[llvm-commits] [llvm] r109696 - /llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
Jim Grosbach
grosbach at apple.com
Wed Jul 28 16:25:44 PDT 2010
Author: grosbach
Date: Wed Jul 28 18:25:44 2010
New Revision: 109696
URL: http://llvm.org/viewvc/llvm-project?rev=109696&view=rev
Log:
ARM mode version of r109693. Remove incorrect substitution pattern for UXTB16. It wrongly assumed the input shift was actually a rotate. rdar://8240138
Modified:
llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
Modified: llvm/trunk/lib/Target/ARM/ARMInstrInfo.td
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMInstrInfo.td?rev=109696&r1=109695&r2=109696&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMInstrInfo.td (original)
+++ llvm/trunk/lib/Target/ARM/ARMInstrInfo.td Wed Jul 28 18:25:44 2010
@@ -1567,8 +1567,12 @@
defm UXTB16 : AI_unary_rrot<0b01101100,
"uxtb16", UnOpFrag<(and node:$Src, 0x00FF00FF)>>;
-def : ARMV6Pat<(and (shl GPR:$Src, (i32 8)), 0xFF00FF),
- (UXTB16r_rot GPR:$Src, 24)>;
+// FIXME: This pattern incorrectly assumes the shl operator is a rotate.
+// The transformation should probably be done as a combiner action
+// instead so we can include a check for masking back in the upper
+// eight bits of the source into the lower eight bits of the result.
+//def : ARMV6Pat<(and (shl GPR:$Src, (i32 8)), 0xFF00FF),
+// (UXTB16r_rot GPR:$Src, 24)>;
def : ARMV6Pat<(and (srl GPR:$Src, (i32 8)), 0xFF00FF),
(UXTB16r_rot GPR:$Src, 8)>;
More information about the llvm-commits
mailing list