1 ; RUN: opt < %s -loop-reduce -S | FileCheck %s 2 ; 3 ; Test LSR's ability to prune formulae that refer to nonexistant 4 ; AddRecs in other loops. 5 ; 6 ; Unable to reduce this case further because it requires LSR to exceed 7 ; ComplexityLimit. 8 ; 9 ; We really just want to ensure that LSR can process this loop without 10 ; finding an unsatisfactory solution and bailing out. I've added 11 ; dummyout, an obvious candidate for postinc replacement so we can 12 ; verify that LSR removes it. 13 14 target 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" 15 target triple = "x86_64-apple-darwin" 16 17 ; CHECK: @test 18 ; CHECK: for.body: 19 ; CHECK: %lsr.iv 20 ; CHECK-NOT: %dummyout 21 ; CHECK: ret 22 define i64 @test(i64 %count, float* nocapture %srcrow, i32* nocapture %destrow) nounwind uwtable ssp { 23 entry: 24 %cmp34 = icmp eq i64 %count, 0 25 br i1 %cmp34, label %for.end29, label %for.body 26 27 for.body: ; preds = %entry, %for.body 28 %dummyiv = phi i64 [ %dummycnt, %for.body ], [ 0, %entry ] 29 %indvars.iv39 = phi i64 [ %indvars.iv.next40, %for.body ], [ 0, %entry ] 30 %dp.036 = phi i32* [ %add.ptr, %for.body ], [ %destrow, %entry ] 31 %p.035 = phi float* [ %incdec.ptr4, %for.body ], [ %srcrow, %entry ] 32 %incdec.ptr = getelementptr inbounds float* %p.035, i64 1 33 %0 = load float* %incdec.ptr, align 4 34 %incdec.ptr2 = getelementptr inbounds float* %p.035, i64 2 35 %1 = load float* %incdec.ptr2, align 4 36 %incdec.ptr3 = getelementptr inbounds float* %p.035, i64 3 37 %2 = load float* %incdec.ptr3, align 4 38 %incdec.ptr4 = getelementptr inbounds float* %p.035, i64 4 39 %3 = load float* %incdec.ptr4, align 4 40 %4 = load i32* %dp.036, align 4 41 %conv5 = fptoui float %0 to i32 42 %or = or i32 %4, %conv5 43 %arrayidx6 = getelementptr inbounds i32* %dp.036, i64 1 44 %5 = load i32* %arrayidx6, align 4 45 %conv7 = fptoui float %1 to i32 46 %or8 = or i32 %5, %conv7 47 %arrayidx9 = getelementptr inbounds i32* %dp.036, i64 2 48 %6 = load i32* %arrayidx9, align 4 49 %conv10 = fptoui float %2 to i32 50 %or11 = or i32 %6, %conv10 51 %arrayidx12 = getelementptr inbounds i32* %dp.036, i64 3 52 %7 = load i32* %arrayidx12, align 4 53 %conv13 = fptoui float %3 to i32 54 %or14 = or i32 %7, %conv13 55 store i32 %or, i32* %dp.036, align 4 56 store i32 %or8, i32* %arrayidx6, align 4 57 store i32 %or11, i32* %arrayidx9, align 4 58 store i32 %or14, i32* %arrayidx12, align 4 59 %add.ptr = getelementptr inbounds i32* %dp.036, i64 4 60 %indvars.iv.next40 = add i64 %indvars.iv39, 4 61 %dummycnt = add i64 %dummyiv, 1 62 %cmp = icmp ult i64 %indvars.iv.next40, %count 63 br i1 %cmp, label %for.body, label %for.cond19.preheader 64 65 for.cond19.preheader: ; preds = %for.body 66 %dummyout = add i64 %dummyiv, 1 67 %rem = and i64 %count, 3 68 %cmp2130 = icmp eq i64 %rem, 0 69 br i1 %cmp2130, label %for.end29, label %for.body23.lr.ph 70 71 for.body23.lr.ph: ; preds = %for.cond19.preheader 72 %8 = and i64 %count, 3 73 br label %for.body23 74 75 for.body23: ; preds = %for.body23, %for.body23.lr.ph 76 %indvars.iv = phi i64 [ 0, %for.body23.lr.ph ], [ %indvars.iv.next, %for.body23 ] 77 %dp.132 = phi i32* [ %add.ptr, %for.body23.lr.ph ], [ %incdec.ptr28, %for.body23 ] 78 %p.131 = phi float* [ %incdec.ptr4, %for.body23.lr.ph ], [ %incdec.ptr24, %for.body23 ] 79 %incdec.ptr24 = getelementptr inbounds float* %p.131, i64 1 80 %9 = load float* %incdec.ptr24, align 4 81 %10 = load i32* %dp.132, align 4 82 %conv25 = fptoui float %9 to i32 83 %or26 = or i32 %10, %conv25 84 store i32 %or26, i32* %dp.132, align 4 85 %indvars.iv.next = add i64 %indvars.iv, 1 86 %incdec.ptr28 = getelementptr inbounds i32* %dp.132, i64 1 87 %exitcond = icmp eq i64 %indvars.iv.next, %8 88 br i1 %exitcond, label %for.end29, label %for.body23 89 90 for.end29: ; preds = %entry, %for.body23, %for.cond19.preheader 91 %result = phi i64 [ 0, %entry ], [ %dummyout, %for.body23 ], [ %dummyout, %for.cond19.preheader ] 92 ret i64 %result 93 } 94
