When a process is started, a small block of memory is set aside to be used as a thread-local memory area. This memory is at the same virtual address for each thread, but is backed by different physical memory. This permits each thread to have a small block of memory that is unique, or local, to that thread.

The thread-local memory area consists of 32 DWORDs (128 bytes), each DWORD being 4 bytes in size. Up to 8 DWORDs (32 bytes) can be requested each time this function is called. If you want to allocate more than 8 DWORDs, you must call this function more than once.

Allocation is by DWORD only. If you want to store a BYTE in the thread-local memory area, you would still allocate a DWORD, then store the BYTE in it.