Skip to contentSkip to conversation

Instruction Execution

Linux-ia64 Mattox, Dave 4 participants 5 messages Oct 17, 2000

lore
Share & tools
lore

Mattox, Dave

author
#0 lore raw

I just got my Itanium Machine in and after installing Linux (no problems with that, good work guys), I tried some timings.

I ran a loop 666,000,000 times and it appears that each bundle is taking 3 seconds to execute. So this means 3 cycles per bundle,
or 1 cycle per instruction. Is the Lion SDV Machine executing instructions serially? If it is, is it Linux setting this, or are the beta chips disabled in some manner?

Thanks in advance for help on this,
Dave Mattox

#1 lore raw
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line · Dave

Dave> I just got my Itanium Machine in and after installing Linux Dave> (no problems with that, good work guys), I tried some timings.

Dave> I ran a loop 666,000,000 times and it appears that each bundle Dave> is taking 3 seconds to execute. So this means 3 cycles per Dave> bundle, or 1 cycle per instruction. Is the Lion SDV Machine Dave> executing instructions serially? If it is, is it Linux setting Dave> this, or are the beta chips disabled in some manner?

Make sure "Multidispersal" is enabled in the advanced chipset configuration menu of the firmware (BIOS).

	--david
#2 lore raw
In reply to Mattox, Dave
Quoted 7 lines · Dave

I ran a loop 666,000,000 times and it appears that each bundle is taking 3 seconds to execute. So this means 3 cycles per bundle,
or 1 cycle per instruction. Is the Lion SDV Machine executing instructions serially? If it is, is it Linux setting this, or are the beta chips disabled in some manner?

ia64 does not do instruction reordering, at least this is what I've heard, and nothing which I have read until now seems to deny this. All insn ordering for faster execution should therefore be done by the compiler.

No URL to give you, sorry...

Signature · 4 lines
Francis Galiegue, [email protected]
"Programming is a race between programmers, who try and make more and more
idiot-proof software, and universe, which produces more and more remarkable
idiots. Until now, universe leads the race"  -- R. Cook
#3 lore raw
In reply to Mattox, Dave

Instruction timing depends on what instructions you are executing, and where the stop bits are. If you are running a loop with FP instructions, or a loop with a lot of serial instruction dependencies, then 1 cycle per instruction is certainly possible. Since you didn't give any assembly code samples, it isn't possible to tell if the loop was poorly optimized or not.

Jim

author
#4 lore raw
In reply to Mattox, Dave

That was the problem, its now working exactly as I expected :)

Thanks for the help everyone,
Dave

-----Original Message-----
From: [email protected] [mailto:[email protected]]On Behalf Of David Mosberger
Sent: Tuesday, October 17, 2000 10:16 AM
To: Mattox, Dave
Cc: [email protected]
Subject: Re: [Linux-ia64] Instruction Execution

Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line

On Tue, 17 Oct 2000 09:51:51 -0700, "Mattox, Dave"

<[email protected]> said:

Dave> I just got my Itanium Machine in and after installing Linux Dave> (no problems with that, good work guys), I tried some timings.

Dave> I ran a loop 666,000,000 times and it appears that each bundle Dave> is taking 3 seconds to execute. So this means 3 cycles per Dave> bundle, or 1 cycle per instruction. Is the Lion SDV Machine Dave> executing instructions serially? If it is, is it Linux setting Dave> this, or are the beta chips disabled in some manner?

Make sure "Multidispersal" is enabled in the advanced chipset configuration menu of the firmware (BIOS).

	--david

_______________________________________________
Linux-IA64 mailing list
[email protected]
http://lists.linuxia64.org/lists/listinfo/linux-ia64