def regenerate_st_stat(pairs: list, cache: str) -> None:
st_targets = [
'system\.cpu\.(ipc::0)',
]
matrix = {}
for p in pairs:
matrix[p] = c.get_stats(pjoin(st_stat_dir, p, 'stats.txt'),
st_targets, re_targets=True)
df = pd.DataFrame(matrix)
df.to_csv(cache, index=True)
df = pd.read_csv(cache, index_col=0)
print(df)
def main():
parser = argparse.ArgumentParser(usage='specify stat directory')
parser.add_argument('-s', '--stat-dir', action='store', required=True,
help='gem5 output directory'
)
parser.add_argument('-o', '--output', action='store',
help='csv to save results'
)
parser.add_argument('--branch', action='store_true')
parser.add_argument('-v', '--verbose', action='store_true',
help='whether output intermediate result'
)
parser.add_argument('-b', '--error-bound', action='store', type=float,
default=0.0,
help='Threshold to output an entry'
)
parser.add_argument('-i', '--ipc-only', action='store_true',
default=0.0,
help='Only extract ipc'
)
parser.add_argument('--st', action='store_true',
help='processing ST stats'
)
parser.add_argument('--pair-filter', action='store', default='',
help='file that filt pairs'
)
parser.add_argument('-f', '--stat-file', action='store',
help='name of stats file', default='stats.txt'
)
parser.add_argument('-l', '--fanout', action='store_true',
help='print fanout'
)
parser.add_argument('-k', '--breakdown', action='store_true',
help='print breakdown'
)
parser.add_argument('--op', action='store_true',
help='print operand busy state'
)
parser.add_argument('--flow', action='store_true',
help='print bandwidth usages'
)
parser.add_argument('-p', '--packet', action='store_true',
help='print type and number of different packets'
)
opt = parser.parse_args()
pairs = c.pairs(opt.stat_dir, return_path=False)
if (opt.pair_filter):
pairs = c.pair_filt(pairs, opt.pair_filter)
paths = c.pair_to_full_path(opt.stat_dir, pairs)
pairs, paths = c.stat_filt(pairs, paths, opt.stat_file)
# paths = c.time_filt(paths)
paths = [pjoin(x, opt.stat_file) for x in paths]
# make_st_stat_cache()
matrix = {}
for pair, path in zip(pairs, paths):
# print(pair)
if opt.ipc_only:
d = c.get_stats(path, ipc_target, re_targets=True)
else:
targets = brief_targets
if opt.branch:
targets += branch_targets
if opt.fanout:
targets += fanout_targets
if opt.breakdown:
targets += breakdown_targets
if opt.op:
targets += operand_targets
if opt.packet:
targets += packet_targets
if opt.flow:
targets += flow_target
d = c.get_stats(path, targets, re_targets=True)
if len(d):
if not opt.st:
matrix[pair] = further_proc(pair, d, opt.verbose)
else:
matrix[pair] = d
if opt.branch:
c.add_branch_mispred(d)
if opt.fanout:
c.add_fanout(d)
if opt.packet:
c.add_packet(d)
df = pd.DataFrame.from_dict(matrix, orient='index')
df = df.sort_index()
df = df.sort_index(1)
df = df.sort_values(['ipc'])
# for x in df.index:
# print(x)
if len(df):
df.loc['mean'] = df.mean()
df = df.fillna(0)
print(df)
if opt.output:
df.to_csv(opt.output, index=True)
width = 0.9
interval = 0.1
rects = []
shift = 0.0
i = 0
num_points = 0
num_configs = len(stat_dirs)
stat_files = [
osp.join(baseline_stat_dir, point, 'stats.txt') for point in points
]
matrix = {}
for point, stat_file in zip(points, stat_files):
d = c.get_stats(stat_file, t.breakdown_targets, re_targets=True)
matrix[point] = d
baseline_df = pd.DataFrame.from_dict(matrix, orient='index')
baseline = baseline_df['queueingD'].values
for config in configs_ordered:
stat_dir = stat_dirs[config]
stat_dir = osp.expanduser(stat_dir)
stat_files = [osp.join(stat_dir, point, 'stats.txt') for point in points]
matrix = {}
for point, stat_file in zip(points, stat_files):
d = c.get_stats(stat_file, t.breakdown_targets, re_targets=True)
matrix[point] = d
df = pd.DataFrame.from_dict(matrix, orient='index')
if num_points == 0:
import common as c
import target_stats as t
stat_dir = osp.expanduser('~/gem5-results-2017/xbar-rand-hint')
benchmarks = [*c.get_spec2017_int(), *c.get_spec2017_fp()]
points = []
for b in benchmarks:
for i in range(0, 3):
points.append(f'{b}_{i}')
stat_files = [osp.join(stat_dir, point, 'stats.txt') for point in points]
matrix = {}
for point, stat_file in zip(points, stat_files):
d = c.get_stats(stat_file, t.packet_targets, re_targets=True)
matrix[point] = d
df = pd.DataFrame.from_dict(matrix, orient='index')
for i in df.columns.values:
if i != 'TotalP':
df[i] = df[i] / df['TotalP']
df.drop('TotalP', axis=1, inplace=True)
names = df.columns.values.tolist()
names.remove('DestOpP')
names.append('DestOpP')
names.remove('KeySrcP')
names.append('KeySrcP')
print(type(names))
print(names)
df = df.reindex(columns=names)
def main():
parser = argparse.ArgumentParser(usage='specify stat directory')
parser.add_argument('-s',
'--stat-dir',
action='store',
required=True,
help='gem5 output directory')
parser.add_argument('-o',
'--output',
action='store',
help='csv to save results')
parser.add_argument('-v',
'--verbose',
action='store_true',
help='whether output intermediate result')
parser.add_argument('-b',
'--error-bound',
action='store',
type=float,
default=0.0,
help='Threshold to output an entry')
parser.add_argument('-i',
'--ipc-only',
action='store_true',
default=0.0,
help='Only extract ipc')
parser.add_argument('--st',
action='store_true',
help='processing ST stats')
parser.add_argument('--pair-filter',
action='store',
default='',
help='file than filt pairs')
opt = parser.parse_args()
pairs = c.pairs(opt.stat_dir, return_path=False)
if (opt.pair_filter):
pairs = c.pair_filt(pairs, opt.pair_filter)
paths = c.pair_to_full_path(opt.stat_dir, pairs)
pairs, paths = c.stat_filt(pairs, paths)
# paths = c.time_filt(paths)
paths = [pjoin(x, 'stats.txt') for x in paths]
# make_st_stat_cache()
matrix = {}
for pair, path in zip(pairs, paths):
if opt.ipc_only:
d = c.get_stats(path, ipc_target, re_targets=True)
else:
#d = c.get_stats(path, branch_targets, re_targets=True)
d = c.get_stats(path, branch_targets, re_targets=True)
if len(d):
if not opt.st:
matrix[pair] = further_proc(pair, d, opt.verbose)
else:
matrix[pair] = d
df = pd.DataFrame.from_dict(matrix, orient='index')
#if df['iew.branchMispredict'] != None:
# df['mpki'] = df['iew.branchMispredict'] / df['Insts'] * 1000
# df['Mispredict Rate'] = df['branchPred.condIncorrect'] / df['branchPred.condPredicted']
pd.set_option('display.max_columns', None)
#print(df.sort_index(1))
try:
df['Misp%'] = df['Misp'] / df['cond'] * 100
df['MPKI'] = df['Misp'] / df['Insts'] * 1000
df['Status'] = 'running'
df.loc[df['Insts'] > 30000000, ['Status']] = 'done'
print(df[['Status', 'ipc', 'indirectMis', 'Misp', 'Misp%', 'MPKI']])
hmean_mpki = len(df['MPKI']) / np.sum(1.0 / df['MPKI'])
hmean_misp = len(df['Misp%']) / np.sum(1.0 / df['Misp%'])
gmean_mpki = df['MPKI'].prod()**(1.0 / len(df['MPKI']))
gmean_misp = df['Misp%'].prod()**(1.0 / len(df['Misp%']))
print('Totally %d tests' % len(df['Status']))
print(
'Mean MPKI is: %.2f, Hmean MPKI is: %.4f, Gmean MPKI is: %.4f' %
(df['MPKI'].mean().round(2), hmean_mpki, gmean_mpki))
print(
'Mean Misp%% is: %.2f, Hmean Misp%% is: %.4f, Gmean Misp%% is: %.4f'
% (df['Misp%'].mean().round(2), hmean_misp, gmean_misp))
print('Mean IPC is: %.4f' % df['ipc'].mean())
df.drop(['Status'], axis=1, inplace=True)
except Exception:
print("No data yet")
exit()
if not opt.st:
errors = df['IPC prediction error'].values
print('Mean: {}'.format(np.mean(np.abs(errors))))
# df.sort_values(['QoS prediction error'], ascending=False, inplace=True)
print(df['overall QoS'][
abs(df['IPC prediction error']) > opt.error_bound])
if opt.output:
df.to_csv(opt.output, index=True)
shift = 0.0
i = 0
num_points = 0
bounds = ['by_bw', 'by_chasing', 'by_crit_ptr']
num_configs = len(bounds)
for bound in bounds:
stat_dir = stat_dirs['Xbar4']
stat_dir = osp.expanduser(stat_dir)
stat_files = [osp.join(stat_dir, point, 'stats.txt') for point in points]
matrix = {}
for point, stat_file in zip(points, stat_files):
d = c.get_stats(stat_file,
t.standard_targets + t.packet_targets,
re_targets=True)
c.add_packet(d)
matrix[point] = d
df = pd.DataFrame.from_dict(matrix, orient='index')
if num_points == 0:
num_points = len(df)
# print(len(df))
tick_starts = np.arange(0, num_points * num_configs,
(width + interval) * num_configs) + shift
# print(tick_starts)
rect = plt.bar(tick_starts,
df[bound].values,
edgecolor=colors[i],
interval = 0.4
rects = []
num_points = 0
num_configs = len(stat_dirs)
dfs = dict()
for config in configs_ordered:
print(config)
stat_dir = stat_dirs[config]
stat_dir = osp.expanduser(stat_dir)
stat_files = [osp.join(stat_dir, point, 'stats.txt') for point in points]
matrix = {}
for point, stat_file in zip(points, stat_files):
d = c.get_stats(stat_file, t.ipc_target, re_targets=True)
matrix[point] = d
df = pd.DataFrame.from_dict(matrix, orient='index')
dfs[config] = df
if num_points == 0:
num_points = len(df)
baseline = 'Xbar4'
dfs['Xbar4'].loc['rel_geo_mean'] = [1.0]
print('Xbar4')
print(dfs['Xbar4'])
for config in configs_ordered:
if config != 'Xbar4':
df = None
count = 0
for point in points:
shift = 0.0
sub_ax = axs[count // n_cols][count % n_cols]
plt.axes(sub_ax)
for j, config in enumerate(configs_ordered):
print(config)
stat_dir = stat_dirs[config]
stat_dir = osp.expanduser(stat_dir)
stat_file = osp.join(stat_dir, point, 'stats.txt')
print(stat_dir)
matrix = {}
d = c.get_stats(stat_file, t.flow_target, re_targets=True)
print(d)
d.pop('WKFlowUsage::total')
d.pop('WKFlowUsage::0')
# cycle to packets
for k in d:
n = int(k.split('::')[1]) * d[k]
d[k] = n
matrix[point] = d
df = pd.DataFrame.from_dict(matrix, orient='index')
if num_x == 0:
num_x = len(df.columns)
# print(len(df))