qemu-devel
[Top][All Lists]
Advanced

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

Re: [Qemu-devel] [PATCH v3 2/5] target/arm: optimize rev16() using extra


From: Aurelien Jarno
Subject: Re: [Qemu-devel] [PATCH v3 2/5] target/arm: optimize rev16() using extract op
Date: Fri, 12 May 2017 21:22:52 +0200
User-agent: NeoMutt/20170113 (1.7.2)

On 2017-05-12 12:05, Richard Henderson wrote:
> On 05/12/2017 11:21 AM, Aurelien Jarno wrote:
> > +    uint64_t mask1 = sf ? 0x00ff00ff00ff00ffull : 0x00ff00ff;
> > +    uint64_t mask2 = sf ? 0xff00ff00ff00ff00ull : 0xff00ff00;
> > +
> > +    tcg_gen_shri_i64(tcg_tmp, tcg_rn, 8);
> > +    tcg_gen_andi_i64(tcg_tmp, tcg_tmp, mask1);
> > +    tcg_gen_shli_i64(tcg_rd, tcg_rn, 8);
> > +    tcg_gen_andi_i64(tcg_rd, tcg_rd, mask2);
> 
> It would probably be better to use a single mask, since they're not free to
> instantiate in a register.  So e.g.
> 
>   TCGv mask = tcg_const_i64(sf ? 0x00ff00ff00ff00ffull : 0x00ff00ff);
>   tcg_gen_shri_i64(tcg_tmp, tcg_rn, 8);
>   tcg_gen_and_i64(tcg_rd, tcg_rn, mask);
>   tcg_gen_and_i64(tcg_tmp, tcg_tmp, mask);
>   tcg_gen_shli_i64(tcg_rd, tcg_rd, 8);

Indeed that improves things a bit for sf=1. For sf=0 though the
constant is never loaded into a register, it is passed to the and
instructions as an immediate.

-- 
Aurelien Jarno                          GPG: 4096R/1DDD8C9B
address@hidden                 http://www.aurel32.net



reply via email to

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