| Message ID | b36e94c5-637b-0cfa-9e7e-76192f175bda@linaro.org |
|---|---|
| State | New |
| Headers | show |
On 07/26/2016 06:10 PM, kugan wrote: > Hi Jeff, > > Thanks for your comments. > > .... > >>> * tree-ssa-reassoc.c (maybe_optimize_range_tests): Check type >>> compatibility. >> I'd kind of like to see some analysis of how we've got a bool here -- >> ISTM it ought to have been converted it to the type of the LHS of the >> PHI when propagated. > > You are right. The problem was with the order of checking tcc_compare > and calling get_ops. We ended up calling get_ops where we should not. > > Bootstrap and regression testing is ongoing. Is this OK for trunk if no > regressions? > > Thanks, > Kugan > > > gcc/testsuite/ChangeLog: > > 2016-07-27 Kugan Vivekanandarajah <kuganv@linaro.org> > > * gcc.dg/torture/pr71994.c: New test. > > gcc/ChangeLog: > > 2016-07-27 Kugan Vivekanandarajah <kuganv@linaro.org> > > * tree-ssa-reassoc.c (maybe_optimize_range_tests): Check tcc_comparison > before calling get_ops. OK if no regressions. jeff
diff --git a/gcc/testsuite/gcc.dg/torture/pr71994.c b/gcc/testsuite/gcc.dg/torture/pr71994.c index e69de29..8f5e92c 100644 --- a/gcc/testsuite/gcc.dg/torture/pr71994.c +++ b/gcc/testsuite/gcc.dg/torture/pr71994.c @@ -0,0 +1,14 @@ +/* PR tree-optimization/71994 */ +/* { dg-do compile } */ +int om, h6; + +void eo (void) +{ + const int tl = 1; + int ln; + + h6 = (om + tl) > 0; + ln = om && (om & h6); + h6 = om; + om = ln < h6; +} diff --git a/gcc/tree-ssa-reassoc.c b/gcc/tree-ssa-reassoc.c index 18cf978..8f2256f 100644 --- a/gcc/tree-ssa-reassoc.c +++ b/gcc/tree-ssa-reassoc.c @@ -3520,10 +3520,10 @@ maybe_optimize_range_tests (gimple *stmt) push into ops the individual range test arguments of the bitwise or resp. and, recursively. */ if (TREE_CODE (rhs) == SSA_NAME - && !get_ops (rhs, code, &ops, - loop_containing_stmt (stmt)) && (TREE_CODE_CLASS (gimple_assign_rhs_code (stmt)) != tcc_comparison) + && !get_ops (rhs, code, &ops, + loop_containing_stmt (stmt)) && has_single_use (rhs)) { /* Otherwise, push the _234 range test itself. */