Hi, For the sake of keeping an already hefty code base reasonable (my filter will become the second largest one if accepted), I don't think that three separate loops (requiring 3 implementations of get_blur(...) and 3 #defines) are a wise idea. However, that part about stride and multiplication are interesting, and I do access the entries sequentially. The speed-up will probably be marginal, but it will only take a few lines of code, so I guess I will implement that. Sincerely, Robert Edele On Thu, 2005-03-03 at 13:35 +0200, Oded Shimon wrote:
On Thursday 03 March 2005 07:49, Yartrebo wrote:
As before, any comments or constructive criticism would be appreciated.
+#define get_part_of_pixel_yv12(mpi, x, y, plane) (*(mpi->planes[(plane)] + (mpi->stride[(plane)] * (y)) + (x)))
You shouldn't be using multiplication in time critical areas. The whole point of stride, is that you have a variable which sits on the line you are currently at, and for each pixel you add 'x', and every line you add 'stride':
uint8_t * tmp = dmpi->planes[0]; int x, y; for (y = 0; y < mpi->h; y++) { for (x = 0; x < mpi->w; x++) { do_stuff(tmp[j]); } tmp += dmpi->stride[0]; }
Repeat for U and V. You should probably use a separate loop for them. (I made the same mistake BTW, even worse :)
Other than that, patch looks good.. Haven't tested it though. BTW, since you seen to like comments so much, you might want to use doxygen comments. They are pretty much the same as you have now, just built for doxygen... Check DOCS/tech/code-documentation.txt .
- ods15
_______________________________________________ MPlayer-dev-eng mailing list MPlayer-dev-eng@mplayerhq.hu http://mplayerhq.hu/mailman/listinfo/mplayer-dev-eng