Abstract: A data processing apparatus contains branch prediction circuitry including a micro branch target buffer, a full branch target buffer and a global history buffer. The branch target buffer entries contain history data which indicates whether or not a number of the following blocks of program instructions, subsequent to and sequential to a block of program instruction identified by that branch target buffer entry containing a branch instruction, do themselves contain any branch instructions. If the history data indicates that the following blocks of program instructions do not contain branches, then the operation of the branch prediction circuitry is suppressed for these following blocks of program instructions so as to save energy.

Abstract: A memory device and a method of operating the memory device are provided. The memory device comprises a plurality of storage units and access control circuitry. The access control is configured to receive an access request and in response to the access request to initiate an access procedure in each of the plurality of storage units. The access control circuitry is configured to receive an access kill signal after the access procedure has been initiated and, in response to the access kill signal, to initiate an access suppression to suppress the access procedure in at least one of the plurality of storage units. Hence, by initiating the access procedures in all storage units in response to the access request, e.g.

Abstract: A single threaded out-of-order processor 2 includes an architected register file 22 and a speculative register file 20. Speculative register allocation circuitry 24 serves to allocate speculative registers for use in accordance with an allocation sequence and taken from a position determined by a tail point. Read suppression circuitry 30 serves to maintain a boundary pointer corresponding to a position within the allocation sequence such that no speculative register more recently allocated within the allocation sequence than that corresponding to the boundary pointer can have a valid register value. The read suppression circuitry 30 serves to suppress read operations for source operands lying within a read-suppression region delimited by the tail point and the boundary pointer. Separate boundary pointers may be maintained for different types of register values, such as integer register values and floating point register values.

Abstract: A data processing system 2 includes a cache hierarchy having a plurality of local cache memories and a shared cache memory 18. State data 30, 32 stored within the shared cache memory 18 on a per cache line basis is used to control whether or not that cache line of data is stored and managed in accordance with non-inclusive operation or inclusive operation of the cache memory system. Snoop transactions are filtered on the basis of data indicating whether or not a cache line of data is unique or non-unique. A switch from non-inclusive operation to inclusive operation may be performed in dependence upon the transaction type of a received transaction requesting a cache line of data.

Abstract: A storage circuit and method are provided for propagating data values across a clock boundary between a first clock domain and a second clock domain. A storage structure is provided with at least one entry, and write circuitry performs write operations in the first clock domain, where each write operation writes a data value into an entry of the storage structure identified by a write pointer. The write circuitry alters the write pointer between each write operation. Write pointer synchronization circuitry then receives the write pointer and synchronizes the write pointer indication to the second clock domain over a predetermined number of clock cycles of the second clock domain. Read circuitry performs read operations in the second clock domain, with each read operation reading a data value from an entry of the storage structure identified by a read pointer.

Abstract: A data processing apparatus 2 contains branch prediction circuitry 10 including a micro branch target buffer 28, a full branch target buffer 30 and a global history buffer 32. The branch target buffer entries 40 contain history data 42, 44 which indicates whether or not a number of the following blocks of program instructions, subsequent to and sequential to a block of program instruction identified by that branch target buffer entry containing a branch instruction, do themselves contain any branch instructions. If the history data 42, 44 indicates that the following blocks of program instructions do not contain branches, then the operation of the branch prediction circuitry 28, 30, 32 is suppressed for these following blocks of program instructions so as to save energy.

Abstract: A single threaded out-of-order processor 2 includes an architected register file 22 and a speculative register file 20. Speculative register allocation circuitry 24 serves to allocate speculative registers for use in accordance with an allocation sequence and taken from a position determined by a tail point. Read suppression circuitry 30 serves to maintain a boundary pointer corresponding to a position within the allocation sequence such that no speculative register more recently allocated within the allocation sequence than that corresponding to the boundary pointer can have a valid register value. The read suppression circuitry 30 serves to suppress read operations for source operands lying within a read-suppression region delimited by the tail point and the boundary pointer. Separate boundary pointers may be maintained for different types of register values, such as integer register values and floating point register values.

Abstract: A memory device and a method of operating the memory device are provided. The memory device comprises a plurality of storage units and access control circuitry. The access control is configured to receive an access request and in response to the access request to initiate an access procedure in each of the plurality of storage units. The access control circuitry is configured to receive an access kill signal after the access procedure has been initiated and, in response to the access kill signal, to initiate an access suppression to suppress the access procedure in at least one of the plurality of storage units. Hence, by initiating the access procedures in all storage units in response to the access request, e.g.

Abstract: A storage circuit and method are provided for propagating data values across a clock boundary between a first clock domain and a second clock domain. A storage structure is provided with at least one entry, and write circuitry performs write operations in the first clock domain, where each write operation writes a data value into an entry of the storage structure identified by a write pointer. The write circuitry alters the write pointer between each write operation. Write pointer synchronisation circuitry then receives the write pointer and synchronises the write pointer indication to the second clock domain over a predetermined number of clock cycles of the second clock domain. Read circuitry performs read operations in the second clock domain, with each read operation reading a data value from an entry of the storage structure identified by a read pointer.

Abstract: A data processing system that manages data hazards at a coherency controller and not at an initiator device is disclosed. Write requests are processed in a two part form, such that a first part is transmitted and when the coherency controller has space to accept data, the data and a state of the data prior to a write are sent as a second part of a write request. When there are copending reads and writes to the same address, writes are stalled by not responding to the first part of a write request and snoop requests received to the address are processed regardless of the fact that the write is pending. When the pending read has completed, the coherency controller will respond to the first part of the write request and the initiator device will complete the write by sending the data and a state indicator following the snoop.

Abstract: A data processing apparatus 2 includes a plurality of transaction sources 8, 10 each including a local cache memory. A shared cache memory 16 stores cache lines of data together with shared cache tag values. Snoop filter circuitry 14 stores snoop filter tag values tracking which cache lines of data are stored within the local cache memories. When a transaction is received for a target cache line of data, then the snoop filter circuitry 14 compares the target tag value with the snoop filter tag values and the shared cache circuitry 16 compares the target tag value with the shared cache tag values. The shared cache circuitry 16 operates in a default non-inclusive mode. The shared cache memory 16 and the snoop filter 14 accordingly behave non-inclusively in respect of data storage within the shared cache memory 16, but inclusively in respect of tag storage given the combined action of the snoop filter tag values and the shared cache tag values.

Abstract: An integrated circuit 2 includes a plurality of transaction sources 6, 8, 10, 12, 14, 16, 18, 20 communicating via a ring-based interconnect 30 with shared caches 22, 24 each having an associated POC/POS 30, 34 and serving as a request servicing circuit. The request servicing circuits have a set of processing resources 36 that may be allocated to different transactions. These processing resources may be allocated either dynamically or statically. Static allocation can be made in dependence upon a selection algorithm. This selection algorithm may use a quality of service value/priority level as one of its input variables. A starvation ratio may also be defined such that lower priority levels are forced to be selected if they are starved of allocation for too long. A programmable mapping may be made between quality of service values and priority levels. The maximum number of processing resources allocated to each priority level may also be programmed.

Abstract: Initiator devices for generating transaction requests and recipient devices for receiving them are disclosed. The recipient devices accept transaction requests where there is available buffer storage for the transaction request. If there is no storage space available an acknowledgement signal generator generates and outputs a reject acknowledgement signal indicating a request has been received but has not been accepted by the recipient device. A credit generator can reserve at least one available storage location in the buffer and generate a credit grant for an initiator device that sent one of the transaction requests that was not accepted by the recipient device. The credit grant indicates to the initiator device that there is at least one reserved storage location, such that a subsequent transaction request from the initiator device will be accepted by the recipient device.

Abstract: A centralised synchronizing device for determining progress of at least a subset of transaction requests that are transmitted through a data processing system. A system synchronizing request is a request generated by one of the plurality of transaction generating devices and queries progress of a subset of the transaction requests. The synchronizing device includes: at least one port to and from the data processing system; a multicast circuitry configured to output a plurality of synchronizing requests for multicast to at least some of the devices within the data processing system where the requests query the progress of the subset of the transaction requests. Gather circuitry collects responses to the requests confirming that the queried progress has occurred at the respective devices. The gather circuitry determines when responses to all of the requests have been received and outputs a response to the system synchronizing request.

Abstract: A centralised synchronising device for determining progress of at least a subset of transaction requests that are transmitted through a data processing system in response to receipt of a system synchronising request, the data processing system having a plurality of devices including a plurality of transaction request generating devices for generating the transaction requests and a plurality of recipient devices for receiving the transaction requests, the synchronising device and at least one interconnect for interconnecting at least some of the devices; wherein the system synchronising request comprising a request generated by one of the plurality of transaction generating devices and querying progress of the at least a subset of the transaction requests; the synchronising device comprising: at least one port for receiving requests from, and outputting requests and responses to, the data processing system; multicast circuitry configured to generate a plurality of synchronising requests in response to recei

Abstract: An integrated circuit 2 includes a plurality of transaction sources 6, 8, 10, 12, 14, 16, 18, 20 communicating via a ring-based interconnect 30 with shared caches 22, 24 each having an associated POC/POS 30, 34 and serving as a request servicing circuit. The request servicing circuits have a set of processing resources 36 that may be allocated to different transactions. These processing resources may be allocated either dynamically or statically. Static allocation can be made in dependence upon a selection algorithm. This selection algorithm may use a quality of service value/priority level as one of its input variables. A starvation ratio may also be defined such that lower priority levels are forced to be selected if they are starved of allocation for too long. A programmable mapping may be made between quality of service values and priority levels. The maximum number of processing resources allocated to each priority level may also be programmed.

Abstract: A data processing system 2 includes a cache hierarchy having a plurality of local cache memories and a shared cache memory 18. State data 30, 32 stored within the shared cache memory 18 on a per cache line basis is used to control whether or not that cache line of data is stored and managed in accordance with non-inclusive operation or inclusive operation of the cache memory system. Snoop transactions are filtered on the basis of data indicating whether or not a cache line of data is unique or non-unique. A switch from non-inclusive operation to inclusive operation may be performed in dependence upon the transaction type of a received transaction requesting a cache line of data.

Abstract: Initiator devices for generating transaction requests and recipient devices for receiving them are disclosed. The recipient devices accept transaction requests where there is available buffer storage for the transaction request. If there is no storage space available an acknowledgement signal generator generates and outputs a reject acknowledgement signal indicating a request has been received but has not been accepted by the recipient device. A credit generator can reserve at least one available storage location in the buffer and generate a credit grant for an initiator device that sent one of the transaction requests that was not accepted by the recipient device. The credit grant indicates to the initiator device that there is at least one reserved storage location, such that a subsequent transaction request from the initiator device will be accepted by the recipient device.

Abstract: A data processing system that manages data hazards at a coherency controller and not at an initiator device is disclosed. The data processing system process write requests in a two part form, such that a first part is transmitted and when the coherency controller has space to accept data it responds to the first part and the data and state of the data prior to the write are sent as a second part of the write request. When there are copending reads and writes to the same address the writes are stalled by the coherency controller by not responding to the first part of the write and the initiator device proceeds to process any snoop requests received to the address of the write regardless of the fact that the write is pending. When the pending read has completed the coherency controller will respond to the first part of the write and the initiator device will complete the write by sending the data and an indicator of the state of the data following the snoop.

Abstract: A data processing apparatus 2 includes a plurality of transaction sources 8, 10 each including a local cache memory. A shared cache memory 16 stores cache lines of data together with shared cache tag values. Snoop filter circuitry 14 stores snoop filter tag values tracking which cache lines of data are stored within the local cache memories. When a transaction is received for a target cache line of data, then the snoop filter circuitry 14 compares the target tag value with the snoop filter tag values and the shared cache circuitry 16 compares the target tag value with the shared cache tag values. The shared cache circuitry 16 operates in a default non-inclusive mode. The shared cache memory 16 and the snoop filter 14 accordingly behave non-inclusively in respect of data storage within the shared cache memory 16, but inclusively in respect of tag storage given the combined action of the snoop filter tag values and the shared cache tag values.