摘要:
Systems and methods may provide a graphics processor that may identify operating conditions under which certain floating point instructions may utilize power to fewer hardware resources compared to when the instructions are executing under other operating conditions. The operating conditions may be determined by examining operands used in a given instruction, including the relative magnitudes of the operands and whether the operands may be taken as equal to certain defined values. The floating point instructions may include instructions for an addition operation, a multiplication operation, a compare operation, and/or a fused multiply-add operation.
摘要:
A system and method for fencing memory accesses. Memory loads can be fenced, or all memory access can be fenced. The system receives a fencing instruction that separates memory access instructions into older accesses and newer accesses. A buffer within the memory ordering unit is allocated to the instruction. The access instructions newer than the fencing instruction are stalled. The older access instructions are gradually retired. When all older memory accesses are retired, the fencing instruction is dispatched from the buffer.
摘要:
Various embodiments may be disclosed that may share a ROM pull down logic circuit among multiple ports of a processing core. The processing core may include an execution unit (EU) having an array of read only memory (ROM) pull down logic storing math functions. The ROM pull down logic circuit may implement single instruction, multiple data (SIMD) operations. The ROM pull down logic circuit may be operatively coupled with each of the multiple ports in a multi-port function sharing arrangement. Sharing the ROM pull down logic circuit reduces the need to duplicate logic and may result in a savings of chip area as well as a savings of power.
摘要:
A method and apparatus for a loop predictor for predicting the end of a loop is disclosed. In one embodiment, the loop predictor may have a predict counter to hold a predict count representing the expected number of times that a predictor stew value will repeat during the execution of a given loop. The loop predictor may also have one or more running counters to hold a count of the times that the stew value has repeated during the execution of the present loop. When the counter values match the predictor may issue a prediction that the loop will end.
摘要:
A cache has an array with single ported cells and is dynamically accessible simultaneously by multiple computing engines. In a further embodiment, the cache also has a tag array including a first address input, a second address input, and a shared mode input, and a data array electrically coupled to the tag array and including a first address input, a second address input, and a shared mode input.
摘要:
A method and apparatus for reducing snoop stall on an external bus. One method of the present invention comprises retrieving an address and a transaction attribute for a bus transaction during a first of a plurality of request phase packets of the bus transaction. Then it is determined whether the bus transaction is a snoopable memory transaction or not. If the bus transaction is a snoopable memory transaction, a snoop probe is dispatched during the first request phase packet of the transaction. Snooping devices are allowed additional bus clocks to respond to the snoop probe, thereby reducing the number of snoop stalls required to be inserted during the bus transaction.
摘要:
A method for charge sharing among data conductors of a bus. The bus has a first data conductor and a corresponding data conductor. The method includes detecting the logic levels on the first data conductor and the corresponding data conductor, and generating a charge sharing signal for sharing charge between the first data conductor and the corresponding data conductor.
摘要:
The present invention discloses a method and apparatus method for efficient utilization of write-combining buffers for a sequence of non-temporal stores to scattered locations. The method comprises: converting the sequence of non-temporal stores to stores to intermediate buffers; and grouping the stores to intermediate buffers into consecutive non-temporal stores. The consecutive non-temporal stores correspond to adjacent memory locations in the write-combining buffers.
摘要:
Briefly, in accordance with one or more embodiments, an apparatus comprises a processor to monitor cache utilization of an application during execution of the application for a workload; and a memory to store cache utilization statistics responsive to the monitored cache utilization. The processor is to determine an optimal cache configuration for the application based at least in part on the cache utilization statistics for the workload such that a smallest amount of cache is turned on for subsequent executions of the workload by the application.
摘要:
A system and method for fencing memory accesses. Memory loads can be fenced, or all memory access can be fenced. The system receives a fencing instruction that separates memory access instructions into older accesses and newer accesses. A buffer within the memory ordering unit is allocated to the instruction. The access instructions newer than the fencing instruction are stalled. The older access instructions are gradually retired. When all older memory accesses are retired, the fencing instruction is dispatched from the buffer.