1; RUN: opt < %s -passes=loop-vectorize,dce,instcombine,simplifycfg -force-vector-interleave=1 -force-vector-width=4 -S -simplifycfg-require-and-preserve-domtree=1 | FileCheck %s 2 3target 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-S128" 4 5; 6; We want to make sure that we are vectorizeing the scalar loop only once 7; even if the pass manager runs the vectorizer multiple times due to inlining. 8 9 10; This test checks that we add metadata to vectorized loops 11; CHECK-LABEL: @_Z4foo1Pii( 12; CHECK: <4 x i32> 13; CHECK: br i1 %{{.*}}, label %{{.*}}, label %vector.body, !llvm.loop [[VEC_LOOP1:![0-9]+]] 14; 15; CHECK: br i1 %{{.*}}, label %{{.*}}, label %for.body.i, !llvm.loop [[SCALAR_LOOP1:![0-9]+]] 16; CHECK: ret 17 18; This test comes from the loop: 19; 20;int foo (int *A, int n) { 21; return std::accumulate(A, A + n, 0); 22;} 23define i32 @_Z4foo1Pii(ptr %A, i32 %n) #0 { 24entry: 25 %idx.ext = sext i32 %n to i64 26 %add.ptr = getelementptr inbounds i32, ptr %A, i64 %idx.ext 27 %cmp3.i = icmp eq i32 %n, 0 28 br i1 %cmp3.i, label %_ZSt10accumulateIPiiET0_T_S2_S1_.exit, label %for.body.i 29 30for.body.i: ; preds = %entry, %for.body.i 31 %__init.addr.05.i = phi i32 [ %add.i, %for.body.i ], [ 0, %entry ] 32 %__first.addr.04.i = phi ptr [ %incdec.ptr.i, %for.body.i ], [ %A, %entry ] 33 %0 = load i32, ptr %__first.addr.04.i, align 4 34 %add.i = add nsw i32 %0, %__init.addr.05.i 35 %incdec.ptr.i = getelementptr inbounds i32, ptr %__first.addr.04.i, i64 1 36 %cmp.i = icmp eq ptr %incdec.ptr.i, %add.ptr 37 br i1 %cmp.i, label %_ZSt10accumulateIPiiET0_T_S2_S1_.exit, label %for.body.i 38 39_ZSt10accumulateIPiiET0_T_S2_S1_.exit: ; preds = %for.body.i, %entry 40 %__init.addr.0.lcssa.i = phi i32 [ 0, %entry ], [ %add.i, %for.body.i ] 41 ret i32 %__init.addr.0.lcssa.i 42} 43 44; This test checks that we don't vectorize loops that are marked with the "width" == 1 metadata. 45; CHECK-LABEL: @_Z4foo2Pii( 46; CHECK-NOT: <4 x i32> 47; CHECK: br i1 %{{.*}}, label %{{.*}}, label %for.body.i, !llvm.loop [[SCALAR_LOOP2:![0-9]+]] 48; CHECK: ret 49define i32 @_Z4foo2Pii(ptr %A, i32 %n) #0 { 50entry: 51 %idx.ext = sext i32 %n to i64 52 %add.ptr = getelementptr inbounds i32, ptr %A, i64 %idx.ext 53 %cmp3.i = icmp eq i32 %n, 0 54 br i1 %cmp3.i, label %_ZSt10accumulateIPiiET0_T_S2_S1_.exit, label %for.body.i 55 56for.body.i: ; preds = %entry, %for.body.i 57 %__init.addr.05.i = phi i32 [ %add.i, %for.body.i ], [ 0, %entry ] 58 %__first.addr.04.i = phi ptr [ %incdec.ptr.i, %for.body.i ], [ %A, %entry ] 59 %0 = load i32, ptr %__first.addr.04.i, align 4 60 %add.i = add nsw i32 %0, %__init.addr.05.i 61 %incdec.ptr.i = getelementptr inbounds i32, ptr %__first.addr.04.i, i64 1 62 %cmp.i = icmp eq ptr %incdec.ptr.i, %add.ptr 63 br i1 %cmp.i, label %_ZSt10accumulateIPiiET0_T_S2_S1_.exit, label %for.body.i, !llvm.loop !0 64 65_ZSt10accumulateIPiiET0_T_S2_S1_.exit: ; preds = %for.body.i, %entry 66 %__init.addr.0.lcssa.i = phi i32 [ 0, %entry ], [ %add.i, %for.body.i ] 67 ret i32 %__init.addr.0.lcssa.i 68} 69 70attributes #0 = { nounwind readonly ssp uwtable "fp-contract-model"="standard" "frame-pointer"="non-leaf" "realign-stack" "relocation-model"="pic" "ssp-buffers-size"="8" } 71 72; CHECK: [[VEC_LOOP1]] = distinct !{[[VEC_LOOP1]], [[MD_IS_VEC:![0-9]+]], [[MD_RT_UNROLL_DIS:![0-9]+]]} 73; CHECK-NEXT: [[MD_IS_VEC:![0-9]+]] = !{!"llvm.loop.isvectorized", i32 1} 74; CHECK-NEXT: [[MD_RT_UNROLL_DIS]] = !{!"llvm.loop.unroll.runtime.disable"} 75; CHECK-NEXT: [[SCALAR_LOOP1]] = distinct !{[[SCALAR_LOOP1]], [[MD_RT_UNROLL_DIS:![0-9]+]], [[MD_IS_VEC]]} 76; CHECK-NEXT: [[SCALAR_LOOP2]] = distinct !{[[SCALAR_LOOP2]], [[VEC_WIDTH_1:![0-9]+]]} 77; CHECK-NEXT: [[VEC_WIDTH_1]] = !{!"llvm.loop.vectorize.width", i32 1} 78 79!0 = !{!0, !1} 80!1 = !{!"llvm.loop.vectorize.width", i32 1} 81