New user self-registration is disabled due to spam. Please email eigen-core-team @ if you need an account.
Before reporting a bug, please make sure that your Eigen version is up-to-date!
Bug 1102 - Linker error in MathFunctions.h when using -Ofast on gcc 4.7.2
Summary: Linker error in MathFunctions.h when using -Ofast on gcc 4.7.2
Alias: None
Product: Eigen
Classification: Unclassified
Component: Core - general (show other bugs)
Version: unspecified
Hardware: All All
: Normal Unknown
Assignee: Nobody
Depends on:
Reported: 2015-10-30 17:55 UTC by Graham Neubig
Modified: 2015-11-02 15:10 UTC (History)
5 users (show)

A simple example that can reproduce the error (413 bytes, application/x-gzip)
2015-10-30 17:55 UTC, Graham Neubig
no flags Details

Description Graham Neubig 2015-10-30 17:55:04 UTC
Created attachment 620 [details]
A simple example that can reproduce the error

I'm having a linker error when compiling a program with the latest development version of Eigen on gcc 4.7.2 (Debian Linux), only when the "-Ofast" option is enabled.

The error seems to be happening because the "isnan" and "isinf" functions in MathFunctions.h that use __builtin_isnan and __builtin_isinf (lines 890-895 in my version) are not inlined, and when Eigen is used in more than one cpp file, linking dies because the functions are multiply defined.

I've attached a simple example that can reproduce the error, and the error is as below:

classb.o: In function `bool Eigen::numext::isnan<long double>(long double const&)':
classb.cpp:(.text+0x0): multiple definition of `bool Eigen::numext::isnan<long double>(long double const&)'
classa.o:classa.cpp:(.text+0x0): first defined here
classb.o: In function `bool Eigen::numext::isnan<double>(double const&)':
classb.cpp:(.text+0x10): multiple definition of `bool Eigen::numext::isnan<double>(double const&)'
classa.o:classa.cpp:(.text+0x10): first defined here
classb.o: In function `bool Eigen::numext::isnan<float>(float const&)':
classb.cpp:(.text+0x20): multiple definition of `bool Eigen::numext::isnan<float>(float const&)'
classa.o:classa.cpp:(.text+0x20): first defined here
classb.o: In function `bool Eigen::numext::isinf<double>(double const&)':
classb.cpp:(.text+0x30): multiple definition of `bool Eigen::numext::isinf<double>(double const&)'
classa.o:classa.cpp:(.text+0x30): first defined here
classb.o: In function `bool Eigen::numext::isinf<float>(float const&)':
classb.cpp:(.text+0x50): multiple definition of `bool Eigen::numext::isinf<float>(float const&)'
classa.o:classa.cpp:(.text+0x50): first defined here
classb.o: In function `bool Eigen::numext::isinf<long double>(long double const&)':
classb.cpp:(.text+0x70): multiple definition of `bool Eigen::numext::isinf<long double>(long double const&)'
classa.o:classa.cpp:(.text+0x70): first defined here
collect2: error: ld returned 1 exit status
Comment 1 Gael Guennebaud 2015-10-30 21:27:15 UTC
Thank you for the detailed report. Fixed:
Comment 2 Silvio Traversaro 2015-11-01 00:19:00 UTC
I don't know if it is directly related, but we are experiencing several problem similar to the one explained this bug. 

Here's  a log from one CI run on Visual Studio 2012 using the latest Eigen from the default branch: 

In particular it is strange the syntax (isinf./) at line 880 of Eigen/src/Core/MathFunctions.h  .
Comment 3 Gael Guennebaud 2015-11-02 15:10:53 UTC
arf, nothing related just a typo

Note You need to log in before you can comment on or make changes to this bug.