117 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_analyze_test_checks.py UTC_ARGS: --version 52; REQUIRES: asserts3; RUN: opt -passes=loop-distribute -enable-loop-distribute \4; RUN: -debug-only=loop-distribute -disable-output 2>&1 %s | FileCheck %s5 6define void @f(ptr noalias %a, ptr noalias %b, ptr noalias %c, ptr noalias %d, i64 %stride) {7; CHECK-LABEL: 'f'8; CHECK-NEXT: LDist: Found a candidate loop: for.body9; CHECK-NEXT: LDist: Backward dependences:10; CHECK-NEXT: Backward:11; CHECK-NEXT: %load.a = load i32, ptr %gep.a, align 4 ->12; CHECK-NEXT: store i32 %mul.a, ptr %gep.a.plus4, align 413; CHECK-NEXT: LDist: Seeded partitions:14; CHECK-NEXT: LDist: Partition 0: (cycle)15; CHECK-NEXT: for.body: %load.a = load i32, ptr %gep.a, align 416; CHECK-NEXT: for.body: %load.b = load i32, ptr %gep.b, align 417; CHECK-NEXT: for.body: store i32 %mul.a, ptr %gep.a.plus4, align 418; CHECK-NEXT: LDist: Partition 1:19; CHECK-NEXT: for.body: %loadD = load i32, ptr %gep.d, align 420; CHECK-NEXT: LDist: Partition 2:21; CHECK-NEXT: for.body: %load.strided.a = load i32, ptr %gep.strided.a, align 422; CHECK-NEXT: LDist: Partition 3:23; CHECK-NEXT: for.body: store i32 %mul.c, ptr %gep.c, align 424; CHECK-NEXT: LDist: Merged partitions:25; CHECK-NEXT: LDist: Partition 0: (cycle)26; CHECK-NEXT: for.body: %load.a = load i32, ptr %gep.a, align 427; CHECK-NEXT: for.body: %load.b = load i32, ptr %gep.b, align 428; CHECK-NEXT: for.body: store i32 %mul.a, ptr %gep.a.plus4, align 429; CHECK-NEXT: LDist: Partition 1:30; CHECK-NEXT: for.body: %loadD = load i32, ptr %gep.d, align 431; CHECK-NEXT: for.body: %load.strided.a = load i32, ptr %gep.strided.a, align 432; CHECK-NEXT: for.body: store i32 %mul.c, ptr %gep.c, align 433; CHECK-NEXT: LDist: Populated partitions:34; CHECK-NEXT: LDist: Partition 0: (cycle)35; CHECK-NEXT: for.body: %load.a = load i32, ptr %gep.a, align 436; CHECK-NEXT: for.body: %load.b = load i32, ptr %gep.b, align 437; CHECK-NEXT: for.body: store i32 %mul.a, ptr %gep.a.plus4, align 438; CHECK-NEXT: for.body: br i1 %exitcond, label %exit, label %for.body39; CHECK-NEXT: for.body: %exitcond = icmp eq i64 %add, 2040; CHECK-NEXT: for.body: %add = add nuw nsw i64 %ind, 141; CHECK-NEXT: for.body: %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]42; CHECK-NEXT: for.body: %mul.a = mul i32 %load.b, %load.a43; CHECK-NEXT: for.body: %gep.a.plus4 = getelementptr inbounds i32, ptr %a, i64 %add44; CHECK-NEXT: for.body: %gep.b = getelementptr inbounds i32, ptr %b, i64 %ind45; CHECK-NEXT: for.body: %gep.a = getelementptr inbounds i32, ptr %a, i64 %ind46; CHECK-NEXT: LDist: Partition 1:47; CHECK-NEXT: for.body: %loadD = load i32, ptr %gep.d, align 448; CHECK-NEXT: for.body: %load.strided.a = load i32, ptr %gep.strided.a, align 449; CHECK-NEXT: for.body: store i32 %mul.c, ptr %gep.c, align 450; CHECK-NEXT: for.body: br i1 %exitcond, label %exit, label %for.body51; CHECK-NEXT: for.body: %exitcond = icmp eq i64 %add, 2052; CHECK-NEXT: for.body: %add = add nuw nsw i64 %ind, 153; CHECK-NEXT: for.body: %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]54; CHECK-NEXT: for.body: %mul.c = mul i32 %loadD, %load.strided.a55; CHECK-NEXT: for.body: %gep.c = getelementptr inbounds i32, ptr %c, i64 %ind56; CHECK-NEXT: for.body: %gep.strided.a = getelementptr inbounds i32, ptr %a, i64 %mul57; CHECK-NEXT: for.body: %mul = mul i64 %ind, %stride58; CHECK-NEXT: for.body: %gep.d = getelementptr inbounds i32, ptr %d, i64 %ind59; CHECK-NEXT: LDist: Distributing loop: for.body60; CHECK-NEXT: LDist: Pointers:61; CHECK-NEXT: LDist: After removing unused Instrs:62; CHECK-NEXT: LDist: Partition 0:63; CHECK-NEXT: for.body.ldist1: ; preds = %for.body.ldist1, %for.body.ph.ldist164; CHECK-NEXT: %ind.ldist1 = phi i64 [ 0, %for.body.ph.ldist1 ], [ %add.ldist1, %for.body.ldist1 ]65; CHECK-NEXT: %gep.a.ldist1 = getelementptr inbounds i32, ptr %a, i64 %ind.ldist166; CHECK-NEXT: %load.a.ldist1 = load i32, ptr %gep.a.ldist1, align 467; CHECK-NEXT: %gep.b.ldist1 = getelementptr inbounds i32, ptr %b, i64 %ind.ldist168; CHECK-NEXT: %load.b.ldist1 = load i32, ptr %gep.b.ldist1, align 469; CHECK-NEXT: %mul.a.ldist1 = mul i32 %load.b.ldist1, %load.a.ldist170; CHECK-NEXT: %add.ldist1 = add nuw nsw i64 %ind.ldist1, 171; CHECK-NEXT: %gep.a.plus4.ldist1 = getelementptr inbounds i32, ptr %a, i64 %add.ldist172; CHECK-NEXT: store i32 %mul.a.ldist1, ptr %gep.a.plus4.ldist1, align 473; CHECK-NEXT: %exitcond.ldist1 = icmp eq i64 %add.ldist1, 2074; CHECK-NEXT: br i1 %exitcond.ldist1, label %for.body.ph, label %for.body.ldist175; CHECK-NEXT: LDist: Partition 1:76; CHECK-NEXT: for.body: ; preds = %for.body, %for.body.ph77; CHECK-NEXT: %ind = phi i64 [ 0, %for.body.ph ], [ %add, %for.body ]78; CHECK-NEXT: %add = add nuw nsw i64 %ind, 179; CHECK-NEXT: %gep.d = getelementptr inbounds i32, ptr %d, i64 %ind80; CHECK-NEXT: %loadD = load i32, ptr %gep.d, align 481; CHECK-NEXT: %mul = mul i64 %ind, %stride82; CHECK-NEXT: %gep.strided.a = getelementptr inbounds i32, ptr %a, i64 %mul83; CHECK-NEXT: %load.strided.a = load i32, ptr %gep.strided.a, align 484; CHECK-NEXT: %mul.c = mul i32 %loadD, %load.strided.a85; CHECK-NEXT: %gep.c = getelementptr inbounds i32, ptr %c, i64 %ind86; CHECK-NEXT: store i32 %mul.c, ptr %gep.c, align 487; CHECK-NEXT: %exitcond = icmp eq i64 %add, 2088; CHECK-NEXT: br i1 %exitcond, label %exit.loopexit1, label %for.body89;90entry:91 br label %for.body92 93for.body: ; preds = %for.body, %entry94 %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]95 %gep.a = getelementptr inbounds i32, ptr %a, i64 %ind96 %load.a = load i32, ptr %gep.a, align 497 %gep.b = getelementptr inbounds i32, ptr %b, i64 %ind98 %load.b = load i32, ptr %gep.b, align 499 %mul.a = mul i32 %load.b, %load.a100 %add = add nuw nsw i64 %ind, 1101 %gep.a.plus4 = getelementptr inbounds i32, ptr %a, i64 %add102 store i32 %mul.a, ptr %gep.a.plus4, align 4103 %gep.d = getelementptr inbounds i32, ptr %d, i64 %ind104 %loadD = load i32, ptr %gep.d, align 4105 %mul = mul i64 %ind, %stride106 %gep.strided.a = getelementptr inbounds i32, ptr %a, i64 %mul107 %load.strided.a = load i32, ptr %gep.strided.a, align 4108 %mul.c = mul i32 %loadD, %load.strided.a109 %gep.c = getelementptr inbounds i32, ptr %c, i64 %ind110 store i32 %mul.c, ptr %gep.c, align 4111 %exitcond = icmp eq i64 %add, 20112 br i1 %exitcond, label %exit, label %for.body113 114exit: ; preds = %for.body115 ret void116}117