llvm/test/tools/llvm-mca/AArch64/Exynos/double-rsqrt.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -march=aarch64 -mcpu=exynos-m3 -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,M3
   3 # RUN: llvm-mca -march=aarch64 -mcpu=exynos-m4 -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,M4
   4 # RUN: llvm-mca -march=aarch64 -mcpu=exynos-m5 -resource-pressure=false < %s | FileCheck %s -check-prefixes=ALL,M5
   5
   6 fsqrt   d30, d30
   7 fmov    d31, #1.00000000
   8 fdiv    d30, d31, d30
   9
  10 # Newton series for 1 / sqrt().
  11 frsqrte d1, d0
  12 fmul    d2, d1, d1
  13 frsqrts d2, d0, d2
  14 fmul    d1, d1, d2
  15 fmul    d2, d1, d1
  16 frsqrts d2, d0, d2
  17 fmul    d1, d1, d2
  18 fmul    d2, d1, d1
  19 frsqrts d0, d0, d2
  20 fmul    d0, d1, d0
  21
  22 # ALL:      Iterations:        100
  23 # ALL-NEXT: Instructions:      1300
  24
  25 # M3-NEXT:  Total Cycles:      3703
  26 # M4-NEXT:  Total Cycles:      3303
  27 # M5-NEXT:  Total Cycles:      3303
  28
  29 # ALL-NEXT: Total uOps:        1300
  30
  31 # ALL:      Dispatch Width:    6
  32
  33 # M3-NEXT:  uOps Per Cycle:    0.35
  34 # M3-NEXT:  IPC:               0.35
  35 # M3-NEXT:  Block RThroughput: 26.0
  36
  37 # M4-NEXT:  uOps Per Cycle:    0.39
  38 # M4-NEXT:  IPC:               0.39
  39 # M4-NEXT:  Block RThroughput: 3.0
  40
  41 # M5-NEXT:  uOps Per Cycle:    0.39
  42 # M5-NEXT:  IPC:               0.39
  43 # M5-NEXT:  Block RThroughput: 3.0
  44
  45 # ALL:      Instruction Info:
  46 # ALL-NEXT: [1]: #uOps
  47 # ALL-NEXT: [2]: Latency
  48 # ALL-NEXT: [3]: RThroughput
  49 # ALL-NEXT: [4]: MayLoad
  50 # ALL-NEXT: [5]: MayStore
  51 # ALL-NEXT: [6]: HasSideEffects (U)
  52
  53 # ALL:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  54
  55 # M3-NEXT:   1      25    26.00                       fsqrt     d30, d30
  56 # M4-NEXT:   1      12    2.25                        fsqrt     d30, d30
  57 # M5-NEXT:   1      12    2.25                        fsqrt     d30, d30
  58
  59 # ALL-NEXT:  1      1     0.33                        fmov      d31, #1.00000000
  60
  61 # M3-NEXT:   1      12    3.25                        fdiv      d30, d31, d30
  62 # M3-NEXT:   1      4     0.50                        frsqrte   d1, d0
  63
  64 # M4-NEXT:   1      12    2.25                        fdiv      d30, d31, d30
  65 # M4-NEXT:   1      3     0.50                        frsqrte   d1, d0
  66
  67 # M5-NEXT:   1      12    2.25                        fdiv      d30, d31, d30
  68 # M5-NEXT:   1      3     0.50                        frsqrte   d1, d0
  69
  70 # ALL-NEXT:  1      3     0.33                        fmul      d2, d1, d1
  71 # ALL-NEXT:  1      4     0.33                        frsqrts   d2, d0, d2
  72 # ALL-NEXT:  1      3     0.33                        fmul      d1, d1, d2
  73 # ALL-NEXT:  1      3     0.33                        fmul      d2, d1, d1
  74 # ALL-NEXT:  1      4     0.33                        frsqrts   d2, d0, d2
  75 # ALL-NEXT:  1      3     0.33                        fmul      d1, d1, d2
  76 # ALL-NEXT:  1      3     0.33                        fmul      d2, d1, d1
  77 # ALL-NEXT:  1      4     0.33                        frsqrts   d0, d0, d2
  78 # ALL-NEXT:  1      3     0.33                        fmul      d0, d1, d0