A compiler for tensor computations on GPUs and other devices, supporting the OpenCL, Level Zero, and SYCL runtime.
Introduction • Documentation • License
The Tiny Tensor Compiler compiles programs written in a domain-specific tensor language to OpenCL-C and provides methods to run these programs on GPU devices and other devices via OpenCL, Level Zero, or SYCL.
The tensor language allows writing device-agnostic tensor programs and a large class of tensor operations can be implemented in terms of loop constructs, subviews, and BLAS-like instructions (e.g. GEMM, GER, or AXPBY). High performance is achieved by providing highly optimized BLAS primitives, in particular through a GEMM code generator, that can be specialized for small GEMMs or uncommon shapes.
The online documentation is available at https://intel.github.io/tiny-tensor-compiler/
