Abstract: An integrated circuit comprises compute nodes arranged in an array; a torus topology network-on-chip interconnecting the compute nodes; and a network extension unit at each end of each row or column of the array, inserted in a network link between two compute nodes. The extension unit has a normal mode establishing the continuity of the network link between the two corresponding compute nodes, and an extension mode dividing the network link in two independent segments that are accessible from outside the integrated circuit.

Abstract: A method for executing instructions on a single-program, multiple-data processor system having a fixed number of execution lanes, including: scheduling a primary instruction for execution with a first wave of multiple data; assigning the first wave to a corresponding primary subset of the execution lanes; scheduling a secondary instruction having a second wave of multiple data, such that the second wave fits in lanes that are unused by the primary subset of lanes; assigning the second wave to a corresponding secondary subset of the lanes; fetching the primary and secondary instructions; configuring the execution lanes such that the primary subset is responsive to the primary instruction and the secondary subset is simultaneously responsive to the secondary instruction; and simultaneously executing the primary and secondary instructions in the execution lanes.

Abstract: A circuit for calculating the fused sum of an addend and product of two multiplicands, the addend and multiplicands being binary floating-point numbers represented in a standardized format as a mantissa and an exponent is provided. The multiplicands are in a lower precision format than the addend, with q>2p, where p and q are respectively the mantissa size of the multiplicand precision format and the addend precision format. The circuit includes a p-bit multiplier receiving the mantissas of the multiplicands; a shift circuit that aligns the mantissa of the addend with the product output by the multiplier based on the exponent values of the addend and multiplicands; and an adder that processes q-bit mantissas, receiving the aligned mantissa of the addend and the product, the input lines of the adder corresponding to the product being completed to the right by lines at 0 to form a q-bit mantissa.

Abstract: The present invention relates to a method for limiting the throughput of a communication in a meshed network, comprising the following steps: allocating fixed paths to communications likely to be established on the network; identifying the communications likely to take a mesh segment; allocating respective throughput quotas to the identified communications such that the sum of these quotas is less than or equal to a nominal throughput of said segment; and measuring the throughput of each communication at the input of the network and suspending the communication when its quota is reached.

Abstract: The present invention relates to a Network on chip comprising a torus matrix of processing elements formed by a juxtaposition of bricks in rows and columns, each brick comprising a longitudinal extra-connection bus segment connecting two terminals situated on opposite transverse edges of the brick on a first axis; two longitudinal intra-connection bus segments connecting circuits of the brick to respective terminals situated on the opposite transverse edges on a second axis symmetrical to the first axis with respect to the center of the brick; a transverse extra-connection bus segment connecting two terminals situated on opposite longitudinal edges of the brick on a third axis; and two transverse intra-connection bus segments connecting circuits of the brick to respective terminals situated on the opposite longitudinal edges on a fourth axis symmetrical to the third axis with respect to the center of the brick.

Abstract: The present invention relates to a torus network comprising a matrix of infrastructure routers, each of which is connected to two other routers belonging to the same row and to two other routers belonging to the same column; and input/output routers, each of which is connected by two internal inputs to two other routers belonging either to the same row, or to the same column, and comprising an external input for supplying the network with data. Each input/output router is devoid of queues for its internal inputs and comprises queues assigned to its external input managed by an arbiter which is configured to also manage the queues of an infrastructure router connected to the input/output router.

Abstract: An integrated circuit comprises compute nodes arranged in an array; a torus topology network-on-chip interconnecting the compute nodes; and a network extension unit at each end of each row or column of the array, inserted in a network link between two compute nodes. The extension unit has a normal mode establishing the continuity of the network link between the two corresponding compute nodes, and an extension mode dividing the network link in two independent segments that are accessible from outside the integrated circuit.

Abstract: The present invention relates to a Network on chip comprising a torus matrix of processing elements formed by a juxtaposition of bricks in rows and columns, each brick comprising a longitudinal extra-connection bus segment connecting two terminals situated on opposite transverse edges of the brick on a first axis; two longitudinal intra-connection bus segments connecting circuits of the brick to respective terminals situated on the opposite transverse edges on a second axis symmetrical to the first axis with respect to the center of the brick; a transverse extra-connection bus segment connecting two terminals situated on opposite longitudinal edges of the brick on a third axis; and two transverse intra-connection bus segments connecting circuits of the brick to respective terminals situated on the opposite longitudinal edges on a fourth axis symmetrical to the third axis with respect to the center of the brick.

Abstract: The present invention relates to a torus network comprising a matrix of infrastructure routers, each of which is connected to two other routers belonging to the same row and to two other routers belonging to the same column; and input/output routers, each of which is connected by two internal inputs to two other routers belonging either to the same row, or to the same column, and comprising an external input for supplying the network with data. Each input/output router is devoid of queues for its internal inputs and comprises queues assigned to its external input managed by an arbiter which is configured to also manage the queues of an infrastructure router connected to the input/output router.

Abstract: The present invention relates to a method for limiting the throughput of a communication in a meshed network, comprising the following steps: allocating fixed paths to communications likely to be established on the network; identifying the communications likely to take a mesh segment; allocating respective throughput quotas to the identified communications such that the sum of these quotas is less than or equal to a nominal throughput of said segment; and measuring the throughput of each communication at the input of the network and suspending the communication when its quota is reached.