1; RUN: llc < %s -march=arm64 -aarch64-neon-syntax=apple | FileCheck %s 2 3; Check that building up a vector w/ only one non-zero lane initializes 4; intelligently. 5define void @one_lane(i32* nocapture %out_int, i32 %skip0) nounwind { 6; CHECK-LABEL: one_lane: 7; CHECK: dup.16b v[[REG:[0-9]+]], wzr 8; CHECK-NEXT: ins.b v[[REG]][0], w1 9; v and q are aliases, and str is preferred against st.16b when possible 10; rdar://11246289 11; CHECK: str q[[REG]], [x0] 12; CHECK: ret 13 %conv = trunc i32 %skip0 to i8 14 %vset_lane = insertelement <16 x i8> <i8 undef, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0, i8 0>, i8 %conv, i32 0 15 %tmp = bitcast i32* %out_int to <4 x i32>* 16 %tmp1 = bitcast <16 x i8> %vset_lane to <4 x i32> 17 store <4 x i32> %tmp1, <4 x i32>* %tmp, align 16 18 ret void 19} 20 21; Check that building a vector from floats doesn't insert an unnecessary 22; copy for lane zero. 23define <4 x float> @foo(float %a, float %b, float %c, float %d) nounwind { 24; CHECK-LABEL: foo: 25; CHECK-NOT: ins.s v0[0], v0[0] 26; CHECK: ins.s v0[1], v1[0] 27; CHECK: ins.s v0[2], v2[0] 28; CHECK: ins.s v0[3], v3[0] 29; CHECK: ret 30 %1 = insertelement <4 x float> undef, float %a, i32 0 31 %2 = insertelement <4 x float> %1, float %b, i32 1 32 %3 = insertelement <4 x float> %2, float %c, i32 2 33 %4 = insertelement <4 x float> %3, float %d, i32 3 34 ret <4 x float> %4 35} 36 37define <8 x i16> @build_all_zero(<8 x i16> %a) #1 { 38; CHECK-LABEL: build_all_zero: 39; CHECK: movz w[[GREG:[0-9]+]], #0xae80 40; CHECK-NEXT: fmov s[[FREG:[0-9]+]], w[[GREG]] 41; CHECK-NEXT: mul.8h v0, v0, v[[FREG]] 42 %b = add <8 x i16> %a, <i16 -32768, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef> 43 %c = mul <8 x i16> %b, <i16 -20864, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef> 44 ret <8 x i16> %c 45} 46 47; There is an optimization in DAG Combiner as following: 48; fold (concat_vectors (BUILD_VECTOR A, B, ...), (BUILD_VECTOR C, D, ...)) 49; -> (BUILD_VECTOR A, B, ..., C, D, ...) 50; This case checks when A,B and C,D are different types, there should be no 51; assertion failure. 52define <8 x i16> @concat_2_build_vector(<4 x i16> %in0) { 53; CHECK-LABEL: concat_2_build_vector: 54; CHECK: movi 55 %vshl_n = shl <4 x i16> %in0, <i16 8, i16 8, i16 8, i16 8> 56 %vshl_n2 = shl <4 x i16> %vshl_n, <i16 9, i16 9, i16 9, i16 9> 57 %shuffle.i = shufflevector <4 x i16> %vshl_n2, <4 x i16> zeroinitializer, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 58 ret <8 x i16> %shuffle.i 59}