It appears to be a recent regression. Compiler Explorer: https://godbolt.org/z/vaY9GEnrz [549] % gcctk -v Using built-in specs. COLLECT_GCC=gcctk COLLECT_LTO_WRAPPER=/local/suz-local/software/local/gcc-trunk/libexec/gcc/x86_64-pc-linux-gnu/14.0.0/lto-wrapper Target: x86_64-pc-linux-gnu Configured with: ../gcc-trunk/configure --disable-bootstrap --enable-checking=yes --prefix=/local/suz-local/software/local/gcc-trunk --enable-sanitizers --enable-languages=c,c++ --disable-werror --enable-multilib Thread model: posix Supported LTO compression algorithms: zlib gcc version 14.0.0 20231028 (experimental) (GCC) [550] % [550] % gcctk -O2 small.c; ./a.out [551] % [551] % gcctk -O3 small.c small.c: In function ‘main’: small.c:3:29: warning: iteration 1 invokes undefined behavior [-Waggressive-loop-optimizations] 3 | for (int b = 0; b < 33; b += 3) { | ~~^~~~ small.c:3:21: note: within this loop 3 | for (int b = 0; b < 33; b += 3) { | ~~^~~~ [552] % ./a.out Aborted [553] % [553] % cat small.c volatile int a; int main() { for (int b = 0; b < 33; b += 3) { if (b > 31) a++; } if (a != 0) __builtin_abort(); return 0; }
Confirmed. -fno-split-loops fixes it ...
According to bisection, f5fb9ff2396fd41fdd2e6d35a412e936d2d42f75 is the first bad commit.
(In reply to Thomas Koenig from comment #2) > According to bisection, f5fb9ff2396fd41fdd2e6d35a412e936d2d42f75 > is the first bad commit. Or, if anybody wants a link, https://gcc.gnu.org/git/?p=gcc.git;a=commit;h=f5fb9ff2396fd41fdd2e6d35a412e936d2d42f75 .
Looks almost the same code as what was in PR 111652 (maybe it is a dup after all).
patch_loop_exit simply copies the original guard code rather than using the analyzed one. Testing a patch.
The master branch has been updated by Richard Biener <rguenth@gcc.gnu.org>: https://gcc.gnu.org/g:2066c29bf8dea87c9810eb261e342d941a6b2059 commit r14-5446-g2066c29bf8dea87c9810eb261e342d941a6b2059 Author: Richard Biener <rguenther@suse.de> Date: Tue Nov 14 14:26:17 2023 +0100 tree-optimization/111233 - loop splitting miscompile The change in r14-2852-gf5fb9ff2396fd4 failed to update patch_loop_exit to compensate for rewriting of a NE/EQ_EXPR to a new code. Fixed with the following. PR tree-optimization/111233 PR tree-optimization/111652 PR tree-optimization/111727 PR tree-optimization/111838 PR tree-optimization/112113 * tree-ssa-loop-split.cc (patch_loop_exit): Get the new guard code instead of the old guard stmt. (split_loop): Adjust. * gcc.dg/torture/pr111233.c: New testcase. * gcc.dg/torture/pr111652.c: Likewise. * gcc.dg/torture/pr111727.c: Likewise. * gcc.dg/torture/pr111838.c: Likewise. * gcc.dg/torture/pr112113.c: Likewise.
Fixed.