1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
|
- It is intended that benchmark_plot.sh be used to drive benchmark.py and plot.py
- There are three modes for use of the script
---------------------------------------------------------------------
--- plot performance for a single architecture with commands like ---
---------------------------------------------------------------------
# to run benchmarks in blas1/*.yaml files and output results in blas1/my_arch/*.csv files
./benchmark_plot.sh --benchmark true --plot false --level1 true --level2 false --level3 false --tag1 my_arch
# to plot data in blas1/my_arch/*.csv files and output in blas1/my_arch/plot_gflops/*.png
./benchmark_plot.sh --benchmark false --plot true --level1 true --level2 false --level3 false --tag1 my_arch
---------------------------------------------------------------------------------------------------------------
--- plot performance relative to theoretical maximum performance for memory bound BLAS1 and BLAS2 functions ---
--- for a single architecture with commands like ---
---------------------------------------------------------------------------------------------------------------
Note that the memory bandwidth in the parameter GBps needs to be defined for the architecture for which
you are running in the file benchmark.py. It is defined for the architectures below:
if machine_spec_dict['arch'] == 'gfx906':
GBps = 1000
elif machine_spec_dict['arch'] == 'gfx908':
GBps = 1100
elif machine_spec_dict['arch'] == 'gfx90a':
GBps = 1600
else:
print("do not know GBps memory bandwidth for ", machine_spec_dict['arch'])
print("add GBps to", sys.argv[0])
print("quitting ", sys.argv[0])
quit(
# to run benchmarks, no need to run it a second time if you have run it above
./benchmark_plot.sh --benchmark true --plot false --level1 true --level2 false --level3 false --tag1 my_arch
# To plot data in blas1/my_arch/*.csv and blas1/my_arch/machine_spec.yaml files and output in blas1/my_arch/plot_vs_theo_max/*.png
./benchmark_plot.sh --benchmark false --plot true --level1 true --level2 false --level3 false --tag1 my_arch --theo_max true
------------------------------------------------------------
--- plot performance of one architecture against another ---
------------------------------------------------------------
# to run benchmarks in blas1/*.yaml files and output results in blas1/my_arch1/*.csv files
./benchmark_plot.sh --benchmark true --plot false --level1 true --level2 false --level3 false --tag1 my_arch1 --perf_vs_perf true
# to run benchmarks in blas1/*.yaml files and output results in blas1/my_arch2/*.csv files
./benchmark_plot.sh --benchmark true --plot false --level1 true --level2 false --level3 false --tag1 my_arch2 --perf_vs_perf true
# to plot performance of data in blas1/my_arch1/*.csv files relative to blas1/my_arch2/*.csv files and output
# plots in blas1/my_arch1_my_arch2/plots_perf_vs_perf/*.png
./benchmark_plot.sh --benchmark false --plot true --level1 true --level2 false --level3 false --tag1 my_arch1 --tag2 my_arch2 --perf_vs_perf true
|
