[Search for users] [Overall Top Noters] [List of all Conferences] [Download this site]

Conference decwet::visual

Title:	Microsoft Visual C++ bug reports and kits
Notice:	Register in Topic 2. 5.Last for latest Kit
Moderator:	DECWET::THOMASN

Created:	Tue May 17 1994
Last Modified:	Fri Jun 06 1997
Last Successful Update:	Fri Jun 06 1997
Number of topics:	521
Total number of notes:	2938

490.0. "Intel vs Axp" by ODIXIE::DWILLIAMS () Thu Apr 03 1997 09:20

    Hello!
    
    I have a military customer who has ported their wargaming engine from
    Unix to NT using Visual C++. They have VC++ 1.0 on the Axp and 4.2 on
    the Intels they are trying. Unfortunately, the 180 Mhz Pentium beats 
    the 433Mhz Axp 4100. Both are running NT Server 3.51 SP4. 
    
    Can the 1.0 VC++ be a problem vs 4.2?
    
    Also, I see in other notes that perhaps other versions of the Axp C++
    compiler may be faster than 4.2. What version of the Axp compiler
    would be best?
    
    Last, What other things in the codes should I look for that would cause
    the Axp to suffer? Can you also mention fixes??
    
    Thanks,
    
    Dave

T.R	Title	User	Personal Name	Date	Lines
490.1		GEMEVN::GLOSSOP	Only the paranoid survive	`Thu Apr 03 1997 09:52`	2
	VC++ 5.0 for Alpha should be available shortly, and should be considerably better. (The difference is several years of work on the compiler backend...)
490.2	probably automatice runtime fixups	DECWET::PETERSON		`Thu Apr 03 1997 10:58`	24
	The oldest Alpha version of VC++ is V2.0 (there is no V1.0) I'm guessing that the alpha kit is actually V4.1 (which is the last widely distributed Alpha version) You can verify by bringing up the Help.About VC++ If it is the V4.1 kit, there are significant optimization improvements for C++ in VC++ V5.0 (available in 1-2 months), but only minor improvements for most C code. With either the V2.0 or V4.1 Alpha kit, the compiled code should still run faster on an alpha 433 than on an Intel 180. I would suspect the performance problem on alpha to come from misaligned data. Have the customer check on alignment faults/second in the NT performance monitor while running the application - there should be zero but I'm guessing there are more than 10/second There is a description of how to inspect your program for possible alignment problems in the VC++ for Risc online documentation (See Porting your application in the Risc Programmer's Guide) If the program uses floating point, verify that the application was NOT compiled with /QAieee. If it was, it is probably encountering runtime floating point fixups that could also make an alpha slower than a pentium.
490.3		BIGUN::nessus.cao.dec.com::Mayne	A wretched hive of scum and villainy	`Thu Apr 03 1997 15:17`	4
	Can you please send me some info about this wargaming engine? Some of my customers might be interested in a look. PJDM ([email protected])
490.4	thanks!	ODIXIE::DWILLIAMS		`Thu Apr 03 1997 18:18`	10
	Thanks, all for the quick responces! I'll go back in armed with this and see where I come out... PJDM, I'll try to find out more about the codes and if they are available... Thanks, Dave
490.5	Use PERFMON	DECWET::MVB	Monty VanderBilt	`Thu May 08 1997 09:34`	4
	You can use PERFMON to check alignment faults. Add the System/Alignment Fixups/Sec counter. You should normally see almost none. More than a few hundred per second is a big performance hit.
490.6		DECCXL::OUELLETTE	mudseason into blackfly season	`Thu May 08 1997 12:44`	9
	To debug alignment faults... Install VC++ V5.0. Enable alignment faults with "axpalign/enable". Reboot. Rerun your program... Unless you've added a call to SetErrorMode, alignment faults will pop up the usual window to allow you to debug (just like an access violation). It's usually easy to fix these problems. R.