Techniques for safely and efficiently enqueueing and dequeueing data on a graphics processor

A method for managing first-in first-out (FIFO) queues in graphics processing, comprising:

• allocating multiple pages of memory to a FIFO queue of multiple FIFO queues;
• and
• executing, via parallel execution of multiple read threads of a graphics processing unit (GPU), a read operation to read data from memory locations in multiple pages of memory corresponding to the FIFO queue, wherein, for a given read thread of the multiple read threads, the read operation comprises:
• determining whether a read page pointer, which points to a page of memory being read, and a write page pointer, which points to a page of memory being written, point to a same page of the memory;
• where the read page pointer and the write page pointer do not point to the same page of the memory:
• reading data from one or more memory locations to which a read allocation pointer points within the memory;
• and
• advancing a read done pointer to the one or more memory locations,
• wherein the read allocation pointer and the read done pointer are common to the FIFO queue;
• and
• where the read page pointer and the write page pointer point to the same page of the memory, indicating an error resulting from execution of the read operation.

A method for managing first-in first-out (FIFO) queues in graphics processing, comprising:

• allocating multiple pages of memory to a FIFO queue of multiple FIFO queues;
• and
• executing, via parallel execution of multiple write threads of a graphics processing unit (GPU), a write operation to write data to memory locations in multiple pages of memory corresponding to the FIFO queue, wherein, for a given write thread of the multiple write threads, the write operation comprises:
• obtaining and advancing one write allocation pointer of multiple write allocation pointers, wherein the multiple write allocation pointers are associated with the FIFO queue, and
• determining whether the one write allocation pointer of the multiple write allocation pointers is greater than one read done pointer of multiple read done pointers maintained for the FIFO queue;
• where the one write allocation pointer is not greater than the one read done pointer:
• writing data to one or more memory locations to which the one write allocation pointer points within the memory;
• and
• where the one write allocation pointer is greater than the one read done pointer:
• determining, based on obtaining a read index and a write index from a snapshot pointer used to avoid race conditions that may occur when reading and/or writing the data, whether to write data to one or more memory locations within the memory.

A method for managing first-in first-out (FIFO) queues in graphics processing, comprising:

• allocating multiple pages of memory to a FIFO queue of multiple FIFO queues;
• and
• executing, via parallel execution of multiple write threads of a graphics processing unit (GPU), a write operation to write data to memory locations in multiple pages of memory corresponding to the FIFO queue, wherein, for a given write thread of the multiple write threads, the write operation comprises:
• determining whether a write page pointer, which points to a page of memory being written to, and a read page pointer, which points to a page of memory being read, point to a same page of the memory;
• where the write page pointer and the read page pointer do not point to the same page of the memory:
• writing data to one or more memory locations to which a write allocation pointer points within the memory;
• and
• advancing a write done pointer to the one or more memory locations, wherein the write allocation pointer and the write done pointer are common to the FIFO queue;
• and
• where the write page pointer and the read page pointer point to the same page of the memory, indicating an error resulting from execution of the write operation.