1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -passes=slp-vectorizer -S -slp-schedule-budget=16 -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck --check-prefix LOBUDGET %s 3; RUN: opt < %s -passes=slp-vectorizer -S -slp-schedule-budget=32 -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck --check-prefix HIBUDGET %s 4 5target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" 6target triple = "x86_64-apple-macosx10.9.0" 7 8; Test if the budget for the scheduling region size works. 9; We test with a reduced budget of 16 which should prevent vectorizing the loads. 10 11declare void @unknown() 12 13define void @test(ptr %a, ptr %b, ptr %c, ptr %d) { 14; LOBUDGET-LABEL: @test( 15; LOBUDGET-NEXT: entry: 16; LOBUDGET-NEXT: [[L0:%.*]] = load float, ptr [[A:%.*]], align 4 17; LOBUDGET-NEXT: [[A1:%.*]] = getelementptr inbounds float, ptr [[A]], i64 1 18; LOBUDGET-NEXT: [[L1:%.*]] = load float, ptr [[A1]], align 4 19; LOBUDGET-NEXT: [[A2:%.*]] = getelementptr inbounds float, ptr [[A]], i64 2 20; LOBUDGET-NEXT: [[L2:%.*]] = load float, ptr [[A2]], align 4 21; LOBUDGET-NEXT: [[A3:%.*]] = getelementptr inbounds float, ptr [[A]], i64 3 22; LOBUDGET-NEXT: [[L3:%.*]] = load float, ptr [[A3]], align 4 23; LOBUDGET-NEXT: [[L00:%.*]] = fadd float [[L0]], [[L0]] 24; LOBUDGET-NEXT: [[L10:%.*]] = fadd float [[L1]], [[L1]] 25; LOBUDGET-NEXT: [[L20:%.*]] = fadd float [[L2]], [[L2]] 26; LOBUDGET-NEXT: [[L30:%.*]] = fadd float [[L3]], [[L3]] 27; LOBUDGET-NEXT: call void @unknown() 28; LOBUDGET-NEXT: call void @unknown() 29; LOBUDGET-NEXT: call void @unknown() 30; LOBUDGET-NEXT: call void @unknown() 31; LOBUDGET-NEXT: call void @unknown() 32; LOBUDGET-NEXT: call void @unknown() 33; LOBUDGET-NEXT: call void @unknown() 34; LOBUDGET-NEXT: call void @unknown() 35; LOBUDGET-NEXT: call void @unknown() 36; LOBUDGET-NEXT: call void @unknown() 37; LOBUDGET-NEXT: call void @unknown() 38; LOBUDGET-NEXT: call void @unknown() 39; LOBUDGET-NEXT: call void @unknown() 40; LOBUDGET-NEXT: call void @unknown() 41; LOBUDGET-NEXT: call void @unknown() 42; LOBUDGET-NEXT: call void @unknown() 43; LOBUDGET-NEXT: call void @unknown() 44; LOBUDGET-NEXT: call void @unknown() 45; LOBUDGET-NEXT: call void @unknown() 46; LOBUDGET-NEXT: call void @unknown() 47; LOBUDGET-NEXT: call void @unknown() 48; LOBUDGET-NEXT: call void @unknown() 49; LOBUDGET-NEXT: call void @unknown() 50; LOBUDGET-NEXT: call void @unknown() 51; LOBUDGET-NEXT: call void @unknown() 52; LOBUDGET-NEXT: call void @unknown() 53; LOBUDGET-NEXT: call void @unknown() 54; LOBUDGET-NEXT: call void @unknown() 55; LOBUDGET-NEXT: store float [[L00]], ptr [[B:%.*]], align 4 56; LOBUDGET-NEXT: [[B1:%.*]] = getelementptr inbounds float, ptr [[B]], i64 1 57; LOBUDGET-NEXT: store float [[L10]], ptr [[B1]], align 4 58; LOBUDGET-NEXT: [[B2:%.*]] = getelementptr inbounds float, ptr [[B]], i64 2 59; LOBUDGET-NEXT: store float [[L20]], ptr [[B2]], align 4 60; LOBUDGET-NEXT: [[B3:%.*]] = getelementptr inbounds float, ptr [[B]], i64 3 61; LOBUDGET-NEXT: store float [[L30]], ptr [[B3]], align 4 62; LOBUDGET-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[C:%.*]], align 4 63; LOBUDGET-NEXT: store <4 x float> [[TMP0]], ptr [[D:%.*]], align 4 64; LOBUDGET-NEXT: ret void 65; 66; HIBUDGET-LABEL: @test( 67; HIBUDGET-NEXT: entry: 68; HIBUDGET-NEXT: [[TMP0:%.*]] = load <4 x float>, ptr [[A:%.*]], align 4 69; HIBUDGET-NEXT: [[TMP1:%.*]] = fadd <4 x float> [[TMP0]], [[TMP0]] 70; HIBUDGET-NEXT: call void @unknown() 71; HIBUDGET-NEXT: call void @unknown() 72; HIBUDGET-NEXT: call void @unknown() 73; HIBUDGET-NEXT: call void @unknown() 74; HIBUDGET-NEXT: call void @unknown() 75; HIBUDGET-NEXT: call void @unknown() 76; HIBUDGET-NEXT: call void @unknown() 77; HIBUDGET-NEXT: call void @unknown() 78; HIBUDGET-NEXT: call void @unknown() 79; HIBUDGET-NEXT: call void @unknown() 80; HIBUDGET-NEXT: call void @unknown() 81; HIBUDGET-NEXT: call void @unknown() 82; HIBUDGET-NEXT: call void @unknown() 83; HIBUDGET-NEXT: call void @unknown() 84; HIBUDGET-NEXT: call void @unknown() 85; HIBUDGET-NEXT: call void @unknown() 86; HIBUDGET-NEXT: call void @unknown() 87; HIBUDGET-NEXT: call void @unknown() 88; HIBUDGET-NEXT: call void @unknown() 89; HIBUDGET-NEXT: call void @unknown() 90; HIBUDGET-NEXT: call void @unknown() 91; HIBUDGET-NEXT: call void @unknown() 92; HIBUDGET-NEXT: call void @unknown() 93; HIBUDGET-NEXT: call void @unknown() 94; HIBUDGET-NEXT: call void @unknown() 95; HIBUDGET-NEXT: call void @unknown() 96; HIBUDGET-NEXT: call void @unknown() 97; HIBUDGET-NEXT: call void @unknown() 98; HIBUDGET-NEXT: store <4 x float> [[TMP1]], ptr [[B:%.*]], align 4 99; HIBUDGET-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[C:%.*]], align 4 100; HIBUDGET-NEXT: store <4 x float> [[TMP2]], ptr [[D:%.*]], align 4 101; HIBUDGET-NEXT: ret void 102; 103entry: 104 ; Don't vectorize these loads (with the reduced budget). 105 %l0 = load float, ptr %a 106 %a1 = getelementptr inbounds float, ptr %a, i64 1 107 %l1 = load float, ptr %a1 108 %a2 = getelementptr inbounds float, ptr %a, i64 2 109 %l2 = load float, ptr %a2 110 %a3 = getelementptr inbounds float, ptr %a, i64 3 111 %l3 = load float, ptr %a3 112 %l00 = fadd float %l0, %l0 113 %l10 = fadd float %l1, %l1 114 %l20 = fadd float %l2, %l2 115 %l30 = fadd float %l3, %l3 116 117 ; some unrelated instructions inbetween to enlarge the scheduling region 118 call void @unknown() 119 call void @unknown() 120 call void @unknown() 121 call void @unknown() 122 call void @unknown() 123 call void @unknown() 124 call void @unknown() 125 call void @unknown() 126 call void @unknown() 127 call void @unknown() 128 call void @unknown() 129 call void @unknown() 130 call void @unknown() 131 call void @unknown() 132 call void @unknown() 133 call void @unknown() 134 call void @unknown() 135 call void @unknown() 136 call void @unknown() 137 call void @unknown() 138 call void @unknown() 139 call void @unknown() 140 call void @unknown() 141 call void @unknown() 142 call void @unknown() 143 call void @unknown() 144 call void @unknown() 145 call void @unknown() 146 147 ; Don't vectorize these stores because their operands are too far away (with 148 ; the reduced budget). 149 store float %l00, ptr %b 150 %b1 = getelementptr inbounds float, ptr %b, i64 1 151 store float %l10, ptr %b1 152 %b2 = getelementptr inbounds float, ptr %b, i64 2 153 store float %l20, ptr %b2 154 %b3 = getelementptr inbounds float, ptr %b, i64 3 155 store float %l30, ptr %b3 156 157 ; But still vectorize the following instructions, because even if the budget 158 ; is exceeded there is a minimum region size. 159 %l4 = load float, ptr %c 160 %c1 = getelementptr inbounds float, ptr %c, i64 1 161 %l5 = load float, ptr %c1 162 %c2 = getelementptr inbounds float, ptr %c, i64 2 163 %l6 = load float, ptr %c2 164 %c3 = getelementptr inbounds float, ptr %c, i64 3 165 %l7 = load float, ptr %c3 166 167 store float %l4, ptr %d 168 %d1 = getelementptr inbounds float, ptr %d, i64 1 169 store float %l5, ptr %d1 170 %d2 = getelementptr inbounds float, ptr %d, i64 2 171 store float %l6, ptr %d2 172 %d3 = getelementptr inbounds float, ptr %d, i64 3 173 store float %l7, ptr %d3 174 175 ret void 176} 177 178