float-formats 0.3.0 → 0.4.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
-SHA1:
-  metadata.gz: e5c6ebefc05d57a1271453eb0ba460e7f6fa9e9d
-  data.tar.gz: 0bb606a66d28fa34cdf03984c72acf0ecdbb4ce3
+SHA256:
+  metadata.gz: a6026d216573d6d4bc1bae38fe5495a26424fb09e306a6ee2abc9d111bc40b0a
+  data.tar.gz: 8dd5a8805026744cf00d40c0b0acbbf06d797b2229de77aacdc10b2e2ebb71e7
 SHA512:
-  metadata.gz: dd8490cef4b2aacb6bb06b7fb5e349acf954055499f2bc52000c126a4095163dca2be630d7f27b60b5426e4e43772d1eae6f099423eead7ecf8b257afc59b23e
-  data.tar.gz: 62ce956a3757fb4719e7cf0572ff4698af641275396ba6b5a5f2d95db641e90e35148adc7499dbe0b8bccd157a682585075efab7a561cf708a4c1e9c9b6c3b7e
+  metadata.gz: 72c9aa4fad90095a7f07ca48d8dc96c1ab0a64e80e57a343a69ae50bb6a65b8aa28e53c71e851d7f6ea809fa2b20206b1c3f847a59efe0908c28976a999eda8a
+  data.tar.gz: 0e49f3bb72c6b21d2ec3c402a175464e8044cae267fcd3d2fcbe45f6536cb472970d2cd18e4a40879cd39012f9195f257eff6ade872898d0f082842d4f68c708

data/History.txt

@@ -1,3 +1,9 @@
+== 0.4.0 2020-04-07
+
+* New formats: BFLOAT16, MSFP8, MSFP9, MSFP10, MSFP11, KHRONOS_VULKAN_UNSIGNED10, KHRONOS_VULKAN_UNSIGNED11
+* New aliases: MBF_EXTENDEND, HFP_SINGLE, HFP_DOUBLE, HFP_EXTENDED
+# Development dependencies updated
+
 == 0.3.0 2015-04-05
 
 * Nio dependency replaced by Numerals
@@ -10,7 +16,7 @@
 == 0.2.0 2009-08-06
 
 *  New redefined interface
-*  The sign in splitted form (sign, significand, exponent) now
+*  The sign in split form (sign, significand, exponent) now
    have the sign as an integer +1,-1 rather than using the
    internal representation (0, 1, etc.)
 *  Arithmetic is possible on FP values
@@ -26,7 +32,7 @@
 *  Allow non-bcd values in fields of BCD formats by passing
    hex values as Strings; allow such values to be used for
    nan/infinity exponents.
-*  Nio 0.2.1 is now reqiured
+*  Nio 0.2.1 is now required
 *  Handle special values (Infinities and NaN) in #from_fmt, #from_number
 *  Add ulp methods to Value and FP classes and to Float
 *  Bug fixes

data/README.md

@@ -1,15 +1,18 @@
-#Introduction
+[![Gem Version](https://badge.fury.io/rb/float-formats.svg)](http://badge.fury.io/rb/float-formats)
+[![Build Status](https://travis-ci.org/jgoizueta/float-formats.svg)](https://travis-ci.org/jgoizueta/float-formats)
+
+# Introduction
 
 Float-Formats is a Ruby package with methods to handle diverse floating-point formats.
 These are some of the things that can be done with it:
 
-* Enconding and decoding numerical values in specific floating point representations.
+* Encoding and decoding numerical values in specific floating point representations.
 * Conversion of floating-point data between different formats.
 * Obtaining properties of floating-point formats (ranges, precision, etc.)
 * Exploring and learning about floating point representations.
 * Definition and testing of new floating-point formats.
 
-#Installation
+# Installation
 
 To install the gem manually:
 
@@ -19,11 +22,11 @@ You can find the code in GitHub:
 
 * http://github.com/jgoizueta/float-formats/
 
-#Predefined formats
+# Predefined formats
 
 A number of common formats are defined as constants in the Flt module:
 
-##IEEE 754-2008
+## IEEE 754-2008
 
 **Binary** floating point representations in little endian order:
 
@@ -44,7 +47,7 @@ A number of common formats are defined as constants in the Flt module:
 Others can be defined with IEEE.interchange_binary and IEEE.interchange_decimal
 (see the IEEE module).
 
-##Legacy
+## Legacy
 
 Formats of historical interest, some of which are found
 in file formats still in use.
@@ -69,13 +72,17 @@ Borland Pascal: (BORLAND48).
 Formats used in the Intel 8051 by the C51 compiler:
 (C51_BCD_FLOAT, C51_BCD_DOUBLE and C51_BCD_LONG_DOUBLE).
 
-##Calculators
+**Minifloats**:
+AI Formats: (BFLOAT16, MSFP8, MSFP9, MSFP10, MSFP11)
+Khronos Vulkan unsigned formats: (KHRONOS_VULKAN_UNSIGNED10, KHRONOS_VULKAN_UNSIGNED11)
+
+## Calculators
 
 Formats used in HP RPL calculators: (RPL, RPL_X),
 HP-71B formats (HP71B, HP71B_X)
 and classic HP 10 digit calculators: (HP_CLASSIC).
 
-#Using the pre-defined formats
+# Using the pre-defined formats
 
     require 'rubygems'
     require 'float-formats'
@@ -111,7 +118,7 @@ Minimum and maximum decimal exponents:
     puts IEEE_binary32.decimal_min_exp                   # -> -37
     puts IEEE_binary32.decimal_max_exp                   # -> 38
 
-##Encode and decode numbers
+## Encode and decode numbers
 
 For each floating-point format class there is a constructor method with the same
 name which can build a floating-point value from a variety of parameters:
@@ -119,7 +126,7 @@ name which can build a floating-point value from a variety of parameters:
 * Using three integers:
   the sign (+1 for +, -1 for -), the significand (coefficient or mantissa)
   and the exponent.
-* From a text numeral (with an optional Nio format specifier)
+* From a text numeral (with an optional `Numerals` format specifier)
 * From a number : converts a numerical value
   to a floating point representation.
 
@@ -148,24 +155,24 @@ Examples:
     puts v.to_bits                                   # -> 00111111111110111100110011001100110011001100110011001100110011001100110011001101
     puts v.to_bits_text(16)                          # -> 3ffbcccccccccccccccd
 
-##Special values:  
+## Special values:
 
 Let's show the decimal expression of some interesting values using
 3 significative digits:
 
-    fmt = Nio::Fmt.mode(:gen,3)  
-    puts IEEE_SINGLE.min_value.to_text(fmt)             # -> 2E-45
-    puts IEEE_SINGLE.min_normalized_value.to_text(fmt)  # -> 1.18E-38
-    puts IEEE_SINGLE.max_value.to_text(fmt)             # -> 3.4E38
-    puts IEEE_SINGLE.epsilon.to_text(fmt)               # -> 1.19E-7
+    fmt = Numerals::Format[mode: :general, rounding: [precision: 3]]
+    puts IEEE_SINGLE.min_value.to_text(fmt)             # -> 1e-45
+    puts IEEE_SINGLE.min_normalized_value.to_text(fmt)  # -> 1.18e-38
+    puts IEEE_SINGLE.max_value.to_text(fmt)             # -> 3.40e38
+    puts IEEE_SINGLE.epsilon.to_text(fmt)               # -> 1.19e-7
 
-##Convert between formats
+## Convert between formats
 
     v = IEEE_EXTENDED.from_text('1.1')
     v = v.convert_to(IEEE_SINGLE)
     v = v.convert_to(IEEE_DEC64)
 
-#Tools for the native floating point format
+# Tools for the native floating point format
 
 This is an optional module to perform conversions and manipulate the native Float format.
 
@@ -182,7 +189,7 @@ This is an optional module to perform conversions and manipulate the native Floa
     puts float_significant_dec(Float::MAX_D)           # -> 2.2250738585072009E-308
     puts float_significant_dec(Float::MIN_N)           # -> 2.2250738585072014E-308
 
-Together with flt/sugar (from Flt) can be use to explore or work with Floats:  
+Together with flt/sugar (from Flt) can be use to explore or work with Floats:
 
     require 'flt/sugar'
 
@@ -196,7 +203,7 @@ Together with flt/sugar (from Flt) can be use to explore or work with Floats:
     puts float_significant_dec(Float::MIN_D.next_plus)  # -> 1.0E-323
     puts float_significant_dec(Float::MAX_D.next_minus) # -> 2.2250738585072004E-308
 
-#Defining new formats
+# Defining new formats
 
 New formats are defined using one of the classes defined in float-formats/classes.rb
 and passing the necessary parameters in a hash to the constructor.
@@ -218,17 +225,18 @@ the first bit, which will be hidden:
 Now we can encode values in this format, decode values, convet to other
 formats, query it's range, etc:
 
-     puts MY_FP('0.1').to_bits_text(16)              # -> 1ee66666
-     puts MY_FP.max_value.to_text(Nio::Fmt.prec(3))  # -> 7.88E115
+Numerals::Format[mode: :general, rounding: [precision: 3]]
+     puts MY_FP('0.1').to_bits_text(16)              # -> 1EE66666
+     puts MY_FP.max_value.to_text                    # -> 7.8804e115
 
 You can look at float-formats/formats.rb to see how the built-in formats
 are defined.
 
-#License
+# License
 
 This code is free to use under the terms of the MIT license.
 
-#References
+# References
 
 [*Floating Point Representations.* C.B. Silio.](http://www.ece.umd.edu/class/enpm607.S2000/fltngpt.pdf)
   Description of formats used in UNIVAC 1100, CDC 6600/7600, PDP-11, IEEE754, IBM360/370
@@ -236,13 +244,13 @@ This code is free to use under the terms of the MIT license.
 [*Floating-Point Formats.* John Savard.](http://www.quadibloc.com/comp/cp0201.htm)
   Description of formats used in VAX and PDF-11
 
-###IEEE754 binary formats
+### IEEE754 binary formats
 
 [*IEEE-754 References.* Christopher Vickery.](http://babbage.cs.qc.edu/courses/cs341/IEEE-754references.html)
 
 [*What Every Computer Scientist Should Know About Floating-Point Arithmetic.* David Goldberg.](http://docs.sun.com/source/806-3568/ncg_goldberg.html)
 
-###DPD/IEEE754r decimal formats
+### DPD/IEEE754r decimal formats
 
 [*Decimal Arithmetic Encoding. Strawman 4d.* Mike Cowlishaw.](http://www2.hursley.ibm.com/decimal/decbits.pdf)
 
@@ -252,9 +260,9 @@ This code is free to use under the terms of the MIT license.
 
 [*DRAFT Standard for Floating-Point Arithmetic P754.* IEEE.](http://www.validlab.com/754R/drafts/archive/2007-10-05.pdf)
 
-###HP 10 digits calculators
+### HP 10 digits calculators
 
-[*HP CPU and Programming*. David G.Hicks.](http://www.hpmuseum.org/techcpu.htm)  
+[*HP CPU and Programming*. David G.Hicks.](http://www.hpmuseum.org/techcpu.htm)
   Description of calculator CPUs from the Museum of HP Calculators.
 
 [*HP 35 ROM step by step.* Jacques Laporte](http://www.jacques-laporte.org/HP35%20ROM.htm)
@@ -263,7 +271,7 @@ This code is free to use under the terms of the MIT license.
 *Scientific Pocket Calculator Extends Range of Built-In Functions.* Eric A. Evett, Paul J. McClellan, Joseph P. Tanzini.
   Hewlett Packard Journal 1983-05 pgs 27-28. Describes format used in HP-15C.
 
-###HP 12 digits calculators
+### HP 12 digits calculators
 
 *Software Internal Design Specification Volume I For the HP-71*. Hewlett Packard.
   Available from http://www.hpmuseum.org/cd/cddesc.htm
@@ -272,30 +280,30 @@ This code is free to use under the terms of the MIT license.
   Excerpted from *RPL: A Mathematical Control Language*. by W. C. Wickes.
   Available at http://www.hpcalc.org/details.php?id=1743
 
-###HP-3000
+### HP-3000
 
 *A Pocket Calculator for Computer Science Professionals.* Eric A. Evett.
   Hewlett Packard Journal 1983-05 pg 37. Describes format used in HP-3000
 
-###IBM
+### IBM
 
 [*IBM Floating Point Architecture.* Wikipedia.](http://en.wikipedia.org/wiki/IBM_Floating_Point_Architecture)
 
 [*The IBM eServer z990 floating-point unit*. G. Gerwig, H. Wetter, E. M. Schwarz, J. Haess, C. A. Krygowski, B. M. Fleischer and M. Kroener.](http://www.research.ibm.com/journal/rd/483/gerwig.html)
 
-###MBF  
+### MBF
 
 [*Microsoft Knowledbase Article 35826*](http://support.microsoft.com/?scid=kb%3Ben-us%3B35826&x=17&y=12)
 
 [*Microsoft MBF2IEEE library*](http://download.microsoft.com/download/vb30/install/1/win98/en-us/mbf2ieee.exe)
 
-###Borland
+### Borland
 
 *An Overview of Floating Point Numbers.* Borland Developer Support Staff
 
 [*Pascal Floating-Point Page.* J R Stockton.](http://www.merlyn.demon.co.uk/pas-real.htm)
 
-###8-bit micros
+### 8-bit micros
 
 This is the MS Basic format (BASIC09 for TRS-80 Color Computer, Dragon),
 also used in the Sinclair Spectrum.
@@ -306,26 +314,26 @@ also used in the Sinclair Spectrum.
 *Sinclair ZX Spectrum / Basic Programming.*. Steven Vickers.
   Chapter 24. http://www.worldofspectrum.org/ZXBasicManual/zxmanchap24.html
 
-###Apple II
+### Apple II
 
 *Floating Point Routines for the 6502* Roy Rankin and Steve Wozniak.
   Dr. Dobb's Journal, August 1976, pages 17-19.
 
-###C51
+### C51
 
 [*Advanced Development System* Franklin Software, Inc.](http://www.fsinc.com/reference/html/com9anm.htm)
 
-###CDC6600
+### CDC6600
 
 *CONTROL DATA 6400/6500/6600 COMPUTER SYSTEMS Reference Manual*
   Manuals available at http://bitsavers.org/
 
-###Cray
+### Cray
 
 *CRAY-1 COMPUTER SYSTEM Hardware Reference Manual*
   See pg 3-20 from 2240004 or pg 4-30 from HR-0808 or pg 4-21 from HP-0032.
   Manuals available at http://bitsavers.org/
 
-###Wang 2200
+### Wang 2200
 
 [*Internal Floating Point Representation*](http://www.wang2200.org/fp_format.html)

data/Rakefile

@@ -17,3 +17,5 @@ Rake::RDocTask.new do |rdoc|
   rdoc.rdoc_files.include('README*')
   rdoc.rdoc_files.include('lib/**/*.rb')
 end
+
+task :default => :test

data/float-formats.gemspec

@@ -18,12 +18,11 @@ Gem::Specification.new do |spec|
   spec.test_files    = spec.files.grep(%r{^(test|spec|features)/})
   spec.require_paths = ["lib"]
 
-  spec.add_dependency 'flt', ">= 1.4.7"
-  spec.add_dependency 'numerals', ">= 0.3.0"
+  spec.add_dependency 'flt', "~> 1.5"
+  spec.add_dependency 'numerals', "~> 0.3"
 
-  spec.add_development_dependency "bundler", "~> 1.6"
-  spec.add_development_dependency "rake"
-  spec.add_development_dependency 'nio', ">= 0.2.4"
+  spec.add_development_dependency "bundler", ">= 1.9"
+  spec.add_development_dependency "rake", ">= 12.3.3"
 
   spec.required_ruby_version = '>= 1.9.3'
 end

data/lib/float-formats/classes.rb

@@ -477,6 +477,10 @@ class FormatBase
     # attr_accessor
   end
 
+  def self.normalized?
+    @normalized
+  end
+
   # compute a power of the radix (base)
   def self.radix_power(n)
     radix**n
@@ -579,7 +583,8 @@ class FormatBase
       precision: significand_digits,
       emin: radix_min_exp(:scientific_significand),
       emax: radix_max_exp(:scientific_significand),
-      rounding:@round || :half_even
+      rounding: @round || :half_even,
+      normalized: @normalized
     )
   end
 
@@ -966,7 +971,7 @@ class FormatBase
 
     def read(numeral, exact_input, approximate_simplified)
       num = num_conversions.read(numeral, exact_input, approximate_simplified)
-      num = num.normalize(@form_class.context) if exact_input
+      num = num.normalize(@form_class.context) if exact_input && @form_class.normalized?
       @form_class.num(num)
     end
 

data/lib/float-formats/formats.rb

@@ -95,7 +95,6 @@ IEEE.interchange_binary :IEEE_binary256_BE, 256, endianness: :big_endian
 IEEE.interchange_binary :IEEE_binary512_BE, 512, endianness: :big_endian
 IEEE.interchange_binary :IEEE_binary1024_BE, 1024, endianness: :big_endian
 
-
 # old names
 IEEE_binaryx = IEEE_binary80
 IEEE_HALF = IEEE_binary16
@@ -130,6 +129,47 @@ IEEE_DEC32 = IEEE_decimal32
 IEEE_DEC64 = IEEE_decimal64
 IEEE_DEC128 = IEEE_decimal128
 
+# Brain Floating Point (bfloat16) - Google Brain 16-bit truncated version of IEEE_binary32
+# Google Cloud TPUs, TensorFlow
+# Intel AI processors: Nervana NNP-L1000, Xeon processors (AVX-512 BF16 extensions)
+# Intel FPGAs
+# ARMv8.6-A
+# AMD ROCm libraries
+
+IEEE.binary :BFLOAT16, significand: 7, exponent: 8, endianness: :little_endian
+IEEE.binary :BFLOAT16_BE, significand: 7, exponent: 8, endianness: :big_endian
+
+# MSFP8-11 Microsoft Floating Point (Project Brainwave)
+
+IEEE.binary :MSFP8, significand: 2, exponent: 5, endianness: :little_endian
+IEEE.binary :MSFP9, significand: 3, exponent: 5, endianness: :little_endian
+IEEE.binary :MSFP10, significand: 4, exponent: 5, endianness: :little_endian
+IEEE.binary :MSFP11, significand: 5, exponent: 5, endianness: :little_endian
+IEEE.binary :MSFP8_BE, significand: 2, exponent: 5, endianness: :big_endian
+IEEE.binary :MSFP9_BE, significand: 3, exponent: 5, endianness: :big_endian
+IEEE.binary :MSFP10_BE, significand: 4, exponent: 5, endianness: :big_endian
+IEEE.binary :MSFP11_BE, significand: 5, exponent: 5, endianness: :big_endian
+
+# Khronos Vulkan unsigned minifloat formats
+
+Flt.define BinaryFormat, :KHRONOS_VULKAN_UNSIGNED11,
+  fields:[:significand, 6, :exponent, 5], # :sign 1
+  bias: 2**(5-1)-1, bias_mode: :scientific_significand,
+  hidden_bit: true,
+  endianness: :little_endian,
+  round: :half_even,
+  gradual_underflow: true,
+  infinity: true, nan: true
+
+Flt.define BinaryFormat, :KHRONOS_VULKAN_UNSIGNED10,
+  fields:[:significand, 5, :exponent, 5], # :sign 1
+  bias: 2**(5-1)-1, bias_mode: :scientific_significand,
+  hidden_bit: true,
+  endianness: :little_endian,
+  round: :half_even,
+  gradual_underflow: true,
+  infinity: true, nan: true
+
 # Excess 128 used by Microsoft Basic in 8-bit micros, Spectrum, ...
 
 Flt.define BinaryFormat, :XS128,
@@ -137,7 +177,6 @@ Flt.define BinaryFormat, :XS128,
   bias: 128, bias_mode: :fractional_significand,
   hidden_bit: true,
   endianness: :big_endian, round: :half_up,
-  endianness: :big_endian,
   gradual_underflow: false, infinity: false, nan: false
 
 # HP-3000 excess 256 format, HP-Tandem...
@@ -181,6 +220,8 @@ Flt.define :MBF_DOUBLE, BinaryFormat,
   endianness: :little_endian,
   gradual_underflow: false, infinity: false, nan: false
 
+MBF_EXTENDEND = XS128
+
 # DEC formats (VAX)
 
 Flt.define :VAX_F, BinaryFormat,
@@ -313,6 +354,10 @@ Flt.define :IBMX, HexadecimalFormat,
   bias: 8192, bias_mode: :fractional_significand,
   endianness: :big_endian
 
+HFP_SINGLE = IBM32
+HFP_DOUBLE = IBM64
+HFP_EXTENDED = IBM128
+
 # Cray-1
 Flt.define :CRAY, BinaryFormat,
   fields:[:significand,48,:exponent,15,:sign,1],

data/lib/float-formats/version.rb

@@ -1,5 +1,5 @@
 module Flt
   module Frmts
-    VERSION = '0.3.0'
+    VERSION = '0.4.0'
   end
 end

metadata

@@ -1,85 +1,71 @@
 --- !ruby/object:Gem::Specification
 name: float-formats
 version: !ruby/object:Gem::Version
-  version: 0.3.0
+  version: 0.4.0
 platform: ruby
 authors:
 - Javier Goizueta
 autorequire: 
 bindir: bin
 cert_chain: []
-date: 2015-04-05 00:00:00.000000000 Z
+date: 2020-04-07 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: flt
   requirement: !ruby/object:Gem::Requirement
     requirements:
-    - - ">="
+    - - "~>"
       - !ruby/object:Gem::Version
-        version: 1.4.7
+        version: '1.5'
   type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
-    - - ">="
+    - - "~>"
       - !ruby/object:Gem::Version
-        version: 1.4.7
+        version: '1.5'
 - !ruby/object:Gem::Dependency
   name: numerals
-  requirement: !ruby/object:Gem::Requirement
-    requirements:
-    - - ">="
-      - !ruby/object:Gem::Version
-        version: 0.3.0
-  type: :runtime
-  prerelease: false
-  version_requirements: !ruby/object:Gem::Requirement
-    requirements:
-    - - ">="
-      - !ruby/object:Gem::Version
-        version: 0.3.0
-- !ruby/object:Gem::Dependency
-  name: bundler
   requirement: !ruby/object:Gem::Requirement
     requirements:
     - - "~>"
       - !ruby/object:Gem::Version
-        version: '1.6'
-  type: :development
+        version: '0.3'
+  type: :runtime
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - "~>"
       - !ruby/object:Gem::Version
-        version: '1.6'
+        version: '0.3'
 - !ruby/object:Gem::Dependency
-  name: rake
+  name: bundler
   requirement: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: '0'
+        version: '1.9'
   type: :development
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: '0'
+        version: '1.9'
 - !ruby/object:Gem::Dependency
-  name: nio
+  name: rake
   requirement: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.2.4
+        version: 12.3.3
   type: :development
   prerelease: false
   version_requirements: !ruby/object:Gem::Requirement
     requirements:
     - - ">="
       - !ruby/object:Gem::Version
-        version: 0.2.4
+        version: 12.3.3
 description: Floating-Point Formats
 email:
 - jgoizueta@gmail.com
@@ -130,8 +116,7 @@ required_rubygems_version: !ruby/object:Gem::Requirement
     - !ruby/object:Gem::Version
       version: '0'
 requirements: []
-rubyforge_project: 
-rubygems_version: 2.2.2
+rubygems_version: 3.1.2
 signing_key: 
 specification_version: 4
 summary: Floating-Point Formats