Hi Ben, thanks for the answer, but I think you misunderstood both questions. Let me re-phrase,
to a) do the standalone benchmarks Richards, Splay, DeltaBlue etc. have self-checking built-in? Meaning is there some kind of results check, whether the benchmark ran correctly? to b) let's say I compile my d8 binary, one time for x86. Now I am running the benchmarks above on 2 different x86 processors (binary is the same!). Would you expect the same executed/retired instruction count? One reason why you wouldn't have the same count would be for instance, if somebody in the benchmark measures time, and there are loops that spin based on that or anything like that. Hope question is more clear now, Thanks, Hendrik Greving On Thursday, May 29, 2014 1:51:01 PM UTC-7, Ben Noordhuis wrote: > > On Thu, May 29, 2014 at 7:53 PM, Hendrik Greving > <[email protected] <javascript:>> wrote: > > Hi, I looked at the code of these benchmarks and kind of have an idea. > > What's unclear to me was a) whether there is some self-checking built-in > and > > b) let's say I am running this between different chip architectures but > the > > same instruction set, let's say an old and a new Intel x86 chip, but > exactly > > the same instruction set, would you expect the same number of > instructions > > being executed? Basically, are there time components built-in that may > vary > > the number of instructions being executed? > > > > Thanks for answers with regards to a) and b) > > As to a), V8 checks the capabilities of the CPU it's running on and > adjusts its code generator accordingly. SSE3 and SSE4 instructions, > for example, are conditionally emitted. So yes, that happens. > > I'm not sure what you mean with b). If you're asking whether V8 > optimizes instruction sequences for specific micro-architectures, like > gcc does with the -march= switch, then I believe the answer is 'no'. > -- -- v8-users mailing list [email protected] http://groups.google.com/group/v8-users --- You received this message because you are subscribed to the Google Groups "v8-users" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. For more options, visit https://groups.google.com/d/optout.
