Module simd::spirv

std::simd, SPIR-V convenience layer

Pure naming/documentation layer over std/simd/simd.h's portable vec<T,N>-based types and functions — LLVM's (experimental) SPIR-V backend lowers vec<T,N> arithmetic directly to real SPIR-V vector instructions (OpTypeVector + OpFAdd/OpFMul/...) from the portable source; verified directly against this compiler's output (--target spirv64-unknown-unknown), not merely assumed from the other backends.

Important caveat, unlike the other four std::simd targets: SPIR-V is a GPU/compute IR (Vulkan, OpenCL), not a CPU ISA — a SPIR-V module is a kernel executed by a GPU driver, with no host libc, so functions like simd_load_f32x4/simd_store_f32x4 in the portable core (which read/write an arbitrary host pointer) and anything using variadic host calls (e.g. printf, used in this library's own doc examples) are meaningful on the CPU targets but not in an actual SPIR-V kernel — a real SPIR-V compute shader gets its data through bound buffer arguments and writes results the same way, per whatever host API (Vulkan/OpenCL) launched it. The arithmetic types and operators below are exactly what a numeric SPIR-V kernel body would use.