Most probably those are problems also on upstream (ioquake3 vanilla) [since renderer is unchanged] or a compiler/compiler options issue.
Are you compiling it with any fancy options? [Since even the 'historical' options in Makefile look a bit doubtful] Lately I went the "trust the gcc programmers safe options" way and just used '-O3', and up to a -march=cpu_type and -fpmath=sse at most. [No cryptic options from the gcc 2.96 era and no 'unroll loops' that many report even may slow it down.]
Though I guess for one to be more confident with compiler options could do a test with a right-out -O0 only. [though theoretically up to a 'just -O3' it's considered 'relatively safe' by gcc]
[btw, I even avoided 'fast math' option since according to gcc manual it's not entered in -O3 [automatically] due to reliability issues; I know it had been tested in q3 for years but the compiler version has advanced a lot, i don't understand why it would be reliable, and there is no considerable gain anyway
I guess time for a question to their mailing list]
This post has been edited by mitsubishi: 19 May 2010 - 01:27 AM