[planned patch] Rotation insns on the H8/300
Tue Aug 8 15:14:00 GMT 2000
I am implementing rotate insns (rotlqi, etc) for the H8/300 target.
Things have been OK so far.
Would the following method be acceptable (to the CVS tree)?
For a rotation by a non-constant, at the RTL generation time, we emit
a loop, repeating a bunch of one-bit rotation. For a rotaion by a
constant, we don't do anything at this time.
At the insn emit time, we are only left with a rotation by a constant.
we emit all assembly insns just like the shift insns are emitted on
the H8/300 port.
As far as the algorithm for a rotation by a constant amount, I am
planning to use a byte swap and a word swap as well as normal rotation
insns. (That is, a endian conversion in HImode ends up with a byte
swap, which was the only thing I wanted to do at first. :-)
More information about the Gcc-patches