xref: /llvm-project/llvm/test/tools/llvm-mca/X86/BdVer2/vbroadcast-operand-latency.s (revision 76fcf900d58826d9f21c0dd7f02b61b4d59c9193)
1a5192187SRoman Lebedev# NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
2a5baf867SRoman Lebedev# RUN: llvm-mca -mtriple=x86_64-unknown-unknown -mcpu=bdver2 -timeline -timeline-max-iterations=3 < %s | FileCheck %s
3a5192187SRoman Lebedev
4a5192187SRoman Lebedevleaq 8(%rsp, %rdi, 2), %rax
5a5192187SRoman Lebedevvbroadcastss (%rax), %ymm0
6a5192187SRoman Lebedev
7a5192187SRoman Lebedev# CHECK:      Iterations:        100
8a5192187SRoman Lebedev# CHECK-NEXT: Instructions:      200
9a5baf867SRoman Lebedev# CHECK-NEXT: Total Cycles:      206
10*76fcf900SRoman Lebedev# CHECK-NEXT: Total uOps:        300
11a5192187SRoman Lebedev
12a5192187SRoman Lebedev# CHECK:      Dispatch Width:    4
13*76fcf900SRoman Lebedev# CHECK-NEXT: uOps Per Cycle:    1.46
14a5baf867SRoman Lebedev# CHECK-NEXT: IPC:               0.97
15a5baf867SRoman Lebedev# CHECK-NEXT: Block RThroughput: 2.0
16a5192187SRoman Lebedev
17a5192187SRoman Lebedev# CHECK:      Instruction Info:
18a5192187SRoman Lebedev# CHECK-NEXT: [1]: #uOps
19a5192187SRoman Lebedev# CHECK-NEXT: [2]: Latency
20a5192187SRoman Lebedev# CHECK-NEXT: [3]: RThroughput
21a5192187SRoman Lebedev# CHECK-NEXT: [4]: MayLoad
22a5192187SRoman Lebedev# CHECK-NEXT: [5]: MayStore
23a5192187SRoman Lebedev# CHECK-NEXT: [6]: HasSideEffects (U)
24a5192187SRoman Lebedev
25a5192187SRoman Lebedev# CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
26*76fcf900SRoman Lebedev# CHECK-NEXT:  1      2     1.00                        leaq	8(%rsp,%rdi,2), %rax
27a5baf867SRoman Lebedev# CHECK-NEXT:  2      6     2.00    *                   vbroadcastss	(%rax), %ymm0
28a5192187SRoman Lebedev
29a5192187SRoman Lebedev# CHECK:      Resources:
30a5baf867SRoman Lebedev# CHECK-NEXT: [0.0] - PdAGLU01
31a5baf867SRoman Lebedev# CHECK-NEXT: [0.1] - PdAGLU01
32a5baf867SRoman Lebedev# CHECK-NEXT: [1]   - PdBranch
33a5baf867SRoman Lebedev# CHECK-NEXT: [2]   - PdCount
34a5baf867SRoman Lebedev# CHECK-NEXT: [3]   - PdDiv
35a5baf867SRoman Lebedev# CHECK-NEXT: [4]   - PdEX0
36a5baf867SRoman Lebedev# CHECK-NEXT: [5]   - PdEX1
37a5baf867SRoman Lebedev# CHECK-NEXT: [6]   - PdFPCVT
38a5baf867SRoman Lebedev# CHECK-NEXT: [7.0] - PdFPFMA
39a5baf867SRoman Lebedev# CHECK-NEXT: [7.1] - PdFPFMA
40a5baf867SRoman Lebedev# CHECK-NEXT: [8.0] - PdFPMAL
41a5baf867SRoman Lebedev# CHECK-NEXT: [8.1] - PdFPMAL
42a5baf867SRoman Lebedev# CHECK-NEXT: [9]   - PdFPMMA
43a5baf867SRoman Lebedev# CHECK-NEXT: [10]  - PdFPSTO
44a5baf867SRoman Lebedev# CHECK-NEXT: [11]  - PdFPU0
45a5baf867SRoman Lebedev# CHECK-NEXT: [12]  - PdFPU1
46a5baf867SRoman Lebedev# CHECK-NEXT: [13]  - PdFPU2
47a5baf867SRoman Lebedev# CHECK-NEXT: [14]  - PdFPU3
48a5baf867SRoman Lebedev# CHECK-NEXT: [15]  - PdFPXBR
49b428b8b2SRoman Lebedev# CHECK-NEXT: [16.0] - PdLoad
50b428b8b2SRoman Lebedev# CHECK-NEXT: [16.1] - PdLoad
51b428b8b2SRoman Lebedev# CHECK-NEXT: [17]  - PdMul
52b428b8b2SRoman Lebedev# CHECK-NEXT: [18]  - PdStore
53a5192187SRoman Lebedev
54a5192187SRoman Lebedev# CHECK:      Resource pressure per iteration:
55b428b8b2SRoman Lebedev# 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*76fcf900SRoman Lebedev# CHECK-NEXT: 0.50   0.50    -      -      -     1.00   1.00    -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -      -
57a5192187SRoman Lebedev
58a5192187SRoman Lebedev# CHECK:      Resource pressure by instruction:
59b428b8b2SRoman Lebedev# 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*76fcf900SRoman Lebedev# CHECK-NEXT:  -      -      -      -      -     1.00   1.00    -      -      -      -      -      -      -      -      -      -      -      -      -      -      -      -     leaq	8(%rsp,%rdi,2), %rax
61dda90323SAndrea Di Biagio# CHECK-NEXT: 0.50   0.50    -      -      -      -      -      -     2.00   2.00    -      -      -      -     1.00   1.00    -      -      -     0.50   0.50    -      -     vbroadcastss	(%rax), %ymm0
62a5192187SRoman Lebedev
63a5192187SRoman Lebedev# CHECK:      Timeline view:
64a5baf867SRoman Lebedev# CHECK-NEXT:                     012
65a5192187SRoman Lebedev# CHECK-NEXT: Index     0123456789
66a5192187SRoman Lebedev
67*76fcf900SRoman Lebedev# CHECK:      [0,0]     DeeER.    . .   leaq	8(%rsp,%rdi,2), %rax
68a5baf867SRoman Lebedev# CHECK-NEXT: [0,1]     DeeeeeeER . .   vbroadcastss	(%rax), %ymm0
69*76fcf900SRoman Lebedev# CHECK-NEXT: [1,0]     DeeE----R . .   leaq	8(%rsp,%rdi,2), %rax
70a5baf867SRoman Lebedev# CHECK-NEXT: [1,1]     .DeeeeeeER. .   vbroadcastss	(%rax), %ymm0
71*76fcf900SRoman Lebedev# CHECK-NEXT: [2,0]     .D=eeE---R. .   leaq	8(%rsp,%rdi,2), %rax
72a5baf867SRoman Lebedev# CHECK-NEXT: [2,1]     . D==eeeeeeER   vbroadcastss	(%rax), %ymm0
73a5192187SRoman Lebedev
74a5192187SRoman Lebedev# CHECK:      Average Wait times (based on the timeline view):
75a5192187SRoman Lebedev# CHECK-NEXT: [0]: Executions
76a5192187SRoman Lebedev# CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
77a5192187SRoman Lebedev# CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
78a5192187SRoman Lebedev# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
79a5192187SRoman Lebedev
80a5192187SRoman Lebedev# CHECK:            [0]    [1]    [2]    [3]
81*76fcf900SRoman Lebedev# CHECK-NEXT: 0.     3     1.3    1.3    2.3       leaq	8(%rsp,%rdi,2), %rax
82*76fcf900SRoman Lebedev# CHECK-NEXT: 1.     3     1.7    1.0    0.0       vbroadcastss	(%rax), %ymm0
83*76fcf900SRoman Lebedev# CHECK-NEXT:        3     1.5    1.2    1.2       <total>
84