Intel has updated its Compute Runtime to include support for the Portable Instruction Set Architecture (PISA). This new virtual ISA and programming model acts as an abstraction layer between compiler output and native GPU machine code.
The primary goal of PISA is to simplify software development by providing a common compiler target. Developers can write generalized code once and finalize it for multiple compatible Intel GPU architectures, reducing the need for architecture-specific code generation.
While PISA allows workloads to run across different GPU generations, compatibility remains dependent on the specific target architecture and the features utilized by the program. It does not guarantee that every workload will run on every Intel GPU.
PISA serves as a new addition to Intel's existing compiler stack. It complements the company's long-standing vISA, which has historically functioned as an intermediate representation for Intel GPU compilers.
