1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -passes=loop-vectorize -hints-allow-reordering=false -S < %s | FileCheck %s --check-prefix=CHECK-ORDERED 3; RUN: opt -passes=loop-vectorize -S < %s | FileCheck %s --check-prefix=CHECK-UNORDERED 4; RUN: opt -passes=loop-vectorize -force-ordered-reductions=false -hints-allow-reordering=false -S < %s | FileCheck %s --check-prefix=CHECK-NOT-VECTORIZED 5; RUN: opt -passes=loop-vectorize -force-ordered-reductions=true -hints-allow-reordering=false -S < %s | FileCheck %s --check-prefix=CHECK-ORDERED 6 7 8target triple = "riscv64" 9 10define float @fadd(ptr noalias nocapture readonly %a, i64 %n) #0 { 11; CHECK-ORDERED-LABEL: @fadd( 12; CHECK-ORDERED-NEXT: entry: 13; CHECK-ORDERED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4 14; CHECK-ORDERED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 15; CHECK-ORDERED: vector.ph: 16; CHECK-ORDERED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 17; CHECK-ORDERED-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 18; CHECK-ORDERED-NEXT: br label [[VECTOR_BODY:%.*]] 19; CHECK-ORDERED: vector.body: 20; CHECK-ORDERED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 21; CHECK-ORDERED-NEXT: [[VEC_PHI:%.*]] = phi float [ 0.000000e+00, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] 22; CHECK-ORDERED-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 23; CHECK-ORDERED-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[TMP0]] 24; CHECK-ORDERED-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i32 0 25; CHECK-ORDERED-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 4 26; CHECK-ORDERED-NEXT: [[TMP3]] = call float @llvm.vector.reduce.fadd.v4f32(float [[VEC_PHI]], <4 x float> [[WIDE_LOAD]]) 27; CHECK-ORDERED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 28; CHECK-ORDERED-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 29; CHECK-ORDERED-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 30; CHECK-ORDERED: middle.block: 31; CHECK-ORDERED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 32; CHECK-ORDERED-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 33; CHECK-ORDERED: scalar.ph: 34; CHECK-ORDERED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 35; CHECK-ORDERED-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP3]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] 36; CHECK-ORDERED-NEXT: br label [[FOR_BODY:%.*]] 37; CHECK-ORDERED: for.body: 38; CHECK-ORDERED-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 39; CHECK-ORDERED-NEXT: [[SUM_07:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ] 40; CHECK-ORDERED-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 41; CHECK-ORDERED-NEXT: [[TMP5:%.*]] = load float, ptr [[ARRAYIDX]], align 4 42; CHECK-ORDERED-NEXT: [[ADD]] = fadd float [[TMP5]], [[SUM_07]] 43; CHECK-ORDERED-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 44; CHECK-ORDERED-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 45; CHECK-ORDERED-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 46; CHECK-ORDERED: for.end: 47; CHECK-ORDERED-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], [[FOR_BODY]] ], [ [[TMP3]], [[MIDDLE_BLOCK]] ] 48; CHECK-ORDERED-NEXT: ret float [[ADD_LCSSA]] 49; 50; CHECK-UNORDERED-LABEL: @fadd( 51; CHECK-UNORDERED-NEXT: entry: 52; CHECK-UNORDERED-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N:%.*]], 4 53; CHECK-UNORDERED-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 54; CHECK-UNORDERED: vector.ph: 55; CHECK-UNORDERED-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], 4 56; CHECK-UNORDERED-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 57; CHECK-UNORDERED-NEXT: br label [[VECTOR_BODY:%.*]] 58; CHECK-UNORDERED: vector.body: 59; CHECK-UNORDERED-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 60; CHECK-UNORDERED-NEXT: [[VEC_PHI:%.*]] = phi <4 x float> [ <float 0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, [[VECTOR_PH]] ], [ [[TMP3:%.*]], [[VECTOR_BODY]] ] 61; CHECK-UNORDERED-NEXT: [[TMP0:%.*]] = add i64 [[INDEX]], 0 62; CHECK-UNORDERED-NEXT: [[TMP1:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[TMP0]] 63; CHECK-UNORDERED-NEXT: [[TMP2:%.*]] = getelementptr inbounds float, ptr [[TMP1]], i32 0 64; CHECK-UNORDERED-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, ptr [[TMP2]], align 4 65; CHECK-UNORDERED-NEXT: [[TMP3]] = fadd <4 x float> [[WIDE_LOAD]], [[VEC_PHI]] 66; CHECK-UNORDERED-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 4 67; CHECK-UNORDERED-NEXT: [[TMP4:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 68; CHECK-UNORDERED-NEXT: br i1 [[TMP4]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 69; CHECK-UNORDERED: middle.block: 70; CHECK-UNORDERED-NEXT: [[TMP5:%.*]] = call float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[TMP3]]) 71; CHECK-UNORDERED-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N]], [[N_VEC]] 72; CHECK-UNORDERED-NEXT: br i1 [[CMP_N]], label [[FOR_END:%.*]], label [[SCALAR_PH]] 73; CHECK-UNORDERED: scalar.ph: 74; CHECK-UNORDERED-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 75; CHECK-UNORDERED-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP5]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[ENTRY]] ] 76; CHECK-UNORDERED-NEXT: br label [[FOR_BODY:%.*]] 77; CHECK-UNORDERED: for.body: 78; CHECK-UNORDERED-NEXT: [[IV:%.*]] = phi i64 [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 79; CHECK-UNORDERED-NEXT: [[SUM_07:%.*]] = phi float [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ], [ [[ADD:%.*]], [[FOR_BODY]] ] 80; CHECK-UNORDERED-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A]], i64 [[IV]] 81; CHECK-UNORDERED-NEXT: [[TMP6:%.*]] = load float, ptr [[ARRAYIDX]], align 4 82; CHECK-UNORDERED-NEXT: [[ADD]] = fadd float [[TMP6]], [[SUM_07]] 83; CHECK-UNORDERED-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 84; CHECK-UNORDERED-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N]] 85; CHECK-UNORDERED-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 86; CHECK-UNORDERED: for.end: 87; CHECK-UNORDERED-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], [[FOR_BODY]] ], [ [[TMP5]], [[MIDDLE_BLOCK]] ] 88; CHECK-UNORDERED-NEXT: ret float [[ADD_LCSSA]] 89; 90; CHECK-NOT-VECTORIZED-LABEL: @fadd( 91; CHECK-NOT-VECTORIZED-NEXT: entry: 92; CHECK-NOT-VECTORIZED-NEXT: br label [[FOR_BODY:%.*]] 93; CHECK-NOT-VECTORIZED: for.body: 94; CHECK-NOT-VECTORIZED-NEXT: [[IV:%.*]] = phi i64 [ 0, [[ENTRY:%.*]] ], [ [[IV_NEXT:%.*]], [[FOR_BODY]] ] 95; CHECK-NOT-VECTORIZED-NEXT: [[SUM_07:%.*]] = phi float [ 0.000000e+00, [[ENTRY]] ], [ [[ADD:%.*]], [[FOR_BODY]] ] 96; CHECK-NOT-VECTORIZED-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, ptr [[A:%.*]], i64 [[IV]] 97; CHECK-NOT-VECTORIZED-NEXT: [[TMP0:%.*]] = load float, ptr [[ARRAYIDX]], align 4 98; CHECK-NOT-VECTORIZED-NEXT: [[ADD]] = fadd float [[TMP0]], [[SUM_07]] 99; CHECK-NOT-VECTORIZED-NEXT: [[IV_NEXT]] = add nuw nsw i64 [[IV]], 1 100; CHECK-NOT-VECTORIZED-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[IV_NEXT]], [[N:%.*]] 101; CHECK-NOT-VECTORIZED-NEXT: br i1 [[EXITCOND_NOT]], label [[FOR_END:%.*]], label [[FOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 102; CHECK-NOT-VECTORIZED: for.end: 103; CHECK-NOT-VECTORIZED-NEXT: [[ADD_LCSSA:%.*]] = phi float [ [[ADD]], [[FOR_BODY]] ] 104; CHECK-NOT-VECTORIZED-NEXT: ret float [[ADD_LCSSA]] 105; 106entry: 107 br label %for.body 108 109for.body: 110 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 111 %sum.07 = phi float [ 0.000000e+00, %entry ], [ %add, %for.body ] 112 %arrayidx = getelementptr inbounds float, ptr %a, i64 %iv 113 %0 = load float, ptr %arrayidx, align 4 114 %add = fadd float %0, %sum.07 115 %iv.next = add nuw nsw i64 %iv, 1 116 %exitcond.not = icmp eq i64 %iv.next, %n 117 br i1 %exitcond.not, label %for.end, label %for.body, !llvm.loop !0 118 119for.end: 120 ret float %add 121} 122 123attributes #0 = { "target-features"="+f,+v" } 124!0 = distinct !{!0, !1, !2} 125!1 = !{!"llvm.loop.vectorize.width", i32 4} 126!2 = !{!"llvm.loop.interleave.count", i32 1} 127