摘要:
A parallel-code optimization system includes a siloed program reference-identifier and an intermediate representation (IR) updater. The siloed program reference identifier determines siloed program references in parallel code, wherein siloed program references are free of cross-thread interference. The IR updater modifies data-flow abstractions based on the identified siloed program references.
摘要:
A parallel-code optimization system includes a Procedural Concurrency Graph (PCG) generator. The PCG generator produces an initial PCG of a computer program including parallel code, and determines a refined PCG from the initial PCG by applying concurrency-type refinements and interference-type refinements to the initial PCG. The initial PCG and the refined PCG include nodes and edges connecting pairs of the nodes. The nodes represent defined procedures in the parallel code, and each edge represents a may-happen-in-parallel relation, and is associated with a set of lvalues that represents the immediate interference between the corresponding pair of nodes.
摘要:
A parallel-code optimization system includes a siloed program reference-identifier and an intermediate representation (IR) updater. The siloed program reference identifier determines siloed program references in parallel code, wherein siloed program references are free of cross-thread interference. The IR updater modifies data-flow abstractions based on the identified siloed program references.
摘要:
Techniques for cross-module in-lining are disclosed. In an embodiment, in-lining is done in conjunction with a 3-phase compiler including a front-end phase, an IPA (Inter-Procedural Analysis) phase, and a back-end phase. The front-end phase processes the source code in various modules and provides the intermediate representations of such source code. The IPA phase determines whether a function should be in-lined, and, if so, provides in-line transformation instructions for the back-end phase to execute. The back-end phase executes the instructions provided by the IPA, which, in effect, transforms the in-lining code.
摘要:
Techniques for partitioning modules into smaller groups are disclosed. In various embodiments, cross-module optimization may be performed on these smaller groups. Further, related modules are grouped together based on a predefined algorithm. Relatedness, or closeness, or affinity, of modules are considered based on various factors including, for example, the number of calls between routines in different modules, possibility of in-lining a first routine in a first module into a second routine in a second module, characteristics of parameters being passed between routines in different modules, etc.