Changeset 806534c for doc/theses/mike_brooks_MMath/plots

Timestamp:

Apr 10, 2026, 5:15:34 PM (3 weeks ago)

Author:

Michael Brooks <mlbrooks@…>

Branches:

master

Children:

d1ccc57

Parents:

17f2a7f4

Message:

refactor list perf analysis with histogram view reusable

Location:

doc/theses/mike_brooks_MMath/plots

Files:

• ListCommon.py (modified) (2 diffs)
• list-mchn-szz.py (modified) (2 diffs)

doc/theses/mike_brooks_MMath/plots/ListCommon.py

@@ -1 +1 @@
 import pandas as pd
 import numpy as np
-import sys
+import math
 import os
 from subprocess import Popen, PIPE
@@ -358 +358 @@
 def gMeanNoOutlr(range):
     return ( range.prod() / range.min() / range.max() ) ** ( 1 / ( range.count() - 2 ) )
+
+
+# The range from 0.9759 to 1.0247 (which is 1.05 x wide) has 1.0 in its centre.
+# This is the bucket with key 0.
+# Logs of values in this bucket go from -0.5 to +0.5.
+# Rounding a log value to the nearest integer gives the key.
+# Exponentiating a key directly gives the centre of its bucket.
+# Exponentiating a key less 0.5 gives the bottom of its bucket.
+# Gnuplot expects the latter.
+
+bucketMin = 0.25
+bucketMax = 4.0
+bucketGrain = 1.05
+bktKeyLo = math.floor( math.log(bucketMin, bucketGrain) )
+bktKeyHi = math.ceil( math.log(bucketMax, bucketGrain) )
+
+def bktKeyOfVal( relDur ):
+    distance = math.log(relDur, bucketGrain)
+    key = round( distance )
+    return key
+
+def bktIxOfVal( relDur ):
+    return bktKeyToIx( bktKeyOfVal( relDur ) )
+
+def botValOfBucketK( key ):
+    return bucketGrain ** ( key - 0.5 )
+
+def topValOfBucketBotVal( botVal ):
+    return bucketGrain * botVal
+
+def bktKeyToIx( key ):
+    return key - bktKeyLo
+
+def bktIxToKey( ix ):
+    return ix + bktKeyLo
+
+def botOfBucketOfVal( relDur ):
+    return botValOfBucketK( bktKeyOfVal( relDur ) )
+
+buckets = [ botValOfBucketK(key) for key in range(bktKeyLo, bktKeyHi) ]
+
+# printSingleDetail
+def printHistos(*,
+    tgtMovement = 'all',
+    tgtPolarity = 'all',
+    tgtAccessor = 'all',
+    tgtInterleave = 0.0,
+    marginalizeOn=['fx'] ):
+
+    # watch out for filtering too early here; need everything sticking around until baselines are applies
+    # ie, maybe I should get rid of all the tgt parms at the pre-benchmark layers
+    timings = getSingleResults(
+        tgtMovement = tgtMovement,
+        tgtPolarity = tgtPolarity,
+        tgtAccessor = tgtAccessor,
+        tgtInterleave = tgtInterleave)
+    timings = getJustCanon( timings,
+                  fxInc = ['cfa-cfa', 'lq-tailq', 'upp-upp', 'lq-list'],
+                  szInc = ['SM', 'ML'],
+                  sExcl = [1] )
+
+
+#    annotateBaselines(timings)
+
+
+    options = timings.groupby(explanations)
+
+    aggregated = options.agg(
+        mean_op_dur_ns = ('mean_op_dur_ns', gMeanNoOutlr)
+    ).reset_index()
+
+
+    annotateBaseline(aggregated, marginalizeOn)
+#    annotateCommonBaselines(aggregated)
+
+
+    # if examining "why CFA slow" need both
+    # - getVariousCfa inplace of getJust Canon
+    # - do annotate-then-filter because baseline needs to stay cfa-tailq-upp
+    # (filter-then-annotate is fine for general cases (where all three canons are included) and good for build time)
+
+
+    c_measure = c('OpDurRel', marginalizeOn)
+    # options = timings.groupby(explanations)
+
+    # aggregated = options.agg(
+    #     **{measure:(measure,gMeanNoOutlr)}
+    # ).reset_index()
+
+    c_measureBkt = 'BUCKET_' + c_measure
+    aggregated[ c_measureBkt ] = aggregated[c_measure].apply( botOfBucketOfVal )
+
+    marggrp = aggregated.groupby(marginalizeOn)
+
+
+    # print(f'measure is {measure}')
+    # print()
+    # print()
+
+    for mkey, mgroup in marggrp:
+#       print(mgroup, file=sys.stderr)
+
+        histo_raw = mgroup[ c_measureBkt ].value_counts()
+        for b in buckets:
+            if b not in histo_raw.keys():
+#                print( f"{b} := 0", file=sys.stderr )
+                histo_raw[b] = 0
+        histo_raw = histo_raw.sort_index()
+
+        histo = histo_raw.rename("count").reset_index()
+        histo = histo.rename(columns={c_measureBkt: "y_lo"})
+        y_lo_col_loc = histo.columns.get_loc("y_lo")
+        histo.insert(y_lo_col_loc + 1, "y_hi", histo["y_lo"].apply(topValOfBucketBotVal))
+
+        header = str.join(', ', mkey)
+        print(f'"{header}"')
+        text = histo.to_csv(header=False, index=False, sep='\t')
+        print(text)
+        print()
+        print()
+
+        # print(f'"{header}" FULL')
+        # text = group.to_csv(header=False, index=True, sep='\t')
+        # print(text)
+        # print()
+        # print()
+
+    # print(f'"RAW"')
+    # text = timings.to_csv(header=False, index=True, sep='\t')
+    # print(text)

doc/theses/mike_brooks_MMath/plots/list-mchn-szz.py

@@ -3 +3 @@
 import os
 import sys
-import math
 
 sys.path.insert(0, os.path.dirname(__file__))
 from ListCommon import *
-
-# The range from 0.9759 to 1.0247 (which is 1.05 x wide) has 1.0 in its centre.
-# This is the bucket with key 0.
-# Logs of values in this bucket go from -0.5 to +0.5.
-# Rounding a log value to the nearest integer gives the key.
-# Exponentiating a key directly gives the centre of its bucket.
-# Exponentiating a key less 0.5 gives the bottom of its bucket.
-# Gnuplot expects the latter.
-
-bucketMin = 0.25
-bucketMax = 4.0
-bucketGrain = 1.05
-bktKeyLo = math.floor( math.log(bucketMin, bucketGrain) )
-bktKeyHi = math.ceil( math.log(bucketMax, bucketGrain) )
-
-def bktKeyOfVal( relDur ):
-    distance = math.log(relDur, bucketGrain)
-    key = round( distance )
-    return key
-
-def bktIxOfVal( relDur ):
-    return bktKeyToIx( bktKeyOfVal( relDur ) )
-
-def botValOfBucketK( key ):
-    return bucketGrain ** ( key - 0.5 )
-
-def topValOfBucketBotVal( botVal ):
-    return bucketGrain * botVal
-
-def bktKeyToIx( key ):
-    return key - bktKeyLo
-
-def bktIxToKey( ix ):
-    return ix + bktKeyLo
-
-def botOfBucketOfVal( relDur ):
-    return botValOfBucketK( bktKeyOfVal( relDur ) )
-
-buckets = [ botValOfBucketK(key) for key in range(bktKeyLo, bktKeyHi) ]
-
-# printSingleDetail
-def printHistos(*,
-    tgtMovement = 'all',
-    tgtPolarity = 'all',
-    tgtAccessor = 'all',
-    tgtInterleave = 0.0,
-    marginalizeOn=['fx'] ):
-
-    # watch out for filtering too early here; need everything sticking around until baselines are applies
-    # ie, maybe I should get rid of all the tgt parms at the pre-benchmark layers
-    timings = getSingleResults(
-        tgtMovement = tgtMovement,
-        tgtPolarity = tgtPolarity,
-        tgtAccessor = tgtAccessor,
-        tgtInterleave = tgtInterleave)
-    timings = getJustCanon( timings,
-                  fxInc = ['cfa-cfa', 'lq-tailq', 'upp-upp', 'lq-list'],
-                  szInc = ['SM', 'ML'],
-                  sExcl = [1] )
-
-
-#    annotateBaselines(timings)
-
-
-    options = timings.groupby(explanations)
-
-    aggregated = options.agg(
-        mean_op_dur_ns = ('mean_op_dur_ns', gMeanNoOutlr)
-    ).reset_index()
-
-
-    annotateBaseline(aggregated, marginalizeOn)
-#    annotateCommonBaselines(aggregated)
-
-
-    # if examining "why CFA slow" need both
-    # - getVariousCfa inplace of getJust Canon
-    # - do annotate-then-filter because baseline needs to stay cfa-tailq-upp
-    # (filter-then-annotate is fine for general cases (where all three canons are included) and good for build time)
-
-
-    c_measure = c('OpDurRel', marginalizeOn)
-    # options = timings.groupby(explanations)
-
-    # aggregated = options.agg(
-    #     **{measure:(measure,gMeanNoOutlr)}
-    # ).reset_index()
-
-    c_measureBkt = 'BUCKET_' + c_measure
-    aggregated[ c_measureBkt ] = aggregated[c_measure].apply( botOfBucketOfVal )
-
-    marggrp = aggregated.groupby(marginalizeOn)
-
-
-    # print(f'measure is {measure}')
-    # print()
-    # print()
-
-    for mkey, mgroup in marggrp:
-#       print(mgroup, file=sys.stderr)
-
-        histo_raw = mgroup[ c_measureBkt ].value_counts()
-        for b in buckets:
-            if b not in histo_raw.keys():
-#                print( f"{b} := 0", file=sys.stderr )
-                histo_raw[b] = 0
-        histo_raw = histo_raw.sort_index()
-
-        histo = histo_raw.rename("count").reset_index()
-        histo = histo.rename(columns={c_measureBkt: "y_lo"})
-        y_lo_col_loc = histo.columns.get_loc("y_lo")
-        histo.insert(y_lo_col_loc + 1, "y_hi", histo["y_lo"].apply(topValOfBucketBotVal))
-
-        header = str.join(', ', mkey)
-        print(f'"{header}"')
-        text = histo.to_csv(header=False, index=False, sep='\t')
-        print(text)
-        print()
-        print()
-
-        # print(f'"{header}" FULL')
-        # text = group.to_csv(header=False, index=True, sep='\t')
-        # print(text)
-        # print()
-        # print()
-
-    # print(f'"RAW"')
-    # text = timings.to_csv(header=False, index=True, sep='\t')
-    # print(text)
-    
 
 printHistos(
@@ -143 +12 @@
     tgtAccessor = 'all',
     marginalizeOn=['machine', 'SizeZone'] )
-