Re: [PATCH v2 02/81] tcg: Replace TCGOP_VECL with TCGOP_TYPE

[Philippe Mathieu-Daudé]

On 8/1/25 22:23, Richard Henderson wrote:
> On 1/8/25 09:39, Philippe Mathieu-Daudé wrote:
> > On 7/1/25 08:59, Richard Henderson wrote:
> > > In the replacement, drop the TCGType - TCG_TYPE_V64 adjustment,
> > > except for the call to tcg_out_vec_op.  Pass type to tcg_gen_op[1-6],
> > > so that all integer opcodes gain the type.
> > >
> > > Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> > > ---
> > >   include/tcg/tcg.h      |   2 +-
> > >   tcg/tcg-internal.h     |  13 ++---
> > >   tcg/optimize.c         |  10 +---
> > >   tcg/tcg-op-ldst.c      |  26 ++++++----
> > >   tcg/tcg-op-vec.c       |   8 +--
> > >   tcg/tcg-op.c           | 113 +++++++++++++++++++++++------------------
> > >   tcg/tcg.c              |  11 ++--
> > >   docs/devel/tcg-ops.rst |  15 +++---
> > >   8 files changed, 105 insertions(+), 93 deletions(-)
> >
> >
> > > @@ -3320,7 +3333,7 @@ void tcg_gen_exit_tb(const TranslationBlock 
> > > *tb, unsigned idx)
> > >           tcg_debug_assert(idx == TB_EXIT_REQUESTED);
> > >       }
> > > -    tcg_gen_op1i(INDEX_op_exit_tb, val);
> > > +    tcg_gen_op1i(INDEX_op_exit_tb, 0, val);
> >
> > Having TCG_TYPE_I32 = 0 is confusing.
> >
> > Should we define TCG_TYPE_VOID = 0 first, shifting all other types
> > (and updating tcg_type_size())?
>
> No.  E.g. tcg/aarch64/tcg-target.c.inc:
>
> /* We're going to re-use TCGType in setting of the SF bit, which controls
>     the size of the operation performed.  If we know the values match, it
>     makes things much cleaner.  */
> QEMU_BUILD_BUG_ON(TCG_TYPE_I32 != 0 || TCG_TYPE_I64 != 1);
>
> Obviously this value is never used for INDEX_op_exit_tb (and others).  
> If you really think it's important, we could add VOID as another value.  
> But I think that might be more confusing than anything else.  So far 
> we're restricted to integer and vector types.  If we add a third 
> classification, we're going to want to adjust all of the checks everywhere.

No need for VOID, I'm good ;)

Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>