Per Richi's suggestions I tweaked the recent tree-ssa-dom.c equivalence code to use int_fits_type_p rather than creating a new integer node and verifying it had the same value as the old node.
At the time I dropped the TYPE_PRECISION check as I thought it was redundant with the int_fits_type_p tests. However int_fits_type_p only checks for a widening precision if it can't prove the constant fits within the bounds of the new type.
Thus we could record an equivalence after a narrowing conversion, which is wrong in this particular case.
Fixed by adding the TYPE_PRECISION checks back. Bootstrapped & regression tested on x86_64-unknown-linux-gnu. Installed.
diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 1e25a82..a3b60c9 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,10 @@ +2013-04-03 Jeff Law <l...@redhat.com> + + PR tree-optimization/56799 + * tree-ssa-dom.c (record_equivalences_from_incoming_edge): Bring + back test for widening conversion erroneously dropped in prior + change. + 2013-04-03 Kyrylo Tkachov <kyrylo.tkac...@arm.com> PR target/56809 @@ -896,6 +903,7 @@ * config/tilegx/tilepro.h (PROFILE_BEFORE_PROLOGUE): Define. 2013-03-25 Jeff Law <l...@redhat.com> + * tree-ssa-dom.c (record_equivalences_from_incoming_edge): Add missing check for INTEGRAL_TYPE_P that was missing due to checking in wrong version of prior patch. diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index 2cc5fa8..dc0b745 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,8 @@ +2013-04-03 Jeff Law <l...@redhat.com> + + PR tree-optimization/56799 + * gcc.c-torture/execute/pr56799.c: New test. + 2013-04-03 Paolo Carlini <paolo.carl...@oracle.com> PR c++/56815 diff --git a/gcc/testsuite/gcc.c-torture/execute/pr56799.c b/gcc/testsuite/gcc.c-torture/execute/pr56799.c new file mode 100644 index 0000000..d9ee26b --- /dev/null +++ b/gcc/testsuite/gcc.c-torture/execute/pr56799.c @@ -0,0 +1,43 @@ + +#include <stdio.h> +typedef struct { int x; int y;} S; +extern int foo(S*); +int hi = 0, lo = 0; + +int main() +{ + S a; + int r; + a.x = (int) 0x00010000; + a.y = 1; + r = foo (&a); + if (r == 2 && lo==0 && hi==1) + { + exit (0); + } + abort (); +} + +typedef unsigned short u16; + +__attribute__ ((noinline)) int foo (S* ptr) +{ + int a = ptr->x; + int c = 0; + u16 b = (u16) a; + if (b != 0) + { + lo = 1; + c += ptr->y; + } + b = a >> 16; + if (b != 0) + { + hi = 1; + c+= ptr->y; + } + c += ptr->y; + return c; +} + + diff --git a/gcc/tree-ssa-dom.c b/gcc/tree-ssa-dom.c index 29d2bb4..d98a646 100644 --- a/gcc/tree-ssa-dom.c +++ b/gcc/tree-ssa-dom.c @@ -1151,9 +1151,15 @@ record_equivalences_from_incoming_edge (basic_block bb) { tree old_rhs = gimple_assign_rhs1 (defstmt); - /* If the constant is in the range of the type of OLD_RHS, - then convert the constant and record the equivalence. */ + /* If the conversion widens the original value and + the constant is in the range of the type of OLD_RHS, + then convert the constant and record the equivalence. + + Note that int_fits_type_p does not check the precision + if the upper and lower bounds are OK. */ if (INTEGRAL_TYPE_P (TREE_TYPE (old_rhs)) + && (TYPE_PRECISION (TREE_TYPE (lhs)) + > TYPE_PRECISION (TREE_TYPE (old_rhs))) && int_fits_type_p (rhs, TREE_TYPE (old_rhs))) { tree newval = fold_convert (TREE_TYPE (old_rhs), rhs);