chiraag / gazelle_mpc Goto Github PK

Is there some tests to reproduce the numbers in the paper?

we con't understand some parameters in your code,such as n_circ.could you please add annotates?

When I try to run the conv1d-benchmark, I get a segmentation fault. This is the command line output:

~/test_gazelle/gazelle_mpc$ ./bin/demo/ahe/conv1d-benchmark
NN Layers Benchmark (ms):
KeyGen (3 keys): 6.5802
Preprocess Vector (2 windows): 0.62092
Preprocess Filter: 0.930723
Segmentation fault (core dumped)

Do you have any idea what goes wrong here? Thanks

PS: I build gazelle as you instruct and the other demo scrips work fine

I understand there does not seem to be any code in this library for loading neural network weights and running inference.

I could implement this myself, but I am unsure how I am supposed to deal with rational numbers, since all the examples work on random integers. Naively, I would store the weights and inputs in integer variables, keep track of a scaling factor for each weight and input, and then finally interpret the result as a fixed point number, where the scaling factor is the product of the scaling factors for each weight.

It would be awesome if someone could confirm, whether this is somewhat correct.

Hi chiraag, I'm trying to compile gazelle on ubuntu 20.04 LTS. But there is a problem when I use make -j8. I have Installed cryptotools successfully. My boost version is 1.64.0. I get the following log.

/home/project/gazelle/gazelle_mpc# make -j8
 -- core:linking bin/lib/libgazelle.so from COREOBJECTS
mkdir -p bin/lib
g++ -std=c++14 -fPIC -fPIC -shared -Wl,--export-dynamic,-z,defs -Lthird_party/cryptoTools/thirdparty/linux/miracl/miracl/source -Lthird_party/cryptoTools/lib -o bin/lib/libgazelle.so src/bin/lib/gc/aescircuits.o src/bin/lib/gc/circuits.o src/bin/lib/gc/common.o src/bin/lib/gc/gates.o src/bin/lib/gc/gazelle_circuits.o src/bin/lib/gc/gc.o src/bin/lib/gc/util.o src/bin/lib/math/automorph.o src/bin/lib/math/bit_twiddle.o src/bin/lib/math/discretegaussiangenerator.o src/bin/lib/math/distributiongenerator.o src/bin/lib/math/nbtheory.o src/bin/lib/math/params.o src/bin/lib/math/transfrm.o src/bin/lib/ot/cot_recv.o src/bin/lib/ot/cot_send.o src/bin/lib/ot/sr_base_ot.o src/bin/lib/ot/tools.o src/bin/lib/pke/conv1d.o src/bin/lib/pke/conv2d.o src/bin/lib/pke/encoding.o src/bin/lib/pke/fv.o src/bin/lib/pke/gemm.o src/bin/lib/pke/layers.o src/bin/lib/pke/mat_mul.o src/bin/lib/pke/square.o src/bin/lib/utils/debug.o src/bin/lib/utils/network.o src/bin/lib/utils/test.o -pthread -lcryptoTools -lmiracl -lboost_system -lboost_thread 
/usr/bin/ld: src/bin/lib/gc/aescircuits.o: in function `_GLOBAL__sub_I_aescircuits.cpp':
aescircuits.cpp:(.text.startup+0x35): undefined reference to `boost::system::generic_category()'
/usr/bin/ld: aescircuits.cpp:(.text.startup+0x3a): undefined reference to `boost::system::generic_category()'
/usr/bin/ld: aescircuits.cpp:(.text.startup+0x3f): undefined reference to `boost::system::system_category()'
......
SocketAdapter.cpp:(.text.startup._GLOBAL__sub_I_SocketAdapter.cpp+0x6): undefined reference to `boost::system::generic_category()'
/usr/bin/ld: SocketAdapter.cpp:(.text.startup._GLOBAL__sub_I_SocketAdapter.cpp+0xb): undefined reference to `boost::system::generic_category()'
/usr/bin/ld: SocketAdapter.cpp:(.text.startup._GLOBAL__sub_I_SocketAdapter.cpp+0x10): undefined reference to `boost::system::system_category()'
/usr/bin/ld: SocketAdapter.cpp:(.text.startup._GLOBAL__sub_I_SocketAdapter.cpp+0x15): undefined reference to `boost::system::system_category()'
/usr/bin/ld: third_party/cryptoTools/thirdparty/linux/miracl/miracl/source/libmiracl.a(mrcore.o): relocation R_X86_64_PC32 against symbol `stdout@@GLIBC_2.2.5' can not be used when making a shared object; recompile with -fPIC
/usr/bin/ld: final link failed: bad value
collect2: error: ld returned 1 exit status
make: *** [Makefile.src:37: bin/lib/libgazelle.so] Error 1

One of the most common solutions is to add -lboost_system flag, but the flag was added.
g++ -std=c++14 -fPIC -fPIC -shared -Wl,--export-dynamic,-z,defs -Lthird_party/cryptoTools/thirdparty/linux/miracl/miracl/source -Lthird_party/cryptoTools/lib -o bin/lib/libgazelle.so src/bin/lib/gc/aescircuits.o src/bin/lib/gc/circuits.o src/bin/lib/gc/common.o src/bin/lib/gc/gates.o src/bin/lib/gc/gazelle_circuits.o src/bin/lib/gc/gc.o src/bin/lib/gc/util.o src/bin/lib/math/automorph.o src/bin/lib/math/bit_twiddle.o src/bin/lib/math/discretegaussiangenerator.o src/bin/lib/math/distributiongenerator.o src/bin/lib/math/nbtheory.o src/bin/lib/math/params.o src/bin/lib/math/transfrm.o src/bin/lib/ot/cot_recv.o src/bin/lib/ot/cot_send.o src/bin/lib/ot/sr_base_ot.o src/bin/lib/ot/tools.o src/bin/lib/pke/conv1d.o src/bin/lib/pke/conv2d.o src/bin/lib/pke/encoding.o src/bin/lib/pke/fv.o src/bin/lib/pke/gemm.o src/bin/lib/pke/layers.o src/bin/lib/pke/mat_mul.o src/bin/lib/pke/square.o src/bin/lib/utils/debug.o src/bin/lib/utils/network.o src/bin/lib/utils/test.o -pthread -lcryptoTools -lmiracl -lboost_system -lboost_thread
Do you have any suggestions? Thank you in advance.

The actual encryption function should multiply the polynomials, but seems the code is performing coefficient-wise multiply

Did you use regular (chosen) OT or correlated OT for the OT extension in your implementation? The code for IKNPSender::send seems to implement regular (chosen) OT, but also labels the OT extension files "cot_recv" and "cot_send", and "cot" is often used to represent correlated OT. cot_send.h also has a set_delta function and a block mDelta variable which I think are used with correlated OT, but the library never seems to use either of these.

Thanks!

We read the codes, but we did not find the way to test your entire network, please show us more information, thanks.

Hi, I think there is a bug in SHLCircuit() and SHRCircuit() functions in src/lib/gc/circuits.cpp:

in SHLCircuit():
for(ui64 i=shift; i<n; i++){
out[i] = in[i-1];
}
should be
for(ui64 i=shift; i<n; i++){
out[i] = in[i-shift];
}

and in SHRCircuit():
for(ui64 i=0; i<(n-shift); i++){
out[i] = in[i+1];
}

should be :
for(ui64 i=0; i<(n-shift); i++){
out[i] = in[i+shift];
}

Otherwise, the shift won't be performed correctly