On 09/11/2011 04:37 AM, Michael Dickens wrote: > Unless Josh changed something recently, the cmake build (with volk > disabled) worked for me under 10.6.8, XCode 3.2.3, 64-bit. It looks > like Ben is doing this testing on 10.5.8, XCode 3.1.4 -- which by > default will use 32-bit for either PPC or Intel compiling, and, > really, getting 64-bit was somewhat of a challenge until 10.6. A > "fat" or "universal" build o 10.5 meant 32-bit PPC and Intel merged > into a single file. IIRC, gcc under 10.5.8 could be coerced into > doing 64-bit, but doing so was much more reliable using Apple's > interface. All of that changed with 10.6, since 64-bit was the > default for any capable processor (and, all processors were Intel, > and mostly legacy ones were 32-bit only), and "universal" meant (and > still means) both 32- and 64-bit Intel compiling (merged). Given > that "we" designed the GR autotools build to work on OSX with both > 32- and 64-bit compiling, starting early in 10.5 days, I'm glad that > that build still works correctly. > > With the above I mind, it would probably help to know from Ben what > "arch" and "uname -a" return. Looking at Josh's pastebin change, I > would bet that CMAKE_SYSTEM_PROCESSOR is "i386" (it is for all Intel > OSX builds, as far as I know). But, under 10.5, > CMAKE_SYSTEM_PROCESSOR should probably be "i386", not "x86_64" -- > but, it really depends on what Ben's system is / provides. - MLD > >
This is only an issue w/ gr-filter where we try to perform arch detection. The way volk is doing it should always be ok :-) I made a branch detect_x86_type on jblum.git, where I try to detect the compiler's bit width in filter. I think its a lot more robust: http://gnuradio.org/cgit/jblum.git/log/?h=detect_x86_type -josh _______________________________________________ Discuss-gnuradio mailing list [email protected] https://lists.gnu.org/mailman/listinfo/discuss-gnuradio
