I would suggest that for most problems a good compiled language (Fortran compilers have beaten very skilled assembly language programmers for about 40 years; I see little reason why C/C++ wouldn't do as well) will beat even the best hand-coded assembler for many classes of problems

Are you making the claim about assembler for programming time or for run time ? Or size ?

If run time and/or size...

Code generation has become better and better. Processors have become better at supporting code generators. Processors have become more complex, so that to get the most out of them you need to worry about the relationship between instructions -- instruction scheduling is hard work for the average human. So these days it's effectively impossible to cost justify assembler vs C -- cycles and RAM are cheap and plentiful, and next year even cheaper and more plentiful.

I suppose with specialist vector processors and such, even FORTRAN could out-instruction-schedule a human. I've no experience with such heavy duty scientific iron.

However, as a bare-knuckle assembler black-belt, I'd have been flabbergasted not to be able to out-run a compiler 20 years ago, let alone 40. I don't want to start an OT religious war... but I must ask if you have evidence for your assertions ?

Does anyone have a section of code that is so time critical it's worth casting into assembler ? I have a hankering to go mano-a-mano with some real code !