132 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<mse>' -polly-print-mse -disable-output < %s | FileCheck %s2; RUN: opt %loadNPMPolly -polly-stmt-granularity=bb '-passes=polly-custom<mse>' -polly-print-mse -pass-remarks-analysis=polly-mse -disable-output < %s 2>&1 | FileCheck %s --check-prefix=MSE3;4; Verify that the accesses are correctly expanded for MemoryKind::Array and MemoryKind::PHI.5; tmp_06_phi is not expanded because it need copy in.6;7; Original source code :8;9; #define Ni 200010; #define Nj 300011;12; void tmp3(double A[Ni], double B[Nj]) {13; int i,j;14; double tmp = 6;15; for (i = 0; i < Ni; i++) {16;17; for(int h = 0; h<Nj; h++)18; B[h] = h;19;20; for(j = 0; j < Nj; j++) {21; for(int k=0; k<Nj; k++) {22; tmp = tmp+i+k+j;23; A[i+j] = tmp*B[k];24; }25; }26; }27; }28;29; Check if the expanded SAI are created except for tmp_06_phi30;31; MSE: MemRef_tmp_06__phi read from its original value.32;33; CHECK-DAG: double MemRef_A_Stmt_for_body11_expanded[10000][10000][10000]; // Element size 834; CHECK-DAG: double MemRef_add16_lcssa__phi_Stmt_for_inc25_expanded[10000][10000]; // Element size 835; CHECK-DAG: double MemRef_B_Stmt_for_body3_expanded[10000][10000]; // Element size 836; CHECK-DAG: double MemRef_tmp_06_Stmt_for_body_expanded[10000]; // Element size 837; CHECK-DAG: double MemRef_add16_lcssa_lcssa__phi_Stmt_for_inc28_expanded[10000]; // Element size 838; CHECK-DAG: double MemRef_tmp_14__phi_Stmt_for_body7_expanded[10000][10000]; // Element size 839; CHECK-DAG: double MemRef_tmp_22__phi_Stmt_for_body11_expanded[10000][10000][10000]; // Element size 840; CHECK-NOT: double MemRef_tmp_06__phi_Stmt_for_body_expanded[10000]; // Element size 841;42; Check if the memory accesses are modified except those of tmp_06_phi43;44; CHECK-NOT: new: { Stmt_for_body[i0] -> MemRef_tmp_06__phi_Stmt_for_body_expanded[i0] };45; CHECK-DAG: new: { Stmt_for_body[i0] -> MemRef_tmp_06_Stmt_for_body_expanded[i0] };46; CHECK-DAG: new: { Stmt_for_body3[i0, i1] -> MemRef_B_Stmt_for_body3_expanded[i0, i1] };47; CHECK-DAG: new: { Stmt_for_end[i0] -> MemRef_tmp_06_Stmt_for_body_expanded[i0] };48; CHECK-DAG: new: { Stmt_for_end[i0] -> MemRef_tmp_14__phi_Stmt_for_body7_expanded[i0, 0] };49; CHECK-DAG: new: { Stmt_for_body7[i0, i1] -> MemRef_tmp_14__phi_Stmt_for_body7_expanded[i0, i1] };50; CHECK-DAG: new: { Stmt_for_body7[i0, i1] -> MemRef_tmp_22__phi_Stmt_for_body11_expanded[i0, i1, 0] };51; CHECK-DAG: new: { Stmt_for_body11[i0, i1, i2] -> MemRef_tmp_22__phi_Stmt_for_body11_expanded[i0, i1, 1 + i2] : i2 <= 9998 };52; CHECK-DAG: new: { Stmt_for_body11[i0, i1, i2] -> MemRef_tmp_22__phi_Stmt_for_body11_expanded[i0, i1, i2] };53; CHECK-DAG: new: { Stmt_for_body11[i0, i1, i2] -> MemRef_B_Stmt_for_body3_expanded[i0, i2] };54; CHECK-DAG: new: { Stmt_for_body11[i0, i1, i2] -> MemRef_A_Stmt_for_body11_expanded[i0, i1, i2] };55; CHECK-DAG: new: { Stmt_for_body11[i0, i1, 9999] -> MemRef_add16_lcssa__phi_Stmt_for_inc25_expanded[i0, i1] };56; CHECK-DAG: new: { Stmt_for_inc25[i0, i1] -> MemRef_tmp_14__phi_Stmt_for_body7_expanded[i0, 1 + i1] : i1 <= 9998 };57; CHECK-DAG: new: { Stmt_for_inc25[i0, i1] -> MemRef_add16_lcssa__phi_Stmt_for_inc25_expanded[i0, i1] };58; CHECK-DAG: new: { Stmt_for_inc25[i0, 9999] -> MemRef_add16_lcssa_lcssa__phi_Stmt_for_inc28_expanded[i0] };59; CHECK-DAG: new: { Stmt_for_inc28[i0] -> MemRef_add16_lcssa_lcssa__phi_Stmt_for_inc28_expanded[i0] };60; CHECK-NOT: new: { Stmt_for_inc28[i0] -> MemRef_tmp_06__phi_Stmt_for_body_expanded[1 + i0] : i0 <= 9998 };61;62target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"63target triple = "x86_64-unknown-linux-gnu"64 65define void @mse(ptr %A, ptr %B) {66entry:67 br label %entry.split68 69entry.split: ; preds = %entry70 br label %for.body71 72for.body: ; preds = %entry.split, %for.inc2873 %indvars.iv15 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next16, %for.inc28 ]74 %tmp.06 = phi double [ 6.000000e+00, %entry.split ], [ %add16.lcssa.lcssa, %for.inc28 ]75 br label %for.body376 77for.body3: ; preds = %for.body, %for.body378 %indvars.iv = phi i64 [ 0, %for.body ], [ %indvars.iv.next, %for.body3 ]79 %0 = trunc i64 %indvars.iv to i3280 %conv = sitofp i32 %0 to double81 %arrayidx = getelementptr inbounds double, ptr %B, i64 %indvars.iv82 store double %conv, ptr %arrayidx, align 883 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 184 %exitcond = icmp ne i64 %indvars.iv.next, 1000085 br i1 %exitcond, label %for.body3, label %for.end86 87for.end: ; preds = %for.body388 br label %for.body789 90for.body7: ; preds = %for.end, %for.inc2591 %indvars.iv11 = phi i64 [ 0, %for.end ], [ %indvars.iv.next12, %for.inc25 ]92 %tmp.14 = phi double [ %tmp.06, %for.end ], [ %add16.lcssa, %for.inc25 ]93 br label %for.body1194 95for.body11: ; preds = %for.body7, %for.body1196 %indvars.iv8 = phi i64 [ 0, %for.body7 ], [ %indvars.iv.next9, %for.body11 ]97 %tmp.22 = phi double [ %tmp.14, %for.body7 ], [ %add16, %for.body11 ]98 %1 = trunc i64 %indvars.iv15 to i3299 %conv12 = sitofp i32 %1 to double100 %add = fadd double %tmp.22, %conv12101 %2 = trunc i64 %indvars.iv8 to i32102 %conv13 = sitofp i32 %2 to double103 %add14 = fadd double %add, %conv13104 %3 = trunc i64 %indvars.iv11 to i32105 %conv15 = sitofp i32 %3 to double106 %add16 = fadd double %add14, %conv15107 %arrayidx18 = getelementptr inbounds double, ptr %B, i64 %indvars.iv8108 %4 = load double, ptr %arrayidx18, align 8109 %mul = fmul double %add16, %4110 %5 = add nuw nsw i64 %indvars.iv11, %indvars.iv15111 %arrayidx21 = getelementptr inbounds double, ptr %A, i64 %5112 store double %mul, ptr %arrayidx21, align 8113 %indvars.iv.next9 = add nuw nsw i64 %indvars.iv8, 1114 %exitcond10 = icmp ne i64 %indvars.iv.next9, 10000115 br i1 %exitcond10, label %for.body11, label %for.inc25116 117for.inc25: ; preds = %for.body11118 %add16.lcssa = phi double [ %add16, %for.body11 ]119 %indvars.iv.next12 = add nuw nsw i64 %indvars.iv11, 1120 %exitcond14 = icmp ne i64 %indvars.iv.next12, 10000121 br i1 %exitcond14, label %for.body7, label %for.inc28122 123for.inc28: ; preds = %for.inc25124 %add16.lcssa.lcssa = phi double [ %add16.lcssa, %for.inc25 ]125 %indvars.iv.next16 = add nuw nsw i64 %indvars.iv15, 1126 %exitcond17 = icmp ne i64 %indvars.iv.next16, 10000127 br i1 %exitcond17, label %for.body, label %for.end30128 129for.end30: ; preds = %for.inc28130 ret void131}132