Pretty much every instruction, save a purely fp one, uses at least one int register t execute and then will increment int_alu_accesses.
Ali On Jul 22, 2011, at 6:26 AM, Iordan Alexandru wrote: > > Thank you for your reply Ali. > > If system.switch_cpus0.iq.int_alu_accesses is not a good counter for the > total > number of INT instructions executed, I have another question. In the example > from my previous e-mail, the number of committed INT instructions > (system.switch_cpus0.commit.int_insts) was significantly greater than the > total > number of INT instructions (total_INT = > system.switch_cpus0.iq.FU_type_0::IntAlu + > system.switch_cpus0.iq.FU_type_0::IntMult + > system.switch_cpus0.iq.FU_type_0::IntDiv). Why is this happening? Should I > use > another counter for total_INT ? > > > > > ----- Original Message ---- > From: Ali Saidi <[email protected]> > To: gem5 users mailing list <[email protected]> > Sent: Fri, July 22, 2011 8:47:45 AM > Subject: Re: [gem5-users] INT and FP counts > > Int alu accesses is counting every register that has at least one integer > operand. > > Ali > > > On Jul 21, 2011, at 5:40 AM, Iordan Alexandru wrote: > >> >> Hello >> >> I am using gem5 to run a MP full-system simulation with Alpha. I am >> interested > >> in profiling some applications in terms of instruction mix and for that >> reason >> I >> >> want to clarify something about the INT and FP instruction counts (total and >> committed). >> >> From my experience there are a two "candidates" for total count of >> instructions >> >> (INT and FP): >> >> system.switch_cpus0.iq.int_alu_accesses 54347417 >> >> system.switch_cpus0.iq.FU_type_0::IntAlu 23500814 >> system.switch_cpus0.iq.FU_type_0::IntMult 4209 >> system.switch_cpus0.iq.FU_type_0::IntDiv 0 >> >> (there are similar stats for FP also) >> >> Since summing all the stats for iq.FU_type_0::Int* does not come even close >> to > >> iq.int_alu_accesses, I am guessing that iq..int_alu_accesses counts the NOPs >> and >> >> the iq.FU_type_0::* does not. Is this true? >> >> Something similar is happening with the committed instructions stats as >> well. >> The reported value (see bellow) is quite large, so again I am guessing that >> NOPs >> >> are included. >> >> system.switch_cpus0.commit.int_insts 50429347 >> >> Can anyone confirm or infirm my assumptions? Are there better suited >> counters >> for INT and FP instructions (total and committed)? >> _______________________________________________ >> gem5-users mailing list >> [email protected] >> http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users >> > > _______________________________________________ > gem5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > > _______________________________________________ > gem5-users mailing list > [email protected] > http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users > _______________________________________________ gem5-users mailing list [email protected] http://m5sim.org/cgi-bin/mailman/listinfo/gem5-users
