The rs6000 port currently has an *lt0_disi define_insn, setting the DI
result to whether the SI argument is negative or not. It turns out the
generic optimisers cannot always figure out in the other cases either
that this is just a shift for us. This patch adds patterns for all
four SI/DI combinations.
Tested on powerpc64-linux {-m32,-m64}; committing to trunk.
Segher
2017-11-03 Segher Boessenkool <seg...@kernel.crashing.org>
* config/rs6000/rs6000.md (*lt0_disi): Delete.
(*lt0_<mode>di, *lt0_<mode>si): New.
---
gcc/config/rs6000/rs6000.md | 16 ++++++++++++----
1 file changed, 12 insertions(+), 4 deletions(-)
diff --git a/gcc/config/rs6000/rs6000.md b/gcc/config/rs6000/rs6000.md
index 18ebe8f..ff79f2d 100644
--- a/gcc/config/rs6000/rs6000.md
+++ b/gcc/config/rs6000/rs6000.md
@@ -3829,11 +3829,19 @@ (define_insn_and_split "*rotl<mode>3_mask_dot2"
; Special case for less-than-0. We can do it with just one machine
; instruction, but the generic optimizers do not realise it is cheap.
-(define_insn "*lt0_disi"
- [(set (match_operand:DI 0 "gpc_reg_operand" "=r")
- (lt:DI (match_operand:SI 1 "gpc_reg_operand" "r")
- (const_int 0)))]
+(define_insn "*lt0_<mode>di"
+ [(set (match_operand:GPR 0 "gpc_reg_operand" "=r")
+ (lt:GPR (match_operand:DI 1 "gpc_reg_operand" "r")
+ (const_int 0)))]
"TARGET_POWERPC64"
+ "srdi %0,%1,63"
+ [(set_attr "type" "shift")])
+
+(define_insn "*lt0_<mode>si"
+ [(set (match_operand:GPR 0 "gpc_reg_operand" "=r")
+ (lt:GPR (match_operand:SI 1 "gpc_reg_operand" "r")
+ (const_int 0)))]
+ ""
"rlwinm %0,%1,1,31,31"
[(set_attr "type" "shift")])
--
1.8.3.1
