tree 6c79fcbfd4b360f8e45c10c272f3abdeb21abf1f
parent 721daa7baedb477808d40a8accdc15b233674f00
author Dillon Sharlet <dsharlet@google.com> 1790981648 -0700
committer XNNPACK Team <xnnpack-github-robot@google.com> 1790981713 -0700

Add AVX-VNNI support for YNNPACK quantized dot product kernels.

- Add arch_flag::avxvnni and CPU feature detection via cpuinfo.
- Configure Bazel and CMake rules to support the x86_avxvnni architecture target.
- Add AVX-VNNI generators for uint8_int2_int32, uint8_int4_int32, and uint8_int8_int32.
- Refactor x86_uint8_int8_int32.py to share a common base class between AVX-VNNI and AVX512-VNNI.
- Register new kernels in kernels.inc and verify against unit tests.

PiperOrigin-RevId: 992584431