Am 15.02.2016 um 11:25 schrieb Mandy Martino:
> why
>
> I1 misses increase, LLi misses increase, LL misses increase, D1 misses
> increase
> though miss rate decrease at this row 0.1% + 0.0% ?
>
> which indicator show the correct number that can show the improvement
> after optimization?
I see you do blocking in the 2nd version. However, the number of data
accesses is 166 million vs. 10 million in your 1st version. I assume
this is because you did not compile with -O2 or -O3 ?
Miss rate is a relative number, based on total number of accesses.
A comparison is meaningless if the number of accesses is so different.
Josef
>
>
> #define min(a,b) (((a)<(b))?(a):(b))
> #define max(a,b) (((a)>(b))?(a):(b))
> int main()
> {
> int x[100][100];
> int y[100][100];
> int z[100][100];
> int i=0;
> int j=0;
> int k=0;
> int N=100;
> int r=0;
> int jj=0;
> int kk=0;
> int B = 5;
> /*
> for(i=0;i<N;++i)
> {
> for(j=0;j<N;++j)
> {
> r=0;
> for(k=0;k<N;++k)
> {
> r=r+y[i][k]*z[k][j];
> }
> x[i][j]=r;
> }
> }
> */
> for(jj=0;jj<N;jj=jj+B)
> for(kk=0;kk<N;kk=kk+B)
> for(i=0;i<N;++i)
> {
> for(j=0;j<min(jj+B,N);++j)
> {
> r=0;
> for(k=kk;k<min(kk+B,N);++k)
> {
> r=r+y[i][k]*z[k][j];
> }
> x[i][j]=x[i][j]+r;
> }
> }
> return 0;
> }
> /*
> for(i=0;i<N;++i)
> {
> for(j=0;j<N;++j)
> {
> r=0;
> for(k=0;k<N;++k)
> {
> r=r+y[i][k]*z[k][j];
> }
> x[i][j]=r;
> }
> }
>
> martin@ubuntu:~$ valgrind --tool=cachegrind ./mar5ti
> ==4602== Cachegrind, a cache and branch-prediction profiler
> ==4602== Copyright (C) 2002-2011, and GNU GPL'd, by Nicholas Nethercote
> et al.
> ==4602== Using Valgrind-3.7.0 and LibVEX; rerun with -h for copyright info
> ==4602== Command: ./mar5ti
> ==4602==
> --4602-- warning: L3 cache found, using its data for the LL simulation.
> ==4602==
> ==4602== I refs: 14,264,184
> ==4602== I1 misses: 689
> ==4602== LLi misses: 684
> ==4602== I1 miss rate: 0.00%
> ==4602== LLi miss rate: 0.00%
> ==4602==
> ==4602== D refs: 10,163,336 (10,117,945 rd + 45,391 wr)
> ==4602== D1 misses: 64,978 ( 64,200 rd + 778 wr)
> ==4602== LLd misses: 2,823 ( 2,063 rd + 760 wr)
> ==4602== D1 miss rate: 0.6% ( 0.6% + 1.7% )
> ==4602== LLd miss rate: 0.0% ( 0.0% + 1.6% )
> ==4602==
> ==4602== LL refs: 65,667 ( 64,889 rd + 778 wr)
> ==4602== LL misses: 3,507 ( 2,747 rd + 760 wr)
> ==4602== LL miss rate: 0.0% ( 0.0% + 1.6% )
>
> for(jj=0;jj<N;jj=jj+B)
> for(kk=0;kk<N;kk=kk+B)
> for(i=0;i<N;++i)
> {
> for(j=0;j<min(jj+B,N);++j)
> {
> r=0;
> for(k=kk;k<min(kk+B,N);++k)
> {
> r=r+y[i][k]*z[k][j];
> }
> x[i][j]=x[i][j]+r;
> }
> }
> martin@ubuntu:~$ valgrind --tool=cachegrind ./mar5ti
> ==4654== Cachegrind, a cache and branch-prediction profiler
> ==4654== Copyright (C) 2002-2011, and GNU GPL'd, by Nicholas Nethercote
> et al.
> ==4654== Using Valgrind-3.7.0 and LibVEX; rerun with -h for copyright info
> ==4654== Command: ./mar5ti
> ==4654==
> --4654-- warning: L3 cache found, using its data for the LL simulation.
> ==4654==
> ==4654== I refs: 265,277,487
> ==4654== I1 misses: 690
> ==4654== LLi misses: 685
> ==4654== I1 miss rate: 0.00%
> ==4654== LLi miss rate: 0.00%
> ==4654==
> ==4654== D refs: 166,275,677 (159,919,965 rd + 6,355,712 wr)
> ==4654== D1 misses: 170,231 ( 170,082 rd + 149 wr)
> ==4654== LLd misses: 2,823 ( 2,688 rd + 135 wr)
> ==4654== D1 miss rate: 0.1% ( 0.1% + 0.0% )
> ==4654== LLd miss rate: 0.0% ( 0.0% + 0.0% )
> ==4654==
> ==4654== LL refs: 170,921 ( 170,772 rd + 149 wr)
> ==4654== LL misses: 3,508 ( 3,373 rd + 135 wr)
> ==4654== LL miss rate: 0.0% ( 0.0% + 0.0% )
>
> */
>
>
> Regards,
>
>
> Martin
>
>
>
> ------------------------------------------------------------------------------
> Site24x7 APM Insight: Get Deep Visibility into Application Performance
> APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
> Monitor end-to-end web transactions and take corrective actions now
> Troubleshoot faster and improve end-user experience. Signup Now!
> http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140
>
>
>
> _______________________________________________
> Valgrind-users mailing list
> Valgrind-users@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/valgrind-users
>
------------------------------------------------------------------------------
Site24x7 APM Insight: Get Deep Visibility into Application Performance
APM + Mobile APM + RUM: Monitor 3 App instances at just $35/Month
Monitor end-to-end web transactions and take corrective actions now
Troubleshoot faster and improve end-user experience. Signup Now!
http://pubads.g.doubleclick.net/gampad/clk?id=272487151&iu=/4140
_______________________________________________
Valgrind-users mailing list
Valgrind-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/valgrind-users
