Efficient data sharing for graphics data processing operations
An apparatus to facilitate efficient data sharing for graphics data processing operations is disclosed. The apparatus includes a processing resource to generate a stream of instructions, an L1 cache communicably coupled to the processing resource and comprising an on-page detector circuit to determine that a set of memory requests in the stream of instructions access a same memory page; and set a marker in a first request of the set of memory requests; and arbitration circuitry communicably coupled to the L1 cache, the arbitration circuitry to route the set of memory requests to memory comprising the memory page and to, in response to receiving the first request with the marker set, remain with the processing resource to process the set of memory requests.
1. An apparatus comprising:
a processing resource of a chiplet; and
an L1 cache communicably coupled to the processing resource and comprising synchronization hardware circuit to:
allocate a first thread group as a member of a super thread group (STG) comprising a collection of thread groups running on the chiplet;
receive from a second thread group within the STG, a request to communicate with the first thread group identified by a thread group identifier (ID) within the STG;
access a routing table to determine a location of the first thread group based on the thread group ID; and
route the request to the determined location of the first thread group using communication links between L1 caches of the chiplet.
2. The apparatus of claim 1 , wherein the processing resource is an execution unit in a graphics processing unit (GPU).
3. The apparatus of claim 1 , wherein the synchronization hardware circuit comprises one or more of a router, a receiver/transmitter (Rx/Tx), or the routing table.
4. The apparatus of claim 3 , wherein the router is to implement an application programming interface (API) defined to allow communication between sub-slices of the chiplet.
5. The apparatus of claim 3 , wherein the router is to communicate with the Rx/Tx to cause communications from the L1 cache to be sent or received over buses of a point-to-point communication network.
6. The apparatus of claim 1 , wherein a kernel is to utilize query functions to inquire an STG ID number, to inquire a number of thread groups in the STG, or to send and receive data from/to thread in a same STG as the kernel using a logical thread number.
7. The apparatus of claim 1 , wherein the chiplet further comprises a chiplet-level cache for sub-slice communication within the chiplet and to enable communication between threads on different sub-slices of the chiplet that are within the STG of the chiplet.
8. The apparatus of claim 7 , wherein the chiplet-level cache comprises an L2 cache that is in communication with a global cache of a base die of the chiplet.
9. The apparatus of claim 1 , wherein the apparatus is at least one of a single instruction multiple data (SIMD) machine or a single instruction multiple thread (SIMT) machine.
10. A method comprising:
allocating, by a synchronization hardware circuit of an L1 cache of a graphics processor, a first thread group as a member of a super thread group (STG) comprising a collection of thread groups running on the graphics processor;
receiving, by the synchronization hardware circuit, from a second thread group within the STG, a request to communicate with the first thread group identified by a thread group identifier (ID) within the STG;
accessing, by the synchronization hardware circuit, a routing table to determine a location of the first thread group based on the thread group ID; and
routing, by the synchronization hardware circuit, the request to the determined location of the first thread group using communication links between L1 caches of the graphics processor.
11. The method of claim 10 , wherein the synchronization hardware circuit comprises one or more of a router, a receiver/transmitter (Rx/Tx), or the routing table.
12. The method of claim 11 , wherein the router is to implement an application programming interface (API) defined to allow communication between sub-slices of the graphics processor.
13. The method of claim 11 , wherein the router is to communicate with the Rx/Tx to cause communicate from the L1 cache to be sent or received over buses of a point-to-point communication network.
14. The method of claim 10 , wherein a kernel is to utilize query functions to inquire an STG ID number, to inquire a number of thread groups in the STG, or to send and receive data from/to thread in a same STG as the kernel using a logical thread number.
15. The method of claim 10 , wherein the graphics processor further comprises a chiplet-level cache for sub-slice communication within the graphics processor and to enable communication between threads on different sub-slices of the graphics processor that are within the STG of the graphics processor.
16. A non-transitory computer-readable medium having instructions stored thereon, which when executed by one or more processors, cause the one or more processors to:
allocate, by a synchronization hardware circuit of an L1 cache of the one or more processors, a first thread group as a member of a super thread group (STG) comprising a collection of thread groups running on the one or more processors;
receive, by the synchronization hardware circuit, from a second thread group within the STG, a request to communicate with the first thread group identified by a thread group identifier (ID) within the STG;
access, by the synchronization hardware circuit, a routing table to determine a location of the first thread group based on the thread group ID; and
route, by the synchronization hardware circuit, the request to the determined location of the first thread group using communication links between L1 caches of the one or more processors.
17. The non-transitory computer-readable medium of claim 16 , wherein the synchronization hardware circuit comprises one or more of a router, a receiver/transmitter (Rx/Tx), or the routing table.
18. The non-transitory computer-readable medium of claim 17 , wherein the router is to implement an application programming interface (API) defined to allow communication between sub-slices of the one or more processors.
19. The non-transitory computer-readable medium of claim 16 , wherein a kernel is to utilize query functions to inquire an STG ID number, to inquire a number of thread groups in the STG, or to send and receive data from/to thread in a same STG as the kernel using a logical thread number.
20. The non-transitory computer-readable medium of claim 16 , wherein the one or more processors further comprise a chiplet-level cache for sub-slice communication within the one or more processors and to enable communication between threads on different sub-slices of the one or more processors that are within the STG of the one or more processors.