-
Notifications
You must be signed in to change notification settings - Fork 3.6k
Closed
Labels
type:rfc-trackingRFC progress tracking. Ref: https://github.com/apache/tvm-rfcsRFC progress tracking. Ref: https://github.com/apache/tvm-rfcs
Description
RFC PR: apache/tvm-rfcs#37
This issue is to track progress for the Arm(R) Ethos(TM)-U Cascading Scheduler
- P1. Add analysis infrastructure for data dependencies between stripes/operators using affine transforms (StripeConfig/Propagator). [microNPU][1] Add affine analysis structures for the cascader #9458
- P2a. Add CascaderGraph to store graph-level information needed by the cascading algorithms (CascaderGraph, Part, Tensor). [microNPU][2a] Add CascaderGraph for cascading analysis #9469
- P2b. Create CascaderGraph from TE graph (including initial Part support for Ethos-U conv2d and 'inline'). [microNPU][2b] Create CascaderGraphs from TE graphs #9471
- P2c. Add performance heuristics to Ethos-U Parts. [microNPU][2c] Add performance modelling to cascader #9778
- P2d. Add further Ethos-U Parts (depthwise/pooling/binary elementwise). [microNPU][2d] Add more Part matchers to cascader #9785
- P3. Plan generation algorithm and associated data structures (Plan/TensorConfig). [microNPU][3] Plan generation for the cascader #9890
- P4. Proposal generation algorithm and associated data structures (Proposal). [microNPU][4] Add the cascader Proposal generator #9959
- P5. Convert Proposals to TE Schedules. [microNPU][5] Convert Proposals to te.Schedules #10062
- P?. Additional patch to integrate cascader block config selection with Ethos-U codegen. [microNPU] Determine block configs using the cascader #10695
- P6. Integrate with Ethos-U codegen. [microNPU] Integrate the cascader #10862
- E2E tests w/o striping : [microNPU] add E2E tests with cascader wo striping #11410
- E2E tests with striping : [microNPU] enable striping for network tests. #11883
Metadata
Metadata
Assignees
Labels
type:rfc-trackingRFC progress tracking. Ref: https://github.com/apache/tvm-rfcsRFC progress tracking. Ref: https://github.com/apache/tvm-rfcs