Tile IR

E435224

Tile IR is an intermediate representation used within the PlaidML machine learning compiler to express and optimize tensor computations across diverse hardware backends.

All labels observed (1)

Label Occurrences
Tile IR canonical 1

How this entity was disambiguated

Statements (39)

Predicate Object
instanceOf compiler IR
intermediate representation
designedFor portability across CPUs
portability across GPUs
portability across specialized accelerators
domain machine learning
enables automatic differentiation support in PlaidML
fusion of tensor operations
hardware-agnostic optimization
lowering to hardware-specific code
memory layout transformations
scheduling optimizations
goal enable cross-platform ML deployment
separate algorithm specification from hardware mapping
hasFeature affine-style indexing of tensors
explicit shape information
loop-based representation of computations
static analysis of tensor shapes
support for broadcasting semantics
support for convolutions
support for elementwise operations
support for matrix multiplications
support for reductions
tensor-centric operations
implementedInContextOf Vertex.AI PlaidML project NERFINISHED
optimizationLevel mid-level IR
partOf PlaidML compiler pipeline
purpose express tensor computations
optimize tensor computations
relatedTo PlaidML Tile language NERFINISHED
supports diverse hardware backends
supportsBackend CPU backends
CUDA-based devices
OpenCL-based devices
usedBy PlaidML code generator NERFINISHED
PlaidML optimizer NERFINISHED
usedFor deep learning workloads
neural network computation graphs
usedIn PlaidML NERFINISHED

How these facts were elicited

Referenced by (1)

Full triples — surface form annotated when it differs from this entity's canonical label.

PlaidML hasComponent Tile IR