]> git.ipfire.org Git - thirdparty/gcc.git/commit
middle-end/13421 - -ftrapv vs. POINTER_DIFF_EXPR
authorRichard Biener <rguenther@suse.de>
Tue, 16 Apr 2024 12:05:35 +0000 (14:05 +0200)
committerRichard Biener <rguenther@suse.de>
Tue, 30 Apr 2024 11:03:55 +0000 (13:03 +0200)
commit667c19de86b33648f5f4599f589a5e02adbb35cb
tree8c72a06586c5519ce16016130a25793b0a77cf9f
parent6c6b70f07208ca14ba783933988c04c6fc2fff42
middle-end/13421 - -ftrapv vs. POINTER_DIFF_EXPR

Currently we expand POINTER_DIFF_EXPR using subv_optab when -ftrapv
(but -fsanitize=undefined does nothing).  That's not consistent
with the behavior of POINTER_PLUS_EXPR which never uses addv_optab
with -ftrapv.  Both are because of the way we select whether to use
the trapping or the non-trapping optab - we look at the result type
of the expression and check

  trapv = INTEGRAL_TYPE_P (type) && TYPE_OVERFLOW_TRAPS (type);

the bugreport correctly complains that -ftrapv affects pointer
subtraction (there's no -ftrapv-pointer).  Now that we have
POINTER_DIFF_EXPR we can honor that appropriately.

The patch moves both POINTER_DIFF_EXPR and POINTER_PLUS_EXPR
handling so they will never consider trapping (or saturating)
optabs.

PR middle-end/13421
* optabs-tree.cc (optab_for_tree_code): Do not consider
{add,sub}v or {us,ss}{add,sub} optabs for POINTER_DIFF_EXPR
or POINTER_PLUS_EXPR.
gcc/optabs-tree.cc