![]() Of course that's just speculation since I haven't studied the asm dumps in shark in any real depth. The overhead from having to deal with this might be whats ending up killing any possible gains. I suspect the autovectorization is probably slower due to memory alignment not being preserved in speed critical areas of the LAME code, which is something I brought up earlier. I guess it's the same old story with the autovectorization (or hand vectorization for that matter) potential of the LAME source code on anything except for ICL. The speed results are pretty much in fact the exact opposite of what they should be. ![]() I didn't try any other compile options, and the difference in speed ranged across a 0.3x spread or so (fastest was like 3.6x on the test file I picked, using -preset standard). Sort of offtopic, but I did some tests compiling 3.96.1 with gcc 4.0 and gcc 3.3 and my results, from fastest to slowest, were:ฤก.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |