Re: [PATCH 2/2] MIPS: Hard-float rounding instructions support

[Xi Ruoyao <xry111@xry111.site>]

On Thu, 2024-01-25 at 21:58 +0800, Junxian Zhu wrote:

> Recently, I'm still trying to do some optimization for MIPS. I want to
> know if I use rounding instructions on MIPS, like |round.l.d|, in 
> functions with parameters input in FPR and results in GPR, like 
> > llrint|or |lrint|, will we still face the INEXACT exception and 
> potentially violate C23?

C23 explicitly says llrint and lrint *raises* FE_INEXACT exception when
they raise no other floating-point exception and the result differs from
the argument.

But you should still do the optimization in GCC, then define
USE_LRINT_BUILTIN etc. in Glibc if GCC is recent enough to support it. 
By doing this in the compiler, users can avoid an overhead to do a
function call (esp. on MIPS with all the "abicall" stuff the overhead is
very large).

-- 
Xi Ruoyao <xry111@xry111.site>
School of Aerospace Science and Technology, Xidian University