<html><head><style type='text/css'>p { margin: 0; }</style></head><body><div style='font-family: Arial; font-size: 12pt; color: #000000'><font class="Apple-style-span" face="'times new roman', 'new york', times, serif"><br>Jon Forrest <jlforrest@berkeley.edu> wrote:<br><br>>On 2/1/2010 7:24 AM, richard.walsh@comcast.net wrote:<br>><br>>> Coming in on this late, but to reduce this work load there is PGI's version<br>>> 10.0 compiler suite which supports accelerator compiler directives. This<br>>> will reduce the coding effort, but probably suffer from the classical<br>>> "if it is<br>>> easy, it won't perform as well" trade-off. My experience is limited, but<br>>> a nice intro can be found at:<br>><br>>I'm not sure how much traction such a thing will get.<br>>Let's say you have a big Fortran program that you want<br>>to port to CUDA. Let's assume you already know where the<br>>program spends its time, so you know which routines<br>>are good candidates for running on the GPU.<br>><br>>Rather than rewriting the whole program in C[++],<br>>wouldn't it be easiest to leave all the non-CUDA<br>>parts of the program in Fortran, and then to call<br>>CUDA routines written in C[++]. Since the CUDA<br>>routines will have to be rewritten anyway, why<br>>write them in a language which would require<br>>purchasing yet another compiler?<br><br></font><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">Mmm ... not sure I understand the response, but perhaps this response</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">was to a different message ... ?? In any case, the PGI software supports</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">accelerator directives for both C and Fortran, so for those languages I do</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">not see a need to rewrite whole applications. The question presented is</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">the same as always, what does the performance-programming effort function</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">look like and how well does your code perform with directives to start</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">with. The PGI models is also hardware generic and the code runs on</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">the CPU in parallel when there is no GPU around I believe. What will</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">gate interest is how well PGI compiler group does at delivering performance</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">and how important portability is to the person developing the code.</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif"><br></font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">HMPP make offers a similar proposition ...</font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif"><br></font></div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">rbw</font></div><div><div><font class="Apple-style-span" face="'times new roman', 'new york', times, serif">_______________________________________________<br>Beowulf mailing list, Beowulf@beowulf.org sponsored by Penguin Computing<br>To change your subscription (digest mode or unsubscribe) visit http://www.beowulf.org/mailman/listinfo/beowulf</font><br></div></div></div></body></html>