LoopVectorizer: Truncate i64 trip counts of i32 phis if necessary

In signed arithmetic we could end up with an i64 trip count for an i32 phi.
Because it is signed arithmetic we know that this is only defined if the i32
does not wrap. It is therefore safe to truncate the i64 trip count to a i32
value.

Fixes PR18049.

llvm-svn: 195787

1 files changed, 39 insertions, 0 deletions

diff --git a/llvm/test/Transforms/LoopVectorize/X86/tripcount.ll b/llvm/test/Transforms/LoopVectorize/X86/tripcount.ll
new file mode 100644
index 00000000000..6b38bacf888
--- /dev/null
+++ b/llvm/test/Transforms/LoopVectorize/X86/tripcount.ll
@@ -0,0 +1,39 @@
+; RUN: opt -S -loop-vectorize -force-vector-width=2 -force-vector-unroll=1 -mcpu=prescott < %s | FileCheck %s
+
+target datalayout = "e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32-n8:16:32-S128"
+target triple = "i386-unknown-freebsd11.0"
+
+@big = external global [0 x i32]
+
+; PR18049
+; We need to truncate the exit count to i32. This is legal because the
+; arithmetic is signed (%inc is nsw).
+
+; CHECK-LABEL: tripcount
+; CHECK: trunc i64 %count to i32
+
+define void @tripcount(i64 %count) {
+entry:
+  %cmp6 = icmp sgt i64 %count, 0
+  br i1 %cmp6, label %for.body.preheader, label %for.end
+
+for.body.preheader:
+  br label %for.body
+
+for.body:
+  %i.07 = phi i32 [ %inc, %for.body ], [ 0, %for.body.preheader ]
+  %arrayidx = getelementptr inbounds [0 x i32]* @big, i32 0, i32 %i.07
+  %0 = load i32* %arrayidx, align 4
+  %neg = xor i32 %0, -1
+  store i32 %neg, i32* %arrayidx, align 4
+  %inc = add nsw i32 %i.07, 1
+  %conv = sext i32 %inc to i64
+  %cmp = icmp slt i64 %conv, %count
+  br i1 %cmp, label %for.body, label %for.end.loopexit
+
+for.end.loopexit:
+  br label %for.end
+
+for.end:
+  ret void
+}