TIFF and LibTiff Mail List Archive


1994.10.06 03:47 "performance with new version on P5", by Kyriakos Georgiou
1994.10.11 16:11 "Re: performance with new version on P5", by Sam Leffler
1994.10.11 20:05 "g4 decomp. improved (3.3beta ALPHA 021)", by Kyriakos Georgiou

1994.10.11 20:05 "g4 decomp. improved (3.3beta ALPHA 021)", by Kyriakos Georgiou

According to my experiments G4 decompression is about 16.4% faster going from 3.3beta ALPHA 015 to 3.3beta ALPHA 021 on a particular test target I am using, but hopefully on most images. Earlier I reported ~3% gains, but it was just after time'ing the reading of just one g4 tiff under the old and new libtiff's (dont shoot me, I know that was bad 8). This time I run a tight loop of 32 images.

I have 32 IEEE Std 167A-1987 Faxcsimilie test chart tiff images which are already scanned. I made two versions of all of them, compressed and uncompressed. I compiled TIFF 3.3beta ALPHA 015 and TIFF 3.3beta ALPHA 021 under a 90Mhz P5 running Linux with 32Mb ram 32Mb swap. I used the pentium gcc with -mpentium -O4. (Makefile.gcc)

Then I recompiled a program that is optimised to the best of my programing ability :^) that reads the test chart tiffs, and evaluates scanning quality (what it does is not important, I just wanted to see the speedup in I/O). Anyway, there is no speedup when the images are compressed (logical), but g4 compression reading is a bit faster, this program runs about 16.4% faster on the same data, with 3.3beta ALPHA 021.

Timings for reading all 32 images:

TIFF 3.3beta ALPHA 015

0.71user 1.65system 0:02.36elapsed 100%CPU
22.03user 0.85system 0:23.09elapsed 99%CPU

TIFF 3.3beta ALPHA 021

0.71user 1.63system 0:02.38elapsed 98%CPU
18.67user 1.00system 0:19.84elapsed 99%CPU