__asm Command

Usually, if you have a good optimizing compiler, it's almost never necessary to do this. Sometimes (pretty often actually) the compiler will create better assembly than a programmer could. If it's purely for learning purposes, though, go ahead and try it out - sometimes it's fun to "compete" with the compiler and see who can come up with the better assembly! [lol]

You can try, but the compilers now are very good, so unless you're an expert I don't think you can even make optimizations at the same level as the compiler, let alone better. Additionally, I think that writing assembly instructions in the middle of a C++ program will turn off some optimizations, because the compiler is supposed to be free to use the registers for example as it pleases. With you getting control of that, you're basically getting into its way.

Thanks! I actually tried a kind of "Benchmark Test" -- I created two functions which added two numbers -- one using assembly. I added numbers a bunch of times, and timed the operation -- the non-assembly-using function was faster...

Inline assembly still is useful - for instance if you want to take advantage of the latest instruction set enhancements (like SSE3), which might not be supported by the compiler yet.
These cases are rare, however.

Quote:

Original post by darookie Inline assembly still is useful - for instance if you want to take advantage of the latest instruction set enhancements (like SSE3), which might not be supported by the compiler yet. These cases are rare, however.

What he said, plus not all processors have the same SIMD intruction set, if you don't want to compile separate exe's one for no SIMD, one for MMX, one for 3dNow, one for SSE, one for SSE2, one for SSE3, then you definitelly have to use inline assembly.

I recently wrote a routine using SSE for matrix multiplication, I was curious as to what the compiler would optimize the non SSE version to, and surprincingly neither VC++ or GCC took advantage of SSE, even though in the GCC case I did passed the -msse flag, so using inine asm at least guarantees you will be using the set of instructions you think you're using.