Providing native support for generic pointers in a graphics processing unit
Systems and methods for supporting generic pointers in hardware of a graphics processing unit (GPU) are provided. In various examples, a GPU includes multiple sub-cores each having a processing resource and a load/store pipeline. The processing resource is operable to receive a memory access message including a pointer and a memory type identifier indicative of the pointer representing a generic pointer. The processing resource is further operable to output a load or store operation to the load/store pipeline based on the memory access message, including computing an address for the load or store operation by adding a base address of a named memory type of a plurality of named memory types referenced by the generic pointer to an offset into a memory of the named memory type. The load/store pipeline is operable to, responsive to receipt of the load or store operation, access the memory at the address.
1. A graphics processing unit (GPU) comprising:
a plurality of sub-cores each including a processing resource and a load/store pipeline;
the processing resource is operable to:
receive a memory access message including a memory type identifier and a pointer, wherein the memory type identifier is indicative of the pointer representing a generic pointer; and
output a load or store operation to the load/store pipeline based on the memory access message, including computing an address for the load or store operation by adding a base address of a named memory type of a plurality of named memory types referenced by the generic pointer to an offset into a memory of the named memory type; and
the load/store pipeline is operable to, responsive to receipt of the load or store operation, access the memory at the address.
2. The GPU of claim 1 , wherein the plurality of named memory types are selected from a group comprising a shared local memory, a private scratch memory, a constant memory, and a global memory.
3. The GPU of claim 2 , further comprising a local range register operable to be programmed by a graphics driver with a local memory generic base address and a size of the shared local memory.
4. The GPU of claim 3 , wherein when the generic pointer is being used to access the shared local memory, a generic linker symbol used as a placeholder by a compiler for the base address within a binary containing the memory access message is patched with the local memory generic base address.
5. The GPU of claim 2 , further comprising a scratch range register operable to be programmed by a graphics driver with a private generic base address and a size of the private scratch memory.
6. The GPU of claim 5 , wherein when the generic pointer is being used to access the private scratch memory, a generic linker symbol used as a placeholder by a compiler for the base address within a binary containing the memory access message is patched with the private generic base address.
7. The GPU of claim 1 , wherein the address comprises a first of a plurality of lane addresses for each of a plurality of Single Instruction Multiple Data (SIMD) or Single Instruction Multiple-Thread (SIMT) operands and the remainder of the plurality of lane addresses are calculated based on a predefined per-lane offset.
8. A method comprising:
receiving, by a processing resource of a sub-core of a plurality of sub-cores of a graphics processing unit (GPU), a memory access message including a memory type identifier and a pointer, wherein the memory type identifier is indicative of the pointer representing a generic pointer;
based on the memory access message, outputting, by the processing resource, a load or store operation to a load/store pipeline of the sub-core, including computing an address for the load or store operation by adding a base address of a named memory type of a plurality of named memory types referenced by the generic pointer to an offset into a memory of the named memory type;
responsive to receipt of the load or store operation, accessing, by the load/store pipeline, the memory at the address.
9. The method of claim 8 , wherein the plurality of named memory types are selected from a group comprising a shared local memory, a private scratch memory, a constant memory, and a global memory.
10. The method of claim 9 , further comprising receiving, by a local range register of the sub-core, a local memory generic base address and a size of the shared local memory.
11. The method of claim 10 , wherein when the generic pointer is being used to access the shared local memory, a binary containing the memory access message is patched by replacing a generic linker symbol that is used as a placeholder by a compiler for the base address with the local memory generic base address.
12. The method of claim 9 , further comprising receiving, by a scratch range register of the sub-core, a private generic base address and a size of the private scratch memory.
13. The method of claim 12 , wherein when the generic pointer is being used to access the private scratch memory, a binary containing the memory access message is patched by replacing a generic linker symbol that is used as a placeholder by a compiler for the base address with the private generic base address.
14. A system comprising:
a central processing unit (CPU); and
a graphics processing unit (GPU) coupled to the CPU, wherein the GPU includes a sub-core including a processing resource and a load/store pipeline;
the processing resource is operable to:
receive a memory access message including a memory type identifier and a pointer, wherein the memory type identifier is indicative of the pointer representing a generic pointer; and
output a load or store operation to the load/store pipeline based on the memory access message, including computing an address for the load or store operation by adding a base address of a named memory type of a plurality of named memory types referenced by the generic pointer to an offset into a memory of the named memory type; and
the load/store pipeline is operable to, responsive to receipt of the load or store operation, access the memory at the address.
15. The system of claim 14 , wherein the plurality of named memory types are selected from a group comprising a shared local memory, a private scratch memory, a constant memory, and a global memory.
16. The system of claim 15 , wherein the sub-core further includes a local range register operable to be programmed by a graphics driver with a local memory generic base address and a size of the shared local memory.
17. The system of claim 16 , wherein when the generic pointer is being used to access the shared local memory, a generic linker symbol used as a placeholder by a compiler for the base address within a binary containing the memory access message is patched with the local memory generic base address.
18. The system of claim 15 , wherein the sub-core further includes a scratch range register operable to be programmed by a graphics driver with a private generic base address and a size of the private scratch memory.
19. The system of claim 18 , wherein when the generic pointer is being used to access the private scratch memory, a generic linker symbol used as a placeholder by a compiler for the base address within a binary containing the memory access message is patched with the private generic base address.
20. The system of claim 14 , wherein the address comprises a first of a plurality of lane addresses for each of a plurality of Single Instruction Multiple Data (SIMD) or Single Instruction Multiple-Thread (SIMT) operands and the remainder of the plurality of lane addresses are calculated based on a predefined per-lane offset.