Message ID | 20190601224719.32872-2-andreas.rheinhardt@gmail.com |
---|---|
State | New |
Headers | show |
On Sun, Jun 02, 2019 at 12:47:15AM +0200, Andreas Rheinhardt wrote: > The reasons are cosmetics and preparation for future patches that will > have even more cases. > > Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> > --- > libavcodec/startcode.c | 37 +++++++++++++++++++------------------ > 1 file changed, 19 insertions(+), 18 deletions(-) This is not the same code, the way memory is read changes, the index is changed to a pointer. Why are these changes done ? And if they are done, they should be benchmarked Thanks
Michael Niedermayer: > On Sun, Jun 02, 2019 at 12:47:15AM +0200, Andreas Rheinhardt wrote: >> The reasons are cosmetics and preparation for future patches that will >> have even more cases. >> >> Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> >> --- >> libavcodec/startcode.c | 37 +++++++++++++++++++------------------ >> 1 file changed, 19 insertions(+), 18 deletions(-) > > This is not the same code, the way memory is read changes, > the index is changed to a pointer. > Why are these changes done ? And if they are done, they should be > benchmarked > > Thanks > I'm sorry. I remember actually doing benchmarks of an earlier version [1] which suddenly was quite faster after switching from index to pointer, so I simply changed it. I must have forgotten to document it in the commit message. - Andreas [1]: https://github.com/mkver/FFmpeg/commits/start_2 Warning: Very raw. IIRC the whole unaligned stuff does not yet work for it and it definitely doesn't work on BE.
diff --git a/libavcodec/startcode.c b/libavcodec/startcode.c index 9efdffe8c6..a55a8fafa6 100644 --- a/libavcodec/startcode.c +++ b/libavcodec/startcode.c @@ -27,31 +27,32 @@ #include "startcode.h" #include "config.h" +#include "libavutil/intreadwrite.h" int ff_startcode_find_candidate_c(const uint8_t *buf, int size) { - int i = 0; + const uint8_t *start = buf, *end = buf + size; + #if HAVE_FAST_UNALIGNED - /* we check i < size instead of i + 3 / 7 because it is - * simpler and there must be AV_INPUT_BUFFER_PADDING_SIZE - * bytes at the end. - */ +#define READ(bitness) AV_RN ## bitness +#define MAIN_LOOP(bitness, mask1, mask2) do { \ + /* we check p < end instead of p + 3 / 7 because it is + * simpler and there must be AV_INPUT_BUFFER_PADDING_SIZE + * bytes at the end. */ \ + for (; buf < end; buf += bitness / 8) \ + if ((~READ(bitness)(buf) & (READ(bitness)(buf) - mask1)) \ + & mask2) \ + break; \ + } while (0) + #if HAVE_FAST_64BIT - while (i < size && - !((~*(const uint64_t *)(buf + i) & - (*(const uint64_t *)(buf + i) - 0x0101010101010101ULL)) & - 0x8080808080808080ULL)) - i += 8; + MAIN_LOOP(64, 0x0101010101010101ULL, 0x8080808080808080ULL); #else - while (i < size && - !((~*(const uint32_t *)(buf + i) & - (*(const uint32_t *)(buf + i) - 0x01010101U)) & - 0x80808080U)) - i += 4; + MAIN_LOOP(32, 0x01010101U, 0x80808080U); #endif #endif - for (; i < size; i++) - if (!buf[i]) + for (; buf < end; buf++) + if (!*buf) break; - return i; + return buf - start; }
The reasons are cosmetics and preparation for future patches that will have even more cases. Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@gmail.com> --- libavcodec/startcode.c | 37 +++++++++++++++++++------------------ 1 file changed, 19 insertions(+), 18 deletions(-)