Abstract: Technologies for accessing pooled accelerator resources over a network fabric are disclosed. In disclosed embodiments, an application hosted by a computing platform accesses remote accelerator resources over a network fabric using protocol multipathing mechanisms. A communication session is established with the remote accelerator resources. The communication session comprises at least two connections. The at least two connections at least include a first connection having or utilizing a first transport layer and a second connection having or utilizing a second transport layer that is different than the first transport layer. Other embodiments may be disclosed and/or claimed.

Abstract: Technologies for pooling accelerators over fabric are disclosed. In the illustrative embodiment, an application may access an accelerator device over an application programming interface (API) and the API can access an accelerator device that is either local or a remote accelerator device that is located on a remote accelerator sled over a network fabric. The API may employ a send queue and a receive queue to send and receive command capsules to and from the accelerator sled.

Abstract: A host fabric interface (HFI) apparatus, including: an HFI to communicatively couple to a fabric; and a remote hardware acceleration (RHA) engine to: query an orchestrator via the fabric to identify a remote resource having an accelerator; and send a remote accelerator request to the remote resource via the fabric.

Abstract: In one embodiment, a data mover accelerator is to receive, from a first agent having a first address space and a first process address space identifier (PASID) to identify the first address space, a first job descriptor comprising a second PASID selector to specify a second PASID to identify a second address space. In response to the first job descriptor, the data mover accelerator is to securely access the first address space and the second address space. Other embodiments are described and claimed.

Abstract: Systems, methods and apparatuses for distributed consistency memory. In some embodiments, the apparatus comprises at least one monitoring circuit to monitor for memory accesses to an address space; at least one a monitoring table to store an identifier of the address space; and at least one hardware core to execute an instruction to enable the monitoring circuit.

Abstract: In one embodiment, an apparatus comprises a fabric controller of a first computing node. The fabric controller is to receive, from a second computing node via a network fabric that couples the first computing node to the second computing node, a request to execute a kernel on a field-programmable gate array (FPGA) of the first computing node; instruct the FPGA to execute the kernel; and send a result of the execution of the kernel to the second computing node via the network fabric.

Abstract: Embodiments of systems, methods, and apparatuses for heterogeneous computing are described. In some embodiments, a hardware heterogeneous scheduler dispatches instructions for execution on one or more plurality of heterogeneous processing elements, the instructions corresponding to a code fragment to be processed by the one or more of the plurality of heterogeneous processing elements, wherein the instructions are native instructions to at least one of the one or more of the plurality of heterogeneous processing elements.

Abstract: In one embodiment, an apparatus comprises a fabric controller of a first computing node. The fabric controller is to receive, from a second computing node via a network fabric that couples the first computing node to the second computing node, a request to execute a kernel on a field-programmable gate array (FPGA) of the first computing node; instruct the FPGA to execute the kernel; and send a result of the execution of the kernel to the second computing node via the network fabric.

Abstract: Techniques for offload device address translation fetching are disclosed. In the illustrative embodiment, a processor of a compute device sends a translation fetch descriptor to an offload device before sending a corresponding work descriptor to the offload device. The offload device can request translations for virtual memory address and cache the corresponding physical addresses for later use. While the offload device is fetching virtual address translations, the compute device can perform other tasks before sending the corresponding work descriptor, including operations that modify the contents of the memory addresses whose translation are being cached. Even if the offload device does not cache the translations, the fetching can warm up the cache in a translation lookaside buffer. Such an approach can reduce the latency overhead that the offload device may otherwise incur in sending memory address translation requests that would be required to execute the work descriptor.

Abstract: Technologies for facilitating remote memory requests in accelerator devices are disclosed. The accelerator device includes circuitry to receive, from a kernel of the present accelerator device, a request through an application programming interface exposed to a high level software language in which the kernel of the present accelerator device is implemented, to establish a logical communication path between the kernel of the present accelerator device and a target accelerator device kernel, based on one or more physical communication paths. The communication protocol supported by the accelerator device may allow kernels operating on the accelerator device to send memory requests for memory locations at remote devices, with the communication protocol performing all of the operations necessary to carry out the memory request.

Abstract: Technologies for facilitating remote memory requests in accelerator devices are disclosed. The accelerator device includes circuitry to receive, from a kernel of the present accelerator device, a request through an application programming interface exposed to a high level software language in which the kernel of the present accelerator device is implemented, to establish a logical communication path between the kernel of the present accelerator device and a target accelerator device kernel, based on one or more physical communication paths. The communication protocol supported by the accelerator device may allow kernels operating on the accelerator device to send memory requests for memory locations at remote devices, with the communication protocol performing all of the operations necessary to carry out the memory request.

Abstract: Embodiments of systems, methods, and apparatuses for heterogeneous computing are described. In some embodiments, a hardware heterogeneous scheduler dispatches instructions for execution on one or more plurality of heterogeneous processing elements, the instructions corresponding to a code fragment to be processed by the one or more of the plurality of heterogeneous processing elements, wherein the instructions are native instructions to at least one of the one or more of the plurality of heterogeneous processing elements.

Abstract: Technologies for accelerated memory lookups include a computing device having a processor and a hardware accelerator. The processor programs the accelerator with a search value, a start pointer, one or more predetermined offsets, and a record length. Each offset may be associated with a pointer type or a value type. The accelerator initializes a memory location at the start pointer and increments the memory location by the offset. The accelerator may read a pointer value from an offset, set the memory location to the pointer value, and repeat for additional offsets. The accelerator may read a value from the offset and compare the value to the search value. If the values match, the accelerator returns the address of the matching value to the processor. If the values do not match, the accelerator searches a next record based on the record length. Other embodiments are described and claimed.

Abstract: Technologies for providing an accelerator device discovery service include a device having circuitry configured to obtain, from a discovery service, availability data indicative of a set of accelerator devices available to assist in the execution of a workload. The circuitry is also configured to select, as a function of the availability data, one or more target accelerator devices to assist in the execution of the workload, and execute the workload with the one or more target accelerator devices.

Abstract: Technologies for providing inter-kernel communication abstraction to support scale-up and scale-out include an accelerator device. The accelerator device includes circuitry to receive, from a kernel of the present accelerator device, a request through an application programming interface exposed to a high level software language in which the kernel of the present accelerator device is implemented, to establish a logical communication path between the kernel of the present accelerator device and a target accelerator device kernel, based on one or more physical communication paths. Additionally, the circuitry is to establish, in response to the request, the logical communication path between the kernel of the present accelerator device and the other accelerator device kernel and communicate data between the kernel of the present accelerator device and the other accelerator device kernel with a unified communication protocol that manages differences between the physical communication paths.

Abstract: Technologies for accessing pooled accelerator resources over a network fabric are disclosed. In disclosed embodiments, an application hosted by a computing platform accesses remote accelerator resources over a network fabric using protocol multipathing mechanisms. A communication session is established with the remote accelerator resources. The communication session comprises at least two connections. The at least two connections at least include a first connection having or utilizing a first transport layer and a second connection having or utilizing a second transport layer that is different than the first transport layer. Other embodiments may be disclosed and/or claimed.

Abstract: Some examples provide a manner of a memory transaction requester to configure a target to recognize a memory address as a non-local or non-shared address. An intermediary between the requester and the target configures a control plane layer of the target to recognize that a memory transaction involving the memory address is to be performed using a direct memory access operation. The intermediary is connected to the requester as a local device or process. After configuration, a memory transaction provided to the target with the configured memory address causes the target to invoke use of the associated direct memory access operation to retrieve content associated with the memory address or write content using a direct memory access operation.

Abstract: Described herein are devices and techniques for distributing application data. A device can communicate with one or more hardware switches. The device can receive, from a software stack, a multicast message including a constraint that indicates how application data is to be distributed. The constraint including a listing of the set of nodes and a number of nodes to which the application data is to be distributed. The device may receive, from the software stack, the application data for distribution to a plurality of nodes. The plurality of nodes being a subset of the set of nodes equaling the number of nodes. The device may select the plurality of nodes from the set of nodes. The device also may distribute a copy of the application data to the plurality of nodes based on the constraint. Also described are other embodiments.

Abstract: An extension of node architecture and proxy requests enables a node to expose memory computation capability to remote nodes. A remote node can request execution of an operation by a remote memory computation resource, and the remote memory computation resource can execute the request locally and return the results of the computation. The node includes processing resources, a fabric interface, and a memory subsystem including a memory computation resource. The local execution of the request by the memory computation resource can reduce latency and bandwidth concerns typical with remote requests.

Abstract: Embodiments of systems, methods, and apparatuses for heterogeneous computing are described. In some embodiments, a hardware heterogeneous scheduler dispatches instructions for execution on one or more plurality of heterogeneous processing elements, the instructions corresponding to a code fragment to be processed by the one or more of the plurality of heterogeneous processing elements, wherein the instructions are native instructions to at least one of the one or more of the plurality of heterogeneous processing elements.