Mercurial > hg > silvet

annotate testdata/timing/results.txt @ 116:91bb029a847a timing

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
Reorder the calculations to match the series of vector operations in the most recent bqvec code, just in case it's the order of vector calculations that is saving the time rather than the avoidance of std::vector
author Chris Cannam
date Wed, 07 May 2014 09:57:19 +0100
parents a6e136aaa202
children f1116eb464f9
rev   line source
Chris@72 1
Chris@72 2 Thinkpad T540p i5-4330M @2.80GHz with 16GB RAM, plugged in
Chris@72 3 Arch Linux, gcc 4.8.2
Chris@72 4 Using sonic-annotator v1.0 (commit:41c4de1e05d8), release build
Chris@72 5
Chris@72 6 Debug flags: -g -fPIC
Chris@72 7 Release flags: -O3 -ffast-math -msse -mfpmath=sse -ftree-vectorize -fPIC
Chris@72 8
Chris@73 9 Release flags for qm-dsp also include -fomit-frame-pointer
Chris@73 10
Chris@73 11 The input file is 1-channel 16-bit PCM at 44100Hz, duration 0m43.5s.
Chris@72 12
Chris@72 13
Chris@72 14 DEBUG/RELEASE:
Chris@72 15
Chris@72 16 commit:ce64d11ef336, release build of Silvet, release build of qm-dsp
Chris@72 17
Chris@73 18 real 1m44.456s
Chris@73 19 user 1m44.343s
Chris@73 20 sys 0m0.210s
Chris@72 21
Chris@72 22 commit:ce64d11ef336, debug build of Silvet, release build of qm-dsp
Chris@72 23
Chris@73 24 real 14m16.124s
Chris@73 25 user 14m16.907s
Chris@73 26 sys 0m0.217s
Chris@72 27
Chris@72 28 commit:ce64d11ef336, release build of Silvet, debug build of qm-dsp
Chris@72 29
Chris@73 30 real 1m55.204s
Chris@73 31 user 1m55.053s
Chris@73 32 sys 0m0.253s
Chris@72 33
Chris@72 34 Subsequent tests use release builds of both.
Chris@72 35
Chris@72 36
Chris@73 37 VAMP FEATURE SUPPRESSION:
Chris@73 38
Chris@75 39 commit:7133f78ccbf6, as commit:ce64d11ef336 but with CQ output feature
Chris@75 40 return commented out
Chris@75 41
Chris@75 42 real 1m46.162s
Chris@75 43 user 1m46.093s
Chris@75 44 sys 0m0.157s
Chris@75 45
Chris@77 46 commit:78a7bf247016, as commit:ce64d11ef336 but with CQ output and FCQ
Chris@75 47 output feature return commented out
Chris@77 48
Chris@77 49 real 1m45.206s
Chris@77 50 user 1m45.153s
Chris@77 51 sys 0m0.147s
Chris@77 52
Chris@77 53 conclusion: no advantage in removing these
Chris@77 54
Chris@77 55
Chris@77 56 DEBUG PRINTOUTS:
Chris@77 57
Chris@77 58 commit:f3bf6503e6c6, as commit:ce64d11ef336 but with debug printouts
Chris@77 59 removed
Chris@77 60
Chris@77 61 real 1m43.744s
Chris@77 62 user 1m43.657s
Chris@77 63 sys 0m0.203s
Chris@77 64
Chris@81 65 conclusion: obviously we want to remove these eventually, but might as
Chris@77 66 well keep in during testing
Chris@77 67
Chris@80 68
Chris@80 69 EM ITERATIONS:
Chris@80 70
Chris@80 71 commit:5314d3361dfb, as commit:ce64d11ef336 but with only 6 EM
Chris@80 72 iterations instead of 12
Chris@80 73
Chris@80 74 real 0m59.055s
Chris@80 75 user 0m58.897s
Chris@80 76 sys 0m0.193s
Chris@80 77
Chris@80 78 conclusion: EM dominates the time taken, not CQ or note forming
Chris@80 79
Chris@81 80
Chris@81 81 CQ DECIMATOR CONFIGURATION:
Chris@81 82
Chris@81 83 Uncommitted revision (because changes are in CQ subrepo) that is as
Chris@81 84 commit:ce64d11ef336 but with resampler SNR=30 and BW=0.04 instead of
Chris@81 85 SNR=60 and BW=0.02
Chris@81 86
Chris@81 87 real 1m43.176s
Chris@81 88 user 1m43.067s
Chris@81 89 sys 0m0.190s
Chris@81 90
Chris@81 91 conclusion: supports the previous test
Chris@81 92
Chris@84 93
Chris@84 94 EM TWEAKS:
Chris@84 95
Chris@84 96 commit:a0dedcbfa628, as commit:ce64d11ef336 but with variables hoisted
Chris@84 97 out of loops and consts added wherever applicable
Chris@84 98
Chris@84 99 real 1m44.548s
Chris@84 100 user 1m44.460s
Chris@84 101 sys 0m0.183s
Chris@84 102
Chris@84 103 conclusion: compiler already knows this stuff
Chris@84 104
Chris@86 105 commit:64b08cc12da0, as commit:ce64d11ef336 but with loops merged so
Chris@86 106 as theoretically to reduce intermediate calculations
Chris@86 107
Chris@86 108 real 3m46.969s
Chris@86 109 user 3m46.850s
Chris@86 110 sys 0m0.220s
Chris@86 111
Chris@87 112 commit:6075e92d63ab, as commit:64b08cc12da0 but with innermost loop
Chris@87 113 reverted to three loops with simple bodies instead of one with a more
Chris@87 114 complex body
Chris@87 115
Chris@87 116 real 1m44.767s
Chris@87 117 user 1m44.490s
Chris@87 118 sys 0m0.190s
Chris@88 119
Chris@89 120 commit:97b77e7cb94c, as commit:6075e92d63ab but with templates stored
Chris@89 121 as doubles instead of floats (doubling the size of the plugin binary)
Chris@89 122
Chris@89 123 real 1m40.135s
Chris@89 124 user 1m39.820s
Chris@89 125 sys 0m0.230s
Chris@89 126