OSX Multibeam OpenCL question

[Urs Echternacht]

All three builds are working correctly based on my tests. I have submitted the binaries to Eric Korpela. If you are curious, and want to play with the builds before he posts them, you can find them here: https://www.dropbox.com/s/hlh9m3z1cl6nlbl/seti%40home.zip?dl=0

Only version of these three i could try is the ATI HD5xxx. First thing i got was an error after the kernels were build successful. All needed info for analysis inside attached package. Guess you did not have ML and Lion in mind when building this app.

[Joe Fox]

Hi Urs, thanks for the debug info. I'll analyze and make adjustments.

[Urs Echternacht]

@JoeFox : TBar has tested also in the meantime and he couldn't run your ati app either. Neither on Yosemite or ML. (EDIT: retest with correct commandline call was successful on Yosemite.)

Awaiting your adjusted builds. If you need more help, some other tests, just ask.

[Joe Fox]

Hi Urs,

Attached are updated builds. The ATI one is now running correctly on an old macbook pro that I found which exhibited the crash you saw.

Thanks
-Joe

[Urs Echternacht]

Hi Urs,

Attached are updated builds. The ATI one is now running correctly on an old macbook pro that I found which exhibited the crash you saw.

Thanks
-Joe

Joe, the second app version of your seti_boinc-hd5xxx fails similarly to the first one on my Mac mini 5.2 (mid 2011 with discrete ATI Radeon HD 6630M onboard) with ML. Again i have attached the debug output and additional a comparison of your weak bound symbols and lazy bound symbols to one of my latest trys to build this app for the Mac. Mine is missing some signals, but has acceptable performance, while yours works correct but inaccepatble slow on Mavericks and Yosemite. If you check the comparison (done with OS X tool dyldinfo) you will find that weak and lazy symbols are somehow switched in the two builds. Also it still shows that your build is using "__DATA  __la_symbol_ptr  0x1002DB0F8 0x00DF libSystem        ___sincos_stret", which apparently is not present before OS X 10.9 (Mavericks). This makes it impossible to run your app on older OS X versions (10.7.4 (Lion) and 10.8.5 (Mountain Lion)).

[Joe Fox]

Hi Urs,

Thanks for the debug info. I changed the targetted SDK to 10.7, which should have removed the reference to sincos_stret, which is an optimization introduced in the compiler for builds targetted at 10.9 and up. I'm not sure how that symbol is still present, but will track down the cause and attach another update once I have verified that it is no longer being introduced.

Thanks.
-Joe

[Raistmer]

Is it possible to get all required changes into build scripts for OS X for committing back to Berkeley's repository?