AOCL-BLAS can now be selected as a vendor for the existing GGML_BLAS backend

[pradeeptrgit]

AOCL-BLAS can now be selected as a vendor for the existing GGML_BLAS backend.

For inference on AMD CPUs, the ZenDNN backend remains the recommended path. This option is for people who want the generic BLAS backend linked against AOCL-BLAS. It applies to large prompt GEMMs. F32 weights go to cblas_sgemm; other types are converted to F32 first. --list-devices reports the device as BLAS: AOCL-BLAS.

Build and run notes, including the amd-libs.cfg setup and BLAS_INCLUDE_DIRS, are in the docs:

• https://github.com/ggml-org/llama.cpp/blob/master/docs/backend/AOCL.md
• https://github.com/ggml-org/llama.cpp/blob/master/docs/build.md#amd-aocl-blas

Change: #29640