Lars Gierth 0bf4e4b9cd cnacl: add mips64 plan 9 år sedan
..
commandline 896627850f oops wrong name 11 år sedan
cpucycles 896627850f oops wrong name 11 år sedan
cpuid 896627850f oops wrong name 11 år sedan
crypto_auth 896627850f oops wrong name 11 år sedan
crypto_box 896627850f oops wrong name 11 år sedan
crypto_core 896627850f oops wrong name 11 år sedan
crypto_hash a3203ade2c More algorithms, also remove accidently added directory 9 år sedan
crypto_hashblocks a3203ade2c More algorithms, also remove accidently added directory 9 år sedan
crypto_onetimeauth dd47cf9ca4 merge in the mips32r2 assembly poly1305 implementation 9 år sedan
crypto_scalarmult a3203ade2c More algorithms, also remove accidently added directory 9 år sedan
crypto_secretbox 896627850f oops wrong name 11 år sedan
crypto_sign 896627850f oops wrong name 11 år sedan
crypto_stream a3203ade2c More algorithms, also remove accidently added directory 9 år sedan
crypto_verify 896627850f oops wrong name 11 år sedan
curvecp 896627850f oops wrong name 11 år sedan
inttypes 896627850f oops wrong name 11 år sedan
node_build 0bf4e4b9cd cnacl: add mips64 plan 9 år sedan
okcompilers c9e5d27f7f cnacl: add arm64 plan 9 år sedan
randombytes 8919badedb First new dev effort 10 år sedan
tests b5923f700e Set return type in `main` to remove OS X warnings 10 år sedan
.gitignore 896627850f oops wrong name 11 år sedan
MACROS 896627850f oops wrong name 11 år sedan
OPERATIONS 896627850f oops wrong name 11 år sedan
PROTOTYPES.c 896627850f oops wrong name 11 år sedan
PROTOTYPES.cpp 896627850f oops wrong name 11 år sedan
README.md 896627850f oops wrong name 11 år sedan
do 896627850f oops wrong name 11 år sedan
measure-anything.c 896627850f oops wrong name 11 år sedan
try-anything.c 896627850f oops wrong name 11 år sedan
version 896627850f oops wrong name 11 år sedan

README.md

cNaCl

If you would like to be confusing, you could pronounce it sea-salt

This is a fork NaCl by Daniel J. Bernstein and Tanja Lange. The build has been ported to cmake so it can be cross compiled and build output is reliable. Since it uses cmake, it could theoretically be built on windows but this has not been tested. It does compile using mingw32.

How do I make this thing work?

mkdir cbuild
cd cbuild
cmake ..
make

Ok now how about cross compiling?

mkdir cbuildw32
cd cbuildw32
cmake -DCMAKE_TOOLCHAIN_FILE=../CMakeWindows.txt ..
make

Why fork?

NaCl builds using a shell script called ./do. This script does compiling, testing, measuring and selection of the best implementation of each algorithm for the given machine. It also generates the header files which will be used.

The problems with ./do are it's slow, it tries compiling with multiple different compiler profiles, it's very platform independent but it doesn't run on Windows and most importantly, with compiling, testing and measuring so tightly bound, it is impossible to cross compile for a different operating system.

How it works

The first time you build for a new ABI, it will trigger the traditional nacl ./do script. What cNaCl does is parse the resulting headers from the ./do build and create a plan so that it can repeat roughly the same build.

If there is already a plan for the given ABI, the build uses this plan and the build is very fast.

Plans are stored in ./cmake/plans/ and I will be adding plans as I find new ones.

What else is new?

There is a problem with the ./do build which prevents it from running on some ARM based machines, this was fixed by adding a more lax method for measuring CPU speed as a fall back.

#EOF#