Re: [Qemu-devel] [PATCH v3.5 2/8] target/s390x: Implement CONVERT UNICODE insns

[Richard Henderson]

On 07/11/2017 08:46 PM, Thomas Huth wrote:
> > +        s1 = cpu_ldub_data_ra(env, addr + 1, ra);
> > +        s2 = cpu_ldub_data_ra(env, addr + 2, ra);
> > +        c = s0 & 0x0f;
> > +        c = (c << 6) | (s1 & 0x3f);
> > +        c = (c << 6) | (s2 & 0x3f);
> > +        /* Fold the byte-by-byte range descriptions in the PoO into
> > +           tests against the complete value.  It disallows encodings
> > +           that could be smaller, and the UTF-16 surrogates.  */
> > +        if (enh_check
> > +            && ((s1 & 0xc0) != 0x80
> > +                || (s2 & 0xc0) != 0x80
> > +                || c < 0x1000
> > +                || (c >= 0xd800 && c <= 0xdfff))) {
> > +            return 2;
> > +        }
> > +    } else if (s0 <= (enh_check ? 0xf4 : 0xf7)) {
> > +        /* four byte character */
> > +        l = 4;
> > +        if (ilen < 4) {
> > +            return 0;
> > +        }
> > +        s1 = cpu_ldub_data_ra(env, addr + 1, ra);
> > +        s2 = cpu_ldub_data_ra(env, addr + 2, ra);
> > +        s3 = cpu_ldub_data_ra(env, addr + 3, ra);
> > +        c = s0 & 0x0f;
> I think you could also use 0x07 instead of 0x0f here. Shouldn't matter
> much due to the s0 <= 0xf7 check above, though.

You're right that it doesn't matter due to the check, but fixed anyway.  It 
makes the pattern wrt 2, 3, and 4 byte encodings more apparent.


r~