On 18 May 2012 17:26, Jeroen Demeyer <jdeme...@cage.ugent.be> wrote: > On 2012-05-18 17:29, Bill Hart wrote: >> I'm not sure how virtual boxes work. Do they emulate the CPU as well as the >> OS? >> >> Unfortunately, there isn't really a lot we can do if virtual box is >> going out of its way to report incorrect cpu parameters through the >> *assembly instruction* cpuid, if that is what is happening (I am not >> sure if it is or not). >> >> Unfortunately, the solution does appear to be to build MPIR with >> -march=native, which apparently works. >> >> If anyone has any deep insight into this, please let us know. Maybe it >> is possible for us to do something about it upstream. I'm keen to fix >> it, I just don't understand the problem well enough. > > I think the following /proc/cpuinfo explains what happens: > > processor : 0 > vendor_id : GenuineIntel > cpu family : 6 > model : 42 > model name : Intel(R) Core(TM) i5-2500K CPU @ 3.30GHz > stepping : 7 > cpu MHz : 3302.482 > cache size : 6144 KB > physical id : 0 > siblings : 4 > core id : 0 > cpu cores : 4 > apicid : 0 > initial apicid : 0 > fpu : yes > fpu_exception : yes > cpuid level : 5 > wp : yes > flags : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca > cmov > pat pse36 clflush mmx fxsr sse sse2 ht syscall nx lm constant_tsc > rep_good nopl pni ssse3 lahf_lm > bogomips : 6604.96 > clflush size : 64 > cache_alignment : 64 > address sizes : 36 bits physical, 48 bits virtual > power management: > > According to vendor/family/model, this is a Sandy Bridge. However, the > CPU features (see flags) are seriously reduced compared to a real Sandy > Bridge. > > Compiling with -march=SOMETHING means that 3 things have to be satisfied: > (1) GCC must understand the command line flag > (2) The assembler must understand the instructions > (3) The processor must support the instructions > > Based on my observations, it seems that MPIR already checks (1) and (2). > With -march=native, (1) and (3) are guaranteed, but not (2). When > checking CFLAGS in MPIR, would it be possible to actually *run* an > executable compiled with -march=SOMETHING, thereby checking (3)? >
Configure does run actual compile tests for the various -march options, I believe. But the problem (3) won't show up unless code specifically designed to compile to instructions that are only supported on that processor is compiled. I don't think this is feasible (and it is an autotools/configure issue anyway, not MPIR). > Alternatively, check the feature flags from CPUID and lower the > architecture based on that. > This would be feasible I think, but basically defeats the entire strategy of selecting CPUs in MPIR. So does virtual box emulate a CPU with reduced features or something? Do they consider this a bug, do you think? Bill. -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
