Calls to copyBuf in both overloads of DeviceArray::fromHost were passing the wrong lengths. They were using size_bytes() which is larger than the staging buffer if end - begin defines a sub-range.
The call to copyBuf in DeviceArray::rangeToHost was passing the wrong length. It was using size_bytes() which is larger than the staging buffer if offset_end - offset_begin defines a sub-range.
The call to copyBuf in DeviceArray::rangeToHost was missing a sizeof(T) factor on src_offset.
copyBufin both overloads ofDeviceArray::fromHostwere passing the wrong lengths. They were usingsize_bytes()which is larger than the staging buffer ifend - begindefines a sub-range.copyBufinDeviceArray::rangeToHostwas passing the wrong length. It was usingsize_bytes()which is larger than the staging buffer ifoffset_end - offset_begindefines a sub-range.copyBufinDeviceArray::rangeToHostwas missing asizeof(T)factor onsrc_offset.