C.N. Arnold. Performance Evaluation of Three Automatic Vectorizer Packages. ICPP, pages 235--242, 1982.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....which asserts the absence of sequentializing dependences. Some other examples of language support include parallel loop constructs such as doall and doacross [Cyt87] intrinsic array primitives, and complex communication primitives such as parallel prefix [Thi92] Compilers such as [AALL93, Arn82, THK93] find data parallelism in loops automatically. Data is often mapped across processors by the programmer, perhaps with some compiler assistance [THK93, CMZ92, Hig93, GOS94, BCKT79, KP96] Pedigree automatically parallelizes a single program for execution across multiple processors. The key ....

C.N. Arnold. Performance Evaluation of Three Automatic Vectorizer Packages. ICPP, pages 235--242, 1982.

....of these machines. These compilers are known as parallelizing compilers. Despite the wealth of research on new restructuring techniques, little work has been done on evaluating their effectiveness. Many early studies measured the success rate of automatic vectorizers on a suite of test loops [4, 6, 7, 10, 23, 26]. The need for more comprehensive studies has been pointed out, and more recent work has measured performance results of automatic parallelizers on a representative set of real programs [8, 13] However, very few papers have reported evaluation measures of individual restructuring techniques [9, ....

Clifford N. Arnold. Performance Evaluation of three Automatic Vectorizer Packages. In International Conference on Parallel Processing, pages 235--242, 1982.

.... implemented using out of order execution in the vector and scalar pipelines [Tan96] Register to Register : All current vector machines are register to register architectures and only a few early vector machines, the CDC Star 100 [HT72] and the TI ASC [Wat72] their successors, the Cyber 205 [Arn82, Arn83, HMD 86] and ETA 10 [Fat89] did have Platform, Tools and Benchmarks 21 memory to memory architectures. Since the register to register model has many clear advantages, as its commercial success shows, we will discard beforehand any further study of memory to memory machines. Among the ....

C. N. Arnold. Performance evaluation of three automatic vectorizer packages. ICPP82, pages 235--242, 1982. 216 Advanced Vector Architectures

....which asserts the absence of sequentializing dependences. Some other examples of language support include parallel loop constructs such as doall and doacross [Cyt87] intrinsic array primitives, and complex communication primitives such as parallel prefix [Thi92] Some compilers, such as [AALL93, Arn82, THK93] find data parallelism in loops automatically. Data is often mapped across processors by the programmer, perhaps with some compiler assistance [THK93, CMZ92, Hig93, GOS94, BCG 94, KP96] Our compiler, PEDIGREE, automatically parallelizes a single program across multiple processors. The ....

C. N. Arnold. Performance Evaluation of Three Automatic Vectorizer Packages. ICPP, 235--242, 1982.

....technique. The difference caused by loop unrolling is significant if we want to use branch probabilities after traditional optimizations have been applied. However, many programmers unroll loops by hand and other programmers use source to source restructuring tools, such as KAP [16] or VAST [2]. The differences evinced by these applications may render the fixed ordering of heuristics ineffective for some programs. 4 Evidence based Branch Prediction In this section, we propose a general framework for program based prediction. Our method, ESP, is generally described as follows. A body of ....

C. N. Arnold. Performance evaluation of three automatic vectorizer packages. Proceedings of the 1982 International Conference on Parallel Processing, pages 235--242, 1982.

....of these machines. These compilers are known as parallelizing compilers. Despite the wealth of research on new restructuring techniques, little work has been done on evaluating their effectiveness. Many early studies measured the success rate of automatic vectorizers on a suite of test loops [1, 2, 3, 7, 17, 18]. The need for more comprehensive studies has been pointed out, and more recent work has measured performance results of automatic parallelizers on a representative set of real programs [5, 10, 22] However, very few papers have reported evaluation measures of individual restructuring techniques ....

Clifford N. Arnold. Performance Evaluation of three Automatic Vectorizer Packages. In International Conference on Parallel Processing, pages 235--242, 1982.

....Table 2 summarizes one of the measurements, which compared the performance of the automatically restructured loops with that of handrestructured loops and also shows the number of loops whose automatic hand optimized performance ratio is higher than the threshold shown in the Table. Arnold [149] reports performance improvements produced by KAP, VAST, and FTN200, the Fortran compiler of the Cyber 200 machines, on 18 Livermore Loops. The measurements were taken on the Cyber 203 and 205 machines. A related study was done by Braswell and Keech [150] who use a set of 90 loops to evaluate ....

.... optimized programs on Cray Y MP [152] Third and fourth line: Improvements over serial program execution on Alliant FX8 [153] Fifth line: manual improvements over serial program execution on Alliant FX8 [154] Study Test Suite Measures Machines Compilers K A P V N T S I F [56] x x simulated [149] x x x x x Cyber 203 5 FTN200, KAP, VAST [158] x x simulated Parafrase [155] x x x simulated Parafrase [144, 143] x x x see Table 1 [150] x x x x Cyber 205 FTN200, KAP, VAST [145] x x see Table 1 [151] x x NAS 160 KAP, VAST [148] x x x x see Table 2 [153] x x x x x Alliant FX 8 KAP, VAST [152] x x ....

Clifford N. Arnold. Performance Evaluation of three Automatic Vectorizer Packages. In Proceedings of Int'l. Conf. on Parallel Processing, pages 235--242, 1982.

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