Content-type: text/html
xieperf - XIE server extension test and demo program
xieperf [-option...]
xieperf
accepts the options listed below:
Specifies which display to use.
Normally,
xieperf
references image files
located in the directory ``images'', which
xieperf
assumes
is located in your current directory. If the ``images'' directory is not in
your current directory, or the file has been renamed, use this option to specify
its location.
Some tests require the reception of an event such as FloNotify
to continue, and may cause
xieperf
to hang should these
events not be received. This option allows the user to specify a timeout value
which, if exceeded, will cause
xieperf
to give up waiting
for an event, and continue on with the next test in sequence. Should an event
timeout, a warning message will be printed to stderr. The default timeout
value is 60 seconds.
Runs the tests in synchronous mode.
Using this option gives the user the ability to run a subset
of the available tests and control the number of times the tests are executed
on an individual basis. This is thought to be especially useful for those
running
xieperf
for demonstration purposes. Using this
option causes
xieperf
to read commands specified in a script
file, or from stdin if
<file>
is ``-''. Tests
are specified by newline-terminated input lines of the form "command
[-reps
n
] [
-repeat
m
]". Characters following and
including "#" are treated as comments. See the
-mkscript
option.
Repeats each test
n
times (by default
each test is run 2 times). This option may be used in script files also, in
which case the script file -repeat overrides the command line option.
Specifies how long in seconds each test should be run (default
5 seconds).
Use a visual with <depth>
planes per pixel (default is
the default visual).
Use a GrayScale visual (default is the default visual).
Use a PseudoColor visual (default is the default visual).
Use a StaticGray visual (default is the default visual).
Use a StaticColor visual (default is the default visual).
Use a TrueColor visual (default is the default visual).
Use a DirectColor visual (default is the default visual).
If
xieperf
must be run in a window manager
environment, use this option to make
xieperf
aware of this.
If specified,
xieperf
will create a window, identical to
the size of the root window, and all further windows created by
xieperf
will be transient popup children of this window. If this
option is omitted,
xieperf
will set the override_redirect
attribute of all windows to ``True'' and will also do evil things such as
calling XInstallColormap. Using this option will cause the window manager
to (hopefully) obey window geometry hints specified by
xieperf.
Display a comprehensive list of techniques, by category, indicating
which of the techniques are supported by the XIE server.
Print test label to screen prior to calling any of the test
code. This allows the user to know which test is executing in case the test
hangs for some reason.
Be verbose when running event and error tests. Also, can be
used to catch and display information on any signals received during execution
of
xieperf. Note that this option is best used in a debugging
situation, or to validate that the error events received by
xieperf
are valid the first time the tests are executed on a new platform.
Run tests which test for event generation.
Run tests which test for error event generation.
Skip test calibration. This may be used when running
xieperf
in situations where execution timing is not important. Execution
times will not be reported by
xieperf
when this option
is enabled. The inner loop repeat count, additionally, is set to a value of
5 (but can be overridden by the
-reps
option).
Runs all tests. This may take a while, depending on the speed
of your machine, and its floating point capabilities. This option is ignored
if a script file is used.
Generate a list of the available tests for the
xieperf
program. In
x11perf, this list is normally displayed
in the usage statement. It was yanked from the usage of
xieperf
because it was too lengthy.
Generate a script file suitable for use with the script option.
If
-repeat
or
-reps
are also specified,
they will be automatically placed at the end of each command in the script.
The script is generated to stderr. See the
-script
command,
above.
Most test flos utilize a photomap resource for a source.
A photomap cache of up to n entries is controlled by
xieperf
to avoid having to constantly reload these images during test initialization.
The default cache size is 4. If a value less than the default is specified,
the cache size will be set to the default.
Generates just the descriptive labels for each test specified.
Use
-all
or
-range
to specify which tests
are included. See
x11perfcomp(1X)
for more details.
Pretend we are running
xieperf
while connected
to a DIS-only capable implementation of XIE. This will cause
xieperf
to execute those tests which only use protocol requests found in
the DIS subset of XIE, and bypass those which are not DIS compatible. If
xieperf
detects a DIS server, it will do this automatically, and
this option is ignored. Use
-all
or
-range
to specify the initial range of tests.
Runs all the tests starting from the specified name
test1
until the name
test2, including
both the specified tests. Some tests, like the event and error tests, also
require the
-errors
or
-events
options to
specified. This option is ignored if a script it used.
Fix the inner loop repetitions to n. This indicates how many
time the photoflo will be executed each time the test is run. This option
is overridden on a per-test basis if specified in a script. Typically,
xieperf
determines the ideal number of reps during each test's calibration
period.
Test generation of events. Requires
-events
option.
Test generation of errors. Requires
-errors
option.
-ColorList
Create and destroy ColorList resource test.
Create and destroy LUT resource test.
Create and destroy Photomap resource test.
Create and destroy ROI resource test.
Create and destroy Photospace test.
Create and destroy Photoflo test.
Query Photomap resource test.
Query ColorList resource test.
Query techniques as specified by test name.
Query photoflo test.
PurgeColorList test.
This tests creates a photoflo which is started and blocks
for data provided by
PutClientData(). Instead of sending
the data, the test uses
XieAbort()
to stop the photoflo,
and then waits for PhotofloDone event to be send by the server. If the test
times out waiting for the event, a error message is sent to stderr.
This test creates a flo of the form ImportClientLUT ->
ExportLUT,
and starts the flo executing.
xieperf
then forks, and the
child process streams the LUT data to the flo using PutClientData, while the
parent blocks in XieAwait. If the flo successfully finishes, XieAwait will
return and the flo state, after query, will indicate that it has completed.
If XieAwait does not complete naturally, or after return from XieAwait the
flo is still active, an error is reported to stderr. Note, on a really slow
machine, it is possible that XieAwait will return before the flo has a chance
to finish. In this case, use the
-timeout
option to increase
the timeout for this test.
ImportClientLUT ->
ExportLUT test.
Flos of the form ImportClientPhoto ->
ExportPhotomap using
various decode techniques, for example, G32D, TIFF2, UncompressedTriple, and
so forth.
ImportClientROI with 10 rectangles.
ImportClientROI with 100 rectangles.
Flos of the form ImportPhotomap ->
ExportPhotomap using various
encode techniques, for example, G32D, TIFF2, UncompressedTriple, and so forth.
Original encoding is shown in left window, image after encoding is shown in
right window.
Two flos, one of the form ImportPhotomap ->
ExportClientPhoto,
and the other of the form ImportClientPhoto ->
ExportPhotomap, where ExportClientPhoto
in the first flo uses various encode techniques, for example, G32D, TIFF2,
UncompressedTriple, and so forth. The image before encoding is displayed in
the left window, while the right window shows the image which was encoded
in the first flo and read back in the second flo.
ExportClientLUT test. LUT is displayed in a histogram window.
ExportClientROI test, 10 ROIs. The ROIs which are sent to
the server are represented by the filled rectangles. The ROIs which are received
back from the server by the client are drawn as white bordered non-filled
rectangles. The resulting output illustrates how the server combined the rectangles
sent to it.
Same as exportclientroi1, except using 100 rectangles.
ExportClientHistogram tests using various images. The histogram
is displayed in a window which overlaps the image.
Same as the ExportClientHistogram test, but using a ROI to
identify the area of interest.
Same as the ExportClientHistogram test, but using a Control
Plane to identify the area of interest.
Test ImportLUT element, LUT size is 256.
ImportPhotomap ->
ExportPhotomap, with source and destination
equal.
ImportPhotomap ->
ExportDrawable, window destination.
ImportROI ->
ExportROI, 10 rectangles, source and destination
ROIs equal.
ImportROI ->
ExportROI, 100 rectangles, source and destination
ROIs equal.
ImportDrawable ->
ExportDrawable, Source is pixmap, destination
is window.
ImportDrawable ->
ExportDrawable, Source and destination is
window.
ImportDrawable ->
ExportDrawable, Destination window obscured
by source window.
ImportDrawable ->
ExportDrawable, Source window obscured by
destination window.
ImportDrawablePlane ->
ExportDrawablePlane, pixmap, source
= destination.
ImportDrawablePlane ->
ExportDrawablePlane, window, source
= destination.
ImportDrawablePlane ->
ExportDrawablePlane, window, source
obscures destination.
ImportDrawablePlane ->
ExportDrawablePlane, window, destination
obscures source.
Constrain HardClip technique test, drawable destination.
Constrain ClipScale technique test, drawable destination.
Constrain HardClip technique test, photomap destination.
Constrain ClipScale technique test, photomap destination.
Boxcar 3x3 convolution test. Smoothing or lowpass filter.
Boxcar 5x5 convolution test. Smoothing or lowpass filter.
LaPlacian 3x3 convolution test. Edge or highpass filter.
LaPlacian 5x5 convolution test. Edge or highpass filter.
LaPlacian 3x3 convolution test, with ROI.
LaPlacian 5x5 convolution test, with ROI.
LaPlacian 3x3 convolution test, with Control Plane.
LaPlacian 5x5 convolution test, with Control Plane.
Various tests which exercise the Math element, some tests
using ROIs and control planes.
Arithmetic element tests, using photomaps as the operands.
Arithmetic element tests, photomap and constant operands.
Arithmetic element tests, using photomaps as the operands,
with ROIs.
Arithmetic element tests, photomap and constant operands,
with ROIs.
Arithmetic element tests, using photomaps as the operands,
with Control Planes.
Arithmetic element tests, photomap and constant operands,
with Control Planes.
Arithmetic element tests, using photomaps as the operands,
unconstrained.
Arithmetic element tests, photomap and constant operands,
unconstrained.
Arithmetic element tests, photomaps as the operands, ROIs,
unconstrained.
Arithmetic element tests, photomap and constant operands,
ROIs, unconstrained.
BandSelect element test. Image input is triple band. If visual
of
xieperf
window is a color visual, then three BandSelect
elements are used to extract the individual bands, they are combined once
again using BandCombine, and displayed using ConvertToIndex. If the visual
is not color, for example, GrayScale or StaticGray, then the flo simply uses
one BandSelect element to extract a single band for display.
BandCombine test. Input bands are made of three separate single
band photomaps. These are combined using a BandCombine element, which is followed
by a BandExtract and ExportDrawable. CCIR 601-1 coefficients.
BandExtract test. Input is a triple band photomap. CCIR 601-1
coefficients. Destination window colormap is gray ramp.
BandExtract test. Input is a triple band photomap. CCIR 601-1
coefficients. Destination window colormap is RGB_BEST_MAP standard colormap.
BandExtract test. Input is a triple band photomap. CCIR 601-1
coefficients. Destination window colormap is RGB_DEFAULT_MAP standard colormap.
Test various Compare operators with dyadic photomap operands.
Test various compare operators with photomap, constant operands.
Test various Compare operators with dyadic photomap operands,
using ROIs.
Test various compare operators with photomap, constant operands,
using ROIs.
Test various Compare operators with dyadic photomap operands,
Control Planes.
Test various compare operators with photomap, constant operands,
Control Planes.
MatchHistogram element tests, using various images and histogram
matching techniques.
A selection of MatchHistogram element tests, with ROIs.
A selection of MatchHistogram element tests, with Control
Planes.
ImportPhotomap, Unconstrain, Constrain(ClipScale), ExportDrawable
test.
PasteUp element tests.
Geometry element tests, including rotations, scales, and mirroring.
NearestNeighbor technique.
Geometry element tests, including rotations, scales, and mirroring.
AntiAlias technique.
Geometry element tests, including rotations, scales, and mirroring.
BilinearInterpolation technique.
Tests to exercise the various FAX decoders and the Geometry
element.
Dither test, ErrorDiffusion dither technique, ExportDrawable.
Dither test, ErrorDiffusion dither technique, ExportDrawablePlane.
Dither test, Ordered(4) dither technique, ExportDrawable.
Dither test, Ordered(4) dither technique, ExportDrawablePlane.
Dither test, Ordered(8) dither technique, ExportDrawable.
Dither test, Ordered(8) dither technique, ExportDrawablePlane.
Dither test, Default dither technique, ExportDrawable.
Dither test, Default dither technique, ExportDrawablePlane.
Logical element, photomap and a constant of 0 as operands,
various operators.
Logical element tests, dyadic photomaps as operands, various
operators.
Logical element, photomap and constant of 0 operands, various
operators, ROIs.
Logical element, dyadic photomaps as operands, various operators,
ROIs.
Logical element, photomap and constant of 0 operands, various
operators, Control Planes.
Logical element, dyadic photomaps as operands, various operators,
Control Planes.
Blend element test. Monadic source, 0.1 source constant. Alpha
constant of 0.5.
Blend element test. Dyadic sources. Alpha constant of 0.5.
Blend test. Monadic source, 0.1 source constant. Alpha constant
of 0.5. ROIs.
Blend element test. Dyadic sources. Alpha constant of 0.5.
Uses ROIs.
Blend test. Monadic source, 0.1 source constant. Alpha constant
of 0.5. Control Plane.
Blend element test. Dyadic sources. Alpha constant of 0.5.
Control Plane.
Blend test. Monadic source, 220 source constant. Alpha plane
is a photomap.
Blend test. Dyadic sources. Alpha plane is a constant 220.
Blend test. Monadic source, 220 source constant. Alpha plane
photomap. ROIs.
Blend test. Dyadic sources. Alpha plane is a constant 220.
ROIs.
Illustrate use of point and Standard colormaps for rendering
triple band images.
These tests are design to perform limited exercising of XIE's
capability of dealing with various encodings of flo source data. The test
init function obtains a photomap using ICP ->
EP. A series of independent
permanent flo pairs, one of the form IP ->
EP, and the other of the basic
form IP ->
ED, are constructed. The encoding parameters for the ExportPhotomap
(EP) element in the first flo are derived from test configuration. The number
of flo pairs created is also dependent upon test configuration. The tests
can be configured so that the test init function will constrain the input
photomap to a specified number of levels, on a per-band basis, so that word-sized
and quad-sized pixels are passed through the flos. Some tests below take advantage
of this. See tests.c for test configuration, and hints on how to add similar
tests.
Simple Point element tests. Drawable destination.
-pointroi1
Simple Point element test which uses ROIs. Drawable destination.
Simple Point element test which uses a Control Plane. Drawable
destination.
Simple Point element test. Photomap destination.
Simple Point element test which uses a ROIs. Photomap destination.
Simple Point element test which uses a Control Plane. Photomap
destination.
Two flographs are created which are the same in structure,
expect for the x and y offsets specified for the ExportDrawable flo elements.
The test init function creates a photoflo based upon one of the two flographs.
The inner loop of the test function uses
XieRedefinePhotoflo()
to alternate between each of the flographs. Make sure that your ineer loop
reps are 2 or greater in order to exercise this test fully (see
-reps).
Test
XieModifyPhotoflo()
by adjust ROI
offsets and size.
Test
XieModifyPhotoflo()
by changing the
LUT input to a Point element.
Test
XieModifyPhotoflo()
by changing ExportDrawable
x and y offsets.
This test creates a rather long flo of arithmetic elements,
each which does nothing more than add 1 to a small image. The test init function
scales the input photomap. The ExportDrawable x and y offset is modified randomly
during each iteration of the test function inner loop.
This test creates a rather long flo of arithmetic elements,
each which does nothing more than add 1 to a large image. Each rep, the Geometry
and ExportDrawable elements at the end of the flo are modified to crop a small
piece of the input into its appropriate place in the larger image.
These tests all basically take an UncompressedTriple image
as input, send it to ConvertFromRGB which converts the image to some configured
colorspace, and then send the converted image on to ConvertToRGB prior to
display. The original image is displayed in the lefthand window, and the image
which has passed through the flo is shown in the righthand window. The goal
of these test is to show that ConvertFromRGB ->
ConvertToRGB is lossless.
ConvertToIndex test, TripleBand BandByPixel.
ConvertToIndex test, TripleBand BandByPlane.
The test init function uses a flo containing ConvertToIndex
to display an image in the left window. The test function uses this drawable
as input to a flo which does ConvertFromIndex ->
ConvertToIndex and sends
the resulting image to the right window. The result should be lossless.
A somewhat large flo which uses control planes, LUTs, Point,
PasteUp, Logical, Constrain, Dither, Geometry, MatchHistogram, BandCombine,
and BandSelect elements. See the Postscript file ``complex.ps'' for a rendition
of the photoflo which is executed.
The xieperf program is based upon R5 x11perf(1X), and while not entirely comprehensive in its coverage of the XIE protocol (see BUGS, below), it is intended to be useful in the evaluation of XIE implementations in the areas of protocol adherence and performance. The xieperf program includes tests which execute each of the protocol requests and photoflo elements specified by revision 5.0 of the XIE protocol. In addition, xieperf provides a set of tests which can be used to validate the detection and transmission of XIE protocol request errors, such as FloMatch, FloValue, and so forth. Finally, xieperf provides a customizable demonstration program for XIE.
A test is made up of three components executed in sequence -- an initialization function, a test function, and an end function. The initialization function is responsible for allocating and populating test resources, such as photomaps and LUTs, and for creating a stored photoflo which will be executed by the test function. The test function, in most cases, simply executes the stored photoflo for a specified number of repetitions. The end function, which is called following the test function, is used primarily to destroy any non-cacheable server resources used by the test, and to free any memory which was dynamically allocated by the client. Some tests, such as -modify1, -await, -abort, and -redefine, perform additional steps within the test function inner loop, as required by the element being tested, or in an attempt to make the test more visually appealing.
Evaluating the performance of individual XIE elements is not as simple as measuring Core X drawing times. The XIE protocol requires elements to be embedded within photoflos in order to be exercised, and the minimum possible photoflo size is two. This implies that it is impossible to measure performance of a single element in isolation -- the time it takes to run the flo depends on what other elements exist in the flo. Extrapolating performance of a single element (or technique) in a flo must be done carefully, on a case-by-case basis, since in general measured element performance depends on input image size, data type, and other factors, all of which can be influenced by upstream flo elements. Note further that the number and type of elements in a flo can be influenced by the visuals available on the display, so even flo-flo comparisons on machines with different visuals must be done with caution.
Many test labels contain an abbreviated pipeline description. For instance. IP/IL/P/ED indicates ImportPhotomap, ImportLUT, Point, and ExportDrawable. Pipelines ending in ED (ExportDrawable) often include hidden elements such as BandExtract, ConvertToIndex, Dither, or Point to match the flo output to the screen visual. Pipelines ending in EP (ExportPhotomap) will result in a blank window.
xieperf
is compatible with
x11perfcomp(1X),
which is used to compare the outputs of different
xieperf
and
x11perf
runs in a nice, tabular format. In
xieperf
you will need to use the
-labels
option
(see
OPTIONS, above), and provide the resulting labels
file to
x11perfcomp(1X)
to obtain correct output. See the
x11perfcomp(1X)
man pages for more details on this.
There are no X defaults used by this program.
There should be a IMAGES environment variable to augment the -images option.
Many tests only scratch the surface of possible test cases. Some of the options available for certain flo elements are either inadequately tested, or ignored altogether. There are insufficient tests for bitonal, large pixel, or triple band tests.
Some of the test names are inconsistently cased, for example, -Abort and -dither1.
Some tests are hopelessly slow when run against machines with slow FPUs.
Bitonal images are for the most part displayed using the ExportDrawable
flo element, however, ExportDrawablePlane would be a better choice.
X(1X),
x11perf(1X),
x11perfcomp(1X)
Syd Logan, AGE Logic, Inc.