-
Notifications
You must be signed in to change notification settings - Fork 3.7k
Closed as not planned
Labels
backend:llvmtir:irsrc/tir/ir, and wrappers in python/tvm/tirsrc/tir/ir, and wrappers in python/tvm/tirtype:rfc-trackingRFC progress tracking. Ref: https://github.com/apache/tvm-rfcsRFC progress tracking. Ref: https://github.com/apache/tvm-rfcs
Description
This issue is to track progress for SVE upstreaming
- P1.
vscalebuiltin and its support in LLVM codegen - [SVE] Add vscale builtin #16484 - P2. Support for scalable lanes in
RampandBroadcastnodes and associated changes inruntime::DataType- [SVE] Change the dtype of Ramp and Broadcast lanes to PrimExpr #16523 - P3. Support string representation of scalable datatype - [SVE] Add support for scalable data type strings #16612
- P4. LLVM codegen support for scalable vectors - [SVE] Add codegen support for scalable buffer accesses #16696
- P5. Support for creating scalable vectors from
LoopVectorizer- [SVE] Support scalable vectors in LoopVectorizer #16782 - P6. Splitting by
vscaledependent expression intir::split- [SVE] Support splitting by vscale intir::splitandte::split#16862 - P7. Splitting by
vscaledependent expression inte::split- [SVE] Support splitting by vscale intir::splitandte::split#16862 - P8. Predication support in
LoopVectorizerand LLVM codegen- P8a. get_active_lane_mask intrinsic - [SVE] Add get_active_lane_mask builtin #16965
- P8b. Representation and creation of buffer-level predicates - [SVE] Add support for representing and creating buffer-level predicates #16966
- P9. Gather load and scatter store support
- P10. SVE specific conv2d schedule - [SVE][TOPI] Add conv2d NHWC hybrid SVE schedule for
arm_cpu#16899
Hzfengsy and lhutton1
Metadata
Metadata
Assignees
Labels
backend:llvmtir:irsrc/tir/ir, and wrappers in python/tvm/tirsrc/tir/ir, and wrappers in python/tvm/tirtype:rfc-trackingRFC progress tracking. Ref: https://github.com/apache/tvm-rfcsRFC progress tracking. Ref: https://github.com/apache/tvm-rfcs