Memory management using a register

A method comprising:

• receiving, by a processing device, a first request to allocate one or more memory blocks of a first plurality of memory blocks of a tensor memory associated with a first memory;
• identifying a consecutive set of a first portion of bits of a first register implemented as a hardware register of the processing device with a first logical state, wherein the first logical state indicates that corresponding memory blocks of the one or more memory blocks are free, wherein the processing device comprises a plurality of multiprocessors, wherein the first register is disposed within a multiprocessor of the plurality of multiprocessors, and wherein the first memory is disposed on the multiprocessor;
• performing an atomic operation to adjust the consecutive set of the first portion of bits of the first register to a second logical state, wherein the atomic operation comprises a read-modify-write operation that executes without interruption by other memory accesses to the first register;
• and
• sending an allocation address comprising an index of the consecutive set of the first portion of bits of the first register to the first request, wherein the allocation address is useable to access the corresponding memory blocks.

A parallel processing unit (PPU) comprising:

• a first memory including a first plurality of memory blocks of a tensor memory;
• a first register implemented as a hardware register configured to maintain a bitmap using a first portion of bits of the first register, wherein each bit of the first portion of bits of the first register represents a corresponding memory block of the first plurality of memory blocks, wherein the PPU comprises a plurality of multiprocessors, wherein the first register is disposed within a multiprocessor of the plurality of multiprocessors, and wherein the first memory is disposed on the multiprocessor, and wherein the PPU is to:
• receive a first request to allocate one or more memory blocks of the first plurality of memory blocks;
• identify a consecutive set of the first portion of bits of the first register with a first logical state, wherein the first logical state indicates that corresponding memory blocks of the one or more memory blocks are free;
• perform an atomic operation to adjust the consecutive set of the first portion of bits of the first register to a second logical state, wherein the atomic operation comprises a read-modify-write operation that executes without interruption by other memory accesses to the first register;
• and
• send an allocation address comprising an index of the consecutive set of the first portion of bits of the first register to the first request, wherein the allocation address is useable to access the corresponding memory blocks.