1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=loop-vectorize -mtriple aarch64-unknown-linux-gnu -mattr=+sve -epilogue-vectorization-force-VF=2 -S | FileCheck %s 3 4; 5; Integer reduction with interleaving & a start value of 5 6; 7define i64 @int_reduction_add(ptr %a, i64 %N) { 8; CHECK-LABEL: @int_reduction_add( 9; CHECK-NEXT: iter.check: 10; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 2 11; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH:%.*]], label [[VECTOR_MAIN_LOOP_ITER_CHECK:%.*]] 12; CHECK: vector.main.loop.iter.check: 13; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 14; CHECK-NEXT: [[TMP1:%.*]] = mul i64 [[TMP0]], 4 15; CHECK-NEXT: [[MIN_ITERS_CHECK1:%.*]] = icmp ult i64 [[N]], [[TMP1]] 16; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK1]], label [[VEC_EPILOG_PH:%.*]], label [[VECTOR_PH:%.*]] 17; CHECK: vector.ph: 18; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 19; CHECK-NEXT: [[TMP3:%.*]] = mul i64 [[TMP2]], 4 20; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]] 21; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 22; CHECK-NEXT: [[TMP18:%.*]] = call i64 @llvm.vscale.i64() 23; CHECK-NEXT: [[TMP19:%.*]] = mul i64 [[TMP18]], 4 24; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 25; CHECK: vector.body: 26; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 27; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <vscale x 2 x i64> [ insertelement (<vscale x 2 x i64> zeroinitializer, i64 5, i32 0), [[VECTOR_PH]] ], [ [[TMP16:%.*]], [[VECTOR_BODY]] ] 28; CHECK-NEXT: [[VEC_PHI2:%.*]] = phi <vscale x 2 x i64> [ zeroinitializer, [[VECTOR_PH]] ], [ [[TMP17:%.*]], [[VECTOR_BODY]] ] 29; CHECK-NEXT: [[TMP4:%.*]] = add i64 [[INDEX]], 0 30; CHECK-NEXT: [[TMP10:%.*]] = getelementptr inbounds i64, ptr [[A:%.*]], i64 [[TMP4]] 31; CHECK-NEXT: [[TMP12:%.*]] = getelementptr inbounds i64, ptr [[TMP10]], i32 0 32; CHECK-NEXT: [[TMP13:%.*]] = call i64 @llvm.vscale.i64() 33; CHECK-NEXT: [[TMP14:%.*]] = mul i64 [[TMP13]], 2 34; CHECK-NEXT: [[TMP15:%.*]] = getelementptr inbounds i64, ptr [[TMP10]], i64 [[TMP14]] 35; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <vscale x 2 x i64>, ptr [[TMP12]], align 8 36; CHECK-NEXT: [[WIDE_LOAD3:%.*]] = load <vscale x 2 x i64>, ptr [[TMP15]], align 8 37; CHECK-NEXT: [[TMP16]] = add <vscale x 2 x i64> [[WIDE_LOAD]], [[VEC_PHI]] 38; CHECK-NEXT: [[TMP17]] = add <vscale x 2 x i64> [[WIDE_LOAD3]], [[VEC_PHI2]] 39; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP19]] 40; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 41; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 42; CHECK: middle.block: 43; CHECK-NEXT: [[BIN_RDX:%.*]] = add <vscale x 2 x i64> [[TMP17]], [[TMP16]] 44; CHECK-NEXT: [[TMP21:%.*]] = call i64 @llvm.vector.reduce.add.nxv2i64(<vscale x 2 x i64> [[BIN_RDX]]) 45; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 46; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[VEC_EPILOG_ITER_CHECK:%.*]] 47; CHECK: vec.epilog.iter.check: 48; CHECK-NEXT: [[N_VEC_REMAINING:%.*]] = sub i64 [[N]], [[N_VEC]] 49; CHECK-NEXT: [[MIN_EPILOG_ITERS_CHECK:%.*]] = icmp ult i64 [[N_VEC_REMAINING]], 2 50; CHECK-NEXT: br i1 [[MIN_EPILOG_ITERS_CHECK]], label [[VEC_EPILOG_SCALAR_PH]], label [[VEC_EPILOG_PH]] 51; CHECK: vec.epilog.ph: 52; CHECK-NEXT: [[VEC_EPILOG_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ], [ 0, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 53; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi i64 [ [[TMP21]], [[VEC_EPILOG_ITER_CHECK]] ], [ 5, [[VECTOR_MAIN_LOOP_ITER_CHECK]] ] 54; CHECK-NEXT: [[N_MOD_VF4:%.*]] = urem i64 [[N]], 2 55; CHECK-NEXT: [[N_VEC5:%.*]] = sub i64 [[N]], [[N_MOD_VF4]] 56; CHECK-NEXT: [[TMP22:%.*]] = insertelement <2 x i64> zeroinitializer, i64 [[BC_MERGE_RDX]], i32 0 57; CHECK-NEXT: br label [[VEC_EPILOG_VECTOR_BODY:%.*]] 58; CHECK: vec.epilog.vector.body: 59; CHECK-NEXT: [[INDEX7:%.*]] = phi i64 [ [[VEC_EPILOG_RESUME_VAL]], [[VEC_EPILOG_PH]] ], [ [[INDEX_NEXT10:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 60; CHECK-NEXT: [[VEC_PHI8:%.*]] = phi <2 x i64> [ [[TMP22]], [[VEC_EPILOG_PH]] ], [ [[TMP26:%.*]], [[VEC_EPILOG_VECTOR_BODY]] ] 61; CHECK-NEXT: [[TMP23:%.*]] = add i64 [[INDEX7]], 0 62; CHECK-NEXT: [[TMP24:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[TMP23]] 63; CHECK-NEXT: [[TMP25:%.*]] = getelementptr inbounds i64, ptr [[TMP24]], i32 0 64; CHECK-NEXT: [[WIDE_LOAD9:%.*]] = load <2 x i64>, ptr [[TMP25]], align 8 65; CHECK-NEXT: [[TMP26]] = add <2 x i64> [[WIDE_LOAD9]], [[VEC_PHI8]] 66; CHECK-NEXT: [[INDEX_NEXT10]] = add nuw i64 [[INDEX7]], 2 67; CHECK-NEXT: [[TMP27:%.*]] = icmp eq i64 [[INDEX_NEXT10]], [[N_VEC5]] 68; CHECK-NEXT: br i1 [[TMP27]], label [[VEC_EPILOG_MIDDLE_BLOCK:%.*]], label [[VEC_EPILOG_VECTOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 69; CHECK: vec.epilog.middle.block: 70; CHECK-NEXT: [[TMP28:%.*]] = call i64 @llvm.vector.reduce.add.v2i64(<2 x i64> [[TMP26]]) 71; CHECK-NEXT: [[CMP_N6:%.*]] = icmp eq i64 [[N]], [[N_VEC5]] 72; CHECK-NEXT: br i1 [[CMP_N6]], label [[FOR_END]], label [[VEC_EPILOG_SCALAR_PH]] 73; CHECK: vec.epilog.scalar.ph: 74; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC5]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 0, [[ITER_CHECK:%.*]] ], [ [[N_VEC]], [[VEC_EPILOG_ITER_CHECK]] ] 75; CHECK-NEXT: [[BC_MERGE_RDX11:%.*]] = phi i64 [ [[TMP28]], [[VEC_EPILOG_MIDDLE_BLOCK]] ], [ 5, [[ITER_CHECK]] ], [ [[TMP21]], [[VEC_EPILOG_ITER_CHECK]] ] 76; CHECK-NEXT: br label [[FOR_BODY:%.*]] 77; CHECK: for.body: 78; CHECK-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 79; CHECK-NEXT: [[SUM:%.*]] = phi i64 [ [[BC_MERGE_RDX11]], [[VEC_EPILOG_SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ] 80; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i64, ptr [[A]], i64 [[IV]] 81; CHECK-NEXT: [[TMP29:%.*]] = load i64, ptr [[ARRAYIDX]], align 8 82; CHECK-NEXT: [[ADD]] = add i64 [[TMP29]], [[SUM]] 83; CHECK-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 84; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 85; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP4:![0-9]+]] 86; CHECK: for.end: 87; CHECK-NEXT: [[ADD_LCSSA:%.*]] = phi i64 [ [[ADD]], [[FOR_BODY]] ], [ [[TMP21]], [[MIDDLE_BLOCK]] ], [ [[TMP28]], [[VEC_EPILOG_MIDDLE_BLOCK]] ] 88; CHECK-NEXT: ret i64 [[ADD_LCSSA]] 89; 90entry: 91 br label %for.body 92 93for.body: 94 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 95 %sum = phi i64 [ 5, %entry ], [ %add, %for.body ] 96 %arrayidx = getelementptr inbounds i64, ptr %a, i64 %iv 97 %0 = load i64, ptr %arrayidx 98 %add = add i64 %0, %sum 99 %iv.next = add nuw nsw i64 %iv, 1 100 %exitcond.not = icmp eq i64 %iv.next, %N 101 br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0 102 103for.end: 104 ret i64 %add 105} 106 107!0 = distinct !{!0, !1, !2} 108!1 = !{!"llvm.loop.interleave.count", i32 2} 109!2 = !{!"llvm.loop.vectorize.scalable.enable", i1 true} 110