1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -loop-vectorize -force-target-instruction-cost=1 -dce -instcombine < %s -S | FileCheck %s 3 4target triple = "aarch64-linux-gnu" 5 6; Test a case where the vectorised induction variable is used to 7; generate a mask: 8; for (long long i = 0; i < n; i++) { 9; if (i & 0x1) 10; a[i] = b[i]; 11; } 12 13define void @cond_ind64(i32* noalias nocapture %a, i32* noalias nocapture readonly %b, i64 %n) #0 { 14; CHECK-LABEL: @cond_ind64( 15; CHECK-NEXT: entry: 16; CHECK-NEXT: [[TMP0:%.*]] = call i64 @llvm.vscale.i64() 17; CHECK-NEXT: [[TMP1:%.*]] = shl i64 [[TMP0]], 2 18; CHECK-NEXT: [[MIN_ITERS_CHECK:%.*]] = icmp ugt i64 [[TMP1]], [[N:%.*]] 19; CHECK-NEXT: br i1 [[MIN_ITERS_CHECK]], label [[SCALAR_PH:%.*]], label [[VECTOR_PH:%.*]] 20; CHECK: vector.ph: 21; CHECK-NEXT: [[TMP2:%.*]] = call i64 @llvm.vscale.i64() 22; CHECK-NEXT: [[TMP3:%.*]] = shl i64 [[TMP2]], 2 23; CHECK-NEXT: [[N_MOD_VF:%.*]] = urem i64 [[N]], [[TMP3]] 24; CHECK-NEXT: [[N_VEC:%.*]] = sub i64 [[N]], [[N_MOD_VF]] 25; CHECK-NEXT: [[TMP4:%.*]] = call <vscale x 4 x i64> @llvm.experimental.stepvector.nxv4i64() 26; CHECK-NEXT: [[TMP5:%.*]] = call i64 @llvm.vscale.i64() 27; CHECK-NEXT: [[TMP6:%.*]] = shl i64 [[TMP5]], 2 28; CHECK-NEXT: [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 4 x i64> poison, i64 [[TMP6]], i64 0 29; CHECK-NEXT: [[DOTSPLAT:%.*]] = shufflevector <vscale x 4 x i64> [[DOTSPLATINSERT]], <vscale x 4 x i64> poison, <vscale x 4 x i32> zeroinitializer 30; CHECK-NEXT: br label [[VECTOR_BODY:%.*]] 31; CHECK: vector.body: 32; CHECK-NEXT: [[INDEX:%.*]] = phi i64 [ 0, [[VECTOR_PH]] ], [ [[INDEX_NEXT:%.*]], [[VECTOR_BODY]] ] 33; CHECK-NEXT: [[VEC_IND:%.*]] = phi <vscale x 4 x i64> [ [[TMP4]], [[VECTOR_PH]] ], [ [[VEC_IND_NEXT:%.*]], [[VECTOR_BODY]] ] 34; CHECK-NEXT: [[TMP7:%.*]] = trunc <vscale x 4 x i64> [[VEC_IND]] to <vscale x 4 x i1> 35; CHECK-NEXT: [[TMP8:%.*]] = getelementptr i32, i32* [[B:%.*]], i64 [[INDEX]] 36; CHECK-NEXT: [[TMP9:%.*]] = bitcast i32* [[TMP8]] to <vscale x 4 x i32>* 37; CHECK-NEXT: [[WIDE_MASKED_LOAD:%.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0nxv4i32(<vscale x 4 x i32>* [[TMP9]], i32 4, <vscale x 4 x i1> [[TMP7]], <vscale x 4 x i32> poison) 38; CHECK-NEXT: [[TMP10:%.*]] = getelementptr i32, i32* [[A:%.*]], i64 [[INDEX]] 39; CHECK-NEXT: [[TMP11:%.*]] = bitcast i32* [[TMP10]] to <vscale x 4 x i32>* 40; CHECK-NEXT: call void @llvm.masked.store.nxv4i32.p0nxv4i32(<vscale x 4 x i32> [[WIDE_MASKED_LOAD]], <vscale x 4 x i32>* [[TMP11]], i32 4, <vscale x 4 x i1> [[TMP7]]) 41; CHECK-NEXT: [[TMP12:%.*]] = call i64 @llvm.vscale.i64() 42; CHECK-NEXT: [[TMP13:%.*]] = shl i64 [[TMP12]], 2 43; CHECK-NEXT: [[INDEX_NEXT]] = add nuw i64 [[INDEX]], [[TMP13]] 44; CHECK-NEXT: [[VEC_IND_NEXT]] = add <vscale x 4 x i64> [[VEC_IND]], [[DOTSPLAT]] 45; CHECK-NEXT: [[TMP14:%.*]] = icmp eq i64 [[INDEX_NEXT]], [[N_VEC]] 46; CHECK-NEXT: br i1 [[TMP14]], label [[MIDDLE_BLOCK:%.*]], label [[VECTOR_BODY]], !llvm.loop [[LOOP0:![0-9]+]] 47; CHECK: middle.block: 48; CHECK-NEXT: [[CMP_N:%.*]] = icmp eq i64 [[N_MOD_VF]], 0 49; CHECK-NEXT: br i1 [[CMP_N]], label [[EXIT:%.*]], label [[SCALAR_PH]] 50; CHECK: scalar.ph: 51; CHECK-NEXT: [[BC_RESUME_VAL:%.*]] = phi i64 [ [[N_VEC]], [[MIDDLE_BLOCK]] ], [ 0, [[ENTRY:%.*]] ] 52; CHECK-NEXT: br label [[FOR_BODY:%.*]] 53; CHECK: for.body: 54; CHECK-NEXT: [[I_08:%.*]] = phi i64 [ [[INC:%.*]], [[FOR_INC:%.*]] ], [ [[BC_RESUME_VAL]], [[SCALAR_PH]] ] 55; CHECK-NEXT: [[AND:%.*]] = and i64 [[I_08]], 1 56; CHECK-NEXT: [[TOBOOL_NOT:%.*]] = icmp eq i64 [[AND]], 0 57; CHECK-NEXT: br i1 [[TOBOOL_NOT]], label [[FOR_INC]], label [[IF_THEN:%.*]] 58; CHECK: if.then: 59; CHECK-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i32, i32* [[B]], i64 [[I_08]] 60; CHECK-NEXT: [[TMP15:%.*]] = load i32, i32* [[ARRAYIDX]], align 4 61; CHECK-NEXT: [[ARRAYIDX1:%.*]] = getelementptr inbounds i32, i32* [[A]], i64 [[I_08]] 62; CHECK-NEXT: store i32 [[TMP15]], i32* [[ARRAYIDX1]], align 4 63; CHECK-NEXT: br label [[FOR_INC]] 64; CHECK: for.inc: 65; CHECK-NEXT: [[INC]] = add nuw nsw i64 [[I_08]], 1 66; CHECK-NEXT: [[EXITCOND_NOT:%.*]] = icmp eq i64 [[INC]], [[N]] 67; CHECK-NEXT: br i1 [[EXITCOND_NOT]], label [[EXIT]], label [[FOR_BODY]], !llvm.loop [[LOOP3:![0-9]+]] 68; CHECK: exit: 69; CHECK-NEXT: ret void 70; 71entry: 72 br label %for.body 73 74for.body: ; preds = %entry, %for.inc 75 %i.08 = phi i64 [ %inc, %for.inc ], [ 0, %entry ] 76 %and = and i64 %i.08, 1 77 %tobool.not = icmp eq i64 %and, 0 78 br i1 %tobool.not, label %for.inc, label %if.then 79 80if.then: ; preds = %for.body 81 %arrayidx = getelementptr inbounds i32, i32* %b, i64 %i.08 82 %0 = load i32, i32* %arrayidx, align 4 83 %arrayidx1 = getelementptr inbounds i32, i32* %a, i64 %i.08 84 store i32 %0, i32* %arrayidx1, align 4 85 br label %for.inc 86 87for.inc: ; preds = %for.body, %if.then 88 %inc = add nuw nsw i64 %i.08, 1 89 %exitcond.not = icmp eq i64 %inc, %n 90 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !0 91 92exit: ; preds = %for.inc 93 ret void 94} 95 96attributes #0 = { "target-features"="+sve" } 97 98!0 = distinct !{!0, !1, !2, !3, !4, !5} 99!1 = !{!"llvm.loop.mustprogress"} 100!2 = !{!"llvm.loop.vectorize.scalable.enable", i1 true} 101!3 = !{!"llvm.loop.vectorize.enable", i1 true} 102!4 = !{!"llvm.loop.vectorize.width", i32 4} 103!5 = !{!"llvm.loop.interleave.count", i32 1} 104