Skip to content

Commit e6dae2e

Browse files
Ceng23333Ceng23333
authored
Issue/506 实现TensorMetaData析构方法
1 parent df8d440 commit e6dae2e

File tree

12 files changed

+2073
-7
lines changed

12 files changed

+2073
-7
lines changed

‎include/infinicore/tensor.hpp‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -27,6 +27,7 @@ struct TensorMetaData {
2727
infiniopTensorDescriptor_t desc;
2828

2929
TensorMetaData(const Shape &shape, const Strides &strides, const DataType &dtype);
30+
~TensorMetaData();
3031
};
3132

3233
struct TensorData {

‎src/infinicore-test/README.md‎

Lines changed: 277 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,277 @@
1+
# InfiniCore Memory Management Test Suite
2+
3+
This test suite provides comprehensive testing for the InfiniCore memory management system, focusing on the critical issues identified in the memory management architecture analysis.
4+
5+
## Overview
6+
7+
The test suite includes six main test categories:
8+
9+
1. **Basic Memory Tests** - Basic allocation, deallocation, and memory operations
10+
2. **Concurrency Tests** - Thread safety and concurrent access testing
11+
3. **Exception Safety Tests** - Exception handling and safety testing
12+
4. **Memory Leak Tests** - Memory leak detection and prevention
13+
5. **Performance Tests** - Performance benchmarks and optimization validation
14+
6. **Stress Tests** - High-load stress testing and edge cases
15+
16+
## Building
17+
18+
### Using XMake (if integrated with main build)
19+
```bash
20+
# From InfiniCore root directory
21+
xmake build infinicore-test
22+
```
23+
24+
## Running Tests
25+
26+
### Run All Tests
27+
```bash
28+
./infinicore-test
29+
```
30+
31+
### Run Specific Test Categories
32+
```bash
33+
# Basic memory tests
34+
./infinicore-test --test basic
35+
36+
# Concurrency tests
37+
./infinicore-test --test concurrency
38+
39+
# Exception safety tests
40+
./infinicore-test --test exception
41+
42+
# Memory leak tests
43+
./infinicore-test --test leak
44+
45+
# Performance tests
46+
./infinicore-test --test performance
47+
48+
# Stress tests
49+
./infinicore-test --test stress
50+
```
51+
52+
### Run with Specific Device
53+
```bash
54+
# Run on CPU
55+
./infinicore-test --cpu
56+
57+
# Run on NVIDIA GPU
58+
./infinicore-test --nvidia
59+
60+
# Run on other devices
61+
./infinicore-test --cambricon
62+
./infinicore-test --ascend
63+
./infinicore-test --metax
64+
./infinicore-test --moore
65+
./infinicore-test --iluvatar
66+
./infinicore-test --kunlun
67+
./infinicore-test --hygon
68+
```
69+
70+
### Customize Test Parameters
71+
```bash
72+
# Run with custom thread count
73+
./infinicore-test --threads 8
74+
75+
# Run with custom iteration count
76+
./infinicore-test --iterations 5000
77+
78+
# Combine options
79+
./infinicore-test --nvidia --test concurrency --threads 16 --iterations 2000
80+
```
81+
82+
## Test Categories
83+
84+
### 1. Basic Memory Tests
85+
Tests fundamental memory operations:
86+
- Memory allocation and deallocation
87+
- Memory size and device properties
88+
- Memory read/write operations
89+
- Pinned memory allocation
90+
- Memory data integrity
91+
92+
### 2. Concurrency Tests
93+
Tests thread safety and concurrent access:
94+
- **Concurrent Allocations**: Multiple threads allocating memory simultaneously
95+
- **Concurrent Device Switching**: Multiple threads switching device contexts
96+
- **Memory Allocation Race**: Race condition testing for memory operations
97+
98+
### 3. Exception Safety Tests
99+
Tests exception handling and safety:
100+
- **Allocation Failure**: Tests behavior when allocation fails
101+
- **Deallocation Exception**: Tests exception safety during deallocation
102+
- **Context Switch Exception**: Tests exception handling during device switching
103+
104+
### 4. Memory Leak Tests
105+
Tests memory leak detection and prevention:
106+
- **Basic Leak Detection**: Basic memory leak detection
107+
- **Cross-Device Leak Detection**: Memory leaks in cross-device scenarios
108+
- **Exception Leak Detection**: Memory leaks during exception handling
109+
110+
### 5. Performance Tests
111+
Tests performance and benchmarks:
112+
- **Allocation Performance**: Memory allocation speed benchmarks
113+
- **Concurrent Performance**: Performance under concurrent load
114+
- **Memory Copy Performance**: Memory copy bandwidth tests
115+
116+
### 6. Stress Tests
117+
Tests high-load scenarios and edge cases:
118+
- **High Frequency Allocations**: Rapid allocation/deallocation cycles
119+
- **Large Memory Allocations**: Large memory block allocation
120+
- **Cross-Device Stress**: Stress testing across multiple devices
121+
122+
## Expected Results
123+
124+
### Critical Issues to Watch For
125+
126+
The tests are designed to detect the critical issues identified in the memory management analysis:
127+
128+
1. **Thread Safety Violations**
129+
- Race conditions in concurrent allocations
130+
- Inconsistent device context switching
131+
- Global state corruption
132+
133+
2. **Memory Leaks**
134+
- Unfreed memory after deallocation
135+
- Cross-device memory not properly cleaned up
136+
- Exception-related memory leaks
137+
138+
3. **Exception Safety Issues**
139+
- Exceptions during allocation causing resource leaks
140+
- Exceptions in destructors causing `std::terminate`
141+
- Incomplete cleanup on exceptions
142+
143+
4. **Performance Issues**
144+
- Slow allocation/deallocation performance
145+
- Poor concurrent performance
146+
- Inefficient memory copy operations
147+
148+
### Performance Thresholds
149+
150+
The tests include performance thresholds:
151+
152+
- **Allocation Performance**: < 100μs per allocation
153+
- **Concurrent Performance**: < 200μs per allocation under load
154+
- **Memory Bandwidth**: > 100 MB/s for memory copies
155+
156+
## Test Output
157+
158+
### Successful Test Run
159+
```
160+
==============================================
161+
InfiniCore Memory Management Test Suite
162+
==============================================
163+
Device: 0
164+
Threads: 4
165+
Iterations: 1000
166+
==============================================
167+
168+
[SUITE] Running: BasicMemoryTest
169+
[TEST] Starting: BasicMemoryTest
170+
[TEST] PASSED: BasicMemoryTest (Duration: 1234μs)
171+
172+
[SUITE] Running: ConcurrencyTest
173+
[TEST] Starting: ConcurrencyTest
174+
[TEST] PASSED: ConcurrencyTest (Duration: 5678μs)
175+
176+
...
177+
178+
==============================================
179+
Test Summary
180+
==============================================
181+
Total Tests: 6
182+
Passed: 6
183+
Failed: 0
184+
Total Time: 12345μs
185+
==============================================
186+
187+
✅ All tests passed!
188+
```
189+
190+
### Failed Test Run
191+
```
192+
[TEST] FAILED: ConcurrencyTest - Concurrent allocation test failed: expected 8000 successes, got 7995 successes and 5 failures
193+
194+
==============================================
195+
Final Results
196+
==============================================
197+
Total Tests: 6
198+
Passed: 5
199+
Failed: 1
200+
==============================================
201+
202+
❌ Some tests failed. Please review the output above.
203+
```
204+
205+
## Debugging Failed Tests
206+
207+
### Common Issues and Solutions
208+
209+
1. **Thread Safety Failures**
210+
- Check for race conditions in global state access
211+
- Verify proper synchronization in allocators
212+
- Review device context switching logic
213+
214+
2. **Memory Leak Failures**
215+
- Check deallocation logic in allocators
216+
- Verify cross-device cleanup mechanisms
217+
- Review exception safety in destructors
218+
219+
3. **Performance Failures**
220+
- Profile allocation/deallocation paths
221+
- Check for unnecessary context switching
222+
- Review memory copy implementations
223+
224+
4. **Exception Safety Failures**
225+
- Verify no-throw guarantees in destructors
226+
- Check exception handling in allocation paths
227+
- Review resource cleanup on exceptions
228+
229+
## Integration with CI/CD
230+
231+
### GitHub Actions Example
232+
```yaml
233+
- name: Run Memory Tests
234+
run: |
235+
cd src/infinicore-test
236+
mkdir build && cd build
237+
cmake ..
238+
make
239+
./infinicore-test --test all
240+
```
241+
242+
### Custom Test Targets
243+
```bash
244+
# Run specific test categories
245+
make test-memory-basic
246+
make test-memory-concurrency
247+
make test-memory-exception
248+
make test-memory-leak
249+
make test-memory-performance
250+
make test-memory-stress
251+
make test-memory-all
252+
```
253+
254+
## Contributing
255+
256+
When adding new tests:
257+
258+
1. Follow the existing test framework pattern
259+
2. Add appropriate error messages and logging
260+
3. Include performance thresholds where applicable
261+
4. Test both success and failure scenarios
262+
5. Update this README with new test descriptions
263+
264+
## Dependencies
265+
266+
- InfiniCore library (infinicore, infiniop, infinirt, infiniccl)
267+
- C++17 compatible compiler
268+
- Threading library (pthread on Linux)
269+
- CMake 3.16+ (for CMake build)
270+
271+
## Notes
272+
273+
- Tests are designed to be deterministic where possible
274+
- Some tests may have timing dependencies
275+
- Performance tests may vary based on system load
276+
- Memory leak detection is basic and may not catch all leaks
277+
- Tests assume proper InfiniCore initialization

0 commit comments

Comments
 (0)