Intel ISPC 1.26 Compiler Delivers Improved ARM Support
Intel's ISPC project as the Implicit SPMD Program Compiler as this C language variant for Single Program, Multiple Data programming on CPUs and GPUs is out with a new release.
Intel ISPC focuses on C-based SPMD programming traditionally for their range of CPUs and GPUs and being able to efficiently exploit features like AVX2 and AVX-512. There's been ARM CPU support too while for this week's ISPC 1.26 release the ARM processor support is much better positioned. In fact, better ARM support is the primary improvement with Intel ISPC 1.26.
ISPC 1.26 when targeting ARM with "--arch=arm" will now map to ARMv8 rather than ARMv7. There are also new CPU definitions for the Arm Cortex A78, A55, A510, A520, A9, and A15. Plus support for newer Apple Silicon devices too.
Intel ISPC 1.26 for ARM also introduces new double-pumped targets of neon-i16x16 and neon-i8x32, dot product operations are now supported using native ARM instructions, and the ARMv8 performance is said to be around 13% faster than prior ISPC releases.
Intel ISPC 1.26 also introduces new generic targets for different CPU architectures, which makes it easier adding RISC-V and other LLVM-supported CPU targets. ISPC 1.26 also brings some language enhancements, various code generation improvements, dropping the Xeon Phi Knights Landing target, and various fixes.
Downloads and more details on this week's ISPC 1.26 release via GitHub.
Intel ISPC focuses on C-based SPMD programming traditionally for their range of CPUs and GPUs and being able to efficiently exploit features like AVX2 and AVX-512. There's been ARM CPU support too while for this week's ISPC 1.26 release the ARM processor support is much better positioned. In fact, better ARM support is the primary improvement with Intel ISPC 1.26.
ISPC 1.26 when targeting ARM with "--arch=arm" will now map to ARMv8 rather than ARMv7. There are also new CPU definitions for the Arm Cortex A78, A55, A510, A520, A9, and A15. Plus support for newer Apple Silicon devices too.
Intel ISPC 1.26 for ARM also introduces new double-pumped targets of neon-i16x16 and neon-i8x32, dot product operations are now supported using native ARM instructions, and the ARMv8 performance is said to be around 13% faster than prior ISPC releases.
Intel ISPC 1.26 also introduces new generic targets for different CPU architectures, which makes it easier adding RISC-V and other LLVM-supported CPU targets. ISPC 1.26 also brings some language enhancements, various code generation improvements, dropping the Xeon Phi Knights Landing target, and various fixes.
Downloads and more details on this week's ISPC 1.26 release via GitHub.
4 Comments