RELEASE 0.2.37 ============== Base ---- ~ StateObject (used for simulator and hardware checkpointing) now has public, instead of package scope variables. + Added support to Simulator and HWSystem for hardware writeback. ~ addProperty and replaceProperty can now be chained. Xilinx ------ - Fixed a few bugs with ram primitives in relation to the multiclock simulator. - Fixed some internally placed cells to not print any placement warnings (since the placement they are warning about has already been verified.) Virtex ------ - Deprecated some functions that are potentially dangerous for users to use (but must be public) or just are things that should never be called anymore (since there are new methods availible.) Viewers ------- ! The default TechMapper for TBone/JAB is now Virtex. ~ Changed the port table so that it only sets the prefered size when it initializes. JDK 1.3 would resize each time the table updated, so now it only resizes initially, and then the user can change the collumns to any width. + Added javadoc comments to the MemoryViewer, BrowserCore classes. Modgen ------ ~ Many mogden modules' tests updated to work correctly with multiclock. - Made tests less verbose when there are no errors. RELEASE 0.2.36 ============== Base ---- ! Multiclock is now the default simulator for everything. ~ Modification so that the Observable test works with multiclock as default. - Fixed an incorrect error message in Cell. ~ Modified the join test to work with multiclock. - Eliminated the erroneous errors printing out from Wires when Jab read them. Logic ----- ~ Modification to LibrarySelfTester to make it work with multiclock as the default. Xilinx ------ + Gave Constant a behavioral model. ~ Modified Xilinx, XC4000, and Virtex flip-flop and latch primitives to pass the tests with the multiclock simulator as default. XC4000 ------ ~ Modified techmap/test/cpu to work with multiclock as the default. jmac ---- + Added @mc_test macro to jmac to allow to have a different test vector for multiclock mode. This is necessary for FF's and latches to pass the tests correctly. + Added @implicit macro to create an implicit interface. This macro automatically handles creating the implicit_interface and the connectImplicitInterface() method. However, implicit ports must still appear in the @interface block as type implicit. Viewers ------- ~ Made some changes in the updating of observers so that there will not be as much updating using the schematic viewers. This significantly speeds up the simmulation time with multiple schematic viewers open. Jab --- ~ Made small changes so that board model GUIs will appear in front of the main Jab window on startup. JHDLOutput ---------- - Fixed Bugzilla #456 (printing null in JHDLPrintWriter). ReadBack -------- + Made it so that ReadBackData.removeReadBack() can be called even if the object has not been fully initialized. + Added writeback support to the readback class suite. RELEASE 0.2.35 ============== Base ---- ! Finished changes to now mostly use Hash tables for the simulator schedules. Build time was reduced in one extreme case from 8 minutes to 15 seconds--that is 32x speed up. Which is a 660x speed up from the original 02 hr 45min build time. Most circuits won't see this drastic speedup, but there should be significant speed up in build time for other big designs. + Added some debug info for multiClock in HWSystem(): printClockSchedules() and printPropagateSchedules(). Modgen ------ - FPAdd test added for the IEEE case, and a fix applied to handle the zero output case correctly. Readback -------- - Handled the problem with NullPointerException caused by warning message in RBSym.toCombinedString(). RELEASE 0.2.34 ============== Base ---- ! Changes made to the simulator to modify how the mangle propagate schedule function mangles. In some cases, this is a 7x speed up. Basically, removed some redundant work that was never being used and made some of the work easier by not construction some intermediary objects. It will require a redesign to make a big difference from this point on. ! Made a couple of changes to GlobalSchedule to improve the time it takes to build schedules for huge gated-clock circuits. A small change resulted in a 20% gain and faster propagate schedule build time. This change mostly affects circuits with multiple gated clocks, although the change applies to all multiClock circuits. The biggest change made only affects circuits that all 100% gated clock. If a circuit is 100% gated, then all step schedules are exactly the same. So, they are only calculated them once, so that it does not take as long to create the schedules. This optimization could apply to any circuits with a gated-to-gated schedule, but it is easier to add to 100% gated circuits. This is a speed up of ~18x . There are probably more things that could be optimized to be faster -- but the "inner" loop is where all of the time is being spent. This is in a inList() call and to make things faster, it may be necessary to not call inList() so many times or make it faster, or maybe thread things ... ~ Changed the behavior of Wire.getSourceCell to resemble that of getSourceCells().getSingleCell() (Bugzilla #445). ~ BV will now remember if it is signed or unsigned, and perform all of it's operations correctly accordingly. + Regression tests for BV also added. Viewers ------- ~ If a port is added to a used defined schematic node now a generic schematic node will be initialized instead so the port will appear in the schematic. + Added some sizing to make the waves viewer compatible with the JDK1.3 TBone ----- + Added error checking for loading a Tbone class. Also updated it's javadoc. RELEASE 0.2.33 ============== Base ---- - Fixed a couple bugs in Simulator, GlobalSchedule, DynamicClockDriver, and Cell that were preventing things from working correctly when both in multiclock and hardware mode at the same time. Virtex ------ + clkdll's now are correctly constructed and netlisted in both multiclock mode, and with gated-clock inputs. + Javadoc added to clkdll (now that there is jmac support for it). - Fixed a bug in clkdll that was causing it to mark itself as notNetlistable in some cases. jmac ---- + jmac now supports '@@' as a literal '@' character. This is useful for adding javadoc to jmac generated files. Viewers ------- + Four-column format for memories now works. RELEASE 0.2.32 ============== Base ---- - Fixed bug in ValuePropagater to use cell.getSystem() instead of getSystem() for calls associated with delete. Synthesis --------- ~ Changed WeightedOpCode to support a String weight instead of an int weight. This is to enable the "default" branch of if and switch statements. Various other small cleanups including cleanup of the print routines. ! Major changes to improve synthesis support. The code is stabilizing now and is only missing one or two major features. The following test cases now synthesize correctly: byucc.jhdl.synth.test.FullAdder byucc.jhdl.synth.test.Delay4 byucc.jhdl.synth.test.SequenceDetector101 These tests have been incoporated into "make test". There is a problem when compiling some test cases with jikes, (ie. SequenceDetector101). This issue should become moot when things are rewritten to not require the -g flag during compilation. Some of the behavioral models of the tests above still do not work in simulation, (although the synthesized structural models do). Justin is working on this now and they should be working soon. Major features that still need to be done: o Loop unrolling (DelayGeneric.java) o Support of clock and propagate in the same class, (should be easy) o Support for function calls outside of current class in behavioral models, (currently only function calls within the same class will work). This will require a change to classparse to eliminate the static file input stream. The testing could also be improved. Currently it only tests that the class passes through synthesis without throwing an exception. It should run test vectors through the behavioral and structural models and verify that they match. (This is done manually -- it just needs to be automated). Modgen ------ ~ Changed "CLK" to "c" in many modgen schematics for consistancy. Viewers ------- - If a port is missing the schematic viewer will only inform the user once instead of everytime the view is refreshed. - Corrected a small case with a null pointer exception. Parsers ------- - EDIF Parser Bug fix (Bugzilla #443). Not all parameters will be treated as ports to be merged now. RELEASE 0.2.31.1 ================ Viewers ------- + Support added to use the waves viewer with a text file outside the jab envirnment. Readback -------- ~ Modified the MRP parsers under Virtex and XC4000 to treat the trimmed logic sections as optional. + Added support for the case when two srl16(e) primitives are packed into a single slice (they share the "M" type in the LL file, like the LUTs in a ram16x1d or a ram32x1s). RELEASE 0.2.31 ============== Base ---- + Added the necessary modifications for Carl-wires to work in multiclock. Unfortunately, they do not work correctly if the parent behavioral model is gated. I can make this work, but it will take a little more effort. - Recovered from the deleting of the get() that was needed to implement gated clock. Re-implemented it in the fashion necessary because of incorrect changes to put and get. The value_prop get is not the ultimate get any more and it does very little usefuel. So these changes re-add gated clock functionality. Also committed four gated clock test benchs so that gated clock cannot be munged without knowing about it. All of these test benches and multiClock pass all tests and complete with no errors. + Played with some commented code and removed something that seemed unneeded. It was placing pullups on the settle schedule if they were pullups that were found between the primary inputs, and the circuit. This is probably not allowable behavior, and it may work normal anyway. It seems like it was superfulous work. ~ Changed the notices printed regarding clock wires, so they make sense with gated clock. Xilinx ------ + Modified Makefile so that you don't get errors when trying to compile for XC4000 or Virtex, but don't have the part of the tree corresponding to the other technology checked out. (i.e. if you're compiling against XC4000, it won't give an error if the Virtex directory doesn't exist.) XC4000 ------ + Added another place call to fdr because the logic (fmap) was not getting plaed. This is a temperary solution because XC4000 should place this logic automatically because it is linked to the fdce. However, any future changes should be transparent to the user. Netlist ------- + Added Javadoc comments to the (CSRC) VHDLNetlister. Modgen ------ + Added new FPAdd. It will now work with Virtex, XC4000, etc. - Empty fp.fpaddPack package no longer put in the jar. Readback -------- + Added the ability for multi-bit state elements to provide a "don't care" readback value by having a readback offset of -1. This applies to RAM entries only, for now. If a -1 is encountered for these state elements, the getBit() method returns a 0. RELEASE 0.2.30 ============== Base ---- ~ Updated base to use JHDLOutput. ~ Made getLeafName() no longer append the "-0" to instance names. ~ Improved the addProperty, getProperties API of wires. - Fixed some other small bugs. - Fixed bug in EDIF when user has JHDL in case sensitive mode but has a case insensitive clash in names. ~ Changed FloatException to WireValueException, so that multiple puts also show up when trying to read a wire (rather than when the put happens). This solves the problem where an unpropagated, unread wire shows up as multiple put. - Fixed a NullPointerException when reading uninit-ed arrays in ExternalUpdateManager. - Fixed a bug with unbound parameters after uniquifying a cell. Virtex ------ + Updated VirtexTechMapper documentation. - Fixed an input in norX and nandX that was apparently being handled wrong. Viewers ------- + Added a menu bar to the waves viewer when running in win false mode. - Fixed a rare null pointer exception in the waves viewer. ~ Updated WavesViewer to grab correct wire properties. - Fixed the problem of waves viewer closing BrowserCore if you switched modes on it. Made sure waves viewer always clears the port table checkmarks when it is closed. - Fixed a small bug in the waves viewer. Also fixed a major bug when running with win false. The tabs were not updating schematic viewer or memory viewer, so there values would never change. (Made a tag HONEYMOON version before these changes that should be used if I broke anything since I will be on my honeymoon and not checking email. -Brian) Parsers ------- + Added "PAD" as a valid token in the LL parser to support the M1.5 Alliance tools. JHDLOutput ---------- + JHDLOutput now has parseLine(String) method (and supporting methods) to provide functionality for configuration commands (i.e., from a config file). + Added support for global defaults in the config file. + Updated JHDLOutput documentation. + Modified file handling methods to facilitate multiple writers to the same file. Also extended ability to configure via command lines. ~ JHDLPrintWriter now appends to pre-existing files by default. RELEASE 0.2.29 ============== General ------- - A few inaccessable test directories that were erroneously included in JHDL.jar will no longer be included. Base ---- ~ Marked pads as non-simulateable, so that simulation after a netlist is possible. (But may still not work in some board models.) This change made down the tree where appropriate. Virtex ------ ~ Modified startup_virtex and ibufg to use the correct clk() ports. ~ Updated several clockable cells to use clk() type ports instead of just in(). Modgen ------ - Pruned some old unused directories in the msd package. Viewers ------- ~ Removing the step button in ViewManager would bring the Jab Browser in front of a board model GUI. Now removing of the step button is only done when neccessary. + Updated the coloring of some nodes for the schematic viewer. + Added a menu option to the schematic viewer to toggle between viewing all cells and viewing only cells that will not be optimized away. + Added keystroke support to dialogue boxes so that they can be manupulated without a mouse. Enter performs the action and ESC cancels. Tab scrolls through components. RELEASE 0.2.28 ============== General ------- + Various updates to JHDLOutput and JHDLPrintWriter. ~ Many references throughout the tree to 'NetList' changed to the more common 'Netlist'. - Cleaned up some of the pedantic warnings issued when compiling with jikes. Base ---- - Fixed bug where clkdll tried to enable behavioral model. + Fixes for making rising/falling gated clock work. ~ Moved the error messages for multiple puts and floats from propagate to Wire.put() and Wire.get(); also added FloatException as a checked exception for the browser version of get (so JAB can display X on a floating wire). + Minor documentation changes in BV. + Added setInstanceNumber to Nameable (for use by scan chains). - When deleting a clock driver it's now taken out of the external_clock_drivers list. ~ Modified the implicit clock algorithm from removing automatic clocks from purely combinational cells to adding clocks to cells with clocked children. In the process, this fixes Bugzilla #377, and #393. + Added initReverse to LinkedList, atBegin to allow for reverse list traversal. Xilinx ------ - 1-bit boolX gates are now handled correctly in both XC4000 and Virtex. + All read-only (non-ExternallyUpdateable) cells in the Xilinx tree (lut and rom primitives) have been updated to extend BasicMemory or XilinxBasicMemoryCL, which are the equivelent of Memory or XilinxMemoryCL but don't implement ExternallyUpdateable. (This fixes a problem with them getting clobbered in the simulator by readback.) ~ Modified Memory and XilinxMemoryCL hierarchy to include BasicMemory and XilinxBasicMemoryCL (to avoid duplication of code). - Removed unused file XCSynchronous. XC4000 ------ - norX, nandX, and xnorX's mapping and placement has been fixed. Virtex ------ ~ xorX and xnorX still don't map and place 100% right (because of some non-trivial Virtex mapping limitations) but mapping and placement has been updated to be better than it was previously. - Fixed a bad clock port in XilinxFD_1. + Added fdcp and fdcpe primitives. jmac ---- + XilinxBasicMemoryCL added to the list of "things that can behavioral models". ~ Modified jmac to accept clk() port designations. Viewers ------- ~ Cells who's outputs are not used are no longer drawn because they will be optimized away. A warning message is given to the user. (This change makes schematics much less messy, and can avoid some of the strange placement of cells up in the corner on top of each other.) ~ Now wire values are blue and if a cell is selected the wire values changes to selected color. When a cell is selected all of the cell changes to red. ~ When a circuit view is brought up it will use the same radix as already opened circuitviews. + Added FloatException handling (floating wires will display a value of X). - If you zoom in too far it keeps you zoomed in to where you were. TBone ----- + Added support for the new clk() type ports. + Added some extra error handling for the building of circuits. JAB --- ~ Modified JAB and related components to catch the new checked FloatException in base, and display X as the value of a wire in those circumstances. - Fixed a bug where two HWSystems were sticking around instead of one. - Fixed a lot of problems with the loading of classes in Jab, and the step button appearing and disappearing. Modgen ------ - Fixed a bug in removing ports in CordicPack's Stage.java + Added constructors to allow for user-provided instance names for accum, booth, delay, downcnt, gt, lt, and upcnt. + Added BV to behavioral models of various modgen modules. Templates --------- - Fixed two bugs in TBWriter with choosing the techmapper and determining port widths. Readback -------- + The initial import of the "generalized" readback code. This readback code has removed all of the FPGA specific information from the readback runtime software. All of this information has been restricted to the symbol table creation code for each architecture. Also, a single ReadBackSymbolWriter is shared by Virtex and XC4000 now and all Xilinx FPGA architectures' symbol table creation code resides in a subdirectory of platforms/util/readback/Xilinx now. - The ram32x1s permutation problem reported in Bugzilla #409 has been resolved. + The initial import of the readBackConstants.h file into CVS. This file only defines RB_SUCCESS and RB_FAILURE for use in JNI programming for readback. In the future it should be used for other readback constants which are important for JNI programming. + Added a more friendly message when .rbentry parsing fails. It suggests a regeneration of the .rbentry file, assuming that the file is corrupt or out of date. - Old unused readback code removed from the tree. RELEASE 0.2.27 ============== General ------- ~ Updated JHDLOutput to no longer prefix output, and to better utilize java.io.Writer in addition to java.io.OutputStream. - Fixed a null pointer exception in JHDLOutput. Base ---- - Fixed a bug in HWSystem that prevented registering clockDrivers that are not in the top level of hierarchy. + Added Cell.getFlatNetlist(), and supporting classes like FlatNetlist in response to Bugzilla #368. ~ Changed instances of "NetList" to the more common "Netlist" in the source code. + Added changes so that MultiClock gated clock schedules can be built faster. + Updated BV: Added multUpper() which returns the most significant bits of a multiplication, fixed some bugs related to longs vs. ints, and thread-safed a static array used internally. Logic ----- + Added ability to request flat netlists from Logic (Bugzilla #403). ~ 'to' is now appended to the placement info of cell when "translate origin" is called. Virtex ------ + Made changes to support the new ramb4_wrapper directory and to allow BlockRamView have explicit or implicit clocks. (These cells are specifically for the use of the EDIF parser. Users should still use RAMB4Single and RAMB4Dual.) Modgen ------ - Made it so cout ports are removed in ADDSUB_FG_CI and ADDSUB_F_CI. - Made it so unused ports are removed in Cordic, Cordicl and CordicPack. JAB --- + Fixed Bugzilla #402 to print stack trace when enabling behavioral model fails. ~ Now when a new class is loaded Jab checks to make sure it exists before unloading old class. TBone ----- - Fixed ambiguous method call in TBone, and made netlist filename generation better. Parsers ------- + EDIFParser can now use all BlockRam types, widthd, etc. RELEASE 0.2.26 ============== Base ---- - Fixed exception handling during reset of the simulation. Logic ----- + A static version of buf_o() added. + concat_o() added (both instance and static versions) - Fixed the special case of 1-bit buswise gates. (They are now mapped to a buffer or inverter as required.) Xilinx ------ ~ Added some checking if getting sink leaf cell, and there are multiple ones, the Warning is only given if the multiple sinks are in different places. XC4000 ------ ~ Changed the check Cell placement so that if it hits a cell that is not placed it continues to check the placement below that level of hierarchy. Virtex ------ - Fixed some bugs in the placement of Virtex's mux41 and mux81 modules' outputs. Viewers ------- + Now doing a memory dump writes to a file with the same radix used in the memory table, instead of always decimal. + Color properties can now be added to wires. ("Jab:color"). + Now the custom BVFormats are remembered and added to the selectable list in MemoryViewer and WavesViewer. + Now the viewers check for a wire property "Jab:format" which is set to a string to describe the BVformat for that wire. TBone ----- - Removed a 'step' so that the first step or reset is controlled by the user. Modgen ------ - Fixed initial reset state problem with upcnt and downcnt. - Modified delay to eliminate annoying placement warnings. ~ accum now places first addsub at (0,0). ~ Modified accum to use BVs in it's behavioral model. + Made a totally revamped testbench for the new floating point adder. RELEASE 0.2.25 ============== Release/Build ------------- - Fixed some incorrect paths in configure.in (for several checks). General ------- - JHDLPrintWriter no longer closes System.out and System.err. ~ setJHDLPrintWriter() used to finalize streams; now it just flushes them. + JHDLPrintWriter now overrides all println() methods. Base ---- ~ Changed method call connect_implicit_ports() to be better named, connectImplicitPorts() (old call will be deprecated). The change was made down the entire tree. + Added isSimulateable() to all cells (whether to simulate), to complement isVisible() (whether to view in JAB) and isNetlistable() (whether to include in netlists). This should be a better way to implement Annotations, but that fix is still forth-coming. + Increased the amount of information stored by ValuePropagaters with regard to sink and source cells. getSinkCells() still returns only the behavioral cells as before, but the new getAllSinkCells() returns everything connected to a wire, including annotations. ~ Modified BooleanFlags to include SIMULATEABLE, and remove some unused flags. ! Modified the clkdll/bufg combination to simulate correctly with the addition of new features in base: isSimulateable(), and getAllSourceCells(). - Fixed a bug that occured in some widths of BV.toBVArray(). This now will handle any arbitrary width. + Added Wire.clearTransitionCount(), and made Cell.getSink/SourceWires public to match Wire.getSink/SourceCells. + Fixed Bugzilla #391 - The DRC before simulation now detects leaf cells with no behaviorial model. + Added a convience method for gated clock in GatedPropScheduleList. getElementCount() - Fixed bug with UpdateObservers. Logic ----- + Changed the field AlreadyRelativelyDerminedPlacementOrigin to placementOriginDetermined; added a few more versions of the pad*() calls. XC4000 ------ + Updated the regression tests to not call System.exit(), and to be more lenient on what directory the test is run from. Virtex ------ ! Modified the clkdll/bufg combination to simulate correctly with the addition of new features in base: isSimulateable(), and getAllSourceCells(). - Fixed placement of mux81 to place each bit of the output only once. Modgen ------ - Various minor changes in modgen code. Viewers ------- - Solved a small problem with implementing the large memory interface for some units that need the BV[] to be initialized. + Added the wire color and text properties, and added the ability to pull up a schematic viewer with a wire and that wire will be selected. Parsers ------- + EdifParser has a new anonymous inner class for case- insensitive input streaming instead of relying on an external input stream class in this package. Templates --------- ! TBWriter was missing an initalization for some arrays--these are now allocated. RELEASE 0.2.24 ============== Release/Build ------------- + Modified the code for setting the JAVA_CPP_FLAGS to handle the IBM's 1.3 JDK Basically, the structure for IBM's JDK is quite different from the others since Java is actually located in "jre/bin" and not just "bin", which configure used to assume. Also, there is not platform dependent includes required for this version of the JDK. Base ---- - Removed the unused OUTE designation of ports (as it was originally coded to assist tri-state back when there were TriStateWires, and INOUT ports work just fine). + Added PortRecord.is{In|Out}put() and Cell.is{In|Out}put(portname) to determine whether an inout port is really an input to a cell (ie. connects to a leaf cell input) - Bugzilla #376. + Added Nameable.caseSensitivity() to allow case-insensitive comparisons of common tasks like connect(), finding a named cell,... This may not currently control every case sensitive aspect of JHDL, but it is a start, and works with the case-insensitive EDIF parser. ~ Modified clock ports to be width 1 only (does a wide clock make sense?). - Fixed a bug when passing "" to Util.isLegalJHDLIdentifier. + Some various initial changes to support gated-clock. - Removed some miscellanious debug println's. - Improved more javadoc stuff, and various scoping inconsistancies. - Fixed Bugzilla #355 - gw() and range() no longer conflict with findNamed(). ~ Greatly improved the javadoc of almost all of base. ~ Modified Structural to give a better complaint when a behavioral model is not available because of user error (such as spelling "propogate"). + Made base more consistent across the board on boolean flag APIs: the version that takes a boolean to set the flag returns the prior value of the flag, in case the user wants to cache it and restore state later. + Added HWSystem.forceUniquify(), a method to be called before building the circuit to force every cell to netlist uniquely, in spite of cellInterfaceDeterminesUniqueNetlist- Structure. + Added ability to get connected port's width. + Added getB() method to get a 1-bit wire value as a boolean. + BV now supports toBVArray() to return a BV[] of the current BV chunked up into the specified size. + Made Node.getSystem() public instead of protected. + Added hasPort() to cell, a boolean check to see if a cell has a port by a given name. Logic ----- ~ Modified the implicit CLK port to be a clk() instead of in(). (This change was made down through the tree where necessary.) FSM --- - Made FsmData be much less verbose. Xilinx ------ + Added wire names to ExplicitClockTest, so that debugging would be easier. - Fixed a typo in xnorX. XC4000 ------ - Fixed Bugzilla #335 and #336 related to pad insertion in XC4000. Virtex ------ ~ Renamed I port in lut1* to I0. + Started adding wrappers for ramb4's for the edit parsers to use. (Although, there are in the process of being re-implemented). - RAMB4DualCheckPointStateObject now correctly implements Serializable so as not to cause a ClassCastException when calling getData(). + Added clockport version of constructors to srl16 and srl16e. Modgen ------ + Made some minor documentation updates on several modules. Viewers ------- + Added BVFormat to Schematic viewer. All windows now havethe same format, but soon there will be a sticky tag to decide that. + Now the user has the option of having the schematic viewer windows share a radix, or use a different radix. + Added the cell name to the title of the circuit view window. + Removed some recursion so that WireBrowser is faster. Parsers ------- ~ Regenerated the syntaxtree with jtb 1.2.1, and modified the makefile to perform regeneration easier. ~ Collapsed common code in the various constructors of EdifParser into one method. ~ Changed Cell and NewJHDLGenerator to better handle case-insensitivity issues while retaining the user's case specifications; and to correctly handle the cell rename construct. + Added the necessary constructors in the parse helper class to fix inheritance issues. - Now checks for null tmp when looking for bus port. - No longer refuses to blow out bus ports if contents of cell are empty. + Added reversed flag in preparation for getting bus ordering correct. + Added case for wires named like abc_123 - before this the bus abc was getting created. We want abc_. + Added synthetic port flag. This keeps from a0 from clobbering a if they are both ports on a cell. + Added ability to specify, on the command line, a list of cells you would like port merging on. + Added routines to clear the mergePortList and remove entries from the mergePortList. + Netlisting now does bus ports int he same order as other tools as in: (port (array (rename a "a[3:0]") 4) (direction INPUT)) + Added feature to allow user to use EdifParser.parse(...) return a SUBcell of the edif file being parsed instead of the default top-level cell. + Case insensitive option now available for EdifParser. To use, call the setCaseSensitive(boolean b) method of EdifParser or use the -i switch on the command line to set to case-insensitive. Netlisters ---------- + Added the cell rename construct into JHDL generated EDIF - Numerous small fixes to the CSRC VHDL netlister: + Added support for extended identifiers for entity name. This fixes problem with __ and '*' showing up. ~ Changed the instance naming convention to _inst_# + To support above, added function toVHDLCompName() - Fixed bug that was having clockdriver show up in the netlist. + Added two calls to setNotNetlistable(). Templates --------- + Added support for CSRC in NewTB (Bugzilla #367). RELEASE 0.2.23 ============== Release ------- ~ Changed the -O addition to JAVACFLAGS to be present only during a release build, as it overrides -g which is useful in debugging. Base ---- - Initializeables are not initialized early enough if they have only constant sources. All constants are resolved in simulator during the initialize() and are not touched later. So, an additional Initializeable call was added to initialize. Also, I initialize initializeables instead of reseting them. + Improved the javadoc for Cell and Wire. + Updated ClockDriver so that it can be constructed without a schedule, but determine its schedule by looking upstream during verifyAndCleanup (not fully implemented yet), and changed related code so that everything still works. - Miscellanious fixes to help some current problems in MultiClock in preparation to commit gated clock. FSM --- - Makefiles now rebuilds javacc files as needed. - FSM tests are not quite as verbose. Logic ----- ~ Logic internally split into two classes, to help manage it's incredible size. LogicGates contains most of the combinational logic and flip-flop constructors. Logic inherits LogicGates, and adds all of the additional calls needed in logic. There is no change in API or usage of Logic for any users, but JHDL developpers need to be aware of this change. + Modified Logic to have static versions of gnd(), vcc(). + Modified Logic to use improvements in ClockDriver, and to quit printing a warning message that a cell was placed twice when iterated place is used (since the deprecation message already appears). Xilinx ------ + Added tests for Constant, vcc, gnd to the LibrarySelfTester. ! Merged XilinxClockDriver and bufg into one class, named bufg, that extends ClockDriver, so that edif parsing, netlisting, and simulation should all use the same class. Modified bufg to match changes in ClockDriver. + Added XilinxClock as an interface for clock driving elements in Xilinx such as bufg, clkdll... + Modified some library cells to be of the correct superclass type, and modified XilinxLatch to not have an implicit clock port. + Modified the techMapper to use the padXX() methods of Logic, and to better handle the insertPads() method with real cells (not just annotations). - Fixed some erroneous explicit clock versions of methods. (Bugzilla #366) - Removed bufg from jmac, as it is now too complex and unusual to be generated. + Support for gated clock in XilinxFD. An implicit clk() port. This clk is also an in port, so it should not cause grief to existing code. Everything else will need to follow later. - Made insertPads no longer "final". + In the Xilinx Memory class, padBits and splitBits are no longer static. This makes it possible to print out the offending cell (hard to find bugs without it when bad EDIF is parsed). This *may* break other code which was using these routines as static methods... XC4000 ------ - Modified the mapper to not print extra warnings about duplicate placements on iterated place, since that already prints a deprecation warning. - Pruned some unused code from FloorPlanCanvas - Removed the duplicate printing of the same error message in RegressionTest. Virtex ------ - Fixed the initalizations strings to be the correct length in the lut primitives. jmac ---- - Removed a redundant import statement generated by jmac. Netlisters ---------- + Improved the javadocs, and used a FileOutputStream in place of BufferedWriter(FileWriter). Viewers ------- - Memory viewer has been modified to load quickly, and in hardware mode it only reads the viewable locations, and therefore reads from memory when the table is adjusted. This should make memory viewer fast. ! If in the schematic viewer, one bit of a bus is selected only that bit will show up red in schematic viewer. This will make bit tracing easier! + There are now proper symbols in the schematic viewer for 4:1 and 8:1 muxes. Modgen ------ - Deprecated Add, Sub and Adsu, as these now are replaced by the Logic add, sub, and addsub. + Updated testModgenNetlists to dynamically determine the directory it was called from, so the netlists go to the correct place regardless of where in the tree the test is run. - Fixed duplicate printing of exceptions caught in modgenTester. RELEASE 0.2.22 ============== Base ---- - Fixed setHardwareMode() bug: Used to set hardwareMode to true when HardwareInterface was implemented even when the argument was false. + LargemMemoryInterface added for those who think the MemoryViewer is slow. + Added the a call to HWSystem to check whether the simulator has been initialized or not called isSimulatorInitialized. Logic ----- - Logic shouldn't whine now about things being placed twice when translate is called. Xilinx ------ ~ Several more annotation only cells have been made visible in the schematic viewer. Virtex ------ + BlockRam's now inherit from a common class, RAMB4, which provides extra functionality, such as automatic INIT property building and simulation initalization. ! RAMB4Single and RAMB4Dual now support constructors that take an extra argument as the last argument before the cell name. This can be an int[128] array containing the contents of the BlockRam, or a String[] array that can be up to 16 Strings corresponding to the INIT_00, INIT_01, and so on properties. Any strings not specified will be set to 0's. Either way it is constructed, the INIT properties and the simulation model will be automaticall set. If you use this method, DO NOT try to use add or replaceProperty! Modgen ------ - Fixed a get() bug in upcnt and downcnt that prevented couters from being reset to values greater than 32 bits. - Fixed problem when in Virtex multArray TechMapping was disabled where And_fmap_g didn't create any calls (because the map call didn't create any cells). I fixed this by encapsulating the instantiation of And_fmap_g into the And_fmap_g.and_map method which returns null when TechMapping is disabled and a And_fmap_g when TechMapping is enabled. ~ Made it so that Add, Sub and Adsu remove their clk_en port if it is not connected. These classes are now deprecated. - Fixed some placement and width bugs in upcnt and downcnt. Viewers ------- + Now if the circuit is reset in the begining of simulation, the correct values are displayed in the waves view. + Added some extra error checking to the WatchInterface. + Some optimizations of the MemoryViewer added for greater speed. Also added functionality to support the LargeMemoryInterface. ~ Hexadecimal is now the standard when initializing the viewers. Jab --- + The netlist command will now netlist the currently selected cell. TBone ----- + Two flags added to TBone so you can choose a techmapper. (usage: -technology ) - Small NullPointerException fixed. - Now catches ArrayIndexOutOfBoundsException and tells the user there is something wrong with his ports matching wires. (Bugzilla #372) Parsers ------- + Added support to the EDIFparser remove pads from circuits with them and to netlist the resulting circuit. Readback -------- + Fixed the formatting in some warning messages in associateBlocksAndLL so that it was more clear what things were missing in the LL file. + Added test to check if the bitstream buffer was null in the performReadback method. If it is, then there are no symbols and readback is not performed. Otherwise, normal readback is performed. This solves a null array error from the JNI code which resulted from not being able to load the .rbentry file. RELEASE 0.2.21 ============== Release ------- ~ Makefiles now will still work even if autoconf is not found. + JHDLOutput and JHDLPrintWriter classes output management classes added. Base ---- + Added the static method getGenericWidth(Wire[]) to be able to do this check lots of places--greatly reduced code in the techmapper and Logic! + Added VisibleAnnotation to make some annotations visible. ! Fixed a carry bug in BV that was making add() and sub() return completely wrong results for BV's over 64 bits in length. ! Made Node.getSystem() public since lots of code needs to be able to find the HWSystem. Code that does it's own "top-level" finding to get the HWSystem or the TestBench should be removed and use this method instead. ~ Returned Simulator to its dominanace of a simulation. Thrust the reset() functionality of jab back into the place where it belonged and fixed a little bug in reset() in the Simulator. Logic ----- + Added warning message that prints if a cell is placed twice + There is now Logic support for 4:1 and 8:1 muxes (the call is just like the regular mux calls, but takes 4 or 8 inputs and 2 or 3 bit select lines). In Virtex, this returns highly optimized muxes that take up a SLICE (4:1) or CLB (8:1). In other technologies, these are not optimized, but just return a multi-level mux structure. Xilinx ------ ~ Some invisible annotations, such as ipad, opad, iopad, startups, and several other blocks are now visible annotations. XC4000 ------ - Removed two old unused XC400-specific regression netlisting tests. - All tests that were randomly pointing to home directories now point the right places. + Now, if there is a placement conflict it will now tell you the output wire as well as the cell that has caused the conflict. (Bugzilla #283). Virtex ------ - Fixed an obscure bug in BlockRams that were causing them to not work in multiclock mode. + RAMB4Dual will now throw an exception if an illegal configuration is given (i.e. widthA > widthB). - Fixed problem with updateState() so that it doesn't cause a null pointer exception when a null int[] is passed in, meaning that the BlockRAM had been optimized away from the circuit and doesn't have any readback data. Modgen ------ + Cleaned up code in arrayMult to make it more readable. Also added a public static variable that can disable the warning message given when the width of the y input is greater than the width of the x input for pipelined dividers. + arrayMult (and it's children) now use relative placement. + Modified things in IntDivide to use the add_cregister and add_pregister call to make code a little more readable. Viewers ------- ~ The memory and waves viewers now use the new BVFormat for all of their data, allowing much greater display customization. ~ Changed the way reset is handled on the display. Values will be correct for reset and it is labeled reset. - Fixed a small out of array bounds problem in MemoryViewer. Jab --- - Cycling or stepping with a number less than one still makes no sense, but it will no longer crash Jab. ~ Changed the way reset is handled in Jab. Previously none of the Jab windows were updated. now all windows are updated and the current reset values are displayed. + Now if the user requests a log of an already existing file he is given the option of overwriting, appending, or cancelling. ~ Modified update so that it checks to see if "this.wire" is null, meaning no wire is hooked up to the port. If no wire is hooked up to the port, a "Not connected" String is returned by update. Templates --------- + NewTB now recognizes inout ports. Parsers ------- + Added a new routing to Cell to automatically generate wrapper designs for EDIF files. + Added parseAndBuildWrapper() to automagically build EDIF wrapper JHDL modules to make it easier to include EDIF in your designs. + Added BuildWrapper to automatically build JHDL wrapper files for EDIF. Also, added ability to strip off pads and bufs. RELEASE 0.2.20 ============== Release ------- ~ Modified the Makefiles (hopefully for the last time) to have better dependencies on MakefileGlobal.in, and to delete only what is created by make or is in the user environment variable USERCRUD (so those who want can have *~ deleted, while those who don't are not forced). + The FSM tests now actually run. ~ Modified the global makefile so that a user only needs . and swing on their classpath, whether it is JAVACFLAGS, CLASSPATH, or JIKESPATH; and the make file appends the top directory on the end as needed. ~ configure now will recognize that you have JAVAC=jikes and will change the compile environment appropriately. The main way to configure with jikes is to use the --with-jikes flag. Base ---- ~ Modified ValuePropagator to allow join (test-bench-created-input, nc()) to work, and updated tb_JoinTest to catch that case. (Bugzilla #339) Xilinx ------ + Added method to allow control over whether top level clock port gets a pad inserted for it or not. Important when the user is trying to have some control over the clock pad w.r.t. DLL's and the like. XC4000 ------ - Fixed bugzilla #290. A new field has been added tothe text field to tell you the name of the last cell you selected. There was a need for this because multiple cells could be placed in one CLB and the names of the cells were displayed on top of each other so now there is a way to see the cells that are part of the CLB's. Virtex ------ - Fixed translation to slices for case when the placement hints were negative in method getRLOCFromPlacementInfo. - Added constructors to lut and ram primitives to allow multiple 1-bit ports to be concatenated into the previously supported single n-bit ports, thus giving better support to the edif parser. ~ BlockRam's now default to an init state of 0. To override this, a user will have to use replaceProperty() instead of addProperty(). - Bug in BlockRam's getMemoryElement() method updated. Modgen ------ + arrayMult can now handle pipedepths > 2 in Virtex. ~ Changed modgenTester so that arrayMult does 5 instead of 10 multipliers to speed of make test a bit. + Added EXAMINE_CI_Virtex for use in arrayMult's CarryCellVirtex when routing resources force a design to get the carry out a Slice above. ~ Modified CarryCellVirtex so it has a cleaner implementation of the signed carry-out bit logic and also so it uses an EXAMINE_CI_Virtex module when necessary. ~ In arrayMult, Changed all buf_o's to joins and changed things to call the new carry_cell constructor. + In arrayMult, added the capability so that if the clk_en is null then the clk_en net is not connected to the flip-flops. Also cleaned up some legacy code. + In arrayMult Started to add some placement disabling hints for case when doing another technology. Also added optimizations so that there are many many less wires left around which means less warnings in ngdbuild. + Added generic logic option so that when the techmapper is not XC4000 or Virtex an unplaced, but logically correct array multiplier is instantiated + Added arrayMultException to arrayMult to make it clear where the Exception is coming form. Also made sure all intermediate values in behavioral model were longs + Added IntDivide to modgen package. RELEASE 0.2.19 ============== Release ------- ~ Both edif and vhdl netlisters are not techmapper independent, and have been moved into the byucc.jhdl.netlisters package. - Fixed a missing package in the jar (Bugzilla #313). Base ---- ~ Modifications to GlobalSchedule and Simulator to support the TriStateExceptionHandler under Multiclock. This is a new feature due to the fact that signal output longevity is only the length of a step and it is difficult to determine if someone is not driving on a TriState bus. ~ Converted TriStateDriver to an interface, so that multi-bit tri-state buffers controlled by a single-bit enable could have a behavioral model. + ExternalUpdateManager now supports writeStateToFile and writeStateToHardware. Simulator was updated to call the correct writeStateToFile function (that's all that's there now). + WriteableHardwareInterface extends HardwareInterface and must be implemented by anyone who can do writeback. ~ Modified getClockValue interface in HWSystem. The current incarnation is shorter and more generic. ~ Modified Property and Parameter to utilize new NameTypePair for more efficient use of memory by sharing common names, while holding unique values. In the process, updated Property to be of any type, instead of just String, Int, and Boolean. + Some modifications in support of Gated Clock. Like clk port, in_clk_port, in various things like wire, cell, flags, portinfo, portblock, etc. Util ---- ~ Improved BVFormat so that you can now enter a more sophisticated format string. Check the comments in the files for the correct format for the format string. Logic ----- + Added a notification message that prints on inferred placement. - Made buf's and wires constructed in pad calls in Logic be consistant with their given naming. XC4000 ------ + Added some print messages to report the CLB usage on a placement conflict. Virtex ------ - Function getMemoryRange in RAMB4's has been corrected to return a memory rainge starting where specified and the number of elements specified. Previously it always began at memory location zero. + Behavioral models were added for all Virtex adders and subtractors. ~ Fixed srl16 and srl16e to be purely structural - Removed the verbose pad insertion messages that cropped up from somewhere. (Bugzilla #322) Modgen ------ - Fixed a placement bug in ramrom. Platforms --------- ~ Made GenericMemory implement MemoryInteface. It was commented out when the switch was made from BigInteger to BV as the return value of choice for MemoryInterface. Also changed GenericProcessingElement so that you can now create a VirtualPort and pass an already created Wire into it. The call will create the VirtualPort as well as add the wire into the wires hashtable using the wire's name as the key. This new form of the method should really only be used if you want to pass a wire directly from a connect call to the user. Viewers ------- ! Both node color and selected color are no longer hard coded. Each node has a variable for color and selected color. Also Schematic Canvas now has a variable add properties, and it is originally false, but if it is manually changes schematic canvas will look through the node list and check the attributes. If it finds Jab:color it changes the node color to the string attribute. Also if it finds Jab:message the string attribute is added on to this node in the test box information. Soon functionality will also be added to add wire attributes also! + Added the function updateInfo() to reduce the code size. + Added more precise error messages for unplaced cells or for conflicting placement when layout viewer is initialized. + More precise exception handleing when layout viewer is initialized with unplaced or conflicting placed cells. - Fixed Bugzilla #309, now it should initially update every cell in the memory table. Then it should only update the part of the table visible to the screen. + Added a function, getHardwareMode(), to determine if a circuit is currently in hardware mode. ~ Memory Viewer now operates in two different modes. If in hardware it loads all the data at once and then doesn't do any loading for scrolling, else it loads only the viewable data and if window changes it again loads only the viewable data, so it is always loading small chunks. Also when hex is chosen only the viewable data is converted and displaed, and all has to be reconverted when view changes. ~ Now only the viewable table information is updated to make memoryViewer faster. Update table takes two additional integer arguments, a starting row and the number of rows! + A 6th collumn has been added to the BrowserMainFrame table that displays the wire attached to the ports. (Bugzilla #287) - Fixed Bugzilla #308. Now ports are drawn correctly for a Virtex sub unit without carry in. Jab --- ~ Now an extra entry for the port table is returned in portWireEntry. The extra entry is the wire name. RELEASE 0.2.18 ============== Release/Build ------------- ! Added byucc/jhdl/MakefileGlobal, and modified the majority of the CVS tree to include this. This allows for better directory recursion with less effort per Makefile, fewer files to configure, and the creation of global variables. In the process, weeded out some old files, and restored Makefile instead of Makefile.in as the preferred CVS file in subdirectories. (Bugzilla #280, #282, and friends.) ~ Added some default ignores to .cvsignore files in most directories. ~ Modified the makefiles to use JAVACFLAGS and JAVAFLAGS instead of the ambiguous JFLAGS, to recompile cleanly after make veryclean, and to not override any variables (like CLASSPATH) set by the user. Base ---- ~ Changed ExternalUpdateManager to allow implementing both Checkpointable and ExternallyUpdateable. (This change is needed for simulation of BlockRams when checkpointing.) + Added ability to bind LONGs (Bugzilla #297). + Added ability to use read and write checkpointing state with streams and custom filenames. + Added javadoc to the Checkpointable interface. + Added TriStateExceptionHandler, for use by the simulator to detect conflicting tri-state drivers. - Removed the unused PullUpDownHelpers class in PropagateManager. ~ Major modifications so that no element is propagated more than once in multiclock. Incremental schedules are not as important since the simulator is being settled after just about every step(). Because of this, a different schedule was necessary to make sure that elements were not propatated extra times. + Added multiclock tests to the regression suite in base. + Added isDrivingOutput to all TriStateDrivers in the tree. + Modified the search for the constructor to try an anonymous version if the named version could not be found (Bugzilla #301). + Added a regression test to check valid JHDL naming. Logic ----- - Removed the "Targeting byucc.jhdl.Xilinx.Virtex" message, since it doesn't display for XC4000, and seems to be a debugging artifact. It might be a good idea to have this message appear when using TBone. + Added a warning in Logic about inferred placement to aid in debugging user placement. Xilinx ------ + Added the missing constructor for named creation by the edif parser (Bugzilla #301). Virtex ------ ~ BlockRams now implement both LargeExternallyUpdateable and Checkpointable, allowing them to be checkpointed correctly in simulation. Because of hardware issues, they may never arbitrarily checkpoint correctly in hardware without using a two cycle-checkpoint-restore technique. - Virtex map() now will throw an exception if a H-LUT is given as a hint string (H-LUT's do not exist in Virtex). Viewers ------- - Removed the integer argument 'Start' for the constructor of MemoryViewTableModel. It was designed for start to always be zero, and no longer needed this arguement. + MemoryViewer has been reworked so that it only updates the viewable part of the table. This should increase the speed in which it runs. TBone ----- ~ Checkpoint state files written out now reflect the actual loaded classname instead of 'Tbone-0_...'. Parsers ------- + Changed to retain primitive instance names. May break if there are primitives out there without constructors that take instance names. Xilinx.vcc and Xilinx.gnd were two such ones I had to fix. Modgen ------ ~ Modified upcnt and downcnt to allow the output to be any width in the structural and behavioral models. ~ Modified ramrom to include 'synch' in its parameter list. RELEASE 0.2.17 ============== Base ---- ! Checkpointing is now completely supported in base, and hooks have been added to allow calling of HWSystem's writeSystemState() and readSystemState(int cycle). All ExternallyUpdateable cells now are handled correctly, and Checkpointable cells have been updated to be handled more like ExternallyUpdateable cells. Full documentation coming soon. - Fixed a bug in BV parsing decimal strings and BigIntegers. + Added Initializeable interface to flag objects that should be initialized on simulator reset. + Base modified to handle LargeExternallyUpdateable objects, objects which have more than 32-bits of state and are to be sampled using readback or some other mechanism. - Changed a break to a continue to fix error in isLegalJHDLIdentifier. (Caution: this may break some of the automated tests - before this fix it was letting illegal identifiers through.) ~ Added instance name to pushHierarchy call so that instance names are retained. Xilinx ------ + Place calls added to fd* primitives that were missing them. + Added 'clr and 'pre' ports to several fd* primitives that were missing all their ports. - Fixed the naming of the 'clr' vs. 'pre' ports in fdp* and friends. ~ Modified the LibrarySelfTester to extend a generic version in Logic. ~ Split ibuf.jmac and obuf.jmac into separate files, since they are referenced by @structure commands which looks for the particular filename. XC4000 ------ + An error box is now displayed if the layout viewer is invoked on a completely unplaced cell. (Bugzilla #265). + Fixed Bugzilla #227. The problem was an infinite loop in the getVisibleWire name method that resulted from an invalid map call. Very obsure bug, but it is fixed. Virtex ------ + Added primitives for ramb4_s?_s? for EDIF import. + The BlockRAM models have been changed to support LargeExternallyUpdateable for readback and checkpointing purposes. Further, BlockRamView has added the addrWidthFromDataWidth(int) so that it no longer uses the same function in the deprecated BlockRam class. RAMB4Single and RAMB4Dual also use the addrWidthFromDataWidtn(int) in BlockRamView instead of their own internal copies to allow for some code sharing. - Fixed simulation problems with srl16 and srl16e. jmac ---- ~ jmac modified to support the following major change: ! jmac tree structure has been altered as follows: byucc/jhdl/util/jmac has the jmac program itself. byucc/jhdl//jmac has the .jmac files. (i.e. byucc/jhdl/Xilinx/XC4000/jmac) ~ Made changes to the constructors to allow for generic width wires to be 1 bit wide as well as the previously supported n bit widths. ~ Modified all .jmac files to be named after at least one of the cells they create (it makes the makefile nicer). Viewers ------- ~ Schematic viewer no longer implements double clicking on a wire to invoke watch , it is now a menu option / keyboard shortcut. Jab --- - If file was not succesfully loaded it returned error saying couldn't load file, and then said first argument which was not always the file. Now it displays the file name instead of the first argument. TBone ----- ! TBone now supports circuit checkpointing via the writeState and readState commands. + Tbone has been modified to handle LargeExternallyUpdateable Cells in the getHardwareState(...) call. Templates --------- ~ Changed NewTB to use the constants in the BooleanFlags interface instead of its own constants (that didn't match up with BooleanFlags). It should now recognize all the input and output ports correctly. Parsers ------- + Added default technology and target flags. + Added additional tests for the EDIF Parser. + Added smart GSR checking (only ignores in some cases). ~ Changed getValidIdentifier() to use Util.makeLegalJHDLIdentifier(). We can now read 3rd generation JHDL-EDIF. RELEASE 0.2.16 ============== Release ------- + More additions and optimizations to the release_build target to speed release compile time. ~ Daily releases are now under the "daily" pointer. Weekly releases will typically be under "latest". General ------- ! Added JHDLException as the superclass of all exceptions created in the JHDL tree, so that versioning information will be added to the exceptions. (Bugzilla #241). Pruned some unused exceptions. - Removed BUGS & GREAT_IDEAS files that are not being used. (We have Bugzilla now.) Base ---- + Added checking for undriven inputs; for now, it is only a warning. + Added UndrivenInputsAllowable, for those rare primitives (like cy4, fmap) that can legally have undriven inputs. ~ Modified ClockDriver to be less hassle to extend. - Fixed an undriven input in JoinParent from nc() to a real input gnd(). + Added programatic testbench threads to HWSystem and the Simulator. + Added checks in enableHWUpdate and disableHWUpdate which make sure externallyUpdateableCells and checkpointableCells are not null before accessing them. If they are null, then they must not have been initialized so updateCellList() is called to initialized these lists. Also, made sure checkpointableCells is initiallized to null, like the externallyUpdateableCells. - Fixed Bugzilla #251, with incorrect net naming in netlists. + Added necessary hooks to find out the next transition, last transition and current value of a particular clock wire. + Added properties on wires to edif (Bugzilla #240), and fixed a bug with integer properties. ~ Updated threaded simulator to imporove counter accuracy and multithread support. + Added replaceProperty() in Cell.java Bugzilla #255 solution. + Modified Property.java to have a toString() function. + Modified PropertyList.java to have a removeProperty() method and to die in putProperty() on duplicate entries Part of Bugzilla #255. - Fixed conditional statement that allowed steps in the cycle block of run() running with threads. + Restored name clash checking with ports (Bugzilla #230) ~ To avoid a class cast exception, ClockDriver will not allow anything to be driver by a clockdriver, but a clockable. This is temporary. The changes in simulator are to make sure that reset is only called once no matter how it is called. + Added hardware mode support for the threaded simulator. Modifications were made to the run(int) and step(int) method calls. + Added pushHierarchyNoImplicitPorts - specifically for EDIF imports. Logic ----- ~ TechMappers modified to have several adder, subtractor, and addsub constructors. This was added to support the newly optimized XC4000 adders. Xilinx ------ + Added UndrivenInputsAllowable, for those rare primitives (like cy4, fmap) that can legally have undriven inputs. - Edif now complains if top-level flags are set incorrectly (Bugzilla #235). - Removed dangling net optimization from edif, because UCF files were depending on nets that got optimized. ~ Reversed the rename string in edif of port arrays to match the internal ordering of bits in edif (Bugzilla #244). XC4000 ------ + Added the ability to construct adders, subtractors and addsub's that do not have extraneous carry in and carry out logic when carry in/out is not used. ~ The linking of fmap slave cells to fdc master cells has been modified. This had to reworked to handle some special cases where fmap placement should be linked to the placement of it's corresponding fdc in synchronous logic. + Added "quiet" mode to the regression tests. + Added ram_synch_shift helper class for correctly modelling lut-based shift registers. Virtex ------ ~ clkdll updated to be handled correctly by the schematic viewer and simulator. ~ Split srl16 and srl16e into two parts to work correctly in simulation. Modgen ------ - Some modgen cells modified to explicitly remove unused ports. + Added a "quiet" mode to most of the modgen tests. Viewers ------- ! Larger designs would run out of room and draw annoying large black boxes. The constant numWireColumns has been increased to allow larger designs to be drawn right! This has been changed so that if it runns out of collumns to route a net it no longer draws an annoying large black box, but rather it routes the net off the screen and pack on the screen. + Started supporting the View Wire Waves... menu option. It had somehow been disabled quite a while ago. - Removed the ability to "view" a leaf cell from the hierarchy viewer. (It doesn't make sense to "view" a leaf cell.) - Fixed Bugzilla #233. The names of long output ports no longer run off the screen in schematic viewer. Also large feedback networks used to run off schematic viewer. Now this wont happen. The changes required moving the variable offset from SchematicNode to SchematicCanvas. All supporting Nodes were updated. Also a function was added to OutNode that returns the length of the port name so the Schematic viewer would be able to check how wide it needs to be to fit those names. ! Added ability to double click on a wire in Schematic Viewer to add that wire to the wave viewer. - Fixed various bugs including the errors in removing the class view and added a wait cursor. + Added a wait cursor to several components. - Fixed bugs with the tabbed pane that wouldn't close. CLI --- - Fixed the NullPointerException when trying to tab complete on a command that does not implement a completion interface. Jab --- + Added threaded support to BrowserCore. TBone ----- - Fixed a few bugs that come up while loading classes. Parsers ------- ! NON-RELEASE_READY Stuff, but still worth mentioning: + Added ability to use the EDIF parser for Virtex. + Adding error checking on passed-in ports. - Eliminated tests which ignore all 'clr' and 'gsr' ports. They caused havoc in real circuits. + Made "Virtex" the target. You make changes by how you call EdifParser.parse(). + Added lots of debug print messages. + Add all properties, not just RLOC's for now. In addition, use replaceProperty() in case there were some defaults there already. This completely replaces the use of place(). RELEASE 0.2.15 ============== Release ------- ! Releases are now coming out in two flavors: daily internal releases, compiled, tested, and jar'ed every day, and full weekly releases that are complete with revised release notes, standard versioning, and are released on the bleeding-edge section on the web page. + Added a tutorial on how to do an internal release. Will add docs on how to do a full release real soon. =) - Removed some dead-wood files from base and Virtex - Removed some stray -ms flags for javac that didn't need to be there anymore (and were making it hard to compile on some machines.) Base ---- ~ In Simulator, step(int no_steps) operates in single clock hardware mode just like it does in the simulator, i.e., it is identical to cycle. In multi-clock mode and hardware mode, the simulator now suggests that "cycle" be used instead. + Support for resetMethodImplementedByUser added to Structural (Bugzilla #208) ~ Added nc() capability into base - nc() now creates a wire that returns true in isSourcelessOk(), so that the simulator can deal with them correctly. Some of the changes include bug 175, that connect() now requires a non-null wire, and that the circuit must connect all ports instead of leaving an unused port null. Also some minor changes in BV documentation, and the addition of getCell() to NodeList. - Fix various bugs that were in bugzilla concerning MultiClock: Floating primary inputs that did not float. Pullup/downs that are not being scheduled. Reset did not happen correctly in conjunction with Jab and Tbone. + Added methods like getOnCount, getMSBOn, reverse to BV, along with constructors that take BigIntegers; touched up documentation (bug 192). - Modified nc() so that base no longer reflects on Logic. - Modified clockDrivers so that user-defined clocks automatically invoke multi-clock mode (bugs 199, 202, 203). - Fixed a bug in Cell relating to testbenches not needing an implicit clock port. + Added ProgramaticTechBench interface for use with the threaded simulator. Logic ----- ~ Modified clockDriver in response to bugs 199, 202, 203. (Also updated the clockDrivers in all children classes down the tree.) + Added a collection of wire manipulation routines originally from Dr. Nelson's brentlogic. + Added pad insertion routines into Logic, and added support to all techmappers. - Fixed the bug where you could relatively place to a cell that hadn't ever been placed, twice. Now the first time is handled correctly, the next time will throw an error. Xilinx ------ ~ Updated all Xilinx techmappers to support nc() now residing in base. - Removed UnconnectedOutputWire, since base now supports nc(). ~ Revised edif in order to recognize the changes with nc(). Edif now optimizes away nets that have fewer than two connections (this one-liner may need to be reverted). + Added XilinxAnn that extends Annotation, and made all Xilinx Annotations now extend XilinxAnn. XC4000 ------ ~ Replaced invisible structure with implementing ConstantSource in cy4_mode. - The RegressionTest now correctly exits with a non-zero status on failure. Virtex ------ ~ Updated the behavioral model of fdc_1_g to match the way that fdc_1 actually works. + Added *_1 to LibrarySelfTester. Previously, none of these were being tested. - Fixed a minor clk bug in regc, regp implementation. - Removed a stray vcc() that was in fdc, fdp and friends. + capture_virtex primitive added. ~ Changed lut primitives to extend XilinxMemoryCL. + Fixed insertPads() for user-defined clocks, default global clock, DLL's, etc. ~ Added propagate() method to clkdll simply get him to show up on the sink list of wires so the insertPads() can do the right thing. Still won't simulate since neddsToBePropagated() (or whatever it is called) still returns false. This is the correct behavior. Modgen ------ ~ Modgen netlist test extensions changed from .gold to .edn to make it possible to diff the two directories instead of diffing each file. - Removed weird "Got here." messages from some of the module tests. Viewers ------- - Fixed wave viewer so that there is no exception thrown when zooming out at the end of the viewer history, also added feature that displays only the cycle number in multiclock mode if the space is too small to display. This also changes the cycle name bar to display vibrant green only on cells that have the complete cycle and step name. - Fixed bug that leaves waves viewer open when loading a new class. - Fixed the load new class method to work correctly when using TBone to load multiple classes. - Fixed -nw window mode to work with multiclock. Jab --- - Repaired Jab and TBone, so that they use the 'correct' behavior during reset and in resetting (which are not the same event.) + Added versionCmd to show the version number from CLI. + Added "quit" to File menu. TBone ----- - Repaired Jab and TBone, so that they use the 'correct' behavior during reset and in resetting (which are not the same event.) + Added a catch for NullPointerExceptions that were previously being passed through. Util ---- + Added the BVFormat class which will take format strings to tell how to format a BV for viewing. Templates --------- ~ Changed TBWriter (for newTB) to use putL instead of put, moving the upper limit of wire widths from the previous 32 to 64. This ought to be updated to use BV, but needs a large rework in order not to have a lot of garbage. RELEASE 0.2.14 ============== Release ------- + Added make release_build target to *quickly* build the entire tree from scratch. Will be updated more in the future to optimize for shorter clean compiles. Please do NOT use this target for trees that are not clean! ! The sythesis package has reached an alpha stage so it's been added to the general JHDL tree. The main access to the synthesis package is the Synthesizable class in Logic. The idea is that you can just extend Synthesizable instead of Logic and code up a behavioral model without any structure. Then, JHDL will automagically create all the structure appropriate for the behavior you described whenever it is needed, (ie. netlisting or disabling behavioral models). The code is Alpha still, (with a capital A). It only supports combinational behavioral models, (just propagate -- no clock) and it probably won't work with everything you throw at it. Also, I haven't tested if netlisting will automatically synthesize the structure. Right now the structure is only synthesized when the behvioral model is disabled. - Bug fixed in cvs-log.pl script that was missing some tags because of a fork resource exhaustion. - byucc/jhdl/apps/CBrowser removed from the release tree, as it's unused. Base ---- + ConstantSource can now be created with an arbitrary width BV. (This functionality also updated down the tree in all affected files.) ~ ConstantSource interface now must implement getConstantValue() - Fixed minor BV bug in toBinaryString() and toHexString() + Added hasBehaviorInClockMethod(), hasBehaviorInPropagateMethod() to Cell, implemented it where needed in base. + toString() added to Simulator, GlobablSchedule. ~ MemoryInterface now uses BV instead of BigInteger. ~ BV.toBigInteger() method is now public. - ClockDriver and all children in tree updated to use protected clock wire member. Added a package scope accessor for base to be able to fetch the clock wire. - Fixed some join bugs relating to parameter order, and interaction with techbenches. (Bugzilla #188) + Test suite for join() added. (Yeah!) ~ Simulator changes: Reset is a little kludges and settlePropagate settles the network correctly now. More work needs to be done to cleanly reset and to remove code that represents bad ideas. Fixed step and cycle counters to update correctly. Added patch to allow GlobalSchedule to accept input from the test_benches clock() function. This may or may not have to stay for compatibility ... but may not be a supported feature in the future. Logic ----- ~ By default, Logic will use XC4000TechMapper as the default target, instead of the previous default of GenericTechMapper. ~ setTechMappingEnabled(boolean) and techMappingEnabled() removed from all TechMappers and added instead to Logic. Xilinx ------ ~ Changed the memory interface so that it uses BitVectors instead of BigIntegers. ~ Made shiftl a protected method instead of package so that the Memories in Virtex could use it. + XilinxFD_1 and XilinxLatch_1 classes added to support negitive edge triggered primitives. + Added pulldown_g to compliment pullup_g. XilinxTechMapper modified to correctly handle wide pulldowns instead of printing and incoherent error message. - Error in pulldown's test vector fixed. Virtex ------ ~ Changed tb_BlockRam to use the new RAMB4Dual instead of the deprecated BlockRam. + All fd*_1 primitives added. - Removed stray ibuf that should have been up in Xilinx. ~ fdc and fdp are now used correctly by the VirtexTechMapper. ~ clkdll modified to not be scheduled, but to correctly show up in schematics. jmac ---- ~ globals.pl modified to support XilinxFD_1 and XilinxLatch_1. Modgen ------ + Added checking for unsupported Virtx arrayMult configuration that sets the netlisters setInsertTechMapHints flags to false. - Removed temporary fix that was in tb_float. - In RamRom, fixed a bug in the dual ported ram in Virtex. + Added support to RamRom for outputs greater than 32 bits. util ---- + Sort.java and Compare.java added for general sorting functionality. Sort.java is intended to provide static sort routines for sorting an array of objects. So far only bubbleSort is supported, but plans for Sort.quickSort are coming soon. Comparator.java is a generic interface for comparing objects. Each sort routine in Sort accepts a Comparator to determine how to order the elements in the array. (Note: this interface is compatible with the Comparator interface in java.util with Java 2. We're just coding this one up for now while we are still using Java 1.1. When we move to Java 2 we should be able to eliminate this class.) Viewers ------- + Multiclock is now supported in all Viewers classes. - Fixed a bug in the interaction between the memory viewer and the new MemoryInterface. - Fixed a bug in MemoryView so that you can now view more than 30 rows. + SchematicCanvas will now display a dialog box indicating that a cell is a leaf cell when it is double clicked to view its children. - Fixed a bug that threw an exception when trying to build mapped cells. Jab --- + Multiclock is now supported in Jab. - Fixed Bugzilla #160. The stack trace is no longer printed if behavioral model exception is thrown. TBone ----- + Multiclock is now supported by TBone. ~ Changed TBone to use Logic's default TechMapper rather than explicitly setting it to GenericTechMapper. - Fixed a bug in LoadClassCommand that wasn't properly catching a NullPpointerException. RELEASE 0.2.13 ============== Base ---- + Modifications to HWSystem to support schedules other than four wide. This does not allow mixed widths, but makes things nicer for multiclock or other simulation situations. + Added Util.makeLegalJHDLIdentifier() at the request of bugzilla #174. ~ Modified bus-wise netlsiting to use correct names (or lack thereof) for nets. + ClockDrivers can now have an input port or not, depending on how they are constructed. - Deleting a ClockDriver no longer leaves a dangling clock wire. + Fixed bugzilla #177, #178 - constants no longer have clk port, show correct value in JAB. Logic ----- + Schedules other than 4 wide are now supported -- Yeah. ~ "Targetting Logic class" message no longer prints unless a different techmapper was set as default. + Added call to construct ClockDriver with no input port. - Removed gnd() from input to default clock driver. ~ Renamed the default clock and clock driver to be less annoying. Xilinx ------ - pullup_g no longer has an erroneous CLK port. XC4000 ------ + New test added to do test of cell transformations. + Additional miscellanious tests added/updated/removed. Modgen ------ + Added a non-zero exit on failure in modgenTester. + Added useful information in the area section of arrayMult. Viewers ------- ~ Changed FloorPlanner and LayoutView to just use whatever TechMapper comes with Logic rather than explicitly setting Logic to use GenericTechMapper. CLI --- + Fixed ClassPathBrowser to use the path.separator Property to determine what character is used to delimit elements in the CLASSPATH. (This is the correct way to do it, as opposed to checking if the os.name property is Windows) RELEASE 0.2.12.2 ================ Release ------- + Missing EDIFparser directory now completely included in JHDL.jar XC4000 ------ - ifdxi fixed by adding ExternallyUpdateable interface. - Fixed a bug that did not place the fmap's associated with fdc's. The leaf source needed to be explicitly placed. ~ Deprecated the TechMapHelper.getSourceCell() method. Now library writers must explicity ask for hierarchical or leaf sources. RELEASE 0.2.12.1 ================ XC4000 ------ - Various fixes the the regression tests. - Split the internal TechMapHelper.getSourceCell() method into different functions. Viewers ------- - Minor synchronization update in GenericFrame. - Fixed the bugs with the incorrect scrolling and apparent non-updating in MemoryViewer, as well as the menus being overlapped by the buttons. Modgen ------ + Virtex documentation added. RELEASE 0.2.12 ============== Base ---- - Fixed NullPointerException in PortBlock. - Fixed NullPointerException in Cell.getImplicitInterface() + Added Cell.getSubCellClass() to aid dynamic hierarchy in avoiding NullPointerExceptions relating to implicit ports. + Added an error message for misspelled CellInterface[] (Bugzilla #152) - Fixed addPorts with different parameter bindings. (Bugzilla #153) Logic ----- + Added Logic.getSubCellClass() to aid dynamic hierarchy in avoiding NullPointerExceptions relating to implicit ports. XC4000 ------ ~ Updated the getTechMapHint() method to handle output wires who's source is a non-placeable cell. Now the function will return the placement hint for the placable source of those wires. + XC4000 Regression tests now run on make test target. Virtex ------ ! BlockRam has been deprecated. It had problems with single-port and one-bit width memories. The classes to use in it's place are called RAMB4Single and RAMB4Dual. These should be instaniated for single and dual ported BlockRams, respectively. Modgen ------ - Added error checking to arrayMult for cases not currently supported by the Xilinx tools in Virtex. - Fixed a bug in virtex version of multCol that made a carry_cell that only had buffers (and thus not placable). + Added a switch to turn off placement hints when running arrayMult under virtex and pipedepth is greater than 1. + Added virtex testing of arrayMult and also turned on techmapping for both XC4K case and Virtex case to test more things when tested. Also increased number of mulitpliers tested from 5 to 10. + In Cordic, added exceptions for unsupported angletypes and cordicmodes. - Fixed the puts in downcnt to null term_cnt wires in reset() and clock(). + Updated booth to work for larger inputs. - Fixed a bug in booth's behavioral model. Jab --- - Added a catch for an exception in no-window mode. Now it prints a somewhat meaningful message instead of a vanilla stack-trace. Viewers ------- + Moved the window start location down to (50,50) so that the title bar does not appear off screen. RELEASE 0.2.11 ============== Base ---- + Observable interface added. This is used for adding execution functionality to user designs that do not extend Observer. + Tests added to exercise the observable interface. + First cut of adding ConstBV. - Modified Parameters to stay around even after cell's construction (Bugzilla #151) ~ Delayed check for 0 ports until the DRC of verifyAndCleanup, so that fmap_g can start with 0 ports and dynamically add them later. Logic ----- - getTechMapHint(Wire) now will work with wires of width greater than 1. ~ Logic calls modified to support generic width gates returning wires with correct widths. Xilinx ------ + Added setPrintTimestamp() to edif and XilinxTechMapper to allow true "golden" netlists. - Generic width gates wider than 5 no longer return an incorrect wire width of 1. (Bugzilla #111) + Test added to test bit-wise boolean gates. edif ---- + Added getInsertTechMapHints() to all the checking of the value set previously by setInsertTechMapHints(). XC4000 ------ + Added setNotVisible(), and isVisible() to fmap_g so that it will not appear in Viewers. ~ fmap_g now extends XilinxCL instead of Logic so that connect_implicit_ports does not create an erroneous clock port in the fmap_g's. ~ cy4 modified for use with the EDIF parser: ~ The C port is now eight 1-bit ports. + Named association has been added since the cy4 is a primitive. + An optional string parameter can be added to cy4 to specify the mode of operation. Virtex ------ + An very helpful error message (i.e. practically a tutorial) is now printed out when map() is attempted for widths wider than 4. jmac ---- - Generic width gates (generate by boolgen and jmac) wider than 5 no longer return an incorrect wire width of 1. (Bugzilla #111) - boolgen modified to be generic enough to make gates for both Xilinx and CSRC ~ jmac move from byucc/jhdl/Xilinx/jmac to byucc/jhdl/util/jmac - jmac now puts the CORRECT location it was generated from in the comment in generated java files. Viewers ------- + Added setNotvisible(), and isVisible() so that fmap_g and other such non-visible cells will not appear in Viewers. - Updated the wave viewer to correctly display cycle numbers and wire values after a hardware mulit-step. Waves and Table modes have been fixed. This update places a column of grey (or ?'s) if a multistep has occured making it much easier to read the wave viewer in hardware mode. Now, both simulation mode and hardware mode match perfectly. - Fixed "cycle" and "go to cycle" fields for new waveViewer in hardware mode. Jab --- - Updated to catch some ClassCastExceptions. - Updated support for -nw mode. Modgen ------ ~ Modified netlist testing to use the new flag in edif to disable timestamping. RELEASE 0.2.10 ============== Release ------- ~ Per suggestions from lab members, slightly modified the "key" for the release note prefix markers, as shown: Examples: + Something added or new. (+). - Bug fixed or something removed. (-). ~ Feature modified or changed. (~). ! Major change or critical fix. (!). ~ Most makefiles now set the CLASSPATH to ${TOPDIR} instead of doing a cd before each call to javac. Base ---- + The total_clock_count is now updated when in hardware mode. This allows querying of the current clock cycle in hardware mode. ~ Modified pushHierarchy(), popHierarchy() to return a cell of the variety specified by the library writer (ie. a dynamic subcell can now be Logic) - Changed the Simulator run( int no_clocks ) method to correctly initialize. This fixes the bug that prevented Simulator from corretly counting cycles in hardware mode. Logic ----- ~ Modified pushHierarchy() to grab a Logic object for the subcell instead of an orthogonal base package cell XC4000 ------ ~ The techmapper function getAsyncoutput() was updated to navigate the extended tree. The tree has changed with th addition of fmap_g. Same for the getSinkLeaf() and the getSourceLeaf(). This should fix Bug 120, mapping luts with ff's attached. + ifdi added to XC4000 library. ! This commit represents some major bug fixes. 1) All map methods now return the mapped cells. That is the fmap or fmap_g that was created. This way map methods can be nested within place methods, or used to get the fmap cells themselves. 2) fmap_g's now have a unique cell name that matches their structure so that cellnameDeterminesUniqueNetlistStructure() can be used on cells that have generic width fmaps inside them. 3) fmap_g now support string hints to space the individual fmaps. These hints are the standard ".f", ".g", and ".h". When these hints are used fmaps within fmap_g will be spaced one per CLB. + Added further XC4000 techmapper tests. They test relative placement and cell transformations. - Fixed getTechMapHint function to return the source of a wire with a width greater than 1. Virtex ------ - ifd_1 and ifdi_1 implementation fixed. + mux41 and mux81 added to Virtex.modules package. These are specialized 4:1 and 8:1 muxes that use the specialized Virtex muxing logic. mux41 uses one slice, minus the FF's. mux81 uses one CLB, minus the FF's. Because some of the muxing logic is in the CLB itself and not in each slice, mux81 can ONLY be placed on CLB boundaries (even x locations). Modgen ------ ~ accum, delay, downcnt, upcnt, and ramrom updated to reflex changes in the VirtexTechmapper. + Modgen tests now include netlisting. Netlists are compared against "golden" netlists that we know work. This is done for both Virtex and XC4000. + downcnt specially optimized for Virtex. It requires 1 column of slices if term_cnt isn't used, otherwise it uses 2 columns. This is required to avoid an asynchronous loop. + Added complete Virtex tests to ModgenTester. ! Modgen is now fully supported by Virtex. Jab --- + BrowserCore.hasWindows() method added. This returns if you're in no-window mode or not. RELEASE 0.2.9 ============= Release ------- + Added new markers to release notes. They are pretty straightforward. Suggestions or comments on this are welcome. Examples: + Something added or new. (+) - Something fixed or removed. (-) ! Major change or critical fix. (!) + Xilinx/Virtex/test added to configure. - Xilinx/XC4000/syms dependancies reordered to compile correctly after a make clean. Base ---- - Made all netlisting related classes public. - Fixed a scoping problem on Cell field wire_array. Logic ----- ! source() call deprecated. Instead, use getSoureCell(), getSourceLeaf(), getSourcePlaceable(), or getSourcePlaceableLeaf(). Almost all user usage of source() can be replaced with getSourceCell(). + Added explicit clock port constructors. (NOTE: clk is always the first wire parameter) + Generic gates updated to have explicit clock constructors. Xilinx ------ + Explicit clock constructors added for all primitives in Xilinx, XC4000, and Virtex. + XilinxTechMapper, XC4000TechMapper, and VirtexTechMapper's all updated to support new getSource...() calls. Virtex ------ ! MAJOR CHANGE: Updated map() and place() to support fmap_g's, and all standard Logic placement. All standard placement and mapping constructs should now work. There are still a few unimplemented features (multi-lut automapping, coherency-checking) that will be added shortly. CSRC ---- + CSRCTechMapper updated to support new getSource...() calls. jmac ---- + Changed jmac to use wires based on target packages. + Token @Wire now accepts strings as wire widths. - Variable $generic_width is now initialized correctly (to 1) by default. Viewers ------- + Histogram viewer added. TBone ----- - Removed some debugging println's + Added tab completion. Modgen ------ + Modgen cells all updated to work completely with Virtex. All placement is not necessarily optimal, but all compile, simulate and netlist correctly. RELEASE 0.2.8 ============= Base ---- - Added getAtomicMapping methods, getBusNet() to BusConnection. - Uncommented several lines in the constructor so that the lists and schedules are not null if deletePropagateable is called before the Simulator is initialized, e.g., as a circuit is being built before simulation. - Nuked reconnect(), because it is too hard to implement correctly. - Fixed parsing error for a one-character non-numeric string. List ---- - Added insertBeforeCurrent() to linked and vectorlists. - Updated the documentation to be more explicit on behavior of the current pointer by method. - Added checking for reads beyond the list end. Viewers ------- - Fixed problem with "view child wires" option. - Fixed problem with cellinfo, wireinfo, and portinfo. TBone ----- - Removed dependancy to parseInt. TechMapper ---------- - Changed the name string to be the last parameter in TechMapper logic calls, to be consistant with everything else in JHDL. Xilinx ------ - Changed implicit clocks to be the first parameter. RELEASE 0.2.7.1 =============== XC4000 ------ - Fixed bad reference to fmap_g that made XC4000 break. RELEASE 0.2.7 ============= Base ---- - Fixed a bug where bind() after addPort() caused arrayOutOfBoundsException. - Added the method reconnect() to complement connect(). - Reworked getShortedPorts because it was non-deterministic with a hashtable (Bugzilla #116) - Changed the schedulePullUpDowns() method so that it would work correctly when the sink of the pullup/pulldown was at the top of the propagate list. The solution just checks for this case. A better fix would be to add insertBeforeCurrent to the linked list class and use that instead of list.prev(), list,insertAfterCurrent(),list.next() in order to insert the pullup/pulldown before the wire which is its sink. - made getArgument() public, since it replaces the deprecated public getGeneric() Xilinx ------ - Explicit clock port added to all clockable primitives. - Clock ports named 'C' changed to 'c' to be consistant. XC4000 ------ - Replaced 'new fmap()' calls with map() calls in the adder and subtractor modules. Virtex ------ - Added missing clkdv port to the clkdll primitive. - NOTE: map() and place() are still not fully implemented. - ifd_1, ifdi_1, and ofde primitives added. jmac ---- - Now calls wires based on package instead of using LogicWire. - Modified jmac so that it would preserve the ordering of ports between the order they are declared in the jmac interface and the order they will appear in the cell constructors. (Previously, all implicit ports appeared at the end.) The nitty-gritty is that now instead of having separate arrays for portnames and implicit_ports, we now put all implicit ports in the portnames, portios arrays. The entry in the portios array is $IMPLICIT rather than $INPUT, $OUTPUT, or $INOUTPUT. (I guess the assumption is that we will only have implicit input ports. This would be easier to expand later if we want implicit output ports.) - Fixed a jmac bug that made structural cells cells call their subcell constructors with wire parameters in the wrong order. Modgen ------ - Added rounder to Modgen library. - More modgen cells updated to support Virtex. - More thorough Virtex modgen tests added to modgenTester (do not all pass, yet, but will when Virtex map() and place() are improved) - Modified upcnt and downcnt so that they can take advantage of XC4000 carrylogic when running that techmapper, but use only logic components when running any other techmapper. - Updated ramrom generator to not call the ramrom_struct and ramrom_clk classes anymore. Also, modified all place calls to use relative placement. When running in virtex mode, placement is disabled in this class, but once map is fixed in virtex, it should be just a matter of removing some flags to enable placement in virtex. Templates --------- - Minor changes to JHDLWriter. - Fixed bug #60 in bugzilla. NewTB shouldn't get confused anymore about where to include the string for "instanceName" in the constructor of the JHDL cell RELEASE 0.2.6.4 =============== Release ------- - Pruned several empty directories from JHDL.jar Edif ---- - Added a switch that will omit rloc strings from the netlist. This switch is called setInsertTechMapHints(boolean b). Virtex ------ - Removed extraneous ibuf.java that was still sticking around. This was moved up to Xilinx quite a while ago, as it is a shared cell with XC4000. - Continued updates on the Virtex techmapper map() and place() calls. Currently, all of place functionality should be there, and most of map(). There is still the limitation of not mapping functions with more than 4 inputs (other than wide-boolean functions). Map also does not (yet) automatically trace back inputs of FFs, nor does it do a complete network check to insure legal fmap's. This will be added soon. - fdc and fdp (formerly generic-width functions that called fdce, ala XC4000) are now fdc_g and fdp_g. Real fdc and fdp primitives have been made. Modgen ------ - All of modgen now passes the modgen tests while targeted to Virtex. Netlisting under Virtex for all of modgen has NOT been fully tested, however. RELESAE 0.2.6.3 =============== Helpers ------- - slaac1v helper added to JHDL.jar Base ---- - BitVector changes: - Added barrelShiftRight and barrelShiftLeft, and setRange. - Fixed minor bug in upperBits. - Added ability to set value to a boolean. - Added ability to set value to an array of BVs. - Changed return type of methods like setBit, clearBit from void to BV. - Updated documentation to reflect changes - The ExternalUpdateManager no longer toggles the HWUpdate flag on ExternallyUpdateable and Checkpointable Cells every cycle. Instead, enableHWUpdate and disableHWUpdate have been added to the class to perform this functionality. These calls must be made when the simulator makes the transition between hardware and simulation modes. - Modified the simulator to call the ExternalUpdateManager's enableHWUpdate and disableHWUpdate methods when transition is made between hardware mode and simulation mode. Further, I added some code in the initialize() method to handle the building of the ExternalUpdateManager's list of ExternallyUpdateable and Checkpointable Cells as well as properly disabling and, if needed, enabing the hardware update of these Cells. These should make the simulator much faster in hardware mode. Edif ---- - Modified edif.java so that netlist-leaf cells (Xilinx primitives) do not use port arrays. Port arrays are still available for non-netlist-leaf cells. Virtex ------ - Virtex techmapper will now use fmap_g's. Modgen ------ - accum, upcnt, downcnt, lt, and gt modified to return false for cellInterfaceDeterminesUniqueNetlistStructure() RELEASE 0.2.6.2 =============== Base ---- - Join bug (Bugzilla #84) fixed. - Multiclock now grabs the correct default clock (Bugzilla #96). The call getSystem().setDefaultClock() has been changed to simply setDefaultClock() which is a method in Cell. - Because of fixes with multiclock, Virtex designs will no longer optimized themselves into oblivion (problem in 0.2.6.1) Logic ----- - Modified all cell transformations to return cells. This was so they could be nested within place methods or each other. They previously returned void. XC4000 ------ - Improved techmapper error messages. Modgen ------ - Modified accum, booth, downcnt, gt, lt, upcnt, and floatMult to be technology independant. RELEASE 0.2.6.1 =============== Base ---- - Fixed a ClassCastException when doing a reverse lookup to a hashtable (Bugzilla #85) in Cell - Explicit clock check commented out and made a slight change to what can be put on the BogusSchedule. - Port names must now be JHDL legal. (Bugizzla #87) - Fixed a NullPointerException when deleting a wire. - Added multi-clock support within the buildstack, so that the default clock affects only subtress. - List.intersect died on attempt to create list by reflection. This has been fixed. (Bugzilla #91) Logic ----- - Fixed a bug in relational port placement that was doing unnecessary error checking on primitive cell ports. Virtex ------ - Added ofdi primitive. - Virtex map() call now supports any function with 1-4 inputs, including "iterated" mapping. No internal checking is done. Viewers ------- - Added HistogramWindow and HistogramDialog that allows data to be viewed as a histogram. Jab --- - Cleaned up "win false" command. TBone ----- - Added support for inout ports. RELEASE 0.2.6 ============= Release ------- - The XC4000 techmapper regression test is now fully integrated into the "make test" target! =) Fsm --- - Added "buildFsm(InputStream s)" which will take the description of the FSM to be built from an input stream. - Added "buildFsmFromString(String s)" which will take the FSM description to build from a string. Logic ----- - Updated the map call that takes a Wire[] to protected from private scope, to support hierarchical fmaps. XC4000 ------ - Added the functionality of hierarchical Fmaps. This affects iterated placement which has been depricated in the common case. All generic-width combinational logic that is mapped together will now be found in a layer of hierarchy called fmap_g which is an instance of MapWrapper in Xilinx/XC4000. - Some changes have been made to the XC4000 techmapper to support the Regression tester (listed below). - The getTechMapHint() function now returns a cell hint regardless of wether the cell placement is locked or not. - In very special cases the Mapper sometimes ignored a hint on a wire in the iteratedPlaceWire() function. (Lines 755 to 764 take care of this oversight.) - The changePacking() function in XC4000PlacementInfo has been modified to accept a boolean and all calls to this function in Mapper have been modified accordingly. - The changePlacement() function now checks to make sure x and y are not unconstrained before it sets the placement. - The changePacking() function now accepts a boolean which will determine wether the cell is locked or not. All function calls in the tree have been modified accordingly. Virtex ------ - Removed redundant primitive, ibuf. (It was already in Xilinx, and causes problems if it's in both places.) - Updated GenericFrame with the getInterp() method. Viewers ------- - Fixed the MemoryViewer dump range button to map to a cli command that prints to the console. This makes writing scripts a little easier but looses some of the special features in the MemoryViewer writeMemoryRange() method. TBone ----- - Changed dependency on base.Flags to base.BooleanFlags. This solves the problem of TBone not recognizing inport or outports in the last release. - Added support for wires larger than 64 bits using BitVector. RELEASE 0.2.5 ============= Base ---- - Removed obsolete files CBus, TBus - Updated Bus to use Wire[] concat-style constructor instead of WireList. - Rewrote the join() algorithm to remove the problem in Bugzilla #61 and to make it faster. - Deprecated the wire constructor that does concatenation via WireList, in favor of one that uses Wire[]. - Made BooleanFlags public so that the constants can be accessed outside of base. - Added equals() and hashCode() to Wire to allow hashing of wires based on identical valuepropagater contents. Logic ----- - Added a Wire[] style concat constructor to replace to WireList style. Edif ---- - The Edif netlister now will use port arrays to declare ports. - Changed getCellNetList or getCellNetlist (change in capitalization) Xilinx ------ - Updated the Xilinx Wire constructors to reflect changes in Base and Logic. XC4000 ------ - RegressionTester ressurected to run a fairly complete test of the XC4000 techmapper. This is not yet run in the complete "test" target because several tests fail (and/or are not yet being run) because of apparent incompatibilities with the newer releases. NewJHDL ------- - Now supports implementing multiple interfaces. - Will now ask whether to extend Logic instead of requiring that it be entered manually every time. - Provides the unnamed constructor to call the "real" constructor (this, btw, will make NewJHDL circuits work with TBone much more readily) - NewTB will import the package containing a JHDL cell isntead of being part of that package itself. - NewTB will ask whether the clock method should print out the inputs as well as the outputs. - NewTB and NewJHDL implementation renamed in the build tree. It's still invoked from the commandline the same way (java NewJHDL ...) RELEASE 0.2.4.1 =============== Jab --- - Fixed setWavesTypeCmd to pass the correct value into the BrowserCore.setWavesType() method. Viewers ------- - Fixed the table waves option to correctly display wire values in table of wave form. - Fixed a problem in waves that displayed values one cycle behind the simulator. - Fixed GenericFreame menus to update immediately when the mouse is released rather than waiting until the event has completely processed. - The name of the memory will now show in the title bar of the memory window. Xilinx ------ - Generic width buffers added, ibuf_g, obuf_g, and obuft_g. These are *VERY* handy when viewing schematics that have hundreds of ibufs and obufs (like almost any board model). XC4000 ------ - ofdi primitive added. Modgen ------ - Added buf() calls around the address lines for every primitive ram call in the RamRom generator so that the address permutations can be determined for readback. RELEASE 0.2.4 ============= Base ---- - Unconnected circuits are all straightened out. There should be no problems with the circuits that are not connected and need not be simulated. - BV.range() made more consistant with other JHDL object's range() methods. - Made lists use Object.equals() instead of == as the comparison method for searching a list. - Fixed bugs in BV: getBit, setBit, toggleBit. Logic ----- - Modified the port placement methods for finding the location of input and output ports so that non-placeable cells are skipped over. This is accomplished by searching through the port records and getting attached wires. Some javadoc comments were also added. - Fixed a NullPointerException in Logic/gates/memories/SyncRam that occurs if init[] is null. Jab --- - Added clearCheckBoxes to facilitate BrowerMainFrame clean-up when the WavesViewer is closed. Viewers ------- - Changed the default GenericFrame location to ensure that JAB does not load off-screen on startup. - Fixed a problem in updateGUI() that would dump a null pointer. - Fixed the "go to cycle" in the waves viewer to hold cursor focus. - Added "jump to location" ability to the memory viewer. - Fixed a problem in the info field. When loading a class from the SchematicViewer it will only display the name of the loading class not the extended class type. - Added viewManager pointer to GenericFrame to help manage closing the proper windows. - Added Tool.getName() method which returns the key that ViewManager uses to store the view in its hashtable. This is now implemented in all classes that use the Tool interface. - Housekeeping upgrade that fixes a lot of little bugs that had to be done all at once: ViewManager now manages all windows correctly. If a window is already being displayed it will bring it to the front when the command that created it is invoked. When the window is closed with the close box or the menu option it is successfully deleted from the hash tables. The wave viewer will now clear the check boxes in the Main Frame when it is closed. It also closes automatically when no wire are watched. - Fixed a bug that arose when ctrl-M was pressed in the schematic viewer. - Fixed various things in WatchInterface to help the ViewManager display and dispose this view correctly. Modgen ------ - Clarified an exception message thrown in floatMult. Templates --------- - Minor modification of some wording in NewJHDL and NewTB. RELEASE 0.2.3.1 =============== Viewers ------- - Fixed deprecated method .show() to .setVisible(true) in ViewManager. - Added another button to the memory viewer so that addresses and data can be viewed in difference bases separately. Templates --------- - Made helpers for NewJHDL and NewTB. Amazingly enough, you may now invoke their powers with a simple "java NewJHDL" or "java NewTB". RELEASE 0.2.3 ============= Base ---- - BitVector (BV) setValue() bugs fixed. - BitVector stringBitWidth() added to determine number of bits to represent string. - BitVector hashCode() added to allow creation of Hashtables of BVs. Logic ----- - Changed the integer constants for Directive directions, (ONTOP, LEFT_OF, etc.) to be small numbers. (This works around a bug in JDK 1.2) - Added versions of ashiftr, shiftr, and shiftl that accept Strings as names for the new wire. - Updated LogicMethodTester to be consistent with latest and greatest JHDL. - Added ConstantTest to the regression testing. Xilinx ------ - XilinxTechMapper updated to allow netlisting even if TechMapping was never enabled. - Added constructors with clock wires for all clocked elements in Xilinx, XC4000, and Virtex libraries. For the ordered-wire constructors, the clock is the last wire. Named association constructors exist with and without enoush string-wire pairs for the clock wires. ArgBlockList constructors should also work with the clock wires but this has not been tested. - Fixed a naming bug in vcc() and gnd() that was breaking readback. - Removed vcc() and gnd() calls that constructed wires from all TechMappers, now all wire construction for vcc() and gnd() is done in Logic - Fixed a bug in that Xilinx was not naming constant cells. jmac ---- - jmac has been updated to emit constructors that support explicit wires for implicit ports. Modgen ------ - Cordic placement updated. - many 'new' calls changed to appropriate Logic calls Templates --------- - NewJHDL generator added to JHDL.jar RELEASE 0.2.2.2 =============== Base ---- - Fixed code to be backward compatible with testbenches that extend Synchronous. - intersect() methods in lists now all use reflection to create the correct type of return list. Logic ----- - Fixed the bounds checking for 64-bit constants. GenericTechMapper ----------------- - Add, AddSub, and Sub fixed to get rid of UnboundParameterException. XC4000 ------ - cy4 blocks changed for more support. cy4 now has named association, a constructor that allows you to pass the mode into cy4, and can now use the string "FORCE-0" or the number "01" in the cy4_mode block. CLI --- - The commandline now resets when an exception is thrown. Viewers ------- - Fixed duplicate error message in popup dialog during InvocationTargetException. Jab --- - BrowserTree now catches a null pointer exception when clicking in the hierarchy pane with no circuit loaded. - BrowserCore & BrowserMainFrame file menu fixed to correctly call loadclass. - loadclass now correctly displays a dialog box. - loadclass now only calls reset if not using TBone (TBone calls this itself when it's ready). - Jab now closes down completely when the close box is clicked. - Cell.verifyAndCleanup() is now called with all loading by reflection. TBone ----- - Fixed a catch statement to correctly report exception through TargetInvocationException. Platforms --------- - Made sure Cell.verifyAndCleanup() is called with all loading by reflection. Readback -------- - ReadBackData optimized for speed and memory size. - Added toString method to Block, which dumps most of the interesting information about the object. - Array indexing bug fixed in RBSym.permuteAddress(). - Fixed reversed MSB problem and hard coded address bits in RBSym.calcPermutedAddress(). - Error messages and exception handling messages are now more clear. - Other various changes. RELEASE 0.2.2.1 =============== CVS --- - Fixed a bug in our auto-tagger that prevented commits of files over two log-lines long from being tagged correctly. Base ---- - The simulator now allows circuits with source pieces to go forth. It did before, but in an odd way. The current method is not better, but gets the job done. The fix applied in 0.2.2 started to remove the need for dont_propagate lists. This fix separates how unscheduled ciruit elements are handled and is stepping forward in a "better" direction. - Fixed the error message when cell_interface is not static. This is needed for internal heirachical cells. - ListTest added to test the lists used in base in the standard regression test. Logic ----- - Fixed the naming for vcc() and gnd(). - Fixed the missing name on one of the add_o calls. - Added static nc() methods. Xilinx ------ - Removed the exception that gets thrown if TechMapping is not enabled before netlisting. Virtex ------ - Added multiclock constructors to BlockRAM. - Added support for PreDefinedSchematic to Virtex Shifter module. Modgen ------ - Updated exception message in upcnt and downcnt to be more helpful. Platforms --------- - platforms/util is no longer included in JHDL.jar, but will be included in all platform specific jar files. Readback -------- - Makefile.in's created for entire readback tree. - More work on platform independant raw readback stuff. See release notes from 0.2.2 RELEASE 0.2.2 ============= Base ---- - The constant schedule is now turned back on. This also will affect other designs because stricter connection rules are enforced. - Added the Uniquify to the IncPropSchedule. This sorts and moves the entire schedule to the first clockschedule in a particular time step. This avoids multiclock problems with ordering of inputs and outputs. - Fixed corner case in LinkedList.delete() that broke when altering deleteCurrent() GenericTechMapper ----------------- - Added support for PreDefinedSchematic. - Shortened port names. Xilinx ------ - Added a boolean to Logic and XilinxFD to decide whether or not to connect implicit ports. Virtex ------ - ifd and ofd added to the Virtex library as primitives. CLI --- - Changes made to CLInterepeterKeyListener so that the new GenericFrame will work properly. Viewers ------- - Modified GenericFrame and all related classes so that you can put tabs inside of any GenericFrame. - Schematic symbols for one-input gates made prettier so they don't look so squashed. Jab --- - Two new commands added. dumpram will dump the contents of any cell that implements MemoryInterface. win sets the window mode of Jab. The default is true. When win is set to false, all panels, such as schematics, waves, and viewers appear in the main frame as tabs. Tbone ----- - Various fixes made to Tbone. Platforms --------- - Initial addition of platform independent, Xilinx part-specific readback handling for board models. Specifically, this is for XC4000 Xilinx FPGAs. The C++ related files are useful for handling some of the JNI related tasks. RELEASE 0.2.1 ============= Base ---- - Added push and pop heirarchy support. This allows having multiple levels of heirarchy inside a single classfile. - LinkedList.deleteCurrent() now leaves the current pointer on a valid node after completion. Logic ----- - Updated the clockDriver Logic call to take an input wire (now the call is "clockDriver(Wire clock_in, Wire clock, String Schedule, String name);"). This input wire will be *completely* ignored during simulation, but at netlist time will be the wire that drives the bufg. Thus, if you are letting JHDL add iobufs for you, it should be a clock port in your design, or, if you are adding them yourself (like in a board model) the clock_in wire should be driven with an ibuf. - Updated GenericClockDriver to work with new clockDriver() method. GenericTechMapper ----------------- - All the faulty GenericTechMapper Logic gates should now all simulate correctly. Xilinx ------ - Fixed versioning inconsistancies in edif. - Fixed the bug where bufgp.java was being generated twice (and as two different things!) - Updated XilinxClockDriver to work with new clockDriver() method. Modgen ------ - Modgentester testbenches updated to extend Logic instead of Synchronous. Jab --- - loadModuleCmd now catches invocationTargetException and prints a proper stack trace for it. RELEASE 0.2 =========== Release ------- - Here it is, finally! JHDL Public Release 0.2! =) RELEASE 0.1.9.1 =============== Release ------- - Okay, this time this is a real release candidate. Are all the bugs fixed now? =) Xilinx ------ - Got rid of two quotation marks in bufgp that were causing problems. XC4000 ------ - Fixed a bug in tagging iteratively placed cells as unique. Virtex ------ - Added calls to modgen.ramrom for ram(), rams(), and rom() in the VirtexTechMapper. - javadoc comments added to BlockRam Modgen ------ - RamRom now supports Virtex. RELEASE 0.1.9 ============= Release ------- - This is called releaes 0.1.9, but it's really 0.2 release candidate 1 in disguise. =) Test it to your hearts content! Base ---- - Added setHeirarchicalUpdate to handle the case where a netlist leaf has structure for simulation, as with the synchronous LUT RAM models (e.g., ram16x1s). The useHWUpdate cell of HWSystem cannot handle this situation, so the call is used to pull off cells from the clock list (and now, their children). XC4000 ------ - Fixed a bug in checkNetworkFan. The Network constructor traces back using getLeafSourceCells, but checkNetworkFan was tracing forward using getSinkCells. This was changed to use getLeafSinkCells. RELEASE 0.1.8.4 =============== Base ---- - Fixed order of getSystem vs. clearing parent pointer when deleting a cell. - Made Wire.delete public, and made Cell.delete more robust when trying to delete same cell twice, avoiding a null pointer exception. - Fixed bug in join() that was causing it not to affect the entire tree. Xilinx ------ - Fixed cell interfaces in netlist to ignore non-netlistable ports. Jab --- - Jab modified so that all windows are destroyed when a new class is loaded. - Changed the loadNewClass method so when you load a testbench after loadint Tbone, everything works properly. You cannot load a non-testbench file after this is done, however, due to the inability to remove modules from CLI. Tbone ----- - Added support for wires up to 64 bits wide. - Fixed problem with trying to load other files after the first load. - Fixed problem with multiple circuits after target command is called. - Fixed vectorfile command to enable vector file input without specified outputs. - Fixed unnecessary loadmodule command. - Fixed several other bugs in initializing the hardware system, the target command, and deletion of floating wires. Modgen ------ - CordicPack placement updated to work with new release. - Sub placement updated to work with new release. - modgenTester modified so that it does not simulate the ramrom generator behaviorally, as this is not currently supported. - RamRom and RamRomPack generator behavioral modeling disabled. RELEASE 0.1.8.3 =============== Base ---- - In LinkedList.java, deleteAll does not delete anything but the current object in the list. This continues as an infinite loop deleting the current over and over, so it is probable that no one calls this function. - Removed more debugging messages that are no longer necessary. CLI --- - Changed CLICommandEvent to fire the unaliased name of the command so that command listeners such as wildforce can correctly interpret a quit command to close the board before exit. Viewers ------- - Added hooks in Schematic to CircuitView to explicitly kill a thread in the Schematic Viewer. This fixes a problem that arises when using native threads with JVM. - Removed some debug print statements that were no longer needed. Tbone ----- - Removed the necessity for including output values in vector files. - When targeting a (new) techmapper, the old circuit will now be replaced by the new circuit. - NullPointerException is no longer thrown when you put a value before stepping the clock. RELEASE 0.1.8.2 =============== Jab --- - Added code to explicitly dispose the BrowserWindow when exiting the program. - Modified method killCurrentWindow to be public so the ViewManager could explicitly kill the waves window. Tbone ----- - Modified buildTbone command so the HWSystem would be initialized when the circuit was built. Platforms --------- - platforms/util is now included in JHDL.jar so that it would not need to be duplicated in every platform's jar, and would be easier to maintain. - Changed loadUserCore() so that it will throw a ClassNotFoundException if it cannot find the class. This way, the board models can catch it and itnogr it when in hardware mode. - Added DropinInterface, which is simply a tag class meant to be used by users to pass information to board models. RELEASE 0.1.8.1 =============== Base ---- - Changed the memory-related getWidth() to the less ambiguous getMemoryWidth(). Changes related to this applied all the way down the tree. Modgen ------ - Cleaned up some extra garbage being generated in some Multiplier cells. Jab --- - portWireEntry split into it's own class for ease of compilation. - Fixed JAB so that wire values in the table get updated when any external step command is used. RELEASE 0.1.8 ============= Release ------- - Added some hooks to CVS so that if you remove a file from CVS the RELEASE_READY tag will be moved onto the new (dead) revision. - Updated the CVS commitlog feature. It's now much more readable for the guy who does the releases. =) - Modgen no longer has it's own Version class, as it is now always released simultaniously with JHDL. Base ---- - Propagate manager has now been reverted to handle constants under multiclock. This does affect constants under non-multiclock, but implicit clock support should allow this to work correctly, so there should be no problem. - Multiclock mode does not create lots of prints now. The traces and everything is quiet except for the clock schedules. This affected many files, but should not affect any functionality. - step outside of multiclock does a cycle. This will allow you to jump back and forth between modes, if necessary. - Added ability to mark cells as not netlistable, and added getNetlistableChildren() to Cell. - Added a check to print a warning if a parent has cellInterfaceDeterminesUniqueNetlistStructure() returning true while one of it's children returns false. - Fixed a null pointer resulting from deleting a port. - Fixed improper changing of a cell's netlist status when there is mixed sharing. Logic ----- - Added clockDriver() call to get a techmapper specific clock driver. - Removed relative placement print statements. - Bus-wise xnor, nand, and nor removed from Logic until they are debugged completely. - Added a switch to enable/disable new placement. Improved documentation on all the new placement added. Xilinx ------ - Added a check to the XC4000 Techmapper to only translate children who own their XC4000PlacementInfo object. - Edif now uses the getNetlistableChildren to ignore cells marked not netlistable. - The netlister now appropriately stops at NetlistLeaf cells. - ramd_prop, ram_prop, and ram_synch updated to be consistant with other JHDL code. - Problems with ram16x1 and ram32x1 resolved. - ifd, ofd, and ofdt no longer flagged as netlist leaf cells. - Many library cells updated to extend the proper classes and have properly marked ports and netlist attributes. - Edif is now case-insensitive for uniquifying cell types. This is done because, although JHDL is case sensitive, not all tools are. - Removed INIT param from all cells that extend Memory, as it was causing primitive cells to be uniquified and netlisted incorrectly. - cellInterfaceDeterminesUniqueNetlistStructure() is now correctly overridden in all library cells. - Placement and Floorplanning all now correctly look at isNetlistLeaf() instead of isLeafCell(). jmac ---- - jmac no longer makes any assumptions about leaf cells. - @isLeaf() will no longer work in jmac files. (It will parse, but isLeafCell() is a final method, so the java file won't compile.) - boolgen no longer uses the @isLeaf() jmac token. Jab --- - Various fixes and clean up to BrowserCore and related files. Tbone ----- - Added netlist, and target commands. Removed build command. Tbone will now work without any need for the build command. Modgen ------ - arrayMult and multCol updated to properly do netlist sharing. - Asynchronous loop bug fixed in ramrom generator. RELEASE 0.1.7.8 =============== Release ------- - Implemented the cvs RELEASE_READY tag. All files that are to be included in the next release *MUST* be tagged RELEASE_READY to be included. - make JHDL.jar will now work on the HP machines. - some better handling of linux/HP issues in configure. Base ---- - Fixed the bug in Cell.getUniqueCellName() that caused a null-pointer exception in some cases in release 0.1.7.7. Xilinx ------ - More library cells added, updated, corrected, fixed, etc. Jab --- - terminalControl.c updated to use JNI API. The JNI shared library now compiles under the IBM implementation of JDK 1.1.8. - Updated TextConsole and BrowserCore to properly use non-blocking I/O. - BrowserMainFrame fixed to correctly run certain commands. Tbone ----- - Several minor bug-fixes. Schematic --------- - Fixed schemeatic drawer so bus names will be correctly reported. - inout ports are now drawn differently (<==>). inout ports are placed on the correct side of the cell in the schematic based on the connection of the port. RELEASE 0.1.7.7 =============== General ------- - Build order changed in some Makefiles to speed up compilation. Base ---- - Modified exceptions and toString methods to make error messages more informitive. - addPort() and addPorts() are now both implemented in Cell. - Fixed security problem with isLeafCell(). This function can no longer be overridden and will always been defined correctly. Use isNetlistLeaf() to define a cell with children as a leaf in the netlist. - Corrected port name clash checking to include implicit ports, as well as notify the user of possible conflicts of capitalization. - Instance name uniquification is now case insensitive. jmac ---- - @isleaf() will no longer call isLeafCell(), but isNetlistLeaf() instead. For most cells, this is the desired behavior. However, If you had previously hardcoded a isNetlistLeaf() in your jmac file, it will need to be updated to use @isleaf() correctly. Xilinx ------ - More primitives added, fixed, updated. - make will now correctly handle remaking of jmac-java files when the jmac script itself has been modified. - Several jmac-java files added to the makefile. They were jmac generated files that were not ever being remade when modified. - Fixed a nasty bug that would allow the incorrect overriding of isLeaf, and also would in some cases incorrectly uniquify some Xilinx primitives into unintelligible gibberish. - Behaviour models for ram16x1s, ram16x1d, ram32x1s have been modified to eliminate asynchronous loops that were not consistent with hardware. Asynchronous loops are now not possible for elements that are wired between the data_out and the data_in of the rams. On the other hand, if the data_out is wired to the address lines of the rams there will still be an asynchronous loop. This is the correct behavior. But, if any of these parts are created through the ramrom in modgen, you will not be using their models since ramrom has its own. Jab --- - Added step button to schematic viewer. Modgen ------ - Fixed several cells that called isLeaf(). RELEASE 0.1.7.6 =============== Logic ----- - Generic logic memories modified to correctly implement MemoryInterface. Jab --- - Jab now correctly loads Tbone on classes that do not implement TestBench or extend a class that does. - Disable behavioral modeling now works correctly from Jab. - Memory viewer will now correctly display in hexidecimal when the display hex button is checked. - updateCellInfoCmd modified to properly display child wires. - behavCmd updated to only print to console frame on an error. - TestBench icons fixed to display correctly in the tree view. - Waves viewer added to display saved waves data. Xilinx ------ - ExternallyUpdateable applied to several library cells that needed it. - Several inconsistancies in the handling of MemoryInterface resolved. Wildstar -------- - Added support for downloading bitstreams using Jab. RELEASE 0.1.7.5 =============== Xilinx ------ - BrowserMemory has been removed. All calls to BrowserMemory should use MemoryInterface instead. All cells that previously called BrowserMemory have been updated. - As a result of the change in port optimization in base, BlockRAM clock ports are no longer optimized away. Modgen ------ - All modgens that used BrowserMemory have been updated to use the new MemoryInterface, instead. Base ---- - Cells that return true on isNetlistLeaf() will no longer have their ports optimized away, even if they have child cells. - Code added that allows deletion of cells. Each cell has the method delete() which will remove it, its children, and all wires created by it or its children from the circuit. The deleted cell should be garbage collected, at least once you reinitialize the simulator (by stepping it) to reflect the smaller circuit. This is useful for board models; they can load a "nice" pe for quiet initialization, then when the user supplies their real pe, the board deletes the place-holder pe and adds the users. It is also a step in the direction of dynamic creation and deletion of heirarchy, essential for GUI creation and editing of circuits (instead of just browsing). RELEASE 0.1.7.4 =============== Base ---- - MultiClock mode is now a part of base. No MultiClock parts are available in any libraries, so this is not yet useable. Several key routine that are used for simulation were changed. - Lists now can have multiple enumeration objects, allowing traversal of the same list simultaniously. - added setPortNotNetlistable(...) and isNotNetlistable(...) to Cell to allow ports to exist for simulation purposes, but not for display or netlisting. Xilinx ------ - OFD and other FF components updated to extend XilinxFD instead of Logic. Several other jmac files updated. - Removed some dangling references to ConstantWire in the XC4000 Techmapper General ------- - 99% of all 'Makefile's converted to 'Makefile.in's to work with configure. All 'Makefiles' pruned. If a 'Makefile' is missing from CVS where you didn't have a 'Makefile.in' in, you'll need to check out the old version and re-commit it. RELEASE 0.1.7.3 =============== Xilinx/Logic ------------ - Almost all library cells now override cellInterfaceDeterminesUniqueNetlistStructure() to return true. This should fix the performance problem introduced in 0.1.7.1 - ofdt was fixed so that it will netlist properly. RELEASE 0.1.7.2 =============== Base ---- - Added a couple of debugging hooks - a method to print out lots of statistics, and a method to register a wire to print out lots of diagnostics whenever that wire's value changed. Xilinx - netlisting ------------------- - Fixed a bug causing a NullPointer during netlisting. RELEASE 0.1.7.1 =============== Base ---- - The default behavior is now that cells' PortBlocks and NetList definitions will not be shared. If you want your cell to share these data structures, *and* you can guarantee that they are sharable. That is, you must guarantee that there is a bound parameter for everything that can change the structure of the circuit. *If* you can guarantee this, and you are willing to accept the painful consequences if you are wrong, then you can override the following method to return true: protected boolean cellInterfaceDeterminesUniqueNetlistStructure() The fear invoked by the length of this method name is intentional. As Dr. Hutchings would say: "You should be required to carry a license to call this method". Note, the Xilinx and Logic libraries are not yet overriding this method. This means that there is a memory and performance hit. This will be fixed in the upcoming release 0.1.7.2. - Lots of bug fixes for netlisting code. Xilinx ------ - Lots of bug fixes for netlisting code. Jab --- - BrowserMainFrame has moved to a new package. Hopefully this will not imapct anyone negatively. RELEASE 0.1.7 ============= Base ---- - Netlisting is now deterministic. This means that schematics should be drawn in a deterministic fashion in JAB and that edif netlists should be "diff-able". - Automatic wire/cell renaming due to name clashes is much improved. In the past you may have seen three wires in one cell named: and_out, and_out-3, and_out-12. You will now see those wires as: and_out, and_out-1, and_out-2. (See also the wire name improvements in Logic) - Base now supports the notion of explicit clock wires. The HWsystem has a new method to register a clock wire: declareExternalClock(Wire) as well as a method for specifying a default clock: setDefaultClock(Wire). For single-clock designs the Logic class should take care of these details for most users. You will now just notice a CLK wire running around your schematics. - Implicit interfaces are now available. That is, you can have a CellInterface[] named implicit_interface as well as cell_interface. Most users should not have to be aware of this change. - Using the old portios[], portnames[], and portwidths[] method of declaring a cell will now generate a diagnostic. This warning message is quite nice as it will give you a properly formatted CellInterface declaration that you can paste into your jhdl code. - Unconnected ports are now removed. That is, if a cell has a port and the wire connected to that port does not connect to any leaf descendants of the cell, then that port will not appear in the NetList for that cell. (This is primarily so that the new implicit CLK port will not appear on combinational cells in schematics or netlists.) Logic ----- - Logic wires will no longer have sequential numbers appended on them. This should make wire names much more nice to look at. (See also the wire name improvements in base). - All Logic cells now have an implicit_interface[] with a 1-bit port named CLK. This port will automatically be wired to the default clock wire. (Also a new default clock wire will be constructed and registered with the HWSystem if one does not already exist.) The new code in base will optimize away this CLK port if it is not being used. - Relative placement is finally here! All of the following kinds of calls are now available: (see the documentation on the web page for more details) place(cellA, RIGHT_OF, cellB); place(cellA, LEFT_OF, cellB); place(cellA, ABOVE, cellB); place(cellA, BELOW, cellB); place(cellA, ON, cellB); place(cellA, EAST_OF, cellB); place(cellA, WEST_OF, cellB); place(cellA, NORTH_OF, cellB); place(cellA, SOUTH_OF, cellB); place(cellA, ON_TOP_OF, cellB); You can still use output wires for placement just as before. Also, port placement is also available, (ie. place this cell to the right of that cell so that this port on this cell lines up with that port on that cell). Please consult Matt for more details on this advanced placement. - With the new relative placement, if you use any of the functions of the form place(..., int x, int y, ...), be aware that you will now place with their origin at the upper-left-hand corner of their bounding box and not at the origin that the cell's designer used. - The following methods can change a cell's layout: scale(cell, int x_scale, int y_scale) rotate(cell, int degrees) translate(cell, int dx, int dy) And of course, you can also use output wires instead of cells. Tbone ----- - Tbone's gui is now completely functional. The dialog box will resize itself if it gets too big to be displayed and scroll boxes will appear to enable full access to the parameters. - A new option has been added to compare a behavioral model with a structural model. Both circuits are instatiated and their outputs are compared. Errors in the outputs are reported. - The vector file format has been revised and is a little more understandable. Now each line contains a list of inputs and, after a colon, a list of outputs that would appear after all values have been propagated. - Another new option has been added to create an output file containing the results of the simulator. This is the same output seen in the CLI console sans all of the CLI commands and error feedback from Tbone. - A brief users manual with good examples has been written. - Jab had been updated to support Tbone. Loaded circuits have access to all of the functionality provided by Jab and Tbone. Tbone (earlier changes) ----------------------- - This is a much imporved version of Tbone with many bug fixes and added features. - It is now very easy to load a class directly from the command line. No line in the vector file is needed to specify port widths or parameter values and the gui can be completely avoided if desired. - To implement this a loadclass command, loadvec command have been added. To specify undefined port widths or parameter values a port command and param command have been added to the command line interface. - A new list command has been added which lists the ports of the loaded class. - Vector files contain only data for the inputs and values to compare to the outputs. Jab --- - Two annoying bugs were squashed. The non-reappearing schematic display is now fixed and the "remove the last" wire nullException error is gone. - If a schematic view is open, Jab now raises the window. - Memory viewer now allows contents to be viewed in hex or decimal and displays more than just six rows RELEASE 0.1.6 ============= Base ---- - Implicit interfaces are now available. This should not impact JHDL users. - There is now a new ConstantSource which is scheduled exactly once. This can be used instead of the ConstantWire class. This should not impact JHDL users. - There is now support to bind() any Java Object. There are also methods for binding the native types int and boolean. - Lots of memory/performance optimizations. (VectorLists instead of LinkedLists etc... Logic ----- - A memory leak was removed. (Old circuits were sticking around when you made a new HWSystem and new TechMapper.) Xilinx ------ - The bug with tri-state library elements has been fixed. XC4000 ------ - The bug with 5-bit Logic gates has been fixed. Virtex ------ - The bug with wide gates with (width %4 == 1) has been fixed. - The BlockRam should now netlist without any modifications to BlockRam.isLeafCell(). The multiple-clock port problem still exists so the netlist is not perfect yet. Jab --- - Joe's new ViewManager stuff is online. This was a decoupling of JHDL visualization tools from Jab. This allows users to bring up visualization tools from outside the Jab environment. Use the ViewManager class to get access to visualization tools from outside of Jab. Because of this change, you may need to change import or other package statements. The affected packages/classes are: Old name New name -------- -------- byucc.jhdl.apps.Jab.ViewManager --> byucc.jhdl.apps.Viewers.ViewManager byucc.jhdl.apps.Jab.Schematic --> byucc.jhdl.apps.Viewers.Schematic byucc.jhdl.apps.Jab.FloorPlan --> byucc.jhdl.apps.Viewers.FloorPlan RELEASE 0.1.5 ============= - Initial release of tri-state support in JHDL. The Logic class now has three new methods: tbuf(), pullup(), and pulldown() which provide support for tri-state buffers as well as pull-up and pull-down resistors. These should be usable with both the XC4000 and Virtex TechMappers. Try them out and let us know of any problems. - Versioning added. If you type java byucc.jhdl.Version you will get the JHDL RELEASE number and the date the release was made. There are also 3 static methods that return strings associated with the Version class: GetVersion(), GetFullVersion(), and GetVersionDate() that can be used by other things in JHDL to get this information. Xilinx ------ - All cells that are common to XC4000 and Virtex have been moved up from the XC4000 and Virtex packages into the Xilinx packages. Hopefully, this change should be invisible to users because the Logic class hides it. If however, you were directly instantiating library cells you may have to the following import statement: import byucc.jhdl.Xilinx.*; (But if you are doing this, couldn't you simply use the Logic methods instead?) Virtex ------ - The BlockRam class is finally publicly visible! I haven't yet added a Logic call that gives access to it so just use "new BlockRam()" for now. There is also still an unresolved issue regarding netlisting, so a little edif hack may be necessary for a while. Please contact Carl if you have any questions. - Wide AND/OR gates, (that is the and() and or() calls that accept a single wire in Logic), are now implemented using the carry-chain in Virtex. Several bug fixes were made to this implementation. XC4000 ------ - The wide AND/OR gates, (that is the and() and or() calls that accept a single wire in Logic), are now placed in a single column rather than a tree -- courtesy of Preston. This should make placment easier and more efficient. modgen ------ - multplier_array_pipe is no more. If you use it, use arrayMult instead. arrayMult now will work for Virtex (no Virtex-specific optimization done and place doesn't work yet so it's not quite 100% like the one in XC4000). - Versioning added. Now if you type java byucc.jhdl.modgen.Version it will print out the release number of modgen and the date that it was created. RELEASE 0.1.4 ============= LayoutView ---------- - The LayoutViewer should finally come up now from the TOOLS menu. It's still very BETA. It doesn't currently have a toolbar so you can't zoom. It also will not come up at all if the cell is not completely placed, (at every level), or if there are any placement conflicts. This can all be fixed eventually, but for now it should be somewhat useful. (It will at least save you the M! backend time.) Jab --- - Several bug fixes. Wildforce --------- - Should now come up correctly without the .jabrc hack. RELEASE 0.1.3 ============= Virtex ------ - Fixed wide AND and OR gates. That is, those which are called as and(wire) and or(wire) from within Logic. The TechMapper will use a Virtex-specialized configuration using the carry-chain to perform logic. Jab --- - Added command-line option '-s command_file' which will do an initial source of the file comand_file. - Added -nw switch to run in a non-graphical mode. There is a special native library which must be available if you want TAB-completion and history support in the non-graphical mode. - Added support for a .jabrc file in your home directory, (or the current directory if your OS does not provide a home directory). This is a command file which will be sourced every time you start Jab, (useful for setting up your favorite aliases). - Since aliases are so easy to set up in .jabrc we got rid of the plethora of pre-existing aliases. Note that loadClass has now been changed to load. - Added support for a ViewManager. This allows access to Jab visualization tools outside of the Jab environment. modgen ------ -Added the byucc.jhdl.modgen.fp package with the modules floatMult and FPAdd. documentation ------------- -Generated new documentation that can be seen from the labs internal web page. Note that the release 0.1 documentation can still be seen from the JHDL web page. The new documentation added the byucc.jhdl.modgen.fp to the API section, added a few tidbits here and there and fixed a few other minor errors in the documentation. RELEASE 0.1.2 ============= Virtex ------ - Added Subtractor and adderSubtractor classes so the sub() and addsub() calls should work from Logic with the VirtexTechMapper. - The intial version of the BlockRam is in place. So far it has no call in Logic. Also, there is an issue with netlisting because the BlockRam has two clocks. For now, the netlist will have to be fixed manually. - Fixed regr(), regre(), regs(), and regse() Jab --- - .jab.recent.classes has been re-implemented. It will now save this file in the user's home directory, if there is one, and in the current directory if not. - The LayoutViewer has been re-instated after being yanked out for the public Release 0.1. It's extremely Beta and your feedback is welcome. - TAB completion support is implemented and is better than ever. Hitting TAB twice will list possible completions. (So far this support is only available for select commands). Base ---- ***NOTE: There are still a few issues with the following two changes. So they have been backed off and will not be available for this release. - A new version of the LinkedList class has been included which should use less memory. This should shave a few MB off of big JHDL designs. - Also, a new version of NodeList has been implemented. this together with the change mentioned above take off about 15% of the memory footprint as compared to release 0.1.1. RELEASE 0.1.1 ============= changes to JAB from version 0.1 ------------------------------- - JAB has been changed to work through a class called CLInpterpreter. All commands issued by JAB, either by the GUI or the CONSOLE, will be passed through this interpreter who then issues the commands to the Core program. This means that all GUI commands are now availible via the console to make it easier to script sessions. - A History buffer has been implemented on the console. You can view all previous commands by typing "history" at the command line, or by using up arrows to cycle through the history buffer. This is similar to tcsh functionality. The source command has been modified so that all commands in the file may be included in the history. This is done using "source -e ". If source is used without the -e only the source command will be added to the history and not the this commands in the file. - The console now does limited name completion with the TAB key. The only commands that currently implement this are cd and watch. more commands will have the abitiy to name complete. For cd and watch, the TAB key will completed on the JHDL heirarchy which should make the Console a more viable option for browsing JHDL circuits. - JAB has been converted to the latest version of swing. This means that the program should be able to work with java 1.2, but this has not been tested since java 1.2 has only been released for win32. Virtex ------ - VirtexTechMapper is included for the first time in this release. XC4000 ------ - Fixed a bug with Wide-Gate placement. - Fixed a bug with ram() parameter orders.