[flang][cuda] Allow DO CONCURRENT in cuf kernel (#124190)
[llvm-project.git] / llvm / test / tools / llvm-mca / AArch64 / Cortex / A55-load-store-alias.s
blob3906975b41f6b464e53f8efc7aeb8a8bb0d907ce
1 # NOTE: Assertions have been autogenerated by utils/update_mca_test_checks.py
2 # RUN: llvm-mca -mtriple=aarch64 -mcpu=cortex-a55 -timeline --iterations=3 -noalias=false < %s | FileCheck %s
4 # PR50483: Execution of loads and stores should not overlap if flag -noalias is set to false.
6 str x1, [x10]
7 str x1, [x10]
8 ldr x2, [x10]
9 nop
10 ldr x2, [x10]
11 ldr x3, [x10]
13 # CHECK: Iterations: 3
14 # CHECK-NEXT: Instructions: 18
15 # CHECK-NEXT: Total Cycles: 31
16 # CHECK-NEXT: Total uOps: 18
18 # CHECK: Dispatch Width: 2
19 # CHECK-NEXT: uOps Per Cycle: 0.58
20 # CHECK-NEXT: IPC: 0.58
21 # CHECK-NEXT: Block RThroughput: 3.0
23 # CHECK: Instruction Info:
24 # CHECK-NEXT: [1]: #uOps
25 # CHECK-NEXT: [2]: Latency
26 # CHECK-NEXT: [3]: RThroughput
27 # CHECK-NEXT: [4]: MayLoad
28 # CHECK-NEXT: [5]: MayStore
29 # CHECK-NEXT: [6]: HasSideEffects (U)
31 # CHECK: [1] [2] [3] [4] [5] [6] Instructions:
32 # CHECK-NEXT: 1 1 1.00 * str x1, [x10]
33 # CHECK-NEXT: 1 1 1.00 * str x1, [x10]
34 # CHECK-NEXT: 1 3 1.00 * ldr x2, [x10]
35 # CHECK-NEXT: 1 1 1.00 * * U nop
36 # CHECK-NEXT: 1 3 1.00 * ldr x2, [x10]
37 # CHECK-NEXT: 1 3 1.00 * ldr x3, [x10]
39 # CHECK: Resources:
40 # CHECK-NEXT: [0.0] - CortexA55UnitALU
41 # CHECK-NEXT: [0.1] - CortexA55UnitALU
42 # CHECK-NEXT: [1] - CortexA55UnitB
43 # CHECK-NEXT: [2] - CortexA55UnitDiv
44 # CHECK-NEXT: [3.0] - CortexA55UnitFPALU
45 # CHECK-NEXT: [3.1] - CortexA55UnitFPALU
46 # CHECK-NEXT: [4] - CortexA55UnitFPDIV
47 # CHECK-NEXT: [5.0] - CortexA55UnitFPMAC
48 # CHECK-NEXT: [5.1] - CortexA55UnitFPMAC
49 # CHECK-NEXT: [6] - CortexA55UnitLd
50 # CHECK-NEXT: [7] - CortexA55UnitMAC
51 # CHECK-NEXT: [8] - CortexA55UnitSt
53 # CHECK: Resource pressure per iteration:
54 # CHECK-NEXT: [0.0] [0.1] [1] [2] [3.0] [3.1] [4] [5.0] [5.1] [6] [7] [8]
55 # CHECK-NEXT: - - 1.00 - - - - - - 3.00 - 2.00
57 # CHECK: Resource pressure by instruction:
58 # CHECK-NEXT: [0.0] [0.1] [1] [2] [3.0] [3.1] [4] [5.0] [5.1] [6] [7] [8] Instructions:
59 # CHECK-NEXT: - - - - - - - - - - - 1.00 str x1, [x10]
60 # CHECK-NEXT: - - - - - - - - - - - 1.00 str x1, [x10]
61 # CHECK-NEXT: - - - - - - - - - 1.00 - - ldr x2, [x10]
62 # CHECK-NEXT: - - 1.00 - - - - - - - - - nop
63 # CHECK-NEXT: - - - - - - - - - 1.00 - - ldr x2, [x10]
64 # CHECK-NEXT: - - - - - - - - - 1.00 - - ldr x3, [x10]
66 # CHECK: Timeline view:
67 # CHECK-NEXT: 0123456789 0
68 # CHECK-NEXT: Index 0123456789 0123456789
70 # CHECK: [0,0] DE . . . . . . str x1, [x10]
71 # CHECK-NEXT: [0,1] .DE . . . . . . str x1, [x10]
72 # CHECK-NEXT: [0,2] . DeeE . . . . . ldr x2, [x10]
73 # CHECK-NEXT: [0,3] . DE . . . . . nop
74 # CHECK-NEXT: [0,4] . .DeeE. . . . . ldr x2, [x10]
75 # CHECK-NEXT: [0,5] . . DeeE . . . . ldr x3, [x10]
76 # CHECK-NEXT: [1,0] . . DE . . . . str x1, [x10]
77 # CHECK-NEXT: [1,1] . . .DE . . . . str x1, [x10]
78 # CHECK-NEXT: [1,2] . . . DeeE . . . ldr x2, [x10]
79 # CHECK-NEXT: [1,3] . . . DE . . . nop
80 # CHECK-NEXT: [1,4] . . . .DeeE. . . ldr x2, [x10]
81 # CHECK-NEXT: [1,5] . . . . DeeE . . ldr x3, [x10]
82 # CHECK-NEXT: [2,0] . . . . DE . . str x1, [x10]
83 # CHECK-NEXT: [2,1] . . . . .DE . . str x1, [x10]
84 # CHECK-NEXT: [2,2] . . . . . DeeE . ldr x2, [x10]
85 # CHECK-NEXT: [2,3] . . . . . DE . nop
86 # CHECK-NEXT: [2,4] . . . . . .DeeE. ldr x2, [x10]
87 # CHECK-NEXT: [2,5] . . . . . . DeeE ldr x3, [x10]
89 # CHECK: Average Wait times (based on the timeline view):
90 # CHECK-NEXT: [0]: Executions
91 # CHECK-NEXT: [1]: Average time spent waiting in a scheduler's queue
92 # CHECK-NEXT: [2]: Average time spent waiting in a scheduler's queue while ready
93 # CHECK-NEXT: [3]: Average time elapsed from WB until retire stage
95 # CHECK: [0] [1] [2] [3]
96 # CHECK-NEXT: 0. 3 0.0 0.0 0.0 str x1, [x10]
97 # CHECK-NEXT: 1. 3 0.0 0.0 0.0 str x1, [x10]
98 # CHECK-NEXT: 2. 3 0.0 0.0 0.0 ldr x2, [x10]
99 # CHECK-NEXT: 3. 3 0.0 0.0 0.0 nop
100 # CHECK-NEXT: 4. 3 0.0 0.0 0.0 ldr x2, [x10]
101 # CHECK-NEXT: 5. 3 0.0 0.0 0.0 ldr x3, [x10]
102 # CHECK-NEXT: 3 0.0 0.0 0.0 <total>