[EarlyCSE] Fix handling of target memory intrinsics for CSE'ing loads.

[oota-llvm.git] / test / Transforms / EarlyCSE / AArch64 / ldstN.ll

diff --git a/test/Transforms/EarlyCSE/AArch64/ldstN.ll b/test/Transforms/EarlyCSE/AArch64/ldstN.ll
new file mode 100644 (file)
index 0000000..cc1af31
--- /dev/null
+++ b/test/Transforms/EarlyCSE/AArch64/ldstN.ll
@@ -0,0 +1,18 @@
+; RUN: opt -S -early-cse < %s | FileCheck %s\r
+target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128"\r
+target triple = "aarch64--linux-gnu"\r
+\r
+declare { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> } @llvm.aarch64.neon.ld4.v4i16.p0v4i16(<4 x i16>*)\r
+\r
+; Although the store and the ld4 are using the same pointer, the\r
+; data can not be reused because ld4 accesses multiple elements.\r
+define { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> } @foo() {\r
+entry:\r
+  store <4 x i16> undef, <4 x i16>* undef, align 8\r
+  %0 = call { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> } @llvm.aarch64.neon.ld4.v4i16.p0v4i16(<4 x i16>* undef)\r
+  ret { <4 x i16>, <4 x i16>, <4 x i16>, <4 x i16> } %0\r
+; CHECK-LABEL: @foo(\r
+; CHECK: store\r
+; CHECK-NEXT: call\r
+; CHECK-NEXT: ret\r
+}\r