Usenet 1994 January/usenetsourcesnewsgroupsinfomagicjanuary1994.iso/sources/misc/volume26/pico/part01/README

Text File | 1991-12-14 | 5.0 KB | 128 lines

pico -- 1st approximation, Byron Rakitzis, 12/10/91.

This is a first stab at an implementation of pico, the picture editor
from Bell Labs described in Gerard Holzman's book "The Digital
Darkroom".

This program is *not* a full-blown implementation of pico. However, I
think the code is interesting enough that someone might be able to do
something quite useful with it without too much effort.

HOW TO BUILD PICO:

Pico runs on SGI's and Sun-4's. Edit the Makefile to get the right
compiler and LDFLAGS. On a Sun4, you will need gcc or Sun's ansi
cc, since pico is written with prototypes and new-style definitions.

HOW TO USE IT:

There are two options to pico, "-n" and "-d". "-n" suppresses the
display output, and "-d" provides an assembly dump of the generated
code after each expression is typed (using "dis" on the SGI and "adb"
on the Sun).

If you type "pico" you are prompted with a ">". You can now try typing
in a C expression involving x and y. The greyscale value of the point
[x,y] is determined by the value of the expression you type. e.g.,

    > x+y

creates a series of diagonal bands, since at x+y == 256 the grey
value wraps from white to black. x*y is striking, as is x^y, x*x+y*y
and many others.

You can refer to the current image as old[x,y], e.g.,

    > old[y,x]

rotates the current image 90 degrees. The "u" command performs one
level of undo, alternating between the current and previous images.

If you create a file that you like, you can write it out (in pgm
format) with the "w" command. Use double quotes to quote any special
characters in the filename. i.e.,

    > w foo
but
    > w "/tmp/foo"

Similarly, "r" can be used to read pgm files (which for now must be
512x512, 8-bit greyscale images).

Once you have a file, you can always refer to it by the shorthand "$n"
where n is the number of the file in the order it was read/written.
Type "f" to see what files you have in memory.

For example, to combine chris and dmr (the two sample bitmaps in the
ftp directory) so that chris fades into dmr from left to right, you can
try:

    > r dmr
    > r chris
    > (x*dmr + (X-x)*chris) / (2*X)
        [this can also be entered as]
    > (x*$1 + (X-x)*$2) / (2*X)

to reflect dmr's face along the line y=256, do:

    y < Y/2 ? dmr[x,Y-y] : dmr

(X refers to the constant 512, the maximum x coordinate. The other
built in constants are Y (obvious) and Z, the maximum brightness level,
255.)

Finally, to quit type "q" or control-D.

HOW IT WORKS:

The program is made up of a parser of C expressions, a simple rd code
generator for the MIPS and SPARC architectures, minimal display support
for the GL (Silicon Graphics) library, and file I/O in pgm format. The
image format is hardcoded at 512*512, 8-bit greyscale. An expression is
compiled straight into an array and then gets called by pico in order
to perform the operations on the image as fast as possible. This pico
is at least an order of magnitude faster (if not more) than the popi
posted to the net a few months back.

To use this program you must be on an SGI or a Sun. The Sun version
currently has no visual support, so you must explicitly write out pgm
files and use "xloadimage" or some similar program to view the pico
output.

THINGS TO DO:

I would consider extending the program in the following ways, though I
don't have the time for it right now:

1) Support different displays. Since the format of pico images is so
simple, 8-bit greyscale, it should be easy to pop up images on just
about any kind of computer with a color display. I have used pico
crudely under X-windows by writing a pgm file out to /tmp and then
running "xloadimage /tmp/pico" in order to view the file.

2) Extend the pico language. Right now only C expressions are
supported.  The only two variables recognized are x and y. This is a
drawback, but as I mentioned, this isn't a full-blown pico. The command
set is limited to r, w, f, u, q. (read, write, showfiles, undo, quit)

3) Add support for polar coordinates, transcendental functions, and so
on. The current implementation works only with the MIPS integer unit,
i.e., add, subtract, multiply, divide and the bit operations. The SPARC
port is even cruder; a function call is made for every multiply, divide
and mod operation, since there is no hardware multiply etc.  A future
version may address this problem by performing strength reduction on
loops, but this is really beyond the scope of the project I started.
(However, I do perform strength reduction on power of 2 multiplies
and divides by substituting a shift.)

4) Support color images. A very interesting project would be to support
arbitrary 32-bit color images, i.e., with alpha channel. This might
actually be of use to people who need to, say, boost the alpha channel
of a particular image file. This also entails supporting "real" image
file formats like rla.

4) Support new architectures. I wrote pico on the MIPS, and spent about
2 afternoons porting it to the SPARC. Consequently, the quality of the
SPARC port is probably quite mediocre. Porting to other RISC machines,
however should be equally straightforward.