ispc Files

ISPC release featuring full backend support for AVX10.2 and APX on the avx10.2dmr and avx10.2nvl targets, new parallel bit deposit/extract functions in the standard library, and several bug fixes. Based on a patched LLVM 23.1.2.

AVX10.2 and APX Support:

• ISPC is now based on the LLVM 23.1.2, which provides full backend support for AVX10.2 and APX (Advanced Performance Extensions). Code generated for the avx10.2dmr (Diamond Rapids) and avx10.2nvl (Nova Lake) targets now benefits from complete AVX10.2 and APX code generation.

Standard Library:

• Parallel bit deposit and extract functions pdep32(), pext32(), pdep64(), and pext64() have been added to the standard library, with uniform and varying overloads operating on uint32 and uint64 values (#1243). pdep deposits the low bits of a value into the set-bit positions of a mask, and pext gathers the bits selected by a mask into the low bits of the result. They use native BMI2 instructions where available and are specialized for compile-time constant masks, so common patterns such as Morton code computation compile to short shift/mask sequences.

New CPU Support:

• --cpu=znver6 has been added for AMD Zen 6 processors.

Dispatcher:

• As a reminder, the multi-target runtime dispatcher selects the avx10.2dmr, avx10.2nvl, and AMX-capable variants based on CPUID only. It does not check whether the OS has enabled APX or AMX state:
• The AMX-capable variants (avx512spr, avx512gnr, avx10.2dmr) are selected whenever the CPU supports AMX. The application is responsible for checking OS support and requesting AMX permission before executing AMX instructions.
• The avx10.2dmr and avx10.2nvl variants are compiled with APX enabled by default. If the application may run on an OS without APX support, disable APX with --opt=disable-apx. The dispatch functions themselves are always compiled without APX.
• When --target=host detects a Nova Lake or Diamond Rapids CPU on a system where the OS has not enabled APX state, ISPC keeps the detected target but disables APX code generation.

Compiler Switches:

• Dependency files generated with -M/-MF now escape spaces and other Make-special characters in filenames, making them compatible with Make and Ninja (#3743). A new -MQ <target> option specifies a dependency rule target that is quoted for Make, while -MT <target> uses the target verbatim.

ISPC Runtime:

• New ispcrtLoadModuleFromMemory() and ispcrtLoadModuleFromMemoryWithOptions() APIs, along with matching ispcrt::Module constructors, allow loading a GPU module from an in-memory SPIR-V (or zebin) buffer (#3900).

Bug Fixes:

• Fixed a miscompile where -- on a varying unsigned integer became a no-op on AArch64 hosts and triggered an assertion for uint64 (#3882).

• Fixed an AVX-512 miscompile in ISPC 1.31.0 where a masked load was sunk below an aliasing masked store at -O1 and -O2, breaking patterns such as element swaps (#3891).

• Fixed a compiler crash when loading a bool short vector through a reference, for example passing a const bool<3>& argument to a function that takes bool<3> by value (#3893).

• Fixed a compiler assertion failure in gather/scatter index analysis with large constant offsets or shift amounts, for example when a gather index is derived from 64-bit hash computations (#3881).

Build System:

• Updated default LLVM version to the released LLVM 23.1.2.
• Added support for building with LLVM 24 (trunk).

Recommended versions of Runtime Dependencies when targeting GPU:

Linux:

• Intel(R) Graphics Compute Runtime https://github.com/intel/compute-runtime/releases/tag/25.13.33276.16
• Level Zero Loader https://github.com/oneapi-src/level-zero/releases/tag/v1.20.2
• Threading Building Blocks (TBB)

Alternatively, you can use a validated gfx driver stack supporting Intel Arc(TM) available at https://dgpu-docs.intel.com/driver/installation.html

Windows:

• Intel(R) Graphics Windows(R) DCH Drivers 32.0.101.8250 https://www.intel.com/content/www/us/en/download/785597/869290/intel-arc-graphics-windows.html
• Level Zero Loader https://github.com/oneapi-src/level-zero/releases/tag/v1.20.2
• OpenCL(TM) Offline Compiler (OCLOC) https://www.intel.com/content/www/us/en/developer/articles/tool/oneapi-standalone-components.html (this is needed for AoT compilation on Windows only)
• Supported GPU platforms: Intel(R) Arc Graphics, 11th-13th Gen Intel(R) Core processor graphics

Components revisions used in GPU-enabled build:

• https://github.com/KhronosGroup/SPIRV-LLVM-Translator/commit/e65dde2
• https://github.com/intel/vc-intrinsics/commit/e2bb84f
• https://github.com/oneapi-src/level-zero/commit/v1.20.2
• https://github.com/llvm/llvm-project/commit/85ac560 (llvmorg-23.1.2) + patches from llvm_patches folder