From owner-bug-octave at bevo dot che dot wisc dot edu Thu Dec 19 15:33:44 1996 Subject: Octave configuration options From: "John W. Eaton" To: "C. Burger" Cc: bug-octave at bevo dot che dot wisc dot edu Date: Thu, 19 Dec 1996 15:31:57 -0600 On 13-Dec-1996, C. Burger wrote: : I would like to discuss some configuration aspects for octave under : Linux. I have compiled octave, especially the prereleases, several : times with varying options during the last months. I came to the : conclusion that the configuration enclosed below gives me optimal : results. : : --enable-dl is crucial for me since I'm using this feature. : --enable-shared and --enable-lite-kernel produce the smallest binaries. : All the shared libs and also the .oct files are stripped. To save : space, I don't install the static libs. : : Performance-wise, I certainly want "-O2 -fomit-frame-pointer" in my : compiler FLAGS. -O3 produces faster code at the cost of larger : binaries. gcc-2.7.2.1 does not seem to suffer from the strength-reduce : bug so this optimization is not suppressed. : : For a pentium system, the proper alignments "-malign-loops=2 : -malign-jumps=2 -malign-functions=2" lead to significantly faster and : smaller binaries. Most of the 10-20% speed increase advertised for the : so-called pentium-enhanced gccs is due to this alignment. These : settings are added to the *cc1 section in gcc's specs file. I found : that cc1plus would also include the *cc1 specs entry but f771 would : not. This means, the proper aligment options have to be manually added : to the FFLAGS here. You are always free to experiment with whatever optimization flags you prefer. I'm not sure that putting this level of detail in the default configuration for all x86 systems is a good idea though. Is it still possible to debug programs compiled with -fomit-frame-pointer? jwe