> I don't understand this proposed change. float is a 32-bit
> datatype, double is a 64-bit datatype. My understanding is that on
> ARM64 floating-point arguments are passed as doubles. What am I
> missing?
What I was missing was the rationale for using only even numbered array
indices for floatRegisters in
ThreadedARM{32,64}FFIPlugin>>ffiCalloutTo:SpecOnStack:in:
I am following the pattern of ThreadedARM32FFIPlugin:
- floats get assigned to the register array slots
[#ffiPushSingleFloat:in:]
- doubles get mem-copied to the register array slots
[#ffiPushDoubleFloat:in:]
C casting tricks are then used to interpret memory properly for FFI
usage.
One could do the other way around.
Case of 6 of one, half a dozen of the other.
This is why we need to leave mouse tracks (er, document) our artifacts.
Hey, this confused the heck out of me!
-KenD
-KenD
Locked
