Abstract: An apparatus to facilitate in-place memory copy during remote data transfer in a heterogeneous compute environment is disclosed. The apparatus includes a processor to receive data via a network interface card (NIC) of a hardware accelerator device; identify a destination address of memory of the hardware accelerator device to write the data; determine that access control bits of the destination address in page tables maintained by a memory management unit (MMU) indicate that memory pages of the destination address are both registered and free; write the data to the memory pages of the destination address; and update the access control bits for memory pages of the destination address to indicate that the memory pages are restricted, wherein setting the access control bits to restricted prevents the NIC and a compute kernel of the hardware accelerator device from accessing the memory pages.

Abstract: An apparatus to facilitate protecting assets of mutually distrustful entities during federated learning training on a remote device is disclosed. The apparatus includes a processor to a processor to: receive, at a trusted execution environment (TEE) hosted by a client platform, an encrypted machine learning (ML) model and a cryptographic message authentication code (MAC) from a model owner platform, wherein the encrypted ML model is encrypted by the model owner platform using homomorphic encryption (HE); verify integrity of the encrypted ML model using the cryptographic MAC and a TEE key established by the processor during remote attestation of the TEE with the model owner platform; perform, in the TEE, training of the encrypted ML model using HE computation on sensor data; and send, to the model owner platform, output of the training comprising updated model parameters of the encrypted ML model, where the output is homomorphically encrypted.

Abstract: An apparatus to facilitate confidential computing in a heterogeneous compute environment including a network-connected hardware accelerator is disclosed. The apparatus includes a processor to provide a first trusted execution environment (TEE) to run an application, and to send, via the application to a user mode driver (UMD) hosted in first the TEE, a command to transfer data of the application to a hardware accelerator device that is connected via network to the application; encrypt and integrity-protect, via the UMD, the data using shared secret data keys and a destination buffer address of the hardware accelerator device to generate encrypted and integrity-protected data, the shared secret data keys established with a remote service in a second TEE operating on an accelerator platform connected to the application; and interface, via the UMD with a local network interface card (NIC), to cause a copy over the network of the encrypted and integrity-protected data.

Abstract: An apparatus to facilitate in-place memory copy during remote data transfer in a heterogeneous compute environment is disclosed. The apparatus includes a processor to receive data via a network interface card (NIC) of a hardware accelerator device; identify a destination address of memory of the hardware accelerator device to write the data; determine that access control bits of the destination address in page tables maintained by a memory management unit (MMU) indicate that memory pages of the destination address are both registered and free; write the data to the memory pages of the destination address; and update the access control bits for memory pages of the destination address to indicate that the memory pages are restricted, wherein setting the access control bits to restricted prevents the NIC and a compute kernel of the hardware accelerator device from accessing the memory pages.

Abstract: A hardware system is disclosed for active-core-based performance boost. In an example aspect, the hardware system includes multiple cores and a power mode manager. Each core can be powered up if active or powered down if inactive. The power mode manager manages a power mode collection including an independent power mode collection and an active-core-dependent power mode collection. The power mode manager includes a software-accessible power mode manager and a hardware-reserved power mode manager. The software-accessible power mode manager provides a power-mode-triggering pathway to enable software to trigger activation of an independent power mode of the independent power mode collection. The hardware-reserved power mode manager excludes the software from being able to trigger activation of a dependent power mode of the active-core-dependent power mode collection and triggers activation of a dependent power mode of the active-core-dependent collection based on a number of active cores of the multiple cores.

Abstract: Aspects of the disclosure are directed to DC power management. A sequencer may be configured to execute a first command, wherein the first command is associated with a unique group tag; compare the unique group tag to a master group tag; determine if an interrupt is detected; lock the master group tag to yield a locked master group tag; execute a second command, wherein the second command is associated with the locked master group tag; determine that an end of commands in the locked master group tag is reached and execute a sequence jump through command to put a processor back to a regular power state.

Abstract: Aspects of the disclosure are directed to DC power management. A sequencer may be configured to execute a first command, wherein the first command is associated with a unique group tag; compare the unique group tag to a master group tag; determine if an interrupt is detected; lock the master group tag to yield a locked master group tag; execute a second command, wherein the second command is associated with the locked master group tag; determine that an end of commands in the locked master group tag is reached and execute a sequence jump through command to put a processor back to a regular power state.

Abstract: An apparatus includes a first circuit configured to receive one or more requests from a plurality of cores. Each of the one or more requests is to enter or to exit one of a plurality of power-down modes. The first circuit further selects one or more of the cores to enter or to exit the requested power-down mode or modes based on inrush current information associated with the power-down modes. A second circuit is configured to effect entering or exiting the requested power-down mode or modes in the selected one or more of the cores.

Abstract: A hardware system is disclosed for active-core-based performance boost. In an example aspect, the hardware system includes multiple cores and a power mode manager. Each core can be powered up if active or powered down if inactive. The power mode manager manages a power mode collection including an independent power mode collection and an active-core-dependent power mode collection. The power mode manager includes a software-accessible power mode manager and a hardware-reserved power mode manager. The software-accessible power mode manager provides a power-mode-triggering pathway to enable software to trigger activation of an independent power mode of the independent power mode collection. The hardware-reserved power mode manager excludes the software from being able to trigger activation of a dependent power mode of the active-core-dependent power mode collection and triggers activation of a dependent power mode of the active-core-dependent collection based on a number of active cores of the multiple cores.

Abstract: Systems and methods are disclosed for improved processor hang detection. An exemplary method comprises setting a timer with a hang threshold value for each of a plurality of processors of a system on a chip (SoC). The hang threshold value represents a time in microseconds. The method further comprising receiving a first heartbeat signal from each of the plurality of processors with detection logic hardware of a hang controller coupled to the plurality of processors and to the timer. The timer is reset for each of the plurality of processors if a second heartbeat signal is received from the corresponding one of the plurality of processors before the timer expires. Alternatively, a hang event notification is generated if the second heartbeat signal is not received from the corresponding one of the plurality of processors before the timer expires.

Abstract: An apparatus includes a first circuit configured to receive one or more requests from a plurality of cores. Each of the one or more requests is to enter or to exit one of a plurality of power-down modes. The first circuit further selects one or more of the cores to enter or to exit the requested power-down mode or modes based on inrush current information associated with the power-down modes. A second circuit is configured to effect entering or exiting the requested power-down mode or modes in the selected one or more of the cores.