[PowerPC] Remove zexts after byte-swapping loads

lhbrx and lwbrx not only load their data with byte swapping, but also clear the
upper 32 bits (at least). As a result, they can be added to the PPCISelDAGToDAG
peephole optimization as frontier instructions for the removal of unnecessary
zero extensions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@225189 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/PowerPC/PPCISelDAGToDAG.cpp b/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
index 7d7f557e433022748f61e074934efdccb85324e7..3e787bedd279d81363ffba7ff5c8c6254db03993 100644 (file)
--- a/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
+++ b/lib/Target/PowerPC/PPCISelDAGToDAG.cpp
@@ -3729,6 +3729,13 @@ static bool PeepholePPC64ZExtGather(SDValue Op32,
     return true;
   }
 
+  // LHBRX and LWBRX always clear the higher-order bits.
+  if (Op32.getMachineOpcode() == PPC::LHBRX ||
+      Op32.getMachineOpcode() == PPC::LWBRX) {
+    ToPromote.insert(Op32.getNode());
+    return true;
+  }
+
   // Next, check for those instructions we can look through.
 
   // Assuming the mask does not wrap around, then the higher-order bits are
@@ -3916,6 +3923,8 @@ void PPCDAGToDAGISel::PeepholePPC64ZExt() {
       case PPC::SRW:       NewOpcode = PPC::SRW8; break;
       case PPC::LI:        NewOpcode = PPC::LI8; break;
       case PPC::LIS:       NewOpcode = PPC::LIS8; break;
+      case PPC::LHBRX:     NewOpcode = PPC::LHBRX8; break;
+      case PPC::LWBRX:     NewOpcode = PPC::LWBRX8; break;
       case PPC::RLWIMI:    NewOpcode = PPC::RLWIMI8; break;
       case PPC::OR:        NewOpcode = PPC::OR8; break;
       case PPC::SELECT_I4: NewOpcode = PPC::SELECT_I8; break;

diff --git a/lib/Target/PowerPC/PPCInstr64Bit.td b/lib/Target/PowerPC/PPCInstr64Bit.td
index 75f4656d01b8d7e9e2b77715b8fe86fc74e3c9d6..29b13f8a53addbe9fb089285635921568b728ef8 100644 (file)
--- a/lib/Target/PowerPC/PPCInstr64Bit.td
+++ b/lib/Target/PowerPC/PPCInstr64Bit.td
@@ -843,6 +843,13 @@ def LDBRX : XForm_1<31,  532, (outs g8rc:$rD), (ins memrr:$src),
                    "ldbrx $rD, $src", IIC_LdStLoad,
                    [(set i64:$rD, (PPClbrx xoaddr:$src, i64))]>, isPPC64;
 
+let mayLoad = 1, hasSideEffects = 0, isCodeGenOnly = 1 in {
+def LHBRX8 : XForm_1<31, 790, (outs g8rc:$rD), (ins memrr:$src),
+                   "lhbrx $rD, $src", IIC_LdStLoad, []>;
+def LWBRX8 : XForm_1<31,  534, (outs g8rc:$rD), (ins memrr:$src),
+                   "lwbrx $rD, $src", IIC_LdStLoad, []>;
+}
+
 let mayLoad = 1, hasSideEffects = 0 in {
 def LDU  : DSForm_1<58, 1, (outs g8rc:$rD, ptr_rc_nor0:$ea_result), (ins memrix:$addr),
                     "ldu $rD, $addr", IIC_LdStLDU,

diff --git a/test/CodeGen/PowerPC/rm-zext.ll b/test/CodeGen/PowerPC/rm-zext.ll
index 8fab38ad4cf0867960ce09ffcf2ff5039503e2fe..291b7919280def4014f5de76e55ca5a4d9aa242c 100644 (file)
--- a/test/CodeGen/PowerPC/rm-zext.ll
+++ b/test/CodeGen/PowerPC/rm-zext.ll
@@ -39,5 +39,35 @@ entry:
 ; CHECK: blr
 }
 
+; Function Attrs: nounwind readnone
+declare i32 @llvm.bswap.i32(i32) #1
+
+; Function Attrs: nounwind readonly
+define zeroext i32 @bs32(i32* nocapture readonly %x) #0 {
+entry:
+  %0 = load i32* %x, align 4
+  %1 = tail call i32 @llvm.bswap.i32(i32 %0)
+  ret i32 %1
+
+; CHECK-LABEL: @bs32
+; CHECK-NOT: rldicl 3, {{[0-9]+}}, 0, 32
+; CHECK: blr
+}
+
+; Function Attrs: nounwind readonly
+define zeroext i16 @bs16(i16* nocapture readonly %x) #0 {
+entry:
+  %0 = load i16* %x, align 2
+  %1 = tail call i16 @llvm.bswap.i16(i16 %0)
+  ret i16 %1
+
+; CHECK-LABEL: @bs16
+; CHECK-NOT: rldicl 3, {{[0-9]+}}, 0, 32
+; CHECK: blr
+}
+
+; Function Attrs: nounwind readnone
+declare i16 @llvm.bswap.i16(i16) #1
+
 attributes #0 = { nounwind readnone }