Add simple tool for transforming sched. results

1 files changed, 165 insertions, 0 deletions

diff --git a/transform.py b/transform.py
new file mode 100755
index 0000000..65d34d4
--- /dev/null
+++ b/transform.py
@@ -0,0 +1,165 @@
+#!/usr/bin/env python
+
+import sys
+
+from plot import decode
+from os.path  import splitext, basename
+
+from glob import glob
+from collections import defaultdict
+import optparse
+
+from util import load_csv_file, write_csv_file
+
+o = optparse.make_option
+
+opts = [
+#    o('-r', '--result-type', action='store', dest='',
+#      type='choice', choices=['hard', 'soft', 'tardiness', 'rel-tardiness'],
+#      help='what data should be emitted?'),
+    ]
+
+defaults = {
+    }
+
+options = None
+
+#G-EDF/testpoint_ucap=13.75_wss=1792_dist=exp-10-10-100_deadlines=implicit_host=ludwig_scheduler=G-EDF.csv
+
+IDX_WSS  = 4
+IDX_UCAP = 5
+
+IDX_HARD_IDLE =  6
+IDX_HARD_LOAD =  7
+
+IDX_SOFT_IDLE =  8
+IDX_SOFT_LOAD = 17
+
+IDX_TARD_MAX_IDLE =  9
+IDX_TARD_AVG_IDLE = 10
+IDX_TARD_STD_IDLE = 12
+
+IDX_TARD_MAX_LOAD = 18
+IDX_TARD_AVG_LOAD = 19
+IDX_TARD_STD_LOAD = 21
+
+IDX_RTARD_MAX_IDLE = 13
+IDX_RTARD_AVG_IDLE = 14
+IDX_RTARD_STD_IDLE = 16
+
+IDX_RTARD_MAX_LOAD = 22
+IDX_RTARD_AVG_LOAD = 23
+IDX_RTARD_STD_LOAD = 25
+
+
+def key(fname):
+    name, ext = splitext(basename(fname))
+    conf = decode(name)
+
+    sched = conf['scheduler']
+
+    if 'quanta' in conf and conf['quanta'] == 'staggered':
+        sched = 'S-' + sched
+
+    d  = 'dist=%s'  % conf['dist']
+    dl = 'deadlines=%s' % conf['deadlines']
+    h  = 'host=%s'  % conf['host']
+
+    return ('_'.join([d, dl, h]), sched, fname)
+
+def get_row(i, key_col,  arrays, cols):
+    row  = []
+    data = arrays[0]
+    keys = [data[i,x] for x in key_col]
+    for idx, data in enumerate(arrays):
+        # make sure we are not combining apples and oranges...
+        d_keys = [data[i,x] for x in key_col]
+        if not d_keys == keys:
+            print 'Bad: missing data sched=%d row=%d %s!=%s' % (idx + 1, i+1, d_keys, keys)
+            assert False
+        row += [data[i,x] for x in cols]
+    return keys + row
+
+def get_table(arrays, key_col, data_col):
+    data = arrays[0]
+    return [get_row(i, key_col, arrays, data_col) for i in xrange(len(data))]
+
+def write_sched(name, components, hard=True):
+    header = ['WSS', 'ucap']
+    arrays = []
+    for (sched, data) in components:
+#        print sched
+        header += ["%s (load)" % sched , "%s (idle)" % sched]
+        arrays.append(data)
+    if hard:
+        data_col = [IDX_HARD_LOAD, IDX_HARD_IDLE]
+    else:
+        data_col = [IDX_SOFT_LOAD, IDX_SOFT_IDLE]
+    result = get_table(arrays, [IDX_WSS, IDX_UCAP], data_col)
+    # sort by WSS, then by ucap
+    result.sort(key=lambda row: (row[0], row[1]))
+    fname = '%s_%s.csv' % ('hard' if hard else 'soft',
+                           name)
+    write_csv_file(fname, result, header=header, width=20)
+
+def write_tardiness(name, components, relative=True):
+    header = ['WSS', 'ucap']
+    arrays = []
+    for (sched, data) in components:
+        header += ["%s (max, load)" % sched,
+                   "%s (avg, load)" % sched,
+                   "%s (std, load)" % sched,
+                   "%s (max, idle)" % sched,
+                   "%s (avg, idle)" % sched,
+                   "%s (std, idle)" % sched]
+        arrays.append(data)
+
+    if relative:
+        data_col = [IDX_RTARD_MAX_LOAD, IDX_RTARD_AVG_LOAD, IDX_RTARD_STD_LOAD,
+                    IDX_RTARD_MAX_IDLE, IDX_RTARD_MAX_IDLE, IDX_RTARD_STD_IDLE]
+    else:
+        data_col = [IDX_TARD_MAX_LOAD, IDX_TARD_AVG_LOAD, IDX_TARD_STD_LOAD,
+                    IDX_TARD_MAX_IDLE, IDX_TARD_MAX_IDLE, IDX_TARD_STD_IDLE]
+
+    result = get_table(arrays, [IDX_WSS, IDX_UCAP], data_col)
+    # sort by WSS, then by ucap
+    result.sort(key=lambda row: (row[0], row[1]))
+    fname = '%s_%s.csv' % ('rel-tard' if relative else 'abs-tard',
+                           name)
+    write_csv_file(fname, result, header=header, width=35)
+
+
+def assemble_results(dir):
+    files = glob(dir + '/*.csv')
+
+    parts = defaultdict(list)
+
+    print 'Organizing %d files...' % len(files)
+    for f in files:
+        k, sched, fname = key(f)
+        parts[k].append((sched, fname))
+
+    for k in parts:
+        # sort by scheduler name
+        parts[k].sort()
+
+    for i, k in enumerate(parts):
+        comment = 1
+        print '[%d/%d] Processing %s' % (i+ 1, len(parts), k)
+        print 'Loading files.'
+        components = [(sched, load_csv_file(fname)) for (sched, fname) in parts[k]]
+
+        print 'Generating output.'
+        write_sched(k, components, hard=True)
+        write_sched(k, components, hard=False)
+        write_tardiness(k, components, relative=True)
+        write_tardiness(k, components, relative=False)
+
+if __name__ == '__main__':
+    parser = optparse.OptionParser(option_list=opts)
+    parser.set_defaults(**defaults)
+    (options, dirs) = parser.parse_args()
+
+    for d in dirs:
+        assemble_results(d)
+