https://gcc.gnu.org/bugzilla/show_bug.cgi?id=117081
--- Comment #20 from Hongtao Liu <liuhongt at gcc dot gnu.org> --- > > W/o more usage of callee-saved registers, callee needs to restore them > before exit which is not needed if more caller-saved register are used. W/ https://gcc.gnu.org/pipermail/gcc-patches/2025-February/675714.html the codegen is like foo: .LFB0: .cfi_startproc pushq %r12 .cfi_def_cfa_offset 16 .cfi_offset 12, -16 movq %rsi, %r12 pushq %rbp .cfi_def_cfa_offset 24 .cfi_offset 6, -24 movq %rdi, %rbp pushq %rbx .cfi_def_cfa_offset 32 .cfi_offset 3, -32 movq %rdx, %rbx subq $32, %rsp .cfi_def_cfa_offset 64 leaq 16(%rsp), %rcx leaq 24(%rsp), %r8 call bar movl %eax, %edx testl %eax, %eax je .L1 vmovsd 16(%rsp), %xmm0 vxorpd %xmm1, %xmm1, %xmm1 vcomisd %xmm1, %xmm0 jbe .L18 vmovsd .LC1(%rip), %xmm1 vcomisd %xmm0, %xmm1 ja .L21 .L18: xorl %edx, %edx .L3: vmovsd 24(%rsp), %xmm0 vxorpd %xmm1, %xmm1, %xmm1 vcomisd %xmm1, %xmm0 jbe .L1 vmovsd .LC1(%rip), %xmm1 vcomisd %xmm0, %xmm1 ja .L22 .L1: addq $32, %rsp .cfi_remember_state .cfi_def_cfa_offset 32 movl %edx, %eax popq %rbx .cfi_def_cfa_offset 24 popq %rbp .cfi_def_cfa_offset 16 popq %r12 .cfi_def_cfa_offset 8 ret .p2align 4,,10 .p2align 3 There're 3 pops, more than best one(2 pops), less than worst one(4 pops), but it looks more reasonable since we do want to keep rsi in callee-saved register to preserve it across call.
