https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78327
Bug ID: 78327
Summary: -Walloca-large-than false positives due to bad range
info for signed integers in [-TYPE_MAX + N, N]
Product: gcc
Version: 7.0
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: msebor at gcc dot gnu.org
Target Milestone: ---
The following is a test case for the problem brought up in the the thread Re:
anti-ranges of signed variables
(https://gcc.gnu.org/ml/gcc/2016-11/msg00029.html).
In the test case below, n's range in function g is [-125, 2], making the
largest value that (n + 128) can evaluate to and alloca be called with 130.
Yet -Walloca-larger-than=200 emits a warning for the function with a note
claiming that the computed value may be as large as 255. That's incorrect.
Recompiling the same test case with -Walloca-larger-than=100 results in a
warning for function h as well (as expected), and with a note correctly
indicating the largest value alloca can be called with there given its range of
[-125, 1]: 129. (The warning for g still says 255.)
c.c:14:3: note: limit is 100 bytes, but argument may be as large as 129
The output of -fdump-tree-vrp confirms that the range GCC computes for n in g
is [0, 255], while in h [3, 129]. The same problem seems to affect all signed
integers.
$ cat c.c && /build/gcc-trunk-svn/gcc/xgcc -B /build/gcc-trunk-svn/gcc -O2 -S
-Wall -Wextra -Walloca-larger-than=200 c.c
void f (void*);
void g (signed char n)
{
if (n < -125 || 2 < n) n = 0;
f (__builtin_alloca (n + 128));
}
void h (signed char n)
{
if (n < -125 || 1 < n) n = 0;
f (__builtin_alloca (n + 128));
}
c.c: In function ‘g’:
c.c:7:3: warning: argument to ‘alloca’ may be too large [-Walloca-larger-than=]
f (__builtin_alloca (n + 128));
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
c.c:7:3: note: limit is 200 bytes, but argument may be as large as 255
