"Lee Rhodes" <lee@xxxxxxxxx> wrote: > > I presume you mean an early FP coprocessor, where the CPU issues the > instruction and the binary data sent to the FP unit may have different byte > orders depending on the origin of the data? Precisely. And whether it uses it or not will depend on whether it is there at start-up. Such designs are staging a (perhaps temporary) come-back - look up Clearspeed and SGI/AMD/Intel's plans for hybrid SMP designs. > Sorry, when I said "compiler" I meant compiler/assembler. Nonetheless, what > do you mean "only for C integers"? The compiler does not actually need to know the layout of floating-point, even in C99 - after all, C90 had to support the designs mentioned above. And a Fortran compiler doesn't need to do so even for integers. There is nothing in the languages that expose those to the program at compile time. > That is certainly true for the byte shuffle operations for conversion. But a > run-time detection of endianness might require a memory operation. I would > have to think about it. Oh, it does. Probably lots of them. But you need do it only once. Thereafter, the shuffling can be incredibly cache-friendly. Regards, Nick Maclaren, University of Cambridge Computing Service, New Museums Site, Pembroke Street, Cambridge CB2 3QH, England. Email: nmm1@xxxxxxxxx Tel.: +44 1223 334761 Fax: +44 1223 334679
