The open source project denoted below has a deficiency when the following types expression, parser and symbol_table are specialised to "float" type then the generated expression is evaluated.
[url removed, login to view]
It seems that the computations are occurring as if the type was "double". This is seen on both 32-bit and 64-bit systems using a wide variety of compilers and systems.
The objective, is to take the stock [url removed, login to view] compile it and note the results for both ExprTk and native expression evaluations using the double type.
Then to recompile the benchmark using the float type, run it and note the results (Native and ExprTk), and then make any changes necessary to the file [url removed, login to view] so as to improve the results seen in the benchmark.
The results seen in the benchmark are the indicators of success. The benchmark will be independently compiled with the changes and without and comparisons made, There is an expectation of demonstrable performance increase of at least 10-15% when using the float type.
The deliverable will undergo a final step of validation, using the benchmark suite denoted below (epsilon set to 1). It is expected that an increase in performance across the board for ExprTkFloat benchmark will be observed.
[url removed, login to view]
As such the outcome is very specific and unambiguous and payment will be made on any deliverable that matches ALL the criteria denoted and achieves the minimum increases as specified previously.
01. Any changes must only contain pure C++ syntax.
02. Only the [url removed, login to view] file must be changed, not other files in the project can be modified.
02. No compiler specific syntax/capabilities/functions/builtins etc are allowed.
03. Only C++03 version of the language is allowed (no C++11 or C++98)
04. Using the stock/unmodified makefile found in the project, ALL changes once applied must compile error and warning free.
05. The changes must compile as per  on at least 3 of the following compilers GCC4.3+,Clang3.1+,Intel9.0+,MSVC
06. For an MSVC compilation the option  is not possible, so a solution with W4 set will be enough.
07. The results must work successfully and consistently on at least the x86 and x86-64 platforms (other platforms welcome)
08. The results must work successfully and consistently on Linux (kernel 2.6+) and Win32 (WinXP+) and Win64
09. The stock version of exprtk_test must compile and run successfully (No errors) with the changes made to [url removed, login to view]
10. No 3rd party solutions, libraries or frameworks will be allowed. Only the standard C++ libraries will allowed.
11. All changes must be yours and yours alone, all rights to those changes will be handed over exclusively without further dependencies on any other entities.
01. Text file detailing each of the changes and why they were required and any other pertinent information, ALL the changes must be valid and none of them can be left unexplained.
02. In  there will also be a listing of the results from running the benchmark using the stock and use the patched versions of [url removed, login to view]
03. A patch from the stock version of [url removed, login to view] with the change as per that will achieve the desired results
04. The patch must fix the underlying issue.
05. The deliverable MUST meet all the requirements set above and in the description and constraints sections.
10 freelancers are bidding on average $482 for this job
I have experience working in Performance domain both at the hardware level and software level while working for AMD. I can help you get this done, I have dropped a private message to you as well. Regards, Jegan