Assembly language is indispensible for 8-bit microcontrollers, but out of place in the 64-bit world. In many cases, the processor architecture may be too complex for a programmer to take advantage of, let alone understand. Between pipelines, caching, and branch prediction, the average assembly language programmer would lag behind an average compiler in code quality, not to mention the other advantages of using a high-level language.
Compilers are indispensible for Intel’s Itanium 2 EPIC architecture. The complexity problem is acute with EPIC, because an optimal code stream will likely need to significantly reorder the instructions to keep all execution units running. This is a straightforward exercise for a compiler, but agony for an assembly language programmer.
The problem is even more acute as more execution units are employed in a design. Though recompiling a program may be trivial, recoding in assembly is tedious at best and error-prone at worst.
System complexity, architecture idiosyncrasies, and timing details make proper instruction ordering important to deliver top system performance. SIMD utilization is another area where a compiler will make a processor shine. The ability of a compiler to exploit powerful hardware like the PowerPC AltiVec becomes even more important because of the difficulty in unrolling a number of embedded loops.
Two major changes have occurred in conjunction with the greater use of 64-bit systems. The first is higher performance of development systems, which allows for more optimizations and more complex optimizations. The second is the use of profile data, which enhances the compiler’s ability to optimize a program through subsequent compilations.
