In vasm copy, in order to move a general purpose register to a floating point register two memory access are being done: one to store an integer value in memory, and the other to load this value as a floating point value. A similar approach is being done in the opposite case, moving a floating point value to a general purpose value.
See:
https://github.com/PPC64/hhvm/blob/next/hphp/runtime/vm/jit/vasm-ppc64.cpp#L119
https://github.com/PPC64/hhvm/blob/next/hphp/runtime/vm/jit/vasm-ppc64.cpp#L125
To avoid memory access, it is possible to use an auxiliary vector register and move from a general purpose register to a vector register and then from the vector register to a floating point register (also valid for moving from a floating point register to a general purpose register). This may be accomplished using mtvsrd and mfvsrd.
In vasm copy, in order to move a general purpose register to a floating point register two memory access are being done: one to store an integer value in memory, and the other to load this value as a floating point value. A similar approach is being done in the opposite case, moving a floating point value to a general purpose value.
See:
https://github.com/PPC64/hhvm/blob/next/hphp/runtime/vm/jit/vasm-ppc64.cpp#L119
https://github.com/PPC64/hhvm/blob/next/hphp/runtime/vm/jit/vasm-ppc64.cpp#L125
To avoid memory access, it is possible to use an auxiliary vector register and move from a general purpose register to a vector register and then from the vector register to a floating point register (also valid for moving from a floating point register to a general purpose register). This may be accomplished using mtvsrd and mfvsrd.