Code generation tests on Linux and Unix compatibles
8 views
Skip to first unread message
Jeffrey Walton
Jun 20, 2016, 3:23:26 AM6/20/16
to Crypto++ Users
Hi Everyone,
I stubbed out some code generation tests at http://github.com/weidai11/cryptopp/blob/master/cryptest.sh#L365 and http://github.com/weidai11/cryptopp/blob/master/cryptest.sh#L636. The idea is to disassemble a function and then look for instruction mnemonics. The principal is sound but its failing in practice because of C++ name mangling.
There's kind of a solution at http://stackoverflow.com/questions/12400105/getting-mangled-name-from-demangled-name . But in practice, its tough for us to build the mangled names on the fly because it depends on what code path is activated, and it moves around from platform to platform.
Are there any suggestions on verifying code generation?
Jeff
I stubbed out some code generation tests at http://github.com/weidai11/cryptopp/blob/master/cryptest.sh#L365 and http://github.com/weidai11/cryptopp/blob/master/cryptest.sh#L636. The idea is to disassemble a function and then look for instruction mnemonics. The principal is sound but its failing in practice because of C++ name mangling.
There's kind of a solution at http://stackoverflow.com/questions/12400105/getting-mangled-name-from-demangled-name . But in practice, its tough for us to build the mangled names on the fly because it depends on what code path is activated, and it moves around from platform to platform.
Are there any suggestions on verifying code generation?
Jeff
Reply all
Reply to author
Forward
0 new messages