1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py 2; RUN: llc < %s -mtriple=riscv64 | FileCheck %s 3 4 5; Make sure we don't emit a pair of shift for the zext in the preheader. We 6; can tell that bit 31 is 0 in the preheader and rely on %n already being 7; sign extended without adding zeros explicitly. 8define void @test1(ptr nocapture noundef %a, i32 noundef signext %n) { 9; CHECK-LABEL: test1: 10; CHECK: # %bb.0: # %entry 11; CHECK-NEXT: blez a1, .LBB0_3 12; CHECK-NEXT: # %bb.1: # %for.body.preheader 13; CHECK-NEXT: slli a1, a1, 2 14; CHECK-NEXT: add a1, a0, a1 15; CHECK-NEXT: .LBB0_2: # %for.body 16; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 17; CHECK-NEXT: lw a2, 0(a0) 18; CHECK-NEXT: addi a2, a2, 4 19; CHECK-NEXT: sw a2, 0(a0) 20; CHECK-NEXT: addi a0, a0, 4 21; CHECK-NEXT: bne a0, a1, .LBB0_2 22; CHECK-NEXT: .LBB0_3: # %for.cond.cleanup 23; CHECK-NEXT: ret 24entry: 25 %cmp3 = icmp sgt i32 %n, 0 26 br i1 %cmp3, label %for.body.preheader, label %for.cond.cleanup 27 28for.body.preheader: ; preds = %entry 29 %wide.trip.count = zext nneg i32 %n to i64 30 br label %for.body 31 32for.cond.cleanup: ; preds = %for.body, %entry 33 ret void 34 35for.body: ; preds = %for.body.preheader, %for.body 36 %indvars.iv = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next, %for.body ] 37 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv 38 %0 = load i32, ptr %arrayidx, align 4 39 %add = add nsw i32 %0, 4 40 store i32 %add, ptr %arrayidx, align 4 41 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 42 %exitcond.not = icmp eq i64 %indvars.iv.next, %wide.trip.count 43 br i1 %exitcond.not, label %for.cond.cleanup, label %for.body 44} 45 46; Make sure we convert the 4294967294 in for.body.preheader.new to -2 based on 47; the upper 33 bits being zero by the dominating condition %cmp3. 48define void @test2(ptr nocapture noundef %a, i32 noundef signext %n) { 49; CHECK-LABEL: test2: 50; CHECK: # %bb.0: # %entry 51; CHECK-NEXT: blez a1, .LBB1_7 52; CHECK-NEXT: # %bb.1: # %for.body.preheader 53; CHECK-NEXT: li a3, 1 54; CHECK-NEXT: andi a2, a1, 1 55; CHECK-NEXT: bne a1, a3, .LBB1_3 56; CHECK-NEXT: # %bb.2: 57; CHECK-NEXT: li a3, 0 58; CHECK-NEXT: j .LBB1_5 59; CHECK-NEXT: .LBB1_3: # %for.body.preheader.new 60; CHECK-NEXT: li a3, 0 61; CHECK-NEXT: andi a1, a1, -2 62; CHECK-NEXT: addi a4, a0, 4 63; CHECK-NEXT: .LBB1_4: # %for.body 64; CHECK-NEXT: # =>This Inner Loop Header: Depth=1 65; CHECK-NEXT: lw a5, -4(a4) 66; CHECK-NEXT: lw a6, 0(a4) 67; CHECK-NEXT: addi a3, a3, 2 68; CHECK-NEXT: addi a5, a5, 4 69; CHECK-NEXT: addi a6, a6, 4 70; CHECK-NEXT: sw a5, -4(a4) 71; CHECK-NEXT: sw a6, 0(a4) 72; CHECK-NEXT: addi a4, a4, 8 73; CHECK-NEXT: bne a1, a3, .LBB1_4 74; CHECK-NEXT: .LBB1_5: # %for.cond.cleanup.loopexit.unr-lcssa 75; CHECK-NEXT: beqz a2, .LBB1_7 76; CHECK-NEXT: # %bb.6: # %for.body.epil 77; CHECK-NEXT: slli a3, a3, 2 78; CHECK-NEXT: add a0, a0, a3 79; CHECK-NEXT: lw a1, 0(a0) 80; CHECK-NEXT: addi a1, a1, 4 81; CHECK-NEXT: sw a1, 0(a0) 82; CHECK-NEXT: .LBB1_7: # %for.cond.cleanup 83; CHECK-NEXT: ret 84entry: 85 %cmp3 = icmp sgt i32 %n, 0 86 br i1 %cmp3, label %for.body.preheader, label %for.cond.cleanup 87 88for.body.preheader: ; preds = %entry 89 %wide.trip.count = zext nneg i32 %n to i64 90 %xtraiter = and i64 %wide.trip.count, 1 91 %0 = icmp eq i32 %n, 1 92 br i1 %0, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body.preheader.new 93 94for.body.preheader.new: ; preds = %for.body.preheader 95 %unroll_iter = and i64 %wide.trip.count, 4294967294 96 br label %for.body 97 98for.cond.cleanup.loopexit.unr-lcssa: ; preds = %for.body, %for.body.preheader 99 %indvars.iv.unr = phi i64 [ 0, %for.body.preheader ], [ %indvars.iv.next.1, %for.body ] 100 %lcmp.mod.not = icmp eq i64 %xtraiter, 0 101 br i1 %lcmp.mod.not, label %for.cond.cleanup, label %for.body.epil 102 103for.body.epil: ; preds = %for.cond.cleanup.loopexit.unr-lcssa 104 %arrayidx.epil = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.unr 105 %1 = load i32, ptr %arrayidx.epil, align 4 106 %add.epil = add nsw i32 %1, 4 107 store i32 %add.epil, ptr %arrayidx.epil, align 4 108 br label %for.cond.cleanup 109 110for.cond.cleanup: ; preds = %for.body.epil, %for.cond.cleanup.loopexit.unr-lcssa, %entry 111 ret void 112 113for.body: ; preds = %for.body, %for.body.preheader.new 114 %indvars.iv = phi i64 [ 0, %for.body.preheader.new ], [ %indvars.iv.next.1, %for.body ] 115 %niter = phi i64 [ 0, %for.body.preheader.new ], [ %niter.next.1, %for.body ] 116 %arrayidx = getelementptr inbounds i32, ptr %a, i64 %indvars.iv 117 %2 = load i32, ptr %arrayidx, align 4 118 %add = add nsw i32 %2, 4 119 store i32 %add, ptr %arrayidx, align 4 120 %indvars.iv.next = or disjoint i64 %indvars.iv, 1 121 %arrayidx.1 = getelementptr inbounds i32, ptr %a, i64 %indvars.iv.next 122 %3 = load i32, ptr %arrayidx.1, align 4 123 %add.1 = add nsw i32 %3, 4 124 store i32 %add.1, ptr %arrayidx.1, align 4 125 %indvars.iv.next.1 = add nuw nsw i64 %indvars.iv, 2 126 %niter.next.1 = add i64 %niter, 2 127 %niter.ncmp.1 = icmp eq i64 %niter.next.1, %unroll_iter 128 br i1 %niter.ncmp.1, label %for.cond.cleanup.loopexit.unr-lcssa, label %for.body 129} 130