Thanks, this is almost there now. It was only the problem with the new version of the move pattern (see below) that stopped this from being "OK with...". The next round should be a formality though.
"Moore, Catherine" <catherine_mo...@mentor.com> writes: > +(define_constraint "Uuw6" > + "@internal > + An unsigned constant of 6 bits." > + (match_operand 0 "uw6_operand")) ", shifted left two places". > + (and (ior (eq_attr "compression" "micromips") > + (eq_attr "compression" "all")) Please use (eq_attr "compression" "micromips,all") instead. > (define_insn "sub<mode>3" > - [(set (match_operand:GPR 0 "register_operand" "=d") > - (minus:GPR (match_operand:GPR 1 "register_operand" "d") > - (match_operand:GPR 2 "register_operand" "d")))] > + [(set (match_operand:GPR 0 "register_operand" "=!u,d") > + (minus:GPR (match_operand:GPR 1 "register_operand" "!u,d") > + (match_operand:GPR 2 "register_operand" "!u,d")))] > "" > - "<d>subu\t%0,%1,%2" > + "@ > + <d>subu\t%0,%1,%2 > + <d>subu\t%0,%1,%2" This change isn't needed. It's OK (and IMO better) to keep a single asm string when the string is the same for all alternatives. > @@ -4362,13 +4400,14 @@ > ;; in FP registers (off by default, use -mdebugh to enable). > > (define_insn "*mov<mode>_internal" > - [(set (match_operand:IMOVE32 0 "nonimmediate_operand" > "=d,d,e,d,m,*f,*f,*d,*m,*d,*z,*a,*d,*B*C*D,*B*C*D,*d,*m") > - (match_operand:IMOVE32 1 "move_operand" > "d,Yd,Yf,m,dJ,*d*J,*m,*f,*f,*z,*d,*J*d,*a,*d,*m,*B*C*D,*B*C*D"))] > + [(set (match_operand:IMOVE32 0 "nonimmediate_operand" > "=d,!u,d,e,!u,!u,!u,!u,!ks,d,ZS,ZV,ZU,ZT,m,*f,*f,*d,*m,*d,*z,*a,*d,*B*C*D,*B*C*D,*d,*m") > + (match_operand:IMOVE32 1 "move_operand" > "d,J,Yd,Yf,Udb7,ZW,ZU,ZT,ZS,m,!ks,!u,!u,!u,dJ,*d*J,*m,*f,*f,*z,*d,*J*d,*a,*d,*m,*B*C*D,*B*C*D"))] This pattern is only for 32-bit moves, so should only use ZS and ZT. (I don't mind keeping the definitions of the ZU...ZW constraints in this patch -- even though they start out unused -- because it's obvious they'll be needed eventually. I'd prefer leaving the 8-bit and 16-move patterns themselves to a different patch though.) Sorry, I should have noticed last time, but alternative 5 (u<-Udb7) should come before alternative 3 (d<-Yd), because d<-Yd includes everything that u<-Udb7 does. > +extern bool m16_based_address_p (rtx, enum machine_mode, int (*)(rtx_def*, > machine_mode)); Line too long: should be split after the first enum machine_mode. Thanks, Richard
