test/tools/llvm-mca/X86/Barcelona/dependency-breaking-pcmpeq.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -timeline -timeline-max-iterations=3 -iterations=1500 < %s | FileCheck %s
   3
   4 # All of the vector packed compares from this test are dependency breaking
   5 # instructions. That means, there is no RAW dependency between any of the
   6 # instructions, and the code can be fully parallelized in hardware.
   7
   8 pcmpeqb %mm0, %mm0
   9 pcmpeqd %mm0, %mm0
  10 pcmpeqw %mm0, %mm0
  11
  12 pcmpeqb %xmm0, %xmm0
  13 pcmpeqd %xmm0, %xmm0
  14 pcmpeqq %xmm0, %xmm0
  15 pcmpeqw %xmm0, %xmm0
  16
  17 # CHECK:      Iterations:        1500
  18 # CHECK-NEXT: Instructions:      10500
  19 # CHECK-NEXT: Total Cycles:      13503
  20 # CHECK-NEXT: Total uOps:        10500
  21
  22 # CHECK:      Dispatch Width:    4
  23 # CHECK-NEXT: uOps Per Cycle:    0.78
  24 # CHECK-NEXT: IPC:               0.78
  25 # CHECK-NEXT: Block RThroughput: 3.0
  26
  27 # CHECK:      Instruction Info:
  28 # CHECK-NEXT: [1]: #uOps
  29 # CHECK-NEXT: [2]: Latency
  30 # CHECK-NEXT: [3]: RThroughput
  31 # CHECK-NEXT: [4]: MayLoad
  32 # CHECK-NEXT: [5]: MayStore
  33 # CHECK-NEXT: [6]: HasSideEffects (U)
  34
  35 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  36 # CHECK-NEXT:  1      3     1.00                        pcmpeqb %mm0, %mm0
  37 # CHECK-NEXT:  1      3     1.00                        pcmpeqd %mm0, %mm0
  38 # CHECK-NEXT:  1      3     1.00                        pcmpeqw %mm0, %mm0
  39 # CHECK-NEXT:  1      1     0.50                        pcmpeqb %xmm0, %xmm0
  40 # CHECK-NEXT:  1      1     0.50                        pcmpeqd %xmm0, %xmm0
  41 # CHECK-NEXT:  1      1     0.50                        pcmpeqq %xmm0, %xmm0
  42 # CHECK-NEXT:  1      1     0.50                        pcmpeqw %xmm0, %xmm0
  43
  44 # CHECK:      Resources:
  45 # CHECK-NEXT: [0]   - SBDivider
  46 # CHECK-NEXT: [1]   - SBFPDivider
  47 # CHECK-NEXT: [2]   - SBPort0
  48 # CHECK-NEXT: [3]   - SBPort1
  49 # CHECK-NEXT: [4]   - SBPort4
  50 # CHECK-NEXT: [5]   - SBPort5
  51 # CHECK-NEXT: [6.0] - SBPort23
  52 # CHECK-NEXT: [6.1] - SBPort23
  53
  54 # CHECK:      Resource pressure per iteration:
  55 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]
  56 # CHECK-NEXT:  -      -      -     4.01    -     2.99    -      -
  57
  58 # CHECK:      Resource pressure by instruction:
  59 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6.0]  [6.1]  Instructions:
  60 # CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqb   %mm0, %mm0
  61 # CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqd   %mm0, %mm0
  62 # CHECK-NEXT:  -      -      -     1.00    -      -      -      -     pcmpeqw   %mm0, %mm0
  63 # CHECK-NEXT:  -      -      -     0.01    -     0.99    -      -     pcmpeqb   %xmm0, %xmm0
  64 # CHECK-NEXT:  -      -      -     0.01    -     0.99    -      -     pcmpeqd   %xmm0, %xmm0
  65 # CHECK-NEXT:  -      -      -     0.01    -     0.99    -      -     pcmpeqq   %xmm0, %xmm0
  66 # CHECK-NEXT:  -      -      -     0.99    -     0.01    -      -     pcmpeqw   %xmm0, %xmm0
  67
  68 # CHECK:      Timeline view:
  69 # CHECK-NEXT:                     0123456789
  70 # CHECK-NEXT: Index     0123456789          0123456789
  71
  72 # CHECK:      [0,0]     DeeeER    .    .    .    .   .   pcmpeqb        %mm0, %mm0
  73 # CHECK-NEXT: [0,1]     D===eeeER .    .    .    .   .   pcmpeqd        %mm0, %mm0
  74 # CHECK-NEXT: [0,2]     D======eeeER   .    .    .   .   pcmpeqw        %mm0, %mm0
  75 # CHECK-NEXT: [0,3]     DeE--------R   .    .    .   .   pcmpeqb        %xmm0, %xmm0
  76 # CHECK-NEXT: [0,4]     .DeE-------R   .    .    .   .   pcmpeqd        %xmm0, %xmm0
  77 # CHECK-NEXT: [0,5]     .D=eE------R   .    .    .   .   pcmpeqq        %xmm0, %xmm0
  78 # CHECK-NEXT: [0,6]     .D==eE-----R   .    .    .   .   pcmpeqw        %xmm0, %xmm0
  79 # CHECK-NEXT: [1,0]     .D========eeeER.    .    .   .   pcmpeqb        %mm0, %mm0
  80 # CHECK-NEXT: [1,1]     . D==========eeeER  .    .   .   pcmpeqd        %mm0, %mm0
  81 # CHECK-NEXT: [1,2]     . D=============eeeER    .   .   pcmpeqw        %mm0, %mm0
  82 # CHECK-NEXT: [1,3]     . D==eE-------------R    .   .   pcmpeqb        %xmm0, %xmm0
  83 # CHECK-NEXT: [1,4]     . D===eE------------R    .   .   pcmpeqd        %xmm0, %xmm0
  84 # CHECK-NEXT: [1,5]     .  D===eE-----------R    .   .   pcmpeqq        %xmm0, %xmm0
  85 # CHECK-NEXT: [1,6]     .  D====eE----------R    .   .   pcmpeqw        %xmm0, %xmm0
  86 # CHECK-NEXT: [2,0]     .  D===============eeeER .   .   pcmpeqb        %mm0, %mm0
  87 # CHECK-NEXT: [2,1]     .  D==================eeeER  .   pcmpeqd        %mm0, %mm0
  88 # CHECK-NEXT: [2,2]     .   D====================eeeER   pcmpeqw        %mm0, %mm0
  89 # CHECK-NEXT: [2,3]     .   D====eE------------------R   pcmpeqb        %xmm0, %xmm0
  90 # CHECK-NEXT: [2,4]     .   D=====eE-----------------R   pcmpeqd        %xmm0, %xmm0
  91 # CHECK-NEXT: [2,5]     .   D======eE----------------R   pcmpeqq        %xmm0, %xmm0
  92 # CHECK-NEXT: [2,6]     .    D======eE---------------R   pcmpeqw        %xmm0, %xmm0
  93
  94 # CHECK:      Average Wait times (based on the timeline view):
  95 # CHECK-NEXT: [0]: Executions
  96 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
  97 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
  98 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
  99
 100 # CHECK:            [0]    [1]    [2]    [3]
 101 # CHECK-NEXT: 0.     3     8.7    0.3    0.0       pcmpeqb      %mm0, %mm0
 102 # CHECK-NEXT: 1.     3     11.3   0.0    0.0       pcmpeqd      %mm0, %mm0
 103 # CHECK-NEXT: 2.     3     14.0   0.0    0.0       pcmpeqw      %mm0, %mm0
 104 # CHECK-NEXT: 3.     3     3.0    0.3    13.0      pcmpeqb      %xmm0, %xmm0
 105 # CHECK-NEXT: 4.     3     3.7    0.0    12.0      pcmpeqd      %xmm0, %xmm0
 106 # CHECK-NEXT: 5.     3     4.3    0.0    11.0      pcmpeqq      %xmm0, %xmm0
 107 # CHECK-NEXT: 6.     3     5.0    0.0    10.0      pcmpeqw      %xmm0, %xmm0