Force riscv builds onto the scalar path

[carlosqwqqwq]

Why

Vc already has a portable Scalar implementation, but simulated or cross-target riscv64 validation from an x86_64 host compiler can still inherit host SSE/AVX preprocessor state. That leads the headers to select x86 implementations even when the build system has already chosen the generic support_dummy.cpp path. This patch makes the Scalar fallback explicit for RISC-V and prevents the host x86 ABI from leaking into simulated riscv64 builds.

What changed

• Update CMakeLists.txt so riscv* is treated as an explicit Scalar-only target instead of a generic unknown warning path.
• Update Vc/global.h so __riscv forces Vc_IMPL_Scalar before any host SSE/AVX auto-detection can run.
• Update Vc/common/vectorabi.h so simulated or cross-target RISC-V builds do not inherit the host __x86_64__ ABI deduction.

Verification

• Ran native CMake configuration with Ninja using Vc_ENABLE_INSTALL=OFF.
• Built the native static library successfully with cmake --build build-codex-native-061 --parallel 4.
• Ran simulated riscv64 CMake configuration with CMAKE_SYSTEM_NAME=Linux, CMAKE_SYSTEM_PROCESSOR=riscv64, CMAKE_TRY_COMPILE_TARGET_TYPE=STATIC_LIBRARY, forced __riscv=1 / __riscv_xlen=64, and Vc_ENABLE_INSTALL=OFF.
• Built the simulated riscv64 static library successfully with cmake --build build-codex-riscv-061 --parallel 4.
• Confirmed from the simulated build.ninja that the build only compiles src/support_dummy.cpp and src/const.cpp, with no support_x86.cpp, cpuid.cpp, trigonometric_SSE2.cpp, sse_sorthelper, or avx_sorthelper sources and no -msse* / -mavx* / -mfma / -mbmi flags.
• Preprocessed Vc/global.h under forced __riscv=1 / __riscv_xlen=64 and confirmed only Vc_IMPL_Scalar=1 is defined.
• Compiled a smoke-test translation unit under forced __riscv=1 / __riscv_xlen=64 and confirmed Vc::float_v::Size == 1 and Vc::VectorAbi::DeduceCompatible<float>::type resolves to Scalar.

Notes

This is a conservative portability patch. It does not implement a RISC-V SIMD or RVV backend. The goal is to ensure that RISC-V targets reliably select the existing Scalar implementation instead of inheriting host x86 vector assumptions during simulated or cross-target builds.

[Axel-Naumann]

We are planning to archive this repo. Please reach out to axel@cern.ch if you want to fork and maintain.