test/tools/llvm-mca/X86/BtVer2/pipes-fpu.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=btver2 -timeline -timeline-max-iterations=2 < %s | FileCheck %s
   3
   4 # VALU0/VALU1
   5 vpmulld     %xmm0, %xmm1, %xmm2
   6 vpand       %xmm0, %xmm1, %xmm2
   7
   8 # VIMUL/STC
   9 vcvttps2dq  %xmm0, %xmm2
  10 vpclmulqdq  $0, %xmm0, %xmm1, %xmm2
  11
  12 # FPA/FPM
  13 vaddps      %xmm0, %xmm1, %xmm2
  14 vsqrtps     %xmm0, %xmm2
  15
  16 # FPA/FPM YMM
  17 vaddps      %ymm0, %ymm1, %ymm2
  18 vsqrtps     %ymm0, %ymm2
  19
  20 # CHECK:      Iterations:        100
  21 # CHECK-NEXT: Instructions:      800
  22 # CHECK-NEXT: Total Cycles:      6306
  23 # CHECK-NEXT: Total uOps:        1200
  24
  25 # CHECK:      Dispatch Width:    2
  26 # CHECK-NEXT: uOps Per Cycle:    0.19
  27 # CHECK-NEXT: IPC:               0.13
  28 # CHECK-NEXT: Block RThroughput: 63.0
  29
  30 # CHECK:      Instruction Info:
  31 # CHECK-NEXT: [1]: #uOps
  32 # CHECK-NEXT: [2]: Latency
  33 # CHECK-NEXT: [3]: RThroughput
  34 # CHECK-NEXT: [4]: MayLoad
  35 # CHECK-NEXT: [5]: MayStore
  36 # CHECK-NEXT: [6]: HasSideEffects (U)
  37
  38 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  39 # CHECK-NEXT:  3      4     2.00                        vpmulld %xmm0, %xmm1, %xmm2
  40 # CHECK-NEXT:  1      1     0.50                        vpand   %xmm0, %xmm1, %xmm2
  41 # CHECK-NEXT:  1      3     1.00                        vcvttps2dq      %xmm0, %xmm2
  42 # CHECK-NEXT:  1      2     1.00                        vpclmulqdq      $0, %xmm0, %xmm1, %xmm2
  43 # CHECK-NEXT:  1      3     1.00                        vaddps  %xmm0, %xmm1, %xmm2
  44 # CHECK-NEXT:  1      21    21.00                       vsqrtps %xmm0, %xmm2
  45 # CHECK-NEXT:  2      3     2.00                        vaddps  %ymm0, %ymm1, %ymm2
  46 # CHECK-NEXT:  2      42    42.00                       vsqrtps %ymm0, %ymm2
  47
  48 # CHECK:      Resources:
  49 # CHECK-NEXT: [0]   - JALU0
  50 # CHECK-NEXT: [1]   - JALU1
  51 # CHECK-NEXT: [2]   - JDiv
  52 # CHECK-NEXT: [3]   - JFPA
  53 # CHECK-NEXT: [4]   - JFPM
  54 # CHECK-NEXT: [5]   - JFPU0
  55 # CHECK-NEXT: [6]   - JFPU1
  56 # CHECK-NEXT: [7]   - JLAGU
  57 # CHECK-NEXT: [8]   - JMul
  58 # CHECK-NEXT: [9]   - JSAGU
  59 # CHECK-NEXT: [10]  - JSTC
  60 # CHECK-NEXT: [11]  - JVALU0
  61 # CHECK-NEXT: [12]  - JVALU1
  62 # CHECK-NEXT: [13]  - JVIMUL
  63
  64 # CHECK:      Resource pressure per iteration:
  65 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12]   [13]
  66 # CHECK-NEXT:  -      -      -     3.00   63.00  6.01   5.99    -      -      -     1.00   1.00   1.00   3.00
  67
  68 # CHECK:      Resource pressure by instruction:
  69 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12]   [13]   Instructions:
  70 # CHECK-NEXT:  -      -      -      -      -     2.00   1.00    -      -      -      -     0.02   0.98   2.00   vpmulld %xmm0, %xmm1, %xmm2
  71 # CHECK-NEXT:  -      -      -      -      -     0.01   0.99    -      -      -      -     0.98   0.02    -     vpand   %xmm0, %xmm1, %xmm2
  72 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -     1.00    -      -      -     vcvttps2dq      %xmm0, %xmm2
  73 # CHECK-NEXT:  -      -      -      -      -     1.00    -      -      -      -      -      -      -     1.00   vpclmulqdq      $0, %xmm0, %xmm1, %xmm2
  74 # CHECK-NEXT:  -      -      -     1.00    -     1.00    -      -      -      -      -      -      -      -     vaddps  %xmm0, %xmm1, %xmm2
  75 # CHECK-NEXT:  -      -      -      -     21.00   -     1.00    -      -      -      -      -      -      -     vsqrtps %xmm0, %xmm2
  76 # CHECK-NEXT:  -      -      -     2.00    -     2.00    -      -      -      -      -      -      -      -     vaddps  %ymm0, %ymm1, %ymm2
  77 # CHECK-NEXT:  -      -      -      -     42.00   -     2.00    -      -      -      -      -      -      -     vsqrtps %ymm0, %ymm2
  78
  79 # CHECK:      Timeline view:
  80 # CHECK-NEXT:                     0123456789          0123456789          0123456789          0
  81 # CHECK-NEXT: Index     0123456789          0123456789          0123456789          0123456789
  82
  83 # CHECK:      [0,0]     DeeeeER   .    .    .    .    .    .    .    .    .    .    .    .    .   vpmulld       %xmm0, %xmm1, %xmm2
  84 # CHECK-NEXT: [0,1]     .DeE--R   .    .    .    .    .    .    .    .    .    .    .    .    .   vpand %xmm0, %xmm1, %xmm2
  85 # CHECK-NEXT: [0,2]     . DeeeER  .    .    .    .    .    .    .    .    .    .    .    .    .   vcvttps2dq    %xmm0, %xmm2
  86 # CHECK-NEXT: [0,3]     . DeeE-R  .    .    .    .    .    .    .    .    .    .    .    .    .   vpclmulqdq    $0, %xmm0, %xmm1, %xmm2
  87 # CHECK-NEXT: [0,4]     .  DeeeER .    .    .    .    .    .    .    .    .    .    .    .    .   vaddps        %xmm0, %xmm1, %xmm2
  88 # CHECK-NEXT: [0,5]     .  DeeeeeeeeeeeeeeeeeeeeeER   .    .    .    .    .    .    .    .    .   vsqrtps       %xmm0, %xmm2
  89 # CHECK-NEXT: [0,6]     .   DeeeE-----------------R   .    .    .    .    .    .    .    .    .   vaddps        %ymm0, %ymm1, %ymm2
  90 # CHECK-NEXT: [0,7]     .    D===================eeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeER .   vsqrtps       %ymm0, %ymm2
  91 # CHECK-NEXT: [1,0]     .    .DeeeeE--------------------------------------------------------R .   vpmulld       %xmm0, %xmm1, %xmm2
  92 # CHECK-NEXT: [1,1]     .    . DeE-----------------------------------------------------------R.   vpand %xmm0, %xmm1, %xmm2
  93 # CHECK-NEXT: [1,2]     .    .  DeeeE--------------------------------------------------------R.   vcvttps2dq    %xmm0, %xmm2
  94 # CHECK-NEXT: [1,3]     .    .  DeeE----------------------------------------------------------R   vpclmulqdq    $0, %xmm0, %xmm1, %xmm2
  95 # CHECK-NEXT: [1,4]     .    .   DeeeE--------------------------------------------------------R   vaddps        %xmm0, %xmm1, %xmm2
  96
  97 # CHECK:      Average Wait times (based on the timeline view):
  98 # CHECK-NEXT: [0]: Executions
  99 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
 100 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
 101 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 102
 103 # CHECK:            [0]    [1]    [2]    [3]
 104 # CHECK-NEXT: 0.     2     1.0    1.0    28.0      vpmulld      %xmm0, %xmm1, %xmm2
 105 # CHECK-NEXT: 1.     2     1.0    1.0    30.5      vpand        %xmm0, %xmm1, %xmm2
 106 # CHECK-NEXT: 2.     2     1.0    1.0    28.0      vcvttps2dq   %xmm0, %xmm2
 107 # CHECK-NEXT: 3.     2     1.0    1.0    29.5      vpclmulqdq   $0, %xmm0, %xmm1, %xmm2
 108 # CHECK-NEXT: 4.     2     1.0    1.0    28.0      vaddps       %xmm0, %xmm1, %xmm2
 109 # CHECK-NEXT: 5.     2     29.5   29.5   0.0       vsqrtps      %xmm0, %xmm2
 110 # CHECK-NEXT: 6.     2     1.0    1.0    45.5      vaddps       %ymm0, %ymm1, %ymm2
 111 # CHECK-NEXT: 7.     2     48.5   48.5   0.0       vsqrtps      %ymm0, %ymm2