On 4/3/19 1:56 AM, Peter Maydell wrote:
> On Wed, 3 Apr 2019 at 18:37, Richard Henderson
>> * extract2_i32/i64 dest, t1, t2, pos
>>
>> For N = {32,64}, extract an N-bit quantity from the concatenation
>> of t2:t1, beginning at pos. The tcg_gen_extract2_* expander allows
>> values 0 <= pos <= N, but will expand 0 and N with mov, so only
>> 1 <= pos <= N-1 will be seen by the host tcg_out_op.
>
> If I'm reading that correctly, it seems to be combining in one sentence
> the behaviour of the TCG API exposed to the front-end (pos can be
> between 0 and N inclusive) with a detail of the API that a backend
> needs to care about (that it can assume it never sees 0 or N).
You're not wrong. ;-P
> I think we should be more careful to keep those separate, because
> a reader of this document is almost always going to care only about
> one or the other, never both at the same time. Perhaps things that
> apply only to the backend end of the interface should go in section 4
> of tcg/README?
Sadly, there's a lot of mix up on that count.
Indeed, the very next paragraph,
> * extrl_i64_i32 t0, t1
>
> For 64-bit hosts only, extract the low 32-bits of input T1 and place it
> into 32-bit output T0. Depending on the host, this may be a simple move,
> or may require additional canonicalization.
is entirely about the "section 4" opcode. The "section 2" function,
tcg_gen_extrl_i64_i32, is expanded correctly for 32-bit hosts as a
simple move from the i32 "sub-temp" of the i64 temp.
Clearly the whole thing should be reorganized, but I'm not sure how best that
should be done.
> At any rate I think they should at least be in different sentences :-)
Now that I can do. ;-)
r~
