Abstract: A system includes a multicore chip configured to perform machine learning (ML) operations. The system also includes a power monitoring module configured to measure power consumption of the multicore chip on a main power rail of the multicore chip. The power monitoring module is further configured to assert a signal in response to the measured power consumption exceeding a first threshold. The power monitoring module is further configured to transmit the asserted signal to a power throttling module to initiate a power throttling for the multicore chip.

Abstract: A system includes a multicore chip configured to perform machine learning (ML) operations. The system also includes a power monitoring module configured to measure power consumption of the multicore chip on a main power rail of the multicore chip. The power monitoring module is further configured to assert a signal in response to the measured power consumption exceeding a first threshold. The power monitoring module is further configured to transmit the asserted signal to a power throttling module to initiate a power throttling for the multicore chip.

Abstract: Self modifying code is detected using a translation lookaside buffer in order to provide cache coherency. The translation lookaside buffer has physical page addresses stored therein over which snoops can be performed using the physical memory address of a store into memory. The translation lookaside buffer includes a content addressable memory which not only provides page translation but provides content addressability based on the physical page addresses stored therein. If a match occurs during a snoop using the translation lookaside buffer, it is possible that an SMC occurred within the page of locations stored in memory associated with the matched physical page addresses. To provide finer granularity than a page of addresses, FINE HIT bits are included with each entry in the cache associating information in the cache to portions of a page within memory.

Abstract: Self modifying code is detected using a translation lookaside buffer in order to provide cache coherency. The translation lookaside buffer has physical page addresses stored therein over which snoops can be performed using the physical memory address of a store into memory. The translation lookaside buffer includes a content addressable memory which not only provides page translation but provides content addressability based on the physical page addresses stored therein. If a match occurs during a snoop using the translation lookaside buffer, it is possible that an SMC occurred within the page of locations stored in memory associated with the matched physical page addresses. To provide finer granularity than a page of addresses, FINE HIT bits are included with each entry in the cache associating information in the cache to portions of a page within memory.

Abstract: An apparatus and a method for embedding a dynamic state machine in a static integrated circuit environment. A static integrated circuit environment which is capable of suspending operation during a power down clock-stopped condition and resuming operation from a stored state at the conclusion of the power down condition is combined with a dynamic state machine featuring dynamic latches embedded in the static integrated circuit environment. The disclosed dynamic state machine is also configured to suspend operation during the power down condition and resume operation after the power down condition from the stored stated. In addition, both the static integrated circuit environment and the embedded dynamic state machine draw minimal power during the power down condition.

Abstract: A micro processor including a bus fraction register with an encoding which when decoded indicates either a bus fraction encoding or a stop clock function, data processing logic that includes a number of units including a bus unit, arranged as an instruction pipeline. The units are clocked by an internal clock running at a first frequency and operating with an I/O bus clocked by an I/O clock running at a second frequency which is a fraction of the first frequency. A stop clock signal is generated upon the condition that the bus fraction register contains the stop clock encoding. A bus unit busy (BBSY) signal line is polled to ensure that all pending bus cycles in the pipeline are completed, the polling being initiated in response to the stop clock signal. A special cycle encoded to indicate the stop clock function is run to inform the units of the microprocessor that the internal and I/O clocks are going to stop toggling.

Abstract: Apparatus for enabling internal data processing logic including a number of units clocked at a first frequency to operate with an external bus operating at a second frequency that is a fraction m/n of said first frequency. A first bus is connected via readers to data latched for data transfer from the number internal units of the data processing logic to the data latches. A second bus is connected via drivers to the data latches for data transfer from internal bus units to the data latches. The data latches are connected to the external bus. A control circuit connected to the readers and drivers controls the readers and drivers to guarantee that sampling is done when logic is stable. The control circuit includes priority logic for determining priority between the units for permitting a high priority unit to transfer data on the external bus.