With vector machines, Fortran yielded (much) faster code because (a) there was a whole lot of work on preprocessors for "dusty deck" code and (b) the lack of pointers and their overlap let the compilers do their stuff. C code needed a lot of #pragma's to let it know that some pointers didn't overlap and so could be optimized.
We found it best to to the front-end and I/O of the code in C and the compute in Fortran. Fastest calculations especially with those tailored libraries.
Fortran certainly has its place, but if you really need to squeeze every bit of performance out of a machine I think assembly might be better and not much harder.
C allowed pointer aliasing, so you always have to refresh reads incase a pojnter modified the value. With any loop, that is just more reads to update dependancies every cycle. When not aliased, you cache value in register. Much later C introduced “restrict” keyword to match fortran performance code. c++ is still fighting this by insisting you work with different types, hence by default you dont get aliasing but that doesnt cover 100% of scenarios.
We found it best to to the front-end and I/O of the code in C and the compute in Fortran. Fastest calculations especially with those tailored libraries.