test/tools/llvm-mca/X86/BtVer2/bottleneck-hints-3.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 -iterations=1500 -noalias=false -timeline -timeline-max-iterations=1 -bottleneck-analysis < %s | FileCheck %s
   3
   4 vmovaps (%rsi), %xmm0
   5 vmovaps %xmm0, (%rdi)
   6 vmovaps 16(%rsi), %xmm0
   7 vmovaps %xmm0, 16(%rdi)
   8 vmovaps 32(%rsi), %xmm0
   9 vmovaps %xmm0, 32(%rdi)
  10 vmovaps 48(%rsi), %xmm0
  11 vmovaps %xmm0, 48(%rdi)
  12
  13 # CHECK:      Iterations:        1500
  14 # CHECK-NEXT: Instructions:      12000
  15 # CHECK-NEXT: Total Cycles:      36003
  16 # CHECK-NEXT: Total uOps:        12000
  17
  18 # CHECK:      Dispatch Width:    2
  19 # CHECK-NEXT: uOps Per Cycle:    0.33
  20 # CHECK-NEXT: IPC:               0.33
  21 # CHECK-NEXT: Block RThroughput: 4.0
  22
  23 # CHECK:      Cycles with backend pressure increase [ 99.89% ]
  24 # CHECK-NEXT: Throughput Bottlenecks:
  25 # CHECK-NEXT:   Resource Pressure       [ 0.00% ]
  26 # CHECK-NEXT:   Data Dependencies:      [ 99.89% ]
  27 # CHECK-NEXT:   - Register Dependencies [ 83.24% ]
  28 # CHECK-NEXT:   - Memory Dependencies   [ 99.89% ]
  29
  30 # CHECK:      Critical sequence based on the simulation:
  31
  32 # CHECK:                    Instruction                                 Dependency Information
  33 # CHECK-NEXT:  +----< 7.    vmovaps     %xmm0, 48(%rdi)
  34 # CHECK-NEXT:  |
  35 # CHECK-NEXT:  |    < loop carried >
  36 # CHECK-NEXT:  |
  37 # CHECK-NEXT:  +----> 0.    vmovaps     (%rsi), %xmm0                     ## MEMORY dependency.
  38 # CHECK-NEXT:  +----> 1.    vmovaps     %xmm0, (%rdi)                     ## REGISTER dependency:  %xmm0
  39 # CHECK-NEXT:  +----> 2.    vmovaps     16(%rsi), %xmm0                   ## MEMORY dependency.
  40 # CHECK-NEXT:  +----> 3.    vmovaps     %xmm0, 16(%rdi)                   ## REGISTER dependency:  %xmm0
  41 # CHECK-NEXT:  +----> 4.    vmovaps     32(%rsi), %xmm0                   ## MEMORY dependency.
  42 # CHECK-NEXT:  +----> 5.    vmovaps     %xmm0, 32(%rdi)                   ## REGISTER dependency:  %xmm0
  43 # CHECK-NEXT:  +----> 6.    vmovaps     48(%rsi), %xmm0                   ## MEMORY dependency.
  44 # CHECK-NEXT:  +----> 7.    vmovaps     %xmm0, 48(%rdi)                   ## REGISTER dependency:  %xmm0
  45 # CHECK-NEXT:  |
  46 # CHECK-NEXT:  |    < loop carried >
  47 # CHECK-NEXT:  |
  48 # CHECK-NEXT:  +----> 0.    vmovaps     (%rsi), %xmm0                     ## MEMORY dependency.
  49
  50 # CHECK:      Instruction Info:
  51 # CHECK-NEXT: [1]: #uOps
  52 # CHECK-NEXT: [2]: Latency
  53 # CHECK-NEXT: [3]: RThroughput
  54 # CHECK-NEXT: [4]: MayLoad
  55 # CHECK-NEXT: [5]: MayStore
  56 # CHECK-NEXT: [6]: HasSideEffects (U)
  57
  58 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  59 # CHECK-NEXT:  1      5     1.00    *                   vmovaps (%rsi), %xmm0
  60 # CHECK-NEXT:  1      1     1.00           *            vmovaps %xmm0, (%rdi)
  61 # CHECK-NEXT:  1      5     1.00    *                   vmovaps 16(%rsi), %xmm0
  62 # CHECK-NEXT:  1      1     1.00           *            vmovaps %xmm0, 16(%rdi)
  63 # CHECK-NEXT:  1      5     1.00    *                   vmovaps 32(%rsi), %xmm0
  64 # CHECK-NEXT:  1      1     1.00           *            vmovaps %xmm0, 32(%rdi)
  65 # CHECK-NEXT:  1      5     1.00    *                   vmovaps 48(%rsi), %xmm0
  66 # CHECK-NEXT:  1      1     1.00           *            vmovaps %xmm0, 48(%rdi)
  67
  68 # CHECK:      Resources:
  69 # CHECK-NEXT: [0]   - JALU0
  70 # CHECK-NEXT: [1]   - JALU1
  71 # CHECK-NEXT: [2]   - JDiv
  72 # CHECK-NEXT: [3]   - JFPA
  73 # CHECK-NEXT: [4]   - JFPM
  74 # CHECK-NEXT: [5]   - JFPU0
  75 # CHECK-NEXT: [6]   - JFPU1
  76 # CHECK-NEXT: [7]   - JLAGU
  77 # CHECK-NEXT: [8]   - JMul
  78 # CHECK-NEXT: [9]   - JSAGU
  79 # CHECK-NEXT: [10]  - JSTC
  80 # CHECK-NEXT: [11]  - JVALU0
  81 # CHECK-NEXT: [12]  - JVALU1
  82 # CHECK-NEXT: [13]  - JVIMUL
  83
  84 # CHECK:      Resource pressure per iteration:
  85 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12]   [13]
  86 # CHECK-NEXT:  -      -      -      -      -      -     4.00   4.00    -     4.00   4.00    -      -      -
  87
  88 # CHECK:      Resource pressure by instruction:
  89 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12]   [13]   Instructions:
  90 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -      -      -      -      -     vmovaps (%rsi), %xmm0
  91 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -     vmovaps %xmm0, (%rdi)
  92 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -      -      -      -      -     vmovaps 16(%rsi), %xmm0
  93 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -     vmovaps %xmm0, 16(%rdi)
  94 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -      -      -      -      -     vmovaps 32(%rsi), %xmm0
  95 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -     vmovaps %xmm0, 32(%rdi)
  96 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -      -      -      -      -     vmovaps 48(%rsi), %xmm0
  97 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -     1.00   1.00    -      -      -     vmovaps %xmm0, 48(%rdi)
  98
  99 # CHECK:      Timeline view:
 100 # CHECK-NEXT:                     0123456789
 101 # CHECK-NEXT: Index     0123456789          0123456
 102
 103 # CHECK:      [0,0]     DeeeeeER  .    .    .    ..   vmovaps   (%rsi), %xmm0
 104 # CHECK-NEXT: [0,1]     D=====eER .    .    .    ..   vmovaps   %xmm0, (%rdi)
 105 # CHECK-NEXT: [0,2]     .D=====eeeeeER .    .    ..   vmovaps   16(%rsi), %xmm0
 106 # CHECK-NEXT: [0,3]     .D==========eER.    .    ..   vmovaps   %xmm0, 16(%rdi)
 107 # CHECK-NEXT: [0,4]     . D==========eeeeeER.    ..   vmovaps   32(%rsi), %xmm0
 108 # CHECK-NEXT: [0,5]     . D===============eER    ..   vmovaps   %xmm0, 32(%rdi)
 109 # CHECK-NEXT: [0,6]     .  D===============eeeeeER.   vmovaps   48(%rsi), %xmm0
 110 # CHECK-NEXT: [0,7]     .  D====================eER   vmovaps   %xmm0, 48(%rdi)
 111
 112 # CHECK:      Average Wait times (based on the timeline view):
 113 # CHECK-NEXT: [0]: Executions
 114 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
 115 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
 116 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 117
 118 # CHECK:            [0]    [1]    [2]    [3]
 119 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       vmovaps      (%rsi), %xmm0
 120 # CHECK-NEXT: 1.     1     6.0    0.0    0.0       vmovaps      %xmm0, (%rdi)
 121 # CHECK-NEXT: 2.     1     6.0    0.0    0.0       vmovaps      16(%rsi), %xmm0
 122 # CHECK-NEXT: 3.     1     11.0   0.0    0.0       vmovaps      %xmm0, 16(%rdi)
 123 # CHECK-NEXT: 4.     1     11.0   0.0    0.0       vmovaps      32(%rsi), %xmm0
 124 # CHECK-NEXT: 5.     1     16.0   0.0    0.0       vmovaps      %xmm0, 32(%rdi)
 125 # CHECK-NEXT: 6.     1     16.0   0.0    0.0       vmovaps      48(%rsi), %xmm0
 126 # CHECK-NEXT: 7.     1     21.0   0.0    0.0       vmovaps      %xmm0, 48(%rdi)
 127 # CHECK-NEXT:        1     11.0   0.1    0.0       <total>