Derek J. DeVries. A vectorizing suif compiler: Implementation and performance. Master's thesis, University of Toronto, 1997.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....multimedia extensions has been limited since application writers are largely restricted to using in line assembly routines or specialized library calls. One solution to this inconvenience is to employ vectorization techniques that have been used to parallelize scientific code for vector machines [11, 18, 19]. Since many multimedia applications are vectorizable, this approach promises good results. However, many important multimedia applications are not vectorizable. Furthermore, complicated loop transformation techniques such as loop fission and scalar expansion are required to parallelize loops that ....

Derek J. DeVries. A Vectorizing SUIF Compiler: Implementation and Performance. Master's thesis, University of Toronto, June 1997.

....to using in line assembly routines or specialized library calls. The problem is exacerbated by inconsistencies among different instruction sets. One solution to this inconvenience is to employ vectorization techniques that have been used to parallelize scientific code for vector machines [7, 16, 17]. Since a number of multimedia applications are vectorizable, this approach promises good results. However, many important multimedia applications are difficult to vectorize. Complicated loop transformation techniques such as loop fission and scalar expansion are required to parallelize loops that ....

Derek J. DeVries. A Vectorizing SUIF Compiler: Implementation and Performance. Master's thesis, University of Toronto, June 1997. 49

....to using in line assembly routines or specialized library calls. The problem is exacerbated by inconsistencies among different instruction sets. One solution to this inconvenience is to employ vectorization techniques that have been used to parallelize scientific code for vector machines [5, 14, 15]. Since a number of multimedia applications are vectorizable, this approach promises good results. However, many important multimedia applications are difficult to vectorize. Complicated loop transformation techniques such as loop fission and scalar expansion are required to parallelize loops that ....

Derek J. DeVries. A Vectorizing SUIF Compiler: Implementation and Performance. Master's thesis, University of Toronto, June 1997.

....that its inner loop has a trip count of 100. We also unrolled expand even though its two triple nested loops expand to 1024 iterations. The unrolled and interchanged vector code for the SVX FDX and LV processors was generated by our VSUIF vectorizing compiler from the source codes described above [11] except for expand. Our vector compiler is unable to automatically vectorize the loop interchanged or loop unrolled versions of expand. The problem is that this benchmark requires an array to be privatized before vectorization can occur and our compiler has no pass to perform privatization for ....

Derek J. DeVries. A Vectorizing SUIF Compiler: Implementation and Performance. Master's thesis, University of Toronto, June 1997.

....2.92 2.72 chroma colorspace composite convolve decrypt.inter decrypt.unroll Arithmetic Geometric Average Mean Figure 6. Processor Performance link the programs with the SGI C compiler. For vector execution, we compile the vectorizable functions with VSUIF, a vectorizing compiler we developed [6]. Because little effort has been put into optimizing the code generated by VSUIF, non vectorizable portions are compiled with the SGI compiler to benefit from its optimizing cabilities. The SGI linker is used to combine the separately compiled object files into an executable. To generate ....

Derek J. DeVries. A Vectorizing SUIF Compiler: Implementation and Performance. Master's thesis, University of Toronto, June 1997.

....use an ideal memory system, and all processors use R10000 latencies and repeat rates [32] Finally, a superscalar executable was generated using SGI s Version 5. 3 C compiler invoked with the O2 flag while a vector executable was generated using a vectorizing compiler that we developed [6]. The vectorizing compiler is based on the SUIF 1.1.0 compiler infrastructure [30] with the machsuif 1.1.0 extensions for specialized code generation and machine dependent optimizations [25] Similar to a superscalar executable, once a vector executable is created, it is binary compatible across a ....

....analyzing and fully understanding these initial results, we expect to apply this knowledge to produce similar performance results on a broader range of larger programs. For space reasons, we present the results for a single application: PGP decryption. Results for the full suite can be found in [6]. 3.1 PGP Decryption Code PGP (Pretty Good Privacy) is a public key encryption software package that was published as freeware in 1991 by Philip Zimmerman [19] In addition to satisfying the general criteria of being non scientific, nonfloating point but highly parallel, PGP was included in our ....

[Article contains additional citation context not shown here]

Derek J. DeVries. A Vectorizing SUIF Compiler: Implementation and Performance. Master's thesis, University of Toronto, June 1997.

No context found.

Derek J. DeVries. A vectorizing suif compiler: Implementation and performance. Master's thesis, University of Toronto, 1997.

No context found.

D. J. DeVries, "A vectorizing suif compiler: Implementation and performance," Master's thesis, University of Toronto, 1997.

No context found.

Derek J. DeVries. A Vectorizing SUIF Compiler: Implementation and Performance. Master's thesis, University of Toronto, June 1997.

No context found.

D. J. DeVries. A vectorizing suif compiler: Implementation and performance. Master's thesis, University of Toronto, 1997.

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