Nested parallel language preprocessor for converting parallel language programs into sequential code

A method for producing machine readable code for execution on a parallel computer having two or more processors, the method comprising:

• with computer executed preprocessing code, converting a nested parallel program written in a nested parallel language to code in a sequential programming language and calls to a message processing interface, the nested parallel program including a split function to split the nested parallel program into recursive function calls capable of being executed in parallel on the processors and the nested parallel program including data parallel operations on a collection oriented data type having data elements distributed across the processors;
• compiling the sequential programming language code into machine executable code;
• linking the compiled sequential programming language code with run-time code that manages the split function, and run-time code that controls message passing functions among the processors in response to the calls to the message passing interface;
• with computer executed preprocessing code, identifying the split function, and generating programming code to assign each of the recursive function calls to a team of processors, such that each team independently executes one of the recursive function calls in a control parallel manner;
• and
• with computer executed preprocessing code, generating programming code to compute computational cost of the two or more recursive function calls, and generating code to compute the number of processors in each processor team assigned to the recursive function calls based on the computational cost of the recursive function calls.

An apparatus comprising:

• a preprocessor for converting a parallel program written in a nested parallel programming language with control parallel and data parallel operations to a sequential programming code and calls to a message passing interface for inter-processor communication on a parallel computer having two or more processors;
• wherein the parallel program includes a split function to split the nested parallel program into recursive function calls capable of being executed in parallel on the processors and further includes data parallel operations on a collection oriented data type having data elements distributed across the processors;
• wherein the preprocessor includes programming code for identifying the split function, for generating programming code to assign each of the recursive function calls to a team of processors, and for distributing a collection oriented data type across the processors in each team, such that each team independently executes one of the recursive function calls in a control parallel manner and the processors in each team operate in a data parallel manner on the collection oriented data type;
• and
• wherein the preprocessor includes programming code for inserting code to compute computational cost of the two or more recursive function calls, and for generating code to compute the number of processors in each processor team assigned to the recursive function calls based on the computational cost of the recursive function calls.