This archive contains a quick & dirty implementation of the IEEE Standard
1180-1990 accuracy test for inverse DCT.  It is not guaranteed to be
correct ... but if you find any bugs, please let me know (by email to
tgl@cs.cmu.edu).

The test harness consists of the C program ieeetest.c and shell script
doieee.  For comparison purposes I have also supplied a copy of jrevdct.c,
the inverse DCT routine from release 4 of the Independent JPEG Group's
free JPEG software.  (jrevdct.c is slightly modified from the IJG release
so that it will compile without the IJG include files.)  jrevdct.c passes
the 1180 test --- or at least, this program thinks so.  jrevdct.out is
the output from a test run.

Note that numerical results may vary somewhat across machines.  This appears
to be mostly due to differing results from the cosine function.


INSTALLATION:
  Check the Makefile, change CC and CFLAGS if needed.  Then say "make".
If your C compiler is non-ANSI, you may need to change includes and/or
function headers.

  To test a different IDCT routine, link with that routine instead of
jrevdct.o.  You will need to modify dct.h and/or ieeetest.c if your
routine's calling convention is not in-place modification of an array
of 64 "short"s.


USAGE:
  The standard test procedure is
	doieee ieeetest >outputfile
Expect it to take a while (almost 80 minutes on my old 68030 box).
Each of the six passes will emit a row of 100 dots as it runs.

You can grep the output for the word FAILS if you just want to know
yea or nay.


LEGAL MUMBO-JUMBO:
  I hereby release the test harness to the public domain.
			Thomas G. Lane, 22 Nov 1993

IMPORTANT: jrevdct.c is NOT public domain, but is copyrighted free software
(not the same thing at all).  It is subject to IJG's distribution terms, which
primarily state that if you incorporate it into a program you must acknowledge
IJG's contribution in your program documentation.  For more details and the
complete IJG software, see the IJG FTP archive at ftp.uu.net, in directory
/graphics/jpeg.