Debug Info Testing: use null instead of an empty string in context field.

[oota-llvm.git] / test / CodeGen / ARM / vector-DAGCombine.ll

diff --git a/test/CodeGen/ARM/vector-DAGCombine.ll b/test/CodeGen/ARM/vector-DAGCombine.ll
index 1a97982eb0a23d67db012f1d5e4794b7cf6b2de1..4221c98424ace2b1a800631512d03b1d1d1e5afe 100644 (file)
--- a/test/CodeGen/ARM/vector-DAGCombine.ll
+++ b/test/CodeGen/ARM/vector-DAGCombine.ll
@@ -80,7 +80,7 @@ declare void @llvm.arm.neon.vst1.v8i8(i8*, <8 x i8>, i32) nounwind
 ; so they are not split up into i32 values.  Radar 8755338.
 define void @i64_buildvector(i64* %ptr, <2 x i64>* %vp) nounwind {
 ; CHECK: i64_buildvector
-; CHECK: vldr.64
+; CHECK: vldr
   %t0 = load i64* %ptr, align 4
   %t1 = insertelement <2 x i64> undef, i64 %t0, i32 0
   store <2 x i64> %t1, <2 x i64>* %vp
@@ -89,7 +89,7 @@ define void @i64_buildvector(i64* %ptr, <2 x i64>* %vp) nounwind {
 
 define void @i64_insertelement(i64* %ptr, <2 x i64>* %vp) nounwind {
 ; CHECK: i64_insertelement
-; CHECK: vldr.64
+; CHECK: vldr
   %t0 = load i64* %ptr, align 4
   %vec = load <2 x i64>* %vp
   %t1 = insertelement <2 x i64> %vec, i64 %t0, i32 0
@@ -99,7 +99,7 @@ define void @i64_insertelement(i64* %ptr, <2 x i64>* %vp) nounwind {
 
 define void @i64_extractelement(i64* %ptr, <2 x i64>* %vp) nounwind {
 ; CHECK: i64_extractelement
-; CHECK: vstr.64
+; CHECK: vstr
   %vec = load <2 x i64>* %vp
   %t1 = extractelement <2 x i64> %vec, i32 0
   store i64 %t1, i64* %ptr
@@ -133,3 +133,94 @@ define i16 @foldBuildVectors() {
   %3 = extractelement <8 x i16> %2, i32 0
   ret i16 %3
 }
+
+; Test that we are generating vrev and vext for reverse shuffles of v8i16
+; shuffles.
+; CHECK: reverse_v8i16
+define void @reverse_v8i16(<8 x i16>* %loadaddr, <8 x i16>* %storeaddr) {
+  %v0 = load <8 x i16>* %loadaddr
+  ; CHECK: vrev64.16
+  ; CHECK: vext.16
+  %v1 = shufflevector <8 x i16> %v0, <8 x i16> undef,
+              <8 x i32> <i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  store <8 x i16> %v1, <8 x i16>* %storeaddr
+  ret void
+}
+
+; Test that we are generating vrev and vext for reverse shuffles of v16i8
+; shuffles.
+; CHECK: reverse_v16i8
+define void @reverse_v16i8(<16 x i8>* %loadaddr, <16 x i8>* %storeaddr) {
+  %v0 = load <16 x i8>* %loadaddr
+  ; CHECK: vrev64.8
+  ; CHECK: vext.8
+  %v1 = shufflevector <16 x i8> %v0, <16 x i8> undef,
+       <16 x i32> <i32 15, i32 14, i32 13, i32 12, i32 11, i32 10, i32 9, i32 8,
+                   i32 7, i32 6, i32 5, i32 4, i32 3, i32 2, i32 1, i32 0>
+  store <16 x i8> %v1, <16 x i8>* %storeaddr
+  ret void
+}
+
+; <rdar://problem/14170854>.
+; vldr cannot handle unaligned loads.
+; Fall back to vld1.32, which can, instead of using the general purpose loads
+; followed by a costly sequence of instructions to build the vector register.
+; CHECK: t3
+; CHECK: vld1.32 {[[REG:d[0-9]+]][0]}
+; CHECK: vld1.32 {[[REG]][1]}
+; CHECK: vmull.u8 q{{[0-9]+}}, [[REG]], [[REG]]
+define <8 x i16> @t3(i8 zeroext %xf, i8* nocapture %sp0, i8* nocapture %sp1, i32* nocapture %outp) {
+entry:
+  %pix_sp0.0.cast = bitcast i8* %sp0 to i32*
+  %pix_sp0.0.copyload = load i32* %pix_sp0.0.cast, align 1
+  %pix_sp1.0.cast = bitcast i8* %sp1 to i32*
+  %pix_sp1.0.copyload = load i32* %pix_sp1.0.cast, align 1
+  %vecinit = insertelement <2 x i32> undef, i32 %pix_sp0.0.copyload, i32 0
+  %vecinit1 = insertelement <2 x i32> %vecinit, i32 %pix_sp1.0.copyload, i32 1
+  %0 = bitcast <2 x i32> %vecinit1 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
+}
+
+; 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
+}
+
+; Make sure vector load is used for all three loads.
+; Lowering to build vector was breaking the single use property of the load of
+;  %pix_sp0.0.copyload.
+; CHECK: t5
+; CHECK: vld1.32 {[[REG1:d[0-9]+]][1]}, [r0]
+; CHECK: vorr [[REG2:d[0-9]+]], [[REG1]], [[REG1]]
+; CHECK: vld1.32 {[[REG1]][0]}, [r1]
+; CHECK: vld1.32 {[[REG2]][0]}, [r2]
+; CHECK: vmull.u8 q{{[0-9]+}}, [[REG1]], [[REG2]]
+define <8 x i16> @t5(i8* nocapture %sp0, i8* nocapture %sp1, i8* nocapture %sp2) {
+entry:
+  %pix_sp0.0.cast = bitcast i8* %sp0 to i32*
+  %pix_sp0.0.copyload = load i32* %pix_sp0.0.cast, align 1
+  %pix_sp1.0.cast = bitcast i8* %sp1 to i32*
+  %pix_sp1.0.copyload = load i32* %pix_sp1.0.cast, align 1
+  %pix_sp2.0.cast = bitcast i8* %sp2 to i32*
+  %pix_sp2.0.copyload = load i32* %pix_sp2.0.cast, align 1
+  %vec = insertelement <2 x i32> undef, i32 %pix_sp0.0.copyload, i32 1
+  %vecinit1 = insertelement <2 x i32> %vec, i32 %pix_sp1.0.copyload, i32 0
+  %vecinit2 = insertelement <2 x i32> %vec, i32 %pix_sp2.0.copyload, i32 0
+  %0 = bitcast <2 x i32> %vecinit1 to <8 x i8>
+  %1 = bitcast <2 x i32> %vecinit2 to <8 x i8>
+  %vmull.i = tail call <8 x i16> @llvm.arm.neon.vmullu.v8i16(<8 x i8> %0, <8 x i8> %1)
+  ret <8 x i16> %vmull.i
+}