On Tue, Jul 21, 2015 at 2:45 PM, H.J. Lu <hongjiu...@intel.com> wrote: > When __builtin_frame_address is used to retrieve the address of the > function stack frame, the frame pointer is always kept, which wastes one > register and 2 instructions. For x86-32, one less register means > significant negative impact on performance. This patch adds a new > builtin function, __builtin_ia32_stack_top, to x86 backend. It > returns the stack address when the function is called. > > Any comments, feedbacks? > > Thanks. > > > H.J. > --- > gcc/ > > PR target/66960 > * config/i386/i386.c (ix86_expand_prologue): Sorry if DRAP is > used and the stack address has been taken. > (ix86_builtins): Add IX86_BUILTIN_STACK_TOP. > (ix86_init_mmx_sse_builtins): Add __builtin_ia32_stack_top. > (ix86_expand_builtin): Handle IX86_BUILTIN_STACK_TOP. > * config/i386/i386.h (machine_function): Add stack_top_taken. > * doc/extend.texi: Document __builtin_ia32_stack_top. >
I got a feedback, suggesting __builtin_stack_top, instead of __builtin_ia32_stack_top. But I don't know if + /* After the prologue, stack top is at -WORD(AP) in the current + frame. */ + emit_insn (gen_rtx_SET (target, + plus_constant (Pmode, arg_pointer_rtx, + -UNITS_PER_WORD))); is true for all backends. If it works on all backends, I can move it to builtins.c. -- H.J.
