In this web content, the author provides a history of the ispc compiler that they developed while working at Intel. They mention the failed Larrabee project, Intel’s attempt at building a high-end GPU, and the issues they faced with vector unit utilization. The author criticizes the auto-vectorization approach used by the Intel compiler team and suggests that a proper programming model is needed to optimize code generation. They highlight the benefits of the SPMD programming model used in GPU programming and express frustration over the lack of understanding from the compiler team. Overall, the content provides insights into the challenges and discussions surrounding vector programming at Intel.
https://pharr.org/matt/blog/2018/04/18/ispc-origins