diff options
Diffstat (limited to 'numpy')
| -rw-r--r-- | numpy/core/setup_common.py | 1 | ||||
| -rw-r--r-- | numpy/core/src/private/templ_common.h.src | 6 |
2 files changed, 5 insertions, 2 deletions
diff --git a/numpy/core/setup_common.py b/numpy/core/setup_common.py index e51797c03..0b18bc6c6 100644 --- a/numpy/core/setup_common.py +++ b/numpy/core/setup_common.py @@ -118,6 +118,7 @@ OPTIONAL_INTRINSICS = [("__builtin_isnan", '5.'), ("__builtin_bswap32", '5u'), ("__builtin_bswap64", '5u'), ("__builtin_expect", '5, 0'), + ("__builtin_mul_overflow", '5, 5, (int*)5'), ("_mm_load_ps", '(float*)0', "xmmintrin.h"), # SSE ("_mm_load_pd", '(double*)0', "emmintrin.h"), # SSE2 ] diff --git a/numpy/core/src/private/templ_common.h.src b/numpy/core/src/private/templ_common.h.src index 8b71f2c3e..dd6c7bf23 100644 --- a/numpy/core/src/private/templ_common.h.src +++ b/numpy/core/src/private/templ_common.h.src @@ -21,13 +21,14 @@ static NPY_INLINE int npy_mul_with_overflow_@name@(@type@ * r, @type@ a, @type@ b) { +#ifdef HAVE___BUILTIN_MUL_OVERFLOW + return __builtin_mul_overflow(a, b, r); +#else const @type@ half_sz = (((@type@)1 << (sizeof(a) * 8 / 2)) - 1); *r = a * b; /* * avoid expensive division on common no overflow case - * could be improved via compiler intrinsics e.g. via clang - * __builtin_mul_with_overflow, gcc __int128 or cpu overflow flags */ if (NPY_UNLIKELY((a | b) >= half_sz) && a != 0 && b > @MAX@ / a) { @@ -35,6 +36,7 @@ npy_mul_with_overflow_@name@(@type@ * r, @type@ a, @type@ b) } return 0; +#endif } /**end repeat**/ |