Abstract: Aspects relate to Input/Output (IO) Memory Management Units (MMUs) that include hardware structures for implementing virtualization. Some implementations allow guests to setup and maintain device IO tables within memory regions to which those guests have been given permissions by a hypervisor. Some implementations provide hardware page table walking capability within the IOMMU, while other implementations provide static tables. Such static tables may be maintained by a hypervisor on behalf of guests. Some implementations reduce a frequency of interrupts or invocation of hypervisor by allowing transactions to be setup by guests without hypervisor involvement within their assigned device IO regions. Devices may communicate with IOMMU to setup the requested memory transaction, and completion thereof may be signaled to the guest without hypervisor involvement. Various other aspects will be evident from the disclosure.

Abstract: Techniques for neural network processing using specialized data representation are disclosed. Input data for manipulation in a layer of a neural network is obtained. The input data includes image data, where the image data is represented in bfloat16 format without loss of precision. The manipulation of the input data is performed on a processor that supports single-precision operations. The input data is converted to a 16-bit reduced floating-point representation, where the reduced floating-point representation comprises an alternative single-precision data representation mode. The input data is manipulated with one or more 16-bit reduced floating-point data elements. The manipulation includes a multiply and add-accumulate operation. The manipulation further includes a unary operation, a binary operation, or a conversion operation. A result of the manipulating is forwarded to a next layer of the neural network.

Abstract: Techniques are disclosed for address manipulation using indices and tags. A first index is generated from bits of a processor program counter, where the first index is used to access a branch predictor bimodal table. A first branch prediction is provided from the bimodal table, based on the first index. The first branch prediction is matched against N tables, where the tables contain prior branch histories, and where: the branch history in table T(N) is of greater length than the branch history of table T(N-1), and the branch history in table T(N-1) is of greater length than the branch history of table T(N-2). A processor address is manipulated using a greatest length of hits of branch prediction matches from the N tables, based on one or more hits occurring. The branch predictor address is manipulated using the first branch prediction from the bimodal table, based on zero hits occurring.

Abstract: Techniques are disclosed for processor synchronization within a reconfigurable computing environment for processor array redundancy. Processing elements are configured within a reconfigurable fabric to implement two or more redundant processors, where the two or more redundant processors are enabled for coincident operation. An agent is loaded on each of the two or more redundant processors, where the agent performs a function requiring data validation. The agent is fired on each of the two or more redundant processors to commence coincident operation. The coincident operation can include a lockstep operation. An output data result from each of the two or more redundant processors is compared to enable a data validation result. The data validation result is propagated. The propagating the data validation result can be based on comparing valid output data or can be based on comparing invalid output data.

Abstract: Techniques are disclosed for address manipulation using indices and tags. A first index is generated from bits of a processor program counter, where the first index is used to access a branch predictor bimodal table. A first branch prediction is provided from the bimodal table, based on the first index. The first branch prediction is matched against N tables, where the tables contain prior branch histories, and where: the branch history in table T(N) is of greater length than the branch history of table T(N-1), and the branch history in table T(N-1) is of greater length than the branch history of table T(N-2). A processor address is manipulated using a greatest length of hits of branch prediction matches from the N tables, based on one or more hits occurring. The branch predictor address is manipulated using the first branch prediction from the bimodal table, based on zero hits occurring.

Abstract: Techniques for mixed-precision data manipulation for neural network data computation are disclosed. A first left group comprising eight bytes of data and a first right group of eight bytes of data are obtained for computation using a processor. A second left group comprising eight bytes of data and a second right group of eight bytes of data are obtained. A sum of products is performed between the first left and right groups and the second left and right groups. The sum of products is performed on bytes of 8-bit integer data. A first result is based on a summation of eight values that are products of the first group's left eight bytes and the second group's left eight bytes. A second result is based on the summation of eight values that are products of the first group's left eight bytes and the second group's right eight bytes. Results are output.

Abstract: Techniques are disclosed for address manipulation using indices and tags. A first index is generated from bits of a processor program counter, where the first index is used to access a branch predictor bimodal table. A first branch prediction is provided from the bimodal table, based on the first index. The first branch prediction is matched against N tables, where the tables contain prior branch histories, and where: the branch history in table T(N) is of greater length than the branch history of table T(N?1), and the branch history in table T(N?1) is of greater length than the branch history of table T(N?2). A processor address is manipulated using a greatest length of hits of branch prediction matches from the N tables, based on one or more hits occurring. The branch predictor address is manipulated using the first branch prediction from the bimodal table, based on zero hits occurring.

Abstract: A binary logic circuit for manipulating an input binary string includes a first stage of a first group of multiplexers arranged to select respective portions of an input binary string and configured to receive a respective first control. A second stage is included in which a plurality of a second group of multiplexers is arranged to select respective portions of the input binary string and configured to receive a respective second control signal. The control signals are provided such that each multiplexer of a second group is configured to select a respective second portion of the first binary string. Control circuitry is configured to generate the first and second control signals such that two or more of the first groups and/or two or more of the second groups of multiplexers are independently controllable.

Abstract: Fault tolerant and fault detecting multi-threaded processors are described. Instructions from a program are executed by both a master thread and a slave thread and execution of the master thread is prioritized. If the master thread stalls or reaches a memory write after having executed a sequence of instructions, the slave thread executes a corresponding sequence of instructions, where at least the first and last instructions in the sequence are the same as the sequence executed by the master thread. When the slave thread reaches the point at which execution of the master thread stopped, the contents of register banks for both the threads are compared, and if they are the same, execution by the master thread is allowed to continue, and any buffered speculative writes are committed to the memory system.

Abstract: A method provides for decoding, in a microprocessor, an instruction into data identifying a first register, a second register, an immediate value, and an opcode identifier. The opcode identifier is interpreted as indicating that an arithmetic operation is to be performed on the first register and the second register, and that the microprocessor is to perform a change of control operation in response to the addition of the first register and the second register causing overflow or underflow. The change of control operation is to a location in a program determined based on the immediate value. A processor can be provided with a decoder and other supporting circuitry to implement such method. Overflow/underflow can be checked on word boundaries of a double-word operation.

Abstract: Methods and systems for improved control of traffic generated by a processor are described. In an embodiment, when a device generates a pre-fetch request for a piece of data or an instruction from a memory hierarchy, the device includes a pre-fetch identifier in the request. This identifier flags the request as a pre-fetch request rather than a non-pre-fetch request, such as a time-critical request. Based on this identifier, the memory hierarchy can then issue an abort response at times of high traffic which suppresses the pre-fetch traffic, as the pre-fetch traffic is not fulfilled by the memory hierarchy. On receipt of an abort response, the device deletes at least a part of any record of the pre-fetch request and if the data/instruction is later required, a new request is issued at a higher priority than the original pre-fetch request.

Abstract: Methods and migration units for use in out-of-order processors for migrating data to register file caches associated with functional units of the processor to satisfy register read operations. The migration unit receives register read operations to be executed for a particular functional unit. The migration unit reviews entries in a register renaming table to determine if the particular functional unit has recently accessed the source register and thus is likely to comprise an entry for the source register in its register file cache. In particular, the register renaming table comprises entries for physical registers that indicate what functional units have accessed the physical register. If the particular functional unit has not accessed the particular physical register the migration unit migrates data to the register file cache associated with the particular functional unit.

Abstract: In an aspect, a processor supports modeless execution of 64 bit and 32 bit instructions. A Load/Store Unit (LSU) decodes an instruction that without explicit opcode data indicating whether the instruction is to operate in a 32 or 64 bit memory address space. LSU treats the instruction either as a 32 or 64 bit instruction in dependence on values in an upper 32 bits of one or more 64 bit operands supplied to create an effective address in memory. In an example, a 4 GB space addressed by 32-bit memory space is divided between upper and lower portions of a 64-bit address space, such that a 32-bit instruction is differentiated from a 64-bit instruction in dependence on whether an upper 32 bits of one or more operands is either all binary 1 or all binary 0. Such a processor may support decoding of different arithmetic instructions for 32-bit and 64-bit operations.

Abstract: A fetch ahead branch target buffer is used by a branch predictor to determine a target address for a branch instruction based on a fetch pointer for a previous fetch bundle, i.e. a fetch bundle which is fetched prior to a fetch bundle which includes the branch instruction. An entry in the fetch ahead branch target buffer corresponds to one branch instruction and comprises a data portion identifying the target address of that branch instruction. In various examples, an entry also comprises a tag portion which stores data identifying the fetch pointer by which the entry is indexed. Branch prediction is performed by matching an index generated using a received fetch pointer to the tag portions to identify a matching entry and then determining the target address for the branch instruction from the data portion of the matching entry.

Abstract: A system and method process atomic instructions. A processor system includes a load store unit (LSU), first and second registers, a memory interface, and a main memory. In response to a load link (LL) instruction, the LSU loads first data from memory into the first register and sets an LL bit (LLBIT) to indicate a sequence of atomic instructions is being executed. The LSU further loads second data from memory into the second register in response to a load (LD) instruction. The LSU places a value of the second register into the memory interface in response to a store conditional coupled (SCX) instruction. When the LLBIT is set and in response to a store (SC) instruction, the LSU places a value of the second register into the memory interface and commits the first and second register values in the memory interface into the main memory when the LLBIT is set.

Abstract: Hardware structures for check pointing a main shift register one or more times which include a circular buffer used to store the data elements most recently shifted onto the main shift register which has an extra data position for each check point and an extra data position for each restorable point in time; an update history shift register which has a data position for each check point which is used to store information indicating whether the circular buffer was updated in a particular clock cycle; a pointer that identifies a subset of the data positions of the circular buffer as active data positions; and check point generation logic that derives each check point by selecting a subset of the active data positions based on the information stored in the update history shift register.

Abstract: Aspects relate to Input/Output (IO) Memory Management Units (MMUs) that include hardware structures for implementing virtualization. Some implementations allow guests to setup and maintain device IO tables within memory regions to which those guests have been given permissions by a hypervisor. Some implementations provide hardware page table walking capability within the IOMMU, while other implementations provide static tables. Such static tables may be maintained by a hypervisor on behalf of guests. Some implementations reduce a frequency of interrupts or invocation of hypervisor by allowing transactions to be setup by guests without hypervisor involvement within their assigned device IO regions. Devices may communicate with IOMMU to setup the requested memory transaction, and completion thereof may be signaled to the guest without hypervisor involvement. Various other aspects will be evident from the disclosure.

Abstract: A processor is configured to identify a branch instruction immediately followed by an architectural delay slot. A single bonded instruction comprising the branch instruction immediately followed by the architectural delay slot is created. The single bonded instruction is loaded into an instruction buffer.

Abstract: A method of sharing a plurality of registers in a shared register pool among a plurality of microprocessor threads begins with a determination that a first instruction to be executed by a microprocessor in a first microprocessor thread requires a first logical register. Next a determination is made that a second instruction to be executed by the microprocessor in a second microprocessor thread requires a second logical register. A first physical register in the shared register pool is allocated to the first microprocessor thread for execution of the first instruction and the first logical register is mapped to the first physical register. A second physical register in the shared register pool is allocated to the second microprocessor thread for execution of the second instruction. Finally, the second logical register is mapped to the second physical register.

Abstract: Techniques for executing a load instruction in a processor are described. In one example, load instructions which are detected to have an offset (or displacement) of zero are sent directly to a data cache, bypassing the address generation stage thereby reducing pipeline length. Load instructions having a nonzero offset can be executed in an address generation stage as is conventional. To avoid conflicts between a current load instruction with zero offset and a previous load instruction with nonzero offset, the current instruction can be rescheduled or sent through a separate dedicated load pipe. An alternative technique permits a load instruction with zero offset to be issued one cycle earlier than it would need to be if it had a nonzero offset, thus reducing load latency.