Another attempt to send a rejected message

[Paul Koning <pkoning at equallogic dot com>]

I sent this message a few minutes ago, but it was rejected by a
defective spam filter at RedHat.  So I'm trying it again, this time
with a different subject string in the hopes that it doesn't trigger
the same bug again.

    paul

--- Begin Message ---

• From: Paul Koning <pkoning at equallogic dot com>
• To: alexr at spies dot com
• Cc: gcc at gcc dot gnu dot org
• Date: Fri, 7 May 2004 09:25:08 -0400
• Subject: Re: MS/CW-style inline assembly for GCC

>>>>> "Alex" == Alex Rosenberg <alexr@spies.com> writes:

 Alex> CW indeed does understand all of the inline asm
 Alex> instructions. It optimizes them and schedules them into the
 Alex> surrounding code. 

That's a terrifying concept.

The one really major design error in the MIPS assembler was its notion
that it would mess around with the code you write.

You shouldn't be writing assembler unless you're prepared to (and
qualified to) write it optimally -- or you're dealing with a case
where it doesn't matter, such as many diagnostics.  But to want the
low level fiddling that assembler lets you do and yet expect the
assembler to "optimize" the result is a completely broken concept.

The tie to the optimizer in GCC is that the asm statement lets you
talk about clobbers, and about where operands should be materialized
so the register allocator can take care of things.  But the asm text
itself is an opaque string.  That is the right way.

	  paul

--- End Message ---