RE: [PATCH 11/15] Hexagon (target/hexagon) circular addressing

[Taylor Simpson]

> -----Original Message-----
> From: Richard Henderson <richard.henderson@linaro.org>
> Sent: Thursday, March 25, 2021 11:34 AM
> To: Taylor Simpson <tsimpson@quicinc.com>; qemu-devel@nongnu.org
> Cc: philmd@redhat.com; ale@rev.ng; Brian Cain <bcain@quicinc.com>
> Subject: Re: [PATCH 11/15] Hexagon (target/hexagon) circular addressing
>
> On 3/24/21 8:50 PM, Taylor Simpson wrote:
> > +/*
> > + * Many instructions will work with just macro redefinitions
> > + * with the caveat that they need a tmp variable to carry a
> > + * value between them.
> > + */
> > +#define fGEN_TCG_tmp(SHORTCODE) \
> > +    do { \
> > +        TCGv tmp = tcg_temp_new(); \
> > +        SHORTCODE; \
> > +        tcg_temp_free(tmp); \
> > +    } while (0)
>
> Wow, this is awfully magical.  I don't even see "tmp" referenced in the
> SHORTCODE.  What's going on here?

Here's an example:L2_loadrub_pci
Semantics:{fEA_REG(RxV); fPM_CIRI(RxV,siV,MuV); fLOAD(1,1,u,EA,RdV);}

The need for tmp is inside the fPM_CIRI macro
#define fPM_CIRI(REG, IMM, MVAL) \
    do { \
        TCGv tcgv_siV = tcg_const_tl(siV); \
        gen_helper_fcircadd(REG, REG, tcgv_siV, MuV, fREAD_CSREG(MuN)); \
        tcg_temp_free(tcgv_siV); \
    } while (0)

Notice the use of fREAD_CSREG
#define fREAD_CSREG(N) (READ_REG(tmp, HEX_REG_CS0 + N))

So, we're using tmp to carry the value read from the CS reg to the 
gen_helper_fcircadd call.


I'll go ahead and move the create/free tmp inside the fPM_CIRI macro.

Thanks,
Taylor