Fix the "test" optimization to recognize "dec" as an add of

negative one, as subtracts of immediates are canonicalized
to adds.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@66180 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index d56e672e1f82e615a758e136de084d321cc143db..33eee7a7368dafe6b6c5aecfa064ea8dff34d3ba 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -5382,13 +5382,19 @@ SDValue X86TargetLowering::EmitTest(SDValue Op, SelectionDAG &DAG) {
            UE = Op.getNode()->use_end(); UI != UE; ++UI)
         if (UI->getOpcode() == ISD::STORE)
           goto default_case;
-      // An add of one will be selected as an INC.
       if (ConstantSDNode *C =
-            dyn_cast<ConstantSDNode>(Op.getNode()->getOperand(1)))
+            dyn_cast<ConstantSDNode>(Op.getNode()->getOperand(1))) {
+        // An add of one will be selected as an INC.
         if (C->getAPIntValue() == 1) {
           Opcode = X86ISD::INC;
           break;
         }
+        // An add of negative one (subtract of one) will be selected as a DEC.
+        if (C->getAPIntValue().isAllOnesValue()) {
+          Opcode = X86ISD::DEC;
+          break;
+        }
+      }
       // Otherwise use a regular EFLAGS-setting add.
       Opcode = X86ISD::ADD;
       break;
@@ -5399,13 +5405,6 @@ SDValue X86TargetLowering::EmitTest(SDValue Op, SelectionDAG &DAG) {
            UE = Op.getNode()->use_end(); UI != UE; ++UI)
         if (UI->getOpcode() == ISD::STORE)
           goto default_case;
-      // A subtract of one will be selected as a DEC.
-      if (ConstantSDNode *C =
-            dyn_cast<ConstantSDNode>(Op.getNode()->getOperand(1)))
-        if (C->getAPIntValue() == 1) {
-          Opcode = X86ISD::DEC;
-          break;
-        }
       // Otherwise use a regular EFLAGS-setting sub.
       Opcode = X86ISD::SUB;
       break;

diff --git a/test/CodeGen/X86/peep-test-0.ll b/test/CodeGen/X86/peep-test-0.ll
new file mode 100644 (file)
index 0000000..a95b564
--- /dev/null
+++ b/test/CodeGen/X86/peep-test-0.ll
@@ -0,0 +1,22 @@
+; RUN: llvm-as < %s | llc -march=x86-64 > %t
+; RUN: not grep cmp %t
+; RUN: not grep test %t
+
+define void @loop(i64 %n, double* nocapture %d) nounwind {
+entry:
+       br label %bb
+
+bb:
+       %indvar = phi i64 [ %n, %entry ], [ %indvar.next, %bb ]
+       %i.03 = add i64 %indvar, %n
+       %0 = getelementptr double* %d, i64 %i.03
+       %1 = load double* %0, align 8
+       %2 = mul double %1, 3.000000e+00
+       store double %2, double* %0, align 8
+       %indvar.next = add i64 %indvar, 1
+       %exitcond = icmp eq i64 %indvar.next, 0
+       br i1 %exitcond, label %return, label %bb
+
+return:
+       ret void
+}

diff --git a/test/CodeGen/X86/peep-test-1.ll b/test/CodeGen/X86/peep-test-1.ll
new file mode 100644 (file)
index 0000000..b4698e3
--- /dev/null
+++ b/test/CodeGen/X86/peep-test-1.ll
@@ -0,0 +1,23 @@
+; RUN: llvm-as < %s | llc -march=x86 > %t
+; RUN: grep dec %t | count 1
+; RUN: not grep test %t
+; RUN: not grep cmp %t
+
+define void @foo(i32 %n, double* nocapture %p) nounwind {
+       br label %bb
+
+bb:
+       %indvar = phi i32 [ 0, %0 ], [ %indvar.next, %bb ]
+       %i.03 = sub i32 %n, %indvar
+       %1 = getelementptr double* %p, i32 %i.03
+       %2 = load double* %1, align 4
+       %3 = mul double %2, 2.930000e+00
+       store double %3, double* %1, align 4
+       %4 = add i32 %i.03, -1
+       %phitmp = icmp slt i32 %4, 0
+       %indvar.next = add i32 %indvar, 1
+       br i1 %phitmp, label %bb, label %return
+
+return:
+       ret void
+}

diff --git a/test/CodeGen/X86/peep-test.ll b/test/CodeGen/X86/peep-test.ll
deleted file mode 100644 (file)
index a95b564..0000000
--- a/test/CodeGen/X86/peep-test.ll
+++ /dev/null
@@ -1,22 +0,0 @@
-; RUN: llvm-as < %s | llc -march=x86-64 > %t
-; RUN: not grep cmp %t
-; RUN: not grep test %t
-
-define void @loop(i64 %n, double* nocapture %d) nounwind {
-entry:
-       br label %bb
-
-bb:
-       %indvar = phi i64 [ %n, %entry ], [ %indvar.next, %bb ]
-       %i.03 = add i64 %indvar, %n
-       %0 = getelementptr double* %d, i64 %i.03
-       %1 = load double* %0, align 8
-       %2 = mul double %1, 3.000000e+00
-       store double %2, double* %0, align 8
-       %indvar.next = add i64 %indvar, 1
-       %exitcond = icmp eq i64 %indvar.next, 0
-       br i1 %exitcond, label %return, label %bb
-
-return:
-       ret void
-}