On 07/30/2013 12:51 PM, DJ Delorie wrote:
First, if you can handle ASHIFT directly in PSImode, that's advantageous
for address calculations.
Yup, I have patterns for 1 and 2 bit shifts, which are all that really
happen. They look remarkably like your patterns, but use a subreg on
0 instead of a truncate on the expression (again, that's what combine
was looking for, so that's what I put in).
And to that I would say, let's evaluate why combine was using the subreg
rather than a truncate. That seems wrong if we've defined
TRULY_NOOP_TRUNCATION properly.
jeff