Ideas: 1. Instead of using llvm-gcc -S -emit-llvm + llvm-as to get to the .rbc file for each.c file, just use llvm- gcc -c -emit-llvm. 2. To get from linked.rbc to linked.bc, we currently invoke llvm-dis, then 'gccas' (a shell script which invokes llvm-dis then opt). It would be much better to just run opt! Both of these would significantly reduce the disk I/O, disk space required, and cpu time requirements for running llvm-test. -Chris
I'm going to try to get this done for tonight's run.
This is now implemented, with these patches: http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070129/043789.html http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070129/043791.html http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070129/043792.html http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20070129/043793.html This significantly speeds up llvm-test, especially on the larger test cases. Even the smaller test suites are sped up by a second or two.