1; RUN: opt -passes=loop-vectorize,dce,instcombine -mtriple aarch64-linux-gnu -mattr=+sve \ 2; RUN: -prefer-predicate-over-epilogue=scalar-epilogue -S %s -o - | FileCheck %s 3 4define void @mloadstore_f32(ptr noalias nocapture %a, ptr noalias nocapture readonly %b, i64 %n) { 5; CHECK-LABEL: @mloadstore_f32 6; CHECK: vector.body: 7; CHECK: %[[LOAD1:.*]] = load <vscale x 4 x float>, ptr 8; CHECK-NEXT: %[[MASK:.*]] = fcmp ogt <vscale x 4 x float> %[[LOAD1]], 9; CHECK-NEXT: %[[GEPA:.*]] = getelementptr float, ptr %a, 10; CHECK-NEXT: %[[LOAD2:.*]] = call <vscale x 4 x float> @llvm.masked.load.nxv4f32.p0(ptr %[[GEPA]], i32 4, <vscale x 4 x i1> %[[MASK]] 11; CHECK-NEXT: %[[FADD:.*]] = fadd <vscale x 4 x float> %[[LOAD1]], %[[LOAD2]] 12; CHECK-NEXT: call void @llvm.masked.store.nxv4f32.p0(<vscale x 4 x float> %[[FADD]], ptr %[[GEPA]], i32 4, <vscale x 4 x i1> %[[MASK]]) 13entry: 14 br label %for.body 15 16for.body: ; preds = %entry, %for.inc 17 %i.011 = phi i64 [ %inc, %for.inc ], [ 0, %entry ] 18 %arrayidx = getelementptr inbounds float, ptr %b, i64 %i.011 19 %0 = load float, ptr %arrayidx, align 4 20 %cmp1 = fcmp ogt float %0, 0.000000e+00 21 br i1 %cmp1, label %if.then, label %for.inc 22 23if.then: ; preds = %for.body 24 %arrayidx3 = getelementptr inbounds float, ptr %a, i64 %i.011 25 %1 = load float, ptr %arrayidx3, align 4 26 %add = fadd float %0, %1 27 store float %add, ptr %arrayidx3, align 4 28 br label %for.inc 29 30for.inc: ; preds = %for.body, %if.then 31 %inc = add nuw nsw i64 %i.011, 1 32 %exitcond.not = icmp eq i64 %inc, %n 33 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !0 34 35exit: ; preds = %for.inc 36 ret void 37} 38 39define void @mloadstore_i32(ptr noalias nocapture %a, ptr noalias nocapture readonly %b, i64 %n) { 40; CHECK-LABEL: @mloadstore_i32 41; CHECK: vector.body: 42; CHECK: %[[LOAD1:.*]] = load <vscale x 4 x i32>, ptr 43; CHECK-NEXT: %[[MASK:.*]] = icmp ne <vscale x 4 x i32> %[[LOAD1]], 44; CHECK-NEXT: %[[GEPA:.*]] = getelementptr i32, ptr %a, 45; CHECK-NEXT: %[[LOAD2:.*]] = call <vscale x 4 x i32> @llvm.masked.load.nxv4i32.p0(ptr %[[GEPA]], i32 4, <vscale x 4 x i1> %[[MASK]] 46; CHECK-NEXT: %[[FADD:.*]] = add <vscale x 4 x i32> %[[LOAD1]], %[[LOAD2]] 47; CHECK-NEXT: call void @llvm.masked.store.nxv4i32.p0(<vscale x 4 x i32> %[[FADD]], ptr %[[GEPA]], i32 4, <vscale x 4 x i1> %[[MASK]]) 48entry: 49 br label %for.body 50 51for.body: ; preds = %entry, %for.inc 52 %i.011 = phi i64 [ %inc, %for.inc ], [ 0, %entry ] 53 %arrayidx = getelementptr inbounds i32, ptr %b, i64 %i.011 54 %0 = load i32, ptr %arrayidx, align 4 55 %cmp1 = icmp ne i32 %0, 0 56 br i1 %cmp1, label %if.then, label %for.inc 57 58if.then: ; preds = %for.body 59 %arrayidx3 = getelementptr inbounds i32, ptr %a, i64 %i.011 60 %1 = load i32, ptr %arrayidx3, align 4 61 %add = add i32 %0, %1 62 store i32 %add, ptr %arrayidx3, align 4 63 br label %for.inc 64 65for.inc: ; preds = %for.body, %if.then 66 %inc = add nuw nsw i64 %i.011, 1 67 %exitcond.not = icmp eq i64 %inc, %n 68 br i1 %exitcond.not, label %exit, label %for.body, !llvm.loop !0 69 70exit: ; preds = %for.inc 71 ret void 72} 73 74!0 = distinct !{!0, !1, !2, !3, !4, !5} 75!1 = !{!"llvm.loop.mustprogress"} 76!2 = !{!"llvm.loop.vectorize.width", i32 4} 77!3 = !{!"llvm.loop.vectorize.scalable.enable", i1 true} 78!4 = !{!"llvm.loop.interleave.count", i32 1} 79!5 = !{!"llvm.loop.vectorize.enable", i1 true} 80