program-order.ll 2.0 KB

1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950515253545556575859606162636465
  1. ; RUN: opt -loop-distribute -S -verify-loop-info -verify-dom-info < %s \
  2. ; RUN: | FileCheck %s
  3. ; Distributing this loop to avoid the dependence cycle would require to
  4. ; reorder S1 and S2 to form the two partitions: {S2} | {S1, S3}. The analysis
  5. ; provided by LoopAccessAnalysis does not allow us to reorder memory
  6. ; operations so make sure we bail on this loop.
  7. ;
  8. ; for (i = 0; i < n; i++) {
  9. ; S1: d = D[i];
  10. ; S2: A[i + 1] = A[i] * B[i];
  11. ; S3: C[i] = d * E[i];
  12. ; }
  13. target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
  14. target triple = "x86_64-apple-macosx10.10.0"
  15. define void @f(i32* noalias %a,
  16. i32* noalias %b,
  17. i32* noalias %c,
  18. i32* noalias %d,
  19. i32* noalias %e) {
  20. entry:
  21. br label %for.body
  22. ; CHECK: entry:
  23. ; CHECK: br label %for.body
  24. ; CHECK: for.body:
  25. ; CHECK: br i1 %exitcond, label %for.end, label %for.body
  26. ; CHECK: for.end:
  27. ; CHECK: ret void
  28. for.body: ; preds = %for.body, %entry
  29. %ind = phi i64 [ 0, %entry ], [ %add, %for.body ]
  30. %arrayidxA = getelementptr inbounds i32, i32* %a, i64 %ind
  31. %loadA = load i32, i32* %arrayidxA, align 4
  32. %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %ind
  33. %loadB = load i32, i32* %arrayidxB, align 4
  34. %mulA = mul i32 %loadB, %loadA
  35. %arrayidxD = getelementptr inbounds i32, i32* %d, i64 %ind
  36. %loadD = load i32, i32* %arrayidxD, align 4
  37. %add = add nuw nsw i64 %ind, 1
  38. %arrayidxA_plus_4 = getelementptr inbounds i32, i32* %a, i64 %add
  39. store i32 %mulA, i32* %arrayidxA_plus_4, align 4
  40. %arrayidxC = getelementptr inbounds i32, i32* %c, i64 %ind
  41. %arrayidxE = getelementptr inbounds i32, i32* %e, i64 %ind
  42. %loadE = load i32, i32* %arrayidxE, align 4
  43. %mulC = mul i32 %loadD, %loadE
  44. store i32 %mulC, i32* %arrayidxC, align 4
  45. %exitcond = icmp eq i64 %add, 20
  46. br i1 %exitcond, label %for.end, label %for.body
  47. for.end: ; preds = %for.body
  48. ret void
  49. }