From ac146657fc2dc63e9856e0090fd0bec4ba9befd3 Mon Sep 17 00:00:00 2001
From: Dan Gohman <gohman@apple.com>
Date: Fri, 20 Nov 2009 01:09:34 +0000
Subject: [PATCH] Teach getSmallConstantTripMultiple about Shl operators.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@89426 91177308-0d34-0410-b5e6-96231b3b80d8
---
 lib/Analysis/LoopInfo.cpp                     |  5 ++++
 .../LoopUnroll/shifted-tripcount.ll           | 28 +++++++++++++++++++
 2 files changed, 33 insertions(+)
 create mode 100644 test/Transforms/LoopUnroll/shifted-tripcount.ll

diff --git a/lib/Analysis/LoopInfo.cpp b/lib/Analysis/LoopInfo.cpp
index 1c614b0e06b..4de756c41b0 100644
--- a/lib/Analysis/LoopInfo.cpp
+++ b/lib/Analysis/LoopInfo.cpp
@@ -243,6 +243,11 @@ unsigned Loop::getSmallConstantTripMultiple() const {
         case BinaryOperator::Mul:
           Result = dyn_cast<ConstantInt>(BO->getOperand(1));
           break;
+        case BinaryOperator::Shl:
+          if (ConstantInt *CI = dyn_cast<ConstantInt>(BO->getOperand(1)))
+            if (CI->getValue().getActiveBits() <= 5)
+              return 1u << CI->getZExtValue();
+          break;
         default:
           break;
         }
diff --git a/test/Transforms/LoopUnroll/shifted-tripcount.ll b/test/Transforms/LoopUnroll/shifted-tripcount.ll
new file mode 100644
index 00000000000..a118a463172
--- /dev/null
+++ b/test/Transforms/LoopUnroll/shifted-tripcount.ll
@@ -0,0 +1,28 @@
+; RUN: opt < %s -loop-unroll -unroll-count=2 -S | FileCheck %s
+
+; LoopUnroll should unroll this loop into one big basic block.
+
+; CHECK: for.body:
+; CHECK: %i.013 = phi i64 [ 0, %entry ], [ %tmp16.1, %for.body ]
+; CHECK: br i1 %exitcond.1, label %for.end, label %for.body
+
+define void @foo(double* nocapture %p, i64 %n) nounwind {
+entry:
+  %mul10 = shl i64 %n, 1                          ; <i64> [#uses=2]
+  br label %for.body
+
+for.body:                                         ; preds = %entry, %for.body
+  %i.013 = phi i64 [ %tmp16, %for.body ], [ 0, %entry ] ; <i64> [#uses=2]
+  %arrayidx7 = getelementptr double* %p, i64 %i.013 ; <double*> [#uses=2]
+  %tmp16 = add i64 %i.013, 1                      ; <i64> [#uses=3]
+  %arrayidx = getelementptr double* %p, i64 %tmp16 ; <double*> [#uses=1]
+  %tmp4 = load double* %arrayidx                  ; <double> [#uses=1]
+  %tmp8 = load double* %arrayidx7                 ; <double> [#uses=1]
+  %mul9 = fmul double %tmp8, %tmp4                ; <double> [#uses=1]
+  store double %mul9, double* %arrayidx7
+  %exitcond = icmp eq i64 %tmp16, %mul10          ; <i1> [#uses=1]
+  br i1 %exitcond, label %for.end, label %for.body
+
+for.end:                                          ; preds = %for.body, %entry
+  ret void
+}