test/tools/llvm-mca/X86/read-after-ld-2.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown  -mcpu=haswell -iterations=100 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=HASWELL
   3
   4 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=broadwell -iterations=100 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=BDWELL
   5
   6 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=100 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=SKYLAKE
   7
   8 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=100 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=ZNVER1
   9
  10 # Code Snippet from "Ithemal: Accurate, Portable and Fast Basic Block Throughput Estimation using Deep Neural Networks"
  11 # Charith Mendis, Saman Amarasinghe, Michael Carbin
  12 add     $1, %edx
  13 vpaddd (%r8), %ymm0, %ymm0
  14 add     $32, %r8
  15 cmp     %edi, %edx
  16
  17 # ALL:          Iterations:        100
  18 # ALL-NEXT:     Instructions:      400
  19
  20 # BDWELL-NEXT:  Total Cycles:      142
  21 # BDWELL-NEXT:  Total uOps:        500
  22
  23 # HASWELL-NEXT: Total Cycles:      143
  24 # HASWELL-NEXT: Total uOps:        500
  25
  26 # SKYLAKE-NEXT: Total Cycles:      110
  27 # SKYLAKE-NEXT: Total uOps:        500
  28
  29 # ZNVER1-NEXT:  Total Cycles:      110
  30 # ZNVER1-NEXT:  Total uOps:        400
  31
  32 # BDWELL:       Dispatch Width:    4
  33 # BDWELL-NEXT:  uOps Per Cycle:    3.52
  34 # BDWELL-NEXT:  IPC:               2.82
  35 # BDWELL-NEXT:  Block RThroughput: 1.3
  36
  37 # HASWELL:      Dispatch Width:    4
  38 # HASWELL-NEXT: uOps Per Cycle:    3.50
  39 # HASWELL-NEXT: IPC:               2.80
  40 # HASWELL-NEXT: Block RThroughput: 1.3
  41
  42 # SKYLAKE:      Dispatch Width:    6
  43 # SKYLAKE-NEXT: uOps Per Cycle:    4.55
  44 # SKYLAKE-NEXT: IPC:               3.64
  45 # SKYLAKE-NEXT: Block RThroughput: 0.8
  46
  47 # ZNVER1:       Dispatch Width:    4
  48 # ZNVER1-NEXT:  uOps Per Cycle:    3.64
  49 # ZNVER1-NEXT:  IPC:               3.64
  50 # ZNVER1-NEXT:  Block RThroughput: 1.0
  51
  52 # ALL:          Timeline view:
  53
  54 # BDWELL-NEXT:                      0123456789
  55 # BDWELL-NEXT:  Index     0123456789          01
  56
  57 # HASWELL-NEXT:                     0123456789
  58 # HASWELL-NEXT: Index     0123456789          012
  59
  60 # SKYLAKE-NEXT:                     0123456789
  61 # SKYLAKE-NEXT: Index     0123456789
  62
  63 # ZNVER1-NEXT:                      0123456789
  64 # ZNVER1-NEXT:  Index     0123456789
  65
  66 # BDWELL:       [0,0]     DeER .    .    .    ..   addl $1, %edx
  67 # BDWELL-NEXT:  [0,1]     DeeeeeeeER.    .    ..   vpaddd       (%r8), %ymm0, %ymm0
  68 # BDWELL-NEXT:  [0,2]     DeE------R.    .    ..   addq $32, %r8
  69 # BDWELL-NEXT:  [0,3]     .DeE-----R.    .    ..   cmpl %edi, %edx
  70 # BDWELL-NEXT:  [1,0]     .DeE-----R.    .    ..   addl $1, %edx
  71 # BDWELL-NEXT:  [1,1]     .DeeeeeeeER    .    ..   vpaddd       (%r8), %ymm0, %ymm0
  72 # BDWELL-NEXT:  [1,2]     . DeE-----R    .    ..   addq $32, %r8
  73 # BDWELL-NEXT:  [1,3]     . DeE-----R    .    ..   cmpl %edi, %edx
  74 # BDWELL-NEXT:  [2,0]     . DeE-----R    .    ..   addl $1, %edx
  75 # BDWELL-NEXT:  [2,1]     .  DeeeeeeeER  .    ..   vpaddd       (%r8), %ymm0, %ymm0
  76 # BDWELL-NEXT:  [2,2]     .  DeE------R  .    ..   addq $32, %r8
  77 # BDWELL-NEXT:  [2,3]     .  DeE------R  .    ..   cmpl %edi, %edx
  78 # BDWELL-NEXT:  [3,0]     .   DeE-----R  .    ..   addl $1, %edx
  79 # BDWELL-NEXT:  [3,1]     .   DeeeeeeeER .    ..   vpaddd       (%r8), %ymm0, %ymm0
  80 # BDWELL-NEXT:  [3,2]     .   DeE------R .    ..   addq $32, %r8
  81 # BDWELL-NEXT:  [3,3]     .    DeE-----R .    ..   cmpl %edi, %edx
  82 # BDWELL-NEXT:  [4,0]     .    DeE-----R .    ..   addl $1, %edx
  83 # BDWELL-NEXT:  [4,1]     .    DeeeeeeeER.    ..   vpaddd       (%r8), %ymm0, %ymm0
  84 # BDWELL-NEXT:  [4,2]     .    .DeE-----R.    ..   addq $32, %r8
  85 # BDWELL-NEXT:  [4,3]     .    .DeE-----R.    ..   cmpl %edi, %edx
  86 # BDWELL-NEXT:  [5,0]     .    .DeE-----R.    ..   addl $1, %edx
  87 # BDWELL-NEXT:  [5,1]     .    . DeeeeeeeER   ..   vpaddd       (%r8), %ymm0, %ymm0
  88 # BDWELL-NEXT:  [5,2]     .    . DeE------R   ..   addq $32, %r8
  89 # BDWELL-NEXT:  [5,3]     .    . DeE------R   ..   cmpl %edi, %edx
  90 # BDWELL-NEXT:  [6,0]     .    .  DeE-----R   ..   addl $1, %edx
  91 # BDWELL-NEXT:  [6,1]     .    .  DeeeeeeeER  ..   vpaddd       (%r8), %ymm0, %ymm0
  92 # BDWELL-NEXT:  [6,2]     .    .  DeE------R  ..   addq $32, %r8
  93 # BDWELL-NEXT:  [6,3]     .    .   DeE-----R  ..   cmpl %edi, %edx
  94 # BDWELL-NEXT:  [7,0]     .    .   DeE-----R  ..   addl $1, %edx
  95 # BDWELL-NEXT:  [7,1]     .    .   DeeeeeeeER ..   vpaddd       (%r8), %ymm0, %ymm0
  96 # BDWELL-NEXT:  [7,2]     .    .    DeE-----R ..   addq $32, %r8
  97 # BDWELL-NEXT:  [7,3]     .    .    DeE-----R ..   cmpl %edi, %edx
  98 # BDWELL-NEXT:  [8,0]     .    .    DeE-----R ..   addl $1, %edx
  99 # BDWELL-NEXT:  [8,1]     .    .    .DeeeeeeeER.   vpaddd       (%r8), %ymm0, %ymm0
 100 # BDWELL-NEXT:  [8,2]     .    .    .DeE------R.   addq $32, %r8
 101 # BDWELL-NEXT:  [8,3]     .    .    .DeE------R.   cmpl %edi, %edx
 102 # BDWELL-NEXT:  [9,0]     .    .    . DeE-----R.   addl $1, %edx
 103 # BDWELL-NEXT:  [9,1]     .    .    . DeeeeeeeER   vpaddd       (%r8), %ymm0, %ymm0
 104 # BDWELL-NEXT:  [9,2]     .    .    . DeE------R   addq $32, %r8
 105 # BDWELL-NEXT:  [9,3]     .    .    .  DeE-----R   cmpl %edi, %edx
 106
 107 # HASWELL:      [0,0]     DeER .    .    .    . .   addl        $1, %edx
 108 # HASWELL-NEXT: [0,1]     DeeeeeeeeER    .    . .   vpaddd      (%r8), %ymm0, %ymm0
 109 # HASWELL-NEXT: [0,2]     DeE-------R    .    . .   addq        $32, %r8
 110 # HASWELL-NEXT: [0,3]     .DeE------R    .    . .   cmpl        %edi, %edx
 111 # HASWELL-NEXT: [1,0]     .DeE------R    .    . .   addl        $1, %edx
 112 # HASWELL-NEXT: [1,1]     .DeeeeeeeeER   .    . .   vpaddd      (%r8), %ymm0, %ymm0
 113 # HASWELL-NEXT: [1,2]     . DeE------R   .    . .   addq        $32, %r8
 114 # HASWELL-NEXT: [1,3]     . DeE------R   .    . .   cmpl        %edi, %edx
 115 # HASWELL-NEXT: [2,0]     . DeE------R   .    . .   addl        $1, %edx
 116 # HASWELL-NEXT: [2,1]     .  DeeeeeeeeER .    . .   vpaddd      (%r8), %ymm0, %ymm0
 117 # HASWELL-NEXT: [2,2]     .  DeE-------R .    . .   addq        $32, %r8
 118 # HASWELL-NEXT: [2,3]     .  DeE-------R .    . .   cmpl        %edi, %edx
 119 # HASWELL-NEXT: [3,0]     .   DeE------R .    . .   addl        $1, %edx
 120 # HASWELL-NEXT: [3,1]     .   DeeeeeeeeER.    . .   vpaddd      (%r8), %ymm0, %ymm0
 121 # HASWELL-NEXT: [3,2]     .   DeE-------R.    . .   addq        $32, %r8
 122 # HASWELL-NEXT: [3,3]     .    DeE------R.    . .   cmpl        %edi, %edx
 123 # HASWELL-NEXT: [4,0]     .    DeE------R.    . .   addl        $1, %edx
 124 # HASWELL-NEXT: [4,1]     .    DeeeeeeeeER    . .   vpaddd      (%r8), %ymm0, %ymm0
 125 # HASWELL-NEXT: [4,2]     .    .DeE------R    . .   addq        $32, %r8
 126 # HASWELL-NEXT: [4,3]     .    .DeE------R    . .   cmpl        %edi, %edx
 127 # HASWELL-NEXT: [5,0]     .    .DeE------R    . .   addl        $1, %edx
 128 # HASWELL-NEXT: [5,1]     .    . DeeeeeeeeER  . .   vpaddd      (%r8), %ymm0, %ymm0
 129 # HASWELL-NEXT: [5,2]     .    . DeE-------R  . .   addq        $32, %r8
 130 # HASWELL-NEXT: [5,3]     .    . DeE-------R  . .   cmpl        %edi, %edx
 131 # HASWELL-NEXT: [6,0]     .    .  DeE------R  . .   addl        $1, %edx
 132 # HASWELL-NEXT: [6,1]     .    .  DeeeeeeeeER . .   vpaddd      (%r8), %ymm0, %ymm0
 133 # HASWELL-NEXT: [6,2]     .    .  DeE-------R . .   addq        $32, %r8
 134 # HASWELL-NEXT: [6,3]     .    .   DeE------R . .   cmpl        %edi, %edx
 135 # HASWELL-NEXT: [7,0]     .    .   DeE------R . .   addl        $1, %edx
 136 # HASWELL-NEXT: [7,1]     .    .   DeeeeeeeeER. .   vpaddd      (%r8), %ymm0, %ymm0
 137 # HASWELL-NEXT: [7,2]     .    .    DeE------R. .   addq        $32, %r8
 138 # HASWELL-NEXT: [7,3]     .    .    DeE------R. .   cmpl        %edi, %edx
 139 # HASWELL-NEXT: [8,0]     .    .    DeE------R. .   addl        $1, %edx
 140 # HASWELL-NEXT: [8,1]     .    .    .DeeeeeeeeER.   vpaddd      (%r8), %ymm0, %ymm0
 141 # HASWELL-NEXT: [8,2]     .    .    .DeE-------R.   addq        $32, %r8
 142 # HASWELL-NEXT: [8,3]     .    .    .DeE-------R.   cmpl        %edi, %edx
 143 # HASWELL-NEXT: [9,0]     .    .    . DeE------R.   addl        $1, %edx
 144 # HASWELL-NEXT: [9,1]     .    .    . DeeeeeeeeER   vpaddd      (%r8), %ymm0, %ymm0
 145 # HASWELL-NEXT: [9,2]     .    .    . DeE-------R   addq        $32, %r8
 146 # HASWELL-NEXT: [9,3]     .    .    .  DeE------R   cmpl        %edi, %edx
 147
 148 # SKYLAKE:      [0,0]     DeER .    .    .   .   addl   $1, %edx
 149 # SKYLAKE-NEXT: [0,1]     DeeeeeeeeER    .   .   vpaddd (%r8), %ymm0, %ymm0
 150 # SKYLAKE-NEXT: [0,2]     DeE-------R    .   .   addq   $32, %r8
 151 # SKYLAKE-NEXT: [0,3]     D=eE------R    .   .   cmpl   %edi, %edx
 152 # SKYLAKE-NEXT: [1,0]     D=eE------R    .   .   addl   $1, %edx
 153 # SKYLAKE-NEXT: [1,1]     .DeeeeeeeeER   .   .   vpaddd (%r8), %ymm0, %ymm0
 154 # SKYLAKE-NEXT: [1,2]     .DeE-------R   .   .   addq   $32, %r8
 155 # SKYLAKE-NEXT: [1,3]     .D=eE------R   .   .   cmpl   %edi, %edx
 156 # SKYLAKE-NEXT: [2,0]     .D=eE------R   .   .   addl   $1, %edx
 157 # SKYLAKE-NEXT: [2,1]     . DeeeeeeeeER  .   .   vpaddd (%r8), %ymm0, %ymm0
 158 # SKYLAKE-NEXT: [2,2]     . DeE-------R  .   .   addq   $32, %r8
 159 # SKYLAKE-NEXT: [2,3]     . D=eE------R  .   .   cmpl   %edi, %edx
 160 # SKYLAKE-NEXT: [3,0]     . D=eE------R  .   .   addl   $1, %edx
 161 # SKYLAKE-NEXT: [3,1]     .  DeeeeeeeeER .   .   vpaddd (%r8), %ymm0, %ymm0
 162 # SKYLAKE-NEXT: [3,2]     .  DeE-------R .   .   addq   $32, %r8
 163 # SKYLAKE-NEXT: [3,3]     .  D=eE------R .   .   cmpl   %edi, %edx
 164 # SKYLAKE-NEXT: [4,0]     .  D=eE------R .   .   addl   $1, %edx
 165 # SKYLAKE-NEXT: [4,1]     .   DeeeeeeeeER.   .   vpaddd (%r8), %ymm0, %ymm0
 166 # SKYLAKE-NEXT: [4,2]     .   DeE-------R.   .   addq   $32, %r8
 167 # SKYLAKE-NEXT: [4,3]     .   D=eE------R.   .   cmpl   %edi, %edx
 168 # SKYLAKE-NEXT: [5,0]     .   D=eE------R.   .   addl   $1, %edx
 169 # SKYLAKE-NEXT: [5,1]     .    DeeeeeeeeER   .   vpaddd (%r8), %ymm0, %ymm0
 170 # SKYLAKE-NEXT: [5,2]     .    DeE-------R   .   addq   $32, %r8
 171 # SKYLAKE-NEXT: [5,3]     .    D=eE------R   .   cmpl   %edi, %edx
 172 # SKYLAKE-NEXT: [6,0]     .    D=eE------R   .   addl   $1, %edx
 173 # SKYLAKE-NEXT: [6,1]     .    .DeeeeeeeeER  .   vpaddd (%r8), %ymm0, %ymm0
 174 # SKYLAKE-NEXT: [6,2]     .    .DeE-------R  .   addq   $32, %r8
 175 # SKYLAKE-NEXT: [6,3]     .    .D=eE------R  .   cmpl   %edi, %edx
 176 # SKYLAKE-NEXT: [7,0]     .    .D=eE------R  .   addl   $1, %edx
 177 # SKYLAKE-NEXT: [7,1]     .    . DeeeeeeeeER .   vpaddd (%r8), %ymm0, %ymm0
 178 # SKYLAKE-NEXT: [7,2]     .    . DeE-------R .   addq   $32, %r8
 179 # SKYLAKE-NEXT: [7,3]     .    . D=eE------R .   cmpl   %edi, %edx
 180 # SKYLAKE-NEXT: [8,0]     .    . D=eE------R .   addl   $1, %edx
 181 # SKYLAKE-NEXT: [8,1]     .    .  DeeeeeeeeER.   vpaddd (%r8), %ymm0, %ymm0
 182 # SKYLAKE-NEXT: [8,2]     .    .  DeE-------R.   addq   $32, %r8
 183 # SKYLAKE-NEXT: [8,3]     .    .  D=eE------R.   cmpl   %edi, %edx
 184 # SKYLAKE-NEXT: [9,0]     .    .  D=eE------R.   addl   $1, %edx
 185 # SKYLAKE-NEXT: [9,1]     .    .   DeeeeeeeeER   vpaddd (%r8), %ymm0, %ymm0
 186 # SKYLAKE-NEXT: [9,2]     .    .   DeE-------R   addq   $32, %r8
 187 # SKYLAKE-NEXT: [9,3]     .    .   D=eE------R   cmpl   %edi, %edx
 188
 189 # ZNVER1:       [0,0]     DeER .    .    .   .   addl   $1, %edx
 190 # ZNVER1-NEXT:  [0,1]     DeeeeeeeeER    .   .   vpaddd (%r8), %ymm0, %ymm0
 191 # ZNVER1-NEXT:  [0,2]     DeE-------R    .   .   addq   $32, %r8
 192 # ZNVER1-NEXT:  [0,3]     D=eE------R    .   .   cmpl   %edi, %edx
 193 # ZNVER1-NEXT:  [1,0]     .DeE------R    .   .   addl   $1, %edx
 194 # ZNVER1-NEXT:  [1,1]     .DeeeeeeeeER   .   .   vpaddd (%r8), %ymm0, %ymm0
 195 # ZNVER1-NEXT:  [1,2]     .DeE-------R   .   .   addq   $32, %r8
 196 # ZNVER1-NEXT:  [1,3]     .D=eE------R   .   .   cmpl   %edi, %edx
 197 # ZNVER1-NEXT:  [2,0]     . DeE------R   .   .   addl   $1, %edx
 198 # ZNVER1-NEXT:  [2,1]     . DeeeeeeeeER  .   .   vpaddd (%r8), %ymm0, %ymm0
 199 # ZNVER1-NEXT:  [2,2]     . DeE-------R  .   .   addq   $32, %r8
 200 # ZNVER1-NEXT:  [2,3]     . D=eE------R  .   .   cmpl   %edi, %edx
 201 # ZNVER1-NEXT:  [3,0]     .  DeE------R  .   .   addl   $1, %edx
 202 # ZNVER1-NEXT:  [3,1]     .  DeeeeeeeeER .   .   vpaddd (%r8), %ymm0, %ymm0
 203 # ZNVER1-NEXT:  [3,2]     .  DeE-------R .   .   addq   $32, %r8
 204 # ZNVER1-NEXT:  [3,3]     .  D=eE------R .   .   cmpl   %edi, %edx
 205 # ZNVER1-NEXT:  [4,0]     .   DeE------R .   .   addl   $1, %edx
 206 # ZNVER1-NEXT:  [4,1]     .   DeeeeeeeeER.   .   vpaddd (%r8), %ymm0, %ymm0
 207 # ZNVER1-NEXT:  [4,2]     .   DeE-------R.   .   addq   $32, %r8
 208 # ZNVER1-NEXT:  [4,3]     .   D=eE------R.   .   cmpl   %edi, %edx
 209 # ZNVER1-NEXT:  [5,0]     .    DeE------R.   .   addl   $1, %edx
 210 # ZNVER1-NEXT:  [5,1]     .    DeeeeeeeeER   .   vpaddd (%r8), %ymm0, %ymm0
 211 # ZNVER1-NEXT:  [5,2]     .    DeE-------R   .   addq   $32, %r8
 212 # ZNVER1-NEXT:  [5,3]     .    D=eE------R   .   cmpl   %edi, %edx
 213 # ZNVER1-NEXT:  [6,0]     .    .DeE------R   .   addl   $1, %edx
 214 # ZNVER1-NEXT:  [6,1]     .    .DeeeeeeeeER  .   vpaddd (%r8), %ymm0, %ymm0
 215 # ZNVER1-NEXT:  [6,2]     .    .DeE-------R  .   addq   $32, %r8
 216 # ZNVER1-NEXT:  [6,3]     .    .D=eE------R  .   cmpl   %edi, %edx
 217 # ZNVER1-NEXT:  [7,0]     .    . DeE------R  .   addl   $1, %edx
 218 # ZNVER1-NEXT:  [7,1]     .    . DeeeeeeeeER .   vpaddd (%r8), %ymm0, %ymm0
 219 # ZNVER1-NEXT:  [7,2]     .    . DeE-------R .   addq   $32, %r8
 220 # ZNVER1-NEXT:  [7,3]     .    . D=eE------R .   cmpl   %edi, %edx
 221 # ZNVER1-NEXT:  [8,0]     .    .  DeE------R .   addl   $1, %edx
 222 # ZNVER1-NEXT:  [8,1]     .    .  DeeeeeeeeER.   vpaddd (%r8), %ymm0, %ymm0
 223 # ZNVER1-NEXT:  [8,2]     .    .  DeE-------R.   addq   $32, %r8
 224 # ZNVER1-NEXT:  [8,3]     .    .  D=eE------R.   cmpl   %edi, %edx
 225 # ZNVER1-NEXT:  [9,0]     .    .   DeE------R.   addl   $1, %edx
 226 # ZNVER1-NEXT:  [9,1]     .    .   DeeeeeeeeER   vpaddd (%r8), %ymm0, %ymm0
 227 # ZNVER1-NEXT:  [9,2]     .    .   DeE-------R   addq   $32, %r8
 228 # ZNVER1-NEXT:  [9,3]     .    .   D=eE------R   cmpl   %edi, %edx
 229
 230 # ALL:          Average Wait times (based on the timeline view):
 231 # ALL-NEXT:     [0]: Executions
 232 # ALL-NEXT:     [1]: Average time spent waiting in a scheduler's queue
 233 # ALL-NEXT:     [2]: Average time spent waiting in a scheduler's queue while ready
 234 # ALL-NEXT:     [3]: Average time elapsed from WB until retire stage
 235
 236 # ALL:                [0]    [1]    [2]    [3]
 237
 238 # BDWELL-NEXT:  0.     10    1.0    0.4    4.5       addl       $1, %edx
 239 # HASWELL-NEXT: 0.     10    1.0    0.4    5.4       addl       $1, %edx
 240 # SKYLAKE-NEXT: 0.     10    1.9    0.1    5.4       addl       $1, %edx
 241 # ZNVER1-NEXT:  0.     10    1.0    0.1    5.4       addl       $1, %edx
 242
 243 # ALL-NEXT:     1.     10    1.0    0.1    0.0       vpaddd     (%r8), %ymm0, %ymm0
 244
 245 # BDWELL-NEXT:  2.     10    1.0    0.4    5.7       addq       $32, %r8
 246 # BDWELL-NEXT:  3.     10    1.0    0.0    5.3       cmpl       %edi, %edx
 247
 248 # HASWELL-NEXT: 2.     10    1.0    0.4    6.7       addq       $32, %r8
 249 # HASWELL-NEXT: 3.     10    1.0    0.0    6.3       cmpl       %edi, %edx
 250
 251 # SKYLAKE-NEXT: 2.     10    1.0    0.1    7.0       addq       $32, %r8
 252 # SKYLAKE-NEXT: 3.     10    2.0    0.0    6.0       cmpl       %edi, %edx
 253
 254 # ZNVER1-NEXT:  2.     10    1.0    0.1    7.0       addq       $32, %r8
 255 # ZNVER1-NEXT:  3.     10    2.0    0.0    6.0       cmpl       %edi, %edx