[PATCH] D39510: [PPC] Use xxbrd to speed up bswap64

[Guozhi Wei via Phabricator via llvm-commits]

This revision was automatically updated to reflect the committed changes.
Closed by commit rL317499: [PPC] Use xxbrd to speed up bswap64 (authored by Carrot).

Changed prior to commit:
  https://reviews.llvm.org/D39510?vs=121343&id=121759#toc

Repository:
  rL LLVM

https://reviews.llvm.org/D39510

Files:
  llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp
  llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
  llvm/trunk/test/CodeGen/PowerPC/bswap64.ll


Index: llvm/trunk/test/CodeGen/PowerPC/bswap64.ll
===================================================================
--- llvm/trunk/test/CodeGen/PowerPC/bswap64.ll
+++ llvm/trunk/test/CodeGen/PowerPC/bswap64.ll
@@ -0,0 +1,13 @@
+; RUN: llc -verify-machineinstrs < %s -mtriple=ppc64le-- -mcpu=pwr9 | FileCheck %s
+
+declare i64 @llvm.bswap.i64(i64)
+
+; CHECK: mtvsrdd
+; CHECK: xxbrd
+; CHECK: mfvsrd
+define i64 @bswap64(i64 %x) {
+entry:
+  %0 = call i64 @llvm.bswap.i64(i64 %x)
+  ret i64 %0
+}
+
Index: llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp
===================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp
+++ llvm/trunk/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -291,14 +291,16 @@
     setOperationAction(ISD::FROUND, MVT::f32, Legal);
   }
 
-  // PowerPC does not have BSWAP
+  // PowerPC does not have BSWAP, but we can use vector BSWAP instruction xxbrd
+  // to speed up scalar BSWAP64.
   // CTPOP or CTTZ were introduced in P8/P9 respectivelly
   setOperationAction(ISD::BSWAP, MVT::i32  , Expand);
-  setOperationAction(ISD::BSWAP, MVT::i64  , Expand);
   if (Subtarget.isISA3_0()) {
+    setOperationAction(ISD::BSWAP, MVT::i64  , Custom);
     setOperationAction(ISD::CTTZ , MVT::i32  , Legal);
     setOperationAction(ISD::CTTZ , MVT::i64  , Legal);
   } else {
+    setOperationAction(ISD::BSWAP, MVT::i64  , Expand);
     setOperationAction(ISD::CTTZ , MVT::i32  , Expand);
     setOperationAction(ISD::CTTZ , MVT::i64  , Expand);
   }
@@ -8675,6 +8677,23 @@
   return Op;
 }
 
+// Lower scalar BSWAP64 to xxbrd.
+SDValue PPCTargetLowering::LowerBSWAP(SDValue Op, SelectionDAG &DAG) const {
+  SDLoc dl(Op);
+  // MTVSRDD
+  Op = DAG.getNode(ISD::BUILD_VECTOR, dl, MVT::v2i64, Op.getOperand(0),
+                   Op.getOperand(0));
+  // XXBRD
+  Op = DAG.getNode(PPCISD::XXREVERSE, dl, MVT::v2i64, Op);
+  // MFVSRD
+  int VectorIndex = 0;
+  if (Subtarget.isLittleEndian())
+    VectorIndex = 1;
+  Op = DAG.getNode(ISD::EXTRACT_VECTOR_ELT, dl, MVT::i64, Op,
+                   DAG.getTargetConstant(VectorIndex, dl, MVT::i32));
+  return Op;
+}
+
 SDValue PPCTargetLowering::LowerSIGN_EXTEND_INREG(SDValue Op,
                                                   SelectionDAG &DAG) const {
   SDLoc dl(Op);
@@ -9146,6 +9165,8 @@
   case ISD::SREM:
   case ISD::UREM:
     return LowerREM(Op, DAG);
+  case ISD::BSWAP:
+    return LowerBSWAP(Op, DAG);
   }
 }
 
Index: llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
===================================================================
--- llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
+++ llvm/trunk/lib/Target/PowerPC/PPCISelLowering.h
@@ -953,6 +953,7 @@
     SDValue LowerINTRINSIC_WO_CHAIN(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerINTRINSIC_VOID(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerREM(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerBSWAP(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSCALAR_TO_VECTOR(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerMUL(SDValue Op, SelectionDAG &DAG) const;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D39510.121759.patch
Type: text/x-patch
Size: 3175 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20171106/062f9315/attachment.bin>