AVX-512: fixed a bug in getLoadStoreRegOpcode() for AVX-512 target

[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 6d586f2426464b632ce8c0cae113ff9f0633cc17..793934e746b6e058f3aee37a2b48b468c8fd539e 100644 (file)
--- a/test/CodeGen/ARM/vector-DAGCombine.ll
+++ b/test/CodeGen/ARM/vector-DAGCombine.ll
@@ -184,3 +184,63 @@ 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
+}
+
+; 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
+}
+
+; <rdar://problem/14989896> Make sure we manage to truncate a vector from an
+; illegal type to a legal type.
+define <2 x i8> @test_truncate(<2 x i128> %in) {
+; CHECK-LABEL: test_truncate:
+; CHECK: mov [[BASE:r[0-9]+]], sp
+; CHECK-NEXT: vld1.32 {[[REG1:d[0-9]+]][0]}, {{\[}}[[BASE]]:32]
+; CHECK-NEXT: add [[BASE2:r[0-9]+]], [[BASE]], #4
+; CHECK-NEXT: vld1.32 {[[REG1]][1]}, {{\[}}[[BASE2]]:32]
+; REG2 Should map on the same Q register as REG1, i.e., REG2 = REG1 - 1, but we
+; cannot express that.
+; CHECK-NEXT: vmov.32 [[REG2:d[0-9]+]][0], r0
+; CHECK-NEXT: vmov.32 [[REG2]][1], r1
+; The Q register used here should match floor(REG1/2), but we cannot express that.
+; CHECK-NEXT: vmovn.i64 [[RES:d[0-9]+]], q{{[0-9]+}}
+; CHECK-NEXT: vmov r0, r1, [[RES]]
+entry:
+  %res = trunc <2 x i128> %in to <2 x i8>
+  ret <2 x i8> %res
+}