Re: [PATCH 4/7] Class-fy partial_die_info

[Yao Qi <qiyaoltc at gmail dot com>]

Simon Marchi <simark@simark.ca> writes:

> Ah, indeed.  I thought that unordered_map would be implemented the
> same way as htab_t, but I see it's not the case.  Doing some quick
> tests on a big binary, it increases the time reading the symbols from
> an average of 37 seconds to an average of 42 seconds.
>
> I understand the different hash table implementation having an impact,
> but I don't really understand how the allocation scheme can have a
> meaningful impact.  The partial_die_info objects are still allocated
> on the obstack, aren't they?  So it's just the space for the table
> itself that isn't on the objstack, but I don't see why that would make
> a difference.

Hi Simon,
We have some perf test cases, in gdb.perf, but they may not cover the
path we are discussing here.  If you want to run them, do these things,

$ cd gdb/testsuite
$ make -j10 build-perf RUNTESTFLAGS="MONSTER=y gmonster1.exp"

// this takes a while to generate many executable files,

$ make  check-perf RUNTESTFLAGS="MONSTER=y gmonster1-null-lookup.exp gmonster1-print-cerr.exp gmonster1-runto-main.exp gmonster1-pervasive-typedef.exp gmonster1-ptype-string.exp gmonster1-select-file.exp"  GDB_PERFTEST_MODE=run

// it takes one hour on my aarch64-linux box

You can get the performance number in perftest.sum.

I run these perf tests, with and without Simon's patch (htab_t ->
std::unordered_map), on aarch64-linux, I don't see speed and space
change.  Again, these existing test cases may not cover the path.

gmonster1:gmonster-null-lookup cpu_time 10-cus 0.0008508 0.0013708
gmonster1:gmonster-null-lookup cpu_time 100-cus 0.0047922 0.0030584
gmonster1:gmonster-null-lookup cpu_time 1000-cus 0.0400274 0.0397188
gmonster1:gmonster-null-lookup cpu_time 10000-cus 0.3885292 0.3862456
gmonster1:gmonster-null-lookup wall_time 10-cus 0.000862598419189 0.00137987136841
gmonster1:gmonster-null-lookup wall_time 100-cus 0.00480117797852 0.00306539535522
gmonster1:gmonster-null-lookup wall_time 1000-cus 0.0400432109833 0.0397333621979
gmonster1:gmonster-null-lookup wall_time 10000-cus 0.388552331924 0.386282110214
gmonster1:gmonster-null-lookup vmsize 10-cus 35900 35896
gmonster1:gmonster-null-lookup vmsize 100-cus 64992 64972
gmonster1:gmonster-null-lookup vmsize 1000-cus 364748 364760
gmonster1:gmonster-null-lookup vmsize 10000-cus 3313284 3313360
gmonster1:gmonster-pervasive-typedef cpu_time 10-cus 0.0682848 0.0737696
gmonster1:gmonster-pervasive-typedef cpu_time 100-cus 0.5843324 0.6391266
gmonster1:gmonster-pervasive-typedef cpu_time 1000-cus 6.061932 6.621443
gmonster1:gmonster-pervasive-typedef cpu_time 10000-cus 62.6619226 67.1514486
gmonster1:gmonster-pervasive-typedef wall_time 10-cus 0.0683585643768 0.0737894058228
gmonster1:gmonster-pervasive-typedef wall_time 100-cus 0.585014867783 0.641395568848
gmonster1:gmonster-pervasive-typedef wall_time 1000-cus 6.07634234428 6.62292981148
gmonster1:gmonster-pervasive-typedef wall_time 10000-cus 62.6738821507 67.2341232777
gmonster1:gmonster-pervasive-typedef vmsize 10-cus 32381 32368
gmonster1:gmonster-pervasive-typedef vmsize 100-cus 41131 41084
gmonster1:gmonster-pervasive-typedef vmsize 1000-cus 129726 129553
gmonster1:gmonster-pervasive-typedef vmsize 10000-cus 1007898 1007878
gmonster1:gmonster-print-cerr cpu_time 10-cus 0.0011148 0.0011168
gmonster1:gmonster-print-cerr cpu_time 100-cus 0.0056498 0.0056002
gmonster1:gmonster-print-cerr cpu_time 1000-cus 0.0502508 0.0948982
gmonster1:gmonster-print-cerr cpu_time 10000-cus 0.4922956 0.4948586
gmonster1:gmonster-print-cerr wall_time 10-cus 0.00112357139587 0.00112380981445
gmonster1:gmonster-print-cerr wall_time 100-cus 0.00565934181213 0.00561075210571
gmonster1:gmonster-print-cerr wall_time 1000-cus 0.0502710342407 0.0949506282806
gmonster1:gmonster-print-cerr wall_time 10000-cus 0.492320823669 0.494928407669
gmonster1:gmonster-print-cerr vmsize 10-cus 32508 32500
gmonster1:gmonster-print-cerr vmsize 100-cus 41308 41300
gmonster1:gmonster-print-cerr vmsize 1000-cus 128944 128948
gmonster1:gmonster-print-cerr vmsize 10000-cus 993324 993316
gmonster1:gmonster-runto-main cpu_time 10-cus 0.0360472 0.0231514
gmonster1:gmonster-runto-main cpu_time 100-cus 0.5829484 0.5236148
gmonster1:gmonster-runto-main cpu_time 1000-cus 9.146062 7.9914552
gmonster1:gmonster-runto-main cpu_time 10000-cus 134.266728 134.1361918
gmonster1:gmonster-runto-main wall_time 10-cus 0.042032957077 0.0313341617584
gmonster1:gmonster-runto-main wall_time 100-cus 0.588344860077 0.530299949646
gmonster1:gmonster-runto-main wall_time 1000-cus 9.15567464828 7.99887242317
gmonster1:gmonster-runto-main wall_time 10000-cus 134.285585785 134.15385747
gmonster1:gmonster-runto-main vmsize 10-cus 32503 32496
gmonster1:gmonster-runto-main vmsize 100-cus 41308 41296
gmonster1:gmonster-runto-main vmsize 1000-cus 128960 128952
gmonster1:gmonster-runto-main vmsize 10000-cus 993336 993296
gmonster1:gmonster-select-file cpu_time 10-cus 0.0972798 0.1028992
gmonster1:gmonster-select-file cpu_time 100-cus 0.886951 0.9285964
gmonster1:gmonster-select-file cpu_time 1000-cus 9.166904 9.7175802
gmonster1:gmonster-select-file cpu_time 10000-cus 92.5872488 95.2634352
gmonster1:gmonster-select-file wall_time 10-cus 0.097380399704 0.103002214432
gmonster1:gmonster-select-file wall_time 100-cus 0.887106800079 0.929784965515
gmonster1:gmonster-select-file wall_time 1000-cus 9.17858901024 9.72046675682
gmonster1:gmonster-select-file wall_time 10000-cus 92.6326120853 95.3189713955
gmonster1:gmonster-select-file vmsize 10-cus 32367 32384
gmonster1:gmonster-select-file vmsize 100-cus 41064 41056
gmonster1:gmonster-select-file vmsize 1000-cus 128694 128703
gmonster1:gmonster-select-file vmsize 10000-cus 993052 993048

-- 
Yao (齐尧)