qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: riscv: How to check floating point support is currently enabled?


From: Ian Jiang
Subject: Re: riscv: How to check floating point support is currently enabled?
Date: Wed, 22 Jan 2020 06:22:02 +0800

Alistair Francis <address@hidden> 于2020年1月21日周二 下午6:50写道:
>
> On Tue, Jan 21, 2020 at 11:12 AM Ian Jiang <address@hidden> wrote:
> >
> > The function riscv_cpu_fp_enabled() is used for checking whether floating 
> > point support is currently enabled. In fact it checks the FS field in the 
> > mstatus MSR.
> >
> > target/riscv/cpu_helper.c
> >  76 bool riscv_cpu_fp_enabled(CPURISCVState *env)
> >  77 {
> >  78     if (env->mstatus & MSTATUS_FS) {
> >  79         return true;
> >  80     }
> >  81
> >  82     return false;
> >  83 }
> >
> > This will cause a problem that the SD bit in mstatus is not set to 1 when  
> > FS in mstatus is modified from '00'b to '11'b with write_mstatus().
>
> Thanks for looking into this.
>
> There are patches on list fixing floating point errors. Can you check
> if this branch fixes any issues you have:
> https://github.com/palmer-dabbelt/qemu/commits/for-master

Yes, there is:
https://github.com/palmer-dabbelt/qemu/commit/82f014671cf057de51c4a577c9e2ad637dcec6f9
Thanks!

>
> >
> > file target/riscv/csr.c, func write_mstatus():
> > 350     dirty = (riscv_cpu_fp_enabled(env) &&
> > 351              ((mstatus & MSTATUS_FS) == MSTATUS_FS)) |
> > 352             ((mstatus & MSTATUS_XS) == MSTATUS_XS);
> > 353     mstatus = set_field(mstatus, MSTATUS_SD, dirty);
> > 354     env->mstatus = mstatus;
> >
> > So checking fields D and F in the misa MSR (bit 3 and bit 5) may be an 
> > better way. That is
> > bool riscv_cpu_fp_enabled(CPURISCVState *env)
> >     if (env->misa & (MISA_F | MISA_F) {
> >         return true;
> >     }
> >     return false;
> > }
>
> This doesn't seem right, just because the HW supports it doesn't mean
> it's enabled.

OK.

>
> Alistair
>
> >
> >
> > --
> > Ian Jiang



reply via email to

[Prev in Thread] Current Thread [Next in Thread]