150 lines · plain
1; RUN: opt %loadNPMPolly -polly-stmt-granularity=bb -polly-optree-normalize-phi=true '-passes=polly-custom<optree>' -polly-print-optree -disable-output < %s | FileCheck %s -match-full-lines2 3target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"4 5define internal fastcc void @kernel_atax(ptr nocapture readonly %A, ptr nocapture readonly %x, ptr nocapture %y, ptr nocapture %tmp) unnamed_addr {6entry:7 br label %entry.split8 9entry.split: ; preds = %entry10 call void @llvm.memset.p0.i64(ptr %y, i8 0, i64 16800, i32 8, i1 false)11 br label %for.body312 13for.body3: ; preds = %for.inc40, %entry.split14 %indvars.iv8 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next9, %for.inc40 ]15 %arrayidx5 = getelementptr inbounds double, ptr %tmp, i64 %indvars.iv816 store double 0.000000e+00, ptr %arrayidx5, align 8, !tbaa !617 br label %for.body818 19for.body8: ; preds = %for.body8, %for.body320 %0 = phi double [ 0.000000e+00, %for.body3 ], [ %add, %for.body8 ]21 %indvars.iv = phi i64 [ 0, %for.body3 ], [ %indvars.iv.next, %for.body8 ]22 %arrayidx14 = getelementptr inbounds [2100 x double], ptr %A, i64 %indvars.iv8, i64 %indvars.iv23 %1 = load double, ptr %arrayidx14, align 8, !tbaa !624 %arrayidx16 = getelementptr inbounds double, ptr %x, i64 %indvars.iv25 %2 = load double, ptr %arrayidx16, align 8, !tbaa !626 %mul = fmul double %1, %227 %add = fadd double %0, %mul28 store double %add, ptr %arrayidx5, align 8, !tbaa !629 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 130 %exitcond = icmp eq i64 %indvars.iv.next, 231 br i1 %exitcond, label %for.end21, label %for.body832 33for.end21: ; preds = %for.body834 br label %for.body2435 36for.body24: ; preds = %for.body24.for.body24_crit_edge, %for.end2137 %3 = phi double [ %add, %for.end21 ], [ %.pre, %for.body24.for.body24_crit_edge ]38 %indvars.iv5 = phi i64 [ 0, %for.end21 ], [ %indvars.iv.next6, %for.body24.for.body24_crit_edge ]39 %arrayidx26 = getelementptr inbounds double, ptr %y, i64 %indvars.iv540 %4 = load double, ptr %arrayidx26, align 8, !tbaa !641 %arrayidx30 = getelementptr inbounds [2100 x double], ptr %A, i64 %indvars.iv8, i64 %indvars.iv542 %5 = load double, ptr %arrayidx30, align 8, !tbaa !643 %mul33 = fmul double %5, %344 %add34 = fadd double %4, %mul3345 store double %add34, ptr %arrayidx26, align 8, !tbaa !646 %indvars.iv.next6 = add nuw nsw i64 %indvars.iv5, 147 %exitcond7 = icmp eq i64 %indvars.iv.next6, 248 br i1 %exitcond7, label %for.inc40, label %for.body24.for.body24_crit_edge49 50for.body24.for.body24_crit_edge: ; preds = %for.body2451 %.pre = load double, ptr %arrayidx5, align 8, !tbaa !652 br label %for.body2453 54for.inc40: ; preds = %for.body2455 %indvars.iv.next9 = add nuw nsw i64 %indvars.iv8, 156 %exitcond10 = icmp eq i64 %indvars.iv.next9, 257 br i1 %exitcond10, label %for.end42, label %for.body358 59for.end42: ; preds = %for.inc4060 ret void61}62 63; Function Attrs: argmemonly nounwind64declare void @llvm.memset.p0.i64(ptr nocapture writeonly, i8, i64, i32, i1)65 66!llvm.module.flags = !{!0}67!llvm.ident = !{!1}68 69!0 = !{i32 1, !"wchar_size", i32 4}70!1 = !{!"clang version 6.0.0 (trunk 312565) (llvm/trunk 312564)"}71!2 = !{!3, !3, i64 0}72!3 = !{!"any pointer", !4, i64 0}73!4 = !{!"omnipotent char", !5, i64 0}74!5 = !{!"Simple C/C++ TBAA"}75!6 = !{!7, !7, i64 0}76!7 = !{!"double", !4, i64 0}77 78; CHECK: Statistics {79; CHECK: Operand trees forwarded: 280; CHECK: Statements with forwarded operand trees: 281; CHECK: }82 83; CHECK-NEXT: After statements {84; CHECK-NEXT: Stmt_for_body385; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]86; CHECK-NEXT: { Stmt_for_body3[i0] -> MemRef_tmp[i0] };87; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]88; CHECK-NEXT: { Stmt_for_body3[i0] -> MemRef1__phi[] };89; CHECK-NEXT: Instructions {90; CHECK-NEXT: store double 0.000000e+00, ptr %arrayidx5, align 8, !tbaa !291; CHECK-NEXT: }92; CHECK-NEXT: Stmt_for_body893; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]94; CHECK-NEXT: { Stmt_for_body8[i0, i1] -> MemRef1__phi[] };95; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 1]96; CHECK-NEXT: { Stmt_for_body8[i0, i1] -> MemRef1__phi[] };97; CHECK-NEXT: new: { Stmt_for_body8[i0, i1] -> MemRef_tmp[i0] };98; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]99; CHECK-NEXT: { Stmt_for_body8[i0, i1] -> MemRef_A[i0, i1] };100; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]101; CHECK-NEXT: { Stmt_for_body8[i0, i1] -> MemRef_x[i1] };102; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]103; CHECK-NEXT: { Stmt_for_body8[i0, i1] -> MemRef_tmp[i0] };104; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]105; CHECK-NEXT: { Stmt_for_body8[i0, i1] -> MemRef_add[] };106; CHECK-NEXT: Instructions {107; CHECK-NEXT: %0 = phi double [ 0.000000e+00, %for.body3 ], [ %add, %for.body8 ]108; CHECK-NEXT: %1 = load double, ptr %arrayidx14, align 8, !tbaa !2109; CHECK-NEXT: %2 = load double, ptr %arrayidx16, align 8, !tbaa !2110; CHECK-NEXT: %mul = fmul double %1, %2111; CHECK-NEXT: %add = fadd double %0, %mul112; CHECK-NEXT: store double %add, ptr %arrayidx5, align 8, !tbaa !2113; CHECK-NEXT: %exitcond = icmp eq i64 %indvars.iv.next, 2114; CHECK-NEXT: }115; CHECK-NEXT: Stmt_for_end21116; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 1]117; CHECK-NEXT: { Stmt_for_end21[i0] -> MemRef_add[] };118; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]119; CHECK-NEXT: { Stmt_for_end21[i0] -> MemRef5__phi[] };120; CHECK-NEXT: Instructions {121; CHECK-NEXT: }122; CHECK-NEXT: Stmt_for_body24123; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 1]124; CHECK-NEXT: { Stmt_for_body24[i0, i1] -> MemRef5__phi[] };125; CHECK-NEXT: new: { Stmt_for_body24[i0, i1] -> MemRef_tmp[i0] };126; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]127; CHECK-NEXT: { Stmt_for_body24[i0, i1] -> MemRef_y[i1] };128; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]129; CHECK-NEXT: { Stmt_for_body24[i0, i1] -> MemRef_A[i0, i1] };130; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]131; CHECK-NEXT: { Stmt_for_body24[i0, i1] -> MemRef_y[i1] };132; CHECK-NEXT: Instructions {133; CHECK-NEXT: %3 = phi double [ %add, %for.end21 ], [ %.pre, %for.body24.for.body24_crit_edge ]134; CHECK-NEXT: %4 = load double, ptr %arrayidx26, align 8, !tbaa !2135; CHECK-NEXT: %5 = load double, ptr %arrayidx30, align 8, !tbaa !2136; CHECK-NEXT: %mul33 = fmul double %5, %3137; CHECK-NEXT: %add34 = fadd double %4, %mul33138; CHECK-NEXT: store double %add34, ptr %arrayidx26, align 8, !tbaa !2139; CHECK-NEXT: %exitcond7 = icmp eq i64 %indvars.iv.next6, 2140; CHECK-NEXT: }141; CHECK-NEXT: Stmt_for_body24_for_body24_crit_edge142; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 1]143; CHECK-NEXT: { Stmt_for_body24_for_body24_crit_edge[i0, i1] -> MemRef5__phi[] };144; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]145; CHECK-NEXT: { Stmt_for_body24_for_body24_crit_edge[i0, i1] -> MemRef_tmp[i0] };146; CHECK-NEXT: Instructions {147; CHECK-NEXT: %.pre = load double, ptr %arrayidx5, align 8, !tbaa !2148; CHECK-NEXT: }149; CHECK-NEXT: }150