llvm/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-prefixes=ALL,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-prefixes=ALL,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-prefixes=ALL,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-prefixes=ALL,SKYLAKE
   6 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=icelake-server -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,SKYLAKE
   7 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=x86-64 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,BARCELONA
   8 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,BDVER2
   9 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,BTVER2
  10 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,ZNVER1
  11 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver2 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,ZNVER2
  12 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver3 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,ZNVER3
  13
  14 vdivps  %xmm0, %xmm1, %xmm1
  15 vaddps  (%rax), %xmm1, %xmm1
  16
  17 # ALL:            Iterations:        1
  18 # ALL-NEXT:       Instructions:      2
  19
  20 # BARCELONA-NEXT: Total Cycles:      20
  21 # BARCELONA-NEXT: Total uOps:        3
  22
  23 # BDVER2-NEXT:    Total Cycles:      17
  24 # BDVER2-NEXT:    Total uOps:        2
  25
  26 # BDWELL-NEXT:    Total Cycles:      17
  27 # BDWELL-NEXT:    Total uOps:        3
  28
  29 # BTVER2-NEXT:    Total Cycles:      25
  30 # BTVER2-NEXT:    Total uOps:        2
  31
  32 # HASWELL-NEXT:   Total Cycles:      19
  33 # HASWELL-NEXT:   Total uOps:        3
  34
  35 # SANDY-NEXT:     Total Cycles:      20
  36 # SANDY-NEXT:     Total uOps:        3
  37
  38 # SKYLAKE-NEXT:   Total Cycles:      18
  39 # SKYLAKE-NEXT:   Total uOps:        3
  40
  41 # ZNVER1-NEXT:    Total Cycles:      20
  42 # ZNVER1-NEXT:    Total uOps:        2
  43
  44 # ZNVER2-NEXT:    Total Cycles:      21
  45 # ZNVER2-NEXT:    Total uOps:        2
  46
  47 # ZNVER3-NEXT:    Total Cycles:      17
  48 # ZNVER3-NEXT:    Total uOps:        2
  49
  50 # BARCELONA:      Dispatch Width:    4
  51 # BARCELONA-NEXT: uOps Per Cycle:    0.15
  52 # BARCELONA-NEXT: IPC:               0.10
  53 # BARCELONA-NEXT: Block RThroughput: 14.0
  54
  55 # BDVER2:         Dispatch Width:    4
  56 # BDVER2-NEXT:    uOps Per Cycle:    0.12
  57 # BDVER2-NEXT:    IPC:               0.12
  58 # BDVER2-NEXT:    Block RThroughput: 5.0
  59
  60 # BDWELL:         Dispatch Width:    4
  61 # BDWELL-NEXT:    uOps Per Cycle:    0.18
  62 # BDWELL-NEXT:    IPC:               0.12
  63 # BDWELL-NEXT:    Block RThroughput: 5.0
  64
  65 # BTVER2:         Dispatch Width:    2
  66 # BTVER2-NEXT:    uOps Per Cycle:    0.08
  67 # BTVER2-NEXT:    IPC:               0.08
  68 # BTVER2-NEXT:    Block RThroughput: 19.0
  69
  70 # HASWELL:        Dispatch Width:    4
  71 # HASWELL-NEXT:   uOps Per Cycle:    0.16
  72 # HASWELL-NEXT:   IPC:               0.11
  73 # HASWELL-NEXT:   Block RThroughput: 7.0
  74
  75 # SANDY:          Dispatch Width:    4
  76 # SANDY-NEXT:     uOps Per Cycle:    0.15
  77 # SANDY-NEXT:     IPC:               0.10
  78 # SANDY-NEXT:     Block RThroughput: 14.0
  79
  80 # SKYLAKE:        Dispatch Width:    6
  81 # SKYLAKE-NEXT:   uOps Per Cycle:    0.17
  82 # SKYLAKE-NEXT:   IPC:               0.11
  83 # SKYLAKE-NEXT:   Block RThroughput: 3.0
  84
  85 # ZNVER1:         Dispatch Width:    4
  86 # ZNVER1-NEXT:    uOps Per Cycle:    0.10
  87 # ZNVER1-NEXT:    IPC:               0.10
  88 # ZNVER1-NEXT:    Block RThroughput: 1.0
  89
  90 # ZNVER2:         Dispatch Width:    4
  91 # ZNVER2-NEXT:    uOps Per Cycle:    0.10
  92 # ZNVER2-NEXT:    IPC:               0.10
  93 # ZNVER2-NEXT:    Block RThroughput: 1.0
  94
  95 # ZNVER3:         Dispatch Width:    6
  96 # ZNVER3-NEXT:    uOps Per Cycle:    0.12
  97 # ZNVER3-NEXT:    IPC:               0.12
  98 # ZNVER3-NEXT:    Block RThroughput: 3.0
  99
 100 # ALL:            Timeline view:
 101
 102 # BARCELONA-NEXT:                     0123456789
 103 # BARCELONA-NEXT: Index     0123456789
 104
 105 # BDVER2-NEXT:                        0123456
 106 # BDVER2-NEXT:    Index     0123456789
 107
 108 # BDWELL-NEXT:                        0123456
 109 # BDWELL-NEXT:    Index     0123456789
 110
 111 # BTVER2-NEXT:                        0123456789
 112 # BTVER2-NEXT:    Index     0123456789          01234
 113
 114 # HASWELL-NEXT:                       012345678
 115 # HASWELL-NEXT:   Index     0123456789
 116
 117 # SANDY-NEXT:                         0123456789
 118 # SANDY-NEXT:     Index     0123456789
 119
 120 # SKYLAKE-NEXT:                       01234567
 121 # SKYLAKE-NEXT:   Index     0123456789
 122
 123 # ZNVER1-NEXT:                        0123456789
 124 # ZNVER1-NEXT:    Index     0123456789
 125
 126 # ZNVER2-NEXT:                        0123456789
 127 # ZNVER2-NEXT:    Index     0123456789          0
 128
 129 # ZNVER3-NEXT:                        0123456
 130 # ZNVER3-NEXT:    Index     0123456789
 131
 132 # BARCELONA:      [0,0]     DeeeeeeeeeeeeeeER  .   vdivps       %xmm0, %xmm1, %xmm1
 133 # BARCELONA-NEXT: [0,1]     D========eeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 134
 135 # BDVER2:         [0,0]     DeeeeeeeeeER   ..   vdivps  %xmm0, %xmm1, %xmm1
 136 # BDVER2-NEXT:    [0,1]     D====eeeeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 137
 138 # BDWELL:         [0,0]     DeeeeeeeeeeeER ..   vdivps  %xmm0, %xmm1, %xmm1
 139 # BDWELL-NEXT:    [0,1]     D======eeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 140
 141 # BTVER2:         [0,0]     DeeeeeeeeeeeeeeeeeeeER  .   vdivps  %xmm0, %xmm1, %xmm1
 142 # BTVER2-NEXT:    [0,1]     D==============eeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 143
 144 # HASWELL:        [0,0]     DeeeeeeeeeeeeeER  .   vdivps        %xmm0, %xmm1, %xmm1
 145 # HASWELL-NEXT:   [0,1]     D=======eeeeeeeeeER   vaddps        (%rax), %xmm1, %xmm1
 146
 147 # SANDY:          [0,0]     DeeeeeeeeeeeeeeER  .   vdivps       %xmm0, %xmm1, %xmm1
 148 # SANDY-NEXT:     [0,1]     D========eeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 149
 150 # SKYLAKE:        [0,0]     DeeeeeeeeeeeER . .   vdivps %xmm0, %xmm1, %xmm1
 151 # SKYLAKE-NEXT:   [0,1]     D=====eeeeeeeeeeER   vaddps (%rax), %xmm1, %xmm1
 152
 153 # ZNVER1:         [0,0]     DeeeeeeeeeeeeeeeER .   vdivps       %xmm0, %xmm1, %xmm1
 154 # ZNVER1-NEXT:    [0,1]     D=======eeeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 155
 156 # ZNVER2:         [0,0]     DeeeeeeeeeeeeeeeER  .   vdivps      %xmm0, %xmm1, %xmm1
 157 # ZNVER2-NEXT:    [0,1]     D========eeeeeeeeeeER   vaddps      (%rax), %xmm1, %xmm1
 158
 159 # ZNVER3:         [0,0]     DeeeeeeeeeeeER ..   vdivps  %xmm0, %xmm1, %xmm1
 160 # ZNVER3-NEXT:    [0,1]     D====eeeeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 161
 162 # ALL:            Average Wait times (based on the timeline view):
 163 # ALL-NEXT:       [0]: Executions
 164 # ALL-NEXT:       [1]: Average time spent waiting in a scheduler's queue
 165 # ALL-NEXT:       [2]: Average time spent waiting in a scheduler's queue while ready
 166 # ALL-NEXT:       [3]: Average time elapsed from WB until retire stage
 167
 168 # ALL:                  [0]    [1]    [2]    [3]
 169 # ALL-NEXT:       0.     1     1.0    1.0    0.0       vdivps   %xmm0, %xmm1, %xmm1
 170
 171 # BARCELONA-NEXT: 1.     1     9.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 172 # BARCELONA-NEXT:        1     5.0    0.5    0.0       <total>
 173
 174 # BDVER2-NEXT:    1.     1     5.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 175 # BDVER2-NEXT:           1     3.0    0.5    0.0       <total>
 176
 177 # BDWELL-NEXT:    1.     1     7.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 178 # BDWELL-NEXT:           1     4.0    0.5    0.0       <total>
 179
 180 # BTVER2-NEXT:    1.     1     15.0   0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 181 # BTVER2-NEXT:           1     8.0    0.5    0.0       <total>
 182
 183 # HASWELL-NEXT:   1.     1     8.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 184 # HASWELL-NEXT:          1     4.5    0.5    0.0       <total>
 185
 186 # SANDY-NEXT:     1.     1     9.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 187 # SANDY-NEXT:            1     5.0    0.5    0.0       <total>
 188
 189 # SKYLAKE-NEXT:   1.     1     6.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 190 # SKYLAKE-NEXT:          1     3.5    0.5    0.0       <total>
 191
 192 # ZNVER1-NEXT:    1.     1     8.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 193 # ZNVER1-NEXT:           1     4.5    0.5    0.0       <total>
 194
 195 # ZNVER2-NEXT:    1.     1     9.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 196 # ZNVER2-NEXT:           1     5.0    0.5    0.0       <total>
 197
 198 # ZNVER3-NEXT:    1.     1     5.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 199 # ZNVER3-NEXT:           1     3.0    0.5    0.0       <total>