This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [committed] PR target/25871: Fix MIPS trampolines for Pmode == DImode


Richard Sandiford wrote:
> I've applied Joshua Kinard's patch for PR 25871, in which the MIPS
> nested function trampolines had 32-bit pointer moves even when
> Pmode == DImode.
> 
> The bug isn't a regression, so I've not applied it to release branches.
> However, the patch seems very safe to me, and should only affect people
> for whom the current code doesn't work, so I'd be happy for it to be
> applied to release branches too.
[snip]
> -#define TRAMPOLINE_TEMPLATE(STREAM)					 \
> -{									 \
> -  fprintf (STREAM, "\t.word\t0x03e00821\t\t# move   $1,$31\n");		\
> +#define TRAMPOLINE_TEMPLATE(STREAM)					\
> +{									\
> +  if (ptr_mode == DImode)						\
> +    fprintf (STREAM, "\t.word\t0x03e0082d\t\t# dmove   $1,$31\n");	\
> +  else									\
> +    fprintf (STREAM, "\t.word\t0x03e00821\t\t# move   $1,$31\n");	\

It probably makes sense to use explicitly "addu" and "daddu" in the
comments, this avoids confusion caused by a non-existing "dmove"
assembler macro.


Thiemo


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]