1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; This is the loop in c++ being vectorize in this file with 3; shuffle reverse 4 5;#pragma clang loop vectorize_width(4, fixed) 6; for (long int i = N - 1; i >= 0; i--) 7; { 8; if (cond[i]) 9; a[i] += 1; 10; } 11 12; The test checks if the mask is being correctly created, reverted and used 13 14; RUN: opt -loop-vectorize -dce -instcombine -mtriple aarch64-linux-gnu -S \ 15; RUN: -prefer-predicate-over-epilogue=scalar-epilogue < %s | FileCheck %s 16 17target datalayout = "e-m:e-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128" 18target triple = "aarch64-unknown-linux-gnu" 19 20define void @vector_reverse_mask_v4i1(double* %a, double* %cond, i64 %N) #0 { 21; CHECK-LABEL: @vector_reverse_mask_v4i1( 22; CHECK-NEXT: entry: 23; CHECK-NEXT: [[CMP7:%.*]] = icmp sgt i64 [[N:%.*]], 0 24; CHECK-NEXT: br i1 [[CMP7]], label [[FOR_BODY_PREHEADER:%.*]], label [[FOR_COND_CLEANUP:%.*]] 25; CHECK: for.body.preheader: 26; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ult i64 [[N]], 8 27; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_MEMCHECK:%.*]] 28; CHECK: vector.memcheck: 29; CHECK-NEXT: [[SCEVGEP:%.*]] = getelementptr double, double* [[A:%.*]], i64 [[N]] 30; CHECK-NEXT: [[SCEVGEP4:%.*]] = getelementptr double, double* [[COND:%.*]], i64 [[N]] 31; CHECK-NEXT: [[BOUND0:%.*]] = icmp ugt double* [[SCEVGEP4]], [[A]] 32; CHECK-NEXT: [[BOUND1:%.*]] = icmp ugt double* [[SCEVGEP]], [[COND]] 33; CHECK-NEXT: [[FOUND_CONFLICT:%.*]] = and i1 [[BOUND0]], [[BOUND1]] 34; CHECK-NEXT: br i1 [[FOUND_CONFLICT]], label [[SCALAR_PH]], label [[VECTOR_PH:%.*]] 35; CHECK: vector.ph: 36; CHECK-NEXT: [[N_VEC:%.*]] = and i64 [[N]], -8 37; CHECK-NEXT: [[IND_END:%.*]] = and i64 [[N]], 7 38; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 39; CHECK: vector.body: 40; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 41; CHECK-NEXT: [[TMP0:%.*]] = xor i64 [[INDEX]], -1 42; CHECK-NEXT: [[TMP1:%.*]] = add i64 [[TMP0]], [[N]] 43; CHECK-NEXT: [[TMP2:%.*]] = getelementptr inbounds double, double* [[COND]], i64 [[TMP1]] 44; CHECK-NEXT: [[TMP3:%.*]] = getelementptr inbounds double, double* [[TMP2]], i64 -3 45; CHECK-NEXT: [[TMP4:%.*]] = bitcast double* [[TMP3]] to <4 x double>* 46; CHECK-NEXT: [[WIDE_LOAD:%.*]] = load <4 x double>, <4 x double>* [[TMP4]], align 8, !alias.scope !0 47; CHECK-NEXT: [[REVERSE:%.*]] = shufflevector <4 x double> [[WIDE_LOAD]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 48; CHECK-NEXT: [[TMP5:%.*]] = getelementptr inbounds double, double* [[TMP2]], i64 -4 49; CHECK-NEXT: [[TMP6:%.*]] = getelementptr inbounds double, double* [[TMP5]], i64 -3 50; CHECK-NEXT: [[TMP7:%.*]] = bitcast double* [[TMP6]] to <4 x double>* 51; CHECK-NEXT: [[WIDE_LOAD6:%.*]] = load <4 x double>, <4 x double>* [[TMP7]], align 8, !alias.scope !0 52; CHECK-NEXT: [[REVERSE7:%.*]] = shufflevector <4 x double> [[WIDE_LOAD6]], <4 x double> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 53; CHECK-NEXT: [[TMP8:%.*]] = fcmp une <4 x double> [[REVERSE]], zeroinitializer 54; CHECK-NEXT: [[TMP9:%.*]] = fcmp une <4 x double> [[REVERSE7]], zeroinitializer 55; CHECK-NEXT: [[TMP10:%.*]] = getelementptr double, double* [[A]], i64 [[TMP1]] 56; CHECK-NEXT: [[TMP11:%.*]] = getelementptr double, double* [[TMP10]], i64 -3 57; CHECK-NEXT: [[REVERSE8:%.*]] = shufflevector <4 x i1> [[TMP8]], <4 x i1> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 58; CHECK-NEXT: [[TMP12:%.*]] = bitcast double* [[TMP11]] to <4 x double>* 59; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0v4f64(<4 x double>* [[TMP12]], i32 8, <4 x i1> [[REVERSE8]], <4 x double> poison), !alias.scope !3, !noalias !0 60; CHECK-NEXT: [[TMP13:%.*]] = getelementptr double, double* [[TMP10]], i64 -4 61; CHECK-NEXT: [[TMP14:%.*]] = getelementptr double, double* [[TMP13]], i64 -3 62; CHECK-NEXT: [[REVERSE10:%.*]] = shufflevector <4 x i1> [[TMP9]], <4 x i1> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0> 63; CHECK-NEXT: [[TMP15:%.*]] = bitcast double* [[TMP14]] to <4 x double>* 64; CHECK-NEXT: [[WIDE_MASKED_LOAD11:%.*]] = call <4 x double> @llvm.masked.load.v4f64.p0v4f64(<4 x double>* [[TMP15]], i32 8, <4 x i1> [[REVERSE10]], <4 x double> poison), !alias.scope !3, !noalias !0 65; CHECK-NEXT: [[TMP16:%.*]] = fadd <4 x double> [[WIDE_MASKED_LOAD]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00> 66; CHECK-NEXT: [[TMP17:%.*]] = fadd <4 x double> [[WIDE_MASKED_LOAD11]], <double 1.000000e+00, double 1.000000e+00, double 1.000000e+00, double 1.000000e+00> 67; CHECK-NEXT: [[TMP18:%.*]] = bitcast double* [[TMP11]] to <4 x double>* 68; CHECK-NEXT: call void @llvm.masked.store.v4f64.p0v4f64(<4 x double> [[TMP16]], <4 x double>* [[TMP18]], i32 8, <4 x i1> [[REVERSE8]]), !alias.scope !3, !noalias !0 69; CHECK-NEXT: [[TMP19:%.*]] = bitcast double* [[TMP14]] to <4 x double>* 70; CHECK-NEXT: call void @llvm.masked.store.v4f64.p0v4f64(<4 x double> [[TMP17]], <4 x double>* [[TMP19]], i32 8, <4 x i1> [[REVERSE10]]), !alias.scope !3, !noalias !0 71; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], 8 72; CHECK-NEXT: [[TMP20:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 73; CHECK-NEXT: br i1 [[TMP20]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP5:![0-9]+]] 74; CHECK: middle.block: 75; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_VEC]], [[N]] 76; CHECK-NEXT: br i1 [[CMP_N]], label [[FOR_COND_CLEANUP_LOOPEXIT:%.*]], label [[SCALAR_PH]] 77; CHECK: scalar.ph: 78; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[IND_END]], [[MIDDLE_BLOCK]] ], [ [[N]], [[FOR_BODY_PREHEADER]] ], [ [[N]], [[VECTOR_MEMCHECK]] ] 79; CHECK-NEXT: br label [[FOR_BODY:%.*]] 80; CHECK: for.cond.cleanup.loopexit: 81; CHECK-NEXT: br label [[FOR_COND_CLEANUP]] 82; CHECK: for.cond.cleanup: 83; CHECK-NEXT: ret void 84; CHECK: for.body: 85; CHECK-NEXT: [[I_08_IN:%.*]] = phi i64 [ [[I_08:%.*]], [[FOR_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 86; CHECK-NEXT: [[I_08]] = add nsw i64 [[I_08_IN]], -1 87; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds double, double* [[COND]], i64 [[I_08]] 88; CHECK-NEXT: [[TMP21:%.*]] = load double, double* [[ARRAYIDX]], align 8 89; CHECK-NEXT: [[TOBOOL:%.*]] = fcmp une double [[TMP21]], 0.000000e+00 90; CHECK-NEXT: br i1 [[TOBOOL]], label [[IF_THEN:%.*]], label [[FOR_INC]] 91; CHECK: if.then: 92; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds double, double* [[A]], i64 [[I_08]] 93; CHECK-NEXT: [[TMP22:%.*]] = load double, double* [[ARRAYIDX1]], align 8 94; CHECK-NEXT: [[ADD:%.*]] = fadd double [[TMP22]], 1.000000e+00 95; CHECK-NEXT: store double [[ADD]], double* [[ARRAYIDX1]], align 8 96; CHECK-NEXT: br label [[FOR_INC]] 97; CHECK: for.inc: 98; CHECK-NEXT: [[CMP:%.*]] = icmp sgt i64 [[I_08_IN]], 1 99; CHECK-NEXT: br i1 [[CMP]], label [[FOR_BODY]], label [[FOR_COND_CLEANUP_LOOPEXIT]], !llvm.loop [[LOOP8:![0-9]+]] 100; 101 102entry: 103 %cmp7 = icmp sgt i64 %N, 0 104 br i1 %cmp7, label %for.body, label %for.cond.cleanup 105 106for.cond.cleanup: ; preds = %for.cond.cleanup, %entry 107 ret void 108 109for.body: ; preds = %for.body, %entry 110 %i.08.in = phi i64 [ %i.08, %for.inc ], [ %N, %entry ] 111 %i.08 = add nsw i64 %i.08.in, -1 112 %arrayidx = getelementptr inbounds double, double* %cond, i64 %i.08 113 %0 = load double, double* %arrayidx, align 8 114 %tobool = fcmp une double %0, 0.000000e+00 115 br i1 %tobool, label %if.then, label %for.inc 116 117if.then: ; preds = %for.body 118 %arrayidx1 = getelementptr inbounds double, double* %a, i64 %i.08 119 %1 = load double, double* %arrayidx1, align 8 120 %add = fadd double %1, 1.000000e+00 121 store double %add, double* %arrayidx1, align 8 122 br label %for.inc 123 124for.inc: ; preds = %for.body, %if.then 125 %cmp = icmp sgt i64 %i.08.in, 1 126 br i1 %cmp, label %for.body, label %for.cond.cleanup, !llvm.loop !0 127} 128 129attributes #0 = {"target-cpu"="generic" "target-features"="+neon,+sve" vscale_range(2,0) } 130 131 132!0 = distinct !{!0, !1, !2, !3, !4, !5} 133!1 = !{!"llvm.loop.mustprogress"} 134!2 = !{!"llvm.loop.vectorize.width", i32 4} 135!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 false} 136!4 = !{!"llvm.loop.vectorize.enable", i1 true} 137!5 = !{!"llvm.loop.interleave.count", i32 2} 138