On 3/24/19 11:11 AM, Segher Boessenkool wrote: > Hi Jeff, > > On Sun, Mar 24, 2019 at 09:20:07AM -0600, Jeff Law wrote: >> + PR rtl-optimization/87761 >> + * regcprop.c (copyprop_hardreg_forward_1): Check may_trap_p on SET, >> + not INSN. Also check RTX_FRAME_RELATED_P. Queue insns for DF rescan >> + as needed. > > Why the RTX_FRAME_RELATED_P addition? You didn't explain it I think. Is it > just a bugfix, or something this patch exposed, something that couldn't > happen before (or was harmless) and now isn't anymore? It's just a bugfix -- one of the embedded targets needed it, I can't offhand remember which -- we had a frame related insn with a REG_UNUSED note which the code deleted causing a fault in the dwarf2cfi bits.
Thank goodness for Jenkins scripts which will build and test all those silly *-elf targets :-) > > Or should this part be backported :-) Shouldn't be needed as the code to remove insns with REG_UNUSED notes is new for gcc-9. jeff