test/tools/llvm-mca/X86/BdVer2/vbroadcast-operand-latency.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
   3
   4 leaq 8(%rsp, %rdi, 2), %rax
   5 vbroadcastss (%rax), %ymm0
   6
   7 # CHECK:      Iterations:        100
   8 # CHECK-NEXT: Instructions:      200
   9 # CHECK-NEXT: Total Cycles:      206
  10 # CHECK-NEXT: Total uOps:        400
  11
  12 # CHECK:      Dispatch Width:    4
  13 # CHECK-NEXT: uOps Per Cycle:    1.94
  14 # CHECK-NEXT: IPC:               0.97
  15 # CHECK-NEXT: Block RThroughput: 2.0
  16
  17 # CHECK:      Instruction Info:
  18 # CHECK-NEXT: [1]: #uOps
  19 # CHECK-NEXT: [2]: Latency
  20 # CHECK-NEXT: [3]: RThroughput
  21 # CHECK-NEXT: [4]: MayLoad
  22 # CHECK-NEXT: [5]: MayStore
  23 # CHECK-NEXT: [6]: HasSideEffects (U)
  24
  25 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  26 # CHECK-NEXT:  2      1     0.50                        leaq    8(%rsp,%rdi,2), %rax
  27 # CHECK-NEXT:  2      6     2.00    *                   vbroadcastss    (%rax), %ymm0
  28
  29 # CHECK:      Resources:
  30 # CHECK-NEXT: [0.0] - PdAGLU01
  31 # CHECK-NEXT: [0.1] - PdAGLU01
  32 # CHECK-NEXT: [1]   - PdBranch
  33 # CHECK-NEXT: [2]   - PdCount
  34 # CHECK-NEXT: [3]   - PdDiv
  35 # CHECK-NEXT: [4]   - PdEX0
  36 # CHECK-NEXT: [5]   - PdEX1
  37 # CHECK-NEXT: [6]   - PdFPCVT
  38 # CHECK-NEXT: [7.0] - PdFPFMA
  39 # CHECK-NEXT: [7.1] - PdFPFMA
  40 # CHECK-NEXT: [8.0] - PdFPMAL
  41 # CHECK-NEXT: [8.1] - PdFPMAL
  42 # CHECK-NEXT: [9]   - PdFPMMA
  43 # CHECK-NEXT: [10]  - PdFPSTO
  44 # CHECK-NEXT: [11]  - PdFPU0
  45 # CHECK-NEXT: [12]  - PdFPU1
  46 # CHECK-NEXT: [13]  - PdFPU2
  47 # CHECK-NEXT: [14]  - PdFPU3
  48 # CHECK-NEXT: [15]  - PdFPXBR
  49 # CHECK-NEXT: [16.0] - PdLoad
  50 # CHECK-NEXT: [16.1] - PdLoad
  51 # CHECK-NEXT: [17]  - PdMul
  52 # CHECK-NEXT: [18]  - PdStore
  53
  54 # CHECK:      Resource pressure per iteration:
  55 # CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16.0] [16.1] [17]   [18]
  56 # CHECK-NEXT: 0.50   0.50    -      -      -     0.50   0.50    -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -      -
  57
  58 # CHECK:      Resource pressure by instruction:
  59 # CHECK-NEXT: [0.0]  [0.1]  [1]    [2]    [3]    [4]    [5]    [6]    [7.0]  [7.1]  [8.0]  [8.1]  [9]    [10]   [11]   [12]   [13]   [14]   [15]   [16.0] [16.1] [17]   [18]   Instructions:
  60 # CHECK-NEXT:  -      -      -      -      -     0.50   0.50    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     leaq     8(%rsp,%rdi,2), %rax
  61 # CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -      -     vbroadcastss     (%rax), %ymm0
  62
  63 # CHECK:      Timeline view:
  64 # CHECK-NEXT:                     012
  65 # CHECK-NEXT: Index     0123456789
  66
  67 # CHECK:      [0,0]     DeER .    . .   leaq    8(%rsp,%rdi,2), %rax
  68 # CHECK-NEXT: [0,1]     DeeeeeeER . .   vbroadcastss    (%rax), %ymm0
  69 # CHECK-NEXT: [1,0]     .DeE----R . .   leaq    8(%rsp,%rdi,2), %rax
  70 # CHECK-NEXT: [1,1]     .DeeeeeeER. .   vbroadcastss    (%rax), %ymm0
  71 # CHECK-NEXT: [2,0]     . DeE----R. .   leaq    8(%rsp,%rdi,2), %rax
  72 # CHECK-NEXT: [2,1]     . D==eeeeeeER   vbroadcastss    (%rax), %ymm0
  73
  74 # CHECK:      Average Wait times (based on the timeline view):
  75 # CHECK-NEXT: [0]: Executions
  76 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
  77 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
  78 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
  79
  80 # CHECK:            [0]    [1]    [2]    [3]
  81 # CHECK-NEXT: 0.     3     1.0    1.0    2.7       leaq 8(%rsp,%rdi,2), %rax
  82 # CHECK-NEXT: 1.     3     1.7    0.7    0.0       vbroadcastss (%rax), %ymm0
  83 # CHECK-NEXT:        3     1.3    0.8    1.3       <total>