llvm/test/tools/llvm-mca/X86/Znver3/mulx-hi-read-advance.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=znver3 -timeline -iterations=1 < %s | FileCheck %s
   3
   4 # PR51495: WriteIMulH reports an incorrect latency for the RM variants of MULX.
   5
   6 # LLVM-MCA-BEGIN
   7 mulxl (%rdi), %eax, %ecx
   8 add %eax, %eax
   9 # LLVM-MCA-END
  10
  11 # LLVM-MCA-BEGIN
  12 mulxq (%rdi), %rax, %rcx
  13 add %rax, %rax
  14 # LLVM-MCA-END
  15
  16 # CHECK:      [0] Code Region
  17
  18 # CHECK:      Iterations:        1
  19 # CHECK-NEXT: Instructions:      2
  20 # CHECK-NEXT: Total Cycles:      11
  21 # CHECK-NEXT: Total uOps:        3
  22
  23 # CHECK:      Dispatch Width:    6
  24 # CHECK-NEXT: uOps Per Cycle:    0.27
  25 # CHECK-NEXT: IPC:               0.18
  26 # CHECK-NEXT: Block RThroughput: 1.0
  27
  28 # CHECK:      Instruction Info:
  29 # CHECK-NEXT: [1]: #uOps
  30 # CHECK-NEXT: [2]: Latency
  31 # CHECK-NEXT: [3]: RThroughput
  32 # CHECK-NEXT: [4]: MayLoad
  33 # CHECK-NEXT: [5]: MayStore
  34 # CHECK-NEXT: [6]: HasSideEffects (U)
  35
  36 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  37 # CHECK-NEXT:  2      8     1.00    *                   mulxl   (%rdi), %eax, %ecx
  38 # CHECK-NEXT:  1      1     0.25                        addl    %eax, %eax
  39
  40 # CHECK:      Resources:
  41 # CHECK-NEXT: [0]   - Zn3AGU0
  42 # CHECK-NEXT: [1]   - Zn3AGU1
  43 # CHECK-NEXT: [2]   - Zn3AGU2
  44 # CHECK-NEXT: [3]   - Zn3ALU0
  45 # CHECK-NEXT: [4]   - Zn3ALU1
  46 # CHECK-NEXT: [5]   - Zn3ALU2
  47 # CHECK-NEXT: [6]   - Zn3ALU3
  48 # CHECK-NEXT: [7]   - Zn3BRU1
  49 # CHECK-NEXT: [8]   - Zn3FPP0
  50 # CHECK-NEXT: [9]   - Zn3FPP1
  51 # CHECK-NEXT: [10]  - Zn3FPP2
  52 # CHECK-NEXT: [11]  - Zn3FPP3
  53 # CHECK-NEXT: [12.0] - Zn3FPP45
  54 # CHECK-NEXT: [12.1] - Zn3FPP45
  55 # CHECK-NEXT: [13]  - Zn3FPSt
  56 # CHECK-NEXT: [14.0] - Zn3LSU
  57 # CHECK-NEXT: [14.1] - Zn3LSU
  58 # CHECK-NEXT: [14.2] - Zn3LSU
  59 # CHECK-NEXT: [15.0] - Zn3Load
  60 # CHECK-NEXT: [15.1] - Zn3Load
  61 # CHECK-NEXT: [15.2] - Zn3Load
  62 # CHECK-NEXT: [16.0] - Zn3Store
  63 # CHECK-NEXT: [16.1] - Zn3Store
  64
  65 # CHECK:      Resource pressure per iteration:
  66 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12.0] [12.1] [13]   [14.0] [14.1] [14.2] [15.0] [15.1] [15.2] [16.0] [16.1]
  67 # CHECK-NEXT:  -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -
  68
  69 # CHECK:      Resource pressure by instruction:
  70 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12.0] [12.1] [13]   [14.0] [14.1] [14.2] [15.0] [15.1] [15.2] [16.0] [16.1] Instructions:
  71 # CHECK-NEXT:  -      -     1.00    -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -     mulxl    (%rdi), %eax, %ecx
  72 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     addl     %eax, %eax
  73
  74 # CHECK:      Timeline view:
  75 # CHECK-NEXT:                     0
  76 # CHECK-NEXT: Index     0123456789
  77
  78 # CHECK:      [0,0]     DeeeeeeeeER   mulxl     (%rdi), %eax, %ecx
  79 # CHECK-NEXT: [0,1]     D=======eER   addl      %eax, %eax
  80
  81 # CHECK:      Average Wait times (based on the timeline view):
  82 # CHECK-NEXT: [0]: Executions
  83 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
  84 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
  85 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
  86
  87 # CHECK:            [0]    [1]    [2]    [3]
  88 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       mulxl        (%rdi), %eax, %ecx
  89 # CHECK-NEXT: 1.     1     8.0    0.0    0.0       addl %eax, %eax
  90 # CHECK-NEXT:        1     4.5    0.5    0.0       <total>
  91
  92 # CHECK:      [1] Code Region
  93
  94 # CHECK:      Iterations:        1
  95 # CHECK-NEXT: Instructions:      2
  96 # CHECK-NEXT: Total Cycles:      11
  97 # CHECK-NEXT: Total uOps:        3
  98
  99 # CHECK:      Dispatch Width:    6
 100 # CHECK-NEXT: uOps Per Cycle:    0.27
 101 # CHECK-NEXT: IPC:               0.18
 102 # CHECK-NEXT: Block RThroughput: 1.0
 103
 104 # CHECK:      Instruction Info:
 105 # CHECK-NEXT: [1]: #uOps
 106 # CHECK-NEXT: [2]: Latency
 107 # CHECK-NEXT: [3]: RThroughput
 108 # CHECK-NEXT: [4]: MayLoad
 109 # CHECK-NEXT: [5]: MayStore
 110 # CHECK-NEXT: [6]: HasSideEffects (U)
 111
 112 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
 113 # CHECK-NEXT:  2      8     1.00    *                   mulxq   (%rdi), %rax, %rcx
 114 # CHECK-NEXT:  1      1     0.25                        addq    %rax, %rax
 115
 116 # CHECK:      Resources:
 117 # CHECK-NEXT: [0]   - Zn3AGU0
 118 # CHECK-NEXT: [1]   - Zn3AGU1
 119 # CHECK-NEXT: [2]   - Zn3AGU2
 120 # CHECK-NEXT: [3]   - Zn3ALU0
 121 # CHECK-NEXT: [4]   - Zn3ALU1
 122 # CHECK-NEXT: [5]   - Zn3ALU2
 123 # CHECK-NEXT: [6]   - Zn3ALU3
 124 # CHECK-NEXT: [7]   - Zn3BRU1
 125 # CHECK-NEXT: [8]   - Zn3FPP0
 126 # CHECK-NEXT: [9]   - Zn3FPP1
 127 # CHECK-NEXT: [10]  - Zn3FPP2
 128 # CHECK-NEXT: [11]  - Zn3FPP3
 129 # CHECK-NEXT: [12.0] - Zn3FPP45
 130 # CHECK-NEXT: [12.1] - Zn3FPP45
 131 # CHECK-NEXT: [13]  - Zn3FPSt
 132 # CHECK-NEXT: [14.0] - Zn3LSU
 133 # CHECK-NEXT: [14.1] - Zn3LSU
 134 # CHECK-NEXT: [14.2] - Zn3LSU
 135 # CHECK-NEXT: [15.0] - Zn3Load
 136 # CHECK-NEXT: [15.1] - Zn3Load
 137 # CHECK-NEXT: [15.2] - Zn3Load
 138 # CHECK-NEXT: [16.0] - Zn3Store
 139 # CHECK-NEXT: [16.1] - Zn3Store
 140
 141 # CHECK:      Resource pressure per iteration:
 142 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12.0] [12.1] [13]   [14.0] [14.1] [14.2] [15.0] [15.1] [15.2] [16.0] [16.1]
 143 # CHECK-NEXT:  -      -     1.00    -     1.00    -     1.00    -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -
 144
 145 # CHECK:      Resource pressure by instruction:
 146 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    [10]   [11]   [12.0] [12.1] [13]   [14.0] [14.1] [14.2] [15.0] [15.1] [15.2] [16.0] [16.1] Instructions:
 147 # CHECK-NEXT:  -      -     1.00    -     1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     1.00    -      -     mulxq    (%rdi), %rax, %rcx
 148 # CHECK-NEXT:  -      -      -      -      -      -     1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     addq     %rax, %rax
 149
 150 # CHECK:      Timeline view:
 151 # CHECK-NEXT:                     0
 152 # CHECK-NEXT: Index     0123456789
 153
 154 # CHECK:      [0,0]     DeeeeeeeeER   mulxq     (%rdi), %rax, %rcx
 155 # CHECK-NEXT: [0,1]     D=======eER   addq      %rax, %rax
 156
 157 # CHECK:      Average Wait times (based on the timeline view):
 158 # CHECK-NEXT: [0]: Executions
 159 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
 160 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
 161 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
 162
 163 # CHECK:            [0]    [1]    [2]    [3]
 164 # CHECK-NEXT: 0.     1     1.0    1.0    0.0       mulxq        (%rdi), %rax, %rcx
 165 # CHECK-NEXT: 1.     1     8.0    0.0    0.0       addq %rax, %rax
 166 # CHECK-NEXT:        1     4.5    0.5    0.0       <total>