This patchset improves folding in cases where input lengths and/or destination sizes may not be constant but are range bound.
Tested on x86_64 with a full bootstrap build and verified that there are no regressions resulting from this patchset. I double-checked that the run was current and I wasn't checking logs of a stale build. I tested builds of bash and wpa_supplicant with this patchset. wpa_supplicant ends up with 30 fewer __memcpy_chk calls, of which 14 are completely optimized away and the rest transformed to __memcpy. In bash, 3 __memcpy_chk calls are optimized away completely in addition to a couple of memmove and strcpy chk variants being transformed into regular calls. Changes from v1: - Use dump_* functions instead of directly using dump_file - Bring back warnings and reduce scope of changes to strncat to only using ranges to determine if call can be simplified to strcat. - Renamed known_safe to known_lower Siddhesh Poyarekar (3): gimple-fold: Transform stp*cpy_chk to str*cpy directly gimple-fold: Use ranges to simplify _chk calls gimple-fold: Use ranges to simplify strncat and snprintf gcc/gimple-fold.c | 343 ++++++++++-------------- gcc/testsuite/gcc.dg/Wobjsize-1.c | 5 +- gcc/testsuite/gcc.dg/fold-stringops-1.c | 23 ++ gcc/testsuite/gcc.dg/fold-stringops-2.c | 63 +++++ gcc/testsuite/gcc.dg/fold-stringops-3.c | 18 ++ 5 files changed, 256 insertions(+), 196 deletions(-) create mode 100644 gcc/testsuite/gcc.dg/fold-stringops-1.c create mode 100644 gcc/testsuite/gcc.dg/fold-stringops-2.c create mode 100644 gcc/testsuite/gcc.dg/fold-stringops-3.c -- 2.31.1
