Uros Bizjak - Re: [PATCH] Force DFP function args into FP registers per the ABI (original) (raw)

This is the mail archive of the gcc-patches@gcc.gnu.orgmailing list for the GCC project.

A later follow on patch will fix up SDmode arguments too. It turns out that SDmode is harder, since (except for POWER6), there is no instruction that will load/store a 32-bit memory location into a FP register without munging the bits (the load/store float insns do an implicit float to/from double conversion since all binary floating values are stored in double precision format within the hardware registers).