[llvm] r191770 - [SystemZ] Extend 32-bit RISBG optimizations to high words

[Richard Sandiford]

Author: rsandifo
Date: Tue Oct  1 09:36:20 2013
New Revision: 191770

URL: http://llvm.org/viewvc/llvm-project?rev=191770&view=rev
Log:
[SystemZ] Extend 32-bit RISBG optimizations to high words

This involves using RISB[LH]G, whereas the equivalent z10 optimization
uses RISBG.

Modified:
    llvm/trunk/lib/Target/SystemZ/SystemZISelDAGToDAG.cpp
    llvm/trunk/test/CodeGen/SystemZ/asm-18.ll

Modified: llvm/trunk/lib/Target/SystemZ/SystemZISelDAGToDAG.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/SystemZ/SystemZISelDAGToDAG.cpp?rev=191770&r1=191769&r2=191770&view=diff
==============================================================================
--- llvm/trunk/lib/Target/SystemZ/SystemZISelDAGToDAG.cpp (original)
+++ llvm/trunk/lib/Target/SystemZ/SystemZISelDAGToDAG.cpp Tue Oct  1 09:36:20 2013
@@ -266,8 +266,8 @@ class SystemZDAGToDAGISel : public Selec
   // Return true on success.
   bool expandRxSBG(RxSBGOperands &RxSBG) const;
 
-  // Return an undefined i64 value.
-  SDValue getUNDEF64(SDLoc DL) const;
+  // Return an undefined value of type VT.
+  SDValue getUNDEF(SDLoc DL, EVT VT) const;
 
   // Convert N to VT, if it isn't already.
   SDValue convertTo(SDLoc DL, EVT VT, SDValue N) const;
@@ -832,15 +832,15 @@ bool SystemZDAGToDAGISel::expandRxSBG(Rx
   }
 }
 
-SDValue SystemZDAGToDAGISel::getUNDEF64(SDLoc DL) const {
-  SDNode *N = CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, DL, MVT::i64);
+SDValue SystemZDAGToDAGISel::getUNDEF(SDLoc DL, EVT VT) const {
+  SDNode *N = CurDAG->getMachineNode(TargetOpcode::IMPLICIT_DEF, DL, VT);
   return SDValue(N, 0);
 }
 
 SDValue SystemZDAGToDAGISel::convertTo(SDLoc DL, EVT VT, SDValue N) const {
   if (N.getValueType() == MVT::i32 && VT == MVT::i64)
     return CurDAG->getTargetInsertSubreg(SystemZ::subreg_l32,
-                                         DL, VT, getUNDEF64(DL), N);
+                                         DL, VT, getUNDEF(DL, MVT::i64), N);
   if (N.getValueType() == MVT::i64 && VT == MVT::i32)
     return CurDAG->getTargetExtractSubreg(SystemZ::subreg_l32, DL, VT, N);
   assert(N.getValueType() == VT && "Unexpected value types");
@@ -880,14 +880,22 @@ SDNode *SystemZDAGToDAGISel::tryRISBGZer
     }
   }  
 
+  unsigned Opcode = SystemZ::RISBG;
+  EVT OpcodeVT = MVT::i64;
+  if (VT == MVT::i32 && Subtarget.hasHighWord()) {
+    Opcode = SystemZ::RISBMux;
+    OpcodeVT = MVT::i32;
+    RISBG.Start &= 31;
+    RISBG.End &= 31;
+  }
   SDValue Ops[5] = {
-    getUNDEF64(SDLoc(N)),
-    convertTo(SDLoc(N), MVT::i64, RISBG.Input),
+    getUNDEF(SDLoc(N), OpcodeVT),
+    convertTo(SDLoc(N), OpcodeVT, RISBG.Input),
     CurDAG->getTargetConstant(RISBG.Start, MVT::i32),
     CurDAG->getTargetConstant(RISBG.End | 128, MVT::i32),
     CurDAG->getTargetConstant(RISBG.Rotate, MVT::i32)
   };
-  N = CurDAG->getMachineNode(SystemZ::RISBG, SDLoc(N), MVT::i64, Ops);
+  N = CurDAG->getMachineNode(Opcode, SDLoc(N), OpcodeVT, Ops);
   return convertTo(SDLoc(N), VT, SDValue(N, 0)).getNode();
 }
 

Modified: llvm/trunk/test/CodeGen/SystemZ/asm-18.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/SystemZ/asm-18.ll?rev=191770&r1=191769&r2=191770&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/SystemZ/asm-18.ll (original)
+++ llvm/trunk/test/CodeGen/SystemZ/asm-18.ll Tue Oct  1 09:36:20 2013
@@ -500,3 +500,28 @@ define i32 @f23(i32 %old) {
                        "=r,h,h,0"(i32 %res1, i32 %and3, i32 %and4)
   ret i32 %res2
 }
+
+; Test RISB[LH]G insertions involving mixtures of high and low registers.
+define i32 @f24(i32 %old) {
+; CHECK-LABEL: f24:
+; CHECK-DAG: risblg [[REG1:%r[0-5]]], %r2, 28, 158, 1
+; CHECK-DAG: risbhg [[REG2:%r[0-5]]], %r2, 24, 158, 29
+; CHECK: stepa %r2, [[REG1]], [[REG2]]
+; CHECK-DAG: risbhg [[REG3:%r[0-5]]], [[REG2]], 25, 159, 62
+; CHECK-DAG: risblg %r2, [[REG2]], 24, 152, 37
+; CHECK: stepb [[REG2]], [[REG3]], %r2
+; CHECK: br %r14
+  %shift1 = shl i32 %old, 1
+  %and1 = and i32 %shift1, 14
+  %shift2 = lshr i32 %old, 3
+  %and2 = and i32 %shift2, 254
+  %res1 = call i32 asm "stepa $1, $2, $3",
+                       "=h,r,r,0"(i32 %old, i32 %and1, i32 %and2)
+  %shift3 = lshr i32 %res1, 2
+  %and3 = and i32 %shift3, 127
+  %shift4 = shl i32 %res1, 5
+  %and4 = and i32 %shift4, 128
+  %res2 = call i32 asm "stepb $1, $2, $3",
+                       "=r,h,h,0"(i32 %res1, i32 %and3, i32 %and4)
+  ret i32 %res2
+}