Abstract: A method and system for optimizing the execution of a software loop is provided. The method involves the determination of an edge in a critical recurrence cycle in the software loop. The edge is a dependency link between two instructions and contains a dependee and a dependent. The dependee is an instruction that produces a result, and the dependent is an instruction that uses the result. The method further involves performing predicate promotion of at least one of the dependee and the dependent if one or more pre-determined conditions are met.

Abstract: Various embodiments of the present invention relate to methods and systems for optimizing an intermediate code in a compilation logic. The intermediate code is optimized by performing reassociation in software loops. The intermediate code includes at least one critical recurrence cycle. The performance of reassociation in software loops can reduce a critical recurrence cycle in them, which can speed up their execution. The subject method can include the determination of one or more critical recurrence cycles in a software loop. The method can also include the determination of at least one edge in a critical recurrence cycle, with respect to which reassociation can be performed, if one or more pre-determined criteria are met. The method can further include performing reassociation of a dependee and a dependent of an edge. In an embodiment, when one or more pre-determined criteria are met, the logic of the software loop is maintained after performing reassociation of the dependee and the dependent of the edge.

Abstract: Disclosed are embodiments of a method and system for calculating an unrolling factor for software loops. The unrolling factor may be calculated by applying a formula that takes into account issue constraints of a processor. The issue constraints may include the total issue width of the processor, and may also include individual issue constraints for each instruction type. The software loop may be unrolled by the calculated unrolling factor and may be software pipelined. Other embodiments are also described and claimed.

Abstract: Disclosed are embodiments of a compiler, methods, and system for resource-aware scheduling of instructions. A list scheduling approach is augmented to take into account resource constraints when determining priority for scheduling of instructions. Other embodiments are also described and claimed.

Abstract: The present invention provides a mechanism that facilitates speculative execution of instructions within software-pipelined loops. In accordance with one embodiment of the invention, a software-pipelined loop is initialized with a speculative instruction deactivated. At least one initiation interval of the software-pipelined loop is executed, and the speculative instruction is activated. Subsequent initiation intervals of the software-pipelined loop are then executed.

Abstract: An efficient method for software-pipelining (SWP) of loops to translate programs, from higher level languages into equivalent object or machine language code for execution on a computer. In one example embodiment, this is accomplished by spilling and filling multiple computed values, in a register, that are live across multiple stages in a software-pipelined loop, using multiple rotating stack memory locations to reduce compiler-time of SWP, and complexity of the implemented SWP.

Abstract: Code restructuring or reordering based on profiling information and memory hierarchy is provided by constructing a Program Execution Graph (PEG) corresponding to a level of the memory hierarchy, partitioning this PEG to reduce estimated memory overhead costs below an upper bound, and constructing a PEG for a next level of the memory hierarchy from the partitioned PEG. The PEG is constructed from control flow and frequency information from a profile of the program to be restructured. The PEG is a weighted undirected graph comprising nodes representing basic blocks and edges representing transfer of control between pairs of basic blocks. The weight of a node is the size of the basic block it represents and the weight of an edge is the frequency of transition between the pair of basic blocs it connects.

Abstract: A method that uses software-pipelining to translate programs, from higher level languages into equivalent object or machine language code for execution on a computer, including sparse arrays/matrices. In one example embodiment, this is accomplished by transforming sparse array matrix source code and software-pipelining the transformed source code to reduce recurrence initiation interval, decrease run time, and enhance performance.

Abstract: A method is provided for processing nested loops that include a modulo-scheduled inner loop within an outer loop. The nested loop is scheduled to execute the epilog stage of the inner loop for a given iteration of the outer loop with the prolog stage of the inner loop for the next iteration of the outer loop. For one embodiment of the invention, this is accomplished by initializing an epilog counter for the inner loop to a value that bypasses draining the software pipeline. This causes the processor to exit the inner loop before it begins draining the inner loop pipeline. The inner loop pipeline is drained during the next iteration of the outer loop, while the inner loop pipeline fills for the next iteration of the outer loop.

Abstract: A method and apparatus are provided for repacking of memory data. For at least one embodiment, data for a plurality of store instructions in a source code program is loaded from memory into the appropriate sub-location of a proxy storage location. The packed data is then written with a single instruction from the proxy storage location into contiguous memory locations.

Abstract: A method is provided for processing nested loops that include a modulo-scheduled inner loop within an outer loop. The nested loop is scheduled to execute the epilog stage of the inner loop for a given iteration of the outer loop with the prolog stage of the inner loop for the next iteration of the outer loop. For one embodiment of the invention, this is accomplished by initializing an epilog counter for the inner loop to a value that bypasses draining the software pipeline. This causes the processor to exit the inner loop before it begins draining the inner loop pipeline. The inner loop pipeline is drained during the next iteration of the outer loop, while the inner loop pipeline fills for the next iteration of the outer loop.

Abstract: The present invention provides a mechanism for implementing compare speculation in software pipelined loops. A data dependency graph (DDG) is generated for a loop that includes a control compare instruction, a compare instruction and a non-speculative instruction that depends directly or indirectly on the compare instruction. A loop-carried edge between the control compare instruction and the compare instruction is replaced by a loop-carried edge between the control compare instruction and the non-speculative instruction. If the compare instruction is speculated when the loop is modulo-scheduled, any load instruction that depends on the compare is converted to a speculative load, and a loop-carried edge is added between the control compare and a check instruction associated with the speculative load. A loop-independent edge is also added between the check instruction and the non-speculative instruction if the non-speculative instruction also depends on the load.

Abstract: The invention is directed to the transformation of software loops having early exit conditions, thereby allowing the loops to be more effectively converted to a single basic block for software pipelining. The invention assigns a predicate register for each early exit condition of the software loop. The predicate registers are set when the corresponding early exit condition is satisfied. In this manner, when the loop terminates the predicate registers can be examined to indicate which early exit conditions were satisfied. The invention produces loops having a lower recurrence II and resource II than conventional techniques.

Abstract: The present invention provides a mechanism that facilitates speculative execution of instructions within software-pipelined loops. In accordance with one embodiment of the invention, a software-pipelined loop is initialized with a speculative instruction deactivated. At least one initiation interval of the software-pipelined loop is executed, and the speculative instruction is activated. Subsequent initiation intervals of the software-pipelined loop are then executed.

Abstract: The present invention provides a mechanism for prefetching array data efficiently from within a loop. A prefetch instruction is parameterized by a register from a set of rotating registers. On each loop iteration, a prefetch is implemented according to the parameterized prefetch instruction, and the address targeted by the prefetch instruction is adjusted. The registers are rotated for each loop iteration, and the prefetch instruction parameterized by the rotating register is adjusted accordingly. The number of iterations between prefetches for a given array is determined by the number of elements in the set of rotating register.

Abstract: Code restructuring or reordering based on profiling information and memory hierarchy is provided by constructing a Program Execution Graph (PEG) corresponding to a level of the memory hierarchy, partitioning this PEG to reduce estimated memory overhead costs below an upper bound, and constructing a PEG for a next level of the memory hierarchy from the partitioned PEG. The PEG is constructed from control flow and frequency information from a profile of the program to be restructured. The PEG is a weighted undirected graph comprising nodes representing basic blocks and edges representing transfer of control between pairs of basic blocks. The weight of a node is the size of the basic block it represents and the weight of an edge is the frequency of transition between the pair of basic blocks it connects. The nodes of the PEG are partitioned or clustered into clusters such that the sum of the weights of the nodes in any cluster is no greater than an upper bound.