Workshop on MLIR ecosystem. The workshop format is similar to that of the LLVM meeting itself with a mix of technical talks and roundtables, with a focus on the MLIR ecosystem. We invite technical talks on MLIR components and use cases as well as downstream projects using or inspired by MLIR. Presentations of work-in-progress, radical design proposals and other wild ideas are encouraged! We are open to considering unconventional session types. This will be an in-person only workshop with focus on active discussions, the sessions will not be recorded.

Pre-registration required.

Agenda:

9:00 9:05 Opening 0:05
9:05 9:30 A Listener-based Approach to Discardable Attribute Conversion/Propagation: Matthias Springer (NVIDIA) 0:25
9:30 10:00 Control flow? In my pass pipeline?: Ivan Butygin (AMD) 0:30
10:00 10:30 BREAK - FOYER 0:30
10:30 11:00 Towards Distributing Workloads in MLIR: Frank Schlimbach 0:30
11:00 11:30 Tensor Operator Set Architecture (TOSA): Udaya Ranga (Arm Ltd), Luke Hutton (Arm Ltd), Peng Sun (Arm Ltd) 0:30
11:30 12:00 Exploring different representations for computation primitives in ML models: Mahesh Ravishankar (AMD) 0:30
12:00 13:00 LUNCH 1:00
13:00 13:30 Data Layout & Target Info-parameterized transforms, schedules & cost models: Rolf Morel (Intel Labs) 0:30
13:30 14:00 Round Tables 0:30
14:00 14:30 Round Tables 0:30
14:30 15:00 BREAK - FOYER 0:30
15:00 15:15 Closing 0:15