Re: port-mips/57680: printf("%.1f") shows wrong results on R3000 mipseb

[Taylor R Campbell <riastradh%NetBSD.org@localhost>]

The following reply was made to PR port-mips/57680; it has been noted by GNATS.

From: Taylor R Campbell <riastradh%NetBSD.org@localhost>
To: RVP <rvp%SDF.ORG@localhost>
Cc: gnats-bugs%netbsd.org@localhost, port-mips-maintainer%netbsd.org@localhost,
	gnats-admin%netbsd.org@localhost, netbsd-bugs%netbsd.org@localhost, tsutsui%ceres.dti.ne.jp@localhost
Subject: Re: port-mips/57680: printf("%.1f") shows wrong results on R3000
	mipseb
Date: Tue, 7 Nov 2023 02:13:05 +0000

 > Date: Tue, 7 Nov 2023 01:29:41 +0000 (UTC)
 > From: RVP <rvp%SDF.ORG@localhost>
 >  
 > __flt_rounds() in src/lib/libc/arch/mips/gen/flt_rounds.c does:
 >  
 >          __asm(".set push; .set noat; cfc1 %0,$31; .set pop" : "=r" (x));
 >  
 > while fegetround() -> __rfs() in src/sys/arch/mips/include/fenv.h only
 > does:
 >  
 >          __asm __volatile("cfc1 %0,$31" : "=r" (__fpsr));
 > [...]
 > So, maybe a ".set push; ...; .set pop" is needed for older MIPS? Can you
 > change __rfs() like that?
 
 It's possible that change is correct, but I doubt it's the culprit.
 
 In the objdump output that tsutsui@ quoted from __dtoa in the broken
 libc, which I confirmed in my own build in the .o and .pico code
 (although I haven't tested my build -- no hardware), there is no use
 of the at register ($1, no overlap with any vN) and there is a nop in
 the cfc1 load delay slot:
 
       50:	4442f800 	cfc1	v0,$31
       54:	00000000 	nop
       58:	30420003 	andi	v0,v0,0x3
 
 So while maybe we ought to set noat and noreorder, and maybe we ought
 to add a nop in the asm, that doesn't seem to be the source of the
 problem.