1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -instcombine -simplifycfg < %s -S -o - | FileCheck %s --check-prefix=CHECK 3; RUN: opt -loop-vectorize -debug-only=loop-vectorize -disable-output < %s 2>&1 | FileCheck %s --check-prefix=CHECK-COST 4; REQUIRES: asserts 5 6target datalayout = "e-m:e-p:32:32-Fi8-i64:64-v128:64:128-a:0:32-n32-S64" 7target triple = "thumbv8.1m.main-arm-none-eabi" 8 9; CHECK-COST-LABEL: test 10; CHECK-COST: LV: Found an estimated cost of 1 for VF 1 For instruction: %or.cond = select i1 %cmp2, i1 true, i1 %cmp3 11; CHECK-COST: LV: Found an estimated cost of 2 for VF 2 For instruction: %or.cond = select i1 %cmp2, i1 true, i1 %cmp3 12; CHECK-COST: LV: Found an estimated cost of 2 for VF 4 For instruction: %or.cond = select i1 %cmp2, i1 true, i1 %cmp3 13 14define float @test(float* nocapture readonly %pA, float* nocapture readonly %pB, i32 %blockSize) #0 { 15; CHECK-LABEL: @test( 16; CHECK-NEXT: entry: 17; CHECK-NEXT: [[CMP_NOT16:%.*]] = icmp eq i32 [[BLOCKSIZE:%.*]], 0 18; CHECK-NEXT: br i1 [[CMP_NOT16]], label [[WHILE_END:%.*]], label [[WHILE_BODY_PREHEADER:%.*]] 19; CHECK: while.body.preheader: 20; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[BLOCKSIZE]], 4 21; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 22; CHECK: vector.ph: 23; CHECK-NEXT: [[N_VEC:%.*]] = and i32 [[BLOCKSIZE]], -4 24; CHECK-NEXT: [[IND_END:%.*]] = getelementptr float, float* [[PA:%.*]], i32 [[N_VEC]] 25; CHECK-NEXT: [[IND_END2:%.*]] = getelementptr float, float* [[PB:%.*]], i32 [[N_VEC]] 26; CHECK-NEXT: [[IND_END4:%.*]] = and i32 [[BLOCKSIZE]], 3 27; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 28; CHECK: vector.body: 29; CHECK-NEXT: [[INDEX:%.*]] = phi i32 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 30; CHECK-NEXT: [[VEC_PHI:%.*]] = phi <4 x float> [ zeroinitializer, [[VECTOR_PH]] ], [ [[PREDPHI:%.*]], [[VECTOR_BODY]] ] 31; CHECK-NEXT: [[NEXT_GEP:%.*]] = getelementptr float, float* [[PA]], i32 [[INDEX]] 32; CHECK-NEXT: [[NEXT_GEP5:%.*]] = getelementptr float, float* [[PB]], i32 [[INDEX]] 33; CHECK-NEXT: [[TMP0:%.*]] = bitcast float* [[NEXT_GEP]] to <4 x float>* 34; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, <4 x float>* [[TMP0]], align 4 35; CHECK-NEXT: [[TMP1:%.*]] = bitcast float* [[NEXT_GEP5]] to <4 x float>* 36; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <4 x float>, <4 x float>* [[TMP1]], align 4 37; CHECK-NEXT: [[TMP2:%.*]] = fcmp fast oeq <4 x float> [[WIDE_LOAD]], zeroinitializer 38; CHECK-NEXT: [[TMP3:%.*]] = fcmp fast oeq <4 x float> [[WIDE_LOAD6]], zeroinitializer 39; CHECK-NEXT: [[TMP4:%.*]] = select <4 x i1> [[TMP2]], <4 x i1> [[TMP3]], <4 x i1> zeroinitializer 40; CHECK-NEXT: [[TMP5:%.*]] = call fast <4 x float> @llvm.fabs.v4f32(<4 x float> [[WIDE_LOAD]]) 41; CHECK-NEXT: [[TMP6:%.*]] = call fast <4 x float> @llvm.fabs.v4f32(<4 x float> [[WIDE_LOAD6]]) 42; CHECK-NEXT: [[TMP7:%.*]] = fadd fast <4 x float> [[TMP6]], [[TMP5]] 43; CHECK-NEXT: [[TMP8:%.*]] = fsub fast <4 x float> [[WIDE_LOAD]], [[WIDE_LOAD6]] 44; CHECK-NEXT: [[TMP9:%.*]] = call fast <4 x float> @llvm.fabs.v4f32(<4 x float> [[TMP8]]) 45; CHECK-NEXT: [[TMP10:%.*]] = fdiv fast <4 x float> [[TMP9]], [[TMP7]] 46; CHECK-NEXT: [[TMP11:%.*]] = select <4 x i1> [[TMP4]], <4 x float> <float -0.000000e+00, float -0.000000e+00, float -0.000000e+00, float -0.000000e+00>, <4 x float> [[TMP10]] 47; CHECK-NEXT: [[PREDPHI]] = fadd fast <4 x float> [[VEC_PHI]], [[TMP11]] 48; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i32 [[INDEX]], 4 49; CHECK-NEXT: [[TMP12:%.*]] = icmp eq i32 [[INDEX_NEXT]], [[N_VEC]] 50; CHECK-NEXT: br i1 [[TMP12]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 51; CHECK: middle.block: 52; CHECK-NEXT: [[TMP13:%.*]] = call fast float @llvm.vector.reduce.fadd.v4f32(float -0.000000e+00, <4 x float> [[PREDPHI]]) 53; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i32 [[N_VEC]], [[BLOCKSIZE]] 54; CHECK-NEXT: br i1 [[CMP_N]], label [[WHILE_END]], label [[SCALAR_PH]] 55; CHECK: scalar.ph: 56; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi float* [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[PA]], [[WHILE_BODY_PREHEADER]] ] 57; CHECK-NEXT: [[BC_RESUME_VAL1:%.*]] = phi float* [ [[IND_END2]], [[MIDDLE_BLOCK]] ], [ [[PB]], [[WHILE_BODY_PREHEADER]] ] 58; CHECK-NEXT: [[BC_RESUME_VAL3:%.*]] = phi i32 [ [[IND_END4]], [[MIDDLE_BLOCK]] ], [ [[BLOCKSIZE]], [[WHILE_BODY_PREHEADER]] ] 59; CHECK-NEXT: [[BC_MERGE_RDX:%.*]] = phi float [ [[TMP13]], [[MIDDLE_BLOCK]] ], [ 0.000000e+00, [[WHILE_BODY_PREHEADER]] ] 60; CHECK-NEXT: br label [[WHILE_BODY:%.*]] 61; CHECK: while.body: 62; CHECK-NEXT: [[PA_ADDR_020:%.*]] = phi float* [ [[INCDEC_PTR:%.*]], [[IF_END:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 63; CHECK-NEXT: [[PB_ADDR_019:%.*]] = phi float* [ [[INCDEC_PTR1:%.*]], [[IF_END]] ], [ [[BC_RESUME_VAL1]], [[SCALAR_PH]] ] 64; CHECK-NEXT: [[BLOCKSIZE_ADDR_018:%.*]] = phi i32 [ [[DEC:%.*]], [[IF_END]] ], [ [[BC_RESUME_VAL3]], [[SCALAR_PH]] ] 65; CHECK-NEXT: [[ACCUM_017:%.*]] = phi float [ [[ACCUM_1:%.*]], [[IF_END]] ], [ [[BC_MERGE_RDX]], [[SCALAR_PH]] ] 66; CHECK-NEXT: [[INCDEC_PTR]] = getelementptr inbounds float, float* [[PA_ADDR_020]], i32 1 67; CHECK-NEXT: [[TMP14:%.*]] = load float, float* [[PA_ADDR_020]], align 4 68; CHECK-NEXT: [[INCDEC_PTR1]] = getelementptr inbounds float, float* [[PB_ADDR_019]], i32 1 69; CHECK-NEXT: [[TMP15:%.*]] = load float, float* [[PB_ADDR_019]], align 4 70; CHECK-NEXT: [[CMP2:%.*]] = fcmp fast une float [[TMP14]], 0.000000e+00 71; CHECK-NEXT: [[CMP3:%.*]] = fcmp fast une float [[TMP15]], 0.000000e+00 72; CHECK-NEXT: [[OR_COND:%.*]] = select i1 [[CMP2]], i1 true, i1 [[CMP3]] 73; CHECK-NEXT: br i1 [[OR_COND]], label [[IF_THEN:%.*]], label [[IF_END]] 74; CHECK: if.then: 75; CHECK-NEXT: [[TMP16:%.*]] = tail call fast float @llvm.fabs.f32(float [[TMP14]]) 76; CHECK-NEXT: [[TMP17:%.*]] = tail call fast float @llvm.fabs.f32(float [[TMP15]]) 77; CHECK-NEXT: [[ADD:%.*]] = fadd fast float [[TMP17]], [[TMP16]] 78; CHECK-NEXT: [[SUB:%.*]] = fsub fast float [[TMP14]], [[TMP15]] 79; CHECK-NEXT: [[TMP18:%.*]] = tail call fast float @llvm.fabs.f32(float [[SUB]]) 80; CHECK-NEXT: [[DIV:%.*]] = fdiv fast float [[TMP18]], [[ADD]] 81; CHECK-NEXT: [[ADD4:%.*]] = fadd fast float [[DIV]], [[ACCUM_017]] 82; CHECK-NEXT: br label [[IF_END]] 83; CHECK: if.end: 84; CHECK-NEXT: [[ACCUM_1]] = phi float [ [[ADD4]], [[IF_THEN]] ], [ [[ACCUM_017]], [[WHILE_BODY]] ] 85; CHECK-NEXT: [[DEC]] = add i32 [[BLOCKSIZE_ADDR_018]], -1 86; CHECK-NEXT: [[CMP_NOT:%.*]] = icmp eq i32 [[DEC]], 0 87; CHECK-NEXT: br i1 [[CMP_NOT]], label [[WHILE_END]], label [[WHILE_BODY]], !llvm.loop [[LOOP2:![0-9]+]] 88; CHECK: while.end: 89; CHECK-NEXT: [[ACCUM_0_LCSSA:%.*]] = phi float [ 0.000000e+00, [[ENTRY:%.*]] ], [ [[ACCUM_1]], [[IF_END]] ], [ [[TMP13]], [[MIDDLE_BLOCK]] ] 90; CHECK-NEXT: ret float [[ACCUM_0_LCSSA]] 91; 92entry: 93 %cmp.not16 = icmp eq i32 %blockSize, 0 94 br i1 %cmp.not16, label %while.end, label %while.body 95 96while.body: ; preds = %entry, %if.end 97 %pA.addr.020 = phi float* [ %incdec.ptr, %if.end ], [ %pA, %entry ] 98 %pB.addr.019 = phi float* [ %incdec.ptr1, %if.end ], [ %pB, %entry ] 99 %blockSize.addr.018 = phi i32 [ %dec, %if.end ], [ %blockSize, %entry ] 100 %accum.017 = phi float [ %accum.1, %if.end ], [ 0.000000e+00, %entry ] 101 %incdec.ptr = getelementptr inbounds float, float* %pA.addr.020, i32 1 102 %0 = load float, float* %pA.addr.020, align 4 103 %incdec.ptr1 = getelementptr inbounds float, float* %pB.addr.019, i32 1 104 %1 = load float, float* %pB.addr.019, align 4 105 %cmp2 = fcmp fast une float %0, 0.000000e+00 106 %cmp3 = fcmp fast une float %1, 0.000000e+00 107 %or.cond = select i1 %cmp2, i1 true, i1 %cmp3 108 br i1 %or.cond, label %if.then, label %if.end 109 110if.then: ; preds = %while.body 111 %2 = tail call fast float @llvm.fabs.f32(float %0) 112 %3 = tail call fast float @llvm.fabs.f32(float %1) 113 %add = fadd fast float %3, %2 114 %sub = fsub fast float %0, %1 115 %4 = tail call fast float @llvm.fabs.f32(float %sub) 116 %div = fdiv fast float %4, %add 117 %add4 = fadd fast float %div, %accum.017 118 br label %if.end 119 120if.end: ; preds = %while.body, %if.then 121 %accum.1 = phi float [ %add4, %if.then ], [ %accum.017, %while.body ] 122 %dec = add i32 %blockSize.addr.018, -1 123 %cmp.not = icmp eq i32 %dec, 0 124 br i1 %cmp.not, label %while.end, label %while.body 125 126while.end: ; preds = %if.end, %entry 127 %accum.0.lcssa = phi float [ 0.000000e+00, %entry ], [ %accum.1, %if.end ] 128 ret float %accum.0.lcssa 129} 130 131declare float @llvm.fabs.f32(float) 132 133attributes #0 = { "target-features"="+mve.fp" } 134