57 lines · plain
1; RUN: llc -mtriple=arm-eabi -mattr=+neon -print-before=post-RA-sched %s -o - 2>&1 \2; RUN: | FileCheck %s3 4define void @vst(ptr %m, [4 x i64] %v) {5entry:6; CHECK: vst:7; CHECK: VST1d64Q killed $r{{[0-9]+}}, 8, $d{{[0-9]+}}, 14, $noreg, implicit killed $q{{[0-9]+}}_q{{[0-9]+}}8 9 %v0 = extractvalue [4 x i64] %v, 010 %v1 = extractvalue [4 x i64] %v, 111 %v2 = extractvalue [4 x i64] %v, 212 %v3 = extractvalue [4 x i64] %v, 313 14 %t0 = bitcast i64 %v0 to <8 x i8>15 %t1 = bitcast i64 %v1 to <8 x i8>16 %t2 = bitcast i64 %v2 to <8 x i8>17 %t3 = bitcast i64 %v3 to <8 x i8>18 19 %s0 = bitcast <8 x i8> %t0 to <1 x i64>20 %s1 = bitcast <8 x i8> %t1 to <1 x i64>21 %s2 = bitcast <8 x i8> %t2 to <1 x i64>22 %s3 = bitcast <8 x i8> %t3 to <1 x i64>23 24 %tmp0 = bitcast <1 x i64> %s2 to i6425 %tmp1 = bitcast <1 x i64> %s3 to i6426 27 %n0 = insertelement <2 x i64> undef, i64 %tmp0, i32 028 %n1 = insertelement <2 x i64> %n0, i64 %tmp1, i32 129 30 call void @llvm.arm.neon.vst4.p0.v1i64(ptr %m, <1 x i64> %s0, <1 x i64> %s1, <1 x i64> %s2, <1 x i64> %s3, i32 8)31 32 call void @bar(<2 x i64> %n1)33 34 ret void35}36 37%struct.__neon_int8x8x4_t = type { <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8> }38define <8 x i8> @vtbx4(ptr %A, ptr %B, ptr %C) nounwind {39; CHECK: vtbx4:40; CHECK: VTBX4 {{.*}}, 14, $noreg, implicit $q{{[0-9]+}}_q{{[0-9]+}}41 %tmp1 = load <8 x i8>, ptr %A42 %tmp2 = load %struct.__neon_int8x8x4_t, ptr %B43 %tmp3 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 044 %tmp4 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 145 %tmp5 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 246 %tmp6 = extractvalue %struct.__neon_int8x8x4_t %tmp2, 347 %tmp7 = load <8 x i8>, ptr %C48 %tmp8 = call <8 x i8> @llvm.arm.neon.vtbx4(<8 x i8> %tmp1, <8 x i8> %tmp3, <8 x i8> %tmp4, <8 x i8> %tmp5, <8 x i8> %tmp6, <8 x i8> %tmp7)49 call void @bar2(%struct.__neon_int8x8x4_t %tmp2, <8 x i8> %tmp8)50 ret <8 x i8> %tmp851}52 53declare void @llvm.arm.neon.vst4.p0.v1i64(ptr, <1 x i64>, <1 x i64>, <1 x i64>, <1 x i64>, i32)54declare <8 x i8> @llvm.arm.neon.vtbx4(<8 x i8>, <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8>, <8 x i8>) nounwind readnone55declare void @bar2(%struct.__neon_int8x8x4_t, <8 x i8>)56declare void @bar(<2 x i64> %arg)57