diff --git a/lib/AlignedAllocator.h b/lib/AlignedAllocator.h index fa001adc..a8b9c53b 100644 --- a/lib/AlignedAllocator.h +++ b/lib/AlignedAllocator.h @@ -130,8 +130,12 @@ public: #ifdef GRID_COMMS_SHMEM pointer allocate(size_type __n, const void* _p= 0) - { + { +#ifdef CRAY _Tp *ptr = (_Tp *) shmem_align(__n*sizeof(_Tp),64); +#else + _Tp *ptr = (_Tp *) shmem_align(64,__n*sizeof(_Tp)); +#endif #ifdef PARANOID_SYMMETRIC_HEAP static void * bcast; static long psync[_SHMEM_REDUCE_SYNC_SIZE]; diff --git a/lib/simd/Grid_avx512.h b/lib/simd/Grid_avx512.h index 45d6d988..62789462 100644 --- a/lib/simd/Grid_avx512.h +++ b/lib/simd/Grid_avx512.h @@ -371,8 +371,14 @@ namespace Optimization { // Some Template specialization // Hack for CLANG until mm512_reduce_add_ps etc... are implemented in GCC and Clang releases +<<<<<<< HEAD #define GNU_CLANG_COMPILER #ifdef GNU_CLANG_COMPILER +======= + +#ifndef __INTEL_COMPILER +#warning "Slow reduction due to incomplete reduce intrinsics" +>>>>>>> develop //Complex float Reduce template<> inline Grid::ComplexF Reduce::operator()(__m512 in){