test/tools/llvm-mca/X86/fma3-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=1 -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=1 -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=1 -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=1 -resource-pressure=false -instruction-info=false -timeline < %s | FileCheck %s -check-prefix=ALL -check-prefix=ZNVER1
   9
  10 vaddps %xmm0, %xmm0, %xmm2
  11 vfmadd213ps (%rdi), %xmm1, %xmm2
  12
  13 # ALL:          Iterations:        1
  14 # ALL-NEXT:     Instructions:      2
  15
  16 # BDWELL-NEXT:  Total Cycles:      13
  17 # BDWELL-NEXT:  Total uOps:        3
  18
  19 # HASWELL-NEXT: Total Cycles:      14
  20 # HASWELL-NEXT: Total uOps:        3
  21
  22 # SKYLAKE-NEXT: Total Cycles:      13
  23 # SKYLAKE-NEXT: Total uOps:        3
  24
  25 # ZNVER1-NEXT:  Total Cycles:      15
  26 # ZNVER1-NEXT:  Total uOps:        2
  27
  28 # BDWELL:       Dispatch Width:    4
  29 # BDWELL-NEXT:  uOps Per Cycle:    0.23
  30 # BDWELL-NEXT:  IPC:               0.15
  31 # BDWELL-NEXT:  Block RThroughput: 1.0
  32
  33 # HASWELL:      Dispatch Width:    4
  34 # HASWELL-NEXT: uOps Per Cycle:    0.21
  35 # HASWELL-NEXT: IPC:               0.14
  36 # HASWELL-NEXT: Block RThroughput: 1.0
  37
  38 # SKYLAKE:      Dispatch Width:    6
  39 # SKYLAKE-NEXT: uOps Per Cycle:    0.23
  40 # SKYLAKE-NEXT: IPC:               0.15
  41 # SKYLAKE-NEXT: Block RThroughput: 1.0
  42
  43 # ZNVER1:       Dispatch Width:    4
  44 # ZNVER1-NEXT:  uOps Per Cycle:    0.13
  45 # ZNVER1-NEXT:  IPC:               0.13
  46 # ZNVER1-NEXT:  Block RThroughput: 1.0
  47
  48 # ALL:          Timeline view:
  49
  50 # BDWELL-NEXT:                      012
  51 # HASWELL-NEXT:                     0123
  52 # SKYLAKE-NEXT:                     012
  53 # ZNVER1-NEXT:                      01234
  54
  55 # ALL-NEXT:     Index     0123456789
  56
  57 # BDWELL:       [0,0]     DeeeER    . .   vaddps        %xmm0, %xmm0, %xmm2
  58 # BDWELL-NEXT:  [0,1]     DeeeeeeeeeeER   vfmadd213ps   (%rdi), %xmm1, %xmm2
  59
  60 # HASWELL:      [0,0]     DeeeER    .  .   vaddps       %xmm0, %xmm0, %xmm2
  61 # HASWELL-NEXT: [0,1]     DeeeeeeeeeeeER   vfmadd213ps  (%rdi), %xmm1, %xmm2
  62
  63 # SKYLAKE:      [0,0]     DeeeeER   . .   vaddps        %xmm0, %xmm0, %xmm2
  64 # SKYLAKE-NEXT: [0,1]     DeeeeeeeeeeER   vfmadd213ps   (%rdi), %xmm1, %xmm2
  65
  66 # ZNVER1:       [0,0]     DeeeER    .   .   vaddps      %xmm0, %xmm0, %xmm2
  67 # ZNVER1-NEXT:  [0,1]     DeeeeeeeeeeeeER   vfmadd213ps (%rdi), %xmm1, %xmm2
  68
  69 # ALL:          Average Wait times (based on the timeline view):
  70 # ALL-NEXT:     [0]: Executions
  71 # ALL-NEXT:     [1]: Average time spent waiting in a scheduler's queue
  72 # ALL-NEXT:     [2]: Average time spent waiting in a scheduler's queue while ready
  73 # ALL-NEXT:     [3]: Average time elapsed from WB until retire stage
  74
  75 # ALL:                [0]    [1]    [2]    [3]
  76 # ALL-NEXT:     0.     1     1.0    1.0    0.0       vaddps     %xmm0, %xmm0, %xmm2
  77 # ALL-NEXT:     1.     1     1.0    0.0    0.0       vfmadd213ps        (%rdi), %xmm1, %xmm2