Intel Graphics Compiler 2.30.1 is now available for this LLVM/Clang-based compiler stack used by the Compute Runtime on Linux and under Windows is used both for graphics and compute.
IGC 2.30.1 brings ongoing enablement work for Xe3P Nova Lake and Crescent Island hardware along with an assortment of other improvements to this open-source compiler stack. Presumably there will be a new Intel Compute Runtime release out shortly for pairing with the Intel Graphics Compiler 2.30.1 milestone. IGC 2.30.1 highlights include:
– Continued preparations for the Crescent Island AI inference accelerator, including adding support for the HF8 format for half-float 8-bit floating point. HF8 is supported by Xe3 and previously confirmed last year while now is wired up for Crescent Island usage.
– Intel Nova Lake P preparations in the IGC compiler.
– Support for rounding modes in correctly-rounded IEEE div and sqrt.
– An option to disable BF16 BIF compilation in order to make the IGC compiler compatible with upstream Clang. The BF16 support for IGC requires patches not yet upstreamed to Clang. If disabling the support, you lose out on the likes of IGC_SPV_INTEL_bfloat16_arithmetic and IGC_SPV_KHR_bfloat16.
– Various ray-tracing changes.
IGC 2.30.1 can be downloaded from GitHub.
