Abstract: An interconnect apparatus provides for the buffering of information in respective transaction buffers according to transaction type. An additional buffer is dynamically assignable to one of the transaction buffers where additional capacity is required by that transaction buffer.

Abstract: An interconnect apparatus provides for the buffering of information among a plurality of retry buffers in an output port. An additional buffer is dynamically assignable to one of the N retry buffer means where additional capacity is required by that retry buffer.

Abstract: An interconnect apparatus includes a transaction packet buffer and control logic. The control logic can be operable sequentially to write transaction packets for transmission to the transaction packet buffer and to transmit the buffered transaction packets in sequence to a destination. The control logic can further be operable on receipt of a control packet indicative of non-receipt by the destination of a transmitted transaction packet to retransmit the non-received transaction packet and transaction packets transmitted from the transaction packet buffer subsequent to the non-received transaction packet.

Abstract: An Infiniband device can be provided. The device can comprise an input port having a serialiser/deserialiser. The serialiser/deserialiser can comprise: a data buffer for storing data from a received serial data stream and for outputting the stored data in parallel groups and a code detector for detecting a predetermined code pattern in the serial data stream and generating a code detection output in response thereto. The serialiser/deserialiser can also comprise a transition detector for detecting transitions in the serial data stream and reconstructing a serial data clock therefrom, and for generating a plurality of parallel data clocks from the serial data clock, each parallel data clock having a different phase. The data buffer can be responsive to the code detection output to adjust a parallel data group start position within the serial data stream and to cause a selection of one of the reduced frequency clocks having a phase corresponding to the adjusted parallel data group start position.

Abstract: A graphic processor having an index processing unit for pre-processing a list of vertices making up a three-dimensional image. The method of pre-processing comprising the following steps. First, decomposing the three-dimensional image into a plurality of primitive elements each defined by a set of vertices, each vertex comprising vertex information stored in a vertex storage area and addressable by a vertex index. Then receiving said vertex indices and creating a set of unique indices identifying a batch of vertices and loading only the vertices corresponding to said unique indices into the vertex storage area. Finally creating transformed primitive elements from transformed vertex information addressed in the vertex storage area using the unique indices.

Abstract: An Infiniband device can be provided. The device can comprise an input port having a serialiser/deserialiser. The serialiser/deserialiser can comprise: a data buffer for storing data from a received serial data stream and for outputting the stored data in parallel groups and a code detector for detecting a predetermined code pattern in the serial data stream and generating a code detection output in response thereto. The serialiser/deserialiser can also comprise a transition detector for detecting transitions in the serial data stream and reconstructing a serial data clock therefrom, and for generating a plurality of parallel data clocks from the serial data clock, each parallel data clock having a different phase. The data buffer can be responsive to the code detection output to adjust a parallel data group start position within the serial data stream and to cause a selection of one of the reduced frequency clocks having a phase corresponding to the adjusted parallel data group start position.

Abstract: A graphic processor having an index processing unit for pre-processing a list of vertices making up a three-dimensional image. The method of pre-processing comprising the following steps. First, decomposing the three-dimensional image into a plurality of primitive elements each defined by a set of vertices, each vertex comprising vertex information stored in a vertex storage area and addressable by a vertex index. Then receiving said vertex indices and creating a set of unique indices identifying a batch of vertices and loading only the vertices corresponding to said unique indices into the vertex storage area. Finally creating transformed primitive elements from transformed vertex information addressed in the vertex storage area using the unique indices.