Patent ReferencesSystem and method for compiling a fine-grained array based source program onto a course-grained hardware Loop scheduler Intelligent loop unrolling Outer loop vectorization Software implemented method for thread-privatizing user-specified global storage objects in parallel computer programs via program transformation High-level loop fusion Patent #: 5822593 InventorAssigneeApplicationNo. 879210 filed on 06/19/1997US Classes:717/154Including analysis of programExaminersPrimary: Lintz, Paul R.Assistant: Chavis, John Q. International ClassG06F 009/45AbstractThe inventive system and method uses two relationships to update the distance vector after the loop strip-mining optimization has been performed by the compiler. The invention applies the original distance vector for the un-stripmined loop and the strip size from the strip-mining optimization to the relationships, and outputs either one or two distance vectors, depending upon whether the distance is a multiple of the strip size. The invention allows subsequent optimization to occur after strip-mining, and eliminates the need of having to normalize the stripmined loop to recompute the subscripts. The invention also eliminates the need to re-analyze the dependencies of the loop. | |