1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc -mtriple=riscv32 -mattr=+v,+d,+zvfh -verify-machineinstrs < %s | FileCheck %s 3; RUN: llc -mtriple=riscv64 -mattr=+v,+d,+zvfh -verify-machineinstrs < %s | FileCheck %s 4 5define <vscale x 4 x i32> @splat_c3_nxv4i32(<vscale x 4 x i32> %v) { 6; CHECK-LABEL: splat_c3_nxv4i32: 7; CHECK: # %bb.0: 8; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma 9; CHECK-NEXT: vrgather.vi v10, v8, 3 10; CHECK-NEXT: vmv.v.v v8, v10 11; CHECK-NEXT: ret 12 %x = extractelement <vscale x 4 x i32> %v, i32 3 13 %ins = insertelement <vscale x 4 x i32> poison, i32 %x, i32 0 14 %splat = shufflevector <vscale x 4 x i32> %ins, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer 15 ret <vscale x 4 x i32> %splat 16} 17 18define <vscale x 4 x i32> @splat_idx_nxv4i32(<vscale x 4 x i32> %v, i64 %idx) { 19; CHECK-LABEL: splat_idx_nxv4i32: 20; CHECK: # %bb.0: 21; CHECK-NEXT: vsetvli a1, zero, e32, m2, ta, ma 22; CHECK-NEXT: vrgather.vx v10, v8, a0 23; CHECK-NEXT: vmv.v.v v8, v10 24; CHECK-NEXT: ret 25 %x = extractelement <vscale x 4 x i32> %v, i64 %idx 26 %ins = insertelement <vscale x 4 x i32> poison, i32 %x, i32 0 27 %splat = shufflevector <vscale x 4 x i32> %ins, <vscale x 4 x i32> poison, <vscale x 4 x i32> zeroinitializer 28 ret <vscale x 4 x i32> %splat 29} 30 31define <vscale x 8 x i16> @splat_c4_nxv8i16(<vscale x 8 x i16> %v) { 32; CHECK-LABEL: splat_c4_nxv8i16: 33; CHECK: # %bb.0: 34; CHECK-NEXT: vsetvli a0, zero, e16, m2, ta, ma 35; CHECK-NEXT: vrgather.vi v10, v8, 4 36; CHECK-NEXT: vmv.v.v v8, v10 37; CHECK-NEXT: ret 38 %x = extractelement <vscale x 8 x i16> %v, i32 4 39 %ins = insertelement <vscale x 8 x i16> poison, i16 %x, i32 0 40 %splat = shufflevector <vscale x 8 x i16> %ins, <vscale x 8 x i16> poison, <vscale x 8 x i32> zeroinitializer 41 ret <vscale x 8 x i16> %splat 42} 43 44define <vscale x 8 x i16> @splat_idx_nxv8i16(<vscale x 8 x i16> %v, i64 %idx) { 45; CHECK-LABEL: splat_idx_nxv8i16: 46; CHECK: # %bb.0: 47; CHECK-NEXT: vsetvli a1, zero, e16, m2, ta, ma 48; CHECK-NEXT: vrgather.vx v10, v8, a0 49; CHECK-NEXT: vmv.v.v v8, v10 50; CHECK-NEXT: ret 51 %x = extractelement <vscale x 8 x i16> %v, i64 %idx 52 %ins = insertelement <vscale x 8 x i16> poison, i16 %x, i32 0 53 %splat = shufflevector <vscale x 8 x i16> %ins, <vscale x 8 x i16> poison, <vscale x 8 x i32> zeroinitializer 54 ret <vscale x 8 x i16> %splat 55} 56 57define <vscale x 2 x half> @splat_c1_nxv2f16(<vscale x 2 x half> %v) { 58; CHECK-LABEL: splat_c1_nxv2f16: 59; CHECK: # %bb.0: 60; CHECK-NEXT: vsetvli a0, zero, e16, mf2, ta, ma 61; CHECK-NEXT: vrgather.vi v9, v8, 1 62; CHECK-NEXT: vmv1r.v v8, v9 63; CHECK-NEXT: ret 64 %x = extractelement <vscale x 2 x half> %v, i32 1 65 %ins = insertelement <vscale x 2 x half> poison, half %x, i32 0 66 %splat = shufflevector <vscale x 2 x half> %ins, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer 67 ret <vscale x 2 x half> %splat 68} 69 70define <vscale x 2 x half> @splat_idx_nxv2f16(<vscale x 2 x half> %v, i64 %idx) { 71; CHECK-LABEL: splat_idx_nxv2f16: 72; CHECK: # %bb.0: 73; CHECK-NEXT: vsetvli a1, zero, e16, mf2, ta, ma 74; CHECK-NEXT: vrgather.vx v9, v8, a0 75; CHECK-NEXT: vmv1r.v v8, v9 76; CHECK-NEXT: ret 77 %x = extractelement <vscale x 2 x half> %v, i64 %idx 78 %ins = insertelement <vscale x 2 x half> poison, half %x, i32 0 79 %splat = shufflevector <vscale x 2 x half> %ins, <vscale x 2 x half> poison, <vscale x 2 x i32> zeroinitializer 80 ret <vscale x 2 x half> %splat 81} 82 83define <vscale x 4 x float> @splat_c3_nxv4f32(<vscale x 4 x float> %v) { 84; CHECK-LABEL: splat_c3_nxv4f32: 85; CHECK: # %bb.0: 86; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma 87; CHECK-NEXT: vrgather.vi v10, v8, 3 88; CHECK-NEXT: vmv.v.v v8, v10 89; CHECK-NEXT: ret 90 %x = extractelement <vscale x 4 x float> %v, i64 3 91 %ins = insertelement <vscale x 4 x float> poison, float %x, i32 0 92 %splat = shufflevector <vscale x 4 x float> %ins, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer 93 ret <vscale x 4 x float> %splat 94} 95 96define <vscale x 4 x float> @splat_idx_nxv4f32(<vscale x 4 x float> %v, i64 %idx) { 97; CHECK-LABEL: splat_idx_nxv4f32: 98; CHECK: # %bb.0: 99; CHECK-NEXT: vsetvli a1, zero, e32, m2, ta, ma 100; CHECK-NEXT: vrgather.vx v10, v8, a0 101; CHECK-NEXT: vmv.v.v v8, v10 102; CHECK-NEXT: ret 103 %x = extractelement <vscale x 4 x float> %v, i64 %idx 104 %ins = insertelement <vscale x 4 x float> poison, float %x, i32 0 105 %splat = shufflevector <vscale x 4 x float> %ins, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer 106 ret <vscale x 4 x float> %splat 107} 108 109define <vscale x 8 x float> @splat_idx_nxv4f32_nxv8f32(<vscale x 4 x float> %v, i64 %idx) { 110; CHECK-LABEL: splat_idx_nxv4f32_nxv8f32: 111; CHECK: # %bb.0: 112; CHECK-NEXT: vsetvli a1, zero, e32, m4, ta, ma 113; CHECK-NEXT: vrgather.vx v12, v8, a0 114; CHECK-NEXT: vmv.v.v v8, v12 115; CHECK-NEXT: ret 116 %x = extractelement <vscale x 4 x float> %v, i64 %idx 117 %ins = insertelement <vscale x 8 x float> poison, float %x, i32 0 118 %splat = shufflevector <vscale x 8 x float> %ins, <vscale x 8 x float> poison, <vscale x 8 x i32> zeroinitializer 119 ret <vscale x 8 x float> %splat 120} 121 122define <vscale x 4 x float> @splat_idx_v4f32_nxv4f32(<4 x float> %v, i64 %idx) { 123; CHECK-LABEL: splat_idx_v4f32_nxv4f32: 124; CHECK: # %bb.0: 125; CHECK-NEXT: vsetvli a1, zero, e32, m2, ta, ma 126; CHECK-NEXT: vrgather.vx v10, v8, a0 127; CHECK-NEXT: vmv.v.v v8, v10 128; CHECK-NEXT: ret 129 %x = extractelement <4 x float> %v, i64 %idx 130 %ins = insertelement <vscale x 4 x float> poison, float %x, i32 0 131 %splat = shufflevector <vscale x 4 x float> %ins, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer 132 ret <vscale x 4 x float> %splat 133} 134 135; Negative test, scale could have a value > 2 136define <8 x float> @splat_idx_nxv4f32_v8f32(<vscale x 4 x float> %v, i64 %idx) { 137; CHECK-LABEL: splat_idx_nxv4f32_v8f32: 138; CHECK: # %bb.0: 139; CHECK-NEXT: vsetivli zero, 1, e32, m2, ta, ma 140; CHECK-NEXT: vslidedown.vx v8, v8, a0 141; CHECK-NEXT: vfmv.f.s fa5, v8 142; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma 143; CHECK-NEXT: vfmv.v.f v8, fa5 144; CHECK-NEXT: ret 145 %x = extractelement <vscale x 4 x float> %v, i64 %idx 146 %ins = insertelement <8 x float> poison, float %x, i32 0 147 %splat = shufflevector <8 x float> %ins, <8 x float> poison, <8 x i32> zeroinitializer 148 ret <8 x float> %splat 149} 150 151define <vscale x 4 x float> @splat_idx_nxv8f32_nxv4f32_constant_0(<vscale x 8 x float> %v) { 152; CHECK-LABEL: splat_idx_nxv8f32_nxv4f32_constant_0: 153; CHECK: # %bb.0: 154; CHECK-NEXT: vsetvli a0, zero, e32, m2, ta, ma 155; CHECK-NEXT: vrgather.vi v10, v8, 0 156; CHECK-NEXT: vmv.v.v v8, v10 157; CHECK-NEXT: ret 158 %x = extractelement <vscale x 8 x float> %v, i64 0 159 %ins = insertelement <vscale x 4 x float> poison, float %x, i32 0 160 %splat = shufflevector <vscale x 4 x float> %ins, <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer 161 ret <vscale x 4 x float> %splat 162} 163 164define <vscale x 4 x i8> @splat_idx_nxv8i8_nxv4i8_constant_0(<vscale x 8 x i8> %v) { 165; CHECK-LABEL: splat_idx_nxv8i8_nxv4i8_constant_0: 166; CHECK: # %bb.0: 167; CHECK-NEXT: vsetvli a0, zero, e8, mf2, ta, ma 168; CHECK-NEXT: vrgather.vi v9, v8, 0 169; CHECK-NEXT: vmv1r.v v8, v9 170; CHECK-NEXT: ret 171 %x = extractelement <vscale x 8 x i8> %v, i64 0 172 %ins = insertelement <vscale x 4 x i8> poison, i8 %x, i32 0 173 %splat = shufflevector <vscale x 4 x i8> %ins, <vscale x 4 x i8> poison, <vscale x 4 x i32> zeroinitializer 174 ret <vscale x 4 x i8> %splat 175} 176 177define <vscale x 4 x i8> @splat_idx_nxv8i8_nxv4i8_constant_3(<vscale x 8 x i8> %v) { 178; CHECK-LABEL: splat_idx_nxv8i8_nxv4i8_constant_3: 179; CHECK: # %bb.0: 180; CHECK-NEXT: vsetvli a0, zero, e8, mf2, ta, ma 181; CHECK-NEXT: vrgather.vi v9, v8, 3 182; CHECK-NEXT: vmv1r.v v8, v9 183; CHECK-NEXT: ret 184 %x = extractelement <vscale x 8 x i8> %v, i64 3 185 %ins = insertelement <vscale x 4 x i8> poison, i8 %x, i32 0 186 %splat = shufflevector <vscale x 4 x i8> %ins, <vscale x 4 x i8> poison, <vscale x 4 x i32> zeroinitializer 187 ret <vscale x 4 x i8> %splat 188} 189 190 191; Negative test, vscale coule be 2 192define <vscale x 4 x i8> @splat_idx_nxv8i8_nxv4i8_constant_15(<vscale x 8 x i8> %v) { 193; CHECK-LABEL: splat_idx_nxv8i8_nxv4i8_constant_15: 194; CHECK: # %bb.0: 195; CHECK-NEXT: vsetivli zero, 1, e8, m1, ta, ma 196; CHECK-NEXT: vslidedown.vi v8, v8, 15 197; CHECK-NEXT: vmv.x.s a0, v8 198; CHECK-NEXT: vsetvli a1, zero, e8, mf2, ta, ma 199; CHECK-NEXT: vmv.v.x v8, a0 200; CHECK-NEXT: ret 201 %x = extractelement <vscale x 8 x i8> %v, i64 15 202 %ins = insertelement <vscale x 4 x i8> poison, i8 %x, i32 0 203 %splat = shufflevector <vscale x 4 x i8> %ins, <vscale x 4 x i8> poison, <vscale x 4 x i32> zeroinitializer 204 ret <vscale x 4 x i8> %splat 205} 206 207define <8 x float> @splat_idx_nxv4f32_v8f32_constant_0(<vscale x 4 x float> %v) { 208; CHECK-LABEL: splat_idx_nxv4f32_v8f32_constant_0: 209; CHECK: # %bb.0: 210; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma 211; CHECK-NEXT: vrgather.vi v10, v8, 0 212; CHECK-NEXT: vmv.v.v v8, v10 213; CHECK-NEXT: ret 214 %x = extractelement <vscale x 4 x float> %v, i64 0 215 %ins = insertelement <8 x float> poison, float %x, i32 0 216 %splat = shufflevector <8 x float> %ins, <8 x float> poison, <8 x i32> zeroinitializer 217 ret <8 x float> %splat 218} 219 220define <8 x float> @splat_idx_nxv4f32_v8f32_constant_7(<vscale x 4 x float> %v) { 221; CHECK-LABEL: splat_idx_nxv4f32_v8f32_constant_7: 222; CHECK: # %bb.0: 223; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma 224; CHECK-NEXT: vrgather.vi v10, v8, 7 225; CHECK-NEXT: vmv.v.v v8, v10 226; CHECK-NEXT: ret 227 %x = extractelement <vscale x 4 x float> %v, i64 7 228 %ins = insertelement <8 x float> poison, float %x, i32 0 229 %splat = shufflevector <8 x float> %ins, <8 x float> poison, <8 x i32> zeroinitializer 230 ret <8 x float> %splat 231} 232 233; Negative test, vscale might be 4 234define <8 x float> @splat_idx_nxv4f32_v8f32_constant_8(<vscale x 4 x float> %v) { 235; CHECK-LABEL: splat_idx_nxv4f32_v8f32_constant_8: 236; CHECK: # %bb.0: 237; CHECK-NEXT: vsetivli zero, 1, e32, m2, ta, ma 238; CHECK-NEXT: vslidedown.vi v8, v8, 8 239; CHECK-NEXT: vfmv.f.s fa5, v8 240; CHECK-NEXT: vsetivli zero, 8, e32, m2, ta, ma 241; CHECK-NEXT: vfmv.v.f v8, fa5 242; CHECK-NEXT: ret 243 %x = extractelement <vscale x 4 x float> %v, i64 8 244 %ins = insertelement <8 x float> poison, float %x, i32 0 245 %splat = shufflevector <8 x float> %ins, <8 x float> poison, <8 x i32> zeroinitializer 246 ret <8 x float> %splat 247} 248