Binary files /tmp/tmpBi8HUl/DDUaiQON8Z/argon2-0~20161029/argon2-specs.pdf and /tmp/tmpBi8HUl/FgaFBMIsIi/argon2-0~20171227/argon2-specs.pdf differ
diff -Nru argon2-0~20161029/CHANGELOG.md argon2-0~20171227/CHANGELOG.md
--- argon2-0~20161029/CHANGELOG.md	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/CHANGELOG.md	2017-12-27 19:56:31.000000000 +0000
@@ -1,3 +1,10 @@
+# 20171227
+* Added ABI version number
+* AVX2/AVX-512F optimizations of BLAMKA
+* Set Argon2 version number from the command line
+* New bindings
+* Minor bug and warning fixes (no security issue)
+
 # 20161029
 
 * Argon2id added
diff -Nru argon2-0~20161029/debian/changelog argon2-0~20171227/debian/changelog
--- argon2-0~20161029/debian/changelog	2018-04-23 20:59:38.000000000 +0000
+++ argon2-0~20171227/debian/changelog	2018-06-06 20:08:10.000000000 +0000
@@ -1,3 +1,12 @@
+argon2 (0~20171227-0.1) unstable; urgency=medium
+
+  * Non-maintainer upload.
+  * New upstream release
+  * ABI bump.
+  * Build udeb without a dependency on pthreads.
+
+ -- Matthew Garrett <mjg59@google.com>  Wed, 06 Jun 2018 13:08:10 -0700
+
 argon2 (0~20161029-2) unstable; urgency=medium
 
   * control: move vcs to salsa gitlab
diff -Nru argon2-0~20161029/debian/control argon2-0~20171227/debian/control
--- argon2-0~20161029/debian/control	2018-04-23 20:42:46.000000000 +0000
+++ argon2-0~20171227/debian/control	2018-06-06 20:08:10.000000000 +0000
@@ -9,10 +9,13 @@
 Vcs-Browser: https://salsa.debian.org/debian/argon2
 Vcs-Git: https://salsa.debian.org/debian/argon2.git
 
-Package: libargon2-0-dev
+Package: libargon2-dev
 Section: libdevel
 Architecture: any
-Depends: libargon2-0 (= ${binary:Version}), ${misc:Depends}
+Depends: libargon2-1 (= ${binary:Version}), ${misc:Depends}
+Provides: libargon2-0-dev
+Replaces: libargon2-0-dev
+Conflicts: libargon2-0-dev
 Multi-Arch: same
 Description: memory-hard hashing function - development files
  Argon2 is a password-hashing function that can be used to hash passwords
@@ -32,6 +35,28 @@
  on argon2.
 
 Package: libargon2-0
+Section: libdevel
+Architecture: any
+Depends: libargon2-1 (= ${binary:Version})
+Description: memory-hard hashing function - runtime library
+ Argon2 is a password-hashing function that can be used to hash passwords
+ for credential storage, key derivation, or other applications.
+ .
+ There are two main versions of Argon2: Argon2i and Argon2d.
+ Argon2i is the safest against side-channel attacks, while Argon2d provides
+ the highest resistance against GPU cracking attacks.
+ .
+ Argon2i and Argon2d are parametrized by:
+  * A time cost, which defines the amount of computation realized and
+    therefore the execution time, given in number of iterations
+  * A memory cost, which defines the memory usage, given in kibibytes
+  * A parallelism degree, which defines the number of parallel threads
+ .
+ This package provides compatibility for packages built against older
+ versions of libargon2.
+
+
+Package: libargon2-1
 Architecture: any
 Pre-Depends: ${misc:Pre-Depends}
 Depends: ${shlibs:Depends}, ${misc:Depends}
@@ -73,7 +98,7 @@
  .
  This package contains the argon2 tool for hashing data on the command-line.
 
-Package: libargon2-0-udeb
+Package: libargon2-1-udeb
 Section: debian-installer
 Package-Type: udeb
 Architecture: any
diff -Nru argon2-0~20161029/debian/libargon2-0-dev.install argon2-0~20171227/debian/libargon2-0-dev.install
--- argon2-0~20161029/debian/libargon2-0-dev.install	2017-07-16 10:59:25.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-0-dev.install	1970-01-01 00:00:00.000000000 +0000
@@ -1,4 +0,0 @@
-#!/usr/bin/dh-exec
-include/argon2.h usr/include
-libargon2.a usr/lib/${DEB_HOST_MULTIARCH}
-libargon2.pc usr/lib/${DEB_HOST_MULTIARCH}/pkgconfig
diff -Nru argon2-0~20161029/debian/libargon2-0-dev.links argon2-0~20171227/debian/libargon2-0-dev.links
--- argon2-0~20161029/debian/libargon2-0-dev.links	2017-07-16 10:59:25.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-0-dev.links	1970-01-01 00:00:00.000000000 +0000
@@ -1,2 +0,0 @@
-#!/usr/bin/dh-exec
-usr/lib/${DEB_HOST_MULTIARCH}/${SONAME} usr/lib/${DEB_HOST_MULTIARCH}/libargon2.so
diff -Nru argon2-0~20161029/debian/libargon2-0.install argon2-0~20171227/debian/libargon2-0.install
--- argon2-0~20161029/debian/libargon2-0.install	2017-07-16 10:59:25.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-0.install	1970-01-01 00:00:00.000000000 +0000
@@ -1,2 +0,0 @@
-#!/usr/bin/dh-exec
-libargon2.so => usr/lib/${DEB_HOST_MULTIARCH}/${SONAME}
diff -Nru argon2-0~20161029/debian/libargon2-0.links argon2-0~20171227/debian/libargon2-0.links
--- argon2-0~20161029/debian/libargon2-0.links	1970-01-01 00:00:00.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-0.links	2018-06-06 20:08:10.000000000 +0000
@@ -0,0 +1,2 @@
+#!/usr/bin/dh-exec
+usr/lib/${DEB_HOST_MULTIARCH}/${SONAME} usr/lib/${DEB_HOST_MULTIARCH}/libargon2.so.0
diff -Nru argon2-0~20161029/debian/libargon2-0.symbols argon2-0~20171227/debian/libargon2-0.symbols
--- argon2-0~20161029/debian/libargon2-0.symbols	2017-07-16 10:59:25.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-0.symbols	1970-01-01 00:00:00.000000000 +0000
@@ -1,23 +0,0 @@
-libargon2.so.0 libargon2-0 #MINVER#
- argon2_ctx@Base 0~20160406~
- argon2_encodedlen@Base 0~20160406~
- argon2_error_message@Base 0~20160406~
- argon2_hash@Base 0~20160406~
- argon2_type2string@Base 0~20161029~
- argon2_verify@Base 0~20160406~
- argon2_verify_ctx@Base 0~20160406~
- argon2d_ctx@Base 0~20160406~
- argon2d_hash_encoded@Base 0~20160406~
- argon2d_hash_raw@Base 0~20160406~
- argon2d_verify@Base 0~20160406~
- argon2d_verify_ctx@Base 0~20160406~
- argon2i_ctx@Base 0~20160406~
- argon2i_hash_encoded@Base 0~20160406~
- argon2i_hash_raw@Base 0~20160406~
- argon2i_verify@Base 0~20160406~
- argon2i_verify_ctx@Base 0~20160406~
- argon2id_ctx@Base 0~20161029~
- argon2id_hash_encoded@Base 0~20161029~
- argon2id_hash_raw@Base 0~20161029~
- argon2id_verify@Base 0~20161029~
- argon2id_verify_ctx@Base 0~20161029~
diff -Nru argon2-0~20161029/debian/libargon2-0-udeb.install argon2-0~20171227/debian/libargon2-0-udeb.install
--- argon2-0~20161029/debian/libargon2-0-udeb.install	2018-04-23 20:30:52.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-0-udeb.install	1970-01-01 00:00:00.000000000 +0000
@@ -1,2 +0,0 @@
-#!/usr/bin/dh-exec
-libargon2.so => usr/lib/${DEB_HOST_MULTIARCH}/${SONAME}
diff -Nru argon2-0~20161029/debian/libargon2-1.install argon2-0~20171227/debian/libargon2-1.install
--- argon2-0~20161029/debian/libargon2-1.install	1970-01-01 00:00:00.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-1.install	2018-06-06 20:08:10.000000000 +0000
@@ -0,0 +1,2 @@
+#!/usr/bin/dh-exec
+libargon2.so.1 => usr/lib/${DEB_HOST_MULTIARCH}/${SONAME}
diff -Nru argon2-0~20161029/debian/libargon2-1.symbols argon2-0~20171227/debian/libargon2-1.symbols
--- argon2-0~20161029/debian/libargon2-1.symbols	1970-01-01 00:00:00.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-1.symbols	2018-06-06 20:08:10.000000000 +0000
@@ -0,0 +1,23 @@
+libargon2.so.1 libargon2-1 #MINVER#
+ argon2_ctx@Base 0~20171227
+ argon2_encodedlen@Base 0~20171227
+ argon2_error_message@Base 0~20171227
+ argon2_hash@Base 0~20171227
+ argon2_type2string@Base 0~20171227
+ argon2_verify@Base 0~20171227
+ argon2_verify_ctx@Base 0~20171227
+ argon2d_ctx@Base 0~20171227
+ argon2d_hash_encoded@Base 0~20171227
+ argon2d_hash_raw@Base 0~20171227
+ argon2d_verify@Base 0~20171227
+ argon2d_verify_ctx@Base 0~20171227
+ argon2i_ctx@Base 0~20171227
+ argon2i_hash_encoded@Base 0~20171227
+ argon2i_hash_raw@Base 0~20171227
+ argon2i_verify@Base 0~20171227
+ argon2i_verify_ctx@Base 0~20171227
+ argon2id_ctx@Base 0~20171227
+ argon2id_hash_encoded@Base 0~20171227
+ argon2id_hash_raw@Base 0~20171227
+ argon2id_verify@Base 0~20171227
+ argon2id_verify_ctx@Base 0~20171227
diff -Nru argon2-0~20161029/debian/libargon2-dev.install argon2-0~20171227/debian/libargon2-dev.install
--- argon2-0~20161029/debian/libargon2-dev.install	1970-01-01 00:00:00.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-dev.install	2018-06-06 20:08:10.000000000 +0000
@@ -0,0 +1,4 @@
+#!/usr/bin/dh-exec
+include/argon2.h usr/include
+libargon2.a usr/lib/${DEB_HOST_MULTIARCH}
+libargon2.pc usr/lib/${DEB_HOST_MULTIARCH}/pkgconfig
diff -Nru argon2-0~20161029/debian/libargon2-dev.links argon2-0~20171227/debian/libargon2-dev.links
--- argon2-0~20161029/debian/libargon2-dev.links	1970-01-01 00:00:00.000000000 +0000
+++ argon2-0~20171227/debian/libargon2-dev.links	2018-06-06 20:08:10.000000000 +0000
@@ -0,0 +1,2 @@
+#!/usr/bin/dh-exec
+usr/lib/${DEB_HOST_MULTIARCH}/${SONAME} usr/lib/${DEB_HOST_MULTIARCH}/libargon2.so
diff -Nru argon2-0~20161029/debian/rules argon2-0~20171227/debian/rules
--- argon2-0~20161029/debian/rules	2018-04-23 20:55:23.000000000 +0000
+++ argon2-0~20171227/debian/rules	2018-06-06 20:08:10.000000000 +0000
@@ -9,24 +9,36 @@
 include /usr/share/dpkg/buildflags.mk
 include /usr/share/dpkg/default.mk
 
+export PKG=libargon2
 export OPTTARGET=$(DEB_TARGET_GNU_CPU)
-export SONAME=libargon2.so.$(SOVER)
-export SOPKG=libargon2-$(SOVER)
-export SOVER=0
+export SONAME=$(PKG).so.$(SOVER)
+export SOPKG=$(PKG)-$(SOVER)
+export SOVER=1
 export DEBPREFIX=0~
 export UPSTREAMVER=$(subst $(DEBPREFIX),,$(DEB_VERSION_UPSTREAM))
+export UDEBLIB=$(SONAME).udeb
 
 %:
 	dh $@
 
+override_dh_auto_build:
+	dh_auto_build -- NO_THREADS=1
+	mv $(SONAME) $(UDEBLIB)
+	dh_auto_build
+
 override_dh_install:
 	dh_install
-	sed -i '/^## /d' debian/${SOPKG}-dev/usr/lib/$(DEB_HOST_MULTIARCH)/pkgconfig/libargon2.pc
-	sed -i s/@HOST_MULTIARCH@/$(DEB_HOST_MULTIARCH)/ debian/${SOPKG}-dev/usr/lib/$(DEB_HOST_MULTIARCH)/pkgconfig/libargon2.pc
-	sed -i s/@UPSTREAM_VER@/$(UPSTREAMVER)/ debian/${SOPKG}-dev/usr/lib/$(DEB_HOST_MULTIARCH)/pkgconfig/libargon2.pc
+	sed -i '/^## /d' debian/${PKG}-dev/usr/lib/$(DEB_HOST_MULTIARCH)/pkgconfig/libargon2.pc
+	sed -i s/@HOST_MULTIARCH@/$(DEB_HOST_MULTIARCH)/ debian/${PKG}-dev/usr/lib/$(DEB_HOST_MULTIARCH)/pkgconfig/libargon2.pc
+	sed -i s/@UPSTREAM_VER@/$(UPSTREAMVER)/ debian/${PKG}-dev/usr/lib/$(DEB_HOST_MULTIARCH)/pkgconfig/libargon2.pc
+	install -D $(UDEBLIB) debian/$(SOPKG)-udeb/usr/lib/$(DEB_HOST_MULTIARCH)/$(SONAME)
+
+override_dh_auto_clean:
+	dh_auto_clean
+	-rm $(UDEBLIB)
 
 override_dh_auto_test:
 	-$(MAKE) -j1 test
 
 override_dh_makeshlibs:
-	dh_makeshlibs -V --add-udeb="libargon2-0-udeb"
+	dh_makeshlibs -V --add-udeb="libargon2-1-udeb"
diff -Nru argon2-0~20161029/include/argon2.h argon2-0~20171227/include/argon2.h
--- argon2-0~20161029/include/argon2.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/include/argon2.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -29,10 +29,13 @@
 /* Symbols visibility control */
 #ifdef A2_VISCTL
 #define ARGON2_PUBLIC __attribute__((visibility("default")))
+#define ARGON2_LOCAL __attribute__ ((visibility ("hidden")))
 #elif _MSC_VER
 #define ARGON2_PUBLIC __declspec(dllexport)
+#define ARGON2_LOCAL
 #else
 #define ARGON2_PUBLIC
+#define ARGON2_LOCAL
 #endif
 
 /*
@@ -267,8 +270,7 @@
                                        const size_t encodedlen);
 
 /**
- * Hashes a password with Argon2i, producing a raw hash by allocating memory at
- * @hash
+ * Hashes a password with Argon2i, producing a raw hash at @hash
  * @param t_cost Number of iterations
  * @param m_cost Sets memory usage to m_cost kibibytes
  * @param parallelism Number of threads and compute lanes
@@ -421,6 +423,7 @@
  * @param parallelism  Number of threads; used to compute lanes
  * @param saltlen  Salt size in bytes
  * @param hashlen  Hash size in bytes
+ * @param type The argon2_type that we want the encoded length for
  * @return  The encoded hash length in bytes
  */
 ARGON2_PUBLIC size_t argon2_encodedlen(uint32_t t_cost, uint32_t m_cost,
diff -Nru argon2-0~20161029/Makefile argon2-0~20171227/Makefile
--- argon2-0~20161029/Makefile	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/Makefile	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
 # Copyright 2015
 # Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
 #
-# You may use this work under the terms of a Creative Commons CC0 1.0 
+# You may use this work under the terms of a Creative Commons CC0 1.0
 # License/Waiver or the Apache Public License 2.0, at your option. The terms of
 # these licenses can be found at:
 #
@@ -19,6 +19,9 @@
 BENCH = bench
 GENKAT = genkat
 
+# Increment on an ABI breaking change
+ABI_VERSION = 1
+
 DIST = phc-winner-argon2
 
 SRC = src/argon2.c src/core.c src/blake2/blake2b.c src/thread.c src/encoding.c
@@ -27,7 +30,14 @@
 SRC_GENKAT = src/genkat.c
 OBJ = $(SRC:.c=.o)
 
-CFLAGS += -std=c89 -pthread -O3 -Wall -g -Iinclude -Isrc
+CFLAGS += -std=c89 -O3 -Wall -g -Iinclude -Isrc
+
+ifeq ($(NO_THREADS), 1)
+CFLAGS += -DARGON2_NO_THREADS
+else
+CFLAGS += -pthread
+endif
+
 CI_CFLAGS := $(CFLAGS) -Werror=declaration-after-statement -D_FORTIFY_SOURCE=2 \
 				-Wextra -Wno-type-limits -Werror -coverage -DTEST_LARGE_RAM
 
@@ -49,17 +59,19 @@
 
 LIB_NAME=argon2
 ifeq ($(KERNEL_NAME), Linux)
-	LIB_EXT := so
+	LIB_EXT := so.$(ABI_VERSION)
 	LIB_CFLAGS := -shared -fPIC -fvisibility=hidden -DA2_VISCTL=1
-	SO_LDFLAGS := -Wl,-soname,libargon2.so.0
+	SO_LDFLAGS := -Wl,-soname,lib$(LIB_NAME).$(LIB_EXT)
+	LINKED_LIB_EXT := so
 endif
 ifeq ($(KERNEL_NAME), $(filter $(KERNEL_NAME),FreeBSD NetBSD OpenBSD))
 	LIB_EXT := so
 	LIB_CFLAGS := -shared -fPIC
 endif
 ifeq ($(KERNEL_NAME), Darwin)
-	LIB_EXT := dylib
+	LIB_EXT := $(ABI_VERSION).dylib
 	LIB_CFLAGS := -dynamiclib -install_name @rpath/lib$(LIB_NAME).$(LIB_EXT)
+	LINKED_LIB_EXT := dylib
 endif
 ifeq ($(findstring CYGWIN, $(KERNEL_NAME)), CYGWIN)
 	LIB_EXT := dll
@@ -88,6 +100,12 @@
 
 LIB_SH := lib$(LIB_NAME).$(LIB_EXT)
 LIB_ST := lib$(LIB_NAME).a
+
+ifdef LINKED_LIB_EXT
+LINKED_LIB_SH := lib$(LIB_NAME).$(LINKED_LIB_EXT)
+endif
+
+
 LIBRARIES = $(LIB_SH) $(LIB_ST)
 HEADERS = include/argon2.h
 
@@ -105,7 +123,7 @@
 
 .PHONY: clean dist format $(GENKAT) all install
 
-all: clean $(RUN) libs 
+all: $(RUN) libs
 libs: $(LIBRARIES)
 
 $(RUN):	        $(SRC) $(SRC_RUN)
@@ -125,7 +143,7 @@
 
 clean:
 		rm -f $(RUN) $(BENCH) $(GENKAT)
-		rm -f $(LIB_SH) $(LIB_ST) kat-argon2* 
+		rm -f $(LIB_SH) $(LIB_ST) kat-argon2*
 		rm -f testcase
 		rm -rf *.dSYM
 		cd src/ && rm -f *.o
@@ -154,8 +172,16 @@
 
 install: $(RUN) libs
 	$(INSTALL) -d $(INST_INCLUDE)
-	$(INSTALL) $(HEADERS) $(INST_INCLUDE)
+	$(INSTALL) -m 0644 $(HEADERS) $(INST_INCLUDE)
 	$(INSTALL) -d $(INST_LIBRARY)
 	$(INSTALL) $(LIBRARIES) $(INST_LIBRARY)
+ifdef LINKED_LIB_SH
+	cd $(INST_LIBRARY) && ln -s $(notdir $(LIB_SH) $(LINKED_LIB_SH))
+endif
 	$(INSTALL) -d $(INST_BINARY)
 	$(INSTALL) $(RUN) $(INST_BINARY)
+
+uninstall:
+	cd $(INST_INCLUDE) && rm -f $(notdir $(HEADERS))
+	cd $(INST_LIBRARY) && rm -f $(notdir $(LIBRARIES) $(LINKED_LIB_SH))
+	cd $(INST_BINARY) && rm -f $(notdir $(RUN))
diff -Nru argon2-0~20161029/man/argon2.1 argon2-0~20171227/man/argon2.1
--- argon2-0~20161029/man/argon2.1	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/man/argon2.1	2017-12-27 19:56:31.000000000 +0000
@@ -45,6 +45,9 @@
 .TP
 .B \-r
 Output only the raw bytes of the hash
+.TP
+.B \-v (10|13)
+Argon2 version (defaults to the most recent version, currently 13)
 
 .SH COPYRIGHT
 This manpage was written by \fBDaniel Kahn Gillmor\fR for the Debian
diff -Nru argon2-0~20161029/README.md argon2-0~20171227/README.md
--- argon2-0~20161029/README.md	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/README.md	2017-12-27 19:56:31.000000000 +0000
@@ -54,7 +54,7 @@
 on your system. To show usage instructions, run
 `./argon2 -h` as
 ```
-Usage:  ./argon2 [-h] salt [-i|-d|-id] [-t iterations] [-m memory] [-p parallelism] [-l hash length] [-e|-r]
+Usage:  ./argon2 [-h] salt [-i|-d|-id] [-t iterations] [-m memory] [-p parallelism] [-l hash length] [-e|-r] [-v (10|13)]
         Password is read from stdin
 Parameters:
         salt            The salt to use, at least 8 characters
@@ -67,6 +67,7 @@
         -l N            Sets hash output length to N bytes (default 32)
         -e              Output only encoded hash
         -r              Output only the raw bytes of the hash
+        -v (10|13)      Argon2 version (defaults to the most recent version, currently 13)
         -h              Print argon2 usage
 ```
 For example, to hash "password" using "somesalt" as a salt and doing 2
@@ -255,13 +256,17 @@
 * [OCaml](https://github.com/Khady/ocaml-argon2) by [@Khady](https://github.com/Khady)
 * [Python (native)](https://pypi.python.org/pypi/argon2), by [@flamewow](https://github.com/flamewow)
 * [Python (ffi)](https://pypi.python.org/pypi/argon2_cffi), by [@hynek](https://github.com/hynek)
+* [Python (ffi, with keyed hashing)](https://github.com/thusoy/porridge), by [@thusoy](https://github.com/thusoy)
+* [R](https://cran.r-project.org/package=argon2) by [@wrathematics](https://github.com/wrathematics)
 * [Ruby](https://github.com/technion/ruby-argon2) by [@technion](https://github.com/technion)
 * [Rust](https://github.com/quininer/argon2-rs) by [@quininer](https://github.com/quininer)
 * [C#/.NET CoreCLR](https://github.com/kmaragon/Konscious.Security.Cryptography) by [@kmaragon](https://github.com/kmaragon)
 * [Perl](https://github.com/Leont/crypt-argon2) by [@leont](https://github.com/Leont)
+* [mruby](https://github.com/Asmod4n/mruby-argon2) by [@Asmod4n](https://github.com/Asmod4n)
+* [Swift](https://github.com/ImKcat/CatCrypto) by [@ImKcat](https://github.com/ImKcat)
 
 
-## Test Suite
+## Test suite
 
 There are two sets of test suites. One is a low level test for the hash
 function, the other tests the higher level API. Both of these are built and
diff -Nru argon2-0~20161029/src/argon2.c argon2-0~20171227/src/argon2.c
--- argon2-0~20161029/src/argon2.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/argon2.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -72,6 +72,10 @@
     instance.threads = context->threads;
     instance.type = type;
 
+    if (instance.threads > instance.lanes) {
+        instance.threads = instance.lanes;
+    }
+
     /* 3. Initialization: Hashing inputs, allocating memory, filling first
      * blocks
      */
@@ -104,6 +108,14 @@
     int result;
     uint8_t *out;
 
+    if (pwdlen > ARGON2_MAX_PWD_LENGTH) {
+        return ARGON2_PWD_TOO_LONG;
+    }
+
+    if (saltlen > ARGON2_MAX_SALT_LENGTH) {
+        return ARGON2_SALT_TOO_LONG;
+    }
+
     if (hashlen > ARGON2_MAX_OUTLEN) {
         return ARGON2_OUTPUT_TOO_LONG;
     }
@@ -245,6 +257,10 @@
     size_t encoded_len;
     uint32_t max_field_len;
 
+    if (pwdlen > ARGON2_MAX_PWD_LENGTH) {
+        return ARGON2_PWD_TOO_LONG;
+    }
+
     if (encoded == NULL) {
         return ARGON2_DECODING_FAIL;
     }
@@ -268,7 +284,7 @@
     }
 
     ctx.pwd = (uint8_t *)pwd;
-    ctx.pwdlen = pwdlen;
+    ctx.pwdlen = (uint32_t)pwdlen;
 
     ret = decode_string(&ctx, encoded, type);
     if (ret != ARGON2_OK) {
diff -Nru argon2-0~20161029/src/bench.c argon2-0~20171227/src/bench.c
--- argon2-0~20161029/src/bench.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/bench.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -45,7 +45,7 @@
 }
 
 /*
- * Benchmarks Argon2 with salt length 16, password length 16, t_cost 1,
+ * Benchmarks Argon2 with salt length 16, password length 16, t_cost 3,
    and different m_cost and threads
  */
 static void benchmark() {
diff -Nru argon2-0~20161029/src/blake2/blake2b.c argon2-0~20171227/src/blake2/blake2b.c
--- argon2-0~20161029/src/blake2/blake2b.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/blake2/blake2b.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
diff -Nru argon2-0~20161029/src/blake2/blake2.h argon2-0~20171227/src/blake2/blake2.h
--- argon2-0~20161029/src/blake2/blake2.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/blake2/blake2.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -18,9 +18,7 @@
 #ifndef PORTABLE_BLAKE2_H
 #define PORTABLE_BLAKE2_H
 
-#include <stddef.h>
-#include <stdint.h>
-#include <limits.h>
+#include <argon2.h>
 
 #if defined(__cplusplus)
 extern "C" {
@@ -69,19 +67,19 @@
 };
 
 /* Streaming API */
-int blake2b_init(blake2b_state *S, size_t outlen);
-int blake2b_init_key(blake2b_state *S, size_t outlen, const void *key,
+ARGON2_LOCAL int blake2b_init(blake2b_state *S, size_t outlen);
+ARGON2_LOCAL int blake2b_init_key(blake2b_state *S, size_t outlen, const void *key,
                      size_t keylen);
-int blake2b_init_param(blake2b_state *S, const blake2b_param *P);
-int blake2b_update(blake2b_state *S, const void *in, size_t inlen);
-int blake2b_final(blake2b_state *S, void *out, size_t outlen);
+ARGON2_LOCAL int blake2b_init_param(blake2b_state *S, const blake2b_param *P);
+ARGON2_LOCAL int blake2b_update(blake2b_state *S, const void *in, size_t inlen);
+ARGON2_LOCAL int blake2b_final(blake2b_state *S, void *out, size_t outlen);
 
 /* Simple API */
-int blake2b(void *out, size_t outlen, const void *in, size_t inlen,
-            const void *key, size_t keylen);
+ARGON2_LOCAL int blake2b(void *out, size_t outlen, const void *in, size_t inlen,
+                         const void *key, size_t keylen);
 
 /* Argon2 Team - Begin Code */
-int blake2b_long(void *out, size_t outlen, const void *in, size_t inlen);
+ARGON2_LOCAL int blake2b_long(void *out, size_t outlen, const void *in, size_t inlen);
 /* Argon2 Team - End Code */
 
 #if defined(__cplusplus)
diff -Nru argon2-0~20161029/src/blake2/blake2-impl.h argon2-0~20171227/src/blake2/blake2-impl.h
--- argon2-0~20161029/src/blake2/blake2-impl.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/blake2/blake2-impl.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
diff -Nru argon2-0~20161029/src/blake2/blamka-round-opt.h argon2-0~20171227/src/blake2/blamka-round-opt.h
--- argon2-0~20161029/src/blake2/blamka-round-opt.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/blake2/blamka-round-opt.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -29,6 +29,8 @@
 #include <x86intrin.h>
 #endif
 
+#if !defined(__AVX512F__)
+#if !defined(__AVX2__)
 #if !defined(__XOP__)
 #if defined(__SSSE3__)
 #define r16                                                                    \
@@ -176,5 +178,294 @@
                                                                                \
         UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1);                         \
     } while ((void)0, 0)
+#else /* __AVX2__ */
 
-#endif
+#include <immintrin.h>
+
+#define rotr32(x)   _mm256_shuffle_epi32(x, _MM_SHUFFLE(2, 3, 0, 1))
+#define rotr24(x)   _mm256_shuffle_epi8(x, _mm256_setr_epi8(3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10, 3, 4, 5, 6, 7, 0, 1, 2, 11, 12, 13, 14, 15, 8, 9, 10))
+#define rotr16(x)   _mm256_shuffle_epi8(x, _mm256_setr_epi8(2, 3, 4, 5, 6, 7, 0, 1, 10, 11, 12, 13, 14, 15, 8, 9, 2, 3, 4, 5, 6, 7, 0, 1, 10, 11, 12, 13, 14, 15, 8, 9))
+#define rotr63(x)   _mm256_xor_si256(_mm256_srli_epi64((x), 63), _mm256_add_epi64((x), (x)))
+
+#define G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+    do { \
+        __m256i ml = _mm256_mul_epu32(A0, B0); \
+        ml = _mm256_add_epi64(ml, ml); \
+        A0 = _mm256_add_epi64(A0, _mm256_add_epi64(B0, ml)); \
+        D0 = _mm256_xor_si256(D0, A0); \
+        D0 = rotr32(D0); \
+        \
+        ml = _mm256_mul_epu32(C0, D0); \
+        ml = _mm256_add_epi64(ml, ml); \
+        C0 = _mm256_add_epi64(C0, _mm256_add_epi64(D0, ml)); \
+        \
+        B0 = _mm256_xor_si256(B0, C0); \
+        B0 = rotr24(B0); \
+        \
+        ml = _mm256_mul_epu32(A1, B1); \
+        ml = _mm256_add_epi64(ml, ml); \
+        A1 = _mm256_add_epi64(A1, _mm256_add_epi64(B1, ml)); \
+        D1 = _mm256_xor_si256(D1, A1); \
+        D1 = rotr32(D1); \
+        \
+        ml = _mm256_mul_epu32(C1, D1); \
+        ml = _mm256_add_epi64(ml, ml); \
+        C1 = _mm256_add_epi64(C1, _mm256_add_epi64(D1, ml)); \
+        \
+        B1 = _mm256_xor_si256(B1, C1); \
+        B1 = rotr24(B1); \
+    } while((void)0, 0);
+
+#define G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+    do { \
+        __m256i ml = _mm256_mul_epu32(A0, B0); \
+        ml = _mm256_add_epi64(ml, ml); \
+        A0 = _mm256_add_epi64(A0, _mm256_add_epi64(B0, ml)); \
+        D0 = _mm256_xor_si256(D0, A0); \
+        D0 = rotr16(D0); \
+        \
+        ml = _mm256_mul_epu32(C0, D0); \
+        ml = _mm256_add_epi64(ml, ml); \
+        C0 = _mm256_add_epi64(C0, _mm256_add_epi64(D0, ml)); \
+        B0 = _mm256_xor_si256(B0, C0); \
+        B0 = rotr63(B0); \
+        \
+        ml = _mm256_mul_epu32(A1, B1); \
+        ml = _mm256_add_epi64(ml, ml); \
+        A1 = _mm256_add_epi64(A1, _mm256_add_epi64(B1, ml)); \
+        D1 = _mm256_xor_si256(D1, A1); \
+        D1 = rotr16(D1); \
+        \
+        ml = _mm256_mul_epu32(C1, D1); \
+        ml = _mm256_add_epi64(ml, ml); \
+        C1 = _mm256_add_epi64(C1, _mm256_add_epi64(D1, ml)); \
+        B1 = _mm256_xor_si256(B1, C1); \
+        B1 = rotr63(B1); \
+    } while((void)0, 0);
+
+#define DIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \
+    do { \
+        B0 = _mm256_permute4x64_epi64(B0, _MM_SHUFFLE(0, 3, 2, 1)); \
+        C0 = _mm256_permute4x64_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
+        D0 = _mm256_permute4x64_epi64(D0, _MM_SHUFFLE(2, 1, 0, 3)); \
+        \
+        B1 = _mm256_permute4x64_epi64(B1, _MM_SHUFFLE(0, 3, 2, 1)); \
+        C1 = _mm256_permute4x64_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
+        D1 = _mm256_permute4x64_epi64(D1, _MM_SHUFFLE(2, 1, 0, 3)); \
+    } while((void)0, 0);
+
+#define DIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \
+    do { \
+        __m256i tmp1 = _mm256_blend_epi32(B0, B1, 0xCC); \
+        __m256i tmp2 = _mm256_blend_epi32(B0, B1, 0x33); \
+        B1 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
+        B0 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
+        \
+        tmp1 = C0; \
+        C0 = C1; \
+        C1 = tmp1; \
+        \
+        tmp1 = _mm256_blend_epi32(D0, D1, 0xCC); \
+        tmp2 = _mm256_blend_epi32(D0, D1, 0x33); \
+        D0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
+        D1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
+    } while(0);
+
+#define UNDIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \
+    do { \
+        B0 = _mm256_permute4x64_epi64(B0, _MM_SHUFFLE(2, 1, 0, 3)); \
+        C0 = _mm256_permute4x64_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
+        D0 = _mm256_permute4x64_epi64(D0, _MM_SHUFFLE(0, 3, 2, 1)); \
+        \
+        B1 = _mm256_permute4x64_epi64(B1, _MM_SHUFFLE(2, 1, 0, 3)); \
+        C1 = _mm256_permute4x64_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
+        D1 = _mm256_permute4x64_epi64(D1, _MM_SHUFFLE(0, 3, 2, 1)); \
+    } while((void)0, 0);
+
+#define UNDIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \
+    do { \
+        __m256i tmp1 = _mm256_blend_epi32(B0, B1, 0xCC); \
+        __m256i tmp2 = _mm256_blend_epi32(B0, B1, 0x33); \
+        B0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
+        B1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
+        \
+        tmp1 = C0; \
+        C0 = C1; \
+        C1 = tmp1; \
+        \
+        tmp1 = _mm256_blend_epi32(D0, D1, 0x33); \
+        tmp2 = _mm256_blend_epi32(D0, D1, 0xCC); \
+        D0 = _mm256_permute4x64_epi64(tmp1, _MM_SHUFFLE(2,3,0,1)); \
+        D1 = _mm256_permute4x64_epi64(tmp2, _MM_SHUFFLE(2,3,0,1)); \
+    } while((void)0, 0);
+
+#define BLAKE2_ROUND_1(A0, A1, B0, B1, C0, C1, D0, D1) \
+    do{ \
+        G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        \
+        DIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \
+        \
+        G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        \
+        UNDIAGONALIZE_1(A0, B0, C0, D0, A1, B1, C1, D1) \
+    } while((void)0, 0);
+
+#define BLAKE2_ROUND_2(A0, A1, B0, B1, C0, C1, D0, D1) \
+    do{ \
+        G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        \
+        DIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        \
+        G1_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        G2_AVX2(A0, A1, B0, B1, C0, C1, D0, D1) \
+        \
+        UNDIAGONALIZE_2(A0, A1, B0, B1, C0, C1, D0, D1) \
+    } while((void)0, 0);
+
+#endif /* __AVX2__ */
+
+#else /* __AVX512F__ */
+
+#include <immintrin.h>
+
+#define ror64(x, n) _mm512_ror_epi64((x), (n))
+
+static __m512i muladd(__m512i x, __m512i y)
+{
+    __m512i z = _mm512_mul_epu32(x, y);
+    return _mm512_add_epi64(_mm512_add_epi64(x, y), _mm512_add_epi64(z, z));
+}
+
+#define G1(A0, B0, C0, D0, A1, B1, C1, D1) \
+    do { \
+        A0 = muladd(A0, B0); \
+        A1 = muladd(A1, B1); \
+\
+        D0 = _mm512_xor_si512(D0, A0); \
+        D1 = _mm512_xor_si512(D1, A1); \
+\
+        D0 = ror64(D0, 32); \
+        D1 = ror64(D1, 32); \
+\
+        C0 = muladd(C0, D0); \
+        C1 = muladd(C1, D1); \
+\
+        B0 = _mm512_xor_si512(B0, C0); \
+        B1 = _mm512_xor_si512(B1, C1); \
+\
+        B0 = ror64(B0, 24); \
+        B1 = ror64(B1, 24); \
+    } while ((void)0, 0)
+
+#define G2(A0, B0, C0, D0, A1, B1, C1, D1) \
+    do { \
+        A0 = muladd(A0, B0); \
+        A1 = muladd(A1, B1); \
+\
+        D0 = _mm512_xor_si512(D0, A0); \
+        D1 = _mm512_xor_si512(D1, A1); \
+\
+        D0 = ror64(D0, 16); \
+        D1 = ror64(D1, 16); \
+\
+        C0 = muladd(C0, D0); \
+        C1 = muladd(C1, D1); \
+\
+        B0 = _mm512_xor_si512(B0, C0); \
+        B1 = _mm512_xor_si512(B1, C1); \
+\
+        B0 = ror64(B0, 63); \
+        B1 = ror64(B1, 63); \
+    } while ((void)0, 0)
+
+#define DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
+    do { \
+        B0 = _mm512_permutex_epi64(B0, _MM_SHUFFLE(0, 3, 2, 1)); \
+        B1 = _mm512_permutex_epi64(B1, _MM_SHUFFLE(0, 3, 2, 1)); \
+\
+        C0 = _mm512_permutex_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
+        C1 = _mm512_permutex_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
+\
+        D0 = _mm512_permutex_epi64(D0, _MM_SHUFFLE(2, 1, 0, 3)); \
+        D1 = _mm512_permutex_epi64(D1, _MM_SHUFFLE(2, 1, 0, 3)); \
+    } while ((void)0, 0)
+
+#define UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1) \
+    do { \
+        B0 = _mm512_permutex_epi64(B0, _MM_SHUFFLE(2, 1, 0, 3)); \
+        B1 = _mm512_permutex_epi64(B1, _MM_SHUFFLE(2, 1, 0, 3)); \
+\
+        C0 = _mm512_permutex_epi64(C0, _MM_SHUFFLE(1, 0, 3, 2)); \
+        C1 = _mm512_permutex_epi64(C1, _MM_SHUFFLE(1, 0, 3, 2)); \
+\
+        D0 = _mm512_permutex_epi64(D0, _MM_SHUFFLE(0, 3, 2, 1)); \
+        D1 = _mm512_permutex_epi64(D1, _MM_SHUFFLE(0, 3, 2, 1)); \
+    } while ((void)0, 0)
+
+#define BLAKE2_ROUND(A0, B0, C0, D0, A1, B1, C1, D1) \
+    do { \
+        G1(A0, B0, C0, D0, A1, B1, C1, D1); \
+        G2(A0, B0, C0, D0, A1, B1, C1, D1); \
+\
+        DIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1); \
+\
+        G1(A0, B0, C0, D0, A1, B1, C1, D1); \
+        G2(A0, B0, C0, D0, A1, B1, C1, D1); \
+\
+        UNDIAGONALIZE(A0, B0, C0, D0, A1, B1, C1, D1); \
+    } while ((void)0, 0)
+
+#define SWAP_HALVES(A0, A1) \
+    do { \
+        __m512i t0, t1; \
+        t0 = _mm512_shuffle_i64x2(A0, A1, _MM_SHUFFLE(1, 0, 1, 0)); \
+        t1 = _mm512_shuffle_i64x2(A0, A1, _MM_SHUFFLE(3, 2, 3, 2)); \
+        A0 = t0; \
+        A1 = t1; \
+    } while((void)0, 0)
+
+#define SWAP_QUARTERS(A0, A1) \
+    do { \
+        SWAP_HALVES(A0, A1); \
+        A0 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A0); \
+        A1 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A1); \
+    } while((void)0, 0)
+
+#define UNSWAP_QUARTERS(A0, A1) \
+    do { \
+        A0 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A0); \
+        A1 = _mm512_permutexvar_epi64(_mm512_setr_epi64(0, 1, 4, 5, 2, 3, 6, 7), A1); \
+        SWAP_HALVES(A0, A1); \
+    } while((void)0, 0)
+
+#define BLAKE2_ROUND_1(A0, C0, B0, D0, A1, C1, B1, D1) \
+    do { \
+        SWAP_HALVES(A0, B0); \
+        SWAP_HALVES(C0, D0); \
+        SWAP_HALVES(A1, B1); \
+        SWAP_HALVES(C1, D1); \
+        BLAKE2_ROUND(A0, B0, C0, D0, A1, B1, C1, D1); \
+        SWAP_HALVES(A0, B0); \
+        SWAP_HALVES(C0, D0); \
+        SWAP_HALVES(A1, B1); \
+        SWAP_HALVES(C1, D1); \
+    } while ((void)0, 0)
+
+#define BLAKE2_ROUND_2(A0, A1, B0, B1, C0, C1, D0, D1) \
+    do { \
+        SWAP_QUARTERS(A0, A1); \
+        SWAP_QUARTERS(B0, B1); \
+        SWAP_QUARTERS(C0, C1); \
+        SWAP_QUARTERS(D0, D1); \
+        BLAKE2_ROUND(A0, B0, C0, D0, A1, B1, C1, D1); \
+        UNSWAP_QUARTERS(A0, A1); \
+        UNSWAP_QUARTERS(B0, B1); \
+        UNSWAP_QUARTERS(C0, C1); \
+        UNSWAP_QUARTERS(D0, D1); \
+    } while ((void)0, 0)
+
+#endif /* __AVX512F__ */
+#endif /* BLAKE_ROUND_MKA_OPT_H */
diff -Nru argon2-0~20161029/src/blake2/blamka-round-ref.h argon2-0~20171227/src/blake2/blamka-round-ref.h
--- argon2-0~20161029/src/blake2/blamka-round-ref.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/blake2/blamka-round-ref.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -21,7 +21,7 @@
 #include "blake2.h"
 #include "blake2-impl.h"
 
-/*designed by the Lyra PHC team */
+/* designed by the Lyra PHC team */
 static BLAKE2_INLINE uint64_t fBlaMka(uint64_t x, uint64_t y) {
     const uint64_t m = UINT64_C(0xFFFFFFFF);
     const uint64_t xy = (x & m) * (y & m);
diff -Nru argon2-0~20161029/src/core.c argon2-0~20171227/src/core.c
--- argon2-0~20161029/src/core.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/core.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -25,7 +25,6 @@
 #endif
 #define VC_GE_2005(version) (version >= 1400)
 
-#include <inttypes.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -246,6 +245,26 @@
     return absolute_position;
 }
 
+/* Single-threaded version for p=1 case */
+static int fill_memory_blocks_st(argon2_instance_t *instance) {
+    uint32_t r, s, l;
+
+    for (r = 0; r < instance->passes; ++r) {
+        for (s = 0; s < ARGON2_SYNC_POINTS; ++s) {
+            for (l = 0; l < instance->lanes; ++l) {
+                argon2_position_t position = {r, l, (uint8_t)s, 0};
+                fill_segment(instance, position);
+            }
+        }
+#ifdef GENKAT
+        internal_kat(instance, r); /* Print all memory blocks */
+#endif
+    }
+    return ARGON2_OK;
+}
+
+#if !defined(ARGON2_NO_THREADS)
+
 #ifdef _WIN32
 static unsigned __stdcall fill_segment_thr(void *thread_data)
 #else
@@ -258,17 +277,13 @@
     return 0;
 }
 
-int fill_memory_blocks(argon2_instance_t *instance) {
+/* Multi-threaded version for p > 1 case */
+static int fill_memory_blocks_mt(argon2_instance_t *instance) {
     uint32_t r, s;
     argon2_thread_handle_t *thread = NULL;
     argon2_thread_data *thr_data = NULL;
     int rc = ARGON2_OK;
 
-    if (instance == NULL || instance->lanes == 0) {
-        rc = ARGON2_THREAD_FAIL;
-        goto fail;
-    }
-
     /* 1. Allocating space for threads */
     thread = calloc(instance->lanes, sizeof(argon2_thread_handle_t));
     if (thread == NULL) {
@@ -342,6 +357,20 @@
     return rc;
 }
 
+#endif /* ARGON2_NO_THREADS */
+
+int fill_memory_blocks(argon2_instance_t *instance) {
+	if (instance == NULL || instance->lanes == 0) {
+	    return ARGON2_INCORRECT_PARAMETER;
+    }
+#if defined(ARGON2_NO_THREADS)
+    return fill_memory_blocks_st(instance);
+#else
+    return instance->threads == 1 ?
+			fill_memory_blocks_st(instance) : fill_memory_blocks_mt(instance);
+#endif
+}
+
 int validate_inputs(const argon2_context *context) {
     if (NULL == context) {
         return ARGON2_INCORRECT_PARAMETER;
@@ -471,8 +500,8 @@
 
 void fill_first_blocks(uint8_t *blockhash, const argon2_instance_t *instance) {
     uint32_t l;
-    /* Make the first and second block in each lane as G(H0||i||0) or
-       G(H0||i||1) */
+    /* Make the first and second block in each lane as G(H0||0||i) or
+       G(H0||1||i) */
     uint8_t blockhash_bytes[ARGON2_BLOCK_SIZE];
     for (l = 0; l < instance->lanes; ++l) {
 
diff -Nru argon2-0~20161029/src/core.h argon2-0~20171227/src/core.h
--- argon2-0~20161029/src/core.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/core.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -20,14 +20,6 @@
 
 #include "argon2.h"
 
-#if defined(_MSC_VER)
-#define ALIGN(n) __declspec(align(16))
-#elif defined(__GNUC__) || defined(__clang)
-#define ALIGN(x) __attribute__((__aligned__(x)))
-#else
-#define ALIGN(x)
-#endif
-
 #define CONST_CAST(x) (x)(uintptr_t)
 
 /**********************Argon2 internal constants*******************************/
@@ -37,6 +29,8 @@
     ARGON2_BLOCK_SIZE = 1024,
     ARGON2_QWORDS_IN_BLOCK = ARGON2_BLOCK_SIZE / 8,
     ARGON2_OWORDS_IN_BLOCK = ARGON2_BLOCK_SIZE / 16,
+    ARGON2_HWORDS_IN_BLOCK = ARGON2_BLOCK_SIZE / 32,
+    ARGON2_512BIT_WORDS_IN_BLOCK = ARGON2_BLOCK_SIZE / 64,
 
     /* Number of pseudo-random values generated by one call to Blake in Argon2i
        to
diff -Nru argon2-0~20161029/src/encoding.c argon2-0~20171227/src/encoding.c
--- argon2-0~20161029/src/encoding.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/encoding.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -288,6 +288,19 @@
         (x) = dec_x;                                                           \
     } while ((void)0, 0)
 
+
+/* Decoding prefix into uint32_t decimal */
+#define DECIMAL_U32(x)                                                         \
+    do {                                                                       \
+        unsigned long dec_x;                                                   \
+        str = decode_decimal(str, &dec_x);                                     \
+        if (str == NULL || dec_x > UINT32_MAX) {                               \
+            return ARGON2_DECODING_FAIL;                                       \
+        }                                                                      \
+        (x) = (uint32_t)dec_x;                                                 \
+    } while ((void)0, 0)
+
+
 /* Decoding base64 into a binary buffer */
 #define BIN(buf, max_len, len)                                                 \
     do {                                                                       \
@@ -315,14 +328,14 @@
 
     /* Reading the version number if the default is suppressed */
     ctx->version = ARGON2_VERSION_10;
-    CC_opt("$v=", DECIMAL(ctx->version));
+    CC_opt("$v=", DECIMAL_U32(ctx->version));
 
     CC("$m=");
-    DECIMAL(ctx->m_cost);
+    DECIMAL_U32(ctx->m_cost);
     CC(",t=");
-    DECIMAL(ctx->t_cost);
+    DECIMAL_U32(ctx->t_cost);
     CC(",p=");
-    DECIMAL(ctx->lanes);
+    DECIMAL_U32(ctx->lanes);
     ctx->threads = ctx->lanes;
 
     CC("$");
diff -Nru argon2-0~20161029/src/encoding.h argon2-0~20171227/src/encoding.h
--- argon2-0~20161029/src/encoding.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/encoding.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
diff -Nru argon2-0~20161029/src/genkat.c argon2-0~20171227/src/genkat.c
--- argon2-0~20161029/src/genkat.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/genkat.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -15,7 +15,6 @@
  * software. If not, they may be obtained at the above URLs.
  */
 
-#include <inttypes.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -116,8 +115,8 @@
                     : ARGON2_QWORDS_IN_BLOCK;
 
             for (j = 0; j < how_many_words; ++j)
-                printf("Block %.4u [%3u]: %016" PRIx64 "\n", i, j,
-                       instance->memory[i].v[j]);
+                printf("Block %.4u [%3u]: %016llx\n", i, j,
+                       (unsigned long long)instance->memory[i].v[j]);
         }
     }
 }
diff -Nru argon2-0~20161029/src/genkat.h argon2-0~20171227/src/genkat.h
--- argon2-0~20161029/src/genkat.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/genkat.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
diff -Nru argon2-0~20161029/src/opt.c argon2-0~20171227/src/opt.c
--- argon2-0~20161029/src/opt.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/opt.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -20,13 +20,95 @@
 #include <stdlib.h>
 
 #include "argon2.h"
-#include "opt.h"
+#include "core.h"
 
 #include "blake2/blake2.h"
 #include "blake2/blamka-round-opt.h"
 
-void fill_block(__m128i *state, const block *ref_block, block *next_block,
-                int with_xor) {
+/*
+ * Function fills a new memory block and optionally XORs the old block over the new one.
+ * Memory must be initialized.
+ * @param state Pointer to the just produced block. Content will be updated(!)
+ * @param ref_block Pointer to the reference block
+ * @param next_block Pointer to the block to be XORed over. May coincide with @ref_block
+ * @param with_xor Whether to XOR into the new block (1) or just overwrite (0)
+ * @pre all block pointers must be valid
+ */
+#if defined(__AVX512F__)
+static void fill_block(__m512i *state, const block *ref_block,
+                       block *next_block, int with_xor) {
+    __m512i block_XY[ARGON2_512BIT_WORDS_IN_BLOCK];
+    unsigned int i;
+
+    if (with_xor) {
+        for (i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++) {
+            state[i] = _mm512_xor_si512(
+                state[i], _mm512_loadu_si512((const __m512i *)ref_block->v + i));
+            block_XY[i] = _mm512_xor_si512(
+                state[i], _mm512_loadu_si512((const __m512i *)next_block->v + i));
+        }
+    } else {
+        for (i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++) {
+            block_XY[i] = state[i] = _mm512_xor_si512(
+                state[i], _mm512_loadu_si512((const __m512i *)ref_block->v + i));
+        }
+    }
+
+    for (i = 0; i < 2; ++i) {
+        BLAKE2_ROUND_1(
+            state[8 * i + 0], state[8 * i + 1], state[8 * i + 2], state[8 * i + 3],
+            state[8 * i + 4], state[8 * i + 5], state[8 * i + 6], state[8 * i + 7]);
+    }
+
+    for (i = 0; i < 2; ++i) {
+        BLAKE2_ROUND_2(
+            state[2 * 0 + i], state[2 * 1 + i], state[2 * 2 + i], state[2 * 3 + i],
+            state[2 * 4 + i], state[2 * 5 + i], state[2 * 6 + i], state[2 * 7 + i]);
+    }
+
+    for (i = 0; i < ARGON2_512BIT_WORDS_IN_BLOCK; i++) {
+        state[i] = _mm512_xor_si512(state[i], block_XY[i]);
+        _mm512_storeu_si512((__m512i *)next_block->v + i, state[i]);
+    }
+}
+#elif defined(__AVX2__)
+static void fill_block(__m256i *state, const block *ref_block,
+                       block *next_block, int with_xor) {
+    __m256i block_XY[ARGON2_HWORDS_IN_BLOCK];
+    unsigned int i;
+
+    if (with_xor) {
+        for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) {
+            state[i] = _mm256_xor_si256(
+                state[i], _mm256_loadu_si256((const __m256i *)ref_block->v + i));
+            block_XY[i] = _mm256_xor_si256(
+                state[i], _mm256_loadu_si256((const __m256i *)next_block->v + i));
+        }
+    } else {
+        for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) {
+            block_XY[i] = state[i] = _mm256_xor_si256(
+                state[i], _mm256_loadu_si256((const __m256i *)ref_block->v + i));
+        }
+    }
+
+    for (i = 0; i < 4; ++i) {
+        BLAKE2_ROUND_1(state[8 * i + 0], state[8 * i + 4], state[8 * i + 1], state[8 * i + 5],
+                       state[8 * i + 2], state[8 * i + 6], state[8 * i + 3], state[8 * i + 7]);
+    }
+
+    for (i = 0; i < 4; ++i) {
+        BLAKE2_ROUND_2(state[ 0 + i], state[ 4 + i], state[ 8 + i], state[12 + i],
+                       state[16 + i], state[20 + i], state[24 + i], state[28 + i]);
+    }
+
+    for (i = 0; i < ARGON2_HWORDS_IN_BLOCK; i++) {
+        state[i] = _mm256_xor_si256(state[i], block_XY[i]);
+        _mm256_storeu_si256((__m256i *)next_block->v + i, state[i]);
+    }
+}
+#else
+static void fill_block(__m128i *state, const block *ref_block,
+                       block *next_block, int with_xor) {
     __m128i block_XY[ARGON2_OWORDS_IN_BLOCK];
     unsigned int i;
 
@@ -61,11 +143,20 @@
         _mm_storeu_si128((__m128i *)next_block->v + i, state[i]);
     }
 }
+#endif
 
 static void next_addresses(block *address_block, block *input_block) {
     /*Temporary zero-initialized blocks*/
+#if defined(__AVX512F__)
+    __m512i zero_block[ARGON2_512BIT_WORDS_IN_BLOCK];
+    __m512i zero2_block[ARGON2_512BIT_WORDS_IN_BLOCK];
+#elif defined(__AVX2__)
+    __m256i zero_block[ARGON2_HWORDS_IN_BLOCK];
+    __m256i zero2_block[ARGON2_HWORDS_IN_BLOCK];
+#else
     __m128i zero_block[ARGON2_OWORDS_IN_BLOCK];
     __m128i zero2_block[ARGON2_OWORDS_IN_BLOCK];
+#endif
 
     memset(zero_block, 0, sizeof(zero_block));
     memset(zero2_block, 0, sizeof(zero2_block));
@@ -87,7 +178,13 @@
     uint64_t pseudo_rand, ref_index, ref_lane;
     uint32_t prev_offset, curr_offset;
     uint32_t starting_index, i;
-    __m128i state[64];
+#if defined(__AVX512F__)
+    __m512i state[ARGON2_512BIT_WORDS_IN_BLOCK];
+#elif defined(__AVX2__)
+    __m256i state[ARGON2_HWORDS_IN_BLOCK];
+#else
+    __m128i state[ARGON2_OWORDS_IN_BLOCK];
+#endif
     int data_independent_addressing;
 
     if (instance == NULL) {
diff -Nru argon2-0~20161029/src/opt.h argon2-0~20171227/src/opt.h
--- argon2-0~20161029/src/opt.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/opt.h	1970-01-01 00:00:00.000000000 +0000
@@ -1,35 +0,0 @@
-/*
- * Argon2 reference source code package - reference C implementations
- *
- * Copyright 2015
- * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
- *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
- * License/Waiver or the Apache Public License 2.0, at your option. The terms of
- * these licenses can be found at:
- *
- * - CC0 1.0 Universal : http://creativecommons.org/publicdomain/zero/1.0
- * - Apache 2.0        : http://www.apache.org/licenses/LICENSE-2.0
- *
- * You should have received a copy of both of these licenses along with this
- * software. If not, they may be obtained at the above URLs.
- */
-
-#ifndef ARGON2_OPT_H
-#define ARGON2_OPT_H
-
-#include "core.h"
-#include <emmintrin.h>
-
-/*
- * Function fills a new memory block and optionally XORs the old block over the new one.
- * Memory must be initialized.
- * @param state Pointer to the just produced block. Content will be updated(!)
- * @param ref_block Pointer to the reference block
- * @param next_block Pointer to the block to be XORed over. May coincide with @ref_block
- * @param with_xor Whether to XOR into the new block (1) or just overwrite (0)
- * @pre all block pointers must be valid
- */
-void fill_block(__m128i *s, const block *ref_block, block *next_block, int with_xor);
-
-#endif /* ARGON2_OPT_H */
diff -Nru argon2-0~20161029/src/ref.c argon2-0~20171227/src/ref.c
--- argon2-0~20161029/src/ref.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/ref.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -20,15 +20,24 @@
 #include <stdlib.h>
 
 #include "argon2.h"
-#include "ref.h"
+#include "core.h"
 
 #include "blake2/blamka-round-ref.h"
 #include "blake2/blake2-impl.h"
 #include "blake2/blake2.h"
 
 
-void fill_block(const block *prev_block, const block *ref_block,
-                block *next_block, int with_xor) {
+/*
+ * Function fills a new memory block and optionally XORs the old block over the new one.
+ * @next_block must be initialized.
+ * @param prev_block Pointer to the previous block
+ * @param ref_block Pointer to the reference block
+ * @param next_block Pointer to the block to be constructed
+ * @param with_xor Whether to XOR into the new block (1) or just overwrite (0)
+ * @pre all block pointers must be valid
+ */
+static void fill_block(const block *prev_block, const block *ref_block,
+                       block *next_block, int with_xor) {
     block blockR, block_tmp;
     unsigned i;
 
diff -Nru argon2-0~20161029/src/ref.h argon2-0~20171227/src/ref.h
--- argon2-0~20161029/src/ref.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/ref.h	1970-01-01 00:00:00.000000000 +0000
@@ -1,35 +0,0 @@
-/*
- * Argon2 reference source code package - reference C implementations
- *
- * Copyright 2015
- * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
- *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
- * License/Waiver or the Apache Public License 2.0, at your option. The terms of
- * these licenses can be found at:
- *
- * - CC0 1.0 Universal : http://creativecommons.org/publicdomain/zero/1.0
- * - Apache 2.0        : http://www.apache.org/licenses/LICENSE-2.0
- *
- * You should have received a copy of both of these licenses along with this
- * software. If not, they may be obtained at the above URLs.
- */
-
-#ifndef ARGON2_REF_H
-#define ARGON2_REF_H
-
-#include "core.h"
-
-/*
- * Function fills a new memory block and optionally XORs the old block over the new one.
- * @next_block must be initialized.
- * @param prev_block Pointer to the previous block
- * @param ref_block Pointer to the reference block
- * @param next_block Pointer to the block to be constructed
- * @param with_xor Whether to XOR into the new block (1) or just overwrite (0)
- * @pre all block pointers must be valid
- */
-void fill_block(const block *prev_block, const block *ref_block,
-                block *next_block, int with_xor);
-
-#endif /* ARGON2_REF_H */
diff -Nru argon2-0~20161029/src/run.c argon2-0~20171227/src/run.c
--- argon2-0~20161029/src/run.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/run.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -17,7 +17,6 @@
 
 #define _GNU_SOURCE 1
 
-#include <inttypes.h>
 #include <stdint.h>
 #include <stdio.h>
 #include <stdlib.h>
@@ -37,8 +36,9 @@
 #define UNUSED_PARAMETER(x) (void)(x)
 
 static void usage(const char *cmd) {
-    printf("Usage:  %s [-h] salt [-i|-d|-id] [-t iterations] [-m memory] "
-           "[-p parallelism] [-l hash length] [-e|-r]\n",
+    printf("Usage:  %s [-h] salt [-i|-d|-id] [-t iterations] "
+           "[-m log2(memory in KiB) | -k memory in KiB] [-p parallelism] "
+           "[-l hash length] [-e|-r] [-v (10|13)]\n",
            cmd);
     printf("\tPassword is read from stdin\n");
     printf("Parameters:\n");
@@ -50,12 +50,16 @@
            T_COST_DEF);
     printf("\t-m N\t\tSets the memory usage of 2^N KiB (default %d)\n",
            LOG_M_COST_DEF);
+    printf("\t-k N\t\tSets the memory usage of N KiB (default %d)\n",
+           1 << LOG_M_COST_DEF);
     printf("\t-p N\t\tSets parallelism to N threads (default %d)\n",
            THREADS_DEF);
     printf("\t-l N\t\tSets hash output length to N bytes (default %d)\n",
            OUTLEN_DEF);
     printf("\t-e\t\tOutput only encoded hash\n");
     printf("\t-r\t\tOutput only the raw bytes of the hash\n");
+    printf("\t-v (10|13)\tArgon2 version (defaults to the most recent version, currently %x)\n",
+            ARGON2_VERSION_NUMBER);
     printf("\t-h\t\tPrint %s usage\n", cmd);
 }
 
@@ -85,12 +89,13 @@
 @type Argon2 type we want to run
 @encoded_only display only the encoded hash
 @raw_only display only the hexadecimal of the hash
+@version Argon2 version
 */
-static void run(uint32_t outlen, char *pwd, char *salt, uint32_t t_cost,
+static void run(uint32_t outlen, char *pwd, size_t pwdlen, char *salt, uint32_t t_cost,
                 uint32_t m_cost, uint32_t lanes, uint32_t threads,
-                argon2_type type, int encoded_only, int raw_only) {
+                argon2_type type, int encoded_only, int raw_only, uint32_t version) {
     clock_t start_time, stop_time;
-    size_t pwdlen, saltlen, encodedlen;
+    size_t saltlen, encodedlen;
     int result;
     unsigned char * out = NULL;
     char * encoded = NULL;
@@ -102,11 +107,10 @@
     }
 
     if (!salt) {
-        clear_internal_memory(pwd, strlen(pwd));
+        clear_internal_memory(pwd, pwdlen);
         fatal("salt missing");
     }
 
-    pwdlen = strlen(pwd);
     saltlen = strlen(salt);
     if(UINT32_MAX < saltlen) {
         fatal("salt is too long");
@@ -116,20 +120,20 @@
 
     out = malloc(outlen + 1);
     if (!out) {
-        clear_internal_memory(pwd, strlen(pwd));
+        clear_internal_memory(pwd, pwdlen);
         fatal("could not allocate memory for output");
     }
 
     encodedlen = argon2_encodedlen(t_cost, m_cost, lanes, (uint32_t)saltlen, outlen, type);
     encoded = malloc(encodedlen + 1);
     if (!encoded) {
-        clear_internal_memory(pwd, strlen(pwd));
+        clear_internal_memory(pwd, pwdlen);
         fatal("could not allocate memory for hash");
     }
 
     result = argon2_hash(t_cost, m_cost, threads, pwd, pwdlen, salt, saltlen,
                          out, outlen, encoded, encodedlen, type,
-                         ARGON2_VERSION_NUMBER);
+                         version);
     if (result != ARGON2_OK)
         fatal(argon2_error_message(result));
 
@@ -171,10 +175,12 @@
     uint32_t threads = THREADS_DEF;
     argon2_type type = Argon2_i; /* Argon2i is the default type */
     int types_specified = 0;
+    int m_cost_specified = 0;
     int encoded_only = 0;
     int raw_only = 0;
+    uint32_t version = ARGON2_VERSION_NUMBER;
     int i;
-    size_t n;
+    size_t pwdlen;
     char pwd[MAX_PASS_LEN], *salt;
 
     if (argc < 2) {
@@ -186,19 +192,14 @@
     }
 
     /* get password from stdin */
-    n = fread(pwd, 1, sizeof pwd - 1, stdin);
-    if(n < 1) {
+    pwdlen = fread(pwd, 1, sizeof pwd, stdin);
+    if(pwdlen < 1) {
         fatal("no password read");
     }
-    if(n == MAX_PASS_LEN-1) {
+    if(pwdlen == MAX_PASS_LEN) {
         fatal("Provided password longer than supported in command line utility");
     }
 
-    pwd[n] = '\0';
-    if (pwd[n - 1] == '\n') {
-        pwd[n - 1] = '\0';
-    }
-
     salt = argv[1];
 
     /* parse options */
@@ -209,6 +210,10 @@
             usage(argv[0]);
             return 1;
         } else if (!strcmp(a, "-m")) {
+            if (m_cost_specified) {
+                fatal("-m or -k can only be used once");
+            }
+            m_cost_specified = 1;
             if (i < argc - 1) {
                 i++;
                 input = strtoul(argv[i], NULL, 10);
@@ -224,6 +229,25 @@
             } else {
                 fatal("missing -m argument");
             }
+        } else if (!strcmp(a, "-k")) {
+            if (m_cost_specified) {
+                fatal("-m or -k can only be used once");
+            }
+            m_cost_specified = 1;
+            if (i < argc - 1) {
+                i++;
+                input = strtoul(argv[i], NULL, 10);
+                if (input == 0 || input == ULONG_MAX) {
+                    fatal("bad numeric input for -k");
+                }
+                m_cost = ARGON2_MIN(input, UINT32_C(0xFFFFFFFF));
+                if (m_cost > ARGON2_MAX_MEMORY) {
+                    fatal("m_cost overflow");
+                }
+                continue;
+            } else {
+                fatal("missing -k argument");
+            }
         } else if (!strcmp(a, "-t")) {
             if (i < argc - 1) {
                 i++;
@@ -273,6 +297,19 @@
             encoded_only = 1;
         } else if (!strcmp(a, "-r")) {
             raw_only = 1;
+        } else if (!strcmp(a, "-v")) {
+            if (i < argc - 1) {
+                i++;
+                if (!strcmp(argv[i], "10")) {
+                    version = ARGON2_VERSION_10;
+                } else if (!strcmp(argv[i], "13")) {
+                    version = ARGON2_VERSION_13;
+                } else {
+                    fatal("invalid Argon2 version");
+                }
+            } else {
+                fatal("missing -v argument");
+            }
         } else {
             fatal("unknown argument");
         }
@@ -287,13 +324,13 @@
 
     if(!encoded_only && !raw_only) {
         printf("Type:\t\t%s\n", argon2_type2string(type, 1));
-        printf("Iterations:\t%" PRIu32 " \n", t_cost);
-        printf("Memory:\t\t%" PRIu32 " KiB\n", m_cost);
-        printf("Parallelism:\t%" PRIu32 " \n", lanes);
+        printf("Iterations:\t%u\n", t_cost);
+        printf("Memory:\t\t%u KiB\n", m_cost);
+        printf("Parallelism:\t%u\n", lanes);
     }
 
-    run(outlen, pwd, salt, t_cost, m_cost, lanes, threads, type,
-       encoded_only, raw_only);
+    run(outlen, pwd, pwdlen, salt, t_cost, m_cost, lanes, threads, type,
+       encoded_only, raw_only, version);
 
     return ARGON2_OK;
 }
diff -Nru argon2-0~20161029/src/test.c argon2-0~20171227/src/test.c
--- argon2-0~20161029/src/test.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/test.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -17,7 +17,6 @@
 
 #include <stdio.h>
 #include <stdint.h>
-#include <inttypes.h>
 #include <stdlib.h>
 #include <string.h>
 #include <time.h>
@@ -81,7 +80,7 @@
              "$argon2i$m=65536,t=2,p=1$c29tZXNhbHQ"
              "$9sTbSlTio3Biev89thdrlKKiCaYsjjYVJxGAL3swxpQ");
 #ifdef TEST_LARGE_RAM
-    hashtest(version, 2, 20, 1, "password", "somesalt",  
+    hashtest(version, 2, 20, 1, "password", "somesalt",
             "9690ec55d28d3ed32562f2e73ea62b02b018757643a2ae6e79528459de8106e9",
             "$argon2i$m=1048576,t=2,p=1$c29tZXNhbHQ"
             "$lpDsVdKNPtMlYvLnPqYrArAYdXZDoq5ueVKEWd6BBuk");
@@ -160,10 +159,10 @@
              "$argon2i$v=19$m=65536,t=2,p=1$c29tZXNhbHQ"
              "$wWKIMhR9lyDFvRz9YTZweHKfbftvj+qf+YFY4NeBbtA");
 #ifdef TEST_LARGE_RAM
-    hashtest(version, 2, 20, 1, "password", "somesalt",  
+    hashtest(version, 2, 20, 1, "password", "somesalt",
              "d1587aca0922c3b5d6a83edab31bee3c4ebaef342ed6127a55d19b2351ad1f41",
-             "$argon2i$v=19$m=1048576,t=2,p=1$c29tZXNhbHQ"  
-             "$0Vh6ygkiw7XWqD7asxvuPE667zQu1hJ6VdGbI1GtH0E");  
+             "$argon2i$v=19$m=1048576,t=2,p=1$c29tZXNhbHQ"
+             "$0Vh6ygkiw7XWqD7asxvuPE667zQu1hJ6VdGbI1GtH0E");
 #endif
     hashtest(version, 2, 18, 1, "password", "somesalt",
              "296dbae80b807cdceaad44ae741b506f14db0959267b183b118f9b24229bc7cb",
diff -Nru argon2-0~20161029/src/thread.c argon2-0~20171227/src/thread.c
--- argon2-0~20161029/src/thread.c	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/thread.c	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -15,6 +15,8 @@
  * software. If not, they may be obtained at the above URLs.
  */
 
+#if !defined(ARGON2_NO_THREADS)
+
 #include "thread.h"
 #if defined(_WIN32)
 #include <windows.h>
@@ -51,3 +53,5 @@
     pthread_exit(NULL);
 #endif
 }
+
+#endif /* ARGON2_NO_THREADS */
diff -Nru argon2-0~20161029/src/thread.h argon2-0~20171227/src/thread.h
--- argon2-0~20161029/src/thread.h	2016-10-29 09:24:03.000000000 +0000
+++ argon2-0~20171227/src/thread.h	2017-12-27 19:56:31.000000000 +0000
@@ -4,7 +4,7 @@
  * Copyright 2015
  * Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves
  *
- * You may use this work under the terms of a Creative Commons CC0 1.0 
+ * You may use this work under the terms of a Creative Commons CC0 1.0
  * License/Waiver or the Apache Public License 2.0, at your option. The terms of
  * these licenses can be found at:
  *
@@ -17,6 +17,9 @@
 
 #ifndef ARGON2_THREAD_H
 #define ARGON2_THREAD_H
+
+#if !defined(ARGON2_NO_THREADS)
+
 /*
         Here we implement an abstraction layer for the simpĺe requirements
         of the Argon2 code. We only require 3 primitives---thread creation,
@@ -43,7 +46,7 @@
  * @param func A function pointer for the thread's entry point. Must not be
  * NULL.
  * @param args Pointer that is passed as an argument to @func. May be NULL.
- * @return 0 if @handle and @func are valid pointers and a thread is successfuly
+ * @return 0 if @handle and @func are valid pointers and a thread is successfully
  * created.
  */
 int argon2_thread_create(argon2_thread_handle_t *handle,
@@ -60,4 +63,5 @@
 */
 void argon2_thread_exit(void);
 
+#endif /* ARGON2_NO_THREADS */
 #endif