crackNum

• Hackage: http://hackage.haskell.org/package/crackNum

• GitHub: http://github.com/LeventErkok/crackNum/

• Latest Hackage released version: 4.2, 2026-08-25

Version 4.2, 2026-08-25

• Documentation only; no functional changes.

• Simplified the README’s download section further. It is now just the three platforms, each linking to the latest release. The per-platform notes (static linking, the macOS quarantine flag, the Windows SmartScreen prompt) and the unpack/PATH instructions are gone from here – every bundle ships a README.txt that covers them for that platform, and it is the first thing you see when you unpack.

• Trimmed the note about needing z3 when installing from Hackage.

Version 4.1, 2026-08-25

• Documentation only; no functional changes.

• Substantially shortened the README. It had grown to fifteen worked examples, all showing the same output fields with different numbers in them, plus a copy of the per-platform install commands that already ship inside every release bundle. The examples are down to four – an encode with a non-default rounding mode, a decode, an E8M0 encode, and a two-lane decode – and the install section now points at the README.txt in the bundle rather than repeating it, so those steps have a single home and cannot drift.

• The “Supported formats” section is now just the flag table. The per-format prose that followed it – the FP4 (E0M3) and E8M0 explanations, the integer flavors, the TF32 bit-count note – is gone; the first three were already stated in the --help output reproduced further down.

• Downloading a prebuilt binary is now the first thing the README talks about, under its own heading, instead of being a sub-section whose point was hidden in a parenthetical. Building from Hackage is presented as the alternative, and each platform bullet links to the latest release.

• Replaced the README’s markdown tables with lists. Hackage renders the README outside the #description pane, and its stylesheet only borders tables inside that pane, so a table there comes out unstyled and the columns run together.

• Dropped the per-platform GUI build instructions from the README; each GUI’s own directory documents itself. GUI/winGUI/README.md is new; GUI/swiftGUI/ and GUI/tclGUI/ already had one. The note about needing wish on Linux went with it – when wish is missing, the binary already prints the apt/dnf/nix command to install it.

• Reworded the cabal Description’s pointer to the releases page: it now leads with not needing a Haskell toolchain, rather than with the word “releases”.

Version 4.0, 2026-08-24

• The Windows GUI now carries the CrackNum icon instead of the stock WinForms one. The artwork is rendered by the same generator that produces the macOS AppIcon.icns, so the two platforms cannot drift apart.

• Note on numbering: 3.32 was tagged and published as a GitHub release, but never uploaded to Hackage. So for anyone installing from Hackage, this is the first version to support Windows – see the 3.32 entry below for what that involves.

Version 3.32, 2026-08-24

• Windows is now a supported platform, with a prebuilt bundle (crackNum-<version>-windows-x86_64.zip) alongside the Mac and Linux ones. It carries the CLI, a native GUI, and a copy of z3, and needs nothing installed: the GUI targets .NET Framework 4.8, which ships with Windows 10 and 11, and z3 brings its own Microsoft C runtime. The binaries are not code-signed, so SmartScreen warns on first run; the bundled README says how to get past it.

• The Windows GUI (GUI/winGUI/) is a native WinForms application rather than a port of the Tcl/Tk script that serves Linux. Reusing the Tcl would have been a far smaller change, but it would have required users to install a Tcl distribution before --gui did anything, and a GUI behind a prerequisite that most people will not clear is worth less than the code it saves.

• --version and the usage text no longer print the executable’s .exe extension on Windows, where getProgName keeps it.

• The golden test suite writes its temporary files with newline translation turned off, so the golds stay LF everywhere. Previously, running the suite on Windows compared CRLF output against LF golds, and --accept there would have rewritten every gold in the process.

Version 3.31, 2026-08-22

• The Hackage description now shows a screenshot of the GUI, and points at the releases page for the prebuilt Mac and Linux bundles. The screenshot was dropped from README.md in the same breath: Hackage renders the README below the description, so keeping it in both would have shown the same image twice on one page.

• The READMEs shipped inside the binary bundles now walk through a single install sequence – unpack, place the files, set PATH, check – instead of splitting the CLI and the GUI into separately numbered steps that had to cross-reference each other. The macOS one leads with the quarantine removal, since nothing runs before that.

Version 3.30, 2026-08-21

• The GUIs now group the floating-point formats by provenance instead of listing all twelve of them in one flat “Float” section. The formats that exist because of machine learning – FP4 (E2M1), FP4 (E0M3), FP8 (E4M3), FP8 (E5M2), FP8 (E8M0), Brain, and TF32 – come first under “AI formats”, followed by the IEEE-754 ones (Half, Single, Double, Quad, and Custom) under “IEEE-754”. Integers now precede words, so the sidebar reads AI formats, IEEE-754, Integer (Signed), Word (Unsigned). Both the macOS (Swift) and the Tcl/Tk GUI are grouped identically. No format was added, removed, or renamed, and the command line is unaffected.

• The “Custom parameters” box in the GUIs is now titled “Custom IEEE-754 float:”, and its heading lines up flush left with “Rounding mode” above it rather than being indented past it. The “(exponent width applies to custom floats)” note is gone, the new title having made it redundant. Note that the “Total width” field also applies to the Custom entries under Integer (Signed) and Word (Unsigned), which take a width but no exponent.

• The format list in the Tcl/Tk GUI now has a vertical scrollbar. With four sections it is taller than the sidebar at the default window size, and without a scrollbar the formats past the bottom were not merely off-screen but unreachable. (The macOS GUI needed no equivalent change; its list already scrolled.)

• Scrollbars in the Tcl/Tk GUI now appear only when there is something to scroll to, rather than always taking up room. This covers the output pane’s pair as well as the new one on the format list.

• Internally, Main.hs has been split into per-topic modules – CrackNum.Types, .Formats, .Options, .Utils, .Output, .GUI, .Decode, and .Encode – leaving Main with just the argument dispatch. It had grown to some 1400 lines holding everything from option parsing to the hand-rolled layouts for the formats that have no IEEE look-alike. This is purely a reorganization: every definition moved verbatim, so there is no change in behavior or output.

Version 3.29, 2026-08-21

• crackNum.vim’s TAB completion no longer offers a hardcoded set of -lN lane counts, since -lN is never valid on its own and always needs pairing with a -f/-i/-w flag; type it directly instead. -i/-w completion now offers the bare flag and prompts for the bit-width separately, rather than baking in a fixed set of widths – crackNum accepts any -iN/-wN, so there was never a complete list to offer.

Version 3.28, 2026-08-21

• Add support for the E8M0 format, via -fe8m0. This is the shared scale of the OCP Microscaling (MX) formats: the value that scales a block of MXFP8/MXFP6/MXFP4 elements, all of which crackNum already knew about. It is the mirror image of FP4 (E0M3): where that format is all significand and no exponent, E8M0 is all exponent and no significand, with no sign bit either. Every value it holds is therefore a power of two, from 2^-127 to 2^127.

  Note that it has no zero and no subnormals – with nothing for the all-zero encoding to mean, it simply denotes 2^-127 – and no infinities. 0xFF is its one and only NaN. Negative inputs are rejected rather than saturated: with no sign bit there is no direction to saturate towards, and clamping would quietly turn a negative into a positive. Values outside the range, infinity included, saturate to the nearest end-point, and anything that is not a power of two rounds according to -r.

Version 3.27, 2026-08-20

• There is now a Linux binary distribution, alongside the macOS one: see the Releases page. It ships the crackNum executable, a copy of z3 (which crackNum shells out to for every operation), the Tcl/Tk GUI script, and a README. Both binaries are statically linked, so there is no libc or distribution requirement: they run as-is on any x86_64 Linux, old or new. The GUI additionally needs wish, which does have to come from your system.

• --gui now also looks for crackNum.tcl next to the crackNum executable itself, after $CRACKNUM_TCL and the PATH but before the cabal data-directory. A binary distribution carries the script alongside the binary, where the data-directory baked in at build time names a path from the build machine that does not exist on the user’s; the GUI now works in such a bundle however it is unpacked or copied, rather than only when the script was also placed on the PATH.

Version 3.26, 2026-08-19

• New flag --list-formats, which prints the floating-point formats -f accepts, one name per line. It is meant for editor integrations, which would otherwise have to hardcode the list and let it go stale; the VIM plugin now asks rather than guesses, and picks up any format added later for free.

• --help now also lists the supported floating-point formats. Both it and the -f error message are generated from a single table, so they cannot disagree.

• Fix the VIM integration, which had been broken since version 3.0. The plugin offered i, w, and f as the choices at its precision prompt, and passed whichever you picked through as-is, so :CrackNum ran crackNum i 0b0110: no leading dash, and no bit-width. Every invocation was rejected, and the quickfix window filled up with crackNum’s usage text instead of an answer. The completion list now offers real flags (-i8, -fhp, …), and anything else crackNum accepts, such as -f3+4 or -l4 -fhp, can be typed in directly.

• The VIM plugin also set grepformat to VIM %m, matching a --vim output mode that was removed in 3.0. Nothing matched that format, so even a correct invocation produced an empty quickfix window. Take crackNum’s output as-is now.

• The VIM plugin now finds the bit-pattern under the cursor itself, instead of relying on <cword>. Verilog notation stops at the quote (64'hdeadbeef came out as just 64), and the value is quoted before it reaches the shell.

Version 3.25, 2026-08-18

• Add support for the FP4 (E0M3) format, via -ffp4e0m3. Unlike every other format crackNum knows about, this one has no exponent bits at all, which makes it a plain 4-bit sign-magnitude integer: a sign bit and a 3-bit magnitude, covering -7 to 7, with both a positive and a negative zero. It has neither NaN nor Inf, so those inputs are rejected; values outside [-7, 7] saturate to the nearest end-point, and fractional inputs round according to -r.

• Report the format the user actually asked for when decoding E4M3 and FP4. Both are modeled by an IEEE look-alike, and the patterns that need no special handling were printed straight from it, so crackNum -ffp4 0b0100 answered 2.0 :: FloatingPoint 2 2 and crackNum -fe4m3 0b00111000 answered 1.0 :: FloatingPoint 4 4. The values were right, only the type name leaked. The patterns that do deviate, and E5M2 throughout, were already correct.

Version 3.24, 2026-08-17

• Add a quad-precision example to the help output. -fqp has always been accepted, but --help never mentioned it, so the only way to find out it existed was to trip over the error message for a bad -f argument.

• Bring the README up to date: document the installation steps (including the z3 requirement), add a table of all supported formats, list the rounding modes, and add worked examples for the FP8, FP4, TF32, and unsigned-word formats. A couple of the existing sample outputs had drifted from what the tool actually prints, and are now regenerated.

Version 3.23, 2026-08-17

• Do not ignore bad flags when --gui is given. A mistyped format, such as crackNum -ft32 4 --gui, used to bring the GUI up with nothing selected, silently swallowing the error the command line would have reported. We now diagnose the flag first, and only launch the GUI if everything checks out.

• Add quad-precision (-fqp) to the format list in both GUIs. It was accepted on the command line, but was missing from the interfaces.

Version 3.22, 2026-08-17

• Fix text alignment in the Tcl/Tk GUI’s entry fields. We asked for the Courier font, which on X11 is an alias that typically resolves to Nimbus Mono PS. Its ascent/descent split is lopsided (9/6 at size 11), and since an entry centers text on the linespace, the glyphs ended up hugging the top of the box with a large gap underneath. We now pick the first monospaced family that is actually installed, preferring ones with sane metrics.

Version 3.21, 2026-08-16

• Add support for TF32 (TensorFloat-32), via -ftf32. This is the 19-bit format with 8 exponent and 10 significand bits, i.e., the exponent range of single precision with the significand of half precision. Note that we crack the 19 architectural bits; hardware typically carries these in a 32-bit container with the remaining bits unused.

Version 3.20, 2026-08-13

• Fix tool lookup in both GUIs: a directory named crackNum (or z3) sitting on the PATH was accepted as the executable, since directories carry the search bit and so look executable. The GUI would then fail with a “permission denied” on that directory. We now require a regular file.

Version 3.19, 2026-08-13

• Fix decoding in the Tcl/Tk GUI: hex and binary input was silently converted to decimal before being handed to crackNum, so entering 0xdeadbeef encoded the value 3735928559 instead of decoding the bit-pattern. Verilog (N'h) input was unaffected.

• Drop the lane count from both GUIs. The number of lanes is inferred from Verilog (N'h) input, and everything else is a single lane, so there was nothing useful for the interface to set.

Version 3.18, 2026-08-13

• Add a Tcl/Tk GUI (GUI/tclGUI/crackNum.tcl) that works on Linux and macOS. When --gui is used on Linux, crackNum now launches this interface (requires wish) instead of erroring out.

• The Tcl/Tk GUI script is now a cabal data-file, so it is installed along with the binary: cabal install crackNum is enough for crackNum --gui to work on Linux, with no PATH setup. To run a different copy of the script, set CRACKNUM_TCL, or put it on your PATH as crackNum.tcl.

• Both front-ends now live under GUI/: the macOS app moved from gui/ to GUI/swiftGUI/, and the Tcl/Tk script to GUI/tclGUI/.

• Encoding a NaN now always displays the canonical quiet-NaN pattern (sign 0, all-ones exponent, leading significand bit set; 0x7FC00000 for a single). SMTLib’s floating-point sort has a single NaN value, so the solver returns an abstract NaN and the concrete bit-pattern shown was whatever the model materialized – which could differ between solver/library versions. The E4M3 path already pinned its NaN this way; the rest now do too.

• Exponent/significand sizes of 1 bit are accepted by -f, but the solver requires at least 2 of each. This now produces a regular error message instead of an uncaught exception with a backtrace.

Version 3.17, 2026-08-10

• Add support for the FP4 (E2M1) format, via -ffp4. Like E4M3, this format deviates from IEEE-754: The all-ones exponent encodes the values 4 and 6, instead of infinity and NaN. Consequently, FP4 can represent neither NaN nor infinity, and finite values outside of [-6, 6] saturate to the end-points.

Version 3.16, 2026-07-24

• Add the --gui option, launching a graphical interface (macOS) for interactively encoding/decoding values.

Version 3.15, 2024-11-09

• Bump up SBV dependence to >= 11.0

Version 3.14, 2024-09-23

• Fix README

Version 3.13, 2024-09-23

• Fix help text

Version 3.12, 2024-04-05

• Fix hexadecimal float parsing for e4m3

Version 3.11, 2024-04-05

• Allow for encoding of hexadecimal floats

Version 3.10, 2024-03-01

• More relaxed parsing for verilog input format

Version 3.9, 2024-02-23

• Fix verilog input format parsing

Version 3.8, 2024-02-21

• Add support for FP8 formats, as decribed in: https://arxiv.org/pdf/2209.05433.pdf

  • E5M2: Which is essentially a synonym for f5+3
  • E4M3: Similar to f4+4, except it does not have infinities and interprets NaN values differently

• Fix a bug in cracking of arbitrary-sized floats, that yielded wrong values for some NaN cases

Version 3.7, 2024-02-15

• Support signaling/quiet indication for decoded NaN values.

• Add support for decoding over multiple lanes. See the -l option.

• Add support for verilog bit-vector notation, e.g., 128’hXXX. If you use this notation, crackNum will automatically infer the number of lanes to crack based on the width given; unless explicitly specified.

Version 3.6, 2024-01-24

• Be more clear when the provided input isn’t a recognizable float, instead of treating it as NaN implicitly. Thanks to Dmitry Blotsky for pointing out the confusion.

Version 3.5, 2024-01-11

• Resolve compilation issues with GHC 9.8 series

Version 3.4, 2023-04-14

• Fix compilation in previous build

Version 3.3, 2023-04-14

• Allow compilation with newer versions of SBV

Version 3.2, 2021-06-30

• Add an explicit note when conversion is exact.

Version 3.1, 2021-03-29

• Fix readme

Version 3.0, 2021-03-29

• A complete rewrite, much simplified, and supporting arbitrary precision floats. Some of the old features and the library are dropped; so if you rely on the library nature of CrackNum, do not upgrade. For other users who merely use crackNum as an executable, the new version is strongly recommended.

Version 2.4, 2020-09-05

• Changes required to compile cleanly with GHC 8.10.2

Version 2.3, 2018-11-17

• Remove dependency on the ieee754 and reinterpret-cast packages. The goal is to remove any FFI dependencies. We now define and export the required utilities directly in the CrackNum package.

Version 2.2, 2018-09-01

• Instead of data-binary-ieee754, use reinterpret-cast package. According to documents, the former is deprecated.

Version 2.1, 2018-07-20

• Support for vi-editor bindings. See the file “crackNum.vim” in the distribution or in the github repo You can put “so ~/.vim/crackNum.vim” (use the correct path!) and have vi crack numbers directly from inside your editor. Simply locate your cursor on a binary/hex stream of digits and type “:CrackNum”. See the “crackNum.vim” file for binding details.

Version 2.0, 2018-03-17

• Import FloatingHex qualified to avoid GHC 8.4.1 compilation issue

Version 1.9, 2017-01-22

• Minor fix to printing of +/-0

Version 1.8, 2017-01-15

• Bump up FloatingHex dependency to >0.4, this enables proper support for large doubles

Version 1.7, 2017-01-14

• Fix a snafu in reading hexadecimal floats

Version 1.6, 2017-01-14

• Add support for hexadecimal-floats. These now work both in toIEEE option as input, and also when printing the values out. (i.e., numbers of the form 0x1.abp-3, etc.)

Version 1.5, 2016-01-23

• Typo fixes; no functionality changes

Version 1.4, 2016-01-17

• Fix NaN nomenclature: Screaming->Signaling
• Add an example to README.md

Version 1.3, 2015-04-11

• Fix docs, github location

Version 1.2, 2015-04-11

• Fix the constant qnan values for SP/DP
• Add conversions from float/double. Much easier to use.
• Better handling of nan values.

Version 1.1, 2015-04-02

• Clean-up the API, examples etc.

Version 1.0, 2015-04-01

• First implementation. Supports HP/SP/DP and signed/unsigned numbers in 8/16/32/64 bits.