Abstract: Techniques and mechanisms for communicating information between clusters of a processor. In an embodiment, a processor core comprises circuitry to identify a condition wherein a first micro-operation (uop) of a uop sequence is to produce a value of an operand, a second uop of the uop sequence is to use the value of the operand, and different respective clusters of processor resources are to execute the first uop and the second uop. Based on the condition, the processor core supplements a strand of uops with a cross-cluster communication uop, wherein the strand comprises one of the first uop or the second uop. In another embodiment, one of the clusters executes the cross-cluster communication uop to provide a value of the operand to another cluster via a cross-cluster network.

Abstract: Techniques and mechanisms for a cache structure to facilitate time-efficient (re)use of an operand value in the execution of a micro-operation (?op). In an embodiment, a processor comprises a physical register file (PRF), and a cache which is coupled between a writeback path and the PRF. The cache functions as a holding buffer that stores operand values recently provided via one or more ports of the processor. A given operand value is available to be accessed in the cache for use in executing a ?op. The operand value is subject to being drained into the PRF, from which it can then be accessed for use in the ?op execution. In another embodiment, a reservation station is configured to track a readiness state of the ?op based on information which specifies that the cache is a currently a repository of the operand value.

Abstract: Techniques for a processor, method and system to implement a vector-integer store data avoidance scheme. An integer logic circuit dispatches an integer store data operation to route integer data to a destination on a first set of connection lines. A vector logic circuit schedules a vector store data operation to transfer vector data to the destination, block dispatch of an integer store data operation to the destination, route higher bits of the vector data via a second set of connection lines to a destination, and route lower bits of the vector data to the integer logic circuit and use the first set of connection lines used for routing of integer data to the destination to route the lower bits of the vector data to the destination.

Abstract: Methods, apparatus, and computer programs are disclosed to decode an integer division instruction. In one embodiment, a method comprises: receiving an instruction to perform an integer division in a set of arithmetic logic circuits, the instruction indicating a dividend and a divisor to obtain a quotient and a remainder; decoding the instruction into microoperations, including a first microoperation to store the quotient in a first register from the dividend and the divisor, and a second microoperation to store the remainder in a second register; renaming registers for the first and second microoperations as a group, wherein for the second microoperation, the renaming provides a first physical register to store a first value of the dividend and a second physical register to store a second value of the quotient obtained through execution of the first microoperation; and executing the first and second microoperations by the set of arithmetic logic circuits.

Abstract: An apparatus and method for efficiently processing denormals on a processor. For example, one embodiment of a processor comprises: a decoder to decode instructions of a program code sequence into mircooperations (uops); a control register to store one or more bits related to denormal processing; uop morphing circuitry to generate or select a first type of FP divide or square root uop when the one or more bits indicate no possibility of denormals and to generate or select a second type of FP divide or square root uop when the one or more bits indicate a possibility of denormals; and execution circuitry to execute the first type of FP divide or square root uop to generate a result or to execute the second type of FP divide or square root uop, handling any denormals in hardware, to generate the result.

Abstract: Disclosed embodiments relate to a method and apparatus for efficient matrix transpose. In one example, a processor to execute a matrix transpose instruction includes fetch circuitry to fetch the matrix transpose instruction specifying a destination matrix and a source matrix having (N×M) elements and (M×N) elements, respectively, a (N×M) load buffer, decode circuitry to decode the fetched matrix transpose instruction, and execution circuitry, responsive to the decoded matrix transpose instruction to, for each row X of M rows of the specified source matrix: fetch and buffer N elements of the row in a load register, and cause the N buffered elements to be written, in the same relative order as in the row, to column X of M columns of the load buffer, and the execution circuitry subsequently to write each of N rows of the load buffer to a same row of the load buffer.

Abstract: An apparatus and method for performing a vector permute.

Abstract: Disclosed embodiments relate to a method and apparatus for efficient matrix transpose. In one example, a processor to execute a matrix transpose instruction includes fetch circuitry to fetch the matrix transpose instruction specifying a destination matrix and a source matrix having (N×M) elements and (M×N) elements, respectively, a (N×M) load buffer, decode circuitry to decode the fetched matrix transpose instruction, and execution circuitry, responsive to the decoded matrix transpose instruction to, for each row X of M rows of the specified source matrix: fetch and buffer N elements of the row in a load register, and cause the N buffered elements to be written, in the same relative order as in the row, to column X of M columns of the load buffer, and the execution circuitry subsequently to write each of N rows of the load buffer to a same row of the load buffer.

Abstract: Embodiments of an invention for consecutive bit error detection and correction are disclosed. In one embodiment, an apparatus includes a storage structure, a second storage structure, a parity checker, an error correction code (ECC) checker, and an error corrector. The first storage structure is to store a plurality of data values, a plurality of parity values, and a plurality of ECC values, each parity value corresponding to one of the plurality of data values, a first bit of each parity value corresponding to a first of a plurality of portions of a corresponding data value, wherein the first of the plurality of portions of the corresponding data value is interleaved with a second of the plurality of portions of the corresponding data value, wherein a second bit of each parity value corresponds to a second of the plurality of portions of the corresponding data value, each ECC value corresponding to one of the plurality of data values.

Abstract: In an embodiment, a fused multiply-add (FMA) circuit is configured to receive a plurality of input data values to perform an FMA instruction on the input data values. The circuit includes a multiplier unit and an adder unit coupled to an output of the multiplier unit, and a control logic to receive the input data values and to reduce switching activity and thus reduce power consumption of one or more components of the circuit based on a value of one or more of the input data values. Other embodiments are described and claimed.

Abstract: Embodiments of an invention for consecutive bit error detection and correction are disclosed. In one embodiment, an apparatus includes a storage structure, a second storage structure, a parity checker, an error correction code (ECC) checker, and an error corrector. The first storage structure is to store a plurality of data values, a plurality of parity values, and a plurality of ECC values, each parity value corresponding to one of the plurality of data values, a first bit of each parity value corresponding to a first of a plurality of portions of a corresponding data value, wherein the first of the plurality of portions of the corresponding data value is interleaved with a second of the plurality of portions of the corresponding data value, wherein a second bit of each parity value corresponds to a second of the plurality of portions of the corresponding data value, each ECC value corresponding to one of the plurality of data values.

Abstract: In an embodiment, a fused multiply-add (FMA) circuit is configured to receive a plurality of input data values to perform an FMA instruction on the input data values. The circuit includes a multiplier unit and an adder unit coupled to an output of the multiplier unit, and a control logic to receive the input data values and to reduce switching activity and thus reduce power consumption of one or more components of the circuit based on a value of one or more of the input data values. Other embodiments are described and claimed.