This patch just changes a QImode (const_int 255) to (const_int -1),
since the canonical form is to sign-extend. As things stand the pattern
trips a new assert added on the wide-int branch.
Tested by building rl78-elf before and after the patch and making sure
that there were no changes in assembly code for gcc.c-torture, gcc.dg
and g++dg. OK to install?
Thanks,
Richard
gcc/
* config/rl78/rl78-expand.md (one_cmplqi2): Use (const_int -1)
rather than (const_int 255).
Index: gcc/config/rl78/rl78-expand.md
===================================================================
--- gcc/config/rl78/rl78-expand.md 2013-10-15 19:24:15.821884272 +0100
+++ gcc/config/rl78/rl78-expand.md 2013-11-07 09:41:50.754226143 +0000
@@ -177,7 +177,7 @@ (define_expand "xorqi3"
(define_expand "one_cmplqi2"
[(set (match_operand:QI 0 "nonimmediate_operand")
(xor:QI (match_operand:QI 1 "general_operand")
- (const_int 255)))
+ (const_int -1)))
]
""
"if (rl78_force_nonfar_2 (operands, gen_one_cmplqi2))
