Path: utzoo!attcan!utgpu!jarvis.csri.toronto.edu!mailrus!tut.cis.ohio-state.edu!ucbvax!decwrl!sun-barr!apple!amdcad!crackle!tim
From: tim@crackle.amd.com (Tim Olson)
Newsgroups: comp.arch
Subject: Re: Register usage [was Re: 80486 vs. 68040 code size]
Message-ID: <25689@amdcad.AMD.COM>
Date: 19 May 89 16:55:11 GMT
References: <959@aber-cs.UUCP>
Sender: news@amdcad.AMD.COM
Reply-To: tim@amd.com (Tim Olson)
Distribution: eunet,world
Organization: Advanced Micro Devices, Inc. Sunnyvale CA
Lines: 32
Summary:
Expires:
Sender:
Followup-To:

In article <959@aber-cs.UUCP> pcg@cs.aber.ac.uk (Piercarlo Grandi) writes:
| Somebody then offered some DATA points; John Mashey has offered his
| recollection that 24-28 is a good number to have on a MIPS (hence their 32),
| and some AMD fellow (Tim Olson?) has posted statistics that, using a static
| analysis, their compiler can on average fill at most 6+7 registers after
| global optimization (I am not clear whether this includes intra expression
							    ^^^^^
Don't you mean "inter" here?

| optimization or not), thus providing an upper bound to the number of *useful*
| registers.

The statistics posted said that, on average, 6.5 global registers (not
live across a function call) and 7.0 local registers (live across a
function call) were used per function. However, this is certainly not an
"upper bound to the number of *useful* registers."  As others have
pointed out, registers can also be used for parameter passing and return
values, emulating a stack cache, etc.

| 	I cannot resist also apologizing, in a different vein :-), to John
| 	Mashey for his opinion that I may not be aware of what has happened
| 	in the latest 15-20 years in compiler research, and for later stating
| 	that I spout falsehoods (numbers, please...); and to Chris Torek's
	^^^^^^^^^^^^^^^^^^^^^^^
That was me.  I sort of flew off the handle there, but you made certain
assertions that *you* have not backed up by any data.



	-- Tim Olson
	Advanced Micro Devices
	(tim@amd.com)