master

GginfungSupport Bisheng kernel generator in Mfuse pipeline

cc3d74f9创建于 22 天前历史提交

文件	最后提交记录	最后更新时间
test_convert_fused_subgraph_to_custom_call.mlir	Support Bisheng kernel generator in Mfuse pipeline	22 天前
test_convert_fused_subgraph_to_custom_call_dynamic.mlir	Support Bisheng kernel generator in Mfuse pipeline	22 天前
test_copy_fused_subgraphs.mlir	feat: add fusion region outlining and DVM call lowering Add three-phase pass pipeline for handling fused subgraphs in MFusion: 1. outline-mfuse-fused-subgraphs: Extract mfuse.fused ops into separate functions 2. copy-fused-subgraphs: Clone outlined functions for DVM backend usage 3. convert-dvm-subgraph-to-mfuse-dvm-call: Replace func.call with mfuse.dvm_call Also add MfuseToDvm conversion pass to lower Mfuse ops to DVM dialect. Key changes: - Add mfuse.dvm_call and mfuse.fused operations to MfuseOps.td - Implement outlining passes with proper attribute handling - Add constant lifting and IO semantics (dvm.load/dvm.store) conversion - Update Python inductor pipeline with complete optimization flow - Add comprehensive test coverage for all new passes	22 天前
test_outline_mfuse_fused_subgraphs.mlir	feat: add fusion region outlining and DVM call lowering Add three-phase pass pipeline for handling fused subgraphs in MFusion: 1. outline-mfuse-fused-subgraphs: Extract mfuse.fused ops into separate functions 2. copy-fused-subgraphs: Clone outlined functions for DVM backend usage 3. convert-dvm-subgraph-to-mfuse-dvm-call: Replace func.call with mfuse.dvm_call Also add MfuseToDvm conversion pass to lower Mfuse ops to DVM dialect. Key changes: - Add mfuse.dvm_call and mfuse.fused operations to MfuseOps.td - Implement outlining passes with proper attribute handling - Add constant lifting and IO semantics (dvm.load/dvm.store) conversion - Update Python inductor pipeline with complete optimization flow - Add comprehensive test coverage for all new passes	22 天前