From d36d30b53c89a1bccde33373021e6cb0a01b8a6a Mon Sep 17 00:00:00 2001
From: Christopher Lamb <christopher.lamb@gmail.com>
Date: Fri, 10 Aug 2007 21:18:25 +0000
Subject: [PATCH] Add 2-addr to 3-addr promotion code that allows 32-bit LEA to
 be used via subregisters when 16-bit LEA is disabled.

llvm-svn: 41007
---
 llvm/lib/Target/X86/X86InstrInfo.cpp          | 57 +++++++++++++++----
 .../test/CodeGen/X86/2007-08-10-LEA16Use32.ll | 23 ++++++++
 2 files changed, 70 insertions(+), 10 deletions(-)
 create mode 100644 llvm/test/CodeGen/X86/2007-08-10-LEA16Use32.ll

diff --git a/llvm/lib/Target/X86/X86InstrInfo.cpp b/llvm/lib/Target/X86/X86InstrInfo.cpp
index d0a0a093bf3e..1dbbce2bb314 100644
--- a/llvm/lib/Target/X86/X86InstrInfo.cpp
+++ b/llvm/lib/Target/X86/X86InstrInfo.cpp
@@ -19,6 +19,7 @@
 #include "X86TargetMachine.h"
 #include "llvm/CodeGen/MachineInstrBuilder.h"
 #include "llvm/CodeGen/LiveVariables.h"
+#include "llvm/CodeGen/SSARegMap.h"
 using namespace llvm;
 
 X86InstrInfo::X86InstrInfo(X86TargetMachine &tm)
@@ -209,17 +210,53 @@ X86InstrInfo::convertToThreeAddress(MachineFunction::iterator &MFI,
   }
   case X86::SHL16ri: {
     assert(MI->getNumOperands() == 3 && "Unknown shift instruction!");
-    if (DisableLEA16) return 0;
+      // NOTE: LEA doesn't produce flags like shift does, but LLVM never uses
+      // the flags produced by a shift yet, so this is safe.
+      unsigned Dest = MI->getOperand(0).getReg();
+      unsigned Src = MI->getOperand(1).getReg();
+      unsigned ShAmt = MI->getOperand(2).getImm();
+      if (ShAmt == 0 || ShAmt >= 4) return 0;
     
-    // NOTE: LEA doesn't produce flags like shift does, but LLVM never uses
-    // the flags produced by a shift yet, so this is safe.
-    unsigned Dest = MI->getOperand(0).getReg();
-    unsigned Src = MI->getOperand(1).getReg();
-    unsigned ShAmt = MI->getOperand(2).getImm();
-    if (ShAmt == 0 || ShAmt >= 4) return 0;
-    
-    NewMI = BuildMI(get(X86::LEA16r), Dest)
-      .addReg(0).addImm(1 << ShAmt).addReg(Src).addImm(0);
+    if (DisableLEA16) {
+      // If 16-bit LEA is disabled, use 32-bit LEA via subregisters.
+      SSARegMap *RegMap = MFI->getParent()->getSSARegMap();
+      unsigned Opc, leaInReg, leaOutReg;
+      MVT::ValueType leaVT;
+      if (TM.getSubtarget<X86Subtarget>().is64Bit()) {
+        Opc = X86::LEA64_32r;
+        leaVT = MVT::i64;
+        leaInReg = RegMap->createVirtualRegister(&X86::GR64RegClass);
+        leaOutReg = RegMap->createVirtualRegister(&X86::GR64RegClass);
+      } else {
+        Opc = X86::LEA32r;
+        leaVT = MVT::i32;
+        leaInReg = RegMap->createVirtualRegister(&X86::GR32RegClass);
+        leaOutReg = RegMap->createVirtualRegister(&X86::GR32RegClass);
+      }
+            
+      MachineInstr *Ins = NULL, *Ext = NULL;
+        
+      Ins = BuildMI(get(X86::INSERT_SUBREG), leaInReg).addReg(Src).addImm(2);
+      Ins->copyKillDeadInfo(MI);
+      
+      NewMI = BuildMI(get(Opc), leaOutReg)
+        .addReg(0).addImm(1 << ShAmt).addReg(leaInReg).addImm(0);
+      
+      Ext = BuildMI(get(X86::EXTRACT_SUBREG), Dest).addReg(leaOutReg).addImm(2);
+      Ext->copyKillDeadInfo(MI);
+      
+      MFI->insert(MBBI, Ins);            // Insert the insert_subreg
+      LV.instructionChanged(MI, NewMI);  // Update live variables
+      LV.addVirtualRegisterKilled(leaInReg, NewMI);
+      MFI->insert(MBBI, NewMI);          // Insert the new inst
+      LV.addVirtualRegisterKilled(leaOutReg, Ext);
+      MFI->insert(MBBI, Ext);            // Insert the extract_subreg
+      
+      return Ext;
+    } else {
+      NewMI = BuildMI(get(X86::LEA16r), Dest)
+        .addReg(0).addImm(1 << ShAmt).addReg(Src).addImm(0);
+    }
     break;
   }
   }
diff --git a/llvm/test/CodeGen/X86/2007-08-10-LEA16Use32.ll b/llvm/test/CodeGen/X86/2007-08-10-LEA16Use32.ll
new file mode 100644
index 000000000000..5bb696f1ea66
--- /dev/null
+++ b/llvm/test/CodeGen/X86/2007-08-10-LEA16Use32.ll
@@ -0,0 +1,23 @@
+; RUN: llvm-as < %s | llc -march=x86 | grep {leal}
+
+@X = global i16 0               ; <i16*> [#uses=1]
+@Y = global i16 0               ; <i16*> [#uses=1]
+
+define void @_Z3fooi(i32 %N) {
+entry:
+        %tmp1019 = icmp sgt i32 %N, 0           ; <i1> [#uses=1]
+        br i1 %tmp1019, label %bb, label %return
+
+bb:             ; preds = %bb, %entry
+        %i.014.0 = phi i32 [ 0, %entry ], [ %indvar.next, %bb ]         ; <i32> [#uses=2]
+        %tmp1 = trunc i32 %i.014.0 to i16               ; <i16> [#uses=2]
+        volatile store i16 %tmp1, i16* @X, align 2
+        %tmp34 = shl i16 %tmp1, 2               ; <i16> [#uses=1]
+        volatile store i16 %tmp34, i16* @Y, align 2
+        %indvar.next = add i32 %i.014.0, 1              ; <i32> [#uses=2]
+        %exitcond = icmp eq i32 %indvar.next, %N                ; <i1> [#uses=1]
+        br i1 %exitcond, label %return, label %bb
+
+return:         ; preds = %bb, %entry
+        ret void
+}
\ No newline at end of file