Changeset e0350e0 for doc/theses/mike_brooks_MMath/plots/string-allocn-attrib.py

Timestamp:

Jun 4, 2025, 1:43:34 PM (4 months ago)

Author:

Michael Brooks <mlbrooks@…>

Branches:

master

Children:

f858ca5

Parents:

c8bdbaf

Message:

Recent rework of string benchmarks

File:

• doc/theses/mike_brooks_MMath/plots/string-allocn-attrib.py (modified) (3 diffs)

doc/theses/mike_brooks_MMath/plots/string-allocn-attrib.py

@@ -14 +14 @@
 import os
 import sys
+from pathlib import Path
 
 sys.path.insert(0, os.path.dirname(__file__))
@@ -19 +20 @@
 
 sizes_i_want = [50, 200] # [20, 50, 100, 200]
+defaultExpansionCfa = 0.2
 
-# assume CFA threshold only run at default value
+metaFilename = os.path.dirname(os.path.abspath(__file__)) + '/string-allocn-attrib-meta.dat'
+catOrder = Path(metaFilename).read_text()
+catOrder = str.split(catOrder)
 
 cfatimings = loadParseTimingData('result-allocate-speed-cfa.csv',
                 xClasNames=['expansion'], xClasDtypes={'expansion':'Float64'},
                 xFactNames=['topIters'], xFactDtypes={'topIters':np.int64})
+cfatimings = cfatimings.groupby('expansion').get_group(defaultExpansionCfa)
 
 cfaattribs = loadParseAttribData('result-allocate-attrib-cfa.ssv')
+cfaattribs = cfaattribs.groupby('expansion').get_group(defaultExpansionCfa)
+
+cfasizings = loadParseSizingData('result-allocate-space-cfa.ssv', xClasNames=['expansion'], xClasDtypes={'expansion':'Float64'})
+cfasizings = cfasizings.groupby('expansion').get_group(defaultExpansionCfa)
 
 stltimings = loadParseTimingData('result-allocate-speed-stl.csv',
@@ -34 +43 @@
 stlattribs = loadParseAttribData('result-allocate-attrib-stl.ssv')
 
+stlsizings = loadParseSizingData('result-allocate-space-stl.ssv', xClasNames=['expansion'], xClasDtypes={'expansion':'Float64'})
+
 timings = pd.concat([cfatimings, stltimings])
 attribs = pd.concat([cfaattribs, stlattribs])
+sizings = pd.concat([cfasizings, stlsizings])
+
+# print("before join", timings.shape[0], attribs.shape[0])
+# print(timings.to_csv(header=True, index=True, sep='\t', na_rep="0"))
+# print(attribs.to_csv(header=True, index=True, sep='\t', na_rep="0"))
+# print(sizings.to_csv(header=True, index=True, sep='\t', na_rep="0"))
 
 combined = pd.merge(
-    left=timings[['sut-platform', 'corpus-meanlen','expansion', 'op-duration-ns']],
-    right=attribs[['sut-platform', 'corpus-meanlen','expansion', 'category', 'fraction']],
-    on=['sut-platform', 'corpus-meanlen','expansion']
+    left=timings[['sut-platform', 'corpus', 'corpus-meanlen-tgt','expansion', 'op-duration-ns']],
+    right=attribs[['sut-platform', 'corpus', 'expansion', 'category', 'fraction']],
+    on=['sut-platform', 'corpus','expansion']
 )
 
+combined = pd.merge(
+    left=combined,
+    right=sizings[['sut-platform', 'corpus','expansion','hw_cur_req_mem(B)']],
+    on=['sut-platform', 'corpus','expansion']
+)
+
+# print("after join", combined.shape[0])
+# print(combined.to_csv(header=True, index=True, sep='\t', na_rep="0"))
+
 combined['cat-duration-ns'] = combined['op-duration-ns'] * combined['fraction']
-combined.drop(columns=['expansion', 'op-duration-ns', 'fraction'], inplace=True)
+combined.drop(columns=['expansion', 'op-duration-ns', 'fraction', 'corpus'], inplace=True)
 
-pvt = combined.pivot( columns='category', values='cat-duration-ns', index=['corpus-meanlen', 'sut-platform'] )
+# print("before summarize", combined.shape[0])
+# print(combined.to_csv(header=True, index=True, sep='\t', na_rep="0"))
 
-desired_dcol_order = ["ctor-dtor", "gc", "malloc-free", "text-import", "harness-leaf", "other"]
-pvt = pvt[desired_dcol_order]
+summary = combined.pivot_table(
+    values=['hw_cur_req_mem(B)','cat-duration-ns'],
+    index=['corpus-meanlen-tgt', 'sut-platform', 'category'],
+    aggfunc={'hw_cur_req_mem(B)':'mean','cat-duration-ns':['mean', 'min', 'max']} )
+summary = summary.reset_index()
+summary.columns = summary.columns.to_flat_index()
+summary.columns = [
+    '-'.join(filter(None, col)).replace(' ', '-')  # replaces space with dash if needed
+    for col in summary.columns.to_flat_index()
+]
 
-filtered = pvt.loc[pvt.index.get_level_values('corpus-meanlen').isin(sizes_i_want)]
+# reorder columns with memory as first value (after sut-platform, which is last key)
+# cols = summary.columns.tolist()
+# cols.remove("hw_cur_req_mem(B)-mean")
+# insert_after = cols.index("sut-platform") + 1
+# cols.insert(insert_after, "hw_cur_req_mem(B)-mean")
+# summary = summary[cols]
 
-print(filtered.to_csv(header=True, index=True, sep='\t', na_rep="0"))
+summary = summary[[
+    'corpus-meanlen-tgt',
+    'sut-platform',
+    'hw_cur_req_mem(B)-mean',
+    'category',
+    'cat-duration-ns-mean',
+    'cat-duration-ns-max',
+    'cat-duration-ns-min']]
 
+# print("after summarize", summary.shape[0])
+# print(summary.to_csv(header=True, index=True, sep='\t', na_rep="0"))
+
+
+
+# Ensure 'category' follows the specified order
+summary['category'] = pd.Categorical(summary['category'], categories=catOrder, ordered=True)
+
+# Sort the DataFrame to prepare for cumulative sum
+summary_sorted = summary.sort_values(by=['corpus-meanlen-tgt', 'sut-platform',  'category'])
+
+# Group by the keys and compute exclusive running total
+summary_sorted['grp-prior-duration-ns'] = (
+    summary_sorted
+    .groupby(['corpus-meanlen-tgt', 'sut-platform'])['cat-duration-ns-mean']
+    .transform(lambda s: s.cumsum().shift(fill_value=0))
+)
+
+summary_sorted = summary_sorted.reset_index(drop=True)
+
+
+# print("after accumulation", summary_sorted.shape[0])
+# print(summary_sorted.to_csv(header=True, index=True, sep='\t', na_rep="0"))
+
+
+filtered = summary_sorted[summary_sorted['corpus-meanlen-tgt'].isin(sizes_i_want)]
+
+# print("after filter", filtered.shape[0])
+
+print(filtered.to_csv(header=True, index=False, sep='\t', na_rep="0"))