This is the mail archive of the
binutils@sourceware.org
mailing list for the binutils project.
Re: [PATCH v2] Work around the NOP issue of Loongson2F
- From: Wu Zhangjin <wuzhangjin at gmail dot com>
- To: Richard Sandiford <rdsandiford at googlemail dot com>
- Cc: Nick Clifton <nickc at redhat dot com>, binutils at sourceware dot org, "Alfred M. Szmidt" <ams at gnu dot org>, zhangfx at lemote dot com, yanh at lemote dot com
- Date: Tue, 17 Nov 2009 18:08:00 +0800
- Subject: Re: [PATCH v2] Work around the NOP issue of Loongson2F
- References: <1258381496-32456-1-git-send-email-wuzhangjin@gmail.com> <87iqdautjc.fsf@firetop.home>
- Reply-to: wuzhangjin at gmail dot com
Hi,
On Mon, 2009-11-16 at 20:03 +0000, Richard Sandiford wrote:
> Why did you remove the md_begin stuff that handled the initialisation
> of nop_insn?
No, I didn't remove the md_begin() stuff, it's still there to init the
INSN_NOP.
> What I was trying to say in my review was that:
>
> (a) Your code to handle _implicit_ nops (in md_begin) looked good.
> I agree with that approach.
>
> (b) The problem with the previous patch was that it didn't handle
> _explicit_ nops.
>
> The new patch handles explicit nops, but doesn't handle implicit nops.
> We need to handle both. Of course, the code in md_begin will need
> to skip the "nop" INSN_MACRO entry, but that should be easy.
>
Only one issue left, that is the nops instroduced by .align 5 or .fill
can not be replaced by our new nops. I have traced the source code, the
INSN_NOP is exactly replaced by "move at,at", but the stuff of .align
and .fill is a little complex, Seems they do not use the INSN_NOP, but
something else. I have tried to analyze the source code about
mips_align(), mips_handle_align() and mips_nop_opcode(), but I can not
easily get the result. any prompt here?
Thanks & Regards,
Wu Zhangjin