https://gcc.gnu.org/bugzilla/show_bug.cgi?id=127988
--- Comment #3 from GCC Commits <cvs-commit at gcc dot gnu.org> --- The trunk branch has been updated by Richard Earnshaw <[email protected]>: https://gcc.gnu.org/g:efcc06ac1adc833d0b104df038d128031ca85e89 commit r17-5635-gefcc06ac1adc833d0b104df038d128031ca85e89 Author: Richard Earnshaw <[email protected]> Date: Tue Oct 20 18:42:08 2026 +0100 arm: tweak limits on minipool placement [PR127988] My previous patch changed a number of the calculations on pool placement, but this wasn't fully accounted for when inserting a forced barrier. The result was that it was possible to find a barrier in the code that had previously been discarded as being too far away. There was also a small memory leak in the pool dumping code in that we weren't destructing the subpools. gcc/ChangeLog: PR target/127988 * config/arm/minipool.cc (minipool::add_forward_ref): Allow for pool alignment when checking if the current fix will fit. (minipool::create_fix_barrier): Ensure we always have a suitable insertion point and fix off-by-one issue when testing if a insn is in range. (place_minipools): Correctly account for padding and the pool skip instruction before inserting a forced barrier. (minipool::dump): Call the destructor on subpools before freeing them.
