Home | History | Annotate | Download | only in X86
      1 ; RUN: opt < %s -loop-vectorize -S -pass-remarks-missed='loop-vectorize' -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s
      2 
      3 ; C/C++ code for tests
      4 ; void test(int *A, int Length) {
      5 ; #pragma clang loop vectorize(enable) interleave(enable)
      6 ;   for (int i = 0; i < Length; i++) {
      7 ;     A[i] = i;
      8 ;     if (A[i] > Length)
      9 ;       break;
     10 ;   }
     11 ; }
     12 
     13 ; void test_disabled(int *A, int Length) {
     14 ; #pragma clang loop vectorize(disable) interleave(disable)
     15 ;   for (int i = 0; i < Length; i++)
     16 ;     A[i] = i;
     17 ; }
     18 
     19 ; void test_array_bounds(int *A, int *B, int Length) {
     20 ; #pragma clang loop vectorize(enable)
     21 ;   for (int i = 0; i < Length; i++)
     22 ;     A[i] = A[B[i]];
     23 ; }
     24 
     25 ; File, line, and column should match those specified in the metadata
     26 ; CHECK: remark: source.cpp:4:5: loop not vectorized: could not determine number of loop iterations
     27 ; CHECK: remark: source.cpp:4:5: loop not vectorized: use -Rpass-analysis=loop-vectorize for more info
     28 ; CHECK: remark: source.cpp:13:5: loop not vectorized: vectorization and interleaving are explicitly disabled, or vectorize width and interleave count are both set to 1
     29 ; CHECK: remark: source.cpp:19:5: loop not vectorized: cannot identify array bounds
     30 ; CHECK: remark: source.cpp:19:5: loop not vectorized: use -Rpass-analysis=loop-vectorize for more info
     31 ; CHECK: warning: source.cpp:19:5: loop not vectorized: failed explicitly specified loop vectorization
     32 
     33 ; CHECK: _Z4testPii
     34 ; CHECK-NOT: x i32>
     35 ; CHECK: ret
     36 
     37 ; CHECK: _Z13test_disabledPii
     38 ; CHECK-NOT: x i32>
     39 ; CHECK: ret
     40 
     41 ; CHECK: _Z17test_array_boundsPiS_i
     42 ; CHECK-NOT: x i32>
     43 ; CHECK: ret
     44 
     45 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
     46 
     47 ; Function Attrs: nounwind optsize ssp uwtable
     48 define void @_Z4testPii(i32* nocapture %A, i32 %Length) #0 !dbg !4 {
     49 entry:
     50   %cmp10 = icmp sgt i32 %Length, 0, !dbg !12
     51   br i1 %cmp10, label %for.body, label %for.end, !dbg !12, !llvm.loop !14
     52 
     53 for.body:                                         ; preds = %entry, %for.body
     54   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
     55   %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv, !dbg !16
     56   %0 = trunc i64 %indvars.iv to i32, !dbg !16
     57   %ld = load i32, i32* %arrayidx, align 4
     58   store i32 %0, i32* %arrayidx, align 4, !dbg !16, !tbaa !18
     59   %cmp3 = icmp sle i32 %ld, %Length, !dbg !22
     60   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !12
     61   %1 = trunc i64 %indvars.iv.next to i32
     62   %cmp = icmp slt i32 %1, %Length, !dbg !12
     63   %or.cond = and i1 %cmp3, %cmp, !dbg !22
     64   br i1 %or.cond, label %for.body, label %for.end, !dbg !22
     65 
     66 for.end:                                          ; preds = %for.body, %entry
     67   ret void, !dbg !24
     68 }
     69 
     70 ; Function Attrs: nounwind optsize ssp uwtable
     71 define void @_Z13test_disabledPii(i32* nocapture %A, i32 %Length) #0 !dbg !7 {
     72 entry:
     73   %cmp4 = icmp sgt i32 %Length, 0, !dbg !25
     74   br i1 %cmp4, label %for.body, label %for.end, !dbg !25, !llvm.loop !27
     75 
     76 for.body:                                         ; preds = %entry, %for.body
     77   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ]
     78   %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv, !dbg !30
     79   %0 = trunc i64 %indvars.iv to i32, !dbg !30
     80   store i32 %0, i32* %arrayidx, align 4, !dbg !30, !tbaa !18
     81   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !25
     82   %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !25
     83   %exitcond = icmp eq i32 %lftr.wideiv, %Length, !dbg !25
     84   br i1 %exitcond, label %for.end, label %for.body, !dbg !25, !llvm.loop !27
     85 
     86 for.end:                                          ; preds = %for.body, %entry
     87   ret void, !dbg !31
     88 }
     89 
     90 ; Function Attrs: nounwind optsize ssp uwtable
     91 define void @_Z17test_array_boundsPiS_i(i32* nocapture %A, i32* nocapture readonly %B, i32 %Length) #0 !dbg !8 {
     92 entry:
     93   %cmp9 = icmp sgt i32 %Length, 0, !dbg !32
     94   br i1 %cmp9, label %for.body.preheader, label %for.end, !dbg !32, !llvm.loop !34
     95 
     96 for.body.preheader:                               ; preds = %entry
     97   br label %for.body, !dbg !35
     98 
     99 for.body:                                         ; preds = %for.body.preheader, %for.body
    100   %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %for.body.preheader ]
    101   %arrayidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv, !dbg !35
    102   %0 = load i32, i32* %arrayidx, align 4, !dbg !35, !tbaa !18
    103   %idxprom1 = sext i32 %0 to i64, !dbg !35
    104   %arrayidx2 = getelementptr inbounds i32, i32* %A, i64 %idxprom1, !dbg !35
    105   %1 = load i32, i32* %arrayidx2, align 4, !dbg !35, !tbaa !18
    106   %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 %indvars.iv, !dbg !35
    107   store i32 %1, i32* %arrayidx4, align 4, !dbg !35, !tbaa !18
    108   %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1, !dbg !32
    109   %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !32
    110   %exitcond = icmp eq i32 %lftr.wideiv, %Length, !dbg !32
    111   br i1 %exitcond, label %for.end.loopexit, label %for.body, !dbg !32, !llvm.loop !34
    112 
    113 for.end.loopexit:                                 ; preds = %for.body
    114   br label %for.end
    115 
    116 for.end:                                          ; preds = %for.end.loopexit, %entry
    117   ret void, !dbg !36
    118 }
    119 
    120 attributes #0 = { nounwind }
    121 
    122 !llvm.dbg.cu = !{!0}
    123 !llvm.module.flags = !{!9, !10}
    124 !llvm.ident = !{!11}
    125 
    126 !0 = distinct !DICompileUnit(language: DW_LANG_C_plus_plus, producer: "clang version 3.5.0", isOptimized: true, runtimeVersion: 6, emissionKind: LineTablesOnly, file: !1, enums: !2, retainedTypes: !2, globals: !2, imports: !2)
    127 !1 = !DIFile(filename: "source.cpp", directory: ".")
    128 !2 = !{}
    129 !4 = distinct !DISubprogram(name: "test", line: 1, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 1, file: !1, scope: !5, type: !6, variables: !2)
    130 !5 = !DIFile(filename: "source.cpp", directory: ".")
    131 !6 = !DISubroutineType(types: !2)
    132 !7 = distinct !DISubprogram(name: "test_disabled", line: 10, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 10, file: !1, scope: !5, type: !6, variables: !2)
    133 !8 = distinct !DISubprogram(name: "test_array_bounds", line: 16, isLocal: false, isDefinition: true, virtualIndex: 6, flags: DIFlagPrototyped, isOptimized: true, unit: !0, scopeLine: 16, file: !1, scope: !5, type: !6, variables: !2)
    134 !9 = !{i32 2, !"Dwarf Version", i32 2}
    135 !10 = !{i32 2, !"Debug Info Version", i32 3}
    136 !11 = !{!"clang version 3.5.0"}
    137 !12 = !DILocation(line: 3, column: 8, scope: !13)
    138 !13 = distinct !DILexicalBlock(line: 3, column: 3, file: !1, scope: !4)
    139 !14 = !{!14, !15, !15}
    140 !15 = !{!"llvm.loop.vectorize.enable", i1 true}
    141 !16 = !DILocation(line: 4, column: 5, scope: !17)
    142 !17 = distinct !DILexicalBlock(line: 3, column: 36, file: !1, scope: !13)
    143 !18 = !{!19, !19, i64 0}
    144 !19 = !{!"int", !20, i64 0}
    145 !20 = !{!"omnipotent char", !21, i64 0}
    146 !21 = !{!"Simple C/C++ TBAA"}
    147 !22 = !DILocation(line: 5, column: 9, scope: !23)
    148 !23 = distinct !DILexicalBlock(line: 5, column: 9, file: !1, scope: !17)
    149 !24 = !DILocation(line: 8, column: 1, scope: !4)
    150 !25 = !DILocation(line: 12, column: 8, scope: !26)
    151 !26 = distinct !DILexicalBlock(line: 12, column: 3, file: !1, scope: !7)
    152 !27 = !{!27, !28, !29}
    153 !28 = !{!"llvm.loop.interleave.count", i32 1}
    154 !29 = !{!"llvm.loop.vectorize.width", i32 1}
    155 !30 = !DILocation(line: 13, column: 5, scope: !26)
    156 !31 = !DILocation(line: 14, column: 1, scope: !7)
    157 !32 = !DILocation(line: 18, column: 8, scope: !33)
    158 !33 = distinct !DILexicalBlock(line: 18, column: 3, file: !1, scope: !8)
    159 !34 = !{!34, !15}
    160 !35 = !DILocation(line: 19, column: 5, scope: !33)
    161 !36 = !DILocation(line: 20, column: 1, scope: !8)
    162