<html> <head> <meta content="text/html; charset=ISO-8859-1" http-equiv="Content-Type"> </head> <body bgcolor="#FFFFFF" text="#000000"> <div class="moz-cite-prefix">Good python script and decrease the ocl_std header file size a lot. 1.There is a typo at backend/src/CMakeLists.txt, set (ocl_blob_c_file ${GBE_SOURCE_DIR}/src/ocl_stdlib_str.c) should be .cpp 2. ocl_vector.h seems to be generated every time in build but the ocl_as.h, ocl_convert.h need to be updated using the update script. I think we can unify the manner here. 3. about the pch file, I think I need to do some investigation in clang to find the exactly not compatible args. I think now you can just use the pch when no buiding args passed, and once there are some extra buiding args, totally fallback to the old manner. On 07/22/2013 04:46 PM, Zhigang Gong wrote: </div> <blockquote cite="mid:20130722084640.GF14543@gongzg-ivb" type="cite"> <pre wrap="">On Fri, Jul 19, 2013 at 05:41:07PM +0800, Zhigang Gong wrote: </pre> <blockquote type="cite"> <pre wrap="">Hi, As Beignet is using the SOA model, and we need to lower down all the vector builtin functions to the scalar version. This type of thing is ideal to use a script to generate all the code according to the OpenCL's spec. I just copy/paste most of the prototypes from the OpenCL spec and put them into builtin_vector_proto.def. Then I wrote a python script to parse the spec and generate all the vector inline functions and I removed all existing duplicate functions in ocl_stdlib.h. Now we have three parts of auto-generated code, the as/convert and this vector related functions. The amount of these three parts are much more than the other code in ocl_stdlib.h. To make the maintainance easier, I split the auto-generated code out of the ocl_stdlib.h, and refine the build process to generate one blob header file at build time. Then I also made a new patch based on Junyan's PCH patch to accelerate the cl kernel build speed. As this patchset is handling tons of those auto-generated code, the patch is relatively big. I just attached the compressed tar ball rather than embed them directly. Waiting for your comments. Thanks. </pre> </blockquote> <pre wrap=""> Sorry, that I sent the wrong attachment in last email. Now correct it here. </pre> <blockquote type="cite"> <pre wrap="">_______________________________________________ Beignet mailing list <a class="moz-txt-link-abbreviated" href="mailto:Beignet@lists.freedesktop.org">Beignet@lists.freedesktop.org</a> <a class="moz-txt-link-freetext" href="http://lists.freedesktop.org/mailman/listinfo/beignet">http://lists.freedesktop.org/mailman/listinfo/beignet</a> </pre> </blockquote> <pre wrap=""> </pre> <fieldset class="mimeAttachmentHeader"></fieldset> <pre wrap="">_______________________________________________ Beignet mailing list <a class="moz-txt-link-abbreviated" href="mailto:Beignet@lists.freedesktop.org">Beignet@lists.freedesktop.org</a> <a class="moz-txt-link-freetext" href="http://lists.freedesktop.org/mailman/listinfo/beignet">http://lists.freedesktop.org/mailman/listinfo/beignet</a> </pre> </blockquote> </body> </html>