A hardware-software co-designed processor includes a front end to decode an instruction, an execution unit to execute the instruction, an auxiliary cache to store auxiliary information for consumption during execution of the instruction, an instruction blender, and a retirement unit to retire the instruction. The auxiliary information may include long immediate values, non-working instructions for emulating an untranslated instruction stream, or execution hints, and is not decoded by the front end. The auxiliary cache includes circuitry to receive the auxiliary information from a binary translator, to store the auxiliary information in the auxiliary cache, and to provide the auxiliary information to the instruction blender prior to execution. The instruction blender includes circuitry to receive the auxiliary information, to blend the instruction with the auxiliary information, and to provide the blended instruction to the execution unit. Use of the auxiliary cache may reduce fetch and decode bandwidth requirements.

A microprocessor splits a fused multiply-accumulate operation of the form A*B+C into first and second multiply-accumulate sub-operations to be performed by a multiplier and an adder. The first sub-operation at least multiplies A and B, and conditionally also accumulates C to the partial products of A and B to generate an unrounded nonredundant sum. The unrounded nonredundant sum is stored in memory shared by the multiplier and adder for an indefinite time period, enabling the multiplier and adder to perform other operations unrelated to the multiply-accumulate operation. The second sub-operation conditionally accumulates C to the unrounded nonredundant sum if C is not already incorporated into the value, and then generates a final rounded result.

Aspects disclosed herein relate to combining instructions to load data from or store data in memory while processing instructions in processors. An exemplary method includes detecting a pattern of pipelined instructions to access memory using a first portion of available bus width and, in response to detecting the pattern, combining the pipelined instructions into a single instruction to access the memory using a second portion of the available bus width that is wider than the first portion. Devices including processors using disclosed aspects may execute currently available software in a more efficient manner without the software being modified.

A method is described for operating a microprocessor, in which a conversion software executed in the microprocessor carries out a binary translation, in the course of which a source instruction that is encoded according to a first instruction-set architecture is translated into a target instruction in a binary manner, which is encoded according to a second instruction-set architecture, and the target instruction translated by the translation software into the second instruction-set architecture being replicated, and in this replicated target instruction a memory area which is to be accessed in the course of the execution of the target instruction is replaced by a second memory area, and the target instruction and the copied target instruction is executed by the microprocessor. With the aid of the method, a temporal redundancy is achieved by a (temporally) parallel execution of the target instruction on a processor core and a local or regional redundancy by a parallel execution of the target instruction on different processor cores.

Implementing a neural network can include receiving a macro instruction for implementing the neural network within a control unit of a neural network processor. The macro instruction can indicate a first data set, a second data set, a macro operation for the neural network, and a mode of operation for performing the macro operation. The macro operation can be automatically initiated using a processing unit of the neural network processor by applying the second data set to the first data set based on the mode of operation.

In one example, a method includes responsive to receiving, by a processing unit, one or more instructions requesting that a first value be moved from a first general purpose register (GPR) to a third GPR and that a second value be moved from a second GPR to a fourth GPR, copying, by an initial logic unit and during a first clock cycle, the first value to an initial pipeline register, copying, by the initial logic and during a second clock cycle, the second value to the initial pipeline register, copying, by a final logic unit and during a third clock cycle, the first value from a final pipeline register to the third GPR, and copying, by the final logic unit and during a fourth clock cycle, the second value from the final pipeline register to the fourth GPR.

Methods and apparatuses relating to processors that contextually optimize instructions at runtime are disclosed. In one embodiment, a processors includes a fetch circuit to fetch an instruction from an instruction storage, a format of the instruction including an opcode, a first source operand identifier, and a second source operand identifier; wherein the instruction storage includes a sequence of sub-optimal instructions preceded by a start-of-sequence instruction and followed by an end-of-sequence instruction. The disclosed processor further includes a decode circuit to decode the instruction, to detect the start-of-sequence instruction and the end-of-sequence instruction, to buffer the sequence of sub-optimal instructions there between, to access a lookup table to identify one or more optimized instructions to substitute for one or more of the sequence of sub-optimal instructions, and to select either the decoded instruction or the sequence of one or more optimized instructions to dispatch to an execution circuit.

In accordance with embodiments disclosed herein, there are provided methods, systems, and apparatuses for scheduling instructions in a multi-strand out-of-order processor. For example, an apparatus for scheduling instructions in a multi-strand out-of-order processor includes an out-of-order instruction fetch unit to retrieve a plurality of interdependent instructions for execution from a multi-strand representation of a sequential program listing; an instruction scheduling unit to schedule the execution of the plurality of interdependent instructions based at least in part on operand synchronization bits encoded within each of the plurality of interdependent instructions; and a plurality of execution units to execute at least a subset of the plurality of interdependent instructions in parallel.

In response to determining an operation is a dependent operation, a mapper of a processor determines the source registers of the operation from which the dependent operation depends. The mapper translates the dependent operation to a new operation that uses as its source operands at least one of the determined source registers and a source register of the dependent operation. The new operation is independent of other pending operations and therefore can be executed without waiting for execution of other operations, thus reducing execution latency.

Techniques are disclosed relating to low-level instruction storage in a graphics unit. In some embodiments, a graphics unit includes execution circuitry, decode circuitry, hazard circuitry, and caching circuitry. In some embodiments the execution circuitry is configured to execute clauses of graphics instructions. In some embodiments, the decode circuitry is configured to receive graphics instructions and a clause identifier for each received graphics instruction and to decode the received graphics instructions. In some embodiments, the hazard circuitry is configured to generate hazard information that specifies dependencies between ones of the decoded graphics instructions in the same clause. In some embodiments, the caching circuitry includes a plurality of entries each configured to store a set of decoded instructions in the same clause and hazard information generated by the decode circuitry for the clause. This may reduce power consumption, in some embodiments, by reducing hazard checking when clauses are executed multiple times.