Patent application number | Description | Published |
20100162262 | Split Scheduler - In an embodiment, a scheduler implements a first dependency array that tracks dependencies on instruction operations (ops) within a distance N of a given op and which are short execution latency ops. Other dependencies are tracked in a second dependency array. The first dependency array may evaluate quickly, to support back-to-back issuance of short execution latency ops and their dependent ops. The second array may evaluate more slowly than the first dependency array. | 06-24-2010 |
20100205384 | Store Hit Load Predictor - In one embodiment, a processor implements a store hit load predictor. The store hit load predictor is configured to monitor fetched ops in the processor, and is configured to detect stores that may have previously caused store hit load events. The store hit load predictor is configured to predict that the store will cause a store hit load event again, and is further configured to monitor subsequent fetched ops for the load. The store hit load predictor may locate the load using, e.g., an offset from the store to the load in the code sequence. In response to locating the load, the store hit load predictor may create a dependency of the load on the store, preventing the load from executing out of order with respect to the store. A store hit load event may be avoided in this fashion, at least in some cases. | 08-12-2010 |
20110289300 | Indirect Branch Target Predictor that Prevents Speculation if Mispredict Is Expected - In one embodiment, a processor implements an indirect branch target predictor to predict target addresses of indirect branch instructions. The indirect branch target predictor may store target addresses generated during previous executions of indirect branches, and may use the stored target addresses as predictions for current indirect branches. The indirect branch target predictor may also store a validation tag corresponding to each stored target address. The validation tag may be compared to similar data corresponding to the current indirect branch being predicted. If the validation tag does not match, the indirect branch is presumed to be mispredicted (since the branch target address actually belongs to a different instruction). The indirect branch target predictor may inhibit speculative execution subsequent to the mispredicted indirect branch until the redirect is signalled for the mispredicted indirect branch. | 11-24-2011 |
20120047332 | Combining Write Buffer with Dynamically Adjustable Flush Metrics - In an embodiment, a combining write buffer is configured to maintain one or more flush metrics to determine when to transmit write operations from buffer entries. The combining write buffer may be configured to dynamically modify the flush metrics in response to activity in the write buffer, modifying the conditions under which write operations are transmitted from the write buffer to the next lower level of memory. For example, in one implementation, the flush metrics may include categorizing write buffer entries as “collapsed.” A collapsed write buffer entry, and the collapsed write operations therein, may include at least one write operation that has overwritten data that was written by a previous write operation in the buffer entry. In another implementation, the combining write buffer may maintain the threshold of buffer fullness as a flush metric and may adjust it over time based on the actual buffer fullness. | 02-23-2012 |
20120290818 | Split Scheduler - In an embodiment, a scheduler implements a first dependency array that tracks dependencies on instruction operations (ops) within a distance N of a given op and which are short execution latency ops. Other dependencies are tracked in a second dependency array. The first dependency array may evaluate quickly, to support back-to-back issuance of short execution latency ops and their dependent ops. The second array may evaluate more slowly than the first dependency array. | 11-15-2012 |
20130103906 | Combining Write Buffer with Dynamically Adjustable Flush Metrics - In an embodiment, a combining write buffer is configured to maintain one or more flush metrics to determine when to transmit write operations from buffer entries. The combining write buffer may be configured to dynamically modify the flush metrics in response to activity in the write buffer, modifying the conditions under which write operations are transmitted from the write buffer to the next lower level of memory. For example, in one implementation, the flush metrics may include categorizing write buffer entries as “collapsed.” A collapsed write buffer entry, and the collapsed write operations therein, may include at least one write operation that has overwritten data that was written by a previous write operation in the buffer entry. In another implementation, the combining write buffer may maintain the threshold of buffer fullness as a flush metric and may adjust it over time based on the actual buffer fullness. | 04-25-2013 |
20130111191 | PROCESSOR INSTRUCTION ISSUE THROTTLING | 05-02-2013 |
20130138931 | MAINTAINING THE INTEGRITY OF AN EXECUTION RETURN ADDRESS STACK - A processor and method for maintaining the integrity of an execution return address stack (RAS). The execution RAS is maintained in an accurate state by storing information regarding branch instructions in a branch information table. The first time a branch instruction is executed, an entry is allocated and populated in the table. If the branch instruction is re-executed, a pointer address is retrieved from the corresponding table entry and the execution RAS pointer is repositioned to the retrieved pointer address. The execution RAS can also be used to restore a speculative RAS due to a mis-speculation. | 05-30-2013 |
20130151823 | NEXT FETCH PREDICTOR TRAINING WITH HYSTERESIS - A system and method for efficient branch prediction. A processor includes two branch predictors. A first branch predictor generates branch prediction data, such as a branch direction and a branch target address. The second branch predictor generates branch prediction data at a later time and with higher prediction accuracy. Control logic may determine whether the branch prediction data from each of the first and the second branch predictors match. If a mismatch occurs, the first predictor may be trained with the branch prediction data generated by the second branch predictor. A stored indication of hysteresis may indicate a given branch instruction exhibits a frequently alternating pattern regarding its branch direction. Such behavior may lead to consistent branch mispredictions due to the training is unable to keep up with the changing branch direction. When such a condition is determined to occur, the control logic may prevent training of the first predictor. | 06-13-2013 |