e.g. can I reverse engineer and extract the assembler code of a windows aplication, and ther compile it back with the flat assembler Thanks in regards.

Maybe you help IDA?

There are surely decompilers that make assembler code, but how to compile again the source code?

I don't know any compiler that does that, but you could create an ollydbg script I guess?

Though I have another question: Why?

Imho, this isn't generally very useful.

If you want to extend/augment and application, by far the easiest and most manageable way is to write your extension in a DLL, and either augment the executable to load your DLL, or use a loader that injects it. From the DLL, you can then patch the necessary locations in the executable, creating 'hooks' that call into your DLL.

There's just too many things that can go wrong with dis/reassembly, especially when dealing with real-world applications.

There is a blind spot every optimizer I have seen (including myself) has. It would be cool to be able to start with the assembly output of Intel's compiler and patch around that blind spot, however Intel's assembly language output seems to be a hodgepodge of 3 or more assembly languages so that it won't assemble with, for e.g. gas or ml64. Given a disassembler this wouldn't be a problem, but disassembly in the general case is not possible because the code could calculate an address and then jump to it. Detecting when an address calculation is occurring and modifying it so that the jump would be to the changed address is not going to happen.

Maybe the only reasonable thing to do in this case would be to write a translator from Intel assembly output to an assembly language that some assembler understands but this would be a lot of work for just a couple of percent improvement in overall performance.

couple of percent is good optimization, but you will hardly manage it with lowlevel optimization. especially optimization on binary code.

Some perl wizard could probably whip up a script to do it pretty quickly...


Depends on the target platform and what the code is for. <5% speedup on a 1+ GHz x86 generally isn't worth it if it's a lot of work, imho.