Yes that is the reason for the original bug but I'm asking about the motivation of the original patch. Why is #0 being special cased when there is code in the ARM backend to generate fconstd's (which is the older syntax for the vmov.f64 instruction). As I said please investigate why that's not getting triggered ? I suspect it is because of a splitter in the backend.
Hi Chung-Lin,
Yes that is the reason for the original bug but I'm asking about the motivation of the original patch. Why is #0 being special cased when there is code in the ARM backend to generate fconstd's (which is the older syntax for the vmov.f64 instruction). As I said please investigate why that's not getting triggered ? I suspect it is because of a splitter in the backend.
cheers
Ramana