There is an often-missed elephant in the compiler optimization living room in the form of poor language design leading to a tremendous amount of wasted human effort. First, effort is demanded of the programmer to write the clever optimized C code in the first place, and then effort is demanded of the compiler writers to recognize the clever optimized C code and compile it down to the native instruction that does the Right Thing.

All of this effort could be eliminated simply by adding a function to the standard library called bitcount. This would make life easier for both the programmer, who could now count bits by calling bitcount, and the compiler writers, who no longer have to write and maintain code that recognizes clever hacks.

This (counting bits) is not an isolated incident either. There is an entire field of academic endeavor devoted to this sort of thing, focusing particularly on recognizing common idioms involved in vector and array processing. All of this could be eliminated at a stroke through better language design. Instead, an entire academic field dons the ball-and-chain of a poorly designed language and then pats itself on the back for getting to the finish line panting and sweating, and people like me who say, "Um, why not just take off the extra weight and use a better designed language?" get marginalized at best and vilified at worst.