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