Re: [Qemu-devel] [PATCH 11/15] tcg: split cpu_set_mxcsr()/cpu_set_fpuc()

[Paolo Bonzini <pbonzini@redhat.com>]

----- Original Message -----
> From: "Zhong Yang" <yang.zhong@intel.com>
> To: "Paolo Bonzini" <pbonzini@redhat.com>
> Cc: qemu-devel@nongnu.org, "anthony xu" <anthony.xu@intel.com>, "a rigo" <a.rigo@virtualopensystems.com>, "Richard
> Henderson" <rth@twiddle.net>, "Thomas Huth" <thuth@redhat.com>
> Sent: Thursday, June 22, 2017 10:03:25 AM
> Subject: Re: [PATCH 11/15] tcg: split cpu_set_mxcsr()/cpu_set_fpuc()
> 
> On Wed, Jun 21, 2017 at 03:15:25PM +0200, Paolo Bonzini wrote:
> > 
> > 
> > On 21/06/2017 12:19, Yang Zhong wrote:
> > > Split the cpu_set_mxcsr()/cpu_set_fpuc() with specific tcg code.
> > > tcg_update_mxcsr()/tcg_set_fpuc() need be implemented in tcg-stub.c
> > > file if tcg is disabled.
> > > 
> > > Signed-off-by: Yang Zhong <yang.zhong@intel.com>
> > > ---
> > >  accel/stubs/tcg-stub.c   |  8 ++++++++
> > >  target/i386/cpu.h        | 15 +++++++++++++--
> > >  target/i386/fpu_helper.c |  8 +++-----
> > >  3 files changed, 24 insertions(+), 7 deletions(-)
> > > 
> > > diff --git a/accel/stubs/tcg-stub.c b/accel/stubs/tcg-stub.c
> > > index dafb1d0..91625a8 100644
> > > --- a/accel/stubs/tcg-stub.c
> > > +++ b/accel/stubs/tcg-stub.c
> > > @@ -75,6 +75,14 @@ void dump_opcount_info(FILE *f, fprintf_function
> > > cpu_fprintf)
> > >  {
> > >  }
> > >  
> > > +void tcg_update_mxcsr(CPUX86State *env)
> > > +{
> > > +}
> > > +
> > > +void tcg_set_fpuc(CPUX86State *env)
> > > +{
> > > +}
> > > +
> > >  void cpu_loop_exit(CPUState *cpu)
> > >  {
> > >      abort();
> > > diff --git a/target/i386/cpu.h b/target/i386/cpu.h
> > > index 8b3b535..229b216 100644
> > > --- a/target/i386/cpu.h
> > > +++ b/target/i386/cpu.h
> > > @@ -1643,8 +1643,19 @@ static inline int32_t x86_get_a20_mask(CPUX86State
> > > *env)
> > >  }
> > >  
> > >  /* fpu_helper.c */
> > > -void cpu_set_mxcsr(CPUX86State *env, uint32_t val);
> > > -void cpu_set_fpuc(CPUX86State *env, uint16_t val);
> > > +void tcg_update_mxcsr(CPUX86State *env);
> > > +void tcg_set_fpuc(CPUX86State *env);
> > > +static inline void cpu_set_mxcsr(CPUX86State *env, uint32_t mxcsr)
> > > +{
> > > +    env->mxcsr = mxcsr;
> > > +    tcg_update_mxcsr(env);
> > 
> > Instead of having to add stubs, please guard the call with "if
> > (tcg_enabled())".  Same below.
> > 
> > Paolo
> >
>   Hello Paolo,
> 
>   Got it! thanks!
> 
>   From your comments + Richard Henderson's comments, i got below guideline
>   (1) accel/stubs/tcg-stub.c
>       This file is target-independent file, do not add x86-specific stubs
>       functions into this file.
> 
>   (2) there are three kind of methods to mask TCG relative function
>       a) if(tcg_enabled())
>       b) if CONFIG_TCG
>       c) stub function
> 
>   Paolo, which one is your prefer?  a) ?  please also make sure your prefer
>   sequence, a) > b) > c) ?

There is also

   d) method in AccelClass

It seldom makes sense, but when it does it's the best.  Apart from this one,
a) > c) > b).

For target-independent code, "c" is more often the solution.  For target-dependent
code, "a" is more common.

Paolo