Menu
▾
▴
Re: [Quantlib-dev] Fwd: Re: QuantLib-1.8 | vc14 | AdaptiveRungeKutta testsuite error
|
From: Ivan A. C. <iva...@gm...> - 2016-09-09 06:30:16
|
3rd attempt, SF blocks attachments.
It failed. Supposedly some other function gets messed up by inline
expansion and that was caught accidentally by the RK test case. Or compiler
ignores the declspec.
--- ./ql18_vc2015_mk2/QuantLib-1.8/ql/math/ode/adaptiverungekutta.hpp
2015-12-21 13:46:03.000000000 +0300
+++ ./ql18_vc2015_mk3/QuantLib-1.8/ql/math/ode/adaptiverungekutta.hpp
2016-09-08 19:46:27.359406500 +0300
@@ -161,7 +161,7 @@
}
template<class T>
- T AdaptiveRungeKutta<T>::operator()(const OdeFct1d& ode,
+ __declspec(noinline) T
AdaptiveRungeKutta<T>::operator()(const OdeFct1d& ode,
const T y1,
const Real x1,
const Real x2) {
@@ -170,7 +170,7 @@
}
template<class T>
- void AdaptiveRungeKutta<T>::rkqs(std::vector<T>& y,
+ __declspec(noinline) void
AdaptiveRungeKutta<T>::rkqs(std::vector<T>& y,
const std::vector<T>& dydx,
Real& x,
const Real htry,
@@ -213,7 +213,7 @@
}
template <class T>
- void AdaptiveRungeKutta<T>::rkck(const std::vector<T>& y,
+ __declspec(noinline) void AdaptiveRungeKutta<T>::rkck(const
std::vector<T>& y,
const std::vector<T>& dydx,
Real x,
const Real h,
^C:\TMP\QL18_VC2015_MK3\QUANTLIB-1.8\TEST-SUITE\QUANTLIBTESTSUITE.CPP
/c /IC:\BIN\BOOST_1_61_0 /I.. /Zi /nologo /W3 /WX- /O2 /Ob2 /Ot /D NDEBUG
/D WIN32 /D _CONSOLE /D _SCL_SECURE_NO_DEPRECATE /D
_CRT_SECURE_NO_DEPRECATE /D _MBCS /GF /Gm- /EHsc /MD /GS /Gy /fp:precise
/Zc:wchar_t /Zc:forScope /Zc:inline /GR /Fo".\BUILD\VC140\X64\RELEASE\\"
/Fd".\BUILD\VC140\X64\RELEASE\VC140.PDB" /Gd /TP
C:\TMP\QL18_VC2015_MK3\QUANTLIB-1.8\TEST-SUITE\QUANTLIBTESTSUITE.CPP
Testing adaptive Runge Kutta...
unknown location : fatal error : in "QuantLib test suite/ode
tests/QuantLib::detail::quantlib_test_case(&OdeTest::testAdaptiveRungeKutta)":
class QuantLib::Error: Stepsize (3.1414) underflow in
AdaptiveRungeKutta::rkqs
c:\tmp\ql18_vc2015_mk3\quantlib-1.8\test-suite\utilities.hpp(74): last
checkpoint
Testing matrix exponential based on ode...
unknown location : fatal error : in "QuantLib test suite/ode
tests/QuantLib::detail::quantlib_test_case(&OdeTest::testMatrixExponential)":
class QuantLib::Error: Too many steps (10000) in AdaptiveRungeKutta
c:\tmp\ql18_vc2015_mk3\quantlib-1.8\test-suite\utilities.hpp(74): last
checkpoint
Testing matrix exponential of a zero matrix based on ode...
On Sep 9, 2016 10:15 AM, "Ivan A. Cherkasov" <iva...@gm...>
wrote:
> ---------- Forwarded message ----------
> From: "Ivan A. Cherkasov" <iva...@gm...>
> Date: Sep 9, 2016 10:10 AM
> Subject: Re: [Quantlib-dev] QuantLib-1.8 | vc14 | AdaptiveRungeKutta
> testsuite error
> To: "Peter Caspers" <pca...@gm...>
> Cc: <qua...@li...>, "Luigi Ballabio" <
> lui...@gm...>
>
> Hello.
> It failed. Please see the attached files.
>
> Supposedly some other function gets messed up by inline expansion and that
> was caught accidentally by the RK test case. Or compiler ignores the
> declspec.
> On Sep 8, 2016 4:44 PM, "Peter Caspers" <pca...@gm...> wrote:
>
>> Yes. Ivan, could you try to put
>>
>> __declspec(noinline)
>>
>>
>> before the rkqs and rkck and operator() functions and see if it works
>> then?
>>
>> On 08 Sep 2016, at 14:22, Luigi Ballabio <lui...@gm...>
>> wrote:
>>
>> Except it's not good if VC14 fails in the default configuration...
>>
>>
>> On Thu, Sep 8, 2016 at 2:19 PM Peter Caspers <pca...@gm...>
>> wrote:
>>
>>> Hi,
>>>
>>> I don’t see an issue with /Ob2 on VC12. This all looks like a bug in the
>>> VC14 - code optimization to me, nothing we can (or should) really fix in
>>> the code?
>>>
>>> Regards
>>> Peter
>>>
>>> On 08 Sep 2016, at 09:19, Ivan A. Cherkasov <iva...@gm...>
>>> wrote:
>>>
>>> Hello.
>>>
>>> This failed for me (CL.command.1.tlog):
>>>
>>> ^C:\TMP\QL18_VC2015\REPRO_RK\REPRO_RK.CPP
>>> /c /IC:\BIN\BOOST_1_61_0 /I"C:\TMP\QL18_VC2015\QUANTLIB-1.8" /Zi
>>> /nologo /W3 /WX- /sdl /O2 /Oi /GL /D NDEBUG /D _CONSOLE /D _UNICODE /D
>>> UNICODE /Gm- /EHsc /MD /GS /Gy /fp:precise /Zc:wchar_t /Zc:forScope
>>> /Zc:inline /Fo"X64\RELEASE\\" /Fd"X64\RELEASE\VC140.PDB" /Gd /TP
>>> C:\TMP\QL18_VC2015\REPRO_RK\REPRO_RK.CPP
>>>
>>> This worked for me:
>>>
>>> ^C:\TMP\QL18_VC2015\REPRO_RK\REPRO_RK.CPP
>>> /c /IC:\BIN\BOOST_1_61_0 /I"C:\TMP\QL18_VC2015\QUANTLIB-1.8" /Zi
>>> /nologo /W3 /WX- /sdl /O2 /Ob0 /Ot /D NDEBUG /D _CONSOLE /D _UNICODE /D
>>> UNICODE /Gm- /EHsc /MD /GS /Gy /fp:precise /Zc:wchar_t /Zc:forScope
>>> /Zc:inline /Fo"X64\RELEASE\\" /Fd"X64\RELEASE\VC140.PDB" /Gd /TP
>>> C:\TMP\QL18_VC2015\REPRO_RK\REPRO_RK.CPP
>>>
>>> I’m using “Version 14.0.25422.01 Update 3”.
>>>
>>> I rebuilt from scratch using QuantLib-1.8.tar.gz and QuantLib_vc14.sln,
>>> the only thing I changed in the project settings is the C/C++ ->
>>> Optimization -> Inline Function Expansion = Disabled (/Ob0). It succeeded,
>>> test suite log attached.
>>> On Sep 7, 2016 10:00 PM, "Peter Caspers" <pca...@gm...> wrote:
>>>
>>>> Can you post the full compiler command line (you find that under C++ /
>>>> Command Line), I can not provoke the error no matter what optimization
>>>> flags I use
>>>>
>>>> /FR".\build\vc120\x64\Release\" /GS /W3 /Gy /Zc:wchar_t /I"..\.." /Zi
>>>> /Gm- /Ox /Ob2 /Fd".\build\vc120\x64\Release\" /fp:fast /D "NDEBUG" /D
>>>> "WIN32" /D "_CONSOLE" /D "_SCL_SECURE_NO_DEPRECATE" /D
>>>> "_CRT_SECURE_NO_DEPRECATE" /D "_MBCS" /fp:except /errorReport:prompt /GF
>>>> /WX- /Zc:forScope /GR /Gd /MD /Fa".\build\vc120\x64\Release\" /EHsc
>>>> /nologo /Fo".\build\vc120\x64\Release\" /Ot
>>>> /Fp".\build\vc120\x64\Release\BermudanSwaption.pch”
>>>>
>>>> On 07 Sep 2016, at 19:59, Ivan A. Cherkasov <iva...@gm...>
>>>> wrote:
>>>>
>>>> Also works with /O2 /Ob0 (w/o inlines)
>>>>
>>>>
>>>> <testsuite.log>
>>>
>>>
>>> ------------------------------------------------------------
>>> ------------------
>>> _______________________________________________
>>> QuantLib-dev mailing list
>>> Qua...@li...
>>> https://lists.sourceforge.net/lists/listinfo/quantlib-dev
>>>
>>
>>
|
