K. Kailas, K. Ebcioglu, and A. Agrawala. CARS: A New Code Generation Framework for Clustered ILP Processors. In Proceedings of the 7th HighPerformance Computer Architecture (HPCA-7), Nuevo Leone, Mexico, 2001, pp. pages 133-143.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....focus on acyclic schedules and perform the cluster assignment and instruction scheduling in two sequential steps [8, 10, 18] Some of them (such as [4] also handle constraints in terms of reduced connectivity between the registers and the functional units. Recently there have been other proposals [19, 24] for solving the cluster assignment and instruction scheduling in a single step. In the context of cyclic schedules (such as modulo scheduling) there have been a few proposals to solve the same problem. Some of them [23] perform the job in two sequential steps (cluster selection and instruction ....

K. Kailas, K. Ebcioglu, and A. Agrawala. Cars: A new code generation framework for clustered ilp processors. In Proc., 7th High-Performance Computer Architecture (HPCA--7), January 2001.

....non partitioned) architectures. A comparison of some of these techniques can be found in [5] There are several works related to acyclic code scheduling for clustered architectures [3, 7, 10, 21, 30] The most closely related work to our ideas include the work of Kailas, Ebcioglu and Agrawala [22]. They proposed an approach to cluster assignment, instruction scheduling and register allocation in a single compilation phase, all based on a list scheduling scheme. The approach taken in [22] differs from the approach presented in this paper in that they target instruction scheduling for ....

....21, 30] The most closely related work to our ideas include the work of Kailas, Ebcioglu and Agrawala [22] They proposed an approach to cluster assignment, instruction scheduling and register allocation in a single compilation phase, all based on a list scheduling scheme. The approach taken in [22] differs from the approach presented in this paper in that they target instruction scheduling for acyclic code and use different cluster assignment heuristics. A number of modulo scheduling approaches, targeting clustered VLIW architectures, have been recently proposed: Nystrom and Eichenberger ....

K. Kailas, K. Ebcioglu, and A. Agrawala. CARS: A New Code Generation Framework for Clustered ILP Processors. In Proc. of the 7th Int. Symposium on High Performance Computer Architecture, pages 133--143, 2001.

....of op1 is moved to register r1 in cluster 2 without being stored in cluster 1. If the cluster id and an extra destination field were attached to the results, the architecture would implement multicast writing the same result to a number of clusters as the sendb operation from the CRB scheme [18]. In the code below operation op1 writes its result to the local register r1 and to the register r1 in cluster 2. The latency of transporting the result to the cluster executing op2 is accounted for by delaying op2 till the third instruction. We used this variant of the extended results model in ....

....the inter cluster communication in all five ICC models. The scheduling unit is the guarded decision tree of basic blocks [23] 24] 11] The guarded decision tree is an acyclic control flow graph without join point, which is a more general case than superblocks [21] or traces [22] Inspired by [18], 19] and [20] we integrated cluster assignment, instruction scheduling, and register allocation in a single phase. Thanks to the integration of the phases our algorithm avoids the well known problem of phase coupling [18] 12] and yields a significantly denser code [19] The core of our ....

[Article contains additional citation context not shown here]

K. Kailas, K. Ebcioglu, et al., "CARS: A New Code Generation Framework for Clustered ILP processors", 7th International Symposium on High Performance Computer Architecture, pp. 133-134, Nuevo Leone Mexico, January 2001.

.... assignment and scheduling together by integrating assignment into a cycle driven list scheduler [23] CARS performs all three tasks assignment, scheduling, and register allocation in one step, by integrating both assignment and register allocation into a mostly cycledriven list scheduler [11]. In all these approaches, however, every decision is irrevocable and nal. In contrast, convergent scheduling provides a general framework that allows decisions to be postponed or reversed, and we have used it to address the phase ordering problem we found in cluster assignment when we have ....

K. Kailas, K. Ebcioglu, and A. K. Agrawala. CARS: A New Code Generation Framework for Clustered ILP Processors. In 7th International Symposium on High Performance Computer Architecture (HPCA), pages 133-143, 2001.

....schedules and perform the cluster assignment and instruction scheduling in two sequential steps [40] 41] 42] Some of them (such as [31] also handle constraints in terms of reduced connectivity between the registers and the functional units. Recently there have been other proposals [43] [44] to solve the cluster assignment and instruction scheduling in a single step. In the context of cyclic schedules (such as modulo scheduling) there have been a few proposals to solve the same problem. Some of them [45] perform the job in two sequential steps (cluster selection and instruction ....

K. Kailas, K. Ebcioglu, and A. Agrawala, "Cars: A new code generation framework for clustered ilp processors," in Proc., 7th High-Performance Computer Architecture (HPCA--7), January 2001.

....focus on acyclic schedules and perform the cluster assignment and instruction scheduling in two sequential steps [8, 10, 18] Some of them (such as [4] also handle constraints in terms of reduced connectivity between the registers and the functional units. Recently there have been other proposals [19, 24] for solving the cluster assignment and instruction scheduling in a single step. In the context of cyclic schedules (such as modulo scheduling) there have been a few proposals to solve the same problem. Some of them [23] perform the job in two sequential steps (cluster selection and instruction ....

K. Kailas, K. Ebcioglu, and A. Agrawala. Cars: A new code generation framework for clustered ilp processors. In Proc., 7th High-Performance Computer Architecture (HPCA--7), January 2001.

....to build a windowed CRB structure that has the same access time of the partitioned register file attached to it. However, CRB may need a larger area than a traditional register file. 3 Compilation scheme 3. 1 Overview of code generation framework We used the cars code generation framework [12, 8] which combines the cluster assignment, register allocation, and instruction scheduling phases to generate efficient code for clustered ILP processors. The input to the code generator is a dependence flow graph (DFG) 13] with nodes representing OPs and directed edges representing data control ....

.... OPs (lines 9 12) Otherwise, we schedule a sendb OP in the current basic block using the operation driven version of cars algorithm (line 5) 8] 4 Experimental results We used the chameleon VLIW research compiler [15] gcc version) with a new back end based on the cars code generation framework [12] for the experimental evaluation of our new partitioned register file scheme. We have developed a cycleaccurate simulator to model the clustered ILP processor with CRB and intercluster buses. We use the compiled simulation approach: each VLIW instruction is instrumented and translated into PowerPC ....

[Article contains additional citation context not shown here]

K. Kailas, K. Ebcioglu, and A. Agrawala, "cars: A New Code Generation Framework for Clustered ILP Processors," in Proceedings of the 7th International Symposium on High-Performance Computer Architecture (HPCA-7), pp. 133--143, 2001.

....that are live across a call OP so that caller save registers will not be allocated to them. We also identify and flag loop invariant DEFs, back edge DEFs and loop join DEFs of nodes in loops. This information will be used by CARS, for example to eliminate copy OPs along the back edges of loops [27, 26]. Physical registers are treated as a resource in CARS. Based on the input parametric description of the machine model, we initialize the register resource structures and their bit vector representations, local resource counters for function units and global resource counters for shared resources ....

....[10] representation so that independent OPs from multiple regions can be scheduled in the same vliws aggregate. Due to lack of space, we shall not further describe the details of operation driven CARS algorithm, tree VLIW scheduling, and scheduling of loops, which may be found elsewhere [27, 26]. 5 3. Implementation We have implemented a code generator based on CARS on top of CHAMELEON [37, 39] VLIW research testbed. The input to CHAMELEON is object code ( o files) produced by a modified version of gcc compiler. An object code translator processes these .o files to generate an ....

K. Kailas, K. Ebcioglu, and A. Agrawala. CARS: A New Code Generation Framework for Clustered ILP Processors. Technical Report UMIACS-TR-

No context found.

K. Kailas, K. Ebcioglu, and A. Agrawala. CARS: A New Code Generation Framework for Clustered ILP Processors. In Proceedings of the 7th HighPerformance Computer Architecture (HPCA-7), Nuevo Leone, Mexico, 2001, pp. pages 133-143.

No context found.

K. Kailas, K. Ebcioglu, and A. K. Agrawala. CARS: A new code generation framework for clustered ILP processors. In Proceedings of the Seventh International Symposium on High-Performance Computer Architecture, pages 133--143, January 2001.

No context found.

K. Kailas, K. Ebcioglu, and A. Agrawala. Cars: A new code generation framework for clustered ilp processors. In Proc., 7th High-Performance Computer Architecture (HPCA--7), January 2001.

No context found.

K. Kailas, K. Ebcioglu, and A. K. Agrawala. CARS: A new code generation framework for clustered ILP processors. In Proceedings of the Seventh International Symposium on High-Performance Computer Architecture, pages 133--143, January 2001.

No context found.

Krishnan Kailas, Kemal Ebcioglu, and Ashok K. Agrawala. CARS: A new code generation framework for clustered ILP processors. In HPCA, pages 133--144, 2001.

No context found.

K. Kailas, K. Ebcioglu, and A. Agrawala. CARS: A New Code Generation Framework for Clustered ILP Processors. In Proc. of the 7th Int. Symposium on High Performance Computer Architecture, pages 133--143, 2001.

No context found.

K. Kailas, K. Ebcioglu and A. Agrawala, "CARS: A New Code Generation Framework for CLustered ILP Processors", in Proc. 7th Int. Symp. on High-Performance Computer Architecture (HPCA-7) , Jan. 2001

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