diff options
Diffstat (limited to 'gcc')
| -rw-r--r-- | gcc/value-query.cc | 20 |
1 files changed, 13 insertions, 7 deletions
diff --git a/gcc/value-query.cc b/gcc/value-query.cc index 9047e27..93609f3 100644 --- a/gcc/value-query.cc +++ b/gcc/value-query.cc @@ -397,14 +397,20 @@ get_range_global (irange &r, tree name) r.set_varying (type); } -// ?? Like above, but only for default definitions of NAME. This is -// so VRP passes using ranger do not start with known ranges, -// otherwise we'd eliminate builtin_unreachables too early because of -// inlining. +// This is where the ranger picks up global info to seed initial +// requests. It is a slightly restricted version of +// get_range_global() above. +// +// The reason for the difference is that we can always pick the +// default definition of an SSA with no adverse effects, but for other +// SSAs, if we pick things up to early, we may prematurely eliminate +// builtin_unreachables. // // Without this restriction, the test in g++.dg/tree-ssa/pr61034.C has -// all of its unreachable calls removed too early. We should -// investigate whether we should just adjust the test above. +// all of its unreachable calls removed too early. +// +// See discussion here: +// https://gcc.gnu.org/pipermail/gcc-patches/2021-June/571709.html value_range gimple_range_global (tree name) @@ -412,7 +418,7 @@ gimple_range_global (tree name) gcc_checking_assert (gimple_range_ssa_p (name)); tree type = TREE_TYPE (name); - if (SSA_NAME_IS_DEFAULT_DEF (name)) + if (SSA_NAME_IS_DEFAULT_DEF (name) || (cfun && cfun->after_inlining)) { value_range vr; get_range_global (vr, name); |