Intel Corporation

Institute for the Development of Emerging Architectures, L.l.c.

Idea Corporation

Po-Hua Chang

A method and apparatus for handling branch instructions contained within a source program includes applying a set of heuristics to classify each of the branch instructions in the source program as either a hard-to-predict type or a simple type of branch. A system implements a multi-heuristic branch predictor comprising a large, relatively simple branch predictor having many entries, to accommodate the majority of branch instructions encountered in a program, and a second, relatively small, sophisticated branch predictor having a few entries. The sophisticated branch predictor predicts the target addresses of the hard-to-predict branches. By mapping hard-to-predict branches to the sophisticated branch predictor, and easy-to-predict branches to the relatively simple branch predictor, overall performance is enhanced.

Method for identifying hard-to-predict branches to enhance processor performance

Method for identifying hard-to-predict branches to enhance processor

A method, performed by a computer, for optimizing a computer program having a plurality of instructions that form a loop. The loop has a first block and a second block. The first block has at least one instruction for using a variable and the second block has an ambiguous definition instruction for defining the variable. A load instruction for loading the variable into a register is inserted prior to the plurality of instructions that form the loop. In addition, at least one instruction for using the variable is replaced with an instruction for using the register. Further, a load instruction for loading the variable into the register is inserted after the ambiguous definition instruction.

Method for optimizing a computer program by moving certain load and

A method of facilitating optimization of computer program code. The code includes instructions for performing a plurality of loops, with each loop including at least one memory reference. The program code is instrumented to cause a memory reference trace to be generated when the program code is executed. The loops in the program code are identified from the trace. For each loop, a plurality of performance parameters relating to the memory references contained within the loop, including dynamic memory usage information and cache statistics, are computed based on the information in the trace. The performance parameters are then applied to a plurality of sets of conditions. Each set of conditions corresponds to one of a plurality of known optimization techniques to determine whether each set of conditions is satisfied for each loop. The optimization techniques include loop transformations and data restructuring techniques. For each set of conditions that is satisfied for a particular loop, an output is generated identifying the optimization technique which corresponds to that set of conditions as a suggested optimization for the loop.

Memory pattern analysis tool for use in optimizing computer program code

In a computer program, a branch instruction selects a prediction heuristic from a plurality of prediction heuristics for predicting whether the branch will be taken during execution of the program by a computer. A current pattern comprises a number of consecutive identical branch decisions for the instruction. A prior pattern comprises a number of consecutive identical prior branch decisions for the instruction, the prior branch decisions occurring prior to the branch decisions comprised by the current pattern. The selected prediction heuristic generates a branch prediction using the current pattern and the prior pattern. The selected prediction heuristic is identified by adding profiling instructions to the program to compute history information for the branch instruction. The profiling instructions input the branch history information to a plurality of prediction heuristics, and each prediction heuristic outputs a prediction of whether the branch instruction will be taken. The program is executed with a sample data set, and the output of each prediction heuristic is compared to the branch decision for the instruction to identify which heuristic most accurately predicts the branch decision for the branch instruction.

Growing community of inventors

Saratoga, CA, United States of America

Po-Hua Chang