[PATCH v11 16/40] c, c++: Use 16 bits for all use of enum rid for more keyword space

[Ken Matsui]

On Thu, Sep 14, 2023 at 10:54 AM Joseph Myers <joseph@codesourcery.com> wrote:
>
> On Wed, 13 Sep 2023, Ken Matsui via Gcc-patches wrote:
>
> > diff --git a/gcc/c/c-parser.h b/gcc/c/c-parser.h
> > index 545f0f4d9eb..eed6deaf0f8 100644
> > --- a/gcc/c/c-parser.h
> > +++ b/gcc/c/c-parser.h
> > @@ -51,14 +51,14 @@ enum c_id_kind {
> >  /* A single C token after string literal concatenation and conversion
> >     of preprocessing tokens to tokens.  */
> >  struct GTY (()) c_token {
> > +  /* If this token is a keyword, this value indicates which keyword.
> > +     Otherwise, this value is RID_MAX.  */
> > +  ENUM_BITFIELD (rid) keyword : 16;
> >    /* The kind of token.  */
> >    ENUM_BITFIELD (cpp_ttype) type : 8;
> >    /* If this token is a CPP_NAME, this value indicates whether also
> >       declared as some kind of type.  Otherwise, it is C_ID_NONE.  */
> >    ENUM_BITFIELD (c_id_kind) id_kind : 8;
> > -  /* If this token is a keyword, this value indicates which keyword.
> > -     Otherwise, this value is RID_MAX.  */
> > -  ENUM_BITFIELD (rid) keyword : 8;
> >    /* If this token is a CPP_PRAGMA, this indicates the pragma that
> >       was seen.  Otherwise it is PRAGMA_NONE.  */
> >    ENUM_BITFIELD (pragma_kind) pragma_kind : 8;
>
> If you want to optimize layout, I'd expect flags to move so it can share
> the same 32-bit unit as the pragma_kind bit-field (not sure if any changes
> should be made to the declaration of flags to maximise the chance of such
> sharing across different host bit-field ABIs).
>
> > diff --git a/gcc/cp/parser.h b/gcc/cp/parser.h
> > index 6cbb9a8e031..3c3c482c6ce 100644
> > --- a/gcc/cp/parser.h
> > +++ b/gcc/cp/parser.h
> > @@ -40,11 +40,11 @@ struct GTY(()) tree_check {
> >  /* A C++ token.  */
> >
> >  struct GTY (()) cp_token {
> > -  /* The kind of token.  */
> > -  enum cpp_ttype type : 8;
> >    /* If this token is a keyword, this value indicates which keyword.
> >       Otherwise, this value is RID_MAX.  */
> > -  enum rid keyword : 8;
> > +  enum rid keyword : 16;
> > +  /* The kind of token.  */
> > +  enum cpp_ttype type : 8;
> >    /* Token flags.  */
> >    unsigned char flags;
> >    /* True if this token is from a context where it is implicitly extern "C" */
>
> You're missing an update to the "3 unused bits." comment further down.
>
> > @@ -988,7 +988,7 @@ struct GTY(()) cpp_hashnode {
> >    unsigned int directive_index : 7;  /* If is_directive,
> >                                          then index into directive table.
> >                                          Otherwise, a NODE_OPERATOR.  */
> > -  unsigned int rid_code : 8;         /* Rid code - for front ends.  */
> > +  unsigned int rid_code : 16;                /* Rid code - for front ends.  */
> >    unsigned int flags : 9;            /* CPP flags.  */
> >    ENUM_BITFIELD(node_type) type : 2; /* CPP node type.  */
>
> You're missing an update to the "5 bits spare." comment further down.
>
> Do you have any figures for the effects on compilation time or memory
> usage from the increase in size of these structures?
>

Here is the benchmark result:

https://github.com/ken-matsui/gsoc23/tree/main/reports/gcc-build

I can see regression for compilation time but not for memory.

* Time: +0.950995%
* Memory: No difference proven at 95.0% confidence

Sincerely,
Ken Matsui

> --
> Joseph S. Myers
> joseph@codesourcery.com