On Thu, Jun 12, 2025 at 10:58 AM Uros Bizjak <ubiz...@gmail.com> wrote: > > On Thu, Jun 12, 2025 at 9:26 AM Cui, Lili <lili....@intel.com> wrote: > > > > > > @@ -7753,8 +7762,12 @@ pro_epilogue_adjust_stack (rtx dest, rtx src, > > > > rtx > > > offset, > > > > add_frame_related_expr = true; > > > > } > > > > > > > > + if (crtl->shrink_wrapped_separate) insn = emit_insn (gen_rtx_SET > > > > + (dest, gen_rtx_PLUS (Pmode, src, addend))); > > > > > > Please use ix86_expand_binary_operator here, it will correctly > > > generate LEA during pro/epilogue generation. Also, please check > > indenting. > > > > > Hi Uros, > > > > I'm confused here, ix86_expand_binary_operator cannot return insn, and > > insn needs to be used later. Another thing is in > > ix86_expand_binary_operator, if !rtx_equal_p (dst, src1), it will generate > > clobber for it. > > Oh, I missed this condition. The original is OK, but please add a > comment why clobber-less LEA is required here.
BTW: Do you need to clobber memory to prevent EBP based memory access to get scheduled around the insn? Uros.
