Repository navigation
Tags
Tags
MIPP v2.2.0: Half-Vectors, LMUL-Changing Functions & API Cleanup - SVE and SVE2 are now supported (fixed vector length, native svbool_t masks, half-vectors), validated in CI on armv9-sve and armv9-sve2. - Half-vectors (d2 / MIPP_LDIV_2) on NEON, SVE and SSE; new low, high, combine, wcvt and ncvt functions (native masked versions on RVV). - ISA list, mipp::ISA enum and mipp_info() derived from generator/simd_ext/. - Native SSE max/min, float<->int cvt (SSE, AVX-512, RVV), AVX-512 unsigned masked load/store, native RVV loadu/storeu/masked memory ops. - gcc-15 for the SVE CI jobs (gcc-14 ICE); gather/scatter tests. - Fixes: integer pow, SVE testz, C prototypes (-Wstrict-prototypes), scalar MIPP_LDIV_2/64BIT/BW. - BREAKING: toreg/tomsk/blend -> torvd/torvm/select, testz_2 -> none, DSL mk: -> v:, pow2(r0, v0) -> pow2(r0), MIPP_ALIGNED_LOADS removed, category converts -> convert. - Documentation and API Explorer rebuilt.
MIPP v2.1.0: Fine-Grained Headers, Object API Polish & API Explorer U… …pgrade - Fine-grained modular headers (--granularity fine|coarse) for atomic includes, eliminating compilation time and compiler memory bloat. - Streamlined C++ Object API (Rvd/Rvm) with multi-LMUL support, operator overloads, simplified masked syntax, and Object/Functional interoperability. - Renamed enum MKIND to VARIANT (U, M, Z, S). - Major MIPP API Explorer upgrade with Compiler Explorer (Godbolt) integration, dual-view algorithm modal, and standalone Flat C99 generator. - Expanded ISA support and fixes for AVX-512, SVE/SVE2, and masked math kernels. - Automated fine-grained validation in GitLab CI pipeline.