Intel’s Implicit SPMD Program Compiler (ISPC) is out with a new feature release in preparing for AVX10.2-enabled processors and bringing other enhancements to this C programming language variant focused on single program, multiple data (SPMD) programming.
ISPC continues to be developed as one of Intel’s routes for SPMD programming across their range of hardware from CPUs to GPUs using C code. With ISPC 1.27 they have introduced support for AVX10.2 and added the new avx10.2-x4, avx10.2-x8, avx10.2-x16, avx10.2-x32, and avx10.2-x64 targets.
ISPC 1.27 also brings full support for element-wise functions on short vectors within its standard library, cross-lane operations for unsigned types, better dot product functionality, and a variety of other enhancements.
On the performance side ISPC 1.27 has improved the performance of masked loads and stores for AVX-512 by an order of magnitude said to be around 10x faster on micro-benchmarks. The packed_store_active2 call on AVX2 is also coming in around 65% faster for int32 and around 45% faster for int64.
Downloads and more details on the Intel ISPC 1.27 release via GitHub.
