| Leonard F. Wisniewski. E#cient Design and Implementation of Permutation Algorithms on the Memory Hierarchy. PhD thesis, Dartmouth College Department of Computer Science, March 1996. 150 |
....Instead it generates a special call to the runtime system, providing the serial parameters and pcoord( dimensions as arguments. For such index expressions, these arguments completely describe the communication. The runtime system can execute such communications using a special purpose algorithm [Wis96, CSW99] In some cases, the index may describe a regular communication only for some parameter values. For example, a torus rotation, 1) n]a; has a serial parameter, n, whose value may not be known until runtime. The ViC compiler recognizes this pattern as a possible regular ....
....permutes the dimensions. XOR matches TRANSPOSE combined with exclusive or. The compiler generates a call to ViC left xor( with dim and xor for each index. When the dimensions are powers of two, this is a BMMC (Bit Matrix Multiply Complement) permutation, for which optimal algorithms are known [Wis96, CSW99] The compiler converts the left index expressions in Figure 6.8 into the runtimesystem calls in Figure 6.11. If the parameters describe a regular communication, the runtime system switches to a special purpose algorithm. The ViC runtime system does not fully support these communications ....
Leonard F. Wisniewski. E#cient Design and Implementation of Permutation Algorithms on the Memory Hierarchy. PhD thesis, Dartmouth College Department of Computer Science, March 1996. 150
Online articles have much greater impact More about CiteSeer.IST Add search form to your site Submit documents Feedback
CiteSeer.IST - Copyright Penn State and NEC