Numerical discrepancy between 2 project configurations in fltpnt ('Debug' vs 'float')
I am running C code compiled and linked in two different configurations of the same 'project' with exactly the same settings (source/header files, include directories, libraries, additional dependencies, etc.). The two configurations are 'Debug' and 'float'; both are in floating point arithmetic, which is selected in preprocessing (a macro that selects fixed or floating point arithmetic is not defined, leading to compiling/linking in floating point). The numerical results differ between the two runs; e.g., an array element that is a complex-number structure (with two float members---real and imag) shows significantly different numeric values after the two runs.
Has anybody come across a similar problem? I would appreciate any insight, or suggestions for where to look---thank you in advance.
Re: Numerical discrepancy between 2 project configurations in fltpnt ('Debug' vs 'float')
Hallo.
Are there any differences in optimisation or preprocessor symbols?
Re: Numerical discrepancy between 2 project configurations in fltpnt ('Debug' vs 'float')
Hello, I was away from the site (lost password; had to see to other tasks), sorry for the silence...
Preprocessor Definitions are "WIN32;_DEBUG;_CONSOLE" in 'Debug' and none in 'float'.
Optimizations are disabled in 'Debug' and "Maximize Speed (/O2)" in 'float'.
Re: Numerical discrepancy between 2 project configurations in fltpnt ('Debug' vs 'flo
Quote:
Originally Posted by D.Mihai
Has anybody come across a similar problem? I would appreciate any insight, or suggestions for where to look---thank you in advance.
Maybe it's a bug in your program.
Regards,
Paul McKenzie
Re: Numerical discrepancy between 2 project configurations in fltpnt ('Debug' vs 'float')
Thank you, I will look again into it.
Regards,
Mihai Ionescu
Re: Numerical discrepancy between 2 project configurations in fltpnt ('Debug' vs 'float')
Quote:
Originally Posted by D.Mihai
I am running C code compiled and linked in two different configurations of the same 'project' with exactly the same settings (source/header files, include directories, libraries, additional dependencies, etc.). The two configurations are 'Debug' and 'float'; both are in floating point arithmetic, which is selected in preprocessing (a macro that selects fixed or floating point arithmetic is not defined, leading to compiling/linking in floating point). The numerical results differ between the two runs; e.g., an array element that is a complex-number structure (with two float members---real and imag) shows significantly different numeric values after the two runs.
Has anybody come across a similar problem? I would appreciate any insight, or suggestions for where to look---thank you in advance.
I suggest reading this FAQ: http://www.codeguru.com/forum/showthread.php?t=323835.
