| Support Bisheng kernel generator in Mfuse pipeline
| 20 天前 |
| Add constant mfusetotorch converter
| 20 天前 |
| Add split
| 20 天前 |
| mfuse: add symbolic-shape consistency trait for all ops
Enforce two symbolic-shape rules on all mfuse ops via VerifySymbolicShapeTrait:
(1) if any input has symbolic shape, every non-static RankedTensor input and
output must have it; (2) when encoding is present, it must match the tensor
rank and static/dynamic shape. Update Split lit tests.
| 20 天前 |
| Add split
| 20 天前 |
| Fix split create fusedop problem
| 20 天前 |
| Add Reduce post split pattern
| 20 天前 |
| Add split
| 20 天前 |
| Add split
| 20 天前 |
| Add split
| 20 天前 |
| Add Reduce post split pattern
| 20 天前 |
| Skip unsupported reduce fusion cases for dvm
| 20 天前 |
| Add split
| 20 天前 |
| mfusion: add ASAN build option and fix issues
- Add CMake ENABLE_ASAN, build.sh -a on|off (forces Debug), and
setup.py passthrough via ENABLE_ASAN env.
- Avoid use-after-free in MatMul fusion patterns: capture Location
before replaceOp when logging; same for split debug paths.
- Prefer MLOG(DEBUG) over LLVM_DEBUG in FusedOpUtils, Split pass,
FuseOpRebuilder; FuseOpSplitter logs via Node::toString.
- Node::toString uses OpPrintingFlags assumeVerified + useLocalScope.
- Update lit: torch i32 as si32 in MfuseToTorch tests; trim RUN lines
(drop -allow-unregistered-dialect / -mlir-print-ir-after-all).
| 20 天前 |
| Fuse numtotensor and cast or single numtotensor to full
| 20 天前 |
| Add constant mfusetotorch converter
| 20 天前 |
| Add split
| 20 天前 |