[Gcl-devel] Re: compiler speedup hack

[Robert Dodier]

hello camm, thanks for the update.

> 1) Not having gcc hit the file system -- this appears to be possible
>    all the way up to the assembly stage.  The assembler (gas) is also
>    bfd based, and cannot therefore output to stdout (file must be
>    seekable).  We actually have the source to gas under binutils, and
>    it is conceivable that we could incoporate it and modify it
>    to write to a block of memory, then use gcc with -S with input and
>    output pipes.  It would be a medium sized project, and I'm not sure
>    what performance gain we'd see given Bob's results below.

it seems like it must be possible to construct something
that acts like a seekable device in memory, and then present
that to gas. it seems like that would be much easier than modifying gas ...

> 2) gcc warm start -- This cmpinclude preprocess is the closest thing I
>    can see at present.  It is too bad that gcc (apparently) does not
>    have any incremental compile ability (even if only to assembler),
>    i.e. read from stdin, special char means flush stdout, etc.  If you
>    run across same or similar, please let me know.

well, i guess i would suggest making a request for the appropriate
feature (warm start / incremental compilation / whatever) to the gcc
developers, in the hope that if the feature is requested often enough
someone will get around to working on it. a slim chance, i know.

regards,
robert dodier