At 2013-09-02 16:22:28,"Tobias Grosser" <tobias at grosser.es> wrote:>On 09/01/2013 08:02 PM, Star Tan wrote: >> Hi all, >> >> >> Attached patch file to update lit config for Cloog. Without it, Polly always skips Cloog testings when we run "make check-polly". > >Dear Star Tan, > >thanks a lot for the patch. It looks very reasonable, but I am wondering >why it was not needed before or what problem it fixes exactly. Could you >add some information about this to the commit message. >I am not sure why it was not needed before; maybe it has never worked well before. The problem is that Polly never executes Cloog specific testcases no matther whether Cloog is found or not. I find this problem because I put a new testcase in test/Cloog/CodeGen/, but it is never executed. @Sebastian, you added the Cloog directory in r169159, including all testcases and the lit.local.cfg. The lit.local.cfg ensures that cloog specific testcases are executed only with CLOOG_FOUND by adding the following lit commands in test/Cloog/lit.local.cfg: cloog = config.root.cloog_found if cloog not in ['TRUE', 'true'] : config.unsupported = True However, there are two problems: First, since the cloog_found is set as "@CLOOG_FOUND@", I think the following "sed" command should be added into Makefile: sed -e "s#@CLOOG_FOUN@#$(CLOOG_FOUND)#g" Unfortunately such command is missed in current Polly. If it is not needed, I am curious how could Polly determine the value of @CLOOG_FOUND@ ? Second, even if we add the "sed" command in Makefile, the config.root.cloog_found would be set as "yes" or null, but Polly currently compares it with "TRUE" or "true", which thus always fails and the Cloog testcases will never be executed. @Sebastian, could you do me a favor to have a review of this problem? FYI, I have re-attached the patch file. Thanks, Star Tan -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130902/a7394d8e/attachment.html> -------------- next part -------------- A non-text attachment was scrubbed... Name: 0001-Update-lit-config-for-Cloog.patch Type: application/octet-stream Size: 2019 bytes Desc: not available URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130902/a7394d8e/attachment.obj>
On 09/02/2013 07:44 AM, Star Tan wrote:> At 2013-09-02 16:22:28,"Tobias Grosser" <tobias at grosser.es> wrote: > >> On 09/01/2013 08:02 PM, Star Tan wrote: >>> Hi all, >>> >>> >>> Attached patch file to update lit config for Cloog. Without it, Polly always skips Cloog testings when we run "make check-polly". >> >> Dear Star Tan, >> >> thanks a lot for the patch. It looks very reasonable, but I am wondering >> why it was not needed before or what problem it fixes exactly. Could you >> add some information about this to the commit message. >> > I am not sure why it was not needed before; maybe it has never worked well before. The problem is that Polly never executes Cloog specific testcases no matther whether Cloog is found or not. I find this problem because I put a new testcase in test/Cloog/CodeGen/, but it is never executed. > @Sebastian, you added the Cloog directory in r169159, including all testcases and the lit.local.cfg. The lit.local.cfg ensures that cloog specific testcases are executed only with CLOOG_FOUND by adding the following lit commands in test/Cloog/lit.local.cfg: > cloog = config.root.cloog_found > if cloog not in ['TRUE', 'true'] : > config.unsupported = True > However, there are two problems: > First, since the cloog_found is set as "@CLOOG_FOUND@", I think the following "sed" command should be added into Makefile: > sed -e "s#@CLOOG_FOUN@#$(CLOOG_FOUND)#g" > Unfortunately such command is missed in current Polly. If it is not needed, I am curious how could Polly determine the value of @CLOOG_FOUND@ ? > Second, even if we add the "sed" command in Makefile, the config.root.cloog_found would be set as "yes" or null, but Polly currently compares it with "TRUE" or "true", which thus always fails and the Cloog testcases will never be executed. > @Sebastian, could you do me a favor to have a review of this problem? FYI, I have re-attached the patch file.This is surprising. Even without your patch my test/lit.site.cfg file contains: config.enable_gpgpu_codegen = "" config.cloog_found = "TRUE" (With CLOOG enabled and GPGPU_codegen disabled). Also, the CLooG test are run (and are failing) as expected. I think it would be good to understand why this different behaviour can be observed. Sebastian, any ideas? Cheers Tobias
At 2013-09-03 00:12:56,"Tobias Grosser" <tobias at grosser.es> wrote:>On 09/02/2013 07:44 AM, Star Tan wrote: >> At 2013-09-02 16:22:28,"Tobias Grosser" <tobias at grosser.es> wrote: >> >>> On 09/01/2013 08:02 PM, Star Tan wrote: >>>> Hi all, >>>> >>>> >>>> Attached patch file to update lit config for Cloog. Without it, Polly always skips Cloog testings when we run "make check-polly". >>> >>> Dear Star Tan, >>> >>> thanks a lot for the patch. It looks very reasonable, but I am wondering >>> why it was not needed before or what problem it fixes exactly. Could you >>> add some information about this to the commit message. >>> >> I am not sure why it was not needed before; maybe it has never worked well before. The problem is that Polly never executes Cloog specific testcases no matther whether Cloog is found or not. I find this problem because I put a new testcase in test/Cloog/CodeGen/, but it is never executed. >> @Sebastian, you added the Cloog directory in r169159, including all testcases and the lit.local.cfg. The lit.local.cfg ensures that cloog specific testcases are executed only with CLOOG_FOUND by adding the following lit commands in test/Cloog/lit.local.cfg: >> cloog = config.root.cloog_found >> if cloog not in ['TRUE', 'true'] : >> config.unsupported = True >> However, there are two problems: >> First, since the cloog_found is set as "@CLOOG_FOUND@", I think the following "sed" command should be added into Makefile: >> sed -e "s#@CLOOG_FOUN@#$(CLOOG_FOUND)#g" >> Unfortunately such command is missed in current Polly. If it is not needed, I am curious how could Polly determine the value of @CLOOG_FOUND@ ? >> Second, even if we add the "sed" command in Makefile, the config.root.cloog_found would be set as "yes" or null, but Polly currently compares it with "TRUE" or "true", which thus always fails and the Cloog testcases will never be executed. >> @Sebastian, could you do me a favor to have a review of this problem? FYI, I have re-attached the patch file. > >This is surprising. Even without your patch my test/lit.site.cfg file >contains: > >config.enable_gpgpu_codegen = "" >config.cloog_found = "TRUE" > >(With CLOOG enabled and GPGPU_codegen disabled). Also, the CLooG test >are run (and are failing) as expected. > >I think it would be good to understand why this different behaviour can >be observed. Sebastian, any ideas? >Interesting! In my computer, it shows as: config.enable_gpgpu_codegen = "@CUDALIB_FOUND@" config.cloog_found = "@CLOOG_FOUND@" The source version is: LLVM: r189730; Polly: r189177; Clang:r189731 Best, Star Tan -------------- next part -------------- An HTML attachment was scrubbed... URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20130903/5282918c/attachment.html>