InterleaveAccessPass: Avoid constructing invalid shuffle masks

Fix a bug where we would construct shufflevector instructions addressing
invalid elements.

Differential Revision: https://reviews.llvm.org/D29313

llvm-svn: 293673

2 files changed, 36 insertions, 0 deletions

diff --git a/llvm/test/Transforms/InterleavedAccess/AArch64/interleaved-accesses.ll b/llvm/test/Transforms/InterleavedAccess/AArch64/interleaved-accesses.ll
index 702dfdbb81a..2a257d49081 100644
--- a/llvm/test/Transforms/InterleavedAccess/AArch64/interleaved-accesses.ll
+++ b/llvm/test/Transforms/InterleavedAccess/AArch64/interleaved-accesses.ll
@@ -547,3 +547,21 @@ define void @store_general_mask_factor3_negativestart(<12 x i32>* %ptr, <32 x i3
   store <12 x i32> %interleaved.vec, <12 x i32>* %ptr, align 4
   ret void
 }
+
+@g = external global <4 x float>
+
+; The following does not give a valid interleaved store
+; NEON-LABEL: define void @no_interleave
+; NEON-NOT: call void @llvm.aarch64.neon.st2
+; NEON: shufflevector
+; NEON: store
+; NEON: ret void
+; NO_NEON-LABEL: define void @no_interleave
+; NO_NEON: shufflevector
+; NO_NEON: store
+; NO_NEON: ret void
+define void @no_interleave(<4 x float> %a0) {
+  %v0 = shufflevector <4 x float> %a0, <4 x float> %a0, <4 x i32> <i32 0, i32 3, i32 7, i32 undef>
+  store <4 x float> %v0, <4 x float>* @g, align 16
+  ret void
+}
diff --git a/llvm/test/Transforms/InterleavedAccess/ARM/interleaved-accesses.ll b/llvm/test/Transforms/InterleavedAccess/ARM/interleaved-accesses.ll
index caaaa21c5a1..21eb8d7a1b0 100644
--- a/llvm/test/Transforms/InterleavedAccess/ARM/interleaved-accesses.ll
+++ b/llvm/test/Transforms/InterleavedAccess/ARM/interleaved-accesses.ll
@@ -626,3 +626,21 @@ define void @store_general_mask_factor3_midstart_pass(<12 x i32>* %ptr, <32 x i3
   store <12 x i32> %interleaved.vec, <12 x i32>* %ptr, align 4
   ret void
 }
+
+@g = external global <4 x float>
+
+; The following does not give a valid interleaved store
+; NEON-LABEL: define void @no_interleave
+; NEON-NOT: call void @llvm.arm.neon.vst2
+; NEON: shufflevector
+; NEON: store
+; NEON: ret void
+; NO_NEON-LABEL: define void @no_interleave
+; NO_NEON: shufflevector
+; NO_NEON: store
+; NO_NEON: ret void
+define void @no_interleave(<4 x float> %a0) {
+  %v0 = shufflevector <4 x float> %a0, <4 x float> %a0, <4 x i32> <i32 0, i32 7, i32 1, i32 undef>
+  store <4 x float> %v0, <4 x float>* @g, align 16
+  ret void
+}