/ 64-bit load over 32-bit stride
| 403 | |
| 404 | //// 64-bit load over 32-bit stride |
| 405 | NPY_FINLINE npyv_s32 npyv_loadn2_till_s32(const npy_int32 *ptr, npy_intp stride, npy_uintp nlane, |
| 406 | npy_int32 fill_lo, npy_int32 fill_hi) |
| 407 | { |
| 408 | assert(nlane > 0); |
| 409 | if (nlane == 1) { |
| 410 | const __m128i vfill = npyv_set_s32(0, 0, fill_lo, fill_hi); |
| 411 | __m128i a = _mm_castpd_si128( |
| 412 | _mm_loadl_pd(_mm_castsi128_pd(vfill), (double*)ptr) |
| 413 | ); |
| 414 | #if NPY_SIMD_GUARD_PARTIAL_LOAD |
| 415 | volatile __m128i workaround = a; |
| 416 | a = _mm_or_si128(workaround, a); |
| 417 | #endif |
| 418 | return a; |
| 419 | } |
| 420 | return npyv_loadn2_s32(ptr, stride); |
| 421 | } |
| 422 | NPY_FINLINE npyv_s32 npyv_loadn2_tillz_s32(const npy_int32 *ptr, npy_intp stride, npy_uintp nlane) |
| 423 | { |
| 424 | assert(nlane > 0); |
nothing calls this directly
no test coverage detected