93 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly<no-default-opts>' -S < %s | FileCheck %s -check-prefix=SEQUENTIAL2; RUN: opt %loadNPMPolly '-passes=polly<no-default-opts>' -polly-ast-detect-parallel -S < %s | FileCheck %s -check-prefix=PARALLEL3target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"4 5; This is a trivially parallel loop. We just use it to ensure that we actually6; emit the right information.7;8; for (i = 0; i < n; i++)9; A[i] = 1;10;11@A = common global [1024 x i32] zeroinitializer12define void @test-one(i64 %n) {13start:14 fence seq_cst15 br label %loop.header16 17loop.header:18 %i = phi i64 [ 0, %start ], [ %i.next, %loop.backedge ]19 %exitcond = icmp ne i64 %i, %n20 br i1 %exitcond, label %loop.body, label %ret21 22loop.body:23 %scevgep = getelementptr [1024 x i32], ptr @A, i64 0, i64 %i24 store i32 1, ptr %scevgep25 br label %loop.backedge26 27loop.backedge:28 %i.next = add nsw i64 %i, 129 br label %loop.header30 31ret:32 fence seq_cst33 ret void34}35 36; SEQUENTIAL-LABEL: @test-one37; SEQUENTIAL-NOT: !llvm.mem.parallel_loop_access38; SEQUENTIAL-NOT: !llvm.access.group39 40; PARALLEL: @test-one41; PARALLEL: store i32 1, ptr %scevgep1, {{[ ._!,a-zA-Z0-9]*}}, !llvm.access.group ![[GROUPID3:[0-9]+]]42; PARALLEL: br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit, !llvm.loop ![[LoopID4:[0-9]+]]43 44 45; This loop has memory dependences that require at least a simple dependence46; analysis to detect the parallelism.47;48; for (i = 0; i < n; i++)49; A[2 * i] = A[2 * i + 1];50;51define void @test-two(i64 %n) {52start:53 fence seq_cst54 br label %loop.header55 56loop.header:57 %i = phi i64 [ 0, %start ], [ %i.next, %loop.backedge ]58 %exitcond = icmp ne i64 %i, %n59 br i1 %exitcond, label %loop.body, label %ret60 61loop.body:62 %loadoffset1 = mul nsw i64 %i, 263 %loadoffset2 = add nsw i64 %loadoffset1, 164 %scevgepload = getelementptr [1024 x i32], ptr @A, i64 0, i64 %loadoffset265 %val = load i32, ptr %scevgepload66 %storeoffset = mul i64 %i, 267 %scevgepstore = getelementptr [1024 x i32], ptr @A, i64 0, i64 %storeoffset68 store i32 %val, ptr %scevgepstore69 br label %loop.backedge70 71loop.backedge:72 %i.next = add nsw i64 %i, 173 br label %loop.header74 75ret:76 fence seq_cst77 ret void78}79 80; SEQUENTIAL-LABEL: @test-two81; SEQUENTIAL-NOT: !llvm.mem.parallel_loop_access82; SEQUENTIAL-NOT: !llvm.access.group83 84; PARALLEL: @test-two85; PARALLEL: %val_p_scalar_ = load i32, ptr %scevgep, {{[ ._!,a-zA-Z0-9]*}}, !llvm.access.group ![[GROUPID8:[0-9]*]]86; PARALLEL: store i32 %val_p_scalar_, ptr %scevgep1, {{[ ._!,a-zA-Z0-9]*}}, !llvm.access.group ![[GROUPID8]]87; PARALLEL: br i1 %polly.loop_cond, label %polly.loop_header, label %polly.loop_exit, !llvm.loop ![[LoopID9:[0-9]*]]88 89; PARALLEL: ![[LoopID4]] = distinct !{![[LoopID4]], ![[PARACC5:[0-9]+]]}90; PARALLEL: ![[PARACC5]] = !{!"llvm.loop.parallel_accesses", ![[GROUPID3]]}91; PARALLEL: ![[LoopID9]] = distinct !{![[LoopID9]], ![[PARACC10:[0-9]+]]}92; PARALLEL: ![[PARACC10]] = !{!"llvm.loop.parallel_accesses", ![[GROUPID8]]}93