On 9/15/24 23:11, Andreas Beckmann wrote:
not a regression in nvidia-nccl, the version in testing nowadays ftbfs
similarily
This is caused by glibc 2.38 which enhanced
/usr/include/aarch64-linux-gnu/bits/math-vector.h
with
...
#if __GNUC_PREREQ(9, 0)
# define __ADVSIMD_VEC_MATH_SUPPORTED
typedef __Float32x4_t __f32x4_t;
typedef __Float64x2_t __f64x2_t;
#elif __glibc_clang_prereq(8, 0)
# define __ADVSIMD_VEC_MATH_SUPPORTED
typedef __attribute__ ((__neon_vector_type__ (4))) float __f32x4_t;
typedef __attribute__ ((__neon_vector_type__ (2))) double __f64x2_t;
#endif
#if __GNUC_PREREQ(10, 0) || __glibc_clang_prereq(11, 0)
# define __SVE_VEC_MATH_SUPPORTED
typedef __SVFloat32_t __sv_f32_t;
typedef __SVFloat64_t __sv_f64_t;
typedef __SVBool_t __sv_bool_t;
#endif
...
This got temporarily disabled for 2.38 (only).
It seems newer toolkit versions "support" these types, so for now we
only need to use the old math-vector.h when compiling stuff with old nvcc.
Andreas
