Post-processor for hatari profiler data
hatari_profile.py [options] <datafiles>
A Python script to post-process data produced by Hatari profiler "save" commands, both for CPU and DSP:
profile save <filename> dspprofile save <filename>
It can provide function level (CPU and DSP) processor usage information for anything Hatari profiler records: instruction counts, processor cycles and depending on processor, also instruction cache misses or cycle differences.
This information can be provided as ASCII list of heaviest functions, as callgraphs (if profile data includes caller information), or as callgrind format export which can be viewed in (Linux) Kcachegrind GUI.
Invoking hatari_profile without arguments lists all of its options.
Regardless of whether profile data contains some symbol information, you should always give script all relevant debug symbols, otherwise costs might not get correctly assigned to symbols preceding those costs.
This post-processes profile data for EmuTOS (ROM), with symbol file having fixed/absolute addresses (-a), shows statistics (-s) and top lists (-t), but limits list output to anything taking over 2% (-l 2) and includes "propagated" subroutine call costs (-p):
------------------------------------------------------------- $ hatari_profile.py -a etos512k.sym -st -l 2 -p etos-boot.txt Hatari profile data processor Parsing absolute symbol address information from etos512k.sym... 1538 lines with 1236 code symbols/addresses parsed, 0 unknown. Parsing profile information from etos-boot.txt... 16993 lines processed with 465 functions. CPU profile information from 'etos-boot.txt': - Hatari v1.6.2+ (Apr 15 2013), OldUAE CPU core Time spent in profile = 4.60740s. Calls: - max = 1800, in ___mulsi3 at 0xe60790, on line 16477 - 16659 in total Executed instructions: - max = 315254, in _timeout_gpip+38 at 0xe06fd8, on line 3237 - 3074165 in total Used cycles: - max = 6311628, in _timeout_gpip+38 at 0xe06fd8, on line 3237 - 36957124 in total Calls: 10.80% 10.88% 1800 ___mulsi3 8.04% 8.16% 1339 _memcpy 5.20% 0.26% 866 _int_timerc 4.61% 768 _call_user_wheel 3.86% 3.86% 643 _min 3.76% 3.82% 627 _sti 3.76% 3.76% 627 _cli 3.30% 3.30% 550 _bcostat4 Executed instructions: 61.58% 61.79% 1892940 _timeout_gpip 8.62% 9.33% 264918 _draw_rect 4.96% 4.98% 152582 _run_calibration 4.06% 4.09% 124944 _blank_out 3.22% 99134 meminit Used cycles: 61.58% 62.01% 22759756 _timeout_gpip 5.69% 6.85% 2104464 _draw_rect 4.29% 1586108 meminit 4.19% 4.22% 1547768 _run_calibration 3.04% 3.07% 1122816 _blank_out 2.88% 2.98% 1065272 _stop_until_interrupt -------------------------------------------------------------
This creates GraphViz callgraph files (-g) and Kcachegrind callgrind data file (-k) from Bad Mood (Doom BSP viewer) profile. Symbols for the program are TEXT section relative (-r), calls to some interrupt routines in the program are ignored as they aren't real calls, callgraph nodes using over 2% are highlighted, and there are some options to limit and simplify the graph:
------------------------------------------------------------- $ hatari_profile.py -r bmsym.sym \ -k -g -p --emph-limit 2.0 --limit 0.5 \ --ignore-to framecounter,new_vbi,stabilizer_b \ --compact --no-leafs --no-intermediate \ badmood-profile.txt Hatari profile data processor Parsing TEXT relative symbol address information from bmsym.sym... 1023 lines with 392 code symbols/addresses parsed, 0 unknown. Parsing profile information from badmood-profile.txt... 1420 lines processed with 63 functions. Ignoring 29 switches to framecounter Ignoring 9 switches to stabilizer_b Ignoring 8 switches to new_vbi Of all 1562 switches, ignored 56 for type(s) ['r', 'u', 'x']. Generating callgrind file 'badmood-profile.cg'... Generating 'badmood-profile-0.dot' DOT callgraph file... Generating 'badmood-profile-1.dot' DOT callgraph file... Generating 'badmood-profile-2.dot' DOT callgraph file... Generating 'badmood-profile-3.dot' DOT callgraph file... CPU profile information from 'badmood-profile.txt': - Hatari v1.6.2+ (Apr 10 2013), WinUAE CPU core -------------------------------------------------------------
Written by Eero Tamminen <oak at helsinkinet fi>.
This program is free software; you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation; either version 2 of the License, or (at your option) any later version.
This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.