I need to use the BFI_INT instruction directly from some optimization. According to the Evergreen docs it performs:
dst = (src1 & src0) | (src2 & ~src0)
But the closest I can find in the IL docs are UBIT_INSERT.
Alas ubit_insert doesn't allow me to specify the bitmask src0 directly, only to the level of widths and offsets. So if I want a mask such as 0x5555aaaa ubit_insert is no use.
Have I overlooked something in the docs? Or is there an undocumented feature I can use to utilize this assembly level operation directly ?