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 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver4 -iterations=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefixes=ALL,ZNVER4
  14
  15 vdivps  %xmm0, %xmm1, %xmm1
  16 vaddps  (%rax), %xmm1, %xmm1
  17
  18 # ALL:            Iterations:        1
  19 # ALL-NEXT:       Instructions:      2
  20
  21 # BARCELONA-NEXT: Total Cycles:      20
  22 # BARCELONA-NEXT: Total uOps:        3
  23
  24 # BDVER2-NEXT:    Total Cycles:      17
  25 # BDVER2-NEXT:    Total uOps:        2
  26
  27 # BDWELL-NEXT:    Total Cycles:      17
  28 # BDWELL-NEXT:    Total uOps:        3
  29
  30 # BTVER2-NEXT:    Total Cycles:      25
  31 # BTVER2-NEXT:    Total uOps:        2
  32
  33 # HASWELL-NEXT:   Total Cycles:      19
  34 # HASWELL-NEXT:   Total uOps:        3
  35
  36 # SANDY-NEXT:     Total Cycles:      20
  37 # SANDY-NEXT:     Total uOps:        3
  38
  39 # SKYLAKE-NEXT:   Total Cycles:      18
  40 # SKYLAKE-NEXT:   Total uOps:        3
  41
  42 # ZNVER1-NEXT:    Total Cycles:      15
  43 # ZNVER1-NEXT:    Total uOps:        2
  44
  45 # ZNVER2-NEXT:    Total Cycles:      16
  46 # ZNVER2-NEXT:    Total uOps:        2
  47
  48 # ZNVER3-NEXT:    Total Cycles:      17
  49 # ZNVER3-NEXT:    Total uOps:        2
  50
  51 # ZNVER4-NEXT:    Total Cycles:      17
  52 # ZNVER4-NEXT:    Total uOps:        2
  53
  54 # BARCELONA:      Dispatch Width:    4
  55 # BARCELONA-NEXT: uOps Per Cycle:    0.15
  56 # BARCELONA-NEXT: IPC:               0.10
  57 # BARCELONA-NEXT: Block RThroughput: 14.0
  58
  59 # BDVER2:         Dispatch Width:    4
  60 # BDVER2-NEXT:    uOps Per Cycle:    0.12
  61 # BDVER2-NEXT:    IPC:               0.12
  62 # BDVER2-NEXT:    Block RThroughput: 5.0
  63
  64 # BDWELL:         Dispatch Width:    4
  65 # BDWELL-NEXT:    uOps Per Cycle:    0.18
  66 # BDWELL-NEXT:    IPC:               0.12
  67 # BDWELL-NEXT:    Block RThroughput: 5.0
  68
  69 # BTVER2:         Dispatch Width:    2
  70 # BTVER2-NEXT:    uOps Per Cycle:    0.08
  71 # BTVER2-NEXT:    IPC:               0.08
  72 # BTVER2-NEXT:    Block RThroughput: 19.0
  73
  74 # HASWELL:        Dispatch Width:    4
  75 # HASWELL-NEXT:   uOps Per Cycle:    0.16
  76 # HASWELL-NEXT:   IPC:               0.11
  77 # HASWELL-NEXT:   Block RThroughput: 7.0
  78
  79 # SANDY:          Dispatch Width:    4
  80 # SANDY-NEXT:     uOps Per Cycle:    0.15
  81 # SANDY-NEXT:     IPC:               0.10
  82 # SANDY-NEXT:     Block RThroughput: 14.0
  83
  84 # SKYLAKE:        Dispatch Width:    6
  85 # SKYLAKE-NEXT:   uOps Per Cycle:    0.17
  86 # SKYLAKE-NEXT:   IPC:               0.11
  87 # SKYLAKE-NEXT:   Block RThroughput: 3.0
  88
  89 # ZNVER1:         Dispatch Width:    4
  90 # ZNVER1-NEXT:    uOps Per Cycle:    0.13
  91 # ZNVER1-NEXT:    IPC:               0.13
  92 # ZNVER1-NEXT:    Block RThroughput: 3.0
  93
  94 # ZNVER2:         Dispatch Width:    4
  95 # ZNVER2-NEXT:    uOps Per Cycle:    0.13
  96 # ZNVER2-NEXT:    IPC:               0.13
  97 # ZNVER2-NEXT:    Block RThroughput: 5.0
  98
  99 # ZNVER3:         Dispatch Width:    6
 100 # ZNVER3-NEXT:    uOps Per Cycle:    0.12
 101 # ZNVER3-NEXT:    IPC:               0.12
 102 # ZNVER3-NEXT:    Block RThroughput: 3.0
 103
 104 # ZNVER4:         Dispatch Width:    6
 105 # ZNVER4-NEXT:    uOps Per Cycle:    0.12
 106 # ZNVER4-NEXT:    IPC:               0.12
 107 # ZNVER4-NEXT:    Block RThroughput: 3.0
 108
 109 # ALL:            Timeline view:
 110
 111 # BARCELONA-NEXT:                     0123456789
 112 # BARCELONA-NEXT: Index     0123456789
 113
 114 # BDVER2-NEXT:                        0123456
 115 # BDVER2-NEXT:    Index     0123456789
 116
 117 # BDWELL-NEXT:                        0123456
 118 # BDWELL-NEXT:    Index     0123456789
 119
 120 # BTVER2-NEXT:                        0123456789
 121 # BTVER2-NEXT:    Index     0123456789          01234
 122
 123 # HASWELL-NEXT:                       012345678
 124 # HASWELL-NEXT:   Index     0123456789
 125
 126 # SANDY-NEXT:                         0123456789
 127 # SANDY-NEXT:     Index     0123456789
 128
 129 # SKYLAKE-NEXT:                       01234567
 130 # SKYLAKE-NEXT:   Index     0123456789
 131
 132 # ZNVER1-NEXT:                        01234
 133 # ZNVER1-NEXT:    Index     0123456789
 134
 135 # ZNVER2-NEXT:                        012345
 136 # ZNVER2-NEXT:    Index     0123456789
 137
 138 # ZNVER3-NEXT:                        0123456
 139 # ZNVER3-NEXT:    Index     0123456789
 140
 141 # ZNVER4-NEXT:                        0123456
 142 # ZNVER4-NEXT:    Index     0123456789
 143
 144 # BARCELONA:      [0,0]     DeeeeeeeeeeeeeeER  .   vdivps       %xmm0, %xmm1, %xmm1
 145 # BARCELONA-NEXT: [0,1]     D========eeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 146
 147 # BDVER2:         [0,0]     DeeeeeeeeeER   ..   vdivps  %xmm0, %xmm1, %xmm1
 148 # BDVER2-NEXT:    [0,1]     D====eeeeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 149
 150 # BDWELL:         [0,0]     DeeeeeeeeeeeER ..   vdivps  %xmm0, %xmm1, %xmm1
 151 # BDWELL-NEXT:    [0,1]     D======eeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 152
 153 # BTVER2:         [0,0]     DeeeeeeeeeeeeeeeeeeeER  .   vdivps  %xmm0, %xmm1, %xmm1
 154 # BTVER2-NEXT:    [0,1]     D==============eeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 155
 156 # HASWELL:        [0,0]     DeeeeeeeeeeeeeER  .   vdivps        %xmm0, %xmm1, %xmm1
 157 # HASWELL-NEXT:   [0,1]     D=======eeeeeeeeeER   vaddps        (%rax), %xmm1, %xmm1
 158
 159 # SANDY:          [0,0]     DeeeeeeeeeeeeeeER  .   vdivps       %xmm0, %xmm1, %xmm1
 160 # SANDY-NEXT:     [0,1]     D========eeeeeeeeeER   vaddps       (%rax), %xmm1, %xmm1
 161
 162 # SKYLAKE:        [0,0]     DeeeeeeeeeeeER . .   vdivps %xmm0, %xmm1, %xmm1
 163 # SKYLAKE-NEXT:   [0,1]     D=====eeeeeeeeeeER   vaddps (%rax), %xmm1, %xmm1
 164
 165 # ZNVER1:         [0,0]     DeeeeeeeeeeER .   vdivps    %xmm0, %xmm1, %xmm1
 166 # ZNVER1-NEXT:    [0,1]     D==eeeeeeeeeeER   vaddps    (%rax), %xmm1, %xmm1
 167
 168 # ZNVER2:         [0,0]     DeeeeeeeeeeER  .   vdivps   %xmm0, %xmm1, %xmm1
 169 # ZNVER2-NEXT:    [0,1]     D===eeeeeeeeeeER   vaddps   (%rax), %xmm1, %xmm1
 170
 171 # ZNVER3:         [0,0]     DeeeeeeeeeeeER ..   vdivps  %xmm0, %xmm1, %xmm1
 172 # ZNVER3-NEXT:    [0,1]     D====eeeeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 173
 174 # ZNVER4:         [0,0]     DeeeeeeeeeeeER ..   vdivps  %xmm0, %xmm1, %xmm1
 175 # ZNVER4-NEXT:    [0,1]     D====eeeeeeeeeeER   vaddps  (%rax), %xmm1, %xmm1
 176
 177 # ALL:            Average Wait times (based on the timeline view):
 178 # ALL-NEXT:       [0]: Executions
 179 # ALL-NEXT:       [1]: Average time spent waiting in a scheduler's queue
 180 # ALL-NEXT:       [2]: Average time spent waiting in a scheduler's queue while ready
 181 # ALL-NEXT:       [3]: Average time elapsed from WB until retire stage
 182
 183 # ALL:                  [0]    [1]    [2]    [3]
 184 # ALL-NEXT:       0.     1     1.0    1.0    0.0       vdivps   %xmm0, %xmm1, %xmm1
 185
 186 # BARCELONA-NEXT: 1.     1     9.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 187 # BARCELONA-NEXT:        1     5.0    0.5    0.0       <total>
 188
 189 # BDVER2-NEXT:    1.     1     5.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 190 # BDVER2-NEXT:           1     3.0    0.5    0.0       <total>
 191
 192 # BDWELL-NEXT:    1.     1     7.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 193 # BDWELL-NEXT:           1     4.0    0.5    0.0       <total>
 194
 195 # BTVER2-NEXT:    1.     1     15.0   0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 196 # BTVER2-NEXT:           1     8.0    0.5    0.0       <total>
 197
 198 # HASWELL-NEXT:   1.     1     8.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 199 # HASWELL-NEXT:          1     4.5    0.5    0.0       <total>
 200
 201 # SANDY-NEXT:     1.     1     9.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 202 # SANDY-NEXT:            1     5.0    0.5    0.0       <total>
 203
 204 # SKYLAKE-NEXT:   1.     1     6.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 205 # SKYLAKE-NEXT:          1     3.5    0.5    0.0       <total>
 206
 207 # ZNVER1-NEXT:    1.     1     3.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 208 # ZNVER1-NEXT:           1     2.0    0.5    0.0       <total>
 209
 210 # ZNVER2-NEXT:    1.     1     4.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 211 # ZNVER2-NEXT:           1     2.5    0.5    0.0       <total>
 212
 213 # ZNVER3-NEXT:    1.     1     5.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 214 # ZNVER3-NEXT:           1     3.0    0.5    0.0       <total>
 215
 216 # ZNVER4-NEXT:    1.     1     5.0    0.0    0.0       vaddps   (%rax), %xmm1, %xmm1
 217 # ZNVER4-NEXT:           1     3.0    0.5    0.0       <total>