… (which is erroring).
…ions. This time for packaging.
… the way benchmarks are run.
This commit adds support for building on MacOS X. There are two main changes needed for os x. * Use a .dylib extension for building, cleaning and installation * Use libtool to record dylib name with @rpath/ prefix The dylib extension was a standard change for os x. The libtool change is needed for the difference in dynamic linking behavior on os x. OS X differs from linux in that it the linker embeds the full path to the shared library when it records the dependency. Therefore to succesfully use -rpath to find the library, the library must say that its name should be found at @rpath/libintel_aes.dylib. The @rpath will be expanded by dyld at runtime to the full path, and an executable can link with a -rpath option to specify the search path.
…MASSACRED this working copy with spurious deletions.
…d twice. New version for this to try to fix build failure on hackage.
goal is to remove possible fragilities in the build process that would limit portability.
…ntation at runtime by checking CPU id.
…tion that caused me to run into a weird link problem. Namely, I moved the file to a subdir and moved the mkAESGen defs from SimpleRNGBench.hs into the GladmanAES.hsc file. I will dig deeper to figure out *which* of those changes caused me to trip the link error.
…es to make way for the compound AES implementation that checks the processor flags. HOWEVER now I'm running into odd link problems. Linking dist/build/benchmark-intel-aes-rng/benchmark-intel-aes-rng ... dist/build/benchmark-intel-aes-rng/benchmark-intel-aes-rng-tmp/Main.o: In function `sqM1_info': (.text+0xbb0a): undefined reference to `intelzmaeszm0zi1zi1_CodecziCryptoziConvertRNG_CRGtoRG0_con_info'
… the .so.
…etter one that modifies the PackageDescription data structure. Also began to make some progress on getting the AESNI package building under Mac OS.
Alas this means that I need to be careful not to check-in the rewritten version :(. Also, the current approach only works if cabal configure and cabal install are run separately.
…Buisson. These can be a separate package later. Right now I don't implement automatic fallback... but this commit is able to time both gladman and the AESNI version.
…d config that requires the build directory be /opt/intel-aes. ld-options/-rpath is the way that I did it. Hypothetically all I need to change from there is to figure out how to get the right paths. Having trouble at the moment accomplishing that...
… to work, including other-modules.
…at produces more readable output at least. I continued to hack on the .cabal file. Right now it's trying to add an extra-lib-dirs to the executable. This doesn't seem to affect the executable search path however....