摘要:
A detector detects at least one kind of dependence in address between instructions executed by at least a processor, the detector being adopted to detect a possibility of presence of the at least one kind of dependence, wherein if the at least one kind of dependence is present in fact, then the detector detects a possibility of presence of the at least one kind of dependence, and if the at least one kind of dependence is not present in fact, then the detector may detect a pseudo presence of the at least one kind of dependence. The detector has an execution history storing unit with a plurality of entries and an address converter for converting an address of a memory access instruction into an entry number, where different addresses may be converted into entry numbers that are the same.
摘要:
When a processor executes a memory operation instruction by means of data dependence speculative execution, a speculative execution result history table which stores history information concerning success/failure results of the speculative execution of memory operation instructions of the past is referred to and thereby whether the speculative execution will succeed or fail is predicted. In the prediction, the target address of the memory operation instruction is converted by a hash function circuit into an entry number of the speculative execution result history table (allowing the existence of aliases), and an entry of the table designated by the entry number is referred to. If the prediction is “success”, the memory operation instruction is executed in out-of-order execution speculatively (with regard to data dependence relationship between the instructions). If the prediction is “failure”, the speculative execution is canceled and the memory operation instruction is executed later in the program order non-speculatively. Whether the speculative execution of the memory operation instructions has succeeded or failed is judged by detecting the data dependence relationship between the memory operation instructions, and the speculative execution result history table is updated taking the judgment into account.
摘要:
A detector detects at least one kind of dependence in address between instructions executed by at least a processor, the detector being adopted to detect a possibility of presence of the at least one kind of dependence, wherein if the at least one kind of dependence is present in fact, then the detector detects a possibility of presence of the at least one kind of dependence, and if the at least one kind of dependence is not present in fact, then the detector may detect a pseudo presence of the at least one kind of dependence. The detector has an execution history storing unit with a plurality of entries and an address converter for converting an address of a memory access instruction into an entry number, where different addresses may be converted into entry numbers that are the same.
摘要:
A detector detects at least one kind of dependence in address between instructions executed by at least a processor, the detector being adopted to detect a possibility of presence of the at least one kind of dependence, wherein if the at least one kind of dependence is present in fact, then the detector detects a possibility of presence of the at least one kind of dependence, and if the at least one kind of dependence is not present in fact, then the detector may detect a pseudo presence of the at least one kind of dependence. The detector has an execution history storing unit with a plurality of entries and an address converter for converting an address of a memory access instruction into an entry number, where different addresses may be converted into entry numbers that are the same.
摘要:
A program is divided into several instruction streams, and each of them is executed as a thread. A thread processor executed the thread. The thread generates another thread, but one thread is controlled to make a fork operation at most once. Each thread is terminated in the order of generations. A thread manager may be shared with the several thread processors or be distributed to the several thread processors. The thread manager includes a thread sequencer and a thread status table. The thread status table manages execution status of each thread processor and parent-child relation. The thread sequencer requests a thread generation and permits its termination in accordance with the content of the thread status table. The thread processor can execute a thread speculatively.
摘要:
A register content inheriting system contributes for realization of register content inheriting with a hardware of simple construction in a multithread multi-processor. Respective thread execution units and physical common register are provided. Using a register mapping table, a register number to be made reference to from each program is placed in the physical common register. Only as required in inheriting of register content, a relationship of the register mapping table is updated. Upon inheriting the content of the register, the content of the register mapping table is copied.
摘要:
A semiconductor integrated circuit of the invention comprises a plurality of cores, and an interconnecting network including adaptors connected to each of the cores and a plurality of routers connecting the adaptors to communicate therebetween. Transmission side adaptors store first delivery information, and control delivery of the request signal to be received from the first core in accordance with the first delivery information. Reception side adaptors store second delivery information, and control delivery of the request signal to be received through the interconnecting network to the second core in accordance with the second delivery information. The first delivery information and the second delivery information are hierarchically set.
摘要:
Provided is a performance optimization system that can identify a case where the impact on performance is large even when the number of cache misses is small. The performance optimization system includes: a required-period-of-time measurement unit that measures a required period of time concerning a to-be-observed access; a required-period-of-time table holding unit that holds a required-period-of-time table that consists of a plurality of table entries in which stored are measured values of the required period of time for each of classification regions produced by dividing a memory region for each of types based on the to-be-observed access to store a measured value of the required period of time; a table entry selection unit that makes a selection as to in which table entry, out of a plurality of table entries for each of the classification regions that make up the required-period-of-time table, the measured value of the required period of time is stored on the basis of the to-be-observed access; and a cache miss observation unit that detects the occurrence of a cache miss associated with the to-be-observed access.
摘要:
A router includes: a flit arrival time management section that records flit arrival time which is the time at which the packet is received for the first time, transmission interval of the packet which are acquired from a control packet transmitted prior to the first transmission of a packet and input and output channels of the control packet and requires a crossbar section for an output channel from which the packet is supposed to be output before the flit arrival time; a switch assignment section that performs arbitration on the output channel request and performs input/output connection relationship setting processing; and a switch assignment verification section that verifies whether a result of the input/output connection relationship setting processing coincides with the actual routing of the packet. The cross bar section performs switching of the arriving packet using a result of the input/output connection relationship processing.
摘要:
A semiconductor integrated circuit of the invention comprises a plurality of cores, and an interconnecting network including adaptors connected to each of the cores and a plurality of routers connecting the adaptors to communicate therebetween. The adaptors keep delivery information indicating a delivery condition of a request signal received from the cores and control delivery of the request signal received from the cores in accordance with the delivery information.