must be NULL. - - - If TYPE == VR_RANGE then MIN holds the minimum value and - MAX holds the maximum value of the range [MIN, MAX]. - - - If TYPE == ANTI_RANGE the variable is known to NOT - take any values in the range [MIN, MAX]. */ - tree min; - tree max; -}; - -typedef struct value_range_def value_range; - - /*--------------------------------------------------------------------------- Tree annotations stored in tree_common.ann ---------------------------------------------------------------------------*/ @@ -619,12 +591,8 @@ bool fold_stmt_inplace (tree); tree widen_bitfield (tree, tree, tree); /* In tree-vrp.c */ -value_range *get_value_range (tree); -void dump_value_range (FILE *, value_range *); -void debug_value_range (value_range *); -void dump_all_value_ranges (FILE *); -void debug_all_value_ranges (void); bool expr_computes_nonzero (tree); +tree vrp_evaluate_conditional (tree, bool); /* In tree-ssa-dom.c */ extern void dump_dominator_optimization_stats (FILE *); diff --git a/gcc/tree-ssa-ccp.c b/gcc/tree-ssa-ccp.c index 4068f4b..3fe8de3 100644 --- a/gcc/tree-ssa-ccp.c +++ b/gcc/tree-ssa-ccp.c @@ -580,7 +580,7 @@ static void ccp_finalize (void) { /* Perform substitutions based on the known constant values. */ - substitute_and_fold (const_val); + substitute_and_fold (const_val, false); free (const_val); } diff --git a/gcc/tree-ssa-copy.c b/gcc/tree-ssa-copy.c index c096a71..2f10726 100644 --- a/gcc/tree-ssa-copy.c +++ b/gcc/tree-ssa-copy.c @@ -894,7 +894,7 @@ fini_copy_prop (void) copy_of[i].value = get_last_copy_of (var); } - substitute_and_fold (copy_of); + substitute_and_fold (copy_of, false); free (cached_last_copy_of); free (copy_of); diff --git a/gcc/tree-ssa-propagate.c b/gcc/tree-ssa-propagate.c index f272aa5..37c644b 100644 --- a/gcc/tree-ssa-propagate.c +++ b/gcc/tree-ssa-propagate.c @@ -773,6 +773,7 @@ struct prop_stats_d { long num_const_prop; long num_copy_prop; + long num_pred_folded; }; static struct prop_stats_d prop_stats; @@ -964,6 +965,11 @@ static void replace_phi_args_in (tree phi, prop_value_t *prop_value) { int i; + bool replaced = false; + tree prev_phi = NULL; + + if (dump_file && (dump_flags & TDF_DETAILS)) + prev_phi = unshare_expr (phi); for (i = 0; i < PHI_NUM_ARGS (phi); i++) { @@ -981,6 +987,7 @@ replace_phi_args_in (tree phi, prop_value_t *prop_value) prop_stats.num_copy_prop++; propagate_value (PHI_ARG_DEF_PTR (phi, i), val); + replaced = true; /* If we propagated a copy and this argument flows through an abnormal edge, update the replacement @@ -991,19 +998,79 @@ replace_phi_args_in (tree phi, prop_value_t *prop_value) } } } + + if (replaced && dump_file && (dump_flags & TDF_DETAILS)) + { + fprintf (dump_file, "Folded PHI node: "); + print_generic_stmt (dump_file, prev_phi, TDF_SLIM); + fprintf (dump_file, " into: "); + print_generic_stmt (dump_file, phi, TDF_SLIM); + fprintf (dump_file, "\n"); + } +} + + +/* If STMT has a predicate whose value can be computed using the value + range information computed by VRP, compute its value and return true. + Otherwise, return false. */ + +static bool +fold_predicate_in (tree stmt) +{ + tree *pred_p = NULL; + tree val; + + if (TREE_CODE (stmt) == MODIFY_EXPR + && COMPARISON_CLASS_P (TREE_OPERAND (stmt, 1))) + pred_p = &TREE_OPERAND (stmt, 1); + else if (TREE_CODE (stmt) == COND_EXPR) + pred_p = &COND_EXPR_COND (stmt); + else + return false; + + val = vrp_evaluate_conditional (*pred_p, true); + if (val) + { + if (dump_file) + { + fprintf (dump_file, "Folding predicate "); + print_generic_expr (dump_file, *pred_p, 0); + fprintf (dump_file, " to "); + print_generic_expr (dump_file, val, 0); + fprintf (dump_file, "\n"); + } + + prop_stats.num_pred_folded++; + *pred_p = val; + return true; + } + + return false; } -/* Perform final substitution and folding of propagated values. */ +/* Perform final substitution and folding of propagated values. + + PROP_VALUE[I] contains the single value that should be substituted + at every use of SSA name N_I. If PROP_VALUE is NULL, no values are + substituted. + + If USE_RANGES_P is true, statements that contain predicate + expressions are evaluated with a call to vrp_evaluate_conditional. + This will only give meaningful results when called from tree-vrp.c + (the information used by vrp_evaluate_conditional is built by the + VRP pass). */ void -substitute_and_fold (prop_value_t *prop_value) +substitute_and_fold (prop_value_t *prop_value, bool use_ranges_p) { basic_block bb; + if (prop_value == NULL && !use_ranges_p) + return; + if (dump_file && (dump_flags & TDF_DETAILS)) - fprintf (dump_file, - "\nSubstituing values and folding statements\n\n"); + fprintf (dump_file, "\nSubstituing values and folding statements\n\n"); memset (&prop_stats, 0, sizeof (prop_stats)); @@ -1013,41 +1080,51 @@ substitute_and_fold (prop_value_t *prop_value) block_stmt_iterator i; tree phi; - /* Propagate our known values into PHI nodes. */ - for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi)) - { - if (dump_file && (dump_flags & TDF_DETAILS)) - { - fprintf (dump_file, "Replaced "); - print_generic_stmt (dump_file, phi, TDF_SLIM); - } - + /* Propagate known values into PHI nodes. */ + if (prop_value) + for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi)) replace_phi_args_in (phi, prop_value); - if (dump_file && (dump_flags & TDF_DETAILS)) - { - fprintf (dump_file, " with "); - print_generic_stmt (dump_file, phi, TDF_SLIM); - fprintf (dump_file, "\n"); - } - } - for (i = bsi_start (bb); !bsi_end_p (i); bsi_next (&i)) { bool replaced_address, did_replace; + tree prev_stmt = NULL; tree stmt = bsi_stmt (i); + /* Ignore ASSERT_EXPRs. They are used by VRP to generate + range information for names and they are discarded + afterwards. */ + if (TREE_CODE (stmt) == MODIFY_EXPR + && TREE_CODE (TREE_OPERAND (stmt, 1)) == ASSERT_EXPR) + continue; + /* Replace the statement with its folded version and mark it folded. */ + did_replace = false; + replaced_address = false; if (dump_file && (dump_flags & TDF_DETAILS)) + prev_stmt = unshare_expr (stmt); + + /* If we have range information, see if we can fold + predicate expressions. */ + if (use_ranges_p) + did_replace = fold_predicate_in (stmt); + + if (prop_value) { - fprintf (dump_file, "Replaced "); - print_generic_stmt (dump_file, stmt, TDF_SLIM); + /* Only replace real uses if we couldn't fold the + statement using value range information (value range + information is not collected on virtuals, so we only + need to check this for real uses). */ + if (!did_replace) + did_replace |= replace_uses_in (stmt, &replaced_address, + prop_value); + + did_replace |= replace_vuses_in (stmt, &replaced_address, + prop_value); } - replaced_address = false; - did_replace = replace_uses_in (stmt, &replaced_address, prop_value); - did_replace |= replace_vuses_in (stmt, &replaced_address, prop_value); + /* If we made a replacement, fold and cleanup the statement. */ if (did_replace) { tree old_stmt = stmt; @@ -1068,13 +1145,15 @@ substitute_and_fold (prop_value_t *prop_value) rhs = get_rhs (stmt); if (TREE_CODE (rhs) == ADDR_EXPR) recompute_tree_invarant_for_addr_expr (rhs); - } - if (dump_file && (dump_flags & TDF_DETAILS)) - { - fprintf (dump_file, " with "); - print_generic_stmt (dump_file, stmt, TDF_SLIM); - fprintf (dump_file, "\n"); + if (dump_file && (dump_flags & TDF_DETAILS)) + { + fprintf (dump_file, "Folded statement: "); + print_generic_stmt (dump_file, prev_stmt, TDF_SLIM); + fprintf (dump_file, " into: "); + print_generic_stmt (dump_file, stmt, TDF_SLIM); + fprintf (dump_file, "\n"); + } } } } @@ -1085,6 +1164,9 @@ substitute_and_fold (prop_value_t *prop_value) prop_stats.num_const_prop); fprintf (dump_file, "Copies propagated: %6ld\n", prop_stats.num_copy_prop); + fprintf (dump_file, "Predicates folded: %6ld\n", + prop_stats.num_pred_folded); } } + #include "gt-tree-ssa-propagate.h" diff --git a/gcc/tree-ssa-propagate.h b/gcc/tree-ssa-propagate.h index f0124f4..f7265fe 100644 --- a/gcc/tree-ssa-propagate.h +++ b/gcc/tree-ssa-propagate.h @@ -73,6 +73,39 @@ struct prop_value_d { typedef struct prop_value_d prop_value_t; +/* Type of value ranges. See value_range_d for a description of these + types. */ +enum value_range_type { VR_UNDEFINED, VR_RANGE, VR_ANTI_RANGE, VR_VARYING }; + +/* Range of values that can be associated with an SSA_NAME after VRP + has executed. */ +struct value_range_d +{ + /* Lattice value represented by this range. */ + enum value_range_type type; + + /* Minimum and maximum values represented by this range. These + values should be interpreted as follows: + + - If TYPE is VR_UNDEFINED or VR_VARYING then MIN and MAX must + be NULL. + + - If TYPE == VR_RANGE then MIN holds the minimum value and + MAX holds the maximum value of the range [MIN, MAX]. + + - If TYPE == ANTI_RANGE the variable is known to NOT + take any values in the range [MIN, MAX]. */ + tree min; + tree max; + + /* Set of SSA names whose value ranges are equivalent to this one. + This set is only valid when TYPE is VR_RANGE or VR_ANTI_RANGE. */ + bitmap equiv; +}; + +typedef struct value_range_d value_range_t; + + /* Call-back functions used by the value propagation engine. */ typedef enum ssa_prop_result (*ssa_prop_visit_stmt_fn) (tree, edge *, tree *); typedef enum ssa_prop_result (*ssa_prop_visit_phi_fn) (tree); @@ -87,6 +120,6 @@ bool stmt_makes_single_load (tree); bool stmt_makes_single_store (tree); prop_value_t *get_value_loaded_by (tree, prop_value_t *); bool replace_uses_in (tree, bool *, prop_value_t *); -void substitute_and_fold (prop_value_t *); +void substitute_and_fold (prop_value_t *, bool); #endif /* _TREE_SSA_PROPAGATE_H */ diff --git a/gcc/tree-vrp.c b/gcc/tree-vrp.c index 16fa58b..e55ef37 100644 --- a/gcc/tree-vrp.c +++ b/gcc/tree-vrp.c @@ -38,8 +38,8 @@ Boston, MA 02111-1307, USA. */ #include "tree-chrec.h" /* Set of SSA names found during the dominator traversal of a - sub-graph in maybe_add_assert_expr. */ -static sbitmap found; + sub-graph in find_assert_locations. */ +static sbitmap found_in_subgraph; /* Loop structure of the program. Used to analyze scalar evolutions inside adjust_range_with_scev. */ @@ -48,20 +48,51 @@ static struct loops *cfg_loops; /* Local functions. */ static int compare_values (tree val1, tree val2); -/* Given a conditional predicate COND that has WHICH as one of its - operands, return the other operand. No error checking is done. - This helper assumes that COND is a comparison and WHICH is one of - its operands. */ - -static inline tree -get_opposite_operand (tree cond, tree which) +/* Location information for ASSERT_EXPRs. Each instance of this + structure describes an ASSERT_EXPR for an SSA name. Since a single + SSA name may have more than one assertion associated with it, these + locations are kept in a linked list attached to the corresponding + SSA name. */ +struct assert_locus_d { - if (TREE_OPERAND (cond, 0) == which) - return TREE_OPERAND (cond, 1); - else - return TREE_OPERAND (cond, 0); -} + /* Basic block where the assertion would be inserted. */ + basic_block bb; + + /* Some assertions need to be inserted on an edge (e.g., assertions + generated by COND_EXPRs). In those cases, BB will be NULL. */ + edge e; + + /* Pointer to the statement that generated this assertion. */ + block_stmt_iterator si; + + /* Predicate code for the ASSERT_EXPR. Must be COMPARISON_CLASS_P. */ + enum tree_code comp_code; + + /* Value being compared against. */ + tree val; + + /* Next node in the linked list. */ + struct assert_locus_d *next; +}; + +typedef struct assert_locus_d *assert_locus_t; + +/* If bit I is present, it means that SSA name N_i has a list of + assertions that should be inserted in the IL. */ +static bitmap need_assert_for; + +/* Array of locations lists where to insert assertions. ASSERTS_FOR[I] + holds a list of ASSERT_LOCUS_T nodes that describe where + ASSERT_EXPRs for SSA name N_I should be inserted. */ +static assert_locus_t *asserts_for; + +/* Set of blocks visited in find_assert_locations. Used to avoid + visiting the same block more than once. */ +static sbitmap blocks_visited; +/* Value range array. After propagation, VR_VALUE[I] holds the range + of values that SSA name N_I may take. */ +static value_range_t **vr_value; /* Given a comparison code, return its opposite. Note that this is *not* the same as inverting its truth value (invert_tree_comparison). Here we @@ -104,12 +135,44 @@ opposite_comparison (enum tree_code code) } -/* Set value range VR to {T, MIN, MAX}. */ +/* Return true if EXPR computes a non-zero value. */ -static inline void -set_value_range (value_range *vr, enum value_range_type t, tree min, tree max) +bool +expr_computes_nonzero (tree expr) +{ + /* Type casts won't change anything, so just strip them. */ + STRIP_NOPS (expr); + + /* Calling alloca, guarantees that the value is non-NULL. */ + if (alloca_call_p (expr)) + return true; + + /* The address of a non-weak symbol is never NULL, unless the user + has requested not to remove NULL pointer checks. */ + if (flag_delete_null_pointer_checks + && TREE_CODE (expr) == ADDR_EXPR + && DECL_P (TREE_OPERAND (expr, 0)) + && !DECL_WEAK (TREE_OPERAND (expr, 0))) + return true; + + /* IOR of any value with a nonzero value will result in a nonzero + value. */ + if (TREE_CODE (expr) == BIT_IOR_EXPR + && integer_nonzerop (TREE_OPERAND (expr, 1))) + return true; + + return false; +} + + +/* Set value range VR to {T, MIN, MAX, EQUIV}. */ + +static void +set_value_range (value_range_t *vr, enum value_range_type t, tree min, + tree max, bitmap equiv) { #if defined ENABLE_CHECKING + /* Check the validity of the range. */ if (t == VR_RANGE || t == VR_ANTI_RANGE) { int cmp; @@ -123,114 +186,171 @@ set_value_range (value_range *vr, enum value_range_type t, tree min, tree max) cmp = compare_values (min, max); gcc_assert (cmp == 0 || cmp == -1 || cmp == -2); } -#endif - if (t == VR_RANGE - && INTEGRAL_TYPE_P (TREE_TYPE (min)) - && min == TYPE_MIN_VALUE (TREE_TYPE (min)) - && max == TYPE_MAX_VALUE (TREE_TYPE (max))) - { - /* Ranges that cover all the possible values for the type decay - to VARYING. */ - vr->type = VR_VARYING; - vr->min = NULL_TREE; - vr->max = NULL_TREE; - return; - } + if (t == VR_UNDEFINED || t == VR_VARYING) + gcc_assert (min == NULL_TREE && max == NULL_TREE); + + if (t == VR_UNDEFINED || t == VR_VARYING) + gcc_assert (equiv == NULL || bitmap_empty_p (equiv)); +#endif vr->type = t; vr->min = min; vr->max = max; + + /* Since updating the equivalence set involves deep copying the + bitmaps, only do it if absolutely necessary. */ + if (vr->equiv == NULL) + vr->equiv = BITMAP_ALLOC (NULL); + + if (equiv != vr->equiv) + { + if (equiv && !bitmap_empty_p (equiv)) + bitmap_copy (vr->equiv, equiv); + else + bitmap_clear (vr->equiv); + } } -/* Similar to set_value_range but return true if any field of VR - changed from its previous value. */ +/* Copy value range FROM into value range TO. */ -static inline bool -update_value_range (value_range *vr, enum value_range_type t, tree min, - tree max) +static inline void +copy_value_range (value_range_t *to, value_range_t *from) { - bool is_new = vr->type != t || vr->min != min || vr->max != max; - if (is_new) - set_value_range (vr, t, min, max); + set_value_range (to, from->type, from->min, from->max, from->equiv); +} - return is_new; + +/* Set value range VR to a non-NULL range of type TYPE. */ + +static inline void +set_value_range_to_nonnull (value_range_t *vr, tree type) +{ + tree zero = build_int_cst (type, 0); + set_value_range (vr, VR_ANTI_RANGE, zero, zero, vr->equiv); +} + + +/* Set value range VR to a NULL range of type TYPE. */ + +static inline void +set_value_range_to_null (value_range_t *vr, tree type) +{ + tree zero = build_int_cst (type, 0); + set_value_range (vr, VR_RANGE, zero, zero, vr->equiv); +} + + +/* Set value range VR to VR_VARYING. */ + +static inline void +set_value_range_to_varying (value_range_t *vr) +{ + vr->type = VR_VARYING; + vr->min = vr->max = NULL_TREE; + if (vr->equiv) + bitmap_clear (vr->equiv); +} + + +/* Set value range VR to VR_UNDEFINED. */ + +static inline void +set_value_range_to_undefined (value_range_t *vr) +{ + vr->type = VR_UNDEFINED; + vr->min = vr->max = NULL_TREE; + if (vr->equiv) + bitmap_clear (vr->equiv); } -/* Return value range information for VAR. Create an empty range if - none existed. */ +/* Return value range information for VAR. Create an empty range + if none existed. */ -value_range * +static value_range_t * get_value_range (tree var) { - value_range *vr; + value_range_t *vr; tree sym; + unsigned ver = SSA_NAME_VERSION (var); - vr = SSA_NAME_VALUE_RANGE (var); + vr = vr_value[ver]; if (vr) return vr; /* Create a default value range. */ - vr = ggc_alloc (sizeof (*vr)); - memset ((void *) vr, 0, sizeof (*vr)); - SSA_NAME_VALUE_RANGE (var) = vr; + vr_value[ver] = vr = xmalloc (sizeof (*vr)); + memset (vr, 0, sizeof (*vr)); - /* If VAR is a default definition for a PARM_DECL, then we have to - assume a VARYING range for it. */ + /* Allocate an equivalence set. */ + vr->equiv = BITMAP_ALLOC (NULL); + + /* If VAR is a default definition, the variable can take any value + in VAR's type. */ sym = SSA_NAME_VAR (var); - if (TREE_CODE (sym) == PARM_DECL && var == var_ann (sym)->default_def) - set_value_range (vr, VR_VARYING, NULL_TREE, NULL_TREE); + if (var == var_ann (sym)->default_def) + set_value_range_to_varying (vr); return vr; } -/* Return true if value range VR involves at least one symbol. */ +/* Update the value range and equivalence set for variable VAR to + NEW_VR. Return true if NEW_VR is different from VAR's previous + value. + + NOTE: This function assumes that NEW_VR is a temporary value range + object created for the sole purpose of updating VAR's range. The + storage used by the equivalence set from NEW_VR will be freed by + this function. Do not call update_value_range when NEW_VR + is the range object associated with another SSA name. */ static inline bool -symbolic_range_p (value_range *vr) +update_value_range (tree var, value_range_t *new_vr) { - return (!is_gimple_min_invariant (vr->min) - || !is_gimple_min_invariant (vr->max)); -} + value_range_t *old_vr; + bool is_new; + + /* Update the value range, if necessary. */ + old_vr = get_value_range (var); + is_new = old_vr->type != new_vr->type + || old_vr->min != new_vr->min + || old_vr->max != new_vr->max + || (old_vr->equiv == NULL && new_vr->equiv) + || (old_vr->equiv && new_vr->equiv == NULL) + || (!bitmap_equal_p (old_vr->equiv, new_vr->equiv)); + if (is_new) + set_value_range (old_vr, new_vr->type, new_vr->min, new_vr->max, + new_vr->equiv); -/* Return true if EXPR computes a non-zero value. */ + BITMAP_FREE (new_vr->equiv); + new_vr->equiv = NULL; -bool -expr_computes_nonzero (tree expr) -{ - /* Type casts won't change anything, so just strip it. */ - STRIP_NOPS (expr); + return is_new; +} - /* Calling alloca, guarantees that the value is non-NULL. */ - if (alloca_call_p (expr)) - return true; - /* The address of a non-weak symbol is never NULL, unless the user - has requested not to remove NULL pointer checks. */ - if (flag_delete_null_pointer_checks - && TREE_CODE (expr) == ADDR_EXPR - && DECL_P (TREE_OPERAND (expr, 0)) - && !DECL_WEAK (TREE_OPERAND (expr, 0))) - return true; +/* Add VAR and VAR's equivalence set to EQUIV. */ - /* IOR of any value with a nonzero value will result in a nonzero - value. */ - if (TREE_CODE (expr) == BIT_IOR_EXPR - && integer_nonzerop (TREE_OPERAND (expr, 1))) - return true; +static void +add_equivalence (bitmap equiv, tree var) +{ + unsigned ver = SSA_NAME_VERSION (var); + value_range_t *vr = vr_value[ver]; - return false; + bitmap_set_bit (equiv, ver); + if (vr && vr->equiv) + bitmap_ior_into (equiv, vr->equiv); } /* Return true if VR is ~[0, 0]. */ static inline bool -range_is_nonnull (value_range *vr) +range_is_nonnull (value_range_t *vr) { return vr->type == VR_ANTI_RANGE && integer_zerop (vr->min) @@ -241,7 +361,7 @@ range_is_nonnull (value_range *vr) /* Return true if VR is [0, 0]. */ static inline bool -range_is_null (value_range *vr) +range_is_null (value_range_t *vr) { return vr->type == VR_RANGE && integer_zerop (vr->min) @@ -249,32 +369,42 @@ range_is_null (value_range *vr) } -/* Set value range VR to a non-NULL range of type TYPE. */ +/* Return true if value range VR involves at least one symbol. */ -static inline void -set_value_range_to_nonnull (value_range *vr, tree type) +static inline bool +symbolic_range_p (value_range_t *vr) { - tree zero = build_int_cst (type, 0); - set_value_range (vr, VR_ANTI_RANGE, zero, zero); + return (!is_gimple_min_invariant (vr->min) + || !is_gimple_min_invariant (vr->max)); } -/* Set value range VR to a NULL range of type TYPE. */ +/* Like expr_computes_nonzero, but this function uses value ranges + obtained so far. */ -static inline void -set_value_range_to_null (value_range *vr, tree type) +static bool +vrp_expr_computes_nonzero (tree expr) { - tree zero = build_int_cst (type, 0); - set_value_range (vr, VR_RANGE, zero, zero); -} + if (expr_computes_nonzero (expr)) + return true; + /* If we have an expression of the form &X->a, then the expression + is nonnull if X is nonnull. */ + if (TREE_CODE (expr) == ADDR_EXPR) + { + tree base = get_base_address (TREE_OPERAND (expr, 0)); -/* Set value range VR to VR_VARYING. */ + if (base != NULL_TREE + && TREE_CODE (base) == INDIRECT_REF + && TREE_CODE (TREE_OPERAND (base, 0)) == SSA_NAME) + { + value_range_t *vr = get_value_range (TREE_OPERAND (base, 0)); + if (range_is_nonnull (vr)) + return true; + } + } -static inline void -set_value_range_to_varying (value_range *vr) -{ - set_value_range (vr, VR_VARYING, NULL_TREE, NULL_TREE); + return false; } @@ -414,6 +544,10 @@ compare_values (tree val1, tree val2) if (!is_gimple_min_invariant (val1) || !is_gimple_min_invariant (val2)) return -2; + /* We cannot compare overflowed values. */ + if (TREE_OVERFLOW (val1) || TREE_OVERFLOW (val2)) + return -2; + if (!POINTER_TYPE_P (TREE_TYPE (val1))) return tree_int_cst_compare (val1, val2); else @@ -449,7 +583,7 @@ compare_values (tree val1, tree val2) -2 if we cannot tell either way. */ static inline int -value_inside_range (tree val, value_range *vr) +value_inside_range (tree val, value_range_t *vr) { int cmp1, cmp2; @@ -469,7 +603,7 @@ value_inside_range (tree val, value_range *vr) intersection. */ static inline bool -value_ranges_intersect_p (value_range *vr0, value_range *vr1) +value_ranges_intersect_p (value_range_t *vr0, value_range_t *vr1) { return (value_inside_range (vr1->min, vr0) == 1 || value_inside_range (vr1->max, vr0) == 1 @@ -478,38 +612,77 @@ value_ranges_intersect_p (value_range *vr0, value_range *vr1) } +/* Return true if VR includes the value zero, false otheriwse. */ + +static inline bool +range_includes_zero_p (value_range_t *vr) +{ + tree zero; + + gcc_assert (vr->type != VR_UNDEFINED + && vr->type != VR_VARYING + && !symbolic_range_p (vr)); + + zero = build_int_cst (TREE_TYPE (vr->min), 0); + return (value_inside_range (zero, vr) == 1); +} + + /* Extract value range information from an ASSERT_EXPR EXPR and store it in *VR_P. */ static void -extract_range_from_assert (value_range *vr_p, tree expr) +extract_range_from_assert (value_range_t *vr_p, tree expr) { - tree var, cond, limit, type; - value_range *var_vr; + tree var, cond, limit, min, max, type; + value_range_t *var_vr, *limit_vr; enum tree_code cond_code; var = ASSERT_EXPR_VAR (expr); cond = ASSERT_EXPR_COND (expr); - cond_code = TREE_CODE (cond); gcc_assert (COMPARISON_CLASS_P (cond)); /* Find VAR in the ASSERT_EXPR conditional. */ - limit = get_opposite_operand (cond, var); - type = TREE_TYPE (limit); + if (var == TREE_OPERAND (cond, 0)) + { + /* If the predicate is of the form VAR COMP LIMIT, then we just + take LIMIT from the RHS and use the same comparison code. */ + limit = TREE_OPERAND (cond, 1); + cond_code = TREE_CODE (cond); + } + else + { + /* If the predicate is of the form LIMIT COMP VAR, then we need + to flip around the comparison code to create the proper range + for VAR. */ + limit = TREE_OPERAND (cond, 0); + cond_code = opposite_comparison (TREE_CODE (cond)); + } + type = TREE_TYPE (limit); gcc_assert (limit != var); - /* For pointer arithmetic, we only keep track of anti-ranges - (NE_EXPR). Notice that we don't need to handle EQ_EXPR in these - cases because assertions with equalities are never generated. - The assert pass generates straight assignments in those cases. */ - if (POINTER_TYPE_P (type) && cond_code != NE_EXPR) + /* For pointer arithmetic, we only keep track of pointer equality + and inequality. */ + if (POINTER_TYPE_P (type) && cond_code != NE_EXPR && cond_code != EQ_EXPR) { set_value_range_to_varying (vr_p); return; } + /* If LIMIT is another SSA name and LIMIT has a range of its own, + try to use LIMIT's range to avoid creating symbolic ranges + unnecessarily. */ + limit_vr = (TREE_CODE (limit) == SSA_NAME) ? get_value_range (limit) : NULL; + + /* LIMIT's range is only interesting if it has any useful information. */ + if (limit_vr + && (limit_vr->type == VR_UNDEFINED + || limit_vr->type == VR_VARYING + || symbolic_range_p (limit_vr))) + limit_vr = NULL; + /* Special handling for integral types with super-types. Some FEs construct integral types derived from other types and restrict the range of values these new types may take. @@ -538,7 +711,9 @@ extract_range_from_assert (value_range *vr_p, tree expr) So, the only sensible thing we can do for now is set the resulting range to VR_VARYING. TODO, would having symbolic -INF and +INF values be worth the trouble? */ - if (TREE_TYPE (type)) + if (TREE_CODE (limit) != SSA_NAME + && INTEGRAL_TYPE_P (type) + && TREE_TYPE (type)) { if (cond_code == LE_EXPR || cond_code == LT_EXPR) { @@ -568,29 +743,125 @@ extract_range_from_assert (value_range *vr_p, tree expr) } } - if (TREE_CODE (cond) == NE_EXPR) - set_value_range (vr_p, VR_ANTI_RANGE, limit, limit); - else if (TREE_CODE (cond) == LE_EXPR) - set_value_range (vr_p, VR_RANGE, TYPE_MIN_VALUE (type), limit); - else if (TREE_CODE (cond) == LT_EXPR) + /* The new range has the same set of equivalences of VAR's range. */ + gcc_assert (vr_p->equiv == NULL); + vr_p->equiv = BITMAP_ALLOC (NULL); + add_equivalence (vr_p->equiv, var); + + /* Extract a new range based on the asserted comparison for VAR and + LIMIT's value range. Notice that if LIMIT has an anti-range, we + will only use it for equality comparisons (EQ_EXPR). For any + other kind of assertion, we cannot derive a range from LIMIT's + anti-range that can be used to describe the new range. For + instance, ASSERT_EXPR . If b_4 is ~[2, 10], + then b_4 takes on the ranges [-INF, 1] and [11, +INF]. There is + no single range for x_2 that could describe LE_EXPR, so we might + as well build the range [b_4, +INF] for it. */ + if (cond_code == EQ_EXPR) + { + enum value_range_type range_type; + + if (limit_vr) + { + range_type = limit_vr->type; + min = limit_vr->min; + max = limit_vr->max; + } + else + { + range_type = VR_RANGE; + min = limit; + max = limit; + } + + set_value_range (vr_p, range_type, min, max, vr_p->equiv); + + /* When asserting the equality VAR == LIMIT and LIMIT is another + SSA name, the new range will also inherit the equivalence set + from LIMIT. */ + if (TREE_CODE (limit) == SSA_NAME) + add_equivalence (vr_p->equiv, limit); + } + else if (cond_code == NE_EXPR) + { + /* As described above, when LIMIT's range is an anti-range and + this assertion is an inequality (NE_EXPR), then we cannot + derive anything from the anti-range. For instance, if + LIMIT's range was ~[0, 0], the assertion 'VAR != LIMIT' does + not imply that VAR's range is [0, 0]. So, in the case of + anti-ranges, we just assert the inequality using LIMIT and + not its anti-range. */ + if (limit_vr == NULL + || limit_vr->type == VR_ANTI_RANGE) + { + min = limit; + max = limit; + } + else + { + min = limit_vr->min; + max = limit_vr->max; + } + + /* If MIN and MAX cover the whole range for their type, then + just use the original LIMIT. */ + if (INTEGRAL_TYPE_P (type) + && min == TYPE_MIN_VALUE (type) + && max == TYPE_MAX_VALUE (type)) + min = max = limit; + + set_value_range (vr_p, VR_ANTI_RANGE, min, max, vr_p->equiv); + } + else if (cond_code == LE_EXPR || cond_code == LT_EXPR) { - tree one = build_int_cst (type, 1); - set_value_range (vr_p, VR_RANGE, TYPE_MIN_VALUE (type), - fold (build (MINUS_EXPR, type, limit, one))); + min = TYPE_MIN_VALUE (type); + + if (limit_vr == NULL || limit_vr->type == VR_ANTI_RANGE) + max = limit; + else + { + /* If LIMIT_VR is of the form [N1, N2], we need to build the + range [MIN, N2] for LE_EXPR and [MIN, N2 - 1] for + LT_EXPR. */ + max = limit_vr->max; + } + + /* For LT_EXPR, we create the range [MIN, MAX - 1]. */ + if (cond_code == LT_EXPR) + { + tree one = build_int_cst (type, 1); + max = fold (build (MINUS_EXPR, type, max, one)); + } + + set_value_range (vr_p, VR_RANGE, min, max, vr_p->equiv); } - else if (TREE_CODE (cond) == GE_EXPR) - set_value_range (vr_p, VR_RANGE, limit, TYPE_MAX_VALUE (type)); - else if (TREE_CODE (cond) == GT_EXPR) + else if (cond_code == GE_EXPR || cond_code == GT_EXPR) { - tree one = build_int_cst (type, 1); - set_value_range (vr_p, VR_RANGE, - fold (build (PLUS_EXPR, type, limit, one)), - TYPE_MAX_VALUE (type)); + max = TYPE_MAX_VALUE (type); + + if (limit_vr == NULL || limit_vr->type == VR_ANTI_RANGE) + min = limit; + else + { + /* If LIMIT_VR is of the form [N1, N2], we need to build the + range [N1, MAX] for GE_EXPR and [N1 + 1, MAX] for + GT_EXPR. */ + min = limit_vr->min; + } + + /* For GT_EXPR, we create the range [MIN + 1, MAX]. */ + if (cond_code == GT_EXPR) + { + tree one = build_int_cst (type, 1); + min = fold (build (PLUS_EXPR, type, min, one)); + } + + set_value_range (vr_p, VR_RANGE, min, max, vr_p->equiv); } else gcc_unreachable (); - /* If VAR already has a known range and the two ranges have a + /* If VAR already had a known range and the two ranges have a non-empty intersection, we can refine the resulting range. Since the assert expression creates an equivalency and at the same time it asserts a predicate, we can take the intersection of @@ -600,8 +871,6 @@ extract_range_from_assert (value_range *vr_p, tree expr) && vr_p->type == VR_RANGE && value_ranges_intersect_p (var_vr, vr_p)) { - tree min, max; - /* Use the larger of the two minimums. */ if (compare_values (vr_p->min, var_vr->min) == -1) min = var_vr->min; @@ -614,7 +883,7 @@ extract_range_from_assert (value_range *vr_p, tree expr) else max = vr_p->max; - set_value_range (vr_p, vr_p->type, min, max); + set_value_range (vr_p, vr_p->type, min, max, vr_p->equiv); } } @@ -633,14 +902,16 @@ extract_range_from_assert (value_range *vr_p, tree expr) always false. */ static void -extract_range_from_ssa_name (value_range *vr, tree var) +extract_range_from_ssa_name (value_range_t *vr, tree var) { - value_range *var_vr = get_value_range (var); + value_range_t *var_vr = get_value_range (var); if (var_vr->type != VR_UNDEFINED && var_vr->type != VR_VARYING) - *vr = *var_vr; + copy_value_range (vr, var_vr); else - set_value_range (vr, VR_RANGE, var, var); + set_value_range (vr, VR_RANGE, var, var, NULL); + + add_equivalence (vr->equiv, var); } @@ -648,12 +919,13 @@ extract_range_from_ssa_name (value_range *vr, tree var) the ranges of each of its operands and the expression code. */ static void -extract_range_from_binary_expr (value_range *vr, tree expr) +extract_range_from_binary_expr (value_range_t *vr, tree expr) { enum tree_code code = TREE_CODE (expr); tree op0, op1, min, max; - value_range vr0, vr1; int cmp; + value_range_t vr0 = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; + value_range_t vr1 = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; /* Not all binary expressions can be applied to ranges in a meaningful way. Handle only arithmetic operations. */ @@ -666,7 +938,12 @@ extract_range_from_binary_expr (value_range *vr, tree expr) && code != EXACT_DIV_EXPR && code != ROUND_DIV_EXPR && code != MIN_EXPR - && code != MAX_EXPR) + && code != MAX_EXPR + && code != TRUTH_ANDIF_EXPR + && code != TRUTH_ORIF_EXPR + && code != TRUTH_AND_EXPR + && code != TRUTH_OR_EXPR + && code != TRUTH_XOR_EXPR) { set_value_range_to_varying (vr); return; @@ -677,48 +954,34 @@ extract_range_from_binary_expr (value_range *vr, tree expr) op0 = TREE_OPERAND (expr, 0); if (TREE_CODE (op0) == SSA_NAME) vr0 = *(get_value_range (op0)); + else if (is_gimple_min_invariant (op0)) + set_value_range (&vr0, VR_RANGE, op0, op0, NULL); else - { - if (is_gimple_min_invariant (op0)) - set_value_range (&vr0, VR_RANGE, op0, op0); - else - set_value_range_to_varying (&vr0); - } + set_value_range_to_varying (&vr0); op1 = TREE_OPERAND (expr, 1); if (TREE_CODE (op1) == SSA_NAME) vr1 = *(get_value_range (op1)); + else if (is_gimple_min_invariant (op1)) + set_value_range (&vr1, VR_RANGE, op1, op1, NULL); else - { - if (is_gimple_min_invariant (op1)) - set_value_range (&vr1, VR_RANGE, op1, op1); - else - set_value_range_to_varying (&vr1); - } + set_value_range_to_varying (&vr1); /* If either range is UNDEFINED, so is the result. */ if (vr0.type == VR_UNDEFINED || vr1.type == VR_UNDEFINED) { - set_value_range (vr, VR_UNDEFINED, NULL_TREE, NULL_TREE); - return; - } - - /* If either range is VARYING, so is the result. */ - if (vr0.type == VR_VARYING || vr1.type == VR_VARYING) - { - set_value_range_to_varying (vr); - return; - } - - /* If the ranges are of different types, the result is VARYING. */ - if (vr0.type != vr1.type) - { - set_value_range_to_varying (vr); + set_value_range_to_undefined (vr); return; } - /* TODO. Refuse to do any symbolic range operations for now. */ - if (symbolic_range_p (&vr0) || symbolic_range_p (&vr1)) + /* Refuse to operate on VARYING ranges, ranges of different kinds + and symbolic ranges. TODO, we may be able to derive anti-ranges + in some cases. */ + if (vr0.type == VR_VARYING + || vr1.type == VR_VARYING + || vr0.type != vr1.type + || symbolic_range_p (&vr0) + || symbolic_range_p (&vr1)) { set_value_range_to_varying (vr); return; @@ -730,17 +993,12 @@ extract_range_from_binary_expr (value_range *vr, tree expr) || POINTER_TYPE_P (TREE_TYPE (op1))) { /* For pointer types, we are really only interested in asserting - whether the expression evaluates to non-NULL. FIXME. We - used to gcc_assert (code == PLUS_EXPR || code == MINUS_EXPR), - but ivopts is generating expressions with pointer - multiplication in them. */ + whether the expression evaluates to non-NULL. FIXME, we used + to gcc_assert (code == PLUS_EXPR || code == MINUS_EXPR), but + ivopts is generating expressions with pointer multiplication + in them. */ if (code == PLUS_EXPR) - { - /* Assume that pointers can never wrap around. FIXME, Is - this always safe? */ - tree zero = build_int_cst (TREE_TYPE (expr), 0); - set_value_range (vr, VR_ANTI_RANGE, zero, zero); - } + set_value_range_to_nonnull (vr, TREE_TYPE (expr)); else { /* Subtracting from a pointer, may yield 0, so just drop the @@ -753,10 +1011,20 @@ extract_range_from_binary_expr (value_range *vr, tree expr) /* For integer ranges, apply the operation to each end of the range and see what we end up with. */ - if (code == PLUS_EXPR - || code == MULT_EXPR - || code == MIN_EXPR - || code == MAX_EXPR) + if (code == TRUTH_ANDIF_EXPR + || code == TRUTH_ORIF_EXPR + || code == TRUTH_AND_EXPR + || code == TRUTH_OR_EXPR + || code == TRUTH_XOR_EXPR) + { + /* Boolean expressions cannot be folded with int_const_binop. */ + min = fold_binary (code, TREE_TYPE (expr), vr0.min, vr1.min); + max = fold_binary (code, TREE_TYPE (expr), vr0.max, vr1.max); + } + else if (code == PLUS_EXPR + || code == MULT_EXPR + || code == MIN_EXPR + || code == MAX_EXPR) { /* For operations that make the resulting range directly proportional to the original ranges, apply the operation to @@ -764,54 +1032,141 @@ extract_range_from_binary_expr (value_range *vr, tree expr) min = int_const_binop (code, vr0.min, vr1.min, 0); max = int_const_binop (code, vr0.max, vr1.max, 0); } - else - { - /* For operations that make the resulting range inversely - proportional to the original ranges (-, /), apply the - operation to the opposite ends of each range. */ - min = int_const_binop (code, vr0.min, vr1.max, 0); - max = int_const_binop (code, vr0.max, vr1.min, 0); - } - - cmp = compare_values (min, max); - if (cmp == -2 || cmp == 1) + else if (code == TRUNC_DIV_EXPR + || code == FLOOR_DIV_EXPR + || code == CEIL_DIV_EXPR + || code == EXACT_DIV_EXPR + || code == ROUND_DIV_EXPR) { - /* If the new range has its limits swapped around (MIN > MAX), - then the operation caused one of them to wrap around, mark - the new range VARYING. */ - set_value_range_to_varying (vr); - } - else - set_value_range (vr, vr0.type, min, max); -} - + tree zero; -/* Like expr_computes_nonzero, but this function uses value ranges - obtained so far. */ + /* Divisions are a bit tricky to handle, depending on the mix of + signs we have in the two range, we will need to divide + different values to get the minimum and maximum values for + the new range. If VR1 includes zero, the result is VARYING. */ + if (range_includes_zero_p (&vr1)) + { + set_value_range_to_varying (vr); + return; + } -static bool -vrp_expr_computes_nonzero (tree expr) -{ - if (expr_computes_nonzero (expr)) - return true; + /* We have three main variations to handle for VR0: all negative + values, all positive values and a mix of negative and + positive. For each of these, we need to consider if VR1 is + all negative or all positive. In total, there are 6 + combinations to handle. */ + zero = build_int_cst (TREE_TYPE (expr), 0); + if (compare_values (vr0.max, zero) == -1) + { + /* VR0 is all negative. */ + if (compare_values (vr1.min, zero) == 1) + { + /* If VR1 is all positive, the new range is obtained + with [VR0.MIN / VR1.MIN, VR0.MAX / VR1.MAX]. */ + min = int_const_binop (code, vr0.min, vr1.min, 0); + max = int_const_binop (code, vr0.max, vr1.max, 0); + } + else + { + /* If VR1 is all negative, the new range is obtained + with [VR0.MAX / VR1.MIN, VR0.MIN / VR1.MAX]. */ + gcc_assert (compare_values (vr1.max, zero) == -1); + min = int_const_binop (code, vr0.max, vr1.min, 0); + max = int_const_binop (code, vr0.min, vr1.max, 0); + } + } + else if (range_includes_zero_p (&vr0)) + { + /* VR0 is a mix of negative and positive values. */ + if (compare_values (vr1.min, zero) == 1) + { + /* If VR1 is all positive, the new range is obtained + with [VR0.MIN / VR1.MIN, VR0.MAX / VR1.MIN]. */ + min = int_const_binop (code, vr0.min, vr1.min, 0); + max = int_const_binop (code, vr0.max, vr1.min, 0); + } + else + { + /* If VR1 is all negative, the new range is obtained + with [VR0.MAX / VR1.MAX, VR0.MIN / VR1.MAX]. */ + gcc_assert (compare_values (vr1.max, zero) == -1); + min = int_const_binop (code, vr0.max, vr1.max, 0); + max = int_const_binop (code, vr0.min, vr1.max, 0); + } + } + else + { + /* VR0 is all positive. */ + gcc_assert (compare_values (vr0.min, zero) == 1); + if (compare_values (vr1.min, zero) == 1) + { + /* If VR1 is all positive, the new range is obtained + with [VR0.MIN / VR1.MAX, VR0.MAX / VR1.MIN]. */ + min = int_const_binop (code, vr0.min, vr1.max, 0); + max = int_const_binop (code, vr0.max, vr1.min, 0); + } + else + { + /* If VR1 is all negative, the new range is obtained + with [VR0.MAX / VR1.MAX, VR0.MIN / VR1.MIN]. */ + gcc_assert (compare_values (vr1.max, zero) == -1); + min = int_const_binop (code, vr0.max, vr1.max, 0); + max = int_const_binop (code, vr0.min, vr1.min, 0); + } + } + } + else if (code == MINUS_EXPR) + { + /* For MINUS_EXPR, apply the operation to the opposite ends of + each range. */ + min = int_const_binop (code, vr0.min, vr1.max, 0); + max = int_const_binop (code, vr0.max, vr1.min, 0); + } + else + gcc_unreachable (); - /* If we have an expression of the form &X->a, then the expression - is nonnull if X is nonnull. */ - if (TREE_CODE (expr) == ADDR_EXPR) + /* If MAX overflowed, then the result depends on whether we are + using wrapping arithmetic or not. */ + if (TREE_OVERFLOW (max)) { - tree base = get_base_address (TREE_OPERAND (expr, 0)); + /* If we are using wrapping arithmetic, set the result to + VARYING. */ + if (flag_wrapv) + { + set_value_range_to_varying (vr); + return; + } - if (base != NULL_TREE - && TREE_CODE (base) == INDIRECT_REF - && TREE_CODE (TREE_OPERAND (base, 0)) == SSA_NAME) + /* Otherwise, set MAX to +INF. */ + max = TYPE_MAX_VALUE (TREE_TYPE (expr)); + } + + /* If MIN overflowed, then the result depends on whether we are + using wrapping arithmetic or not. */ + if (TREE_OVERFLOW (min)) + { + /* If we are using wrapping arithmetic, set the result to + VARYING. */ + if (flag_wrapv) { - value_range *vr = get_value_range (TREE_OPERAND (base, 0)); - if (range_is_nonnull (vr)) - return true; + set_value_range_to_varying (vr); + return; } + + /* Otherwise, set MIN to -INF. */ + min = TYPE_MIN_VALUE (TREE_TYPE (expr)); } - return false; + cmp = compare_values (min, max); + if (cmp == -2 || cmp == 1) + { + /* If the new range has its limits swapped around (MIN > MAX), + then the operation caused one of them to wrap around, mark + the new range VARYING. */ + set_value_range_to_varying (vr); + } + else + set_value_range (vr, vr0.type, min, max, NULL); } @@ -819,51 +1174,53 @@ vrp_expr_computes_nonzero (tree expr) the range of its operand and the expression code. */ static void -extract_range_from_unary_expr (value_range *vr, tree expr) +extract_range_from_unary_expr (value_range_t *vr, tree expr) { enum tree_code code = TREE_CODE (expr); tree min, max, op0; - value_range vr0; int cmp; + value_range_t vr0 = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; + + /* Refuse to operate on certain unary expressions for which we + cannot easily determine a resulting range. */ + if (code == FIX_TRUNC_EXPR + || code == FIX_CEIL_EXPR + || code == FIX_FLOOR_EXPR + || code == FIX_ROUND_EXPR + || code == FLOAT_EXPR + || code == BIT_NOT_EXPR + || code == NON_LVALUE_EXPR + || code == CONJ_EXPR) + { + set_value_range_to_varying (vr); + return; + } /* Get value ranges for the operand. For constant operands, create a new value range with the operand to simplify processing. */ op0 = TREE_OPERAND (expr, 0); if (TREE_CODE (op0) == SSA_NAME) vr0 = *(get_value_range (op0)); + else if (is_gimple_min_invariant (op0)) + set_value_range (&vr0, VR_RANGE, op0, op0, NULL); else - { - if (is_gimple_min_invariant (op0)) - set_value_range (&vr0, VR_RANGE, op0, op0); - else - set_value_range_to_varying (&vr0); - } + set_value_range_to_varying (&vr0); /* If VR0 is UNDEFINED, so is the result. */ if (vr0.type == VR_UNDEFINED) { - set_value_range (vr, VR_UNDEFINED, NULL_TREE, NULL_TREE); + set_value_range_to_undefined (vr); return; } - /* If VR0 is VARYING, so is the result. */ - if (vr0.type == VR_VARYING) - { - set_value_range_to_varying (vr); - return; - } - - /* TODO. Refuse to do any symbolic range operations for now. */ - if (symbolic_range_p (&vr0)) - { - set_value_range_to_varying (vr); - return; - } - - /* If the operand is neither a pointer nor an integral type, set the - range to VARYING. TODO, we may set the range to non-zero. */ - if (!INTEGRAL_TYPE_P (TREE_TYPE (op0)) - && !POINTER_TYPE_P (TREE_TYPE (op0))) + /* Refuse to operate on varying and symbolic ranges. Also, if the + operand is neither a pointer nor an integral type, set the + resulting range to VARYING. TODO, in some cases we may be able + to derive anti-ranges (like non-zero values). */ + if (vr0.type == VR_VARYING + || (!INTEGRAL_TYPE_P (TREE_TYPE (op0)) + && !POINTER_TYPE_P (TREE_TYPE (op0))) + || symbolic_range_p (&vr0)) { set_value_range_to_varying (vr); return; @@ -899,8 +1256,52 @@ extract_range_from_unary_expr (value_range *vr, tree expr) /* Apply the operation to each end of the range and see what we end up with. */ - min = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min); - max = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max); + if (code == NEGATE_EXPR + && !TYPE_UNSIGNED (TREE_TYPE (expr))) + { + /* Negating an anti-range doesn't really do anything to it. The + new range will also not take on the same range of values + excluded by the original anti-range. */ + if (vr0.type == VR_ANTI_RANGE) + { + copy_value_range (vr, &vr0); + return; + } + + /* NEGATE_EXPR flips the range around. */ + min = (vr0.max == TYPE_MAX_VALUE (TREE_TYPE (expr))) + ? TYPE_MIN_VALUE (TREE_TYPE (expr)) + : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max); + + max = (vr0.min == TYPE_MIN_VALUE (TREE_TYPE (expr))) + ? TYPE_MAX_VALUE (TREE_TYPE (expr)) + : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min); + } + else if (code == ABS_EXPR + && !TYPE_UNSIGNED (TREE_TYPE (expr))) + { + /* ABS_EXPR may flip the range around, if the original range + included negative values. */ + min = (vr0.min == TYPE_MIN_VALUE (TREE_TYPE (expr))) + ? TYPE_MAX_VALUE (TREE_TYPE (expr)) + : fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min); + + max = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max); + + /* If the range was reversed, swap MIN and MAX. */ + if (compare_values (min, max) == 1) + { + tree t = min; + min = max; + max = t; + } + } + else + { + /* Otherwise, operate on each end of the range. */ + min = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.min); + max = fold_unary_to_constant (code, TREE_TYPE (expr), vr0.max); + } cmp = compare_values (min, max); if (cmp == -2 || cmp == 1) @@ -911,15 +1312,35 @@ extract_range_from_unary_expr (value_range *vr, tree expr) set_value_range_to_varying (vr); } else - set_value_range (vr, vr0.type, min, max); + set_value_range (vr, vr0.type, min, max, NULL); +} + + +/* Extract range information from a comparison expression EXPR based + on the range of its operand and the expression code. */ + +static void +extract_range_from_comparison (value_range_t *vr, tree expr) +{ + tree val = vrp_evaluate_conditional (expr, false); + if (val) + { + /* Since this expression was found on the RHS of an assignment, + its type may be different from _Bool. Convert VAL to EXPR's + type. */ + val = fold_convert (TREE_TYPE (expr), val); + set_value_range (vr, VR_RANGE, val, val, vr->equiv); + } + else + set_value_range_to_varying (vr); } /* Try to compute a useful range out of expression EXPR and store it - in *VR_P. */ + in *VR. */ static void -extract_range_from_expr (value_range *vr, tree expr) +extract_range_from_expr (value_range_t *vr, tree expr) { enum tree_code code = TREE_CODE (expr); @@ -927,14 +1348,21 @@ extract_range_from_expr (value_range *vr, tree expr) extract_range_from_assert (vr, expr); else if (code == SSA_NAME) extract_range_from_ssa_name (vr, expr); - else if (TREE_CODE_CLASS (code) == tcc_binary) + else if (TREE_CODE_CLASS (code) == tcc_binary + || code == TRUTH_ANDIF_EXPR + || code == TRUTH_ORIF_EXPR + || code == TRUTH_AND_EXPR + || code == TRUTH_OR_EXPR + || code == TRUTH_XOR_EXPR) extract_range_from_binary_expr (vr, expr); else if (TREE_CODE_CLASS (code) == tcc_unary) extract_range_from_unary_expr (vr, expr); + else if (TREE_CODE_CLASS (code) == tcc_comparison) + extract_range_from_comparison (vr, expr); else if (vrp_expr_computes_nonzero (expr)) set_value_range_to_nonnull (vr, TREE_TYPE (expr)); - else if (TREE_CODE (expr) == INTEGER_CST) - set_value_range (vr, VR_RANGE, expr, expr); + else if (is_gimple_min_invariant (expr)) + set_value_range (vr, VR_RANGE, expr, expr, NULL); else set_value_range_to_varying (vr); } @@ -945,7 +1373,7 @@ extract_range_from_expr (value_range *vr, tree expr) for VAR. If so, update VR with the new limits. */ static void -adjust_range_with_scev (value_range *vr, struct loop *l, tree var) +adjust_range_with_scev (value_range_t *vr, struct loop *l, tree var) { tree init, step, chrec; bool init_is_max; @@ -975,6 +1403,10 @@ adjust_range_with_scev (value_range *vr, struct loop *l, tree var) if (INTEGRAL_TYPE_P (TREE_TYPE (step)) && TYPE_UNSIGNED (TREE_TYPE (step))) return; + /* Do not adjust ranges when using wrapping arithmetic. */ + if (flag_wrapv) + return; + /* If STEP is negative, then INIT is the maximum value the range will take. Otherwise, INIT is the minimum value. */ init_is_max = (tree_int_cst_sgn (step) < 0); @@ -985,9 +1417,11 @@ adjust_range_with_scev (value_range *vr, struct loop *l, tree var) /* For VARYING or UNDEFINED ranges, just about anything we get from scalar evolutions should be better. */ if (init_is_max) - set_value_range (vr, VR_RANGE, TYPE_MIN_VALUE (TREE_TYPE (init)), init); + set_value_range (vr, VR_RANGE, TYPE_MIN_VALUE (TREE_TYPE (init)), + init, vr->equiv); else - set_value_range (vr, VR_RANGE, init, TYPE_MAX_VALUE (TREE_TYPE (init))); + set_value_range (vr, VR_RANGE, init, TYPE_MAX_VALUE (TREE_TYPE (init)), + vr->equiv); } else if (vr->type == VR_RANGE) { @@ -1024,23 +1458,24 @@ adjust_range_with_scev (value_range *vr, struct loop *l, tree var) } } - set_value_range (vr, VR_RANGE, min, max); + set_value_range (vr, VR_RANGE, min, max, vr->equiv); } } /* Given two numeric value ranges VR0, VR1 and a comparison code COMP: - - Return BOOLEAN_TRUE_NODE if VR0 COMP VR1 always returns true for all the - values in the ranges. + - Return BOOLEAN_TRUE_NODE if VR0 COMP VR1 always returns true for + all the values in the ranges. - Return BOOLEAN_FALSE_NODE if the comparison always returns false. - - Return NULL_TREE if it is not always possible to determine the value of - the comparison. */ + - Return NULL_TREE if it is not always possible to determine the + value of the comparison. */ + static tree -compare_ranges (enum tree_code comp, value_range *vr0, value_range *vr1) +compare_ranges (enum tree_code comp, value_range_t *vr0, value_range_t *vr1) { /* VARYING or UNDEFINED ranges cannot be compared. */ if (vr0->type == VR_VARYING @@ -1069,7 +1504,7 @@ compare_ranges (enum tree_code comp, value_range *vr0, value_range *vr1) if (vr0->type == VR_RANGE) { /* To simplify processing, make VR0 the anti-range. */ - value_range *tmp = vr0; + value_range_t *tmp = vr0; vr0 = vr1; vr1 = tmp; } @@ -1087,7 +1522,7 @@ compare_ranges (enum tree_code comp, value_range *vr0, value_range *vr1) operands around and change the comparison code. */ if (comp == GT_EXPR || comp == GE_EXPR) { - value_range *tmp; + value_range_t *tmp; comp = (comp == GT_EXPR) ? LT_EXPR : LE_EXPR; tmp = vr0; vr0 = vr1; @@ -1162,13 +1597,13 @@ compare_ranges (enum tree_code comp, value_range *vr0, value_range *vr1) /* Given a value range VR, a value VAL and a comparison code COMP, return - BOOLEAN_TRUE_NODE if VR COMP VR1 always returns true for all the + BOOLEAN_TRUE_NODE if VR COMP VAL always returns true for all the values in VR. Return BOOLEAN_FALSE_NODE if the comparison always returns false. Return NULL_TREE if it is not always possible to determine the value of the comparison. */ static tree -compare_range_with_value (enum tree_code comp, value_range *vr, tree val) +compare_range_with_value (enum tree_code comp, value_range_t *vr, tree val) { if (vr->type == VR_VARYING || vr->type == VR_UNDEFINED) return NULL_TREE; @@ -1271,8 +1706,18 @@ compare_range_with_value (enum tree_code comp, value_range *vr, tree val) /* Debugging dumps. */ +void dump_value_range (FILE *, value_range_t *); +void debug_value_range (value_range_t *); +void dump_all_value_ranges (FILE *); +void debug_all_value_ranges (void); +void dump_vr_equiv (FILE *, bitmap); +void debug_vr_equiv (bitmap); + + +/* Dump value range VR to FILE. */ + void -dump_value_range (FILE *file, value_range *vr) +dump_value_range (FILE *file, value_range_t *vr) { if (vr == NULL) fprintf (file, "[]"); @@ -1280,11 +1725,43 @@ dump_value_range (FILE *file, value_range *vr) fprintf (file, "UNDEFINED"); else if (vr->type == VR_RANGE || vr->type == VR_ANTI_RANGE) { + tree type = TREE_TYPE (vr->min); + fprintf (file, "%s[", (vr->type == VR_ANTI_RANGE) ? "~" : ""); - print_generic_expr (file, vr->min, 0); + + if (INTEGRAL_TYPE_P (type) + && !TYPE_UNSIGNED (type) + && vr->min == TYPE_MIN_VALUE (type)) + fprintf (file, "-INF"); + else + print_generic_expr (file, vr->min, 0); + fprintf (file, ", "); - print_generic_expr (file, vr->max, 0); + + if (INTEGRAL_TYPE_P (type) + && vr->max == TYPE_MAX_VALUE (type)) + fprintf (file, "+INF"); + else + print_generic_expr (file, vr->max, 0); + fprintf (file, "]"); + + if (vr->equiv) + { + bitmap_iterator bi; + unsigned i, c = 0; + + fprintf (file, " EQUIVALENCES: { "); + + EXECUTE_IF_SET_IN_BITMAP (vr->equiv, 0, i, bi) + { + print_generic_expr (file, ssa_name (i), 0); + fprintf (file, " "); + c++; + } + + fprintf (file, "} (%u elements)", c); + } } else if (vr->type == VR_VARYING) fprintf (file, "VARYING"); @@ -1296,7 +1773,7 @@ dump_value_range (FILE *file, value_range *vr) /* Dump value range VR to stderr. */ void -debug_value_range (value_range *vr) +debug_value_range (value_range_t *vr) { dump_value_range (stderr, vr); } @@ -1311,12 +1788,11 @@ dump_all_value_ranges (FILE *file) for (i = 0; i < num_ssa_names; i++) { - tree var = ssa_name (i); - if (var && SSA_NAME_VALUE_RANGE (var)) + if (vr_value[i]) { - print_generic_expr (file, var, 0); + print_generic_expr (file, ssa_name (i), 0); fprintf (file, ": "); - dump_value_range (file, SSA_NAME_VALUE_RANGE (var)); + dump_value_range (file, vr_value[i]); fprintf (file, "\n"); } } @@ -1334,10 +1810,6 @@ debug_all_value_ranges (void) } -/*--------------------------------------------------------------------------- - Value Range Propagation ----------------------------------------------------------------------------*/ - /* Given a COND_EXPR COND of the form 'V OP W', and an SSA name V, create a new SSA name N and return the assertion assignment 'V = ASSERT_EXPR '. */ @@ -1352,17 +1824,8 @@ build_assert_expr_for (tree cond, tree v) if (COMPARISON_CLASS_P (cond)) { - /* Build N = ASSERT_EXPR . As a special case, if the - conditional is an EQ_EXPR (V == Z), just build the assignment - N = Z. */ - if (TREE_CODE (cond) == EQ_EXPR) - { - tree other = get_opposite_operand (cond, v); - assertion = build (MODIFY_EXPR, TREE_TYPE (v), n, other); - } - else - assertion = build (MODIFY_EXPR, TREE_TYPE (v), n, - build (ASSERT_EXPR, TREE_TYPE (v), v, cond)); + tree a = build (ASSERT_EXPR, TREE_TYPE (v), v, cond); + assertion = build (MODIFY_EXPR, TREE_TYPE (v), n, a); } else if (TREE_CODE (cond) == TRUTH_NOT_EXPR) { @@ -1403,16 +1866,26 @@ fp_predicate (tree expr) } -/* Return an expression predicate that represents the range of values - that can be taken by operand OP after STMT executes. */ +/* If the range of values taken by OP can be inferred after STMT executes, + return the comparison code (COMP_CODE_P) and value (VAL_P) that + describes the inferred range. Return true if a range could be + inferred. */ -static tree -infer_value_range (tree stmt, tree op) +static bool +infer_value_range (tree stmt, tree op, enum tree_code *comp_code_p, tree *val_p) { + *val_p = NULL_TREE; + *comp_code_p = ERROR_MARK; + /* Do not attempt to infer anything in names that flow through abnormal edges. */ if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op)) - return NULL_TREE; + return false; + + /* Similarly, don't infer anything from statements that may throw + exceptions. */ + if (tree_could_throw_p (stmt)) + return false; if (POINTER_TYPE_P (TREE_TYPE (op))) { @@ -1424,16 +1897,17 @@ infer_value_range (tree stmt, tree op) { /* We can only assume that a pointer dereference will yield non-NULL if -fdelete-null-pointer-checks is enabled. */ - tree null = build_int_cst (TREE_TYPE (op), 0); - tree t = build (NE_EXPR, boolean_type_node, op, null); - return t; + *val_p = build_int_cst (TREE_TYPE (op), 0); + *comp_code_p = NE_EXPR; + return true; } } - return NULL_TREE; + return false; } +#if 0 /* Return true if OP is the result of an ASSERT_EXPR that tests the same condition as COND. */ @@ -1606,14 +2080,594 @@ maybe_add_assert_expr (basic_block bb) edge_iterator ei; edge e; - FOR_EACH_EDGE (e, ei, bb->succs) - if (!(e->flags & EDGE_ABNORMAL)) - { - tree t = build_assert_expr_for (cond, op); - bsi_insert_on_edge (e, t); - added = true; - break; - } + FOR_EACH_EDGE (e, ei, bb->succs) + if (!(e->flags & EDGE_ABNORMAL)) + { + tree t = build_assert_expr_for (cond, op); + bsi_insert_on_edge (e, t); + added = true; + break; + } + } + } + + /* Remember the last statement of the block. */ + last = stmt; + } + + /* Step 3. If BB's last statement is a conditional expression + involving integer operands, recurse into each of the sub-graphs + rooted at BB to determine if we need to add ASSERT_EXPRs. + Notice that we only care about the first operand of the + conditional. Adding assertions for both operands may actually + hinder VRP. FIXME, add example. */ + if (last + && TREE_CODE (last) == COND_EXPR + && !fp_predicate (COND_EXPR_COND (last)) + && !ZERO_SSA_OPERANDS (last, SSA_OP_USE)) + { + edge e; + edge_iterator ei; + tree op, cond; + basic_block son; + ssa_op_iter iter; + + cond = COND_EXPR_COND (last); + + /* Get just the first use operand. */ + FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) + break; + gcc_assert (op != NULL); + + /* Do not attempt to infer anything in names that flow through + abnormal edges. */ + if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op)) + return false; + + /* Remove the COND_EXPR operand from the FOUND bitmap. + Otherwise, when we finish traversing each of the sub-graphs, + we won't know whether the variables were found in the + sub-graphs or if they had been found in a block upstream from + BB. */ + RESET_BIT (found, SSA_NAME_VERSION (op)); + + /* Look for uses of the operands in each of the sub-graphs + rooted at BB. We need to check each of the outgoing edges + separately, so that we know what kind of ASSERT_EXPR to + insert. */ + FOR_EACH_EDGE (e, ei, bb->succs) + { + /* If BB strictly dominates the sub-graph at E->DEST, + recurse into it. */ + if (e->dest != bb + && dominated_by_p (CDI_DOMINATORS, e->dest, bb)) + added |= maybe_add_assert_expr (e->dest); + + /* Once we traversed the sub-graph, check if any block inside + used either of the predicate's operands. If so, add the + appropriate ASSERT_EXPR. */ + if (TEST_BIT (found, SSA_NAME_VERSION (op))) + { + /* We found a use of OP in the sub-graph rooted at + E->DEST. Add an ASSERT_EXPR according to whether + E goes to THEN_CLAUSE or ELSE_CLAUSE. */ + tree c, t; + + if (e->flags & EDGE_TRUE_VALUE) + c = unshare_expr (cond); + else if (e->flags & EDGE_FALSE_VALUE) + c = invert_truthvalue (cond); + else + gcc_unreachable (); + + t = build_assert_expr_for (c, op); + bsi_insert_on_edge (e, t); + added = true; + } + } + + /* Finally, mark all the COND_EXPR operands as found. */ + SET_BIT (found, SSA_NAME_VERSION (op)); + + /* Recurse into the dominator children of BB that are not BB's + immediate successors. Note that we have already visited BB's + other dominator children above. */ + for (son = first_dom_son (CDI_DOMINATORS, bb); + son; + son = next_dom_son (CDI_DOMINATORS, son)) + { + if (find_edge (bb, son) == NULL) + added |= maybe_add_assert_expr (son); + } + } + else + { + /* Step 4. Recurse into the dominator children of BB. */ + basic_block son; + + for (son = first_dom_son (CDI_DOMINATORS, bb); + son; + son = next_dom_son (CDI_DOMINATORS, son)) + added |= maybe_add_assert_expr (son); + } + + return added; +} +#endif + + +void dump_asserts_for (FILE *, tree); +void debug_asserts_for (tree); +void dump_all_asserts (FILE *); +void debug_all_asserts (void); + +/* Dump all the registered assertions for NAME to FILE. */ + +void +dump_asserts_for (FILE *file, tree name) +{ + assert_locus_t loc; + + fprintf (file, "Assertions to be inserted for "); + print_generic_expr (file, name, 0); + fprintf (file, "\n"); + + loc = asserts_for[SSA_NAME_VERSION (name)]; + while (loc) + { + fprintf (file, "\t"); + print_generic_expr (file, bsi_stmt (loc->si), 0); + fprintf (file, "\n\tBB #%d", loc->bb->index); + if (loc->e) + { + fprintf (file, "\n\tEDGE %d->%d", loc->e->src->index, + loc->e->dest->index); + dump_edge_info (file, loc->e, 0); + } + fprintf (file, "\n\tPREDICATE: "); + print_generic_expr (file, name, 0); + fprintf (file, " %s ", tree_code_name[(int)loc->comp_code]); + print_generic_expr (file, loc->val, 0); + fprintf (file, "\n\n"); + loc = loc->next; + } + + fprintf (file, "\n"); +} + + +/* Dump all the registered assertions for NAME to stderr. */ + +void +debug_asserts_for (tree name) +{ + dump_asserts_for (stderr, name); +} + + +/* Dump all the registered assertions for all the names to FILE. */ + +void +dump_all_asserts (FILE *file) +{ + unsigned i; + bitmap_iterator bi; + + fprintf (file, "\nASSERT_EXPRs to be inserted\n\n"); + EXECUTE_IF_SET_IN_BITMAP (need_assert_for, 0, i, bi) + dump_asserts_for (file, ssa_name (i)); + fprintf (file, "\n"); +} + + +/* Dump all the registered assertions for all the names to stderr. */ + +void +debug_all_asserts (void) +{ + dump_all_asserts (stderr); +} + + +/* If NAME doesn't have an ASSERT_EXPR registered for asserting + 'NAME COMP_CODE VAL' at a location that dominates block BB or + E->DEST, then register this location as a possible insertion point + for ASSERT_EXPR . + + BB, E and SI provide the exact insertion point for the new + ASSERT_EXPR. If BB is NULL, then the ASSERT_EXPR is to be inserted + on edge E. Otherwise, if E is NULL, the ASSERT_EXPR is inserted on + BB. If SI points to a COND_EXPR or a SWITCH_EXPR statement, then E + must not be NULL. */ + +static void +register_new_assert_for (tree name, + enum tree_code comp_code, + tree val, + basic_block bb, + edge e, + block_stmt_iterator si) +{ + assert_locus_t n, loc, last_loc; + bool found; + basic_block dest_bb; + +#if defined ENABLE_CHECKING + gcc_assert (bb == NULL || e == NULL); + + if (e == NULL) + gcc_assert (TREE_CODE (bsi_stmt (si)) != COND_EXPR + && TREE_CODE (bsi_stmt (si)) != SWITCH_EXPR); +#endif + + /* The new assertion A will be inserted at BB or E. We need to + determine if the new location is dominated by a previously + registered location for A. If we are doing an edge insertion, + assume that A will be inserted at E->DEST. Note that this is not + necessarily true. + + If E is a critical edge, it will be split. But even if E is + split, the new block will dominate the same set of blocks that + E->DEST dominates. + + The reverse, however, is not true, blocks dominated by E->DEST + will not be dominated by the new block created to split E. So, + if the insertion location is on a critical edge, we will not use + the new location to move another assertion previously registered + at a block dominated by E->DEST. */ + dest_bb = (bb) ? bb : e->dest; + + /* If NAME already has an ASSERT_EXPR registered for COMP_CODE and + VAL at a block dominating DEST_BB, then we don't need to insert a new + one. Similarly, if the same assertion already exists at a block + dominated by DEST_BB and the new location is not on a critical + edge, then update the existing location for the assertion (i.e., + move the assertion up in the dominance tree). + + Note, this is implemented as a simple linked list because there + should not be more than a handful of assertions registered per + name. If this becomes a performance problem, a table hashed by + COMP_CODE and VAL could be implemented. */ + loc = asserts_for[SSA_NAME_VERSION (name)]; + last_loc = loc; + found = false; + while (loc) + { + if (loc->comp_code == comp_code + && (loc->val == val + || operand_equal_p (loc->val, val, 0))) + { + /* If the assertion NAME COMP_CODE VAL has already been + registered at a basic block that dominates DEST_BB, then + we don't need to insert the same assertion again. Note + that we don't check strict dominance here to avoid + replicating the same assertion inside the same basic + block more than once (e.g., when a pointer is + dereferenced several times inside a block). + + An exception to this rule are edge insertions. If the + new assertion is to be inserted on edge E, then it will + dominate all the other insertions that we may want to + insert in DEST_BB. So, if we are doing an edge + insertion, don't do this dominance check. */ + if (e == NULL + && dominated_by_p (CDI_DOMINATORS, dest_bb, loc->bb)) + return; + + /* Otherwise, if E is not a critical edge and DEST_BB + dominates the existing location for the assertion, move + the assertion up in the dominance tree by updating its + location information. */ + if ((e == NULL || !EDGE_CRITICAL_P (e)) + && dominated_by_p (CDI_DOMINATORS, loc->bb, dest_bb)) + { + loc->bb = dest_bb; + loc->e = e; + loc->si = si; + return; + } + } + + /* Update the last node of the list and move to the next one. */ + last_loc = loc; + loc = loc->next; + } + + /* If we didn't find an assertion already registered for + NAME COMP_CODE VAL, add a new one at the end of the list of + assertions associated with NAME. */ + n = xmalloc (sizeof (*n)); + n->bb = dest_bb; + n->e = e; + n->si = si; + n->comp_code = comp_code; + n->val = val; + n->next = NULL; + + if (last_loc) + last_loc->next = n; + else + asserts_for[SSA_NAME_VERSION (name)] = n; + + bitmap_set_bit (need_assert_for, SSA_NAME_VERSION (name)); +} + + +/* Try to register an edge assertion for SSA name NAME on edge E for + the conditional jump pointed by SI. Return true if an assertion + for NAME could be registered. */ + +static bool +register_edge_assert_for (tree name, edge e, block_stmt_iterator si) +{ + tree val, stmt; + enum tree_code comp_code; + + stmt = bsi_stmt (si); + + /* Do not attempt to infer anything in names that flow through + abnormal edges. */ + if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (name)) + return false; + + /* If NAME was not found in the sub-graph reachable from E, then + there's nothing to do. */ + if (!TEST_BIT (found_in_subgraph, SSA_NAME_VERSION (name))) + return false; + + /* We found a use of NAME in the sub-graph rooted at E->DEST. + Register an assertion for NAME according to the value that NAME + takes on edge E. */ + if (TREE_CODE (stmt) == COND_EXPR) + { + /* If BB ends in a COND_EXPR then NAME then we should insert + the original predicate on EDGE_TRUE_VALUE and the + opposite predicate on EDGE_FALSE_VALUE. */ + tree cond = COND_EXPR_COND (stmt); + bool is_else_edge = (e->flags & EDGE_FALSE_VALUE) != 0; + + /* Predicates may be a single SSA name or NAME OP VAL. */ + if (cond == name) + { + /* If the predicate is a name, it must be NAME, in which + case we create the predicate NAME == true or + NAME == false accordingly. */ + comp_code = EQ_EXPR; + val = (is_else_edge) ? boolean_false_node : boolean_true_node; + } + else + { + /* Otherwise, we have a comparison of the form NAME COMP VAL + or VAL COMP NAME. */ + if (name == TREE_OPERAND (cond, 1)) + { + /* If the predicate is of the form VAL COMP NAME, flip + COMP around because we need to register NAME as the + first operand in the predicate. */ + comp_code = opposite_comparison (TREE_CODE (cond)); + val = TREE_OPERAND (cond, 0); + } + else + { + /* The comparison is of the form NAME COMP VAL, so the + comparison code remains unchanged. */ + comp_code = TREE_CODE (cond); + val = TREE_OPERAND (cond, 1); + } + + /* If we are inserting the assertion on the ELSE edge, we + need to invert the sign comparison. */ + if (is_else_edge) + comp_code = invert_tree_comparison (comp_code, 0); + } + } + else + { + /* FIXME. Handle SWITCH_EXPR. */ + gcc_unreachable (); + } + + register_new_assert_for (name, comp_code, val, NULL, e, si); + return true; +} + + +static bool find_assert_locations (basic_block bb); + +/* Determine whether the outgoing edges of BB should receive an + ASSERT_EXPR for each of the operands of BB's last statement. The + last statement of BB must be a COND_EXPR or a SWITCH_EXPR. + + If any of the sub-graphs rooted at BB have an interesting use of + the predicate operands, an assert location node is added to the + list of assertions for the corresponding operands. */ + +static bool +find_conditional_asserts (basic_block bb) +{ + bool need_assert; + block_stmt_iterator last_si; + tree op, last; + edge_iterator ei; + edge e; + ssa_op_iter iter; + + need_assert = false; + last_si = bsi_last (bb); + last = bsi_stmt (last_si); + + /* Look for uses of the operands in each of the sub-graphs + rooted at BB. We need to check each of the outgoing edges + separately, so that we know what kind of ASSERT_EXPR to + insert. */ + FOR_EACH_EDGE (e, ei, bb->succs) + { + if (e->dest == bb) + continue; + + /* Remove the COND_EXPR operands from the FOUND_IN_SUBGRAPH bitmap. + Otherwise, when we finish traversing each of the sub-graphs, we + won't know whether the variables were found in the sub-graphs or + if they had been found in a block upstream from BB. */ + FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) + RESET_BIT (found_in_subgraph, SSA_NAME_VERSION (op)); + + /* Traverse the strictly dominated sub-graph rooted at E->DEST + to determine if any of the operands in the conditional + predicate are used. */ + if (e->dest != bb) + need_assert |= find_assert_locations (e->dest); + + /* Register the necessary assertions for each operand in the + conditional predicate. */ + FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) + need_assert |= register_edge_assert_for (op, e, last_si); + } + + /* Finally, indicate that we have found the operands in the + conditional. */ + FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) + SET_BIT (found_in_subgraph, SSA_NAME_VERSION (op)); + + return need_assert; +} + + +/* Traverse all the statements in block BB looking for statements that + may generate useful assertions for the SSA names in their operand. + If a statement produces a useful assertion A for name N_i, then the + list of assertions already generated for N_i is scanned to + determine if A is actually needed. + + If N_i already had the assertion A at a location dominating the + current location, then nothing needs to be done. Otherwise, the + new location for A is recorded instead. + + 1- For every statement S in BB, all the variables used by S are + added to bitmap FOUND_IN_SUBGRAPH. + + 2- If statement S uses an operand N in a way that exposes a known + value range for N, then if N was not already generated by an + ASSERT_EXPR, create a new assert location for N. For instance, + if N is a pointer and the statement dereferences it, we can + assume that N is not NULL. + + 3- COND_EXPRs are a special case of #2. We can derive range + information from the predicate but need to insert different + ASSERT_EXPRs for each of the sub-graphs rooted at the + conditional block. If the last statement of BB is a conditional + expression of the form 'X op Y', then + + a) Remove X and Y from the set FOUND_IN_SUBGRAPH. + + b) If the conditional is the only entry point to the sub-graph + corresponding to the THEN_CLAUSE, recurse into it. On + return, if X and/or Y are marked in FOUND_IN_SUBGRAPH, then + an ASSERT_EXPR is added for the corresponding variable. + + c) Repeat step (b) on the ELSE_CLAUSE. + + d) Mark X and Y in FOUND_IN_SUBGRAPH. + + For instance, + + if (a == 9) + b = a; + else + b = c + 1; + + In this case, an assertion on the THEN clause is useful to + determine that 'a' is always 9 on that edge. However, an assertion + on the ELSE clause would be unnecessary. + + 4- If BB does not end in a conditional expression, then we recurse + into BB's dominator children. + + At the end of the recursive traversal, every SSA name will have a + list of locations where ASSERT_EXPRs should be added. When a new + location for name N is found, it is registered by calling + register_new_assert_for. That function keeps track of all the + registered assertions to prevent adding unnecessary assertions. + For instance, if a pointer P_4 is dereferenced more than once in a + dominator tree, only the location dominating all the dereference of + P_4 will receive an ASSERT_EXPR. + + If this function returns true, then it means that there are names + for which we need to generate ASSERT_EXPRs. Those assertions are + inserted by process_assert_insertions. + + TODO. Handle SWITCH_EXPR. */ + +static bool +find_assert_locations (basic_block bb) +{ + block_stmt_iterator si; + tree last, phi; + bool need_assert; + basic_block son; + + if (TEST_BIT (blocks_visited, bb->index)) + return false; + + SET_BIT (blocks_visited, bb->index); + + need_assert = false; + + /* Traverse all PHI nodes in BB marking used operands. */ + for (phi = phi_nodes (bb); phi; phi = PHI_CHAIN (phi)) + { + use_operand_p arg_p; + ssa_op_iter i; + + FOR_EACH_PHI_ARG (arg_p, phi, i, SSA_OP_USE) + { + tree arg = USE_FROM_PTR (arg_p); + if (TREE_CODE (arg) == SSA_NAME) + { + gcc_assert (is_gimple_reg (PHI_RESULT (phi))); + SET_BIT (found_in_subgraph, SSA_NAME_VERSION (arg)); + } + } + } + + /* Traverse all the statements in BB marking used names and looking + for statements that may infer assertions for their used operands. */ + last = NULL_TREE; + for (si = bsi_start (bb); !bsi_end_p (si); bsi_next (&si)) + { + tree stmt, op; + ssa_op_iter i; + + stmt = bsi_stmt (si); + + /* See if we can derive an assertion for any of STMT's operands. */ + FOR_EACH_SSA_TREE_OPERAND (op, stmt, i, SSA_OP_USE) + { + tree value; + enum tree_code comp_code; + + /* Mark OP in bitmap FOUND_IN_SUBGRAPH. If STMT is inside + the sub-graph of a conditional block, when we return from + this recursive walk, our parent will use the + FOUND_IN_SUBGRAPH bitset to determine if one of the + operands it was looking for was present in the sub-graph. */ + SET_BIT (found_in_subgraph, SSA_NAME_VERSION (op)); + + /* If OP is used only once, namely in this STMT, don't + bother creating an ASSERT_EXPR for it. Such an + ASSERT_EXPR would do nothing but increase compile time. + Experiments show that with this simple check, we can save + more than 20% of ASSERT_EXPRs. */ + if (has_single_use (op)) + continue; + + /* If OP is used in such a way that we can infer a value + range for it, and we don't find a previous assertion for + it, create a new assertion location node for OP. */ + if (infer_value_range (stmt, op, &comp_code, &value)) + { + register_new_assert_for (op, comp_code, value, bb, NULL, si); + need_assert = true; } } @@ -1621,103 +2675,112 @@ maybe_add_assert_expr (basic_block bb) last = stmt; } - /* Step 3. If BB's last statement is a conditional expression + /* If BB's last statement is a conditional expression involving integer operands, recurse into each of the sub-graphs - rooted at BB to determine if we need to add ASSERT_EXPRs. - Notice that we only care about the first operand of the - conditional. Adding assertions for both operands may actually - hinder VRP. FIXME, add example. */ + rooted at BB to determine if we need to add ASSERT_EXPRs. */ if (last && TREE_CODE (last) == COND_EXPR && !fp_predicate (COND_EXPR_COND (last)) && !ZERO_SSA_OPERANDS (last, SSA_OP_USE)) + need_assert |= find_conditional_asserts (bb); + + /* Recurse into the dominator children of BB. */ + for (son = first_dom_son (CDI_DOMINATORS, bb); + son; + son = next_dom_son (CDI_DOMINATORS, son)) + need_assert |= find_assert_locations (son); + + return need_assert; +} + + +/* Create an ASSERT_EXPR for NAME and insert it in the location + indicated by LOC. Return true if we made any edge insertions. */ + +static bool +process_assert_insertions_for (tree name, assert_locus_t loc) +{ + /* Build the comparison expression NAME_i COMP_CODE VAL. */ + tree stmt, cond, assert_expr; + edge_iterator ei; + edge e; + + cond = build (loc->comp_code, boolean_type_node, name, loc->val); + assert_expr = build_assert_expr_for (cond, name); + + if (loc->e) { - edge e; - edge_iterator ei; - tree op, cond; - basic_block son; - ssa_op_iter iter; - - cond = COND_EXPR_COND (last); + /* We have been asked to insert the assertion on an edge. This + is used only by COND_EXPR and SWITCH_EXPR assertions. */ +#if defined ENABLE_CHECKING + gcc_assert (TREE_CODE (bsi_stmt (loc->si)) == COND_EXPR + || TREE_CODE (bsi_stmt (loc->si)) == SWITCH_EXPR); +#endif - /* Get just the first use operand. */ - FOR_EACH_SSA_TREE_OPERAND (op, last, iter, SSA_OP_USE) - break; - gcc_assert (op != NULL); + bsi_insert_on_edge (loc->e, assert_expr); + return true; + } - /* Do not attempt to infer anything in names that flow through - abnormal edges. */ - if (SSA_NAME_OCCURS_IN_ABNORMAL_PHI (op)) - return false; + /* Otherwise, we can insert right after LOC->SI iff the + statement must not be the last statement in the block. */ + stmt = bsi_stmt (loc->si); + if (!stmt_ends_bb_p (stmt)) + { + bsi_insert_after (&loc->si, assert_expr, BSI_SAME_STMT); + return false; + } - /* Remove the COND_EXPR operand from the FOUND bitmap. - Otherwise, when we finish traversing each of the sub-graphs, - we won't know whether the variables were found in the - sub-graphs or if they had been found in a block upstream from - BB. */ - RESET_BIT (found, SSA_NAME_VERSION (op)); + /* If STMT must be the last statement in BB, we can only insert new + assertions on the non-abnormal edge out of BB. Note that since + STMT is not control flow, there may only be one non-abnormal edge + out of BB. */ + FOR_EACH_EDGE (e, ei, loc->bb->succs) + if (!(e->flags & EDGE_ABNORMAL)) + { + bsi_insert_on_edge (e, assert_expr); + return true; + } - /* Look for uses of the operands in each of the sub-graphs - rooted at BB. We need to check each of the outgoing edges - separately, so that we know what kind of ASSERT_EXPR to - insert. */ - FOR_EACH_EDGE (e, ei, bb->succs) - { - /* If BB strictly dominates the sub-graph at E->DEST, - recurse into it. */ - if (e->dest != bb - && dominated_by_p (CDI_DOMINATORS, e->dest, bb)) - added |= maybe_add_assert_expr (e->dest); + gcc_unreachable (); +} - /* Once we traversed the sub-graph, check if any block inside - used either of the predicate's operands. If so, add the - appropriate ASSERT_EXPR. */ - if (TEST_BIT (found, SSA_NAME_VERSION (op))) - { - /* We found a use of OP in the sub-graph rooted at - E->DEST. Add an ASSERT_EXPR according to whether - E goes to THEN_CLAUSE or ELSE_CLAUSE. */ - tree c, t; - if (e->flags & EDGE_TRUE_VALUE) - c = unshare_expr (cond); - else if (e->flags & EDGE_FALSE_VALUE) - c = invert_truthvalue (cond); - else - gcc_unreachable (); +/* Process all the insertions registered for every name N_i registered + in NEED_ASSERT_FOR. The list of assertions to be inserted are + found in ASSERTS_FOR[i]. */ - t = build_assert_expr_for (c, op); - bsi_insert_on_edge (e, t); - added = true; - } - } +static void +process_assert_insertions (void) +{ + unsigned i; + bitmap_iterator bi; + bool update_edges_p = false; + int num_asserts = 0; - /* Finally, mark all the COND_EXPR operands as found. */ - SET_BIT (found, SSA_NAME_VERSION (op)); + if (dump_file && (dump_flags & TDF_DETAILS)) + dump_all_asserts (dump_file); - /* Recurse into the dominator children of BB that are not BB's - immediate successors. Note that we have already visited BB's - other dominator children above. */ - for (son = first_dom_son (CDI_DOMINATORS, bb); - son; - son = next_dom_son (CDI_DOMINATORS, son)) + EXECUTE_IF_SET_IN_BITMAP (need_assert_for, 0, i, bi) + { + assert_locus_t loc = asserts_for[i]; + gcc_assert (loc); + + while (loc) { - if (find_edge (bb, son) == NULL) - added |= maybe_add_assert_expr (son); + assert_locus_t next = loc->next; + update_edges_p |= process_assert_insertions_for (ssa_name (i), loc); + free (loc); + loc = next; + num_asserts++; } } - else - { - /* Step 4. Recurse into the dominator children of BB. */ - basic_block son; - for (son = first_dom_son (CDI_DOMINATORS, bb); - son; - son = next_dom_son (CDI_DOMINATORS, son)) - added |= maybe_add_assert_expr (son); - } + if (update_edges_p) + bsi_commit_edge_inserts (); - return added; + if (dump_file && (dump_flags & TDF_STATS)) + fprintf (dump_file, "\nNumber of ASSERT_EXPR expressions inserted: %d\n\n", + num_asserts); } @@ -1760,19 +2823,26 @@ insert_range_assertions (void) edge_iterator ei; bool update_ssa_p; - found = sbitmap_alloc (num_ssa_names); - sbitmap_zero (found); + found_in_subgraph = sbitmap_alloc (num_ssa_names); + sbitmap_zero (found_in_subgraph); + + blocks_visited = sbitmap_alloc (last_basic_block); + sbitmap_zero (blocks_visited); + + need_assert_for = BITMAP_ALLOC (NULL); + asserts_for = xmalloc (num_ssa_names * sizeof (assert_locus_t)); + memset (asserts_for, 0, num_ssa_names * sizeof (assert_locus_t)); calculate_dominance_info (CDI_DOMINATORS); update_ssa_p = false; FOR_EACH_EDGE (e, ei, ENTRY_BLOCK_PTR->succs) - if (maybe_add_assert_expr (e->dest)) + if (find_assert_locations (e->dest)) update_ssa_p = true; if (update_ssa_p) { - bsi_commit_edge_inserts (); + process_assert_insertions (); update_ssa (TODO_update_ssa_no_phi); } @@ -1782,11 +2852,27 @@ insert_range_assertions (void) dump_function_to_file (current_function_decl, dump_file, dump_flags); } - sbitmap_free (found); + sbitmap_free (found_in_subgraph); + free (asserts_for); + BITMAP_FREE (need_assert_for); } -/* Convert range assertion expressions into copies. FIXME, explain why. */ +/* Convert range assertion expressions into the implied copies. + + FIXME, this will eventually lead to copy propagation removing the + names that had useful range information attached to them. For + instance, if we had the assertion N_i = ASSERT_EXPR 3>, + then N_i will have the range [3, +INF]. + + However, by converting the assertion into the implied copy + operation N_i = N_j, we will then copy-propagate N_j into the uses + of N_i and lose the range information. We may want to hold on to + ASSERT_EXPRs a little while longer as the ranges could be used in + things like jump threading. + + The problem with keeping ASSERT_EXPRs around is that passes after + VRP need to handle them appropriately. */ static void remove_range_assertions (void) @@ -1843,15 +2929,13 @@ stmt_interesting_for_vrp (tree stmt) is worth running (i.e. if we found any statements that could benefit from range information). */ -static bool +static void vrp_initialize (void) { basic_block bb; - bool do_vrp; - /* If we don't find any ASSERT_EXPRs in the code, there's no point - running VRP. */ - do_vrp = false; + vr_value = xmalloc (num_ssa_names * sizeof (value_range_t *)); + memset (vr_value, 0, num_ssa_names * sizeof (value_range_t *)); FOR_EACH_BB (bb) { @@ -1884,16 +2968,10 @@ vrp_initialize (void) } else { - if (TREE_CODE (stmt) == MODIFY_EXPR - && TREE_CODE (TREE_OPERAND (stmt, 1)) == ASSERT_EXPR) - do_vrp = true; - DONT_SIMULATE_AGAIN (stmt) = false; } } } - - return do_vrp; } @@ -1914,10 +2992,9 @@ vrp_visit_assignment (tree stmt, tree *output_p) && (INTEGRAL_TYPE_P (TREE_TYPE (lhs)) || POINTER_TYPE_P (TREE_TYPE (lhs)))) { - value_range *vr, new_vr; struct loop *l; - - vr = get_value_range (lhs); + value_range_t new_vr = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; + extract_range_from_expr (&new_vr, rhs); /* If STMT is inside a loop, we may be able to know something @@ -1926,16 +3003,16 @@ vrp_visit_assignment (tree stmt, tree *output_p) if (cfg_loops && (l = loop_containing_stmt (stmt))) adjust_range_with_scev (&new_vr, l, lhs); - if (update_value_range (vr, new_vr.type, new_vr.min, new_vr.max)) + if (update_value_range (lhs, &new_vr)) { *output_p = lhs; if (dump_file && (dump_flags & TDF_DETAILS)) { - fprintf (dump_file, "Found new range "); - dump_value_range (dump_file, &new_vr); - fprintf (dump_file, " for "); + fprintf (dump_file, "Found new range for "); print_generic_expr (dump_file, lhs, 0); + fprintf (dump_file, ": "); + dump_value_range (dump_file, &new_vr); fprintf (dump_file, "\n\n"); } @@ -1948,7 +3025,7 @@ vrp_visit_assignment (tree stmt, tree *output_p) return SSA_PROP_NOT_INTERESTING; } - /* Every other statements produces no useful ranges. */ + /* Every other statement produces no useful ranges. */ FOR_EACH_SSA_TREE_OPERAND (def, stmt, iter, SSA_OP_DEF) set_value_range_to_varying (get_value_range (def)); @@ -1956,43 +3033,226 @@ vrp_visit_assignment (tree stmt, tree *output_p) } -/* Given a conditional predicate COND, try to determine if COND yields - true or false based on the value ranges of its operands. */ +/* Compare all the value ranges for names equivalent to VAR with VAL + using comparison code COMP. Return the same value returned by + compare_range_with_value. */ + +static tree +compare_name_with_value (enum tree_code comp, tree var, tree val) +{ + bitmap_iterator bi; + unsigned i; + bitmap e; + tree retval, t; + + t = retval = NULL_TREE; + + /* Get the set of equivalences for VAR. */ + e = get_value_range (var)->equiv; + + /* Add VAR to its own set of equivalences so that VAR's value range + is processed by this loop (otherwise, we would have to replicate + the body of the loop just to check VAR's value range). */ + bitmap_set_bit (e, SSA_NAME_VERSION (var)); + + EXECUTE_IF_SET_IN_BITMAP (e, 0, i, bi) + { + value_range_t equiv_vr = *(vr_value[i]); + + /* If name N_i does not have a valid range, use N_i as its own + range. This allows us to compare against names that may + have N_i in their ranges. */ + if (equiv_vr.type == VR_VARYING || equiv_vr.type == VR_UNDEFINED) + { + equiv_vr.type = VR_RANGE; + equiv_vr.min = ssa_name (i); + equiv_vr.max = ssa_name (i); + } + + t = compare_range_with_value (comp, &equiv_vr, val); + if (t) + { + /* All the ranges should compare the same against VAL. */ + gcc_assert (retval == NULL || t == retval); + retval = t; + } + } + + /* Remove VAR from its own equivalence set. */ + bitmap_clear_bit (e, SSA_NAME_VERSION (var)); + + if (retval) + return retval; + + /* We couldn't find a non-NULL value for the predicate. */ + return NULL_TREE; +} + + +/* Given a comparison code COMP and names N1 and N2, compare all the + ranges equivalent to N1 against all the ranges equivalente to N2 + to determine the value of N1 COMP N2. Return the same value + returned by compare_ranges. */ static tree -vrp_evaluate_conditional (tree cond) +compare_names (enum tree_code comp, tree n1, tree n2) +{ + tree t, retval; + bitmap e1, e2; + bitmap_iterator bi1, bi2; + unsigned i1, i2; + + /* Compare the ranges of every name equivalent to N1 against the + ranges of every name equivalent to N2. */ + e1 = get_value_range (n1)->equiv; + e2 = get_value_range (n2)->equiv; + + /* Add N1 and N2 to their own set of equivalences to avoid + duplicating the body of the loop just to check N1 and N2 + ranges. */ + bitmap_set_bit (e1, SSA_NAME_VERSION (n1)); + bitmap_set_bit (e2, SSA_NAME_VERSION (n2)); + + /* If the equivalence sets have a common intersection, then the two + names can be compared without checking their ranges. */ + if (bitmap_intersect_p (e1, e2)) + { + bitmap_clear_bit (e1, SSA_NAME_VERSION (n1)); + bitmap_clear_bit (e2, SSA_NAME_VERSION (n2)); + + return (comp == EQ_EXPR || comp == GE_EXPR || comp == LE_EXPR) + ? boolean_true_node + : boolean_false_node; + } + + /* Otherwise, compare all the equivalent ranges. First, add N1 and + N2 to their own set of equivalences to avoid duplicating the body + of the loop just to check N1 and N2 ranges. */ + EXECUTE_IF_SET_IN_BITMAP (e1, 0, i1, bi1) + { + value_range_t vr1 = *(vr_value[i1]); + + /* If the range is VARYING or UNDEFINED, use the name itself. */ + if (vr1.type == VR_VARYING || vr1.type == VR_UNDEFINED) + { + vr1.type = VR_RANGE; + vr1.min = ssa_name (i1); + vr1.max = ssa_name (i1); + } + + t = retval = NULL_TREE; + EXECUTE_IF_SET_IN_BITMAP (e2, 0, i2, bi2) + { + value_range_t vr2 = *(vr_value[i2]); + + if (vr2.type == VR_VARYING || vr2.type == VR_UNDEFINED) + { + vr2.type = VR_RANGE; + vr2.min = ssa_name (i2); + vr2.max = ssa_name (i2); + } + + t = compare_ranges (comp, &vr1, &vr2); + if (t) + { + /* All the ranges in the equivalent sets should compare + the same. */ + gcc_assert (retval == NULL || t == retval); + retval = t; + } + } + + if (retval) + { + bitmap_clear_bit (e1, SSA_NAME_VERSION (n1)); + bitmap_clear_bit (e2, SSA_NAME_VERSION (n2)); + return retval; + } + } + + /* None of the equivalent ranges are useful in computing this + comparison. */ + bitmap_clear_bit (e1, SSA_NAME_VERSION (n1)); + bitmap_clear_bit (e2, SSA_NAME_VERSION (n2)); + return NULL_TREE; +} + + +/* Given a conditional predicate COND, try to determine if COND yields + true or false based on the value ranges of its operands. Return + BOOLEAN_TRUE_NODE if the conditional always evaluates to true, + BOOLEAN_FALSE_NODE if the conditional always evaluates to false, and, + NULL if the conditional cannot be evaluated at compile time. + + If USE_EQUIV_P is true, the ranges of all the names equivalent with + the operands in COND are used when trying to compute its value. + This is only used during final substitution. During propagation, + we only check the range of each variable and not its equivalents. */ + +tree +vrp_evaluate_conditional (tree cond, bool use_equiv_p) { gcc_assert (TREE_CODE (cond) == SSA_NAME || TREE_CODE_CLASS (TREE_CODE (cond)) == tcc_comparison); if (TREE_CODE (cond) == SSA_NAME) { - /* For SSA names, only return a truth value if the range is - known and contains exactly one value. */ - value_range *vr = SSA_NAME_VALUE_RANGE (cond); - if (vr && vr->type == VR_RANGE && vr->min == vr->max) + value_range_t *vr; + tree retval; + + if (use_equiv_p) + retval = compare_name_with_value (NE_EXPR, cond, boolean_false_node); + else + { + value_range_t *vr = get_value_range (cond); + retval = compare_range_with_value (NE_EXPR, vr, boolean_false_node); + } + + /* If COND has a known boolean range, return it. */ + if (retval) + return retval; + + /* Otherwise, if COND has a symbolic range of exactly one value, + return it. */ + vr = get_value_range (cond); + if (vr->type == VR_RANGE && vr->min == vr->max) return vr->min; } else { - /* For comparisons, evaluate each operand and compare their - ranges. */ - tree op0, op1; - value_range *vr0, *vr1; - - op0 = TREE_OPERAND (cond, 0); - vr0 = (TREE_CODE (op0) == SSA_NAME) ? get_value_range (op0) : NULL; + tree op0 = TREE_OPERAND (cond, 0); + tree op1 = TREE_OPERAND (cond, 1); - op1 = TREE_OPERAND (cond, 1); - vr1 = (TREE_CODE (op1) == SSA_NAME) ? get_value_range (op1) : NULL; + /* We only deal with integral and pointer types. */ + if (!INTEGRAL_TYPE_P (TREE_TYPE (op0)) + && !POINTER_TYPE_P (TREE_TYPE (op0))) + return NULL_TREE; - if (vr0 && vr1) - return compare_ranges (TREE_CODE (cond), vr0, vr1); - else if (vr0 && vr1 == NULL) - return compare_range_with_value (TREE_CODE (cond), vr0, op1); - else if (vr0 == NULL && vr1) - return compare_range_with_value (opposite_comparison (TREE_CODE (cond)), - vr1, op0); + if (use_equiv_p) + { + if (TREE_CODE (op0) == SSA_NAME && TREE_CODE (op1) == SSA_NAME) + return compare_names (TREE_CODE (cond), op0, op1); + else if (TREE_CODE (op0) == SSA_NAME) + return compare_name_with_value (TREE_CODE (cond), op0, op1); + else if (TREE_CODE (op1) == SSA_NAME) + return compare_name_with_value ( + opposite_comparison (TREE_CODE (cond)), op1, op0); + } + else + { + value_range_t *vr0, *vr1; + + vr0 = (TREE_CODE (op0) == SSA_NAME) ? get_value_range (op0) : NULL; + vr1 = (TREE_CODE (op1) == SSA_NAME) ? get_value_range (op1) : NULL; + + if (vr0 && vr1) + return compare_ranges (TREE_CODE (cond), vr0, vr1); + else if (vr0 && vr1 == NULL) + return compare_range_with_value (TREE_CODE (cond), vr0, op1); + else if (vr0 == NULL && vr1) + return compare_range_with_value ( + opposite_comparison (TREE_CODE (cond)), vr1, op0); + } } /* Anything else cannot be computed statically. */ @@ -2033,15 +3293,55 @@ vrp_visit_cond_stmt (tree stmt, edge *taken_edge_p) fprintf (dump_file, "\t"); print_generic_expr (dump_file, use, 0); fprintf (dump_file, ": "); - dump_value_range (dump_file, SSA_NAME_VALUE_RANGE (use)); + dump_value_range (dump_file, vr_value[SSA_NAME_VERSION (use)]); } fprintf (dump_file, "\n"); } /* Compute the value of the predicate COND by checking the known - ranges of each of its operands. */ - val = vrp_evaluate_conditional (cond); + ranges of each of its operands. + + Note that we cannot evaluate all the equivalent ranges here + because those ranges may not yet be final and with the current + propagation strategy, we cannot determine when the value ranges + of the names in the equivalence set have changed. + + For instance, given the following code fragment + + i_5 = PHI <8, i_13> + ... + i_14 = ASSERT_EXPR + if (i_14 == 1) + ... + + Assume that on the first visit to i_14, i_5 has the temporary + range [8, 8] because the second argument to the PHI function is + not yet executable. We derive the range ~[0, 0] for i_14 and the + equivalence set { i_5 }. So, when we visit 'if (i_14 == 1)' for + the first time, since i_14 is equivalent to the range [8, 8], we + determine that the predicate is always false. + + On the next round of propagation, i_13 is determined to be + VARYING, which causes i_5 to drop down to VARYING. So, another + visit to i_14 is scheduled. In this second visit, we compute the + exact same range and equivalence set for i_14, namely ~[0, 0] and + { i_5 }. But we did not have the previous range for i_5 + registered, so vrp_visit_assignment thinks that the range for + i_14 has not changed. Therefore, the predicate 'if (i_14 == 1)' + is not visited again, which stops propagation from visiting + statements in the THEN clause of that if(). + + To properly fix this we would need to keep the previous range + value for the names in the equivalence set. This way we would've + discovered that from one visit to the other i_5 changed from + range [8, 8] to VR_VARYING. + + However, fixing this apparent limitation may not be worth the + additional checking. Testing on several code bases (GCC, DLV, + MICO, TRAMP3D and SPEC2000) showed that doing this results in + 4 more predicates folded in SPEC. */ + val = vrp_evaluate_conditional (cond, false); if (val) *taken_edge_p = find_taken_edge (bb_for_stmt (stmt), val); @@ -2108,11 +3408,11 @@ vrp_visit_stmt (tree stmt, edge *taken_edge_p, tree *output_p) union of VR0 and VR1. */ static void -vrp_meet (value_range *vr0, value_range *vr1) +vrp_meet (value_range_t *vr0, value_range_t *vr1) { if (vr0->type == VR_UNDEFINED) { - *vr0 = *vr1; + copy_value_range (vr0, vr1); return; } @@ -2130,13 +3430,6 @@ vrp_meet (value_range *vr0, value_range *vr1) if (vr1->type == VR_VARYING) { - *vr0 = *vr1; - return; - } - - /* If either is a symbolic range, drop to VARYING. */ - if (symbolic_range_p (vr0) || symbolic_range_p (vr1)) - { set_value_range_to_varying (vr0); return; } @@ -2147,28 +3440,46 @@ vrp_meet (value_range *vr0, value_range *vr1) union of both ranges. */ if (value_ranges_intersect_p (vr0, vr1)) { + int cmp; tree min, max; - min = vr0->min; - max = vr0->max; - /* The lower limit of the new range is the minimum of the - two ranges. */ - if (compare_values (vr0->min, vr1->min) == 1) + two ranges. If they cannot be compared, the result is + VARYING. */ + cmp = compare_values (vr0->min, vr1->min); + if (cmp == 0 || cmp == 1) min = vr1->min; + else if (cmp == -1) + min = vr0->min; + else + { + set_value_range_to_varying (vr0); + return; + } - /* The upper limit of the new range is the maximum of the - two ranges. */ - if (compare_values (vr0->max, vr1->max) == -1) + /* Similarly, the upper limit of the new range is the + maximum of the two ranges. If they cannot be compared, + the result is VARYING. */ + cmp = compare_values (vr0->max, vr1->max); + if (cmp == 0 || cmp == -1) max = vr1->max; + else if (cmp == 1) + max = vr0->max; + else + { + set_value_range_to_varying (vr0); + return; + } - set_value_range (vr0, vr0->type, min, max); + /* The resulting set of equivalencies is the intersection of + the two sets. */ + if (vr0->equiv && vr1->equiv && vr0->equiv != vr1->equiv) + bitmap_and_into (vr0->equiv, vr1->equiv); + + set_value_range (vr0, vr0->type, min, max, vr0->equiv); } else - { - /* The two ranges don't intersect, set the result to VR_VARYING. */ - set_value_range_to_varying (vr0); - } + goto no_meet; } else if (vr0->type == VR_ANTI_RANGE && vr1->type == VR_ANTI_RANGE) { @@ -2176,28 +3487,49 @@ vrp_meet (value_range *vr0, value_range *vr1) if (compare_values (vr0->min, vr1->min) == 0 && compare_values (vr0->max, vr1->max) == 0 && compare_values (vr0->min, vr0->max) == 0) - /* Nothing to do. */ ; + { + /* The resulting set of equivalencies is the intersection of + the two sets. */ + if (vr0->equiv && vr1->equiv && vr0->equiv != vr1->equiv) + bitmap_and_into (vr0->equiv, vr1->equiv); + } else - set_value_range_to_varying (vr0); + goto no_meet; } else if (vr0->type == VR_ANTI_RANGE || vr1->type == VR_ANTI_RANGE) { - /* A range [VAL1, VAL2] and an anti-range ~[VAL3, VAL4] meet - only if the ranges have an empty intersection. The result of - the meet operation is the anti-range. */ - if (!value_ranges_intersect_p (vr0, vr1)) + /* A numeric range [VAL1, VAL2] and an anti-range ~[VAL3, VAL4] + meet only if the ranges have an empty intersection. The + result of the meet operation is the anti-range. */ + if (!symbolic_range_p (vr0) + && !symbolic_range_p (vr1) + && !value_ranges_intersect_p (vr0, vr1)) { if (vr1->type == VR_ANTI_RANGE) - *vr0 = *vr1; + copy_value_range (vr0, vr1); } else - set_value_range_to_varying (vr0); + goto no_meet; } else gcc_unreachable (); + + return; + +no_meet: + /* The two range VR0 and VR1 do not meet. Before giving up and + setting the result to VARYING, see if we can at least derive a + useful anti-range. */ + if (!symbolic_range_p (vr0) + && !range_includes_zero_p (vr0) + && !symbolic_range_p (vr1) + && !range_includes_zero_p (vr1)) + set_value_range_to_nonnull (vr0, TREE_TYPE (vr0->min)); + else + set_value_range_to_varying (vr0); } - + /* Visit all arguments for PHI node PHI that flow through executable edges. If a valid value range can be derived from all the incoming value ranges, set a new range for the LHS of PHI. */ @@ -2207,8 +3539,10 @@ vrp_visit_phi_node (tree phi) { int i; tree lhs = PHI_RESULT (phi); - value_range *lhs_vr = get_value_range (lhs); - value_range vr_result = *lhs_vr; + value_range_t *lhs_vr = get_value_range (lhs); + value_range_t vr_result = { VR_UNDEFINED, NULL_TREE, NULL_TREE, NULL }; + + copy_value_range (&vr_result, lhs_vr); if (dump_file && (dump_flags & TDF_DETAILS)) { @@ -2231,7 +3565,7 @@ vrp_visit_phi_node (tree phi) if (e->flags & EDGE_EXECUTABLE) { tree arg = PHI_ARG_DEF (phi, i); - value_range vr_arg; + value_range_t vr_arg; if (TREE_CODE (arg) == SSA_NAME) vr_arg = *(get_value_range (arg)); @@ -2240,6 +3574,7 @@ vrp_visit_phi_node (tree phi) vr_arg.type = VR_RANGE; vr_arg.min = arg; vr_arg.max = arg; + vr_arg.equiv = NULL; } if (dump_file && (dump_flags & TDF_DETAILS)) @@ -2259,10 +3594,7 @@ vrp_visit_phi_node (tree phi) } if (vr_result.type == VR_VARYING) - { - set_value_range_to_varying (lhs_vr); - return SSA_PROP_VARYING; - } + goto varying; /* To prevent infinite iterations in the algorithm, derive ranges when the new value is slightly bigger or smaller than the @@ -2291,20 +3623,22 @@ vrp_visit_phi_node (tree phi) VARYING. */ if (vr_result.min == TYPE_MIN_VALUE (TREE_TYPE (vr_result.min)) && vr_result.max == TYPE_MAX_VALUE (TREE_TYPE (vr_result.max))) - { - set_value_range_to_varying (lhs_vr); - return SSA_PROP_VARYING; - } + goto varying; } } /* If the new range is different than the previous value, keep iterating. */ - if (update_value_range (lhs_vr, vr_result.type, vr_result.min, vr_result.max)) + if (update_value_range (lhs, &vr_result)) return SSA_PROP_INTERESTING; /* Nothing changed, don't add outgoing edges. */ return SSA_PROP_NOT_INTERESTING; + + /* No match found. Set the LHS to VARYING. */ +varying: + set_value_range_to_varying (lhs_vr); + return SSA_PROP_VARYING; } @@ -2313,8 +3647,9 @@ vrp_visit_phi_node (tree phi) static void vrp_finalize (void) { - basic_block bb; - int num_pred_folded = 0; + size_t i; + prop_value_t *single_val_range; + bool do_value_subst_p; if (dump_file) { @@ -2323,33 +3658,42 @@ vrp_finalize (void) fprintf (dump_file, "\n"); } - FOR_EACH_BB (bb) + /* We may have ended with ranges that have exactly one value. Those + values can be substituted as any other copy/const propagated + value using substitute_and_fold. */ + single_val_range = xmalloc (num_ssa_names * sizeof (*single_val_range)); + memset (single_val_range, 0, num_ssa_names * sizeof (*single_val_range)); + + do_value_subst_p = false; + for (i = 0; i < num_ssa_names; i++) + if (vr_value[i] + && vr_value[i]->type == VR_RANGE + && vr_value[i]->min == vr_value[i]->max) + { + single_val_range[i].value = vr_value[i]->min; + do_value_subst_p = true; + } + + if (!do_value_subst_p) { - tree last = last_stmt (bb); - if (last && TREE_CODE (last) == COND_EXPR) - { - tree val = vrp_evaluate_conditional (COND_EXPR_COND (last)); - if (val) - { - if (dump_file) - { - fprintf (dump_file, "Folding predicate "); - print_generic_expr (dump_file, COND_EXPR_COND (last), 0); - fprintf (dump_file, " to "); - print_generic_expr (dump_file, val, 0); - fprintf (dump_file, "\n"); - } - - num_pred_folded++; - COND_EXPR_COND (last) = val; - update_stmt (last); - } - } + /* We found no single-valued ranges, don't waste time trying to + do single value substitution in substitute_and_fold. */ + free (single_val_range); + single_val_range = NULL; } - if (dump_file && (dump_flags & TDF_STATS)) - fprintf (dump_file, "\nNumber of predicates folded: %d\n\n", - num_pred_folded); + substitute_and_fold (single_val_range, true); + + /* Free allocated memory. */ + for (i = 0; i < num_ssa_names; i++) + if (vr_value[i]) + { + BITMAP_FREE (vr_value[i]->equiv); + free (vr_value[i]); + } + + free (single_val_range); + free (vr_value); } @@ -2362,6 +3706,34 @@ vrp_finalize (void) This is essentially an SSA-CCP pass modified to deal with ranges instead of constants. + While propagating ranges, we may find that two or more SSA name + have equivalent, though distinct ranges. For instance, + + 1 x_9 = p_3->a; + 2 p_4 = ASSERT_EXPR + 3 if (p_4 == q_2) + 4 p_5 = ASSERT_EXPR ; + 5 endif + 6 if (q_2) + + In the code above, pointer p_5 has range [q_2, q_2], but from the + code we can also determine that p_5 cannot be NULL and, if q_2 had + a non-varying range, p_5's range should also be compatible with it. + + These equivalencies are created by two expressions: ASSERT_EXPR and + copy operations. Since p_5 is an assertion on p_4, and p_4 was the + result of another assertion, then we can use the fact that p_5 and + p_4 are equivalent when evaluating p_5's range. + + Together with value ranges, we also propagate these equivalencies + between names so that we can take advantage of information from + multiple ranges when doing final replacement. Note that this + equivalency relation is transitive but not symmetric. + + In the example above, p_5 is equivalent to p_4, q_2 and p_3, but we + cannot assert that q_2 is equivalent to p_5 because q_2 may be used + in contexts where that assertion does not hold (e.g., in line 6). + TODO, the main difference between this pass and Patterson's is that we do not propagate edge probabilities. We only compute whether edges can be taken or not. That is, instead of having a spectrum @@ -2378,11 +3750,9 @@ execute_vrp (void) if (cfg_loops) scev_initialize (cfg_loops); - if (vrp_initialize ()) - { - ssa_propagate (vrp_visit_stmt, vrp_visit_phi_node); - vrp_finalize (); - } + vrp_initialize (); + ssa_propagate (vrp_visit_stmt, vrp_visit_phi_node); + vrp_finalize (); if (cfg_loops) { diff --git a/gcc/tree.h b/gcc/tree.h index c9fdb93..ced72b1 100644 --- a/gcc/tree.h +++ b/gcc/tree.h @@ -1338,17 +1338,12 @@ struct tree_exp GTY(()) #define SSA_NAME_VALUE(N) \ SSA_NAME_CHECK (N)->ssa_name.value_handle -/* Range information for SSA_NAMEs. */ -#define SSA_NAME_VALUE_RANGE(N) \ - SSA_NAME_CHECK (N)->ssa_name.value_range - /* Auxiliary pass-specific data. */ #define SSA_NAME_AUX(N) \ SSA_NAME_CHECK (N)->ssa_name.aux #ifndef _TREE_FLOW_H struct ptr_info_def; -struct value_range_def; #endif @@ -1386,9 +1381,6 @@ struct tree_ssa_name GTY(()) as well. */ tree value_handle; - /* Value range information. */ - struct value_range_def *value_range; - /* Auxiliary information stored with the ssa name. */ PTR GTY((skip)) aux; @@ -3601,6 +3593,7 @@ extern bool tree_swap_operands_p (tree, tree, bool); extern enum tree_code swap_tree_comparison (enum tree_code); extern bool ptr_difference_const (tree, tree, HOST_WIDE_INT *); +extern enum tree_code invert_tree_comparison (enum tree_code, bool); /* In builtins.c */ extern tree fold_builtin (tree, tree, bool); -- 2.7.4