It is better to not depend on CSE to share multiplies due to IV insertion.

This testcase checks that only one mul is present in the output code, as it
should be.

llvm-svn: 22701

1 files changed, 28 insertions, 0 deletions

diff --git a/llvm/test/Regression/Transforms/LoopStrengthReduce/share_code_in_preheader.ll b/llvm/test/Regression/Transforms/LoopStrengthReduce/share_code_in_preheader.ll
new file mode 100644
index 00000000000..532fe820ac4
--- /dev/null
+++ b/llvm/test/Regression/Transforms/LoopStrengthReduce/share_code_in_preheader.ll
@@ -0,0 +1,28 @@
+; RUN: llvm-as < %s | opt -loop-reduce | llvm-dis | grep mul | wc -l | grep 1
+; LSR should not make two copies of the Q*L expression in the preheader!
+
+sbyte %test(sbyte* %A, sbyte* %B, int %L, int %Q, int %N) {
+entry:
+	%tmp.6 = mul int %Q, %L		; <int> [#uses=1]
+	%N = cast int %N to uint		; <uint> [#uses=1]
+	br label %no_exit
+
+no_exit:		; preds = %no_exit, %no_exit.preheader
+	%indvar = phi uint [ 0, %entry], [ %indvar.next, %no_exit ]		; <uint> [#uses=2]
+	%Sum.0.0 = phi sbyte [ 0, %entry], [ %tmp.21, %no_exit ]		; <sbyte> [#uses=1]
+	%indvar = cast uint %indvar to int		; <int> [#uses=1]
+	%N_addr.0.0 = sub int %N, %indvar		; <int> [#uses=1]
+	%tmp.8 = add int %N_addr.0.0, %tmp.6		; <int> [#uses=2]
+	%tmp.9 = getelementptr sbyte* %A, int %tmp.8		; <sbyte*> [#uses=1]
+	%tmp.10 = load sbyte* %tmp.9		; <sbyte> [#uses=1]
+	%tmp.17 = getelementptr sbyte* %B, int %tmp.8		; <sbyte*> [#uses=1]
+	%tmp.18 = load sbyte* %tmp.17		; <sbyte> [#uses=1]
+	%tmp.19 = sub sbyte %tmp.10, %tmp.18		; <sbyte> [#uses=1]
+	%tmp.21 = add sbyte %tmp.19, %Sum.0.0		; <sbyte> [#uses=2]
+	%indvar.next = add uint %indvar, 1		; <uint> [#uses=2]
+	%exitcond = seteq uint %indvar.next, %N		; <bool> [#uses=1]
+	br bool %exitcond, label %loopexit, label %no_exit
+
+loopexit:
+	ret sbyte %tmp.21
+}