Menu
▾
▴
Re: [brlcad-devel] OpenCL GPGPU Raytracing
|
From: Abhishek V. <abh...@gm...> - 2018-04-02 01:47:25
|
Hello, I’m having trouble building the code with PCL. I’ve added the PCL 'find_package and include_directories’ in the CMakeLists (of the top parent directory) and during the build process, it keeps throwing some error related to struct _xmlBuf (while scanning the dependencies of target xml in file brlcad-7.2.4/misc/tools/xmltools/libxml/src/HTMLparser.c - screenshot attached). But. when I remove the PCL part from the CMakeLists, the code gets built. I’ll keep trying to resolve this issue. However, it would seem you have to review my proposal without the patch. Warm regards, Abhishek > On Mar 29, 2018, at 2:23 AM, Christopher Sean Morrison <br...@ma...> wrote: > > Jingchun, > > Are you currently working on a code patch? > > If you’re not, I would very much recommend it for you because it's difficult to tell from your proposal how adept you will be at working with BRL-CAD’s code and what practical OpenCL experience you have. Perhaps you can try converting the HYP primitive or implement an OpenCL unit test for rt_boolweave() or something else from Vasco’s comprehensive OpenCL to-do list. If you can submit something before Monday, that would be helpful. > > That goes for everyone who submitted a GSoC proposal. Please try to have all code patches and pull requests submitted as soon as possible if you’re providing supplementary productive coding evidence along with your proposal. Thank you to everyone that has worked so hard! It’s going to be a competitive selection. > > Cheers! > Sean > > > >> On Mar 26, 2018, at 9:40 PM, Vasco Alexandre da Silva Costa <vas...@gm... <mailto:vas...@gm...>> wrote: >> >> The TODO list is just a suggestion based on our knowledge (i.e. that >> of the maintainers) given the current status of OpenCL librt. In other >> words, we already know those are open issues, so we don't need to be >> convinced of the need for those changes. :-) >> Everyone is welcome to propose their own (new) changes, as long as >> they provide a proper rationale for them. >> >> We should have started this proposal discussion process earlier though... >> >> On Mon, Mar 26, 2018 at 9:41 PM, Jingchun Wang <jin...@uc... <mailto:jin...@uc...>> wrote: >>> Hi Vasco, >>> >>> Thank you for your comments. Yes I’d like to cooperate with one else and include OpenCL code optimization into plan according to .cl files given. >>> For items in TODO list, are they assigned to certain individual or open to any contributors? I mean, should I select ones from the list or choose something new into my Detailed Plan? >>> >>> Regards, >>> Jingchun >>> >>>> On Mar 26, 2018, at 2:33 PM, Vasco Alexandre da Silva Costa <vas...@gm... <mailto:vas...@gm...>> wrote: >>>> >>>> Ok, I've now managed to see your draft proposal (thanks for granting >>>> read access). >>>> >>>> You seem to have done your homework with regards to the CG aspects of >>>> the problem. However we typically expect implementation and planning >>>> detail more than CG theory in the proposal. You don't seem to have had >>>> enough time to look into BRL-CAD (source code) proper though. There >>>> already is a basic OpenCL rendering pipeline implementation with >>>> several primitives, the thing is, there are still more primitives to >>>> port and several further code optimizations could be done to improve >>>> performance in several stages of the rendering pipelines. >>>> >>>> I have a recently up to date summary of the status of the OpenCL >>>> raytracer code here: >>>> http://brlcad.org/wiki/User:Vasco.costa <http://brlcad.org/wiki/User:Vasco.costa> >>>> >>>> Since someone else is interested in the primitives work as well this >>>> year, we could split that work among you both, but I get that you have >>>> experience with HPC code optimization? In that case, you could include >>>> code optimization in your work plan. For example you could work on >>>> some of these elements of the TODO list I link to above: >>>> >>>> - Split struct hit in common.cl into two structs. One for shot() >>>> results and another for norm() results. This will reduce the amount of >>>> memory used to store temporary results between stages. (EASY) >>>> - Execute prefix sums and reductions in clt_frame() on the device to >>>> eliminate round-trips. (MEDIUM) >>>> - Refactor code so single-hit and multi-hit don't require recompiling >>>> all the sources twice. >>>> - Cache the compiled binaries so things aren't recompiled on every >>>> launch. (EASY) >>>> - Don't intersect primitives twice. This requires a dynamic memory >>>> allocator. (HARD) >>>> - Smarter kernel scheduling for reduce thread divergence. For example >>>> coallesce all the quadric intersections. (MEDIUM) >>>> - Spatial partitioning (perhaps a hybrid Kd-tree) in order to have >>>> early exit on scenes with high depth complexity. (HARD) >>>> >>>> The current OpenCL librt code is mostly under this part of the source code tree: >>>> https://svn.code.sf.net/p/brlcad/code/brlcad/trunk/src/librt/ >>>> It's in the .cl files and primitives/primitive_util.c >>>> >>>> You can learn more about how to use BRL-CAD with the quick reference card: >>>> http://brlcad.org/w/images/5/52/MGED_Quick_Reference_Card.pdf >>>> >>>> With regards to OpenCL, since you know C/C++ you can read these resources: >>>> https://www.khronos.org/files/opencl-1-2-quick-reference-card.pdf >>>> https://yosefk.com/blog/simd-simt-smt-parallelism-in-nvidia-gpus.html >>>> >>>> Regards, >>>> >>>> On Mon, Mar 26, 2018 at 8:08 PM, Vasco Alexandre da Silva Costa >>>> <vas...@gm...> wrote: >>>>> Hello, >>>>> You have uploaded your proposal to the GSoC website but you haven't >>>>> given us permissions to view your draft proposal so we can comment on >>>>> it before the deadline. >>>>> >>>>> We also expect students to introduce themselves to the community (in >>>>> our case this is typically done via this project mailing-list or the >>>>> project IRC chat channel). In this way we can get to know you and >>>>> discuss your work proposal to increase your chances of a smooth GSoC >>>>> experience. >>>>> >>>>> These steps are summarized here: >>>>> http://brlcad.org/wiki/Summer_of_Code/Checklist >>>>> >>>>> Also, we typically request that students provide a working proof that >>>>> they can modify the codebase, in your case, this can be as simple as >>>>> providing a patch to port one of the easier to port quadrics >>>>> primitives, like HYP (Hyperboloid), from ANSI C to OpenCL. It >>>>> shouldn't be too hard to do since you can use one of the other already >>>>> ported primitives (like ELL) as a template. You basically need to port >>>>> the ray-primitive intersection (shot) code from C to OpenCL and to >>>>> provide the glue code which packs and unpacks that primitive from the >>>>> C side to the OpenCL side. >>>>> >>>>> (sorry for re-sending this e-mail as I didn't put the mailing-list in >>>>> the CC properly) >>>>> >>>>> Regards, >>>>> >>>>> -- >>>>> Vasco Alexandre da Silva Costa >>>>> PhD in Computer Engineering (Computer Graphics) >>>>> Instituto Superior Técnico/University of Lisbon, Portugal >>>> >>>> >>>> >>>> -- >>>> Vasco Alexandre da Silva Costa >>>> PhD in Computer Engineering (Computer Graphics) >>>> Instituto Superior Técnico/University of Lisbon, Portugal >>> >> >> >> >> -- >> Vasco Alexandre da Silva Costa >> PhD in Computer Engineering (Computer Graphics) >> Instituto Superior Técnico/University of Lisbon, Portugal >> >> ------------------------------------------------------------------------------ >> Check out the vibrant tech community on one of the world's most >> engaging tech sites, Slashdot.org <http://slashdot.org/>! http://sdm.link/slashdot <http://sdm.link/slashdot> >> _______________________________________________ >> BRL-CAD Developer mailing list >> brl...@li... <mailto:brl...@li...> >> https://lists.sourceforge.net/lists/listinfo/brlcad-devel <https://lists.sourceforge.net/lists/listinfo/brlcad-devel> > ------------------------------------------------------------------------------ > Check out the vibrant tech community on one of the world's most > engaging tech sites, Slashdot.org! http://sdm.link/slashdot_______________________________________________ > BRL-CAD Developer mailing list > brl...@li... > https://lists.sourceforge.net/lists/listinfo/brlcad-devel |