[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Gcl-devel] Re: [Gcl-commits] gcl/cmpnew cmpmain.lsp
From: |
Mike Thomas |
Subject: |
Re: [Gcl-devel] Re: [Gcl-commits] gcl/cmpnew cmpmain.lsp |
Date: |
Thu, 25 Jul 2002 11:08:50 +1000 |
Hi there.
> 1) Are these compiler flags portable to different
> compilers/assemblers?
In general I suppose you can't count on it:
> Specifically solaris?
Using this test file:
===============
int sumit (int a, int b)
{
return (a + b );
}
===============
on Sparc Solaris with "cc -S -c t.c" produces a pile of junk including the
following extract:
================
! File t.c:
! 1 int sumit (int a, int b)
! 2 {
! 3 return (a + b );
ld [%fp+68],%l0
ld [%fp+72],%l1
add %l0,%l1,%l0
st %l0,[%fp-4]
ba .L13
nop
! block 2
ba .L13
nop
! block 3
.L17:
.L18:
.L13:
ld [%fp-4],%l0
mov %l0,%i0
jmp %i7+8
restore
.size sumit,(.-sumit)
.align 8
==================
This isn't really a dissassembly, but in theory you can pass -S to the
assembler for a dissassembler using "-Wa,-S" and it amounts to the same
thing I would expect - I just couldn't get this to work when I tried it
today.
>
> 2) Having biased myself by getting accustomized to objdump output, I'd
> likt to hear what people think the advantages are of this output
> vis a vis objdump. The obvious one is the interleaving of the C
> source. But I wanted to know if people think relying on objedump
> in general is a problem. Isn't it (at least) as widely available
> as the availability of the compiler flag combo used below?
Gcc is usually distributed with objdump so this is probably true. Objdump
doesn't seem to be present on the Solaris machine I did the above test on
however.
The other advantage is that it reduces the problem down to just one
sub-process call plus some LISP filter code.
All-in-all, the interleaving is the only real advantage from my perspective.
> 3) We still need some kind of option to disassemble to turn on or off
> the -g, and restructure the output accordingly, I'd think.
Agreed. As mentioned in some overenthusiastic four-in-the-morning email a
week ago, I'll be happy to look at this once I extract myself from the BFD
and ANSI quagmire, if no-one else wants to do it, and we are agreed it is a
good move overall.
Cheers
Mike Thomas.