[ARM][ISel] Improve the lowering of vector loads.

When vectors are built from a single value, the ARM lowering issues a
scalar_to_vector node.
This node is then always morphed into a move from the general purpose unit to
the vector unit.
When the value comes from a load, this can be simplified into a vector load to
the right lane.

This patch changes the lowering of insert_vector_elt to expose a vector
friendly pattern in this situation.

This is a step toward fixing <rdar://problem/14170854>.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186999 91177308-0d34-0410-b5e6-96231b3b80d8

1 files changed, 14 insertions, 0 deletions

diff --git a/test/CodeGen/ARM/vector-DAGCombine.ll b/test/CodeGen/ARM/vector-DAGCombine.ll
index 6d586f2426..3e138199e6 100644
--- a/test/CodeGen/ARM/vector-DAGCombine.ll
+++ b/test/CodeGen/ARM/vector-DAGCombine.ll
@@ -184,3 +184,17 @@ entry:
 
 ; Function Attrs: nounwind readnone
 declare <8 x i16> @llvm.arm.neon.vmullu.v8i16(<8 x i8>, <8 x i8>)
+
+; Check that (insert_vector_elt (load)) => (vector_load).
+; Thus, check that scalar_to_vector do not interfer with that.
+define <8 x i16> @t4(i8* nocapture %sp0) {
+; CHECK: t4
+; CHECK: vld1.32 {{{d[0-9]+}}[0]}, [r0]
+entry:
+  %pix_sp0.0.cast = bitcast i8* %sp0 to i32*
+  %pix_sp0.0.copyload = load i32* %pix_sp0.0.cast, align 1
+  %vec = insertelement <2 x i32> undef, i32 %pix_sp0.0.copyload, i32 0
+  %0 = bitcast <2 x i32> %vec to <8 x i8>
+  %vmull.i = tail call <8 x i16> @llvm.arm.neon.vmullu.v8i16(<8 x i8> %0, <8 x i8> %0)
+  ret <8 x i16> %vmull.i
+}