This is the mail archive of the
gcc-help@gcc.gnu.org
mailing list for the GCC project.
Re: nop..in gcc-4.4.2. :: reason for requirement
- From: Ian Lance Taylor <iant at google dot com>
- To: isha Kandaswamy <isha dot kandaswamy at gmail dot com>
- To: gcc-help at gcc dot gnu dot org
- Date: Wed, 18 Nov 2009 07:33:16 -0800
- Subject: Re: nop..in gcc-4.4.2. :: reason for requirement
- References: <1794f0100911180240x3029ed04o47337473d9768a51@mail.gmail.com>
isha Kandaswamy <isha.kandaswamy@gmail.com> writes:
> When I was compiling gcc 4.4.2 for a certain machine description, I realized
> that it needed a nop pattern in the md. This was being called by cfglayout.c
> in the function fixup_reorder_chain. This call was not made in the previous
> version of gcc I was using namely gcc-4.3.1. My doubt is whay do we need the
> nop pattern in gcc-4.4.2 and not in gcc-4.3.1. I also have another doubt.
> What does the function copy_to_reg do? Why does it require a movsi pattern
> in the machine descriptions ?
I've Bcc'ed almost everybody. Please don't send messages to a huge
group of people like this. Please don't send messages to both
gcc@gcc.gnu.org and gcc-patches@gcc.gnu.org. Thanks.
New versions of gcc impose new requirements on backends. That's just
the way it works. The gen_nop call in cfglayout.c is to improve
debuggability of code compiled with -O0, by giving debuggers a place
to set a breakpoint on an edge which otherwise contains no
instructions.
The copy_to_reg function copies a value into a register. It requires
a movMM insn where MM is the mode of the value. In general a gcc
backend must have a movMM insn for all supported modes.
Ian