Platform interface layer and protocol for accelerators

1. An accelerator apparatus, comprising:

a programmable region capable of being programmed to provide an accelerator function unit (AFU); and

a platform interface layer (PIL) to communicatively couple to the AFU via an intra-accelerator protocol, and to provide multiplexed communication with a processor via a plurality of platform interconnect interfaces, wherein the PIL is to provide abstracted communication services for the AFU to communicate with the processor,

wherein the PIL is to support cache hints from the AFU, wherein supporting cache hints comprises receiving a cache hint from the AFU for a transaction, selecting a platform interconnect for the transaction, and handling the cache hint contextually according to a capability of the platform interconnect.

2. The accelerator apparatus of claim 1 , wherein the intra-accelerator protocol is a core cache interface version P (CCI-P) protocol.

3. The accelerator apparatus of claim 1 , wherein the plurality of platform interconnect interfaces comprises a first low-latency interface and a second high-bandwidth interface, wherein the PIL is configured to select a preferred platform interconnect interface for a data transaction.

4. The accelerator apparatus of claim 1 , wherein the plurality of platform interconnect interfaces comprises a first cache-coherent interface, and a second non-cache-coherent interface, wherein the PIL is to select a preferred platform interconnect interface for a data transaction.

5. The accelerator apparatus of claim 4 , wherein the first cache-coherent interface is an ultra-path interconnect (UPI) interface.

6. The accelerator apparatus of claim 4 , wherein the second non-cache-coherent interface is a peripheral component interconnect express (PCIe) interface.

7. The accelerator apparatus of claim 1 , wherein the PIL is to provide to the AFU a non-ordered memory model.

8. The accelerator apparatus of claim 7 , wherein the PIL is to support non-posted writes and write fences.

9. The accelerator apparatus of claim 1 , wherein the PIL is to provide a plurality of virtual channels to the AFU.

10. The accelerator apparatus of claim 9 , wherein the plurality of virtual channels comprise a first virtual channel tuned for low latency, and a second virtual channel tuned for high bandwidth bursts.

11. The accelerator apparatus of claim 1 , wherein the PIL is to provide a burst mode to guarantee that a burst request does not cross a page boundary.

12. The accelerator apparatus of claim 1 , wherein the PIL is to support an address width matching a processor virtual address width via a platform interconnect interface of the plurality of platform interconnect interfaces, wherein the intra-accelerator protocol is to be agnostic of virtual or physical addressing.

13. The accelerator apparatus of claim 1 , wherein the PIL is further to provide power management of the accelerator apparatus.

14. The accelerator apparatus of claim 1 , wherein the PIL is further to provide thermal management of the accelerator apparatus.

15. The accelerator apparatus of claim 1 , wherein the accelerator apparatus comprises an FPGA, the PIL comprises a first region of the FPGA, and the AFU comprises a second region of the FPGA.

16. The accelerator apparatus of claim 1 , wherein the PIL comprises an intellectual property block.

17. The accelerator apparatus of claim 1 , wherein the PIL comprises a co-processor.

18. One or more tangible, non-transitory computer-readable mediums having stored thereon a plurality of instructions that, when executed, causes a computing device to:

provide a platform interface layer (PIL) for an accelerator apparatus, the PIL to communicatively couple to an accelerator function unit (AFU) via an intra-accelerator protocol, and to provide multiplexed communication with a processor via a plurality of platform interconnect interfaces, wherein the PIL is to provide abstracted communication services for the AFU to communicate with the processor.

19. The one or more tangible, non-transitory computer-readable mediums of claim 18 , wherein the intra-accelerator protocol is a core cache interface version P (CCI-P) protocol.

20. The one or more tangible, non-transitory computer-readable mediums of claim 18 , wherein the plurality of platform interconnect interfaces comprises a first low-latency interface and a second high-bandwidth interface, wherein the PIL is configured to select a preferred platform interconnect interface for a data transaction.

21. The one or more tangible, non-transitory computer-readable mediums of claim 18 , wherein the plurality of platform interconnect interfaces comprises a first cache-coherent interface, and a second non-cache-coherent interface, wherein the PIL is to select a preferred platform interconnect interface for a data transaction.

22. The one or more tangible, non-transitory computer-readable mediums of claim 21 , wherein the first cache-coherent interface is an ultra-path interconnect (UPI) interface.

23. The one or more tangible, non-transitory computer-readable mediums of claim 21 , wherein the second non-cache-coherent interface is a peripheral component interconnect express (PCIe) interface.

24. A method of providing a platform interface layer (PIL) for an accelerator apparatus, comprising:

communicatively coupling to an accelerator function unit (AFU) via an intra-accelerator protocol;

providing multiplexed communication with a processor via a plurality of platform interconnect interfaces, comprising providing abstracted communication services for the AFU to communicate with the processor,

wherein the PIL is to support cache hints from the AFU, wherein supporting cache hints comprises receiving a cache hint from the AFU for a transaction, selecting a platform interconnect for the transaction, and handling the cache hint contextually according to a capability of the platform interconnect.