[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[lmi-commits] [lmi] valyuta/002 1fbef51 13/14: Make it even faster

From: Greg Chicares
Subject: [lmi-commits] [lmi] valyuta/002 1fbef51 13/14: Make it even faster
Date: Thu, 17 Sep 2020 16:27:16 -0400 (EDT)

branch: valyuta/002
commit 1fbef5167563c9a24f08a3f2592188412890129b
Author: Gregory W. Chicares <gchicares@sbcglobal.net>
Commit: Gregory W. Chicares <gchicares@sbcglobal.net>

    Make it even faster
    $wine ./lmi_cli_shared.exe --accept --data_path=/opt/lmi/data --selftest
    Test speed:
      naic, no solve      : 6.618e-02 s mean;     65698 us least of  16 runs
      naic, specamt solve : 1.177e-01 s mean;    117107 us least of   9 runs
      naic, ee prem solve : 1.089e-01 s mean;    107848 us least of  10 runs
      finra, no solve     : 2.400e-02 s mean;     22353 us least of  42 runs
      finra, specamt solve: 7.148e-02 s mean;     68895 us least of  15 runs
      finra, ee prem solve: 6.582e-02 s mean;     63941 us least of  16 runs
    same, but 64-bit:
      naic, no solve      : 2.744e-02 s mean;     27096 us least of  37 runs
      naic, specamt solve : 4.420e-02 s mean;     43700 us least of  23 runs
      naic, ee prem solve : 4.180e-02 s mean;     40617 us least of  24 runs
      finra, no solve     : 1.422e-02 s mean;     13251 us least of  71 runs
      finra, specamt solve: 2.908e-02 s mean;     28452 us least of  35 runs
      finra, ee prem solve: 2.768e-02 s mean;     27004 us least of  37 runs
    This still isn't as fast as 'master' [see 'README.branch.patch' for
    recent measurements]; but, taking the third of six scenarios as the
    benchmark, the penalty is only
      (107848 / 84947) - 1
    which is about twenty-seven percent, for 32 bit; or, for 64 bit:
      (40617 / 37410) - 1
    which is only about nine percent.
    And a change from 32-bit master to 64-bit valyuta/002 actually _reduces_
    the calculation time by
      (40617 / 84947) - 1
    about fifty-two percent.
    What about accuracy? The full lmi system test [using mostly proprietary
    products] doesn't quite run with this commit, but only because of
    spurious regressions like this:
    4415.97 GrossPmts[Month]
    4415.97 EeGrossPmts[Month] + ErGrossPmts[Month]
    4000 EeGrossPmts[Month]
    415.97 ErGrossPmts[Month]
    0 Year
    0 Month
    [ihs_avmly.cpp : 1294]
    Assertion \
      'GrossPmts[Month] == EeGrossPmts[Month] + ErGrossPmts[Month]' failed.
    But, with the purely superficial changes in the next (throwaway) commit,
    the system test succeeds (it exhibits a number of small regressions that
    can all be justified).
 currency.hpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/currency.hpp b/currency.hpp
index a7ef89b..86c36cf 100644
--- a/currency.hpp
+++ b/currency.hpp
@@ -43,7 +43,7 @@
 // Similar restrictions apply to these macros, but they'd be tedious
 // to write. Use them only with floating-point dollars.
 #   if defined __GNUC__

reply via email to

[Prev in Thread] Current Thread [Next in Thread]