optimize comparisons against cttz/ctlz/ctpop, patch by Alastair Lynn!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@92745 91177308-0d34-0410-b5e6-96231b3b80d8

2 files changed, 49 insertions, 5 deletions

diff --git a/lib/Transforms/InstCombine/InstCombineCompares.cpp b/lib/Transforms/InstCombine/InstCombineCompares.cpp
index 90ab4f4..abbc89b 100644
--- a/lib/Transforms/InstCombine/InstCombineCompares.cpp
+++ b/lib/Transforms/InstCombine/InstCombineCompares.cpp
@@ -1418,11 +1418,33 @@ Instruction *InstCombiner::visitICmpInstWithInstAndIntCst(ICmpInst &ICI,
       }
     } else if (IntrinsicInst *II = dyn_cast<IntrinsicInst>(LHSI)) {
       // Handle icmp {eq|ne} <intrinsic>, intcst.
-      if (II->getIntrinsicID() == Intrinsic::bswap) {
+      switch (II->getIntrinsicID()) {
+      case Intrinsic::bswap:
         Worklist.Add(II);
         ICI.setOperand(0, II->getOperand(1));
         ICI.setOperand(1, ConstantInt::get(II->getContext(), RHSV.byteSwap()));
         return &ICI;
+      case Intrinsic::ctlz:
+      case Intrinsic::cttz:
+        // ctz(A) == bitwidth(a)  ->  A == 0 and likewise for !=
+        if (RHSV == RHS->getType()->getBitWidth()) {
+          Worklist.Add(II);
+          ICI.setOperand(0, II->getOperand(1));
+          ICI.setOperand(1, ConstantInt::get(RHS->getType(), 0));
+          return &ICI;
+        }
+        break;
+      case Intrinsic::ctpop:
+        // popcount(A) == 0  ->  A == 0 and likewise for !=
+        if (RHS->isZero()) {
+          Worklist.Add(II);
+          ICI.setOperand(0, II->getOperand(1));
+          ICI.setOperand(1, RHS);
+          return &ICI;
+        }
+        break;
+      default:
+      	break;
       }
     }
   }
diff --git a/test/Transforms/InstCombine/intrinsics.ll b/test/Transforms/InstCombine/intrinsics.ll
index f899fc0..c63475c 100644
--- a/test/Transforms/InstCombine/intrinsics.ll
+++ b/test/Transforms/InstCombine/intrinsics.ll
@@ -6,6 +6,8 @@ declare %overflow.result @llvm.uadd.with.overflow.i8(i8, i8)
 declare %overflow.result @llvm.umul.with.overflow.i8(i8, i8)
 declare double @llvm.powi.f64(double, i32) nounwind readonly
 declare i32 @llvm.cttz.i32(i32) nounwind readnone
+declare i32 @llvm.ctlz.i32(i32) nounwind readnone
+declare i32 @llvm.ctpop.i32(i32) nounwind readnone
 declare i8 @llvm.ctlz.i8(i8) nounwind readnone
 
 define i8 @test1(i8 %A, i8 %B) {
@@ -99,8 +101,7 @@ entry:
 ; CHECK: volatile store double %V
 }
 
-define i32 @cttz(i32 %a)
-{
+define i32 @cttz(i32 %a) {
 entry:
   %or = or i32 %a, 8
   %and = and i32 %or, -8
@@ -111,8 +112,7 @@ entry:
 ; CHECK-NEXT: ret i32 3
 }
 
-define i8 @ctlz(i8 %a)
-{
+define i8 @ctlz(i8 %a) {
 entry:
   %or = or i8 %a, 32
   %and = and i8 %or, 63
@@ -122,3 +122,25 @@ entry:
 ; CHECK-NEXT: entry:
 ; CHECK-NEXT: ret i8 2
 }
+
+define void @cmp.simplify(i32 %a, i32 %b, i1* %c) {
+entry:
+  %lz = tail call i32 @llvm.ctlz.i32(i32 %a) nounwind readnone
+  %lz.cmp = icmp eq i32 %lz, 32
+  volatile store i1 %lz.cmp, i1* %c
+  %tz = tail call i32 @llvm.cttz.i32(i32 %a) nounwind readnone
+  %tz.cmp = icmp ne i32 %tz, 32
+  volatile store i1 %tz.cmp, i1* %c
+  %pop = tail call i32 @llvm.ctpop.i32(i32 %b) nounwind readnone
+  %pop.cmp = icmp eq i32 %pop, 0
+  volatile store i1 %pop.cmp, i1* %c
+  ret void
+; CHECK: @cmp.simplify
+; CHECK-NEXT: entry:
+; CHECK-NEXT: %lz.cmp = icmp eq i32 %a, 0
+; CHECK-NEXT: volatile store i1 %lz.cmp, i1* %c
+; CHECK-NEXT: %tz.cmp = icmp ne i32 %a, 0
+; CHECK-NEXT: volatile store i1 %tz.cmp, i1* %c
+; CHECK-NEXT: %pop.cmp = icmp eq i32 %b, 0
+; CHECK-NEXT: volatile store i1 %pop.cmp, i1* %c
+}