1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt < %s -loop-vectorize -force-vector-interleave=1 -force-vector-width=4 -dce -instcombine -S | FileCheck %s 3; RUN: opt < %s -loop-vectorize -disable-basic-aa -S -pass-remarks-analysis='loop-vectorize' 2>&1 | FileCheck %s -check-prefix=FORCED_OPTSIZE 4 5target 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" 6 7; Make sure we vectorize this loop: 8; int foo(float *a, float *b, int n) { 9; for (int i=0; i<n; ++i) 10; a[i] = b[i] * 3; 11; } 12 13define i32 @foo(float* nocapture %a, float* nocapture %b, i32 %n) nounwind uwtable ssp { 14; CHECK-LABEL: @foo( 15; CHECK-NEXT: entry: 16; CHECK-NEXT: [[CMP6:%.*]] = icmp sgt i32 [[N:%.*]], 0, [[DBG4:!dbg !.*]] 17; CHECK-NEXT: br i1 [[CMP6]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_END:%.*]], [[DBG4]] 18; CHECK: for.body.preheader: 19; CHECK-NEXT: [[TMP0:%.*]] = add i32 [[N]], -1, [[DBG9:!dbg !.*]] 20; CHECK-NEXT: [[TMP1:%.*]] = zext i32 [[TMP0]] to i64, [[DBG9]] 21; CHECK-NEXT: [[TMP2:%.*]] = add nuw nsw i64 [[TMP1]], 1, [[DBG9]] 22; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i32 [[TMP0]], 3, [[DBG9]] 23; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]], [[DBG9]] 24; CHECK: vector.memcheck: 25; CHECK-NEXT: [[TMP3:%.*]] = add i32 [[N]], -1, [[DBG9]] 26; CHECK-NEXT: [[TMP4:%.*]] = zext i32 [[TMP3]] to i64, [[DBG9]] 27; CHECK-NEXT: [[TMP5:%.*]] = add nuw nsw i64 [[TMP4]], 1, [[DBG9]] 28; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr float, float* [[A:%.*]], i64 [[TMP5]], [[DBG9]] 29; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr float, float* [[B:%.*]], i64 [[TMP5]], [[DBG9]] 30; CHECK-NEXT: [[BOUND0:%.*]] = icmp ugt float* [[SCEVGEP4]], [[A]], [[DBG9]] 31; CHECK-NEXT: [[BOUND1:%.*]] = icmp ugt float* [[SCEVGEP]], [[B]], [[DBG9]] 32; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]], [[DBG9]] 33; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]], [[DBG9]] 34; CHECK: vector.ph: 35; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[TMP2]], 8589934588, [[DBG9]] 36; CHECK-NEXT: br label [[VECTOR_BODY:%.*]], [[DBG9]] 37; CHECK: vector.body: 38; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ], [[DBG9]] 39; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDEX]], [[DBG9]] 40; CHECK-NEXT: [[TMP7:%.*]] = bitcast float* [[TMP6]] to <4 x float>*, [[DBG9]] 41; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x float>, <4 x float>* [[TMP7]], align 4, [[DBG9]], !alias.scope !10 42; CHECK-NEXT: [[TMP8:%.*]] = fmul <4 x float> [[WIDE_LOAD]], <float 3.000000e+00, float 3.000000e+00, float 3.000000e+00, float 3.000000e+00>, [[DBG9]] 43; CHECK-NEXT: [[TMP9:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDEX]], [[DBG9]] 44; CHECK-NEXT: [[TMP10:%.*]] = bitcast float* [[TMP9]] to <4 x float>*, [[DBG9]] 45; CHECK-NEXT: store <4 x float> [[TMP8]], <4 x float>* [[TMP10]], align 4, [[DBG9]], !alias.scope !13, !noalias !10 46; CHECK-NEXT: [[INDEX_NEXT]] = add i64 [[INDEX]], 4, [[DBG9]] 47; CHECK-NEXT: [[TMP11:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]], [[DBG9]] 48; CHECK-NEXT: br i1 [[TMP11]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], [[DBG9]], [[LOOP15:!llvm.loop !.*]] 49; CHECK: middle.block: 50; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[TMP2]], [[N_VEC]], [[DBG9]] 51; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_END_LOOPEXIT:%.*]], label [[SCALAR_PH]], [[DBG9]] 52; CHECK: scalar.ph: 53; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[FOR_BODY_PREHEADER]] ], [ 0, [[VECTOR_MEMCHECK]] ], [[DBG9]] 54; CHECK-NEXT: br label [[FOR_BODY:%.*]], [[DBG9]] 55; CHECK: for.body: 56; CHECK-NEXT: [[INDVARS_IV:%.*]] = phi i64 [ [[INDVARS_IV_NEXT:%.*]], [[FOR_BODY]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ], [[DBG9]] 57; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds float, float* [[B]], i64 [[INDVARS_IV]], [[DBG9]] 58; CHECK-NEXT: [[TMP12:%.*]] = load float, float* [[ARRAYIDX]], align 4, [[DBG9]] 59; CHECK-NEXT: [[MUL:%.*]] = fmul float [[TMP12]], 3.000000e+00, [[DBG9]] 60; CHECK-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds float, float* [[A]], i64 [[INDVARS_IV]], [[DBG9]] 61; CHECK-NEXT: store float [[MUL]], float* [[ARRAYIDX2]], align 4, [[DBG9]] 62; CHECK-NEXT: [[INDVARS_IV_NEXT]] = add i64 [[INDVARS_IV]], 1, [[DBG9]] 63; CHECK-NEXT: [[LFTR_WIDEIV:%.*]] = trunc i64 [[INDVARS_IV_NEXT]] to i32, [[DBG9]] 64; CHECK-NEXT: [[EXITCOND:%.*]] = icmp eq i32 [[LFTR_WIDEIV]], [[N]], [[DBG9]] 65; CHECK-NEXT: br i1 [[EXITCOND]], label [[FOR_END_LOOPEXIT]], label [[FOR_BODY]], [[DBG9]], [[LOOP17:!llvm.loop !.*]] 66; CHECK: for.end.loopexit: 67; CHECK-NEXT: br label [[FOR_END]], [[DBG18:!dbg !.*]] 68; CHECK: for.end: 69; CHECK-NEXT: ret i32 undef, [[DBG18]] 70; 71entry: 72 %cmp6 = icmp sgt i32 %n, 0, !dbg !6 73 br i1 %cmp6, label %for.body, label %for.end, !dbg !6 74 75for.body: ; preds = %entry, %for.body 76 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body ], [ 0, %entry ], !dbg !7 77 %arrayidx = getelementptr inbounds float, float* %b, i64 %indvars.iv, !dbg !7 78 %0 = load float, float* %arrayidx, align 4, !dbg !7 79 %mul = fmul float %0, 3.000000e+00, !dbg !7 80 %arrayidx2 = getelementptr inbounds float, float* %a, i64 %indvars.iv, !dbg !7 81 store float %mul, float* %arrayidx2, align 4, !dbg !7 82 %indvars.iv.next = add i64 %indvars.iv, 1, !dbg !7 83 %lftr.wideiv = trunc i64 %indvars.iv.next to i32, !dbg !7 84 %exitcond = icmp eq i32 %lftr.wideiv, %n, !dbg !7 85 br i1 %exitcond, label %for.end, label %for.body, !dbg !7 86 87for.end: ; preds = %for.body, %entry 88 ret i32 undef, !dbg !8 89} 90 91; Make sure that we try to vectorize loops with a runtime check if the 92; dependency check fails. 93 94; CHECK-LABEL: test_runtime_check 95; CHECK: <4 x float> 96define void @test_runtime_check(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n) { 97entry: 98 br label %for.body 99 100for.body: 101 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 102 %ind.sum = add i64 %iv, %offset 103 %arr.idx = getelementptr inbounds float, float* %a, i64 %ind.sum 104 %l1 = load float, float* %arr.idx, align 4 105 %ind.sum2 = add i64 %iv, %offset2 106 %arr.idx2 = getelementptr inbounds float, float* %a, i64 %ind.sum2 107 %l2 = load float, float* %arr.idx2, align 4 108 %m = fmul fast float %b, %l2 109 %ad = fadd fast float %l1, %m 110 store float %ad, float* %arr.idx, align 4 111 %iv.next = add nuw nsw i64 %iv, 1 112 %exitcond = icmp eq i64 %iv.next, %n 113 br i1 %exitcond, label %loopexit, label %for.body 114 115loopexit: 116 ret void 117} 118 119; Check we do not generate runtime checks if we found a known dependence preventing 120; vectorization. In this case, it is a read of c[i-1] followed by a write of c[i]. 121; The runtime checks would always fail. 122 123; void test_runtime_check2(float *a, float b, unsigned offset, unsigned offset2, unsigned n, float *c) { 124; for (unsigned i = 1; i < n; i++) { 125; a[i+o1] += a[i+o2] + b; 126; c[i] = c[i-1] + b; 127; } 128; } 129; 130; CHECK-LABEL: test_runtime_check2 131; CHECK-NOT: <4 x float> 132define void @test_runtime_check2(float* %a, float %b, i64 %offset, i64 %offset2, i64 %n, float* %c) { 133entry: 134 br label %for.body 135 136for.body: 137 %iv = phi i64 [ 0, %entry ], [ %iv.next, %for.body ] 138 %ind.sum = add i64 %iv, %offset 139 %arr.idx = getelementptr inbounds float, float* %a, i64 %ind.sum 140 %l1 = load float, float* %arr.idx, align 4 141 %ind.sum2 = add i64 %iv, %offset2 142 %arr.idx2 = getelementptr inbounds float, float* %a, i64 %ind.sum2 143 %l2 = load float, float* %arr.idx2, align 4 144 %m = fmul fast float %b, %l2 145 %ad = fadd fast float %l1, %m 146 store float %ad, float* %arr.idx, align 4 147 %c.ind = add i64 %iv, -1 148 %c.idx = getelementptr inbounds float, float* %c, i64 %c.ind 149 %lc = load float, float* %c.idx, align 4 150 %vc = fadd float %lc, 1.0 151 %c.idx2 = getelementptr inbounds float, float* %c, i64 %iv 152 store float %vc, float* %c.idx2 153 %iv.next = add nuw nsw i64 %iv, 1 154 %exitcond = icmp eq i64 %iv.next, %n 155 br i1 %exitcond, label %loopexit, label %for.body 156 157loopexit: 158 ret void 159} 160 161; CHECK: !9 = !DILocation(line: 101, column: 1, scope: !{{.*}}) 162 163define dso_local void @forced_optsize(i64* noalias nocapture readonly %x_p, i64* noalias nocapture readonly %y_p, i64* noalias nocapture %z_p) minsize optsize { 164; 165; FORCED_OPTSIZE: remark: <unknown>:0:0: loop not vectorized: runtime pointer checks needed. Enable vectorization of this loop with '#pragma clang loop vectorize(enable)' when compiling with -Os/-Oz 166; FORCED_OPTSIZE-LABEL: @forced_optsize( 167; FORCED_OPTSIZE-NOT: vector.body: 168; 169entry: 170 br label %for.body 171 172for.cond.cleanup: 173 ret void 174 175for.body: 176 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 177 %arrayidx = getelementptr inbounds i64, i64* %x_p, i64 %indvars.iv 178 %0 = load i64, i64* %arrayidx, align 8 179 %arrayidx2 = getelementptr inbounds i64, i64* %y_p, i64 %indvars.iv 180 %1 = load i64, i64* %arrayidx2, align 8 181 %add = add nsw i64 %1, %0 182 %arrayidx4 = getelementptr inbounds i64, i64* %z_p, i64 %indvars.iv 183 store i64 %add, i64* %arrayidx4, align 8 184 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 185 %exitcond = icmp eq i64 %indvars.iv.next, 128 186 br i1 %exitcond, label %for.cond.cleanup, label %for.body, !llvm.loop !12 187} 188 189!llvm.module.flags = !{!0, !1} 190!llvm.dbg.cu = !{!9} 191!0 = !{i32 2, !"Dwarf Version", i32 4} 192!1 = !{i32 2, !"Debug Info Version", i32 3} 193 194!2 = !{} 195!3 = !DISubroutineType(types: !2) 196!4 = !DIFile(filename: "test.cpp", directory: "/tmp") 197!5 = distinct !DISubprogram(name: "foo", scope: !4, file: !4, line: 99, type: !3, isLocal: false, isDefinition: true, scopeLine: 100, flags: DIFlagPrototyped, isOptimized: false, unit: !9, retainedNodes: !2) 198!6 = !DILocation(line: 100, column: 1, scope: !5) 199!7 = !DILocation(line: 101, column: 1, scope: !5) 200!8 = !DILocation(line: 102, column: 1, scope: !5) 201!9 = distinct !DICompileUnit(language: DW_LANG_C99, producer: "clang", 202 file: !10, 203 isOptimized: true, flags: "-O2", 204 splitDebugFilename: "abc.debug", emissionKind: 2) 205!10 = !DIFile(filename: "path/to/file", directory: "/path/to/dir") 206!11 = !{i32 2, !"Debug Info Version", i32 3} 207!12 = distinct !{!12, !13, !14} 208!13 = !{!"llvm.loop.vectorize.width", i32 2} 209!14 = !{!"llvm.loop.vectorize.enable", i1 true} 210