[lmi] Can linux-perf illuminate this problem?

[Greg Chicares]

Vadim--I seem to have done something awful in a commit that I've
pushed only to this branch, but not to origin/master:

https://git.savannah.nongnu.org/cgit/lmi.git/commit/?h=odd/eraseme
| commit        d1ae1eaeef2ed1a1ddd15b47f19703d6a26b1e5d
| Use currency type for max non-MEC payment
| This commit has a disastrous effect on performance as measured by
| 'make cli_timing': for the first ('naic, no solve') scenario, the
| penalty is 14% for i686 and 30% for x86_64. How can that be?

I pushed that branch to origin so that I could easily switch to it
in a chroot that's configured to use linux-perf. I can get 'perf'
to tell me what I already know--that AccountValue::DoMonthDR() is
where the problem is--but can it tell me which statements are
responsible?

I used this command:

$LD_LIBRARY_PATH=.:/opt/lmi/bin:/opt/lmi/local/gcc_x86_64-pc-linux-gnu/lib/:/srv/cache_for_lmi/perf_ln
 /srv/cache_for_lmi/perf_ln/perf_4.19 record --freq=max --call-graph=lbr 
/opt/lmi/bin/lmi_cli_shared --accept --data_path=/opt/lmi/data --selftest

to collect data: first, with the offending commit; then, again,
without the offending commit.

This command:
  $LD_LIBRARY_PATH=.:/srv/cache_for_lmi/perf_ln 
/srv/cache_for_lmi/perf_ln/perf_4.19 diff
tells me:

# Baseline  Delta Abs  Shared Object         Symbol                             
                  >
# ........  .........  ....................  
.....................................................>
#
    14.90%    -10.35%  liblmi.so             [.] AccountValue::DoMonthDR

but doesn't let me drill down.

This command:
  $LD_LIBRARY_PATH=.:/srv/cache_for_lmi/perf_ln 
/srv/cache_for_lmi/perf_ln/perf_4.19 report -i /opt/lmi/src/lmi/perf.data.old
and a similar one without '.old' at the end show me a detailed
profile with (perf.data.old) and without (perf.data) the offending
commit. Truncating right after the AccountValue::DoMonthDR() line:

Samples: 77K of event 'cycles:ppp', Event count (approx.): 18328223965
  Children      Self  Command         Shared Object            Symbol
+   84.36%     0.00%  lmi_cli_shared  liblmi.so                [.] IllusVal::run
+   75.26%     0.22%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::RunOneCell
+   70.84%     0.00%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::Solve
+   70.83%     0.02%  lmi_cli_shared  liblmi.so                [.] 
decimal_root<SolveHelper>
+   70.40%     0.06%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::SolveTest
+   62.39%     0.37%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::IncrementBOM
+   59.99%    14.90%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::DoMonthDR

+   82.77%     0.00%  lmi_cli_shared  liblmi.so                [.] IllusVal::run
+   72.69%     0.20%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::RunOneCell
+   70.13%     0.00%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::Solve
+   70.12%     0.02%  lmi_cli_shared  liblmi.so                [.] 
decimal_root<SolveHelper>
+   69.59%     0.07%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::SolveTest
+   58.00%     0.44%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::IncrementBOM
+   55.08%     4.55%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::DoMonthDR

shows that the 14.90% vs 4.55% performance regression is in that
function. I can drill down into both...

-   59.99%    14.90%  lmi_cli_shared  liblmi.so                [.] 
AccountValue::DoMonthDR
   - 45.09% AccountValue::DoMonthDR
        14.10% AccountValue::TxSpecAmtChange@plt
      + 4.36% AccountValue::TxSetDeathBft
      + 3.84% AccountValue::InitializeMonth
      + 3.10% Irc7702A::MaxNonMecPremium
      + 2.70% Irc7702A::MaxNecessaryPremium
        1.81% AccountValue::TxSetCoiCharge@plt
      + 1.42% AccountValue::TxSetBOMAV

   - 50.53% AccountValue::DoMonthDR
        16.73% AccountValue::TxSpecAmtChange@plt
      + 5.12% AccountValue::TxSetDeathBft
      + 4.29% AccountValue::InitializeMonth
      + 3.30% Irc7702A::MaxNonMecPremium
      + 2.79% Irc7702A::MaxNecessaryPremium
        2.11% AccountValue::TxSetCoiCharge@plt
      + 1.73% AccountValue::TxSetBOMAV

...but that only seems to tell me how much time is spent in each
function called by AccountValue::DoMonthDR(). However, I suspect
that I've done something atrocious inside DoMonthDR() itself.
Can 'perf' help me find that?