test/tools/llvm-mca/X86/BdVer2/dependency-breaking-pcmpgt.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 -iterations=1500 < %s | FileCheck %s
   3
   4 # perf stat reports an IPC of 2.00 for this block of code.
   5
   6 # All of the vector packed compares from this test are zero idioms.  These zero
   7 # idioms are all detected and removed by the register renamer.  That means, no
   8 # uOp is executed, and there is no RAW dependency for any of the packed
   9 # compares.
  10
  11 vpcmpgtb %xmm0, %xmm0, %xmm1
  12 vpcmpgtw %xmm1, %xmm1, %xmm2
  13 vpcmpgtd %xmm2, %xmm2, %xmm3
  14 vpcmpgtq %xmm3, %xmm3, %xmm0
  15
  16 # CHECK:      Iterations:        1500
  17 # CHECK-NEXT: Instructions:      6000
  18 # CHECK-NEXT: Total Cycles:      1504
  19 # CHECK-NEXT: Total uOps:        6000
  20
  21 # CHECK:      Dispatch Width:    4
  22 # CHECK-NEXT: uOps Per Cycle:    3.99
  23 # CHECK-NEXT: IPC:               3.99
  24 # CHECK-NEXT: Block RThroughput: 1.0
  25
  26 # CHECK:      Instruction Info:
  27 # CHECK-NEXT: [1]: #uOps
  28 # CHECK-NEXT: [2]: Latency
  29 # CHECK-NEXT: [3]: RThroughput
  30 # CHECK-NEXT: [4]: MayLoad
  31 # CHECK-NEXT: [5]: MayStore
  32 # CHECK-NEXT: [6]: HasSideEffects (U)
  33
  34 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  35 # CHECK-NEXT:  1      0     0.25                        vpcmpgtb        %xmm0, %xmm0, %xmm1
  36 # CHECK-NEXT:  1      0     0.25                        vpcmpgtw        %xmm1, %xmm1, %xmm2
  37 # CHECK-NEXT:  1      0     0.25                        vpcmpgtd        %xmm2, %xmm2, %xmm3
  38 # CHECK-NEXT:  1      2     1.00                        vpcmpgtq        %xmm3, %xmm3, %xmm0
  39
  40 # CHECK:      Resources:
  41 # CHECK-NEXT: [0.0] - PdAGLU01
  42 # CHECK-NEXT: [0.1] - PdAGLU01
  43 # CHECK-NEXT: [1]   - PdBranch
  44 # CHECK-NEXT: [2]   - PdCount
  45 # CHECK-NEXT: [3]   - PdDiv
  46 # CHECK-NEXT: [4]   - PdEX0
  47 # CHECK-NEXT: [5]   - PdEX1
  48 # CHECK-NEXT: [6]   - PdFPCVT
  49 # CHECK-NEXT: [7.0] - PdFPFMA
  50 # CHECK-NEXT: [7.1] - PdFPFMA
  51 # CHECK-NEXT: [8.0] - PdFPMAL
  52 # CHECK-NEXT: [8.1] - PdFPMAL
  53 # CHECK-NEXT: [9]   - PdFPMMA
  54 # CHECK-NEXT: [10]  - PdFPSTO
  55 # CHECK-NEXT: [11]  - PdFPU0
  56 # CHECK-NEXT: [12]  - PdFPU1
  57 # CHECK-NEXT: [13]  - PdFPU2
  58 # CHECK-NEXT: [14]  - PdFPU3
  59 # CHECK-NEXT: [15]  - PdFPXBR
  60 # CHECK-NEXT: [16.0] - PdLoad
  61 # CHECK-NEXT: [16.1] - PdLoad
  62 # CHECK-NEXT: [17]  - PdMul
  63 # CHECK-NEXT: [18]  - PdStore
  64
  65 # CHECK:      Resource pressure per iteration:
  66 # 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]
  67 # CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00   1.00    -      -     0.50   0.50    -      -      -      -      -      -      -
  68
  69 # CHECK:      Resource pressure by instruction:
  70 # 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:
  71 # CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtb %xmm0, %xmm0, %xmm1
  72 # CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtw %xmm1, %xmm1, %xmm2
  73 # CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     vpcmpgtd %xmm2, %xmm2, %xmm3
  74 # CHECK-NEXT:  -      -      -      -      -      -      -      -      -      -     1.00   1.00    -      -     0.50   0.50    -      -      -      -      -      -      -     vpcmpgtq %xmm3, %xmm3, %xmm0
  75
  76 # CHECK:      Timeline view:
  77 # CHECK-NEXT: Index     0123456
  78
  79 # CHECK:      [0,0]     DR   ..   vpcmpgtb      %xmm0, %xmm0, %xmm1
  80 # CHECK-NEXT: [0,1]     DR   ..   vpcmpgtw      %xmm1, %xmm1, %xmm2
  81 # CHECK-NEXT: [0,2]     DR   ..   vpcmpgtd      %xmm2, %xmm2, %xmm3
  82 # CHECK-NEXT: [0,3]     DeeER..   vpcmpgtq      %xmm3, %xmm3, %xmm0
  83 # CHECK-NEXT: [1,0]     .D--R..   vpcmpgtb      %xmm0, %xmm0, %xmm1
  84 # CHECK-NEXT: [1,1]     .D--R..   vpcmpgtw      %xmm1, %xmm1, %xmm2
  85 # CHECK-NEXT: [1,2]     .D--R..   vpcmpgtd      %xmm2, %xmm2, %xmm3
  86 # CHECK-NEXT: [1,3]     .DeeER.   vpcmpgtq      %xmm3, %xmm3, %xmm0
  87 # CHECK-NEXT: [2,0]     . D--R.   vpcmpgtb      %xmm0, %xmm0, %xmm1
  88 # CHECK-NEXT: [2,1]     . D--R.   vpcmpgtw      %xmm1, %xmm1, %xmm2
  89 # CHECK-NEXT: [2,2]     . D--R.   vpcmpgtd      %xmm2, %xmm2, %xmm3
  90 # CHECK-NEXT: [2,3]     . DeeER   vpcmpgtq      %xmm3, %xmm3, %xmm0
  91
  92 # CHECK:      Average Wait times (based on the timeline view):
  93 # CHECK-NEXT: [0]: Executions
  94 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
  95 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
  96 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
  97
  98 # CHECK:            [0]    [1]    [2]    [3]
  99 # CHECK-NEXT: 0.     3     0.0    0.0    1.3       vpcmpgtb     %xmm0, %xmm0, %xmm1
 100 # CHECK-NEXT: 1.     3     0.0    0.0    1.3       vpcmpgtw     %xmm1, %xmm1, %xmm2
 101 # CHECK-NEXT: 2.     3     0.0    0.0    1.3       vpcmpgtd     %xmm2, %xmm2, %xmm3
 102 # CHECK-NEXT: 3.     3     1.0    1.0    0.0       vpcmpgtq     %xmm3, %xmm3, %xmm0
 103 # CHECK-NEXT:        3     0.3    0.3    1.0       <total>