Arm32 and arm64 can spill to FP registers, thanks to this code:

ifdef ARM
     // ARM has support for moving 64bit values between a pair of
     // integer registers and a double register
     idealreg2spillmask[Op_RegL]->OR(*idealreg2regmask[Op_RegD]);
     idealreg2spillmask[Op_RegD]->OR(*idealreg2regmask[Op_RegL]);
#endif

however, this increases and FP register pressure and VMOV is slower than a load or store on arm64 (and probably armv7 arm32) because of the FP pipeline.