[cfe-dev] Driver: Default CPUs

[James Molloy]

Hi Eric,

 

IMO you should specify the minimum cpu you want to run on via

-march and the one you want to tune for via -mtune (not that we really
support

doing this in the llvm backend anyhow) and that -mcpu should basically

die in a fire. Historically though the cpu is automagically set from the
architecture

targeted and I think we should continue that and use default tuning to

some processor of that sort.

 

OK, I give in, I'll find a way of getting the defaults generated from ARM.td
(perhaps by #including ARMGenSubtarget.inc?) instead of deleting it
outright.

 

I may delay this though until after the conference in case people have
different ideas about the direction of the driver.

 

Cheers,

 

James

 

From: Eric Christopher [mailto:echristo at apple.com] 
Sent: 15 October 2011 02:38
To: James Molloy
Cc: cfe-dev at cs.uiuc.edu
Subject: Re: [cfe-dev] Driver: Default CPUs

 

 

On Oct 12, 2011, at 2:40 AM, James Molloy wrote:





The arguments I've heard so far are:
 *  Mine, that there should be no default CPU selected if the user doesn't
select it himself. I feel it adds a hidden option that the user is unaware
of, and depending on what that default is may cause inferior performance on
the CPU the user actually runs it on.
 * Jim's (? I forget who responded ?), that there *should* be a default CPU
all the time, but perhaps some pseudo-CPUs could be added that exhibit a
blended scheduling. For example for ARM, a "v7" pseudo-cpu could be added
that has an itinerary that performs well on all v7 cores. By the way, I
think ARM is the only high-visibility target with such a difference between
cores that this matters. I assume MIPS and PowerPC are similar though?



 

What do people think? Do people care?

 

I think I hate the arm notion of "cpu" in the output file. Only arch and
tune should

matter ;)

 

MIPS and (IIRC) powerpc don't work this way. For MIPS -march is the minimum

ISA you're compiling for and if it's a specific processor then by default
will

tune for that unless you want something else. There's no real idea of a
"generic"

cpu for tuning or scheduling itinerary, you'd usually tune for whatever
processor

is the most important.

 

So it looks like this decision is mostly ARM related. For ease of
compatibility

I'd suggest doing whatever gcc does, but I don't really think it matters
that

much. 

 

IMO you should specify the minimum cpu you want to run on via

-march and the one you want to tune for via -mtune (not that we really
support

doing this in the llvm backend anyhow) and that -mcpu should basically

die in a fire. Historically though the cpu is automagically set from the
architecture

targeted and I think we should continue that and use default tuning to

some processor of that sort.

 

-eric
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/cfe-dev/attachments/20111017/a826a0a2/attachment.html>