Work in progress: this is a wiki post, everyone is welcome to modify it directly
See the previous published edition.
Welcome to the twenty-eight issue of the MLIR (bi)Weekly, a newsletter covering developments in MLIR, and related projects in the ecosystem. MLIR (bi)Weekly is brought to you by a collective effort of contributors, we welcome your contributions!
- The Vector dialect now has
vector.storeops. These operations model contiguous vector loads and stores from/to memory and allow representing vector loads and stores that couldn’t be represented using
std.storeops. They will facilitate the progressive lowering of both Affine vector loads/stores and Vector transfer reads/writes. D96185 [mlir][Vector] Introduce ‘vector.load’ and ‘vector.store’ ops (llvm.org)
In the Ecosystem
Flang, the LLVM Fortran Compiler
IREE : An Experimental MLIR Execution Environment
mlir-npcomp: Prototype for compiling numpy programs
- GlobalizeObjectGraph tranformation to convert object graphs to a more easily analyzable form
- add ability to annotate TorchScript classes. Currently used for public/private annotations, but eventually will be important for providing other information to the compiler.
- properly model “derefinement” which is an impedance mismatch between TorchScript and MLIR.
- BERT imports and lowers. Resnet in progress.
TFRT: A New TensorFlow Runtime
CIRCT : Circuit IR Compilers and Tools aka ‘MLIR for hardware’
- ESI’s Co-simulation feature is v1 feature complete. [Demonstration PR]
Slides are online for most talks, here are the MLIR ones:
- Moving LLVM’s code generator to MLIR framework
- Classical Loop Nest Transformation Framework on MLIR
- LTO and Data Layout Optimisations in MLIR
- COMET: Domain Specific Compilation for Heterogenous Targets