1; RUN: opt %loadNPMPolly -passes=polly-codegen -S \ 2; RUN: -polly-codegen-add-debug-printing \ 3; RUN: -polly-ignore-aliasing < %s | FileCheck %s 4 5; #define N 10 6; void foo(float A[restrict], double B[restrict], char C[restrict], 7; int D[restrict], long E[restrict]) { 8; for (long i = 0; i < N; i++) 9; A[i] += B[i] + C[i] + D[i] + E[i]; 10; } 11; 12; int main() { 13; float A[N]; 14; double B[N]; 15; char C[N]; 16; int D[N]; 17; long E[N]; 18; 19; for (long i = 0; i < N; i++) { 20; __sync_synchronize(); 21; A[i] = B[i] = C[i] = D[i] = E[i] = 42; 22; } 23; 24; foo(A, B, C, D, E); 25; 26; return A[8]; 27; } 28 29; CHECK: @0 = private unnamed_addr addrspace(4) constant [11 x i8] c"Load from \00" 30; CHECK: @1 = private unnamed_addr addrspace(4) constant [3 x i8] c": \00" 31; CHECK: @2 = private unnamed_addr addrspace(4) constant [2 x i8] c"\0A\00" 32; CHECK: @3 = private unnamed_addr constant [12 x i8] c"%s%ld%s%f%s\00" 33; CHECK: @4 = private unnamed_addr addrspace(4) constant [11 x i8] c"Load from \00" 34; CHECK: @5 = private unnamed_addr addrspace(4) constant [3 x i8] c": \00" 35; CHECK: @6 = private unnamed_addr addrspace(4) constant [2 x i8] c"\0A\00" 36; CHECK: @7 = private unnamed_addr constant [13 x i8] c"%s%ld%s%ld%s\00" 37; CHECK: @8 = private unnamed_addr addrspace(4) constant [11 x i8] c"Load from \00" 38; CHECK: @9 = private unnamed_addr addrspace(4) constant [3 x i8] c": \00" 39; CHECK: @10 = private unnamed_addr addrspace(4) constant [2 x i8] c"\0A\00" 40; CHECK: @11 = private unnamed_addr constant [13 x i8] c"%s%ld%s%ld%s\00" 41; CHECK: @12 = private unnamed_addr addrspace(4) constant [11 x i8] c"Load from \00" 42; CHECK: @13 = private unnamed_addr addrspace(4) constant [3 x i8] c": \00" 43; CHECK: @14 = private unnamed_addr addrspace(4) constant [2 x i8] c"\0A\00" 44; CHECK: @15 = private unnamed_addr constant [13 x i8] c"%s%ld%s%ld%s\00" 45; CHECK: @16 = private unnamed_addr addrspace(4) constant [11 x i8] c"Load from \00" 46; CHECK: @17 = private unnamed_addr addrspace(4) constant [3 x i8] c": \00" 47; CHECK: @18 = private unnamed_addr addrspace(4) constant [2 x i8] c"\0A\00" 48; CHECK: @19 = private unnamed_addr constant [12 x i8] c"%s%ld%s%f%s\00" 49; CHECK: @20 = private unnamed_addr addrspace(4) constant [11 x i8] c"Store to \00" 50; CHECK: @21 = private unnamed_addr addrspace(4) constant [3 x i8] c": \00" 51; CHECK: @22 = private unnamed_addr addrspace(4) constant [2 x i8] c"\0A\00" 52; CHECK: @23 = private unnamed_addr constant [12 x i8] c"%s%ld%s%f%s\00" 53 54; CHECK: %0 = shl nuw nsw i64 %polly.indvar, 3 55; CHECK: %scevgep = getelementptr i8, ptr %B, i64 %0 56; CHECK: %tmp3_p_scalar_ = load double, ptr %scevgep, align 8, !alias.scope !2, !noalias !5 57; CHECK: %1 = ptrtoint ptr %scevgep to i64 58; CHECK: %2 = call i32 (...) @printf(ptr @3, ptr addrspace(4) @0, i64 %1, ptr addrspace(4) @1, double %tmp3_p_scalar_, ptr addrspace(4) @2) 59; CHECK: %3 = call i32 @fflush(ptr null) 60; CHECK: %scevgep1 = getelementptr i8, ptr %C, i64 %polly.indvar 61; CHECK: %tmp5_p_scalar_ = load i8, ptr %scevgep1, align 1, !alias.scope !10, !noalias !11 62; CHECK: %4 = ptrtoint ptr %scevgep1 to i64 63; CHECK: %5 = sext i8 %tmp5_p_scalar_ to i64 64; CHECK: %6 = call i32 (...) @printf(ptr @7, ptr addrspace(4) @4, i64 %4, ptr addrspace(4) @5, i64 %5, ptr addrspace(4) @6) 65; CHECK: %7 = call i32 @fflush(ptr null) 66; CHECK: %p_tmp6 = sitofp i8 %tmp5_p_scalar_ to double 67; CHECK: %p_tmp7 = fadd double %tmp3_p_scalar_, %p_tmp6 68; CHECK: %8 = shl nuw nsw i64 %polly.indvar, 2 69; CHECK: %scevgep2 = getelementptr i8, ptr %D, i64 %8 70; CHECK: %tmp9_p_scalar_ = load i32, ptr %scevgep2, align 4, !alias.scope !12, !noalias !13 71; CHECK: %9 = ptrtoint ptr %scevgep2 to i64 72; CHECK: %10 = sext i32 %tmp9_p_scalar_ to i64 73; CHECK: %11 = call i32 (...) @printf(ptr @11, ptr addrspace(4) @8, i64 %9, ptr addrspace(4) @9, i64 %10, ptr addrspace(4) @10) 74; CHECK: %12 = call i32 @fflush(ptr null) 75; CHECK: %p_tmp10 = sitofp i32 %tmp9_p_scalar_ to double 76; CHECK: %p_tmp11 = fadd double %p_tmp7, %p_tmp10 77; CHECK: %13 = shl nuw nsw i64 %polly.indvar, 3 78; CHECK: %scevgep3 = getelementptr i8, ptr %E, i64 %13 79; CHECK: %tmp13_p_scalar_ = load i64, ptr %scevgep3, align 8, !alias.scope !14, !noalias !15 80; CHECK: %14 = ptrtoint ptr %scevgep3 to i64 81; CHECK: %15 = call i32 (...) @printf(ptr @15, ptr addrspace(4) @12, i64 %14, ptr addrspace(4) @13, i64 %tmp13_p_scalar_, ptr addrspace(4) @14) 82; CHECK: %16 = call i32 @fflush(ptr null) 83; CHECK: %p_tmp14 = sitofp i64 %tmp13_p_scalar_ to double 84; CHECK: %p_tmp15 = fadd double %p_tmp11, %p_tmp14 85; CHECK: %17 = shl nuw nsw i64 %polly.indvar, 2 86; CHECK: %scevgep4 = getelementptr i8, ptr %A, i64 %17 87; CHECK: %tmp17_p_scalar_ = load float, ptr %scevgep4, align 4, !alias.scope !16, !noalias !17 88; CHECK: %18 = ptrtoint ptr %scevgep4 to i64 89; CHECK: %19 = fpext float %tmp17_p_scalar_ to double 90; CHECK: %20 = call i32 (...) @printf(ptr @19, ptr addrspace(4) @16, i64 %18, ptr addrspace(4) @17, double %19, ptr addrspace(4) @18) 91; CHECK: %21 = call i32 @fflush(ptr null) 92; CHECK: %p_tmp18 = fpext float %tmp17_p_scalar_ to double 93; CHECK: %p_tmp19 = fadd double %p_tmp18, %p_tmp15 94; CHECK: %p_tmp20 = fptrunc double %p_tmp19 to float 95; CHECK: %22 = ptrtoint ptr %scevgep4 to i64 96; CHECK: %23 = fpext float %p_tmp20 to double 97; CHECK: %24 = call i32 (...) @printf(ptr @23, ptr addrspace(4) @20, i64 %22, ptr addrspace(4) @21, double %23, ptr addrspace(4) @22) 98; CHECK: %25 = call i32 @fflush(ptr null) 99 100target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 101 102define void @foo(ptr noalias %A, ptr noalias %B, ptr noalias %C, ptr noalias %D, ptr noalias %E) { 103bb: 104 br label %bb1 105 106bb1: ; preds = %bb21, %bb 107 %i.0 = phi i64 [ 0, %bb ], [ %tmp22, %bb21 ] 108 %exitcond = icmp ne i64 %i.0, 10 109 br i1 %exitcond, label %bb2, label %bb23 110 111bb2: ; preds = %bb1 112 %tmp = getelementptr inbounds double, ptr %B, i64 %i.0 113 %tmp3 = load double, ptr %tmp, align 8 114 %tmp4 = getelementptr inbounds i8, ptr %C, i64 %i.0 115 %tmp5 = load i8, ptr %tmp4, align 1 116 %tmp6 = sitofp i8 %tmp5 to double 117 %tmp7 = fadd double %tmp3, %tmp6 118 %tmp8 = getelementptr inbounds i32, ptr %D, i64 %i.0 119 %tmp9 = load i32, ptr %tmp8, align 4 120 %tmp10 = sitofp i32 %tmp9 to double 121 %tmp11 = fadd double %tmp7, %tmp10 122 %tmp12 = getelementptr inbounds i64, ptr %E, i64 %i.0 123 %tmp13 = load i64, ptr %tmp12, align 8 124 %tmp14 = sitofp i64 %tmp13 to double 125 %tmp15 = fadd double %tmp11, %tmp14 126 %tmp16 = getelementptr inbounds float, ptr %A, i64 %i.0 127 %tmp17 = load float, ptr %tmp16, align 4 128 %tmp18 = fpext float %tmp17 to double 129 %tmp19 = fadd double %tmp18, %tmp15 130 %tmp20 = fptrunc double %tmp19 to float 131 store float %tmp20, ptr %tmp16, align 4 132 br label %bb21 133 134bb21: ; preds = %bb2 135 %tmp22 = add nsw i64 %i.0, 1 136 br label %bb1 137 138bb23: ; preds = %bb1 139 ret void 140} 141 142define i32 @main() { 143bb: 144 %A = alloca [10 x float], align 16 145 %B = alloca [10 x double], align 16 146 %C = alloca [10 x i8], align 1 147 %D = alloca [10 x i32], align 16 148 %E = alloca [10 x i64], align 16 149 br label %bb1 150 151bb1: ; preds = %bb7, %bb 152 %i.0 = phi i64 [ 0, %bb ], [ %tmp8, %bb7 ] 153 %exitcond = icmp ne i64 %i.0, 10 154 br i1 %exitcond, label %bb2, label %bb9 155 156bb2: ; preds = %bb1 157 fence seq_cst 158 %tmp = getelementptr inbounds [10 x i64], ptr %E, i64 0, i64 %i.0 159 store i64 42, ptr %tmp, align 8 160 %tmp3 = getelementptr inbounds [10 x i32], ptr %D, i64 0, i64 %i.0 161 store i32 42, ptr %tmp3, align 4 162 %tmp4 = getelementptr inbounds [10 x i8], ptr %C, i64 0, i64 %i.0 163 store i8 42, ptr %tmp4, align 1 164 %tmp5 = getelementptr inbounds [10 x double], ptr %B, i64 0, i64 %i.0 165 store double 4.200000e+01, ptr %tmp5, align 8 166 %tmp6 = getelementptr inbounds [10 x float], ptr %A, i64 0, i64 %i.0 167 store float 4.200000e+01, ptr %tmp6, align 4 168 br label %bb7 169 170bb7: ; preds = %bb2 171 %tmp8 = add nsw i64 %i.0, 1 172 br label %bb1 173 174bb9: ; preds = %bb1 175 %tmp10 = getelementptr inbounds [10 x float], ptr %A, i64 0, i64 0 176 %tmp11 = getelementptr inbounds [10 x double], ptr %B, i64 0, i64 0 177 %tmp12 = getelementptr inbounds [10 x i8], ptr %C, i64 0, i64 0 178 %tmp13 = getelementptr inbounds [10 x i32], ptr %D, i64 0, i64 0 179 %tmp14 = getelementptr inbounds [10 x i64], ptr %E, i64 0, i64 0 180 call void @foo(ptr %tmp10, ptr %tmp11, ptr %tmp12, ptr %tmp13, ptr %tmp14) 181 %tmp15 = getelementptr inbounds [10 x float], ptr %A, i64 0, i64 8 182 %tmp16 = load float, ptr %tmp15, align 16 183 %tmp17 = fptosi float %tmp16 to i32 184 ret i32 %tmp17 185} 186