Grid
Grid copied to clipboard
ARM NEON is broken
Reminder to self to fix this.
Hi @paboyle I fixed it in 18ce23a, this was a trivial change. It passes the SIMD unit test. #431 still needs to be merged for macOS + M1 compilation, although I am not sure how it affects other platforms.
Is there possibly a description available of the compilation error this problem originally generated? I am currently attempting to run the latest version of Grid on the ARM-GPU machine at UoLeicester and I am encountering nvidia compilation errors reported from arm_neon.h (below). I am wondering if it possibly related to issue #430 ? Thanks. .... CXX qcd/action/fermion/instantiation/WilsonFermionInstantiation.o CXX qcd/action/fermion/instantiation/WilsonKernelsInstantiation.o /opt/nvidia/hpc_sdk/Linux_aarch64/23.5/compilers/include/arm_neon.h(1743): error: identifier "__builtin_neon_vabdq_v" is undefined __ret = (uint8x16_t) __builtin_neon_vabdq_v((int8x16_t)__p0, (int8x16_t)__p1, 48); ^ etc ..