1 ; RUN: llc < %s -mtriple=arm64-eabi -aarch64-neon-syntax=apple | FileCheck %s 2 3 ; Check that building a vector from floats doesn't insert an unnecessary 4 ; copy for lane zero. 5 define <4 x float> @foo(float %a, float %b, float %c, float %d) nounwind { 6 ; CHECK-LABEL: foo: 7 ; CHECK-NOT: mov.s v0[0], v0[0] 8 ; CHECK: mov.s v0[1], v1[0] 9 ; CHECK: mov.s v0[2], v2[0] 10 ; CHECK: mov.s v0[3], v3[0] 11 ; CHECK: ret 12 %1 = insertelement <4 x float> undef, float %a, i32 0 13 %2 = insertelement <4 x float> %1, float %b, i32 1 14 %3 = insertelement <4 x float> %2, float %c, i32 2 15 %4 = insertelement <4 x float> %3, float %d, i32 3 16 ret <4 x float> %4 17 } 18 19 define <8 x i16> @build_all_zero(<8 x i16> %a) #1 { 20 ; CHECK-LABEL: build_all_zero: 21 ; CHECK: mov w[[GREG:[0-9]+]], #44672 22 ; CHECK-NEXT: fmov s[[FREG:[0-9]+]], w[[GREG]] 23 ; CHECK-NEXT: mul.8h v0, v0, v[[FREG]] 24 %b = add <8 x i16> %a, <i16 -32768, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef> 25 %c = mul <8 x i16> %b, <i16 -20864, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef, i16 undef> 26 ret <8 x i16> %c 27 } 28 29 ; There is an optimization in DAG Combiner as following: 30 ; fold (concat_vectors (BUILD_VECTOR A, B, ...), (BUILD_VECTOR C, D, ...)) 31 ; -> (BUILD_VECTOR A, B, ..., C, D, ...) 32 ; This case checks when A,B and C,D are different types, there should be no 33 ; assertion failure. 34 define <8 x i16> @concat_2_build_vector(<4 x i16> %in0) { 35 ; CHECK-LABEL: concat_2_build_vector: 36 ; CHECK: movi 37 %vshl_n = shl <4 x i16> %in0, <i16 8, i16 8, i16 8, i16 8> 38 %vshl_n2 = shl <4 x i16> %vshl_n, <i16 9, i16 9, i16 9, i16 9> 39 %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> 40 ret <8 x i16> %shuffle.i 41 } 42
