xref: /llvm-project/llvm/test/tools/llvm-mca/X86/BdVer2/memcpy-like-test.s (revision 5578ec32f9c4fef46adce52a2e3d22bf409b3d2c)
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 -iterations=100 -timeline -timeline-max-iterations=1 < %s | FileCheck %s
3a5192187SRoman Lebedev
4a5192187SRoman Lebedevvmovaps (%rsi), %xmm0
5a5192187SRoman Lebedevvmovaps %xmm0, (%rdi)
6a5192187SRoman Lebedevvmovaps 16(%rsi), %xmm0
7a5192187SRoman Lebedevvmovaps %xmm0, 16(%rdi)
8a5192187SRoman Lebedevvmovaps 32(%rsi), %xmm0
9a5192187SRoman Lebedevvmovaps %xmm0, 32(%rdi)
10a5192187SRoman Lebedevvmovaps 48(%rsi), %xmm0
11a5192187SRoman Lebedevvmovaps %xmm0, 48(%rdi)
12a5192187SRoman Lebedev
13a5192187SRoman Lebedev# CHECK:      Iterations:        100
14a5192187SRoman Lebedev# CHECK-NEXT: Instructions:      800
159db0e725SRoman Lebedev# CHECK-NEXT: Total Cycles:      806
16a5192187SRoman Lebedev# CHECK-NEXT: Total uOps:        800
17a5192187SRoman Lebedev
18a5192187SRoman Lebedev# CHECK:      Dispatch Width:    4
199db0e725SRoman Lebedev# CHECK-NEXT: uOps Per Cycle:    0.99
209db0e725SRoman Lebedev# CHECK-NEXT: IPC:               0.99
219db0e725SRoman Lebedev# CHECK-NEXT: Block RThroughput: 8.0
22a5192187SRoman Lebedev
23a5192187SRoman Lebedev# CHECK:      Instruction Info:
24a5192187SRoman Lebedev# CHECK-NEXT: [1]: #uOps
25a5192187SRoman Lebedev# CHECK-NEXT: [2]: Latency
26a5192187SRoman Lebedev# CHECK-NEXT: [3]: RThroughput
27a5192187SRoman Lebedev# CHECK-NEXT: [4]: MayLoad
28a5192187SRoman Lebedev# CHECK-NEXT: [5]: MayStore
29a5192187SRoman Lebedev# CHECK-NEXT: [6]: HasSideEffects (U)
30a5192187SRoman Lebedev
31a5192187SRoman Lebedev# CHECK:      [1]    [2]    [3]    [4]    [5]    [6]    Instructions:
329db0e725SRoman Lebedev# CHECK-NEXT:  1      5     1.50    *                   vmovaps	(%rsi), %xmm0
339db0e725SRoman Lebedev# CHECK-NEXT:  1      1     1.50           *            vmovaps	%xmm0, (%rdi)
349db0e725SRoman Lebedev# CHECK-NEXT:  1      5     1.50    *                   vmovaps	16(%rsi), %xmm0
359db0e725SRoman Lebedev# CHECK-NEXT:  1      1     1.50           *            vmovaps	%xmm0, 16(%rdi)
369db0e725SRoman Lebedev# CHECK-NEXT:  1      5     1.50    *                   vmovaps	32(%rsi), %xmm0
379db0e725SRoman Lebedev# CHECK-NEXT:  1      1     1.50           *            vmovaps	%xmm0, 32(%rdi)
389db0e725SRoman Lebedev# CHECK-NEXT:  1      5     1.50    *                   vmovaps	48(%rsi), %xmm0
399db0e725SRoman Lebedev# CHECK-NEXT:  1      1     1.50           *            vmovaps	%xmm0, 48(%rdi)
40a5192187SRoman Lebedev
41a5192187SRoman Lebedev# CHECK:      Resources:
42a5baf867SRoman Lebedev# CHECK-NEXT: [0.0] - PdAGLU01
43a5baf867SRoman Lebedev# CHECK-NEXT: [0.1] - PdAGLU01
44a5baf867SRoman Lebedev# CHECK-NEXT: [1]   - PdBranch
45a5baf867SRoman Lebedev# CHECK-NEXT: [2]   - PdCount
46a5baf867SRoman Lebedev# CHECK-NEXT: [3]   - PdDiv
47a5baf867SRoman Lebedev# CHECK-NEXT: [4]   - PdEX0
48a5baf867SRoman Lebedev# CHECK-NEXT: [5]   - PdEX1
49a5baf867SRoman Lebedev# CHECK-NEXT: [6]   - PdFPCVT
50a5baf867SRoman Lebedev# CHECK-NEXT: [7.0] - PdFPFMA
51a5baf867SRoman Lebedev# CHECK-NEXT: [7.1] - PdFPFMA
52a5baf867SRoman Lebedev# CHECK-NEXT: [8.0] - PdFPMAL
53a5baf867SRoman Lebedev# CHECK-NEXT: [8.1] - PdFPMAL
54a5baf867SRoman Lebedev# CHECK-NEXT: [9]   - PdFPMMA
55a5baf867SRoman Lebedev# CHECK-NEXT: [10]  - PdFPSTO
56a5baf867SRoman Lebedev# CHECK-NEXT: [11]  - PdFPU0
57a5baf867SRoman Lebedev# CHECK-NEXT: [12]  - PdFPU1
58a5baf867SRoman Lebedev# CHECK-NEXT: [13]  - PdFPU2
59a5baf867SRoman Lebedev# CHECK-NEXT: [14]  - PdFPU3
60a5baf867SRoman Lebedev# CHECK-NEXT: [15]  - PdFPXBR
61b428b8b2SRoman Lebedev# CHECK-NEXT: [16.0] - PdLoad
62b428b8b2SRoman Lebedev# CHECK-NEXT: [16.1] - PdLoad
63b428b8b2SRoman Lebedev# CHECK-NEXT: [17]  - PdMul
64b428b8b2SRoman Lebedev# CHECK-NEXT: [18]  - PdStore
65a5192187SRoman Lebedev
66a5192187SRoman Lebedev# CHECK:      Resource pressure per iteration:
67b428b8b2SRoman 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]
689db0e725SRoman Lebedev# CHECK-NEXT: 8.00   8.00    -      -      -      -      -      -     6.00   6.00    -      -      -     4.00   2.00   2.00   6.00   6.00    -     6.00   6.00    -     4.00
69a5192187SRoman Lebedev
70a5192187SRoman Lebedev# CHECK:      Resource pressure by instruction:
71b428b8b2SRoman 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:
729db0e725SRoman Lebedev# CHECK-NEXT: 2.97   0.03    -      -      -      -      -      -      -     3.00    -      -      -      -      -     1.00    -      -      -      -     3.00    -      -     vmovaps	(%rsi), %xmm0
739db0e725SRoman Lebedev# CHECK-NEXT:  -     1.00    -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     3.00    -      -      -      -     1.00   vmovaps	%xmm0, (%rdi)
749db0e725SRoman Lebedev# CHECK-NEXT: 0.03   2.97    -      -      -      -      -      -     3.00    -      -      -      -      -     1.00    -      -      -      -     3.00    -      -      -     vmovaps	16(%rsi), %xmm0
759db0e725SRoman Lebedev# CHECK-NEXT: 0.01   0.99    -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     3.00    -      -      -      -      -     1.00   vmovaps	%xmm0, 16(%rdi)
769db0e725SRoman Lebedev# CHECK-NEXT:  -     3.00    -      -      -      -      -      -      -     3.00    -      -      -      -      -     1.00    -      -      -      -     3.00    -      -     vmovaps	32(%rsi), %xmm0
779db0e725SRoman Lebedev# CHECK-NEXT: 0.99   0.01    -      -      -      -      -      -      -      -      -      -      -     1.00    -      -      -     3.00    -      -      -      -     1.00   vmovaps	%xmm0, 32(%rdi)
789db0e725SRoman Lebedev# CHECK-NEXT: 3.00    -      -      -      -      -      -      -     3.00    -      -      -      -      -     1.00    -      -      -      -     3.00    -      -      -     vmovaps	48(%rsi), %xmm0
799db0e725SRoman Lebedev# CHECK-NEXT: 1.00    -      -      -      -      -      -      -      -      -      -      -      -     1.00    -      -     3.00    -      -      -      -      -     1.00   vmovaps	%xmm0, 48(%rdi)
80a5192187SRoman Lebedev
81a5192187SRoman Lebedev# CHECK:      Timeline view:
829db0e725SRoman Lebedev# CHECK-NEXT:                     0123
83a5192187SRoman Lebedev# CHECK-NEXT: Index     0123456789
84a5192187SRoman Lebedev
85a5baf867SRoman Lebedev# CHECK:      [0,0]     DeeeeeER  .  .   vmovaps	(%rsi), %xmm0
869db0e725SRoman Lebedev# CHECK-NEXT: [0,1]     D======eER.  .   vmovaps	%xmm0, (%rdi)
879db0e725SRoman Lebedev# CHECK-NEXT: [0,2]     DeeeeeE--R.  .   vmovaps	16(%rsi), %xmm0
889db0e725SRoman Lebedev# CHECK-NEXT: [0,3]     D=======eER  .   vmovaps	%xmm0, 16(%rdi)
899db0e725SRoman Lebedev# CHECK-NEXT: [0,4]     .D==eeeeeER  .   vmovaps	32(%rsi), %xmm0
909db0e725SRoman Lebedev# CHECK-NEXT: [0,5]     .D========eER.   vmovaps	%xmm0, 32(%rdi)
919db0e725SRoman Lebedev# CHECK-NEXT: [0,6]     .D==eeeeeE--R.   vmovaps	48(%rsi), %xmm0
929db0e725SRoman Lebedev# CHECK-NEXT: [0,7]     .D=========eER   vmovaps	%xmm0, 48(%rdi)
93a5192187SRoman Lebedev
94a5192187SRoman Lebedev# CHECK:      Average Wait times (based on the timeline view):
95a5192187SRoman Lebedev# CHECK-NEXT: [0]: Executions
96a5192187SRoman Lebedev# CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
97a5192187SRoman Lebedev# CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
98a5192187SRoman Lebedev# CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
99a5192187SRoman Lebedev
100a5192187SRoman Lebedev# CHECK:            [0]    [1]    [2]    [3]
101a5192187SRoman Lebedev# CHECK-NEXT: 0.     1     1.0    1.0    0.0       vmovaps	(%rsi), %xmm0
1029db0e725SRoman Lebedev# CHECK-NEXT: 1.     1     7.0    1.0    0.0       vmovaps	%xmm0, (%rdi)
1039db0e725SRoman Lebedev# CHECK-NEXT: 2.     1     1.0    1.0    2.0       vmovaps	16(%rsi), %xmm0
104*5578ec32SAndrea Di Biagio# CHECK-NEXT: 3.     1     8.0    1.0    0.0       vmovaps	%xmm0, 16(%rdi)
1059db0e725SRoman Lebedev# CHECK-NEXT: 4.     1     3.0    3.0    0.0       vmovaps	32(%rsi), %xmm0
1069db0e725SRoman Lebedev# CHECK-NEXT: 5.     1     9.0    1.0    0.0       vmovaps	%xmm0, 32(%rdi)
1079db0e725SRoman Lebedev# CHECK-NEXT: 6.     1     3.0    3.0    2.0       vmovaps	48(%rsi), %xmm0
108*5578ec32SAndrea Di Biagio# CHECK-NEXT: 7.     1     10.0   1.0    0.0       vmovaps	%xmm0, 48(%rdi)
109*5578ec32SAndrea Di Biagio# CHECK-NEXT:        1     5.3    1.5    0.5       <total>
110