For some time I didn't need assembler, so I forgot some of my skills about AT&T. However, when I started my Ph.D. one of my professors asked me to optimize some common algorithms using MMX. Thus I had to go back to the old days that I used AT&T assembler embedded in gcc. After that, I have made comparisons using both AT&T and Intel, and also Intel Intrinsics (a C library which allows to use assembler functions directly in C). I have found that gcc + AT&T is the fastest of the three options... so I have no doubts, I will use gcc + AT&T, rather than Intel C Compiler + Intel Assembler or Intel Intrinsics.
Have you found the same difficulties as me? Tell us your experiences...

