llvm/test/tools/llvm-mca/X86/variable-blend-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=sandybridge -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,SANDY
   3
   4 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=ivybridge -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,IVY
   5
   6 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown  -mcpu=haswell -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,HASWELL
   7
   8 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=broadwell -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,BDWELL
   9
  10 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,SKYLAKE
  11
  12 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=icelake-server -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,SKYLAKE
  13
  14 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,BDVER2
  15
  16 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,BTVER2
  17
  18 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver1 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,ZNVER1
  19
  20 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver2 -iterations=1 -timeline -instruction-info=false -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,ZNVER2
  21
  22 vaddps %xmm0, %xmm0, %xmm2
  23 vblendvps %xmm1, (%rdi), %xmm2, %xmm3
  24
  25 # ALL:          Iterations:        1
  26 # ALL-NEXT:     Instructions:      2
  27
  28 # BDVER2-NEXT:  Total Cycles:      10
  29 # BDVER2-NEXT:  Total uOps:        2
  30
  31 # BDWELL-NEXT:  Total Cycles:      10
  32 # BDWELL-NEXT:  Total uOps:        4
  33
  34 # BTVER2-NEXT:  Total Cycles:      11
  35 # BTVER2-NEXT:  Total uOps:        4
  36
  37 # HASWELL-NEXT: Total Cycles:      11
  38 # HASWELL-NEXT: Total uOps:        4
  39
  40 # IVY-NEXT:     Total Cycles:      11
  41 # IVY-NEXT:     Total uOps:        4
  42
  43 # SANDY-NEXT:   Total Cycles:      11
  44 # SANDY-NEXT:   Total uOps:        4
  45
  46 # SKYLAKE-NEXT: Total Cycles:      11
  47 # SKYLAKE-NEXT: Total uOps:        4
  48
  49 # ZNVER1-NEXT:  Total Cycles:      11
  50 # ZNVER1-NEXT:  Total uOps:        2
  51
  52 # ZNVER2-NEXT:  Total Cycles:      11
  53 # ZNVER2-NEXT:  Total uOps:        2
  54
  55 # BDVER2:       Dispatch Width:    4
  56 # BDVER2-NEXT:  uOps Per Cycle:    0.20
  57 # BDVER2-NEXT:  IPC:               0.20
  58 # BDVER2-NEXT:  Block RThroughput: 1.5
  59
  60 # BDWELL:       Dispatch Width:    4
  61 # BDWELL-NEXT:  uOps Per Cycle:    0.40
  62 # BDWELL-NEXT:  IPC:               0.20
  63 # BDWELL-NEXT:  Block RThroughput: 2.0
  64
  65 # BTVER2:       Dispatch Width:    2
  66 # BTVER2-NEXT:  uOps Per Cycle:    0.36
  67 # BTVER2-NEXT:  IPC:               0.18
  68 # BTVER2-NEXT:  Block RThroughput: 2.0
  69
  70 # HASWELL:      Dispatch Width:    4
  71 # HASWELL-NEXT: uOps Per Cycle:    0.36
  72 # HASWELL-NEXT: IPC:               0.18
  73 # HASWELL-NEXT: Block RThroughput: 2.0
  74
  75 # IVY:          Dispatch Width:    4
  76 # IVY-NEXT:     uOps Per Cycle:    0.36
  77 # IVY-NEXT:     IPC:               0.18
  78 # IVY-NEXT:     Block RThroughput: 1.0
  79
  80 # SANDY:        Dispatch Width:    4
  81 # SANDY-NEXT:   uOps Per Cycle:    0.36
  82 # SANDY-NEXT:   IPC:               0.18
  83 # SANDY-NEXT:   Block RThroughput: 1.0
  84
  85 # SKYLAKE:      Dispatch Width:    6
  86 # SKYLAKE-NEXT: uOps Per Cycle:    0.36
  87 # SKYLAKE-NEXT: IPC:               0.18
  88 # SKYLAKE-NEXT: Block RThroughput: 0.7
  89
  90 # ZNVER1:       Dispatch Width:    4
  91 # ZNVER1-NEXT:  uOps Per Cycle:    0.18
  92 # ZNVER1-NEXT:  IPC:               0.18
  93 # ZNVER1-NEXT:  Block RThroughput: 0.5
  94
  95 # ZNVER2:       Dispatch Width:    4
  96 # ZNVER2-NEXT:  uOps Per Cycle:    0.18
  97 # ZNVER2-NEXT:  IPC:               0.18
  98 # ZNVER2-NEXT:  Block RThroughput: 0.5
  99
 100 # BDVER2:       Timeline view:
 101 # BDVER2-NEXT:  Index     0123456789
 102
 103 # BDWELL:       Timeline view:
 104 # BDWELL-NEXT:  Index     0123456789
 105
 106 # BTVER2:       Timeline view:
 107 # BTVER2-NEXT:                      0
 108 # BTVER2-NEXT:  Index     0123456789
 109
 110 # HASWELL:      Timeline view:
 111 # HASWELL-NEXT:                     0
 112 # HASWELL-NEXT: Index     0123456789
 113
 114 # IVY:          Timeline view:
 115 # IVY-NEXT:                         0
 116 # IVY-NEXT:     Index     0123456789
 117
 118 # SANDY:        Timeline view:
 119 # SANDY-NEXT:                       0
 120 # SANDY-NEXT:   Index     0123456789
 121
 122 # SKYLAKE:      Timeline view:
 123 # SKYLAKE-NEXT:                     0
 124 # SKYLAKE-NEXT: Index     0123456789
 125
 126 # ZNVER1:       Timeline view:
 127 # ZNVER1-NEXT:                      0
 128 # ZNVER1-NEXT:  Index     0123456789
 129
 130 # ZNVER2:       Timeline view:
 131 # ZNVER2-NEXT:                      0
 132 # ZNVER2-NEXT:  Index     0123456789
 133
 134 # BDVER2:       [0,0]     DeeeeeER .   vaddps   %xmm0, %xmm0, %xmm2
 135 # BDVER2-NEXT:  [0,1]     DeeeeeeeER   vblendvps        %xmm1, (%rdi), %xmm2, %xmm3
 136
 137 # BDWELL:       [0,0]     DeeeER   .   vaddps   %xmm0, %xmm0, %xmm2
 138 # BDWELL-NEXT:  [0,1]     DeeeeeeeER   vblendvps        %xmm1, (%rdi), %xmm2, %xmm3
 139
 140 # BTVER2:       [0,0]     DeeeER    .   vaddps  %xmm0, %xmm0, %xmm2
 141 # BTVER2-NEXT:  [0,1]     .DeeeeeeeER   vblendvps       %xmm1, (%rdi), %xmm2, %xmm3
 142
 143 # HASWELL:      [0,0]     DeeeER    .   vaddps  %xmm0, %xmm0, %xmm2
 144 # HASWELL-NEXT: [0,1]     DeeeeeeeeER   vblendvps       %xmm1, (%rdi), %xmm2, %xmm3
 145
 146 # IVY:          [0,0]     DeeeER    .   vaddps  %xmm0, %xmm0, %xmm2
 147 # IVY-NEXT:     [0,1]     DeeeeeeeeER   vblendvps       %xmm1, (%rdi), %xmm2, %xmm3
 148
 149 # SANDY:        [0,0]     DeeeER    .   vaddps  %xmm0, %xmm0, %xmm2
 150 # SANDY-NEXT:   [0,1]     DeeeeeeeeER   vblendvps       %xmm1, (%rdi), %xmm2, %xmm3
 151
 152 # SKYLAKE:      [0,0]     DeeeeER   .   vaddps  %xmm0, %xmm0, %xmm2
 153 # SKYLAKE-NEXT: [0,1]     DeeeeeeeeER   vblendvps       %xmm1, (%rdi), %xmm2, %xmm3
 154
 155 # ZNVER1:       [0,0]     DeeeER    .   vaddps  %xmm0, %xmm0, %xmm2
 156 # ZNVER1-NEXT:  [0,1]     DeeeeeeeeER   vblendvps       %xmm1, (%rdi), %xmm2, %xmm3
 157
 158 # ZNVER2:       [0,0]     DeeeER    .   vaddps  %xmm0, %xmm0, %xmm2
 159 # ZNVER2-NEXT:  [0,1]     DeeeeeeeeER   vblendvps       %xmm1, (%rdi), %xmm2, %xmm3
 160
 161 # ALL:          Average Wait times (based on the timeline view):
 162 # ALL-NEXT:     [0]: Executions
 163 # ALL-NEXT:     [1]: Average time spent waiting in a scheduler's queue
 164 # ALL-NEXT:     [2]: Average time spent waiting in a scheduler's queue while ready
 165 # ALL-NEXT:     [3]: Average time elapsed from WB until retire stage
 166
 167 # ALL:                [0]    [1]    [2]    [3]
 168 # ALL-NEXT:     0.     1     1.0    1.0    0.0       vaddps     %xmm0, %xmm0, %xmm2
 169
 170 # BDVER2-NEXT:  1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 171 # BDVER2-NEXT:         1     1.0    0.5    0.0       <total>
 172
 173 # BDWELL-NEXT:  1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 174 # BDWELL-NEXT:         1     1.0    0.5    0.0       <total>
 175
 176 # BTVER2-NEXT:  1.     1     1.0    1.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 177 # BTVER2-NEXT:         1     1.0    1.0    0.0       <total>
 178
 179 # HASWELL-NEXT: 1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 180 # HASWELL-NEXT:        1     1.0    0.5    0.0       <total>
 181
 182 # IVY-NEXT:     1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 183 # IVY-NEXT:            1     1.0    0.5    0.0       <total>
 184
 185 # SANDY-NEXT:   1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 186 # SANDY-NEXT:          1     1.0    0.5    0.0       <total>
 187
 188 # SKYLAKE-NEXT: 1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 189 # SKYLAKE-NEXT:        1     1.0    0.5    0.0       <total>
 190
 191 # ZNVER1-NEXT:  1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 192 # ZNVER1-NEXT:         1     1.0    0.5    0.0       <total>
 193
 194 # ZNVER2-NEXT:  1.     1     1.0    0.0    0.0       vblendvps  %xmm1, (%rdi), %xmm2, %xmm3
 195 # ZNVER2-NEXT:         1     1.0    0.5    0.0       <total>