------- Comment #15 from amodra at bigpond dot net dot au 2006-03-24 14:23 ------- Appears to be the peephole in rs6000.md immediately after this comment: ;; after inserting conditional returns we can sometimes have ;; unnecessary register moves.
(subreg:DF (reg:DI 3 3 [128]) 0) is a valid gpc_reg_operand. Try this (totally untested) patch. Index: gcc/config/rs6000/rs6000.md =================================================================== --- gcc/config/rs6000/rs6000.md (revision 112347) +++ gcc/config/rs6000/rs6000.md (working copy) @@ -9771,18 +9771,20 @@ "operands[0] = widen_memory_access (operands[0], V2DFmode, 0); operands[1] = gen_rtx_REG (V2DFmode, REGNO (operands[1]));") -;; after inserting conditional returns we can sometimes have +;; After inserting conditional returns we can sometimes have ;; unnecessary register moves. Unfortunately we cannot have a ;; modeless peephole here, because some single SImode sets have early ;; clobber outputs. Although those sets expand to multi-ppc-insn ;; sequences, using get_attr_length here will smash the operands ;; array. Neither is there an early_cobbler_p predicate. +;; Disallow subregs for E500 so we don't munge frob_di_df_2. (define_peephole2 [(set (match_operand:DF 0 "gpc_reg_operand" "") (match_operand:DF 1 "any_operand" "")) (set (match_operand:DF 2 "gpc_reg_operand" "") (match_dup 0))] - "peep2_reg_dead_p (2, operands[0])" + "!(TARGET_E500_DOUBLE && GET_CODE (operands[2]) == SUBREG) + && peep2_reg_dead_p (2, operands[0])" [(set (match_dup 2) (match_dup 1))]) (define_peephole2 -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26459