test/tools/llvm-mca/X86/SkylakeClient/bottleneck-analysis.s

   1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
   2 # RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=skylake  -bottleneck-analysis < %s | FileCheck %s
   3
   4 .LBB0_4:
   5   vmovups       (%rsi,%rax,2), %xmm0
   6   vpermilps     $255, %xmm0, %xmm7
   7   vmulps        -24(%rsp), %xmm7, %xmm8
   8   vpermilps     $170, %xmm0, %xmm6
   9   vpermilps     $85, %xmm0, %xmm5
  10   vbroadcastss  %xmm0, %xmm0
  11   vfmadd231ps   %xmm9, %xmm6, %xmm8
  12   vfmadd213ps   %xmm8, %xmm10, %xmm5
  13   vfmadd213ps   %xmm5, %xmm11, %xmm0
  14   vfmadd213ps   %xmm0, %xmm12, %xmm4
  15   vfmadd213ps   %xmm4, %xmm13, %xmm1
  16   vmovaps       %xmm7, %xmm4
  17   vfmadd213ps   %xmm1, %xmm14, %xmm2
  18   vmovaps       %xmm6, %xmm1
  19   vfmadd213ps   %xmm2, %xmm15, %xmm3
  20   vpermilps     $170, %xmm3, %xmm0
  21   vmovups       %xmm3, (%rdx,%rax)
  22   vpermilps     $255, %xmm3, %xmm2
  23   addq  $16, %rax
  24   decl  %ecx
  25   vmovaps       %xmm0, %xmm3
  26   jne   .LBB0_4
  27
  28 # CHECK:      Iterations:        100
  29 # CHECK-NEXT: Instructions:      2200
  30 # CHECK-NEXT: Total Cycles:      1039
  31 # CHECK-NEXT: Total uOps:        2400
  32
  33 # CHECK:      Dispatch Width:    6
  34 # CHECK-NEXT: uOps Per Cycle:    2.31
  35 # CHECK-NEXT: IPC:               2.12
  36 # CHECK-NEXT: Block RThroughput: 6.0
  37
  38 # CHECK:      Cycles with backend pressure increase [ 92.69% ]
  39 # CHECK-NEXT: Throughput Bottlenecks:
  40 # CHECK-NEXT:   Resource Pressure       [ 46.78% ]
  41 # CHECK-NEXT:   - SKLPort0  [ 14.24% ]
  42 # CHECK-NEXT:   - SKLPort1  [ 14.24% ]
  43 # CHECK-NEXT:   - SKLPort5  [ 46.49% ]
  44 # CHECK-NEXT:   - SKLPort6  [ 8.66% ]
  45 # CHECK-NEXT:   Data Dependencies:      [ 64.97% ]
  46 # CHECK-NEXT:   - Register Dependencies [ 64.97% ]
  47 # CHECK-NEXT:   - Memory Dependencies   [ 0.00% ]
  48
  49 # CHECK:      Critical sequence based on the simulation:
  50
  51 # CHECK:                    Instruction                                 Dependency Information
  52 # CHECK-NEXT:  +----< 18.   addq        $16, %rax
  53 # CHECK-NEXT:  |
  54 # CHECK-NEXT:  |    < loop carried >
  55 # CHECK-NEXT:  |
  56 # CHECK-NEXT:  +----> 0.    vmovups     (%rsi,%rax,2), %xmm0              ## REGISTER dependency:  %rax
  57 # CHECK-NEXT:  |      1.    vpermilps   $255, %xmm0, %xmm7
  58 # CHECK-NEXT:  |      2.    vmulps      -24(%rsp), %xmm7, %xmm8
  59 # CHECK-NEXT:  +----> 3.    vpermilps   $170, %xmm0, %xmm6                ## REGISTER dependency:  %xmm0
  60 # CHECK-NEXT:  |      4.    vpermilps   $85, %xmm0, %xmm5
  61 # CHECK-NEXT:  |      5.    vbroadcastss        %xmm0, %xmm0
  62 # CHECK-NEXT:  +----> 6.    vfmadd231ps %xmm9, %xmm6, %xmm8       ## REGISTER dependency:  %xmm6
  63 # CHECK-NEXT:  +----> 7.    vfmadd213ps %xmm8, %xmm10, %xmm5      ## REGISTER dependency:  %xmm8
  64 # CHECK-NEXT:  +----> 8.    vfmadd213ps %xmm5, %xmm11, %xmm0      ## REGISTER dependency:  %xmm5
  65 # CHECK-NEXT:  +----> 9.    vfmadd213ps %xmm0, %xmm12, %xmm4      ## REGISTER dependency:  %xmm0
  66 # CHECK-NEXT:  +----> 10.   vfmadd213ps %xmm4, %xmm13, %xmm1      ## REGISTER dependency:  %xmm4
  67 # CHECK-NEXT:  |      11.   vmovaps     %xmm7, %xmm4
  68 # CHECK-NEXT:  +----> 12.   vfmadd213ps %xmm1, %xmm14, %xmm2      ## REGISTER dependency:  %xmm1
  69 # CHECK-NEXT:  |      13.   vmovaps     %xmm6, %xmm1
  70 # CHECK-NEXT:  +----> 14.   vfmadd213ps %xmm2, %xmm15, %xmm3      ## REGISTER dependency:  %xmm2
  71 # CHECK-NEXT:  +----> 15.   vpermilps   $170, %xmm3, %xmm0                ## REGISTER dependency:  %xmm3
  72 # CHECK-NEXT:  |      16.   vmovups     %xmm3, (%rdx,%rax)
  73 # CHECK-NEXT:  |      17.   vpermilps   $255, %xmm3, %xmm2
  74 # CHECK-NEXT:  |      18.   addq        $16, %rax
  75 # CHECK-NEXT:  |      19.   decl        %ecx
  76 # CHECK-NEXT:  +----> 20.   vmovaps     %xmm0, %xmm3                      ## REGISTER dependency:  %xmm0
  77 # CHECK-NEXT:         21.   jne .LBB0_4
  78
  79 # CHECK:      Instruction Info:
  80 # CHECK-NEXT: [1]: #uOps
  81 # CHECK-NEXT: [2]: Latency
  82 # CHECK-NEXT: [3]: RThroughput
  83 # CHECK-NEXT: [4]: MayLoad
  84 # CHECK-NEXT: [5]: MayStore
  85 # CHECK-NEXT: [6]: HasSideEffects (U)
  86
  87 # CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
  88 # CHECK-NEXT:  1      6     0.50    *                   vmovups (%rsi,%rax,2), %xmm0
  89 # CHECK-NEXT:  1      1     1.00                        vpermilps       $255, %xmm0, %xmm7
  90 # CHECK-NEXT:  2      10    0.50    *                   vmulps  -24(%rsp), %xmm7, %xmm8
  91 # CHECK-NEXT:  1      1     1.00                        vpermilps       $170, %xmm0, %xmm6
  92 # CHECK-NEXT:  1      1     1.00                        vpermilps       $85, %xmm0, %xmm5
  93 # CHECK-NEXT:  1      1     1.00                        vbroadcastss    %xmm0, %xmm0
  94 # CHECK-NEXT:  1      4     0.50                        vfmadd231ps     %xmm9, %xmm6, %xmm8
  95 # CHECK-NEXT:  1      4     0.50                        vfmadd213ps     %xmm8, %xmm10, %xmm5
  96 # CHECK-NEXT:  1      4     0.50                        vfmadd213ps     %xmm5, %xmm11, %xmm0
  97 # CHECK-NEXT:  1      4     0.50                        vfmadd213ps     %xmm0, %xmm12, %xmm4
  98 # CHECK-NEXT:  1      4     0.50                        vfmadd213ps     %xmm4, %xmm13, %xmm1
  99 # CHECK-NEXT:  1      1     0.33                        vmovaps %xmm7, %xmm4
 100 # CHECK-NEXT:  1      4     0.50                        vfmadd213ps     %xmm1, %xmm14, %xmm2
 101 # CHECK-NEXT:  1      1     0.33                        vmovaps %xmm6, %xmm1
 102 # CHECK-NEXT:  1      4     0.50                        vfmadd213ps     %xmm2, %xmm15, %xmm3
 103 # CHECK-NEXT:  1      1     1.00                        vpermilps       $170, %xmm3, %xmm0
 104 # CHECK-NEXT:  2      1     1.00           *            vmovups %xmm3, (%rdx,%rax)
 105 # CHECK-NEXT:  1      1     1.00                        vpermilps       $255, %xmm3, %xmm2
 106 # CHECK-NEXT:  1      1     0.25                        addq    $16, %rax
 107 # CHECK-NEXT:  1      1     0.25                        decl    %ecx
 108 # CHECK-NEXT:  1      1     0.33                        vmovaps %xmm0, %xmm3
 109 # CHECK-NEXT:  1      1     0.50                        jne     .LBB0_4
 110
 111 # CHECK:      Resources:
 112 # CHECK-NEXT: [0]   - SKLDivider
 113 # CHECK-NEXT: [1]   - SKLFPDivider
 114 # CHECK-NEXT: [2]   - SKLPort0
 115 # CHECK-NEXT: [3]   - SKLPort1
 116 # CHECK-NEXT: [4]   - SKLPort2
 117 # CHECK-NEXT: [5]   - SKLPort3
 118 # CHECK-NEXT: [6]   - SKLPort4
 119 # CHECK-NEXT: [7]   - SKLPort5
 120 # CHECK-NEXT: [8]   - SKLPort6
 121 # CHECK-NEXT: [9]   - SKLPort7
 122
 123 # CHECK:      Resource pressure per iteration:
 124 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]
 125 # CHECK-NEXT:  -      -     5.52   5.53   1.01   1.03   1.00   6.02   2.93   0.96
 126
 127 # CHECK:      Resource pressure by instruction:
 128 # CHECK-NEXT: [0]    [1]    [2]    [3]    [4]    [5]    [6]    [7]    [8]    [9]    Instructions:
 129 # CHECK-NEXT:  -      -      -      -     0.04   0.96    -      -      -      -     vmovups     (%rsi,%rax,2), %xmm0
 130 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermilps   $255, %xmm0, %xmm7
 131 # CHECK-NEXT:  -      -     0.03   0.97   0.96   0.04    -      -      -      -     vmulps      -24(%rsp), %xmm7, %xmm8
 132 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermilps   $170, %xmm0, %xmm6
 133 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermilps   $85, %xmm0, %xmm5
 134 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vbroadcastss        %xmm0, %xmm0
 135 # CHECK-NEXT:  -      -     0.95   0.05    -      -      -      -      -      -     vfmadd231ps %xmm9, %xmm6, %xmm8
 136 # CHECK-NEXT:  -      -     0.50   0.50    -      -      -      -      -      -     vfmadd213ps %xmm8, %xmm10, %xmm5
 137 # CHECK-NEXT:  -      -     0.92   0.08    -      -      -      -      -      -     vfmadd213ps %xmm5, %xmm11, %xmm0
 138 # CHECK-NEXT:  -      -     0.95   0.05    -      -      -      -      -      -     vfmadd213ps %xmm0, %xmm12, %xmm4
 139 # CHECK-NEXT:  -      -     0.51   0.49    -      -      -      -      -      -     vfmadd213ps %xmm4, %xmm13, %xmm1
 140 # CHECK-NEXT:  -      -     0.52   0.48    -      -      -      -      -      -     vmovaps     %xmm7, %xmm4
 141 # CHECK-NEXT:  -      -     0.49   0.51    -      -      -      -      -      -     vfmadd213ps %xmm1, %xmm14, %xmm2
 142 # CHECK-NEXT:  -      -     0.04   0.95    -      -      -     0.01    -      -     vmovaps     %xmm6, %xmm1
 143 # CHECK-NEXT:  -      -     0.51   0.49    -      -      -      -      -      -     vfmadd213ps %xmm2, %xmm15, %xmm3
 144 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermilps   $170, %xmm3, %xmm0
 145 # CHECK-NEXT:  -      -      -      -     0.01   0.03   1.00    -      -     0.96   vmovups     %xmm3, (%rdx,%rax)
 146 # CHECK-NEXT:  -      -      -      -      -      -      -     1.00    -      -     vpermilps   $255, %xmm3, %xmm2
 147 # CHECK-NEXT:  -      -      -      -      -      -      -      -     1.00    -     addq        $16, %rax
 148 # CHECK-NEXT:  -      -     0.04   0.01    -      -      -     0.01   0.94    -     decl        %ecx
 149 # CHECK-NEXT:  -      -     0.05   0.95    -      -      -      -      -      -     vmovaps     %xmm0, %xmm3
 150 # CHECK-NEXT:  -      -     0.01    -      -      -      -      -     0.99    -     jne .LBB0_4