image_compressor_pack 0.1.3 → 1.0.0.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA1:
-  metadata.gz: 41d51c703869f6e07e06d5511cb0b727ee7f6f0b
-  data.tar.gz: e04c08e3d305e66ebd9e972a35860a6311d5b61d
+  metadata.gz: 1d606ee12c458166ec47cd3361c4a7471794bba5
+  data.tar.gz: dc04a1f472898c69bf4948019edd4eaf631e4902
 SHA512:
-  metadata.gz: 8da03caf5543628f0064df3093aa3eb32eae36645a1be0437931f81901bff46254c913cab5fef2406f640e339382be3b4c52bcc98ea7152b1b83a882d129c8ed
-  data.tar.gz: 5b8cc83fd4cd80b9aad33d89cb1e3aa35da053edccbf36ffe99bce26c70c6b3832ce14e92c0d49bbc13bcbc0979d204fd059fb85fcfeaf0876aa7ea82cbd80f6
+  metadata.gz: 5ebb60121f6caa9ae3da7d1abd1fdea2b55272cd53fa8539df39635ce1055ccb940d28ed7f5490e73a35867516304d7269c1c7330cf2bf73659306d9f8f6a617
+  data.tar.gz: 5484d315f7b21f52c05283a4e0454f5d77b4c227d0532fb510772a80a0fdfc4a489680b061db464962cc857a3f9b101a8d4bf8a7058127f4d340cb878d512a37

data/.travis.yml

@@ -6,17 +6,20 @@ matrix:
     - rvm: 2.2
       os: osx
       osx_image: xcode7.3
+    - rvm: 2.3.2
+      os: osx
+      osx_image: xcode8.1
     - rvm: jruby-9.0.5.0
       jdk: oraclejdk8
-    - rvm: 2.3.1
+    - rvm: 2.3.2
       dist: trusty
       sudo: true
-    - rvm: 2.2
-      os: osx
-      osx_image: xcode6.4
     - rvm: system
       os: osx
       osx_image: xcode7.3
+    - rvm: system
+      os: osx
+      osx_image: xcode8.1
     - rvm: 1.9.3
   allow_failures:
     - rvm: 1.9.3

data/CHANGELOG.md

@@ -0,0 +1,4 @@
+### v1.0.0.0, v1.0.0.1 - 19.11.2016
+
+  * Updated nasm, lcms2, libpng, pngquant, advancecomp, pngcrush, jpegoptim.
+  * Added the dist files back to the repo.

data/README.md

@@ -1,4 +1,6 @@
 # Image Compressor Pack
+[![Gem Version](https://badge.fury.io/rb/image_compressor_pack.svg)](https://badge.fury.io/rb/image_compressor_pack)
+[![Build Status](https://travis-ci.org/ignisf/image_compressor_pack.svg?branch=master)](https://travis-ci.org/ignisf/image_compressor_pack)
 
 A source distribution of a bunch of lossy and lossless image optimisation
 utilities for use with `image_optim`. Created because I didn't trust the
@@ -7,16 +9,37 @@ compilation from source.
 
 ## Installation
 
-Add this line to your application's Gemfile:
+The gem is distributed in both source and binary form. The binary version
+contain statically linked executables for Linux and FreeBSD and dynamically
+linked ones for Mac OS X.
+
+### Using the source version
+
+To compile from source, add this line to your application's Gemfile:
 
 ```ruby
-gem 'image_compressor_pack', github: 'ignisf/image_compressor_pack'
+gem 'image_compressor_pack'
 ```
 
 And then execute:
 
     $ bundle
 
+### Versioning
+
+Starting with `image_compressor_pack` `1.0.0.0`, all even point releases contain
+only a source-based distribution, while odd point releases contain both a
+source-based distribution *and* binary distributions. However both point
+releases correspond to the *exact* underlying code. The only difference is the
+version number.
+
+This way, the most recent version of the gem always has binary distributions,
+but if you don't want to use the binaries, you can always "lock in" your
+dependency a single point version down, forcing it to compile from source.
+
+So for example, `1.0.0.1` contains all the binary distributions, while `1.0.0.0`
+is the exact same code, but contain only a source-based distribution.
+
 ## Usage
 
 For use in Rails with `image_optim`. Just adding it to your `Gemfile` should be

data/Rakefile

@@ -59,15 +59,15 @@ rescue LoadError
 end
 
 namespace :build do
-  ['x86_64-linux', 'x86-linux', 'x86_64-freebsd10'].each do |arch|
+  ['x86_64-linux', 'x86-linux', 'x86_64-freebsd10', 'x86_64-freebsd11'].each do |arch|
     desc "build binary gem for #{arch}"
     task arch do
       arch_dir = Pathname(__FILE__).dirname.join("release/#{arch}")
       Dir.chdir(arch_dir) do
         ENV['RUBYLIB'] = nil # https://github.com/mitchellh/vagrant/issues/6158
         sh "vagrant up"
-        # sh "vagrant ssh -c 'rm -rf ~/image_compressor_pack'"
-        system "vagrant ssh -c 'git clone /image_compressor_pack/.git ~/image_compressor_pack'"
+        sh "vagrant ssh -c 'rm -rf ~/image_compressor_pack'"
+        sh "vagrant ssh -c 'git clone /image_compressor_pack/.git ~/image_compressor_pack'"
         sh "vagrant ssh -c 'cd ~/image_compressor_pack && bundle install --path vendor/bundle'"
         sh "cat ~/.ssh/gem-private_key.pem | vagrant ssh -c 'cat > ~/.ssh/gem-private_key.pem'"
         sh "vagrant ssh -c 'cd ~/image_compressor_pack && bundle exec rake binary'"

data/image_compressor_pack.gemspec

@@ -16,8 +16,7 @@ Gem::Specification.new do |spec|
   spec.homepage      = "https://github.com/ignisf/image_compressor_pack"
   spec.license       = "MIT"
 
-  spec.files         = `git ls-files -z`.split("\x0").reject { |f| f.match(%r{^(test|spec|features)/}) }
-  spec.files        += `git ls-files -o -z ports/archives`.split("\x0")
+  spec.files         = `git ls-files -z`.split("\x0").reject { |f| f.match(%r{^(test|spec|features|release|bin)/}) }
   spec.require_paths = ["lib"]
   spec.extensions    = ["ext/image_compressor_pack/extconf.rb"]
 

data/lib/image_compressor_pack/dynamically_linked_recipes.yml

@@ -1,8 +1,8 @@
 nasm:
-  version: 2.12.01
+  version: 2.12.02
   files:
-    - :url: http://www.nasm.us/pub/nasm/releasebuilds/2.12.01/nasm-2.12.01.tar.gz
-      :sha256: 4cab94c041ae85740b47d92d0a25dd2325f05b2378dcf6fac4d09c49c62ee481
+    - :url: http://www.nasm.us/pub/nasm/releasebuilds/2.12.02/nasm-2.12.02.tar.gz
+      :sha256: b7346d827b282ccf70608fe90b9bcf27effb1ddbd1bcba8a5ab4be95284fd01b
 zlib:
   version: 1.2.8
   files:
@@ -10,15 +10,16 @@ zlib:
       :md5: 44d667c142d7cda120332623eab69f40
   configure_options: ['--static']
 lcms2:
-  version: 2.7
+  version: 2.8
   files:
-    - :url: http://download.sourceforge.net/lcms/lcms2-2.7.tar.gz
-      :sha256: 4524234ae7de185e6b6da5d31d6875085b2198bc63b1211f7dde6e2d197d6a53
+    - :url: http://downloads.sourceforge.net/project/lcms/lcms/2.8/lcms2-2.8.tar.gz
+      :sha256: 66d02b229d2ea9474e62c2b6cd6720fde946155cd1d0d2bffdab829790a0fb22
+  configure_options: ['--enable-static', '--disable-shared', '--without-jpeg', '--without-tiff']
 libpng:
-  version: 1.6.21
+  version: 1.6.26
   files:
-    - :url: http://download.sourceforge.net/libpng/libpng-1.6.21.tar.gz
-      :md5: aca36ec8e0a3b406a5912243bc243717
+    - :url: http://downloads.sourceforge.net/project/libpng/libpng16/1.6.26/libpng-1.6.26.tar.gz
+      :md5: 236cd975520fc1f34cc0b8f0e615f7a0
   patch_files:
     - 0001-Do-not-build-binary-utilities.patch
   configure_options: ['--enable-static', '--disable-shared', '--without-binconfigs', '--disable-unversioned-libpng-pc', '--disable-unversioned-libpng-config']
@@ -37,10 +38,10 @@ optipng:
       :sha256: 4870631fcbd3825605f00a168b8debf44ea1cda8ef98a73e5411eee97199be80
   configure_options: ['-with-system-libs']
 pngquant:
-  version: 2.7.1
+  version: 2.8.0
   files:
-    - :url: https://github.com/pornel/pngquant/archive/2.7.1.tar.gz
-      :sha256: e8645bea07ef255c102f053d3566ac00e64005aa0e68a2c9f00bc39e575f1dfd
+    - :url: https://pngquant.org/pngquant-2.8.0-src.tar.gz
+      :sha1: f4428793dbc439a12bc67c7a585aa2fb9b86b349
   patch_files:
     - 0001-Work-around-mini-portile-s-configure-invocation.patch
     - 0002-Add-default-LDFLAGS.patch
@@ -51,10 +52,12 @@ pngquant:
     - 0007-Use-cc-instead-of-gcc-by-default.patch
   configure_options: []
 advancecomp:
-  version: 1.20
+  version: 1.22
   files:
-    - :url: https://github.com/amadvance/advancecomp/releases/download/v1.20/advancecomp-1.20.tar.gz
-      :sha256: 590a447cfc7ab3a37ec707e13967a0046a81a888c561ebaff5415b1e946da67b
+    - :url: https://github.com/amadvance/advancecomp/releases/download/v1.22/advancecomp-1.22.tar.gz
+      :sha256: 31dec84049b919176d14070b8e6f97e086a0ddcd24ff1b83cf498245ece6607d
+  patch_files:
+    - 0001-Add-missing-libdeflate-header-file.patch
 gifsicle:
   version: 1.88
   files:
@@ -72,19 +75,19 @@ jhead:
     - 0003-Make-the-configure-script-set-the-DESTDIR.patch
     - 0004-Make-the-makefile-install-to-the-proper-location.patch
 pngcrush:
-  version: 1.8.1
+  version: 1.8.10
   files:
-    - :url: http://downloads.sourceforge.net/project/pmt/pngcrush/1.8.1/pngcrush-1.8.1.tar.gz
-      :sha256: ab299feb0dd77d2e84a043cf773302ad6323733d9b82e8ce76468dd024c0ad63
+    - :url: http://downloads.sourceforge.net/project/pmt/pngcrush/1.8.10/pngcrush-1.8.10.tar.gz
+      :sha256: ed8dc4759d5067ebf53a2a5188eff1e8ad10262737cf50516cccf8c60d220b6d
   patch_files:
     - 0001-Add-an-install-task.patch
     - 0002-Add-a-configure-script.patch
     - 0003-Use-cc-by-default-instead-of-gcc.patch
 jpegoptim:
-  version: 1.4.3
+  version: 1.4.4
   files:
-    - :url: http://www.kokkonen.net/tjko/src/jpegoptim-1.4.3.tar.gz
-      :sha256: 233d4ae09273cb977e162671f4767be7ef5d96e8c1888d3ed4aa70c4dac1a34c
+    - :url: http://www.kokkonen.net/tjko/src/jpegoptim-1.4.4.tar.gz
+      :sha256: 89a03ea2dfe9483dfb2e009a2851be0befb0deb7cb45c04550bcc91e1e88c1b2
   patch_files:
     - 0001-Link-lm-after-ljpeg.patch
 jpeg-archive:

data/lib/image_compressor_pack/statically_linked_recipes.yml

@@ -1,25 +1,26 @@
 nasm:
-  version: 2.12.01
+  version: 2.12.02
   files:
-    - :url: http://www.nasm.us/pub/nasm/releasebuilds/2.12.01/nasm-2.12.01.tar.gz
-      :sha256: 4cab94c041ae85740b47d92d0a25dd2325f05b2378dcf6fac4d09c49c62ee481
+    - :url: http://www.nasm.us/pub/nasm/releasebuilds/2.12.02/nasm-2.12.02.tar.gz
+      :sha256: b7346d827b282ccf70608fe90b9bcf27effb1ddbd1bcba8a5ab4be95284fd01b
   configure_options: ['LDFLAGS=-static']
 zlib:
   version: 1.2.8
   files:
-    - :url: http://download.sourceforge.net/libpng/zlib-1.2.8.tar.gz
+    - :url: http://downloads.sourceforge.net/project/libpng/zlib/1.2.8/zlib-1.2.8.tar.gz
       :md5: 44d667c142d7cda120332623eab69f40
   configure_options: ['--static']
 lcms2:
-  version: 2.7
+  version: 2.8
   files:
-    - :url: http://download.sourceforge.net/lcms/lcms2-2.7.tar.gz
-      :sha256: 4524234ae7de185e6b6da5d31d6875085b2198bc63b1211f7dde6e2d197d6a53
+    - :url: http://downloads.sourceforge.net/project/lcms/lcms/2.8/lcms2-2.8.tar.gz
+      :sha256: 66d02b229d2ea9474e62c2b6cd6720fde946155cd1d0d2bffdab829790a0fb22
+  configure_options: ['--enable-static', '--disable-shared', '--without-jpeg', '--without-tiff']
 libpng:
-  version: 1.6.21
+  version: 1.6.26
   files:
-    - :url: http://download.sourceforge.net/libpng/libpng-1.6.21.tar.gz
-      :md5: aca36ec8e0a3b406a5912243bc243717
+    - :url: http://downloads.sourceforge.net/project/libpng/libpng16/1.6.26/libpng-1.6.26.tar.gz
+      :md5: 236cd975520fc1f34cc0b8f0e615f7a0
   patch_files:
     - 0001-Do-not-build-binary-utilities.patch
   configure_options: ['--enable-static', '--disable-shared', '--without-binconfigs', '--disable-unversioned-libpng-pc', '--disable-unversioned-libpng-config']
@@ -41,10 +42,10 @@ optipng:
   patch_files:
     - 0001-Allow-passing-LDFLAGS-as-configure-arg.patch
 pngquant:
-  version: 2.7.1
+  version: 2.8.0
   files:
-    - :url: https://github.com/pornel/pngquant/archive/2.7.1.tar.gz
-      :sha256: e8645bea07ef255c102f053d3566ac00e64005aa0e68a2c9f00bc39e575f1dfd
+    - :url: https://pngquant.org/pngquant-2.8.0-src.tar.gz
+      :sha1: f4428793dbc439a12bc67c7a585aa2fb9b86b349
   patch_files:
     - 0001-Work-around-mini-portile-s-configure-invocation.patch
     - 0002-Add-default-LDFLAGS.patch
@@ -54,10 +55,12 @@ pngquant:
     - 0007-Use-cc-instead-of-gcc-by-default.patch
   configure_options: ['--extra-ldflags=-lpng']
 advancecomp:
-  version: 1.20
+  version: 1.22
   files:
-    - :url: https://github.com/amadvance/advancecomp/releases/download/v1.20/advancecomp-1.20.tar.gz
-      :sha256: 590a447cfc7ab3a37ec707e13967a0046a81a888c561ebaff5415b1e946da67b
+    - :url: https://github.com/amadvance/advancecomp/releases/download/v1.22/advancecomp-1.22.tar.gz
+      :sha256: 31dec84049b919176d14070b8e6f97e086a0ddcd24ff1b83cf498245ece6607d
+  patch_files:
+    - 0001-Add-missing-libdeflate-header-file.patch
   configure_options: ['LDFLAGS=-static']
 gifsicle:
   version: 1.88
@@ -77,20 +80,20 @@ jhead:
     - 0004-Make-the-makefile-install-to-the-proper-location.patch
     - 0005-Produce-a-static-binary.patch
 pngcrush:
-  version: 1.8.1
+  version: 1.8.10
   files:
-    - :url: http://downloads.sourceforge.net/project/pmt/pngcrush/1.8.1/pngcrush-1.8.1.tar.gz
-      :sha256: ab299feb0dd77d2e84a043cf773302ad6323733d9b82e8ce76468dd024c0ad63
+    - :url: http://downloads.sourceforge.net/project/pmt/pngcrush/1.8.10/pngcrush-1.8.10.tar.gz
+      :sha256: ed8dc4759d5067ebf53a2a5188eff1e8ad10262737cf50516cccf8c60d220b6d
   patch_files:
     - 0001-Add-an-install-task.patch
     - 0002-Add-a-configure-script.patch
     - 0003-Use-cc-by-default-instead-of-gcc.patch
     - 0004-Produce-a-static-binary.patch
 jpegoptim:
-  version: 1.4.3
+  version: 1.4.4
   files:
-    - :url: http://www.kokkonen.net/tjko/src/jpegoptim-1.4.3.tar.gz
-      :sha256: 233d4ae09273cb977e162671f4767be7ef5d96e8c1888d3ed4aa70c4dac1a34c
+    - :url: http://www.kokkonen.net/tjko/src/jpegoptim-1.4.4.tar.gz
+      :sha256: 89a03ea2dfe9483dfb2e009a2851be0befb0deb7cb45c04550bcc91e1e88c1b2
   patch_files:
     - 0001-Link-lm-after-ljpeg.patch
   configure_options: ['LDFLAGS=-static']

data/lib/image_compressor_pack/version.rb

@@ -1,3 +1,3 @@
 module ImageCompressorPack
-  VERSION = "0.1.3"
+  VERSION = "1.0.0.0"
 end

data/ports/patches/advancecomp/0001-Add-missing-libdeflate-header-file.patch

@@ -0,0 +1,305 @@
+From ad7bbef7da9a6dd7121366e8552085a0e14170f1 Mon Sep 17 00:00:00 2001
+From: Petko Bordjukov <bordjukov@gmail.com>
+Date: Sat, 19 Nov 2016 12:27:16 +0200
+Subject: [PATCH] Add missing libdeflate header file
+
+---
+ libdeflate/crc32_impl.h | 286 ++++++++++++++++++++++++++++++++++++++++++++++++
+ 1 file changed, 286 insertions(+)
+ create mode 100644 libdeflate/crc32_impl.h
+
+diff --git a/libdeflate/crc32_impl.h b/libdeflate/crc32_impl.h
+new file mode 100644
+index 0000000..625bc18
+--- /dev/null
++++ b/libdeflate/crc32_impl.h
+@@ -0,0 +1,286 @@
++/*
++ * crc32_impl.h
++ *
++ * Copyright 2016 Eric Biggers
++ *
++ * Permission is hereby granted, free of charge, to any person
++ * obtaining a copy of this software and associated documentation
++ * files (the "Software"), to deal in the Software without
++ * restriction, including without limitation the rights to use,
++ * copy, modify, merge, publish, distribute, sublicense, and/or sell
++ * copies of the Software, and to permit persons to whom the
++ * Software is furnished to do so, subject to the following
++ * conditions:
++ *
++ * The above copyright notice and this permission notice shall be
++ * included in all copies or substantial portions of the Software.
++ *
++ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
++ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES
++ * OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
++ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
++ * HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY,
++ * WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
++ * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
++ * OTHER DEALINGS IN THE SOFTWARE.
++ */
++
++/*
++ * CRC-32 folding with PCLMULQDQ.
++ *
++ * The basic idea is to repeatedly "fold" each 512 bits into the next 512 bits,
++ * producing an abbreviated message which is congruent the original message
++ * modulo the generator polynomial G(x).
++ *
++ * Folding each 512 bits is implemented as eight 64-bit folds, each of which
++ * uses one carryless multiplication instruction.  It's expected that CPUs may
++ * be able to execute some of these multiplications in parallel.
++ *
++ * Explanation of "folding": let A(x) be 64 bits from the message, and let B(x)
++ * be 95 bits from a constant distance D later in the message.  The relevant
++ * portion of the message can be written as:
++ *
++ *	M(x) = A(x)*x^D + B(x)
++ *
++ * ... where + and * represent addition and multiplication, respectively, of
++ * polynomials over GF(2).  Note that when implemented on a computer, these
++ * operations are equivalent to XOR and carryless multiplication, respectively.
++ *
++ * For the purpose of CRC calculation, only the remainder modulo the generator
++ * polynomial G(x) matters:
++ *
++ *	M(x) mod G(x) = (A(x)*x^D + B(x)) mod G(x)
++ *
++ * Since the modulo operation can be applied anywhere in a sequence of additions
++ * and multiplications without affecting the result, this is equivalent to:
++ *
++ *	M(x) mod G(x) = (A(x)*(x^D mod G(x)) + B(x)) mod G(x)
++ *
++ * For any D, 'x^D mod G(x)' will be a polynomial with maximum degree 31, i.e.
++ * a 32-bit quantity.  So 'A(x) * (x^D mod G(x))' is equivalent to a carryless
++ * multiplication of a 64-bit quantity by a 32-bit quantity, producing a 95-bit
++ * product.  Then, adding (XOR-ing) the product to B(x) produces a polynomial
++ * with the same length as B(x) but with the same remainder as 'A(x)*x^D +
++ * B(x)'.  This is the basic fold operation with 64 bits.
++ *
++ * Note that the carryless multiplication instruction PCLMULQDQ actually takes
++ * two 64-bit inputs and produces a 127-bit product in the low-order bits of a
++ * 128-bit XMM register.  This works fine, but care must be taken to account for
++ * "bit endianness".  With the CRC version implemented here, bits are always
++ * ordered such that the lowest-order bit represents the coefficient of highest
++ * power of x and the highest-order bit represents the coefficient of the lowest
++ * power of x.  This is backwards from the more intuitive order.  Still,
++ * carryless multiplication works essentially the same either way.  It just must
++ * be accounted for that when we XOR the 95-bit product in the low-order 95 bits
++ * of a 128-bit XMM register into 128-bits of later data held in another XMM
++ * register, we'll really be XOR-ing the product into the mathematically higher
++ * degree end of those later bits, not the lower degree end as may be expected.
++ *
++ * So given that caveat and the fact that we process 512 bits per iteration, the
++ * 'D' values we need for the two 64-bit halves of each 128 bits of data are:
++ *
++ *	D = (512 + 95) - 64	 for the higher-degree half of each 128 bits,
++ *				 i.e. the lower order bits in the XMM register
++ *
++ *	D = (512 + 95) - 128	 for the lower-degree half of each 128 bits,
++ *				 i.e. the higher order bits in the XMM register
++ *
++ * The required 'x^D mod G(x)' values were precomputed.
++ *
++ * When <= 512 bits remain in the message, we finish up by folding across
++ * smaller distances.  This works similarly; the distance D is just different,
++ * so different constant multipliers must be used.  Finally, once the remaining
++ * message is just 64 bits, it is is reduced to the CRC-32 using Barrett
++ * reduction (explained later).
++ *
++ * For more information see the original paper from Intel:
++ *	"Fast CRC Computation for Generic Polynomials Using PCLMULQDQ Instruction"
++ *	December 2009
++ *	http://www.intel.com/content/dam/www/public/us/en/documents/white-papers/fast-crc-computation-generic-polynomials-pclmulqdq-paper.pdf
++ */
++static u32 ATTRIBUTES
++FUNCNAME_ALIGNED(u32 remainder, const __m128i *p, size_t vec_count)
++{
++	/* Constants precomputed by gen_crc32_multipliers.c.  Do not edit! */
++	const __v2di multipliers_4 = (__v2di){ 0x8F352D95, 0x1D9513D7 };
++	const __v2di multipliers_2 = (__v2di){ 0xF1DA05AA, 0x81256527 };
++	const __v2di multipliers_1 = (__v2di){ 0xAE689191, 0xCCAA009E };
++	const __v2di final_multiplier = (__v2di){ 0xB8BC6765 };
++	const __m128i mask32 = (__m128i)(__v4si){ 0xFFFFFFFF };
++	const __v2di barrett_reduction_constants =
++			(__v2di){ 0x00000001F7011641, 0x00000001DB710641 };
++
++	const __m128i * const end = p + vec_count;
++	const __m128i * const end512 = p + (vec_count & ~3);
++	__m128i x0, x1, x2, x3;
++
++	/*
++	 * Account for the current 'remainder', i.e. the CRC of the part of the
++	 * message already processed.  Explanation: rewrite the message
++	 * polynomial M(x) in terms of the first part A(x), the second part
++	 * B(x), and the length of the second part in bits |B(x)| >= 32:
++	 *
++	 *	M(x) = A(x)*x^|B(x)| + B(x)
++	 *
++	 * Then the CRC of M(x) is:
++	 *
++	 *	CRC(M(x)) = CRC(A(x)*x^|B(x)| + B(x))
++	 *	          = CRC(A(x)*x^32*x^(|B(x)| - 32) + B(x))
++	 *	          = CRC(CRC(A(x))*x^(|B(x)| - 32) + B(x))
++	 *
++	 * Note: all arithmetic is modulo G(x), the generator polynomial; that's
++	 * why A(x)*x^32 can be replaced with CRC(A(x)) = A(x)*x^32 mod G(x).
++	 *
++	 * So the CRC of the full message is the CRC of the second part of the
++	 * message where the first 32 bits of the second part of the message
++	 * have been XOR'ed with the CRC of the first part of the message.
++	 */
++	x0 = *p++;
++	x0 ^= (__m128i)(__v4si){ remainder };
++
++	if (p > end512) /* only 128, 256, or 384 bits of input? */
++		goto _128_bits_at_a_time;
++	x1 = *p++;
++	x2 = *p++;
++	x3 = *p++;
++
++	/* Fold 512 bits at a time */
++	for (; p != end512; p += 4) {
++		__m128i y0, y1, y2, y3;
++
++		y0 = p[0];
++		y1 = p[1];
++		y2 = p[2];
++		y3 = p[3];
++
++		/*
++		 * Note: the immediate constant for PCLMULQDQ specifies which
++		 * 64-bit halves of the 128-bit vectors to multiply:
++		 *
++		 * 0x00 means low halves (higher degree polynomial terms for us)
++		 * 0x11 means high halves (lower degree polynomial terms for us)
++		 */
++		y0 ^= _mm_clmulepi64_si128(x0, multipliers_4, 0x00);
++		y1 ^= _mm_clmulepi64_si128(x1, multipliers_4, 0x00);
++		y2 ^= _mm_clmulepi64_si128(x2, multipliers_4, 0x00);
++		y3 ^= _mm_clmulepi64_si128(x3, multipliers_4, 0x00);
++		y0 ^= _mm_clmulepi64_si128(x0, multipliers_4, 0x11);
++		y1 ^= _mm_clmulepi64_si128(x1, multipliers_4, 0x11);
++		y2 ^= _mm_clmulepi64_si128(x2, multipliers_4, 0x11);
++		y3 ^= _mm_clmulepi64_si128(x3, multipliers_4, 0x11);
++
++		x0 = y0;
++		x1 = y1;
++		x2 = y2;
++		x3 = y3;
++	}
++
++	/* Fold 512 bits => 128 bits */
++	x2 ^= _mm_clmulepi64_si128(x0, multipliers_2, 0x00);
++	x3 ^= _mm_clmulepi64_si128(x1, multipliers_2, 0x00);
++	x2 ^= _mm_clmulepi64_si128(x0, multipliers_2, 0x11);
++	x3 ^= _mm_clmulepi64_si128(x1, multipliers_2, 0x11);
++	x3 ^= _mm_clmulepi64_si128(x2, multipliers_1, 0x00);
++	x3 ^= _mm_clmulepi64_si128(x2, multipliers_1, 0x11);
++	x0 = x3;
++
++_128_bits_at_a_time:
++	while (p != end) {
++		/* Fold 128 bits into next 128 bits */
++		x1 = *p++;
++		x1 ^= _mm_clmulepi64_si128(x0, multipliers_1, 0x00);
++		x1 ^= _mm_clmulepi64_si128(x0, multipliers_1, 0x11);
++		x0 = x1;
++	}
++
++	/* Now there are just 128 bits left, stored in 'x0'. */
++
++	/*
++	 * Fold 128 => 96 bits.  This also implicitly appends 32 zero bits,
++	 * which is equivalent to multiplying by x^32.  This is needed because
++	 * the CRC is defined as M(x)*x^32 mod G(x), not just M(x) mod G(x).
++	 */
++	x0 = _mm_srli_si128(x0, 8) ^
++	     _mm_clmulepi64_si128(x0, multipliers_1, 0x10);
++
++	/* Fold 96 => 64 bits */
++	x0 = _mm_srli_si128(x0, 4) ^
++	     _mm_clmulepi64_si128(x0 & mask32, final_multiplier, 0x00);
++
++        /*
++	 * Finally, reduce 64 => 32 bits using Barrett reduction.
++	 *
++	 * Let M(x) = A(x)*x^32 + B(x) be the remaining message.  The goal is to
++	 * compute R(x) = M(x) mod G(x).  Since degree(B(x)) < degree(G(x)):
++	 *
++	 *	R(x) = (A(x)*x^32 + B(x)) mod G(x)
++	 *	     = (A(x)*x^32) mod G(x) + B(x)
++	 *
++	 * Then, by the Division Algorithm there exists a unique q(x) such that:
++	 *
++	 *	A(x)*x^32 mod G(x) = A(x)*x^32 - q(x)*G(x)
++	 *
++	 * Since the left-hand side is of maximum degree 31, the right-hand side
++	 * must be too.  This implies that we can apply 'mod x^32' to the
++	 * right-hand side without changing its value:
++	 *
++	 *	(A(x)*x^32 - q(x)*G(x)) mod x^32 = q(x)*G(x) mod x^32
++	 *
++	 * Note that '+' is equivalent to '-' in polynomials over GF(2).
++	 *
++	 * We also know that:
++	 *
++	 *	              / A(x)*x^32 \
++	 *	q(x) = floor (  ---------  )
++	 *	              \    G(x)   /
++	 *
++	 * To compute this efficiently, we can multiply the top and bottom by
++	 * x^32 and move the division by G(x) to the top:
++	 *
++	 *	              / A(x) * floor(x^64 / G(x)) \
++	 *	q(x) = floor (  -------------------------  )
++	 *	              \           x^32            /
++	 *
++	 * Note that floor(x^64 / G(x)) is a constant.
++	 *
++	 * So finally we have:
++	 *
++	 *	                          / A(x) * floor(x^64 / G(x)) \
++	 *	R(x) = B(x) + G(x)*floor (  -------------------------  )
++	 *	                          \           x^32            /
++	 */
++	x1 = x0;
++	x0 = _mm_clmulepi64_si128(x0 & mask32, barrett_reduction_constants, 0x00);
++	x0 = _mm_clmulepi64_si128(x0 & mask32, barrett_reduction_constants, 0x10);
++	return _mm_cvtsi128_si32(_mm_srli_si128(x0 ^ x1, 4));
++}
++
++/*
++ * Fast CRC-32 implementation for x86_64 processors that have the carryless
++ * multiplication extension (PCLMUL).
++ *
++ * Note: on unaligned ends of the buffer, we fall back to crc32_slice1() instead
++ * of crc32_slice8() because only a few bytes need to be processed, so a smaller
++ * table is preferable.
++ */
++static u32 ATTRIBUTES
++FUNCNAME(u32 remainder, const u8 *buffer, size_t nbytes)
++{
++	if ((uintptr_t)buffer & 15) {
++		size_t n = MIN(nbytes, -(uintptr_t)buffer & 15);
++		remainder = crc32_slice1(remainder, buffer, n);
++		buffer += n;
++		nbytes -= n;
++	}
++	if (nbytes >= 16) {
++		remainder = FUNCNAME_ALIGNED(remainder, (const __m128i *)buffer,
++					     nbytes / 16);
++		buffer += nbytes & ~15;
++		nbytes &= 15;
++	}
++	return crc32_slice1(remainder, buffer, nbytes);
++}
++
++#undef FUNCNAME
++#undef FUNCNAME_ALIGNED
++#undef ATTRIBUTES
+-- 
+2.10.2
+