On Tue, 5 Mar 2019 at 09:14, gérard Calliet <[email protected]> wrote:
> Hello, > > Again I have some performance issues with simh. > > I built it on OpenVMS Itanium. It runs on a process with enought > memory, it takes 1 of the 8 CPU Itanium with 100%, no page faults, and > 100 BIO by second. > > On the emulated VAX we are running processes CPU intensive (old Ada > compiler), they take 100% of the CPU, also very few page fault, very few > IO. > > And the CPU time is about 2 or 3 times the CPU time on the hardware, and > so elapsed time 3 or 4 times. (For my understanding, it is difficult to > think the same stream of instructions can use different CPU times > between hardware and SIMH). > > Where can I do something to improve performance? Are there parameters > (or constant in compilation) in SIMH which could help? And the specific > context is processes doing almost only CPU computing. > > I'm pretty sure I said this the last time you asked, but compile SIMH with profile-directed feedback. I'm not a VMS person so I know very little about the compiler there, but I took a quick look at the manual and it appears to be a supported option. In my testing, profile-directed feedback from a real-life workload increased SIMH's speed significantly, up to 66% if memory serves. -Henry
_______________________________________________ Simh mailing list [email protected] http://mailman.trailing-edge.com/mailman/listinfo/simh
