Support ACEv1 outer product instructions
Adds support for ACEv1 outer product instructions, using a naming convention similar to VNNI dot product instructions.
This commit introduces support for ACEv1 outer product instructions by adding new intrinsics, builtin types, and expanding existing ones. The naming convention for these instructions mirrors that of VNNI dot product instructions found in AVX/AVX512/AVX10, ensuring consistency for developers working with these advanced instruction sets. New tests are included to verify the functionality.
In Details
Introduces ACEv1 outer product intrinsics, mapping them to UNSPECV_TOP2BF16PS, UNSPECV_TOP4BSSD, etc. in sse.md. Leverages existing patterns for dot product instructions, adapting them for the new ACEv1 outer product operations. New builtin types and expanded builtins in i386-builtin-types.def and i386-builtin.def facilitate their use via C/C++ intrinsics. Testing covers compilation and execution of these new instructions.
- ACEv1
- AI Compute Extensions version 1, a new x86 instruction set architecture designed for AI and compute-intensive workloads.
- intrinsics
- Special functions that allow C/C++ code to directly invoke specific machine instructions, often providing performance benefits over traditional function calls.
- VNNI
- Vector Neural Network Instructions, a set of x86 instructions designed to accelerate deep learning inference by performing operations like dot products on packed data.
- AVX/AVX512/AVX10
- Advanced Vector Extensions, a set of x86 instruction set extensions that increase the width of CPU registers for vectorized processing. AVX512 and AVX10 offer even wider registers and more advanced capabilities.