On Thu, 24 Feb 2000, Horst von Brand wrote:
> Artur Skawina <skawina@geocities.com> said:
>
> [...]
>
> > actually somebody updated the code a few months ago. it used to be
> > broken for newer gccs (wrong asm constraints). i have that patch
> > marked 2-look-at, but i think it also went into 2.3 later.
> > the other issue is that some of the functions are huge, it's unobvious
> > that inlining them is a good idea, and apparetnly there are some
> > doubts wrt their correctness.
>
> New inline versions of string and mem functions (in particular for ia32)
> went into the gcc main branch recently. I'll plead again for wasting the
> in-kernel definitions, as (a) the compiler can handle builtins more
> efficiently than inline asm (f.ex. interleave them with the surrounding
> code if useful, optimize them in the context of their use as the compiler
> knows their inards (which it doesn't for inline asm)), and (b) compiler
> builtins will get much wider exposure, thus reducing the number and
> severity of bugs in that code, and finally (c) you get it for free.
egcs-20000207-20000214.diff
17809 + /* If we get here, just use the library routine. */
17810 + emit_library_call (gen_rtx (SYMBOL_REF, Pmode, "memcpy"), 0, VOIDmod 17811 + operands[0], Pmode, operands[1], Pmode, operands[2],
17812 + SImode);
17813 + }
Unfortunately, we seem to get there in aic7xxx.c.. result is no link.
(we didn't get a free builtin memcpy() along with the generated call:)
-Mike
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Tue Feb 29 2000 - 21:00:11 EST