Jeff, I only let his drop during phase 4 because it contains changes that you _might_ want to have in GCC16. None of these are rising to the level of having triggered trouble in the real world (yet) and we only came across them as part of some fold-mem-offsets work we have in-flight for GCC17.
I'll merge whatever receives an OK and resend the rest for GCC17 together with updates to the FMO pass. Cheers, Philipp. On Tue, 10 Mar 2026 at 15:06, Jeffrey Law <[email protected]> wrote: > > > > On 2/26/2026 11:31 AM, Philipp Tomsich wrote: > > This series contains some minor fixes and improvements for the > > fold-mem-offsets pass that we'd like to still land into GCC16. > > > > > > Philipp Tomsich (6): > > fold-mem-offsets: Fix wrong mode in memory_address_addr_space_p > > fold-mem-offsets: Replace boolean parameters with enum in > > fold_offsets_1 > > fold-mem-offsets: Clarify why do_analysis pre-marks the memory insn > > fold-mem-offsets: Move RISC-V size-optimization workaround to the > > backend > > fold-mem-offsets: Store fold_mem_info by value in the hash_map > > fold-mem-offsets: Make get_fold_mem_root static > Given we're in stage4 (regression fixes) we probably need to be pretty > cautious here. > > The first patch likely fixes a correctness issue and latent > regressions. So I think it's reasonable to include now. > > The rest look like cleanups/improvements to me. So I'd tend to > recommend deferring those to gcc-17. > > Jeff
