β版の XL fortran for MacOS Xと fink経由でインストールした g77で姫野ベンチを実行してみました,
なお,コンパイルオプションは, ここを参考にしました.
xlf -qthreaded -O5 -qtune=auto -qarch=auto -qunroll=auto -qextname=dtime himenoBMTxp_s.fで, a.outを実行してみると,
mimax= 129 mjmax= 65 mkmax= 65 imax= 128 jmax= 64 kmax= 64 Start rehearsal measurement process. Measure the performance in 3 times. MFLOPS: 133.5218658 time(s): 0.3700000048 0.3292582231E-02 Now, start the actual measurement process. The loop will be excuted in 486 times. This will take about one minute. Wait for a while. Loop executed for 486 times Gosa : 0.9902104503E-03 MFLOPS: 133.0557098 time(s): 60.14999771 Score based on Pentium III 600MHz : 1.606177092でした.
g77 -O5 -funroll-loops ./himenoBMTxp_s.f
で a.outを実行すると,
mimax= 129 mjmax= 65 mkmax= 65
imax= 128 jmax= 64 kmax= 64
Start rehearsal measurement process.
Measure the performance in 3 times.
MFLOPS: 112.279747 time(s): 0.439999998 0.00328862783
Now, start the actual measurement process.
The loop will be excuted in 409 times.
This will take about one minute.
Wait for a while.
Loop executed for 409 times
Gosa : 0.00112893619
MFLOPS: 118.100784 time(s): 57.0299988
Score based on Pentium III 600MHz : 1.42564929
xlfには劣るものの,思ったよりは優秀でした.というか, xlfが Altivec効いていないのかな???
mimax= 129 mjmax= 65 mkmax= 65
imax= 128 jmax= 64 kmax= 64
Start rehearsal measurement process.
Measure the performance in 3 times.
MFLOPS: 411.6924 time(s): 0.1200000 3.2946540E-03
Now, start the actual measurement process.
The loop will be excuted in 1500 times.
This will take about one minute.
Wait for a while.
Loop executed for 1500 times
Gosa : 2.0512370E-04
MFLOPS: 414.8034 time(s): 59.55000
Score based on Pentium III 600MHz : 5.007285
がっくし....やっぱ G5じゃないと勝てないか....

このホームページに関するお問い合わせは
へ戻る
宇都宮大学のホームページ