[Info-vax] Some VMS/x86 perf test data from WASD maintainer.

[David Jones]

On Thursday, April 27, 2023 at 7:05:29 AM UTC-4, Volker Halle wrote:
> Jan-Erik Söderholm schrieb am Donnerstag, 27. April 2023 um 11:24:46 UTC+2: 
> 
> After now also running a VSI OpenVMS x86-64 E9.2-1 system (using VMware Workstation Player 16) and installing the C compiler, I have tried to 'calibrate' the procedure to obtain the VUPS values for x86-64 using the PRIME_SIEVE.C tool (calculating prime numbers up to 10^10). The CPU_MULTIPLIER value for x86_64 should be set to 8 - you can easily edit this procedure to add (at the right place): $ IF F$GETSYI("ARCH_NAME").EQS."x86_64" THEN $ cpu_multiplier=8 
> 
> The resulting VUPS values for x86-64 are now more realistic: 5900 VUPS on an i5-9600K CPU @3.7 GHz 
> 
> On this OpenVMS x86_64 system, PRIME_SIEVE needs 19.34 sec to calculate primes up to 10^10. On an ES45 Model 2B (EV68CB 1250 MHz) system it took 49.04 seconds, that system was measured at 2414 VUPS. 
> 
> See previous discussion in: https://groups.google.com/g/comp.os.vms/c/ouhZ7C7bgoQ/m/_iEqNzm1AwAJ 

I use the old Byte Magazine benchmark from the 1990s (bytemark, normalized to a 90Mhz Pentium). Running in on
my laptop under VirtualBox with an Intel i5-1235U gets an integer score 6-7 times higher than my DS10L/466 Alpha.
Interestingly, the floating point score is barely higher. Running the posted sieve code on both, the laptop is 11 times
faster.

At some point, I'll install visual studio on the laptop and see what difference the compiler makes. The current VSI
compiler is doing some optimization (unlike the cross compiler), as building with /opt=level=0 does slow it down.