test/tools/llvm-mca/X86/read-after-ld-1.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=sandybridge -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=SANDY
   3 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=haswell -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=HASWELL
   4 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=broadwell -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDWELL
   5 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=SKYLAKE
   6 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BARCELONA
   7 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDVER2
   8 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BTVER2
   9 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=ZNVER1
  10
  11 vdivps  %xmm0, %xmm1, %xmm1
  12 vaddps  (%rax), %xmm1, %xmm1
  13
  14 # ALL:            Iterations:        1
  15 # ALL-NEXT:       Instructions:      2
  16
  17 # BARCELONA-NEXT: Total Cycles:      20
  18 # BARCELONA-NEXT: Total uOps:        3
  19
  20 # BDVER2-NEXT:    Total Cycles:      17
  21 # BDVER2-NEXT:    Total uOps:        2
  22
  23 # BDWELL-NEXT:    Total Cycles:      17
  24 # BDWELL-NEXT:    Total uOps:        3
  25
  26 # BTVER2-NEXT:    Total Cycles:      25
  27 # BTVER2-NEXT:    Total uOps:        2
  28
  29 # HASWELL-NEXT:   Total Cycles:      19
  30 # HASWELL-NEXT:   Total uOps:        3
  31
  32 # SANDY-NEXT:     Total Cycles:      20
  33 # SANDY-NEXT:     Total uOps:        3
  34
  35 # SKYLAKE-NEXT:   Total Cycles:      18
  36 # SKYLAKE-NEXT:   Total uOps:        3
  37
  38 # ZNVER1-NEXT:    Total Cycles:      20
  39 # ZNVER1-NEXT:    Total uOps:        2
  40
  41 # BARCELONA:      Dispatch Width:    4
  42 # BARCELONA-NEXT: uOps Per Cycle:    0.15
  43 # BARCELONA-NEXT: IPC:               0.10
  44 # BARCELONA-NEXT: Block RThroughput: 14.0
  45
  46 # BDVER2:         Dispatch Width:    4
  47 # BDVER2-NEXT:    uOps Per Cycle:    0.12
  48 # BDVER2-NEXT:    IPC:               0.12
  49 # BDVER2-NEXT:    Block RThroughput: 5.0
  50
  51 # BDWELL:         Dispatch Width:    4
  52 # BDWELL-NEXT:    uOps Per Cycle:    0.18
  53 # BDWELL-NEXT:    IPC:               0.12
  54 # BDWELL-NEXT:    Block RThroughput: 5.0
  55
  56 # BTVER2:         Dispatch Width:    2
  57 # BTVER2-NEXT:    uOps Per Cycle:    0.08
  58 # BTVER2-NEXT:    IPC:               0.08
  59 # BTVER2-NEXT:    Block RThroughput: 19.0
  60
  61 # HASWELL:        Dispatch Width:    4
  62 # HASWELL-NEXT:   uOps Per Cycle:    0.16
  63 # HASWELL-NEXT:   IPC:               0.11
  64 # HASWELL-NEXT:   Block RThroughput: 7.0
  65
  66 # SANDY:          Dispatch Width:    4
  67 # SANDY-NEXT:     uOps Per Cycle:    0.15
  68 # SANDY-NEXT:     IPC:               0.10
  69 # SANDY-NEXT:     Block RThroughput: 14.0
  70
  71 # SKYLAKE:        Dispatch Width:    6
  72 # SKYLAKE-NEXT:   uOps Per Cycle:    0.17
  73 # SKYLAKE-NEXT:   IPC:               0.11
  74 # SKYLAKE-NEXT:   Block RThroughput: 3.0
  75
  76 # ZNVER1:         Dispatch Width:    4
  77 # ZNVER1-NEXT:    uOps Per Cycle:    0.10
  78 # ZNVER1-NEXT:    IPC:               0.10
  79 # ZNVER1-NEXT:    Block RThroughput: 1.0
  80
  81 # ALL:            Timeline view:
  82
  83 # BARCELONA-NEXT:                     0123456789
  84 # BARCELONA-NEXT: Index     0123456789
  85
  86 # BDVER2-NEXT:                        0123456
  87 # BDVER2-NEXT:    Index     0123456789
  88
  89 # BDWELL-NEXT:                        0123456
  90 # BDWELL-NEXT:    Index     0123456789
  91
  92 # BTVER2-NEXT:                        0123456789
  93 # BTVER2-NEXT:    Index     0123456789          01234
  94
  95 # HASWELL-NEXT:                       012345678
  96 # HASWELL-NEXT:   Index     0123456789
  97
  98 # SANDY-NEXT:                         0123456789
  99 # SANDY-NEXT:     Index     0123456789
 100
 101 # SKYLAKE-NEXT:                       01234567
 102 # SKYLAKE-NEXT:   Index     0123456789
 103
 104 # ZNVER1-NEXT:                        0123456789
 105 # ZNVER1-NEXT:    Index     0123456789
 106
 107 # BARCELONA:      [0,0]     DeeeeeeeeeeeeeeER  .   vdivps       %xmm0, %xmm1, %xmm1
 108 # BARCELONA-NEXT: [0,1]     D========eeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 109
 110 # BDVER2:         [0,0]     DeeeeeeeeeER   ..   vdivps  %xmm0, %xmm1, %xmm1
 111 # BDVER2-NEXT:    [0,1]     D====eeeeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 112
 113 # BDWELL:         [0,0]     DeeeeeeeeeeeER ..   vdivps  %xmm0, %xmm1, %xmm1
 114 # BDWELL-NEXT:    [0,1]     D======eeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 115
 116 # BTVER2:         [0,0]     DeeeeeeeeeeeeeeeeeeeER  .   vdivps  %xmm0, %xmm1, %xmm1
 117 # BTVER2-NEXT:    [0,1]     D==============eeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 118
 119 # HASWELL:        [0,0]     DeeeeeeeeeeeeeER  .   vdivps        %xmm0, %xmm1, %xmm1
 120 # HASWELL-NEXT:   [0,1]     D=======eeeeeeeeeER   vaddps        (%rax), %xmm1, %xmm1
 121
 122 # SANDY:          [0,0]     DeeeeeeeeeeeeeeER  .   vdivps       %xmm0, %xmm1, %xmm1
 123 # SANDY-NEXT:     [0,1]     D========eeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 124
 125 # SKYLAKE:        [0,0]     DeeeeeeeeeeeER . .   vdivps %xmm0, %xmm1, %xmm1
 126 # SKYLAKE-NEXT:   [0,1]     D=====eeeeeeeeeeER   vaddps (%rax), %xmm1, %xmm1
 127
 128 # ZNVER1:         [0,0]     DeeeeeeeeeeeeeeeER .   vdivps       %xmm0, %xmm1, %xmm1
 129 # ZNVER1-NEXT:    [0,1]     D=======eeeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 130
 131 # ALL:            Average Wait times (based on the timeline view):
 132 # ALL-NEXT:       [0]: Executions
 133 # ALL-NEXT:       [1]: Average time spent waiting in a scheduler's queue
 134 # ALL-NEXT:       [2]: Average time spent waiting in a scheduler's queue while ready
 135 # ALL-NEXT:       [3]: Average time elapsed from WB until retire stage
 136
 137 # ALL:                  [0]    [1]    [2]    [3]
 138 # ALL-NEXT:       0.     1     1.0    1.0    0.0       vdivps   %xmm0, %xmm1, %xmm1
 139
 140 # BARCELONA-NEXT: 1.     1     9.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 141 # BDVER2-NEXT:    1.     1     5.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 142 # BDWELL-NEXT:    1.     1     7.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 143 # BTVER2-NEXT:    1.     1     15.0   0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 144 # HASWELL-NEXT:   1.     1     8.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 145 # SANDY-NEXT:     1.     1     9.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 146 # SKYLAKE-NEXT:   1.     1     6.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 147 # ZNVER1-NEXT:    1.     1     8.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1