Home | History | Annotate | Download | only in LoopUnroll
      1 ; RUN: opt -S -unroll-runtime -loop-unroll < %s | FileCheck %s
      2 
      3 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"
      4 
      5 ;; Check that we don't emit expensive instructions to compute trip
      6 ;; counts when unrolling loops.
      7 
      8 define i32 @test(i64 %v12, i8* %array, i64* %loc) {
      9 ; CHECK-LABEL: @test(
     10 ; CHECK-NOT: udiv
     11 entry:
     12   %step = load i64, i64* %loc, !range !0
     13   br label %loop
     14 
     15 loop:                                           ; preds = %entry, %loop
     16   %k.015 = phi i64 [ %v15, %loop ], [ %v12, %entry ]
     17   %v14 = getelementptr inbounds i8, i8* %array, i64 %k.015
     18   store i8 0, i8* %v14
     19   %v15 = add nuw nsw i64 %k.015, %step
     20   %v16 = icmp slt i64 %v15, 8193
     21   br i1 %v16, label %loop, label %loopexit
     22 
     23 loopexit:                             ; preds = %loop
     24   ret i32 0
     25 }
     26 
     27 ;; Though SCEV for loop tripcount contains division,
     28 ;; it shouldn't be considered expensive, since the division already
     29 ;; exists in the code and we don't need to expand it once more.
     30 ;; Thus, it shouldn't prevent us from unrolling the loop.
     31 
     32 define i32 @test2(i64* %loc, i64 %conv7) {
     33 ; CHECK-LABEL: @test2(
     34 ; CHECK: udiv
     35 ; CHECK: udiv
     36 ; CHECK-NOT: udiv
     37 ; CHECK-LABEL: for.body
     38 entry:
     39   %rem0 = load i64, i64* %loc, align 8
     40   %ExpensiveComputation = udiv i64 %rem0, 42 ; <<< Extra computations are added to the trip-count expression
     41   br label %bb1
     42 bb1:
     43   %div11 = udiv i64 %ExpensiveComputation, %conv7
     44   %cmp.i38 = icmp ugt i64 %div11, 1
     45   %div12 = select i1 %cmp.i38, i64 %div11, i64 1
     46   br label %for.body
     47 for.body:
     48   %rem1 = phi i64 [ %rem0, %bb1 ], [ %rem2, %for.body ]
     49   %k1 = phi i64 [ %div12, %bb1 ], [ %dec, %for.body ]
     50   %mul1 = mul i64 %rem1, 48271
     51   %rem2 = urem i64 %mul1, 2147483647
     52   %dec = add i64 %k1, -1
     53   %cmp = icmp eq i64 %dec, 0
     54   br i1 %cmp, label %exit, label %for.body
     55 exit:
     56   %rem3 = phi i64 [ %rem2, %for.body ]
     57   store i64 %rem3, i64* %loc, align 8
     58   ret i32 0
     59 }
     60 
     61 !0 = !{i64 1, i64 100}
     62