https://zenn.dev/tanakmura/articles/litex_linux_ae3feff0b48ede これで説明した make.py を vivado で実行
zen2
- 259.386455704 seconds time elapsed
- 11218.255071 [J]
alderlake P core
- 148.949631452 seconds time elapsed
- 4708.693756 [J]
alderlake E core
- 306.793298195 seconds time elapsed
- 3176.195971 [J]
alderlake が圧倒的にはやいし、消費電力も少ない。
htop で見てるかぎり、numactl で指定しないと、今のLinux では P-core, E-core はランダムに振られてるように見える。
# zen2
Performance counter stats for './make.py --build --board=arty':
430,014.53 msec task-clock # 1.658 CPUs utilized
2,060,997 context-switches # 4.793 K/sec
2,882 cpu-migrations # 6.702 /sec
5,255,087 page-faults # 12.221 K/sec
1,615,079,784,254 cycles # 3.756 GHz (83.47%)
172,481,625,622 stalled-cycles-frontend # 10.68% frontend cycles idle (83.43%)
397,100,079,468 stalled-cycles-backend # 24.59% backend cycles idle (83.47%)
2,007,296,528,712 instructions # 1.24 insn per cycle
# 0.20 stalled cycles per insn (83.43%)
423,176,696,430 branches # 984.099 M/sec (83.47%)
3,807,224,597 branch-misses # 0.90% of all branches (83.41%)
259.386455704 seconds time elapsed
406.331909000 seconds user
22.268063000 seconds sys
11218.255071 [J]
# alderlake
# numctl -C 0-15 あり
Performance counter stats for './make.py --build --board=arty':
231,008.79 msec task-clock # 1.551 CPUs utilized
1,943,789 context-switches # 8.414 K/sec
2,286 cpu-migrations # 9.896 /sec
5,195,518 page-faults # 22.491 K/sec
1,095,514,908,546 cycles # 4.742 GHz
1,910,343,647,050 instructions # 1.74 insn per cycle
403,393,095,844 branches # 1.746 G/sec
3,108,278,190 branch-misses # 0.77% of all branches
148.949631452 seconds time elapsed
220.634331000 seconds user
12.618163000 seconds sys
4708.693756 [J]
# alderlake
# numactl -C 16-19 あり
Performance counter stats for './make.py --build --board=arty':
441,542.57 msec task-clock # 1.439 CPUs utilized
1,121,146 context-switches # 2.539 K/sec
108,792 cpu-migrations # 246.391 /sec
5,192,256 page-faults # 11.759 K/sec
1,636,629,230,219 cycles # 3.707 GHz
1,896,456,531,800 instructions # 1.16 insn per cycle
400,549,399,290 branches # 907.159 M/sec
4,498,261,811 branch-misses # 1.12% of all branches
306.793298195 seconds time elapsed
428.333948000 seconds user
14.598149000 seconds sys
3176.195971 [J]