MARCO GSRC Calibrating Achievable Design Theme GTX

GTX User Manual

revision 2.1, December 18, 2001

	Andrew Caldwell, Andrew B. Kahng, Igor Markov, Mike Oliver and Dirk Stroobandt

Contents

The user manual explains step by step how you can perform studies in GTX. An example study shows the various possibilities GTX offers (just follow all the steps of this manual). This user manual assumes that you know about the major concepts of GTX. If you don't, please read the GTX framework document first. The user manual also assumes you do not wish to change the models (parameters or rules) but merely use them for performing studies. For developing your own models, please refer to the developer manual.

II.a. Loading Files

 From the "File" menu, select "Load Supplied" (for data that came pre-packaged) or "Load User" (for data created by you), then "Parameter File", "Rule File", "Rule Chain File", "Value File" or "GTX File".

• Parameter files have a ".prm" extension and contain the definitions of parameters (data) used within a module.
• Rule files have a ".rul" extension and contain the definitions of rules (models) that are known within the module.
• Rule chain files have a ".chn" extension and contain a list of rules needed for a specific study.
• Value files have a ".val" extension and contain parameter value allocations.
• GTX files have a ".gtx" extension and contain a list of parameter, rule, rule chain and value files that together comprise all information needed for a single study (or a set of studies). They allow you to load all those files in a single action.
• To unload the files, permitting the loading of a new study without interference, go to Edit->Clear Engine State.

Example step 1: Load the file BACPAC.gtx--from the File menu, choose "Load Supplied" followed by "GTX File", and double-click on the icon for BACPAC.gtx.  You should see something like this screenshot. (Go to next step)
(Special note to Windows users:  the BACPAC study will not run correctly unless you have Perl installed.  There is a version called ActivePerl which is downloadable without charge from www.activestate.com.  However you can choose a different study from the supplied studies and run through the example steps, choosing different parameter names as appropriate; most steps will still make sense.)

II.b. The Five Basic Views

 The GTX interface consists of two primary windows (screen shot):

• The list window (on the left): displays lists of items.
• The detail window (on the right): displays information on a particular item selected from the list window.

1. Parameters view
2. Rules view
3. Rule chain view
4. Values view
5. Graph view

The rules view contains a listing of all rules currently loaded into GTX. Selecting a rule from the list displays its detailed information (name, inputs, output, description, reference, body - the relation between inputs and output) in the detail window. Unless you are editing the rule, you may ignore the "Avail outputs" box and the "Available" box in the "Inputs" section.  Their function is explained in the Developer manual: Rules section.

The rule chain view supports viewing, editing and creating rule chains. Rule chains are inference sequences - subsets of the list of all rules that specify how to calculate values for output parameters from input parameters. The rule chain view allows interactive editing by adding and removing rules from the chain. A description of this process follows in the next section.

The values view displays computed parameter values for the current rule chain and specifications for the current study. In the detail window, users can select which parameters to 'watch' (in the sense of remembering their values on given inputs) while the study is executed, and specify plotting options. For more information see Section V Executing a study

The graph view displays in tree form the rule graph, which may be viewed from the point of view of rules or of parameters or both at once, and either "forward" (so that progressing down the tree goes from inputs to outputs) or (more usually) "backward", going from outputs to the rules that compute them or the values that are inputs to those rules.  See Section VIII Using the Graph View.

 Rule chains are sequences of rules used to compute values for parameters. A rule chain is completely described by the set of rules included in it. Parameters which are inputs to a rule in the rule chain and are not computed by a rule in the chain are primary inputs. Parameters computed by a rule in the chain are the chain's outputs .

 The rule chain view allows viewing and editing the rules in the current rule chain, as well as creating a new one. Either click on the "Chain" button at the bottom of the list window or select Rule Chain from the View menu.

 Editing a rule chain (or creating one from scratch)

 Rule chains are created from the "top down". First you decide what you want to compute, then you select a rule that computes it. If the current rule chain is empty, then in the rule-chain view, all available parameters will be listed in the column in the middle of the screen (under "Uncomputed input"). If the rule chain is not empty, then by selecting one of the rules, you will see in the middle column a list of all inputs to the rule that are not currently computed by the chain.

 If you select one of these parameters, you will see in the right-hand column (under "Rules computing selected param") a list of all available rules that compute that parameter. By selecting one of these and choosing "Add" (or by double-clicking the rule), the rule will be added to the chain in topological order.

 The rule chain can be extended in a more general way. To compute some parameter which is not an input to the current rule chain, choose the "Show All" button at the bottom of the middle column. This will show all parameters which are not computed by the current rule chain; you can then select one and add a rule computing it.

 To remove a rule from the chain, double-click on the rule in the list window (left-most list).

 If you select a rule in the list window, and then change to the rules view (via the "Rules" button), the highlighted rule will remain selected. The detail window now provides all available information on that rule.

Example step 2:

• Go to the "rule chain view" (click on the "Chain" button at the bottom of the list window) and select the rule "BACPAC::dp_lgate".
• The "rule view" (click on the "Rules" button at the bottom of the list window) shows all information on this rule.
• Return to the "rule chain view" and double-click the rule "BACPAC::dp_lgate" (this removes this rule from the chain). The parameter "dp_lgate" is now an uncomputed input to the chain and therefore shows up in the middle column.
• Select the parameter "dp_lgate" in the middle column. The rule that computes this parameter ("BACPAC::dp_lgate") now appears in the right column as the only rule that computes this parameter.
• Now load the files "Fisher_250.prm" and "SUSPENS.prm" (from the "File" menu, select "Load Supplied", then "Parameter File"; double-click on the file "Fisher_250.prm".  Repeat for "SUSPENS.prm".  This is necessary only to prevent warnings about implicitly-created parameters.
• Now load the files "Fisher.rul" and "SUSPENS.rul" (as above, except using File->Load Supplied->Rule File).
• Reselect the parameter "dp_lgate" in the middle column. Now four rules appear in the right column. They all compute the selected parameter.  (See screenshot.)  Double-click "FISHER::dp_lgate" in the right column to select this rule for the chain. The middle column no longer contains dp_lgate, which is now computed by a rule in the chain.

The above example shows how easy it is to interchange rules from different modules in GTX.  This may also be done in the graph view.

(Go to next step)

 To see the input and output parameters of the chain, click on the "Vals" button at the lower left, or choose "Values in Chain" under the "View" menu.

 In the values view, the list window displays the current value of each parameter associated with the current rule chain. The top window contains the primary inputs to the rule chain (parameters which are inputs to a rule in the chain, and not output of any rule in the chain). The bottom window contains the outputs (parameters that are output to a rule in the chain). Most often you will be editing values of parameters that are inputs to the current rule chain.  Some inputs may already have values, either because they have default values in their parameter definitions or because they have previously been edited; these will show up in the "Values" column in the box at the top left (the one headed by "Input Name  |  Value  |  Units").

 To change a value or values, select the parameter whose values you wish to edit.  The value(s), if any, will appear in the top portion of the right half of the screen.  If there are multiple values, they will be separated by commas.

 Enter the new values in the top right (to the left of the "Commit" button).

• If you wish to enter multiple values (useful for "sweeping"), separate them by commas.
• For numeric parameters only (i.e. doubles and ints), you may also enter ranges of values in the format  lower ; upper ; step, where lower is the (inclusive) lower bound, upper is the (exclusive) upper bound, and step is the step size (must be positive).  You can use multiple ranges, or mix and match ranges with single values, by separating them with commas.

 Although inputs to the current chain are the most useful values to edit, you may edit the values of any other parameter by going to the "Parameters" view (click the "Pars" button at the lower left or choose "Parameters" from the "View" menu).  You may choose any parameter and follow the procedure above to edit its value(s).

(Relatively) new feature:  You can also view/edit the value of a parameter from any view, by right-clicking on its name and choosing "Edit Value".  This applies in most locations where the parameter name appears, but not in the body of a rule nor in the "Avail Outputs"/"Available Inputs" of a rule; also, a right-click on a rule name that happens to be the same as the parameter name will bring up options appropriate to the rule, not the parameter.

 You can always restore the default values loaded with the parameter file by selecting "Restore Defaults Permissively" or "Restore Defaults Destructively" from the "Edit" menu. The latter option restores all default values, the former one only the default values of parameters in the current study

Example step 3:

• Go to the "values view" (click on the "Vals" button at the bottom of the list window) and select the input parameter "k_Rent_chip".
• Double-click the value box (to the left of the "Commit" button) and enter (type) the new range of values "0.7, 0.3; 0.8; 0.06, 0.5, 0.8". Clicking the "Commit" button (or just hitting "Enter") will enter the values and automatically sort them.  (See screenshot.)

 A study in GTX is the application of values to the primary inputs of the current rule chain to compute values for its outputs. All information about the current study is available from the "Values" view (click the "Vals" button at the lower left, or choose "Values in Chain" from the "View" menu).

 In order to evaluate the rule chain, all primary inputs must be assigned values. The values of output parameters will be updated by evaluating the rule chain.
 
 

V.a. Sweeping

Sweeping is evaluating a rule chain for multiple values of the same input parameter(s). Sweeping can be done by simply giving multiple values to some input parameters and evaluating the rule chain (see further). All outputs that directly or indirectly depend on input parameters that have multiple values will also have multiple values (one for every input parameter combination) after the evaluation of the chain.

V.b.Evaluating a chain

Make sure that all inputs to the current chain have values (look in the box at the top of the left portion of the screen, the one headed by "Input Name  |  Value  |  Units ", and follow the procedure above to edit the values).  If you wish to sweep over multiple values of one or more of these inputs, simply enter multiple values for them as explained above.

To see the output values given these inputs, choose the "Apply Rule Chain" button in the lower portion of the right half of the screen.  All outputs of the chain will now be given their values, and they will appear in the lower box on the left-hand side of the screen (the one headed by "Output Name | Value | Units").  If there are many values for a given output parameter you may not be able to see them all; highlight the parameter and look in the text box to the left of the "Commit" button.

Example step 4:

• Click the button "Apply Rule Chain" to evaluate the rule chain (since we added multiple values for the parameter "k_Rent_chip", the GTX engine will automatically perform a sweep over all input value combinations).
• Select the output parameter "C_int_global" to see that it indeed contains multiple values due to the sweeping.

V.c.Tracing a particular output

 If you wish to view what happens to one or more particular outputs as the inputs vary over their given values, you must let the program know before running the calculation so that it can allocate memory to record that information.  Find the output(s) you wish to trace in the "Available:" box on the right side of the lower-right pane.  Double-click the output(s), or highlight it and choose "Add".

Now choose the "Apply Rule Chain" button.  When the calculation completes, select the output you wish to view (i.e. highlight it either in the "Outputs to trace" box or in the outputs list in the left pane) and choose the "View Sweep Trace" button.  This will bring up a box showing the trace of the output with respect to the varying inputs.

Example step 5:

• Select the parameter "f_cycle_stat" as an output to trace (double-click on the parameter in the "Available Output" box in the detail window).
• Evaluate the chain (click on "Apply Rule Chain").
• Select "f_cycle_stat" in the "Output to trace" box and click "View Sweep Trace". This will give you a window with the minimum and maximum values obtained for "f_cycle_stat" and the values of the input parameters (here "k_Rent_chip") that lead to them. Also the average value and standard deviation is presented. At the end, all input value combinations and the corresponding output value are listed.

V.d.Plotting a particular output

 Add the output you wish to plot to "Output to trace" as described above and choose the "Apply Rule Chain" button.  Now choose the "Plot Sweep Trace" button.  This produces a plot of the outputs with respect to the inputs having multiple values.

• To obtain a plot of a sweep trace, exactly one or two of the inputs must be allowed to vary over their multiple values.  (If no inputs have multiple values then there is nothing to plot; if more than two have multiple values then the plot must be restricted so that it is not high-dimensional to view.)
• Therefore, if you have two or more inputs with multiple values, then when you choose the "Plot Sweep Trace" button, you will see a dialog box entitled "Choose sweep to plot". The inputs having multiple values will appear in the list "PIs that vary".
• If there are exactly two PIs that vary and you wish to plot against both of them, simply choose the "OK" button. If you want a plot with only a single independent variable, double-click on the other varying PI. Its name will appear in the list in the middle, and its values will appear in the list on the right. Select one of these values by highlighting it.
• Similarly, if you have more than two PIs that vary, you will have to fix values for all but (exactly) one or two of them. You fix a value for a PI in the manner described in the previous paragraph. When you have chosen values for all the PIs whose values you wish to fix, choose the OK button, and the plot will be generated.

 When the plot window is hidden by another window and then restored to view, it may appear blank. In this case, choose one of the options "Plot1", "Plot2", "Plot3", "Plot4" or "Plot5" under the "File" menu.

Example step 6:

• Click the button "Plot Sweep Trace" to plot "f_cycle_stat" versus "k_Rent_chip". (choose "Plot 1 in the "File" menu to refresh the window).  (View resulting plot).
• Change the value of the input parameter "num_stage_lgate" to "12;18;2" (range from 12 to 18 in steps of 2) (select "num_stage_lgate" in the input part of the list window, change the value in the "value box" and click on "Commit").
• Again select the parameter "f_cycle_stat" as an output to trace (double-click on the parameter in the "Available Output" box in the detail window) and re-evaluate the chain (click on "Apply Rule Chain"). Wait until the calculation is done and the new values appear in the "outputs box" of the list window.
• Select "f_cycle_stat" in the "Output to trace" box and click "Plot Sweep Trace". This will give you the window where you can choose the parameters to plot against. Simply hit "OK" for a 3D plot against both parameters, select one for a 2D plot.
• Choosing "File", "Plot 2" or other plots allows you to view the plot from different angles.

 A constraint is a special kind of rule.  Its output parameter must always be the built-in boolean parameter CONSTRAINT, and its body must be a bool expression.  It is used in a special way: whenever its output evaluates to false, the computation with the current set of input values is terminated and its outputs are not added to the values of the output parameters.
 
 

VI.a. Adding constraints

 To add a constraint rule to the available rules, create a new rule, making sure that the output is CONSTRAINT (for convenience, CONSTRAINT is the first output parameter listed):

• To create a new rule, choose "New" under the "File" menu, then "Rule". A dialog box will appear with the same fields as in the rule-view described above.
• You must give the rule a name in the "Rule Name" box (it will not parse without it); this name should contain no whitespace or special characters (underscore is fine).
• You must select the output parameter to be "CONSTRAINT". The available loaded parameters are listed in "Avail outputs"; double-click "CONSTRAINT" and it will appear in the "Output" box.
• You must specify the body of the rule.  The syntax for this is described in the grammar document. For constraints, simply put the condition which the input parameter(s) should obey.
• Any parameters referenced in the body of the rule must be selected as inputs in the "Inputs" section on the right side of the dialog box.  All available parameters (except for the selected output and all currently-selected inputs) appear in the "Avail" box at the top.  To add one to the "Selected" box, double-click it, or highlight it and choose "Add".  To remove one from the "Selected" box, double-click it, or highlight it and choose "Remove".
• In addition to the above mandatory fields, you may also give a description of the rule in the "Description" box and/or a literature reference in the "Reference" box.

• Choose "New" under the "File" menu, then "Rule" and give your rule a name.
• Choose "CONSTRAINT" as its output (double-click on it in the "Avail outputs" box) and "k_Rent_global" as the only input (double-click on it in the "Available inputs" box).
• Type "k_Rent_global < 0.6" in the body of the rule and hit "OK".

(Go to next step)
 
 
 

CONSTRAINT is a special exception to the rule that only one rule in a given rule chain may compute a given output; you may have as many constraint rules as you like in the chain (for this reason CONSTRAINT will always be available when you choose the "Show All" button; an ordinary parameter would not show up if the rule chain already computed it).

VI.b. Minimizing or maximizing a particular output subject to constraints

 The output trace described above in "Tracing a particular output" will take into account only combinations of input values for which all constraints evaluate to true.  At the top of the box created when you choose the "View Sweep Trace" button are the minimum and maximum values of the selected output (subject to all constraints) and the combination of input values that resulted in the minimum and maximum values.

Example step 8:

• Add your rule with the constraint to the chain (choose "Chain" for the chain view, click the "Show All" button, select "CONSTRAINT" in the middle window and double-click your rule in the right window).
• Go back to the values view (click the "Vals" button), select the parameter "f_cycle_stat" as an output to trace (double-click on the parameter in the "Available Output" box in the detail window) and re-evaluate the chain (click on "Apply Rule Chain"). Wait until the calculation is done and the new values appear in the "outputs box" of the list window.
• Select "f_cycle_stat" in the "Output to trace" box and click "View Sweep Trace". You will see that not all the values for the input parameter "k_Rent_chip" are listed (those larger than 0.66 are not). The missing values resulted in the constraint not being met and were discarded.

 There are several ways of keeping the results of your study for further use:
 
 

VII.a. Saving parameters, rules, rule chain and values

 From the File menu, select "Save All" or "Save Delta" then Parameters, Rules, Rule Chain or Values.

There are three options for saving:

1. Save All saves all parameters in GTX into a single file (same for rules, rule chain and values)
2. Save Delta only applies to parameters and rules and saves only those parameters or rules that are new or have changed since they were loaded
3. Save as .gtx file performs a "save all" on parameters, rules, rule chain and values and puts a link to all those files in a ".gtx" file.

 After sweeping you can save the sweep trace to file instead of showing it in a dialog box by clicking on the "Trace To File" button in the values view. You will be prompted for a file name (type the name and hit "OK") and your file will contain exactly the same information as in the "View sweep trace" dialog box. You can also click on the button "Raw Output to File" (you will again be prompted for a file name). This file will only contain a list of all output values (in the same order as they are in the sweep trace) without any other information.

 You can also save your plots to file and print them (from the GTX Sweep Trace Plot, choose "File", "Print"; you might have to set the printer settings first).

Example step 9:

• Output your trace to file (click the "Trace To File" button, choose a file name and hit "OK") and have a look at the file that is created.
• Save a raw output to file (click the "Raw Output to File" button, choose a file name and hit "OK") and have a look at the file that is created.

VIII. Using the graph view

The graph view is a powerful new feature for viewing relationships between rules, between parameters, or among rules and parameters.

You have several options for viewing the graph.  These are controlled by the radio buttons "Params" and "Rules" and the checkboxes "bipartite", "forward" and "calculability".
 

• "Params"/"Rules" radio button controls whether the graph is presented with parameters or rules considered as primary.  The default is "Params".
• "bipartite" checkbox determines whether both rules and parameters are displayed in the same graph.  If "bipartite" is not checked, then the "Params"/"Rules" radio button controls whether rules or parameters are displayed.  By default, "bipartite" is checked.
• "forward" checkbox determines whether, by going downward and to the right in the tree, you are going from inputs to outputs, or from outputs to inputs.  If "forward" is checked, you are going from inputs to outputs.  By default, "forward" is not checked, because it is common to know an output parameter and want to know which rules calculate it or which other parameters are used to calculate it.
• "calculability" checkbox determines whether calculability information is displayed in the graph.  A "calculable" parameter is one that either has a value or can be calculated by GTX with the information it currently has, using the current rule chain.  A "calculable" rule is one whose output parameter would be calculable, assuming the rule were in the rule chain.  When "calculability" is checked, calculable rules and parameters display with green dots on the left; uncalculable rules and parameters, with red circles.  By default, "calculability" is not checked.

Example step 10:

• Go to the "Graph" view by clicking the "Graph" button at lower left, or by View->Graph of Rules and Params.
• Scroll down until you find the parameter "dp_lgate".  Click on the plus sign to its left, expanding the node.  The four rules you see in blue in the expansion are the rules available to compute dp_lgate.  The rule FISHER::dp_lgate is in bold characters; this indicates that it is the rule in the current chain used to compute dp_lgate.
• Expand the nodes BACPAC::dp_lgate and FISHER::dp_lgate to see the parameters input to them.  If you expand also the node dA_chip_logic under the rule FISHER::dp_lgate, you will see something like the screenshot; note that dA_chip_logic is computed by BACPAC::dA_chip_logic, which (being displayed in bold characters), is in the current chain.
• You can edit the rule chain from here:  Let's say you decide you don's want dp_lgate to be computed by the FISHER rule after all.  Right-click on FISHER::dp_lgate and choose "Remove from chain".  Note that FISHER::dp_lgate is no longer in bold.
• Click the "calculabilty" checkbox.  Note that there is now a red circle to the left of "dp_lgate".  That means that dp_lgate is not calculable; GTX does not have enough information to determine its value.  That's because you removed the rule that calculated it.
• Right-click on SUSPENS::INTERMEDIATE_dp_lgate, and choose "Add to chain".  The red circle by dp_lgate remains, because the rule SUSPENS::INTERMEDIATE_dp_lgate is itself uncalculable (has a red circle).  Expand the node SUSPENS::INTERMEDIATE_dp_lgate to find out why.  Note that one of its inputs, dl_int_required_pu_dp_udp, is uncalculable.  Expand that node, and note that it is computed by the rule SUSPENS::dl_int_required_pu_dp_udp, which is calculable (has a green dot), but is not in the chain (is not bold).  Right-click on SUSPENS::dl_int_required_pu_dp_udp and choose "Add to chain".
• Now the dots turn green, going all the way back to the root node dp_lgate, indicating that the parameter is now calculable.  You can verify this by going back to the "Vals" view and clicking on "Apply Rule Chain".