PR7458: Try commuting Thumb2 instruction operands to put them into 2-address

form so they can be narrowed to 16-bit instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106762 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/ARM/Thumb2SizeReduction.cpp b/lib/Target/ARM/Thumb2SizeReduction.cpp
index 4ad21c49a8aec547e7045c5ee4c7a77782058c9c..ba392f36d946424fab03cc0fbc615b943dab1acb 100644 (file)
--- a/lib/Target/ARM/Thumb2SizeReduction.cpp
+++ b/lib/Target/ARM/Thumb2SizeReduction.cpp
@@ -451,11 +451,18 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
   if (ReduceLimit2Addr != -1 && ((int)Num2Addrs >= ReduceLimit2Addr))
     return false;
 
-  const TargetInstrDesc &TID = MI->getDesc();
   unsigned Reg0 = MI->getOperand(0).getReg();
   unsigned Reg1 = MI->getOperand(1).getReg();
-  if (Reg0 != Reg1)
-    return false;
+  if (Reg0 != Reg1) {
+    // Try to commute the operands to make it a 2-address instruction.
+    unsigned CommOpIdx1, CommOpIdx2;
+    if (!TII->findCommutedOpIndices(MI, CommOpIdx1, CommOpIdx2) ||
+        CommOpIdx1 != 1 || MI->getOperand(CommOpIdx2).getReg() != Reg0)
+      return false;
+    MachineInstr *CommutedMI = TII->commuteInstruction(MI);
+    if (!CommutedMI)
+      return false;
+  }
   if (Entry.LowRegs2 && !isARMLowRegister(Reg0))
     return false;
   if (Entry.Imm2Limit) {
@@ -484,6 +491,7 @@ Thumb2SizeReduce::ReduceTo2Addr(MachineBasicBlock &MBB, MachineInstr *MI,
 
   bool HasCC = false;
   bool CCDead = false;
+  const TargetInstrDesc &TID = MI->getDesc();
   if (TID.hasOptionalDef()) {
     unsigned NumOps = TID.getNumOperands();
     HasCC = (MI->getOperand(NumOps-1).getReg() == ARM::CPSR);

diff --git a/test/CodeGen/Thumb2/thumb2-eor.ll b/test/CodeGen/Thumb2/thumb2-eor.ll
index b7e276673c4219bce725c399393999e1b191fb08..116a1a3519aab40fa4adf67e8caa5b68fb120533 100644 (file)
--- a/test/CodeGen/Thumb2/thumb2-eor.ll
+++ b/test/CodeGen/Thumb2/thumb2-eor.ll
@@ -9,11 +9,18 @@ define i32 @f1(i32 %a, i32 %b) {
 
 define i32 @f2(i32 %a, i32 %b) {
 ; CHECK: f2:
-; CHECK: eor.w r0, r1, r0
+; CHECK: eors r0, r1
     %tmp = xor i32 %b, %a
     ret i32 %tmp
 }
 
+define i32 @f2b(i32 %a, i32 %b, i32 %c) {
+; CHECK: f2b:
+; CHECK: eor.w r0, r1, r2
+    %tmp = xor i32 %b, %c
+    ret i32 %tmp
+}
+
 define i32 @f3(i32 %a, i32 %b) {
 ; CHECK: f3:
 ; CHECK: eor.w r0, r0, r1, lsl #5

diff --git a/test/CodeGen/Thumb2/thumb2-uxtb.ll b/test/CodeGen/Thumb2/thumb2-uxtb.ll
index 541191400b1cce60095861050d5ca58bc086e4e2..724dbbddb12cd3fcacb18527cc6e23519a99926f 100644 (file)
--- a/test/CodeGen/Thumb2/thumb2-uxtb.ll
+++ b/test/CodeGen/Thumb2/thumb2-uxtb.ll
@@ -87,7 +87,7 @@ define i32 @test10(i32 %p0) {
 ; CHECK: and.w r0, r1, r0, lsr #7
 ; CHECK: lsrs  r1, r0, #5
 ; CHECK: uxtb16  r1, r1
-; CHECK: orr.w r0, r1, r0
+; CHECK: orrs r0, r1
 
        %tmp1 = lshr i32 %p0, 7         ; <i32> [#uses=1]
        %tmp2 = and i32 %tmp1, 16253176         ; <i32> [#uses=2]