diamond.ll 4.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109
  1. ; RUN: opt < %s -basicaa -slp-vectorizer -S -mtriple=x86_64-apple-macosx10.8.0 -mcpu=corei7-avx | FileCheck %s
  2. 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"
  3. target triple = "x86_64-apple-macosx10.8.0"
  4. ; int foo(int * restrict B, int * restrict A, int n, int m) {
  5. ; B[0] = n * A[0] + m * A[0];
  6. ; B[1] = n * A[1] + m * A[1];
  7. ; B[2] = n * A[2] + m * A[2];
  8. ; B[3] = n * A[3] + m * A[3];
  9. ; return 0;
  10. ; }
  11. ; CHECK-LABEL: @foo(
  12. ; CHECK: load <4 x i32>
  13. ; CHECK: mul <4 x i32>
  14. ; CHECK: store <4 x i32>
  15. ; CHECK: ret
  16. define i32 @foo(i32* noalias nocapture %B, i32* noalias nocapture %A, i32 %n, i32 %m) #0 {
  17. entry:
  18. %0 = load i32, i32* %A, align 4
  19. %mul238 = add i32 %m, %n
  20. %add = mul i32 %0, %mul238
  21. store i32 %add, i32* %B, align 4
  22. %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 1
  23. %1 = load i32, i32* %arrayidx4, align 4
  24. %add8 = mul i32 %1, %mul238
  25. %arrayidx9 = getelementptr inbounds i32, i32* %B, i64 1
  26. store i32 %add8, i32* %arrayidx9, align 4
  27. %arrayidx10 = getelementptr inbounds i32, i32* %A, i64 2
  28. %2 = load i32, i32* %arrayidx10, align 4
  29. %add14 = mul i32 %2, %mul238
  30. %arrayidx15 = getelementptr inbounds i32, i32* %B, i64 2
  31. store i32 %add14, i32* %arrayidx15, align 4
  32. %arrayidx16 = getelementptr inbounds i32, i32* %A, i64 3
  33. %3 = load i32, i32* %arrayidx16, align 4
  34. %add20 = mul i32 %3, %mul238
  35. %arrayidx21 = getelementptr inbounds i32, i32* %B, i64 3
  36. store i32 %add20, i32* %arrayidx21, align 4
  37. ret i32 0
  38. }
  39. ; int extr_user(int * restrict B, int * restrict A, int n, int m) {
  40. ; B[0] = n * A[0] + m * A[0];
  41. ; B[1] = n * A[1] + m * A[1];
  42. ; B[2] = n * A[2] + m * A[2];
  43. ; B[3] = n * A[3] + m * A[3];
  44. ; return A[0];
  45. ; }
  46. ; CHECK-LABEL: @extr_user(
  47. ; CHECK: load <4 x i32>
  48. ; CHECK: store <4 x i32>
  49. ; CHECK: extractelement <4 x i32>
  50. ; CHECK-NEXT: ret
  51. define i32 @extr_user(i32* noalias nocapture %B, i32* noalias nocapture %A, i32 %n, i32 %m) {
  52. entry:
  53. %0 = load i32, i32* %A, align 4
  54. %mul238 = add i32 %m, %n
  55. %add = mul i32 %0, %mul238
  56. store i32 %add, i32* %B, align 4
  57. %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 1
  58. %1 = load i32, i32* %arrayidx4, align 4
  59. %add8 = mul i32 %1, %mul238
  60. %arrayidx9 = getelementptr inbounds i32, i32* %B, i64 1
  61. store i32 %add8, i32* %arrayidx9, align 4
  62. %arrayidx10 = getelementptr inbounds i32, i32* %A, i64 2
  63. %2 = load i32, i32* %arrayidx10, align 4
  64. %add14 = mul i32 %2, %mul238
  65. %arrayidx15 = getelementptr inbounds i32, i32* %B, i64 2
  66. store i32 %add14, i32* %arrayidx15, align 4
  67. %arrayidx16 = getelementptr inbounds i32, i32* %A, i64 3
  68. %3 = load i32, i32* %arrayidx16, align 4
  69. %add20 = mul i32 %3, %mul238
  70. %arrayidx21 = getelementptr inbounds i32, i32* %B, i64 3
  71. store i32 %add20, i32* %arrayidx21, align 4
  72. ret i32 %0 ;<--------- This value has multiple users
  73. }
  74. ; In this example we have an external user that is not the first element in the vector.
  75. ; CHECK-LABEL: @extr_user1(
  76. ; CHECK: load <4 x i32>
  77. ; CHECK: store <4 x i32>
  78. ; CHECK: extractelement <4 x i32>
  79. ; CHECK-NEXT: ret
  80. define i32 @extr_user1(i32* noalias nocapture %B, i32* noalias nocapture %A, i32 %n, i32 %m) {
  81. entry:
  82. %0 = load i32, i32* %A, align 4
  83. %mul238 = add i32 %m, %n
  84. %add = mul i32 %0, %mul238
  85. store i32 %add, i32* %B, align 4
  86. %arrayidx4 = getelementptr inbounds i32, i32* %A, i64 1
  87. %1 = load i32, i32* %arrayidx4, align 4
  88. %add8 = mul i32 %1, %mul238
  89. %arrayidx9 = getelementptr inbounds i32, i32* %B, i64 1
  90. store i32 %add8, i32* %arrayidx9, align 4
  91. %arrayidx10 = getelementptr inbounds i32, i32* %A, i64 2
  92. %2 = load i32, i32* %arrayidx10, align 4
  93. %add14 = mul i32 %2, %mul238
  94. %arrayidx15 = getelementptr inbounds i32, i32* %B, i64 2
  95. store i32 %add14, i32* %arrayidx15, align 4
  96. %arrayidx16 = getelementptr inbounds i32, i32* %A, i64 3
  97. %3 = load i32, i32* %arrayidx16, align 4
  98. %add20 = mul i32 %3, %mul238
  99. %arrayidx21 = getelementptr inbounds i32, i32* %B, i64 3
  100. store i32 %add20, i32* %arrayidx21, align 4
  101. ret i32 %1 ;<--------- This value has multiple users
  102. }