未验证 提交 541e9119 编写于 作者: J jed 提交者: GitHub

Merge pull request #107 from Yanghello/truncate3

add compile option for truncate3
......@@ -61,6 +61,8 @@ option(USE_AES_NI "Compile with AES NI" ON)
option(USE_OPENMP "Compile with OpenMP" ON)
option(USE_ABY3_TRUNC1 "Compile with ABY3 truncate 1 algorithm" OFF)
########################### the project build part ###############################
message(STATUS "Using paddlepaddle installation of ${paddle_version}")
message(STATUS "paddlepaddle include directory: ${PADDLE_INCLUDE}")
......@@ -84,6 +86,10 @@ if (USE_OPENMP)
find_package(OpenMP REQUIRED)
endif(USE_OPENMP)
if (USE_ABY3_TRUNC1)
add_compile_definitions(USE_ABY3_TRUNC1)
endif(USE_ABY3_TRUNC1)
add_subdirectory(core/privc3)
add_subdirectory(core/paddlefl_mpc/mpc_protocol)
add_subdirectory(core/paddlefl_mpc/operators)
......
......@@ -191,6 +191,9 @@ public:
void max_pooling(FixedPointTensor* ret,
BooleanTensor<T>* pos = nullptr) const;
static void truncate(const FixedPointTensor* op, FixedPointTensor* ret,
size_t scaling_factor);
private:
static inline std::shared_ptr<CircuitContext> aby3_ctx() {
......@@ -201,9 +204,6 @@ private:
return paddle::mpc::ContextHolder::tensor_factory();
}
static void truncate(const FixedPointTensor* op, FixedPointTensor* ret,
size_t scaling_factor);
template<typename MulFunc>
static void mul_trunc(const FixedPointTensor<T, N>* lhs,
const FixedPointTensor<T, N>* rhs,
......
......@@ -21,7 +21,6 @@
#include "prng.h"
namespace aby3 {
template<typename T, size_t N>
FixedPointTensor<T, N>::FixedPointTensor(TensorAdapter<T>* share_tensor[2]) {
// TODO: check tensors' shapes
......@@ -166,6 +165,7 @@ void FixedPointTensor<T, N>::mul(const FixedPointTensor<T, N>* rhs,
mul_trunc(this, rhs, ret, &TensorAdapter<T>::mul);
}
#ifdef USE_ABY3_TRUNC1 //use aby3 trunc1
template<typename T, size_t N>
void FixedPointTensor<T, N>::truncate(const FixedPointTensor<T, N>* op,
FixedPointTensor<T, N>* ret,
......@@ -208,7 +208,20 @@ void FixedPointTensor<T, N>::truncate(const FixedPointTensor<T, N>* op,
return;
}
// Protocol. `truncate3`
#else // use truncate3
// Protocol. `truncate3` (illustrated for data type T = int64_t)
// motivation:
// truncates in aby3 may cause msb error with small probability
// the reason is that before rishft op, its masked value e.g., x' - r' may overflow in int64_t
// so that, in `truncate3`, we limit r' in (-2^62, 2^62) to avoid the problem.
// notice:
// when r' is contrainted in (-2^62, 2^62),
// the SD (statistical distance) of x' - r' between this
// and r' in Z_{2^64} is equal to |X| / (2^63 + |X|)
// detail protocol:
// P2 randomly generates r' \in (-2^62, 2^62), randomly generates r'_0, r_0, r_1 in Z_{2^64},
// P2 compute r'_1 = r' - r'_0, r_2 = r'/2^N - r_0 - r_1, let x2 = r_2
// P2 send r_0, r'_0 to P0, send r_1, r'_1 to P1
......@@ -217,7 +230,7 @@ void FixedPointTensor<T, N>::truncate(const FixedPointTensor<T, N>* op,
// P0 set x0 = r_0
// P0, P1, P2 invoke reshare() with inputs x0, x1, x2 respectively.
template<typename T, size_t N>
void FixedPointTensor<T, N>::truncate3(const FixedPointTensor<T, N>* op,
void FixedPointTensor<T, N>::truncate(const FixedPointTensor<T, N>* op,
FixedPointTensor<T, N>* ret,
size_t scaling_factor) {
if (scaling_factor == 0) {
......@@ -231,23 +244,9 @@ void FixedPointTensor<T, N>::truncate3(const FixedPointTensor<T, N>* op,
temp.emplace_back(
tensor_factory()->template create<T>(op->shape()));
}
// r', contraint in (-2^62, 2^62)
// notice : when r' is contrainted in (-2^62, 2^62),
// the SD (statistical distance) of x - r' between this
// and r' in Z_{2^64} is equal to |X| / (2^63 + |X|)
// according to http://yuyu.hk/files/ho2.pdf
// r'
aby3_ctx()->template gen_random_private(*temp[0]);
int64_t contraint_upper = ~((uint64_t) 1 << 62);
int64_t contraint_low = (uint64_t) 1 << 62;
std::for_each(temp[0]->data(), temp[0]->data() + temp[0]->numel(),
[&contraint_upper, &contraint_low] (T& a) {
// contraint -2^62 < a < 2^62
if (a >= 0) {
a &= contraint_upper;
} else {
a |= contraint_low;
}
});
temp[0]->rshift(1, temp[0].get());
//r'_0, r'_1
aby3_ctx()->template gen_random_private(*temp[1]);
......@@ -307,6 +306,7 @@ void FixedPointTensor<T, N>::truncate3(const FixedPointTensor<T, N>* op,
tensor_carry_in->scaling_factor() = N;
ret->add(tensor_carry_in.get(), ret);
}
#endif //USE_ABY3_TRUNC1
template<typename T, size_t N>
template<typename MulFunc>
......@@ -345,7 +345,7 @@ void FixedPointTensor<T, N>::mul_trunc(const FixedPointTensor<T, N>* lhs,
temp->copy(ret_no_trunc->_share[0]);
reshare(temp.get(), ret_no_trunc->_share[1]);
truncate3(ret_no_trunc.get(), ret, N);
truncate(ret_no_trunc.get(), ret, N);
}
template<typename T, size_t N>
......@@ -360,7 +360,7 @@ void FixedPointTensor<T, N>::mul(const TensorAdapter<T>* rhs,
_share[0]->mul(rhs, temp->_share[0]);
_share[1]->mul(rhs, temp->_share[1]);
truncate3(temp.get(), ret, rhs->scaling_factor());
truncate(temp.get(), ret, rhs->scaling_factor());
}
template<typename T, size_t N>
......@@ -404,7 +404,7 @@ void FixedPointTensor<T, N>::mat_mul(const TensorAdapter<T>* rhs,
FixedPointTensor<T, N>* ret) const {
_share[0]->mat_mul(rhs, ret->_share[0]);
_share[1]->mat_mul(rhs, ret->_share[1]);
truncate3(ret, ret, rhs->scaling_factor());
truncate(ret, ret, rhs->scaling_factor());
}
template< typename T, size_t N>
......@@ -831,7 +831,7 @@ void FixedPointTensor<T, N>::long_div(const FixedPointTensor<T, N>* rhs,
}
for (size_t i = 1; i <= N; ++i) {
truncate3(&abs_rhs, &sub_rhs, i);
truncate(&abs_rhs, &sub_rhs, i);
abs_lhs.gt(&sub_rhs, &cmp_res);
cmp_res.mul(&sub_rhs, &sub_rhs);
cmp_res.lshift(N - i, &cmp_res);
......@@ -1184,7 +1184,7 @@ void FixedPointTensor<T, N>::inverse_square_root(const FixedPointTensor* op,
std::shared_ptr<FixedPointTensor<T, N>> x2 =
std::make_shared<FixedPointTensor<T, N>>(temp[2].get(), temp[3].get());
// x2 = 0.5 * op
truncate3(op, x2.get(), 1);
truncate(op, x2.get(), 1);
assign_to_tensor(y->mutable_share(0), (T)(x0 * pow(2, N)));
assign_to_tensor(y->mutable_share(1), (T)(x0 * pow(2, N)));
......
......@@ -1267,6 +1267,7 @@ TEST_F(FixedTensorTest, mulfixed) {
EXPECT_TRUE(test_fixedt_check_tensor_eq(out0.get(), &result));
}
#ifndef USE_ABY3_TRUNC1 //use aby3 trunc1
TEST_F(FixedTensorTest, mulfixed_multi_times) {
std::vector<size_t> shape = {100000, 1};
......@@ -1327,6 +1328,7 @@ TEST_F(FixedTensorTest, mulfixed_multi_times) {
EXPECT_TRUE(test_fixedt_check_tensor_eq(out1.get(), out2.get()));
EXPECT_TRUE(test_fixedt_check_tensor_eq(out0.get(), &result));
}
#endif
TEST_F(FixedTensorTest, mulfixed_overflow) {
......@@ -3435,4 +3437,124 @@ TEST_F(FixedTensorTest, inv_sqrt_test) {
}
#ifdef USE_ABY3_TRUNC1 //use aby3 trunc1
TEST_F(FixedTensorTest, truncate1_msb_incorrect) {
std::vector<size_t> shape = { 1 };
std::shared_ptr<TensorAdapter<int64_t>> sl[3] = { gen(shape), gen(shape), gen(shape) };
std::shared_ptr<TensorAdapter<int64_t>> sout[6] = { gen(shape), gen(shape), gen(shape),
gen(shape), gen(shape), gen(shape)};
// lhs = 6 = 1 + 2 + 3, share before truncate
// zero share 0 = (1 << 62) + (1 << 62) - (1 << 63)
sl[0]->data()[0] = ((int64_t) 3 << 32) - ((uint64_t) 1 << 63);
sl[1]->data()[0] = ((int64_t) 2 << 32) + ((int64_t) 1 << 62);
sl[2]->data()[0] = ((int64_t) 1 << 32) + ((int64_t) 1 << 62);
auto pr = gen(shape);
// rhs = 15
pr->data()[0] = 6 << 16;
pr->scaling_factor() = 16;
Fix64N16 fl0(sl[0].get(), sl[1].get());
Fix64N16 fl1(sl[1].get(), sl[2].get());
Fix64N16 fl2(sl[2].get(), sl[0].get());
Fix64N16 fout0(sout[0].get(), sout[1].get());
Fix64N16 fout1(sout[2].get(), sout[3].get());
Fix64N16 fout2(sout[4].get(), sout[5].get());
auto p = gen(shape);
_t[0] = std::thread(
[&] () {
g_ctx_holder::template run_with_context(
_exec_ctx.get(), _mpc_ctx[0], [&](){
Fix64N16::truncate(&fl0, &fout0, 16);
fout0.reveal_to_one(0, p.get());
});
}
);
_t[1] = std::thread(
[&] () {
g_ctx_holder::template run_with_context(
_exec_ctx.get(), _mpc_ctx[1], [&](){
Fix64N16::truncate(&fl1, &fout1, 16);
fout1.reveal_to_one(0, nullptr);
});
}
);
_t[2] = std::thread(
[&] () {
g_ctx_holder::template run_with_context(
_exec_ctx.get(), _mpc_ctx[2], [&](){
Fix64N16::truncate(&fl2, &fout2, 16);
fout2.reveal_to_one(0, nullptr);
});
}
);
for (auto &t: _t) {
t.join();
}
// failed: result is not close to 6
EXPECT_GT(std::abs((p->data()[0] >> 16) - 6), 1000);
}
#else
TEST_F(FixedTensorTest, truncate3_msb_correct) {
std::vector<size_t> shape = { 1 };
std::shared_ptr<TensorAdapter<int64_t>> sl[3] = { gen(shape), gen(shape), gen(shape) };
std::shared_ptr<TensorAdapter<int64_t>> sout[6] = { gen(shape), gen(shape), gen(shape),
gen(shape), gen(shape), gen(shape)};
// lhs = 6 = 1 + 2 + 3, share before truncate
// zero share 0 = (1 << 62) + (1 << 62) - (1 << 63)
sl[0]->data()[0] = ((int64_t) 3 << 32) - ((uint64_t) 1 << 63);
sl[1]->data()[0] = ((int64_t) 2 << 32) + ((int64_t) 1 << 62);
sl[2]->data()[0] = ((int64_t) 1 << 32) + ((int64_t) 1 << 62);
auto pr = gen(shape);
// rhs = 15
pr->data()[0] = 6 << 16;
pr->scaling_factor() = 16;
Fix64N16 fl0(sl[0].get(), sl[1].get());
Fix64N16 fl1(sl[1].get(), sl[2].get());
Fix64N16 fl2(sl[2].get(), sl[0].get());
Fix64N16 fout0(sout[0].get(), sout[1].get());
Fix64N16 fout1(sout[2].get(), sout[3].get());
Fix64N16 fout2(sout[4].get(), sout[5].get());
auto p = gen(shape);
_t[0] = std::thread(
[&] () {
g_ctx_holder::template run_with_context(
_exec_ctx.get(), _mpc_ctx[0], [&](){
Fix64N16::truncate(&fl0, &fout0, 16);
fout0.reveal_to_one(0, p.get());
});
}
);
_t[1] = std::thread(
[&] () {
g_ctx_holder::template run_with_context(
_exec_ctx.get(), _mpc_ctx[1], [&](){
Fix64N16::truncate(&fl1, &fout1, 16);
fout1.reveal_to_one(0, nullptr);
});
}
);
_t[2] = std::thread(
[&] () {
g_ctx_holder::template run_with_context(
_exec_ctx.get(), _mpc_ctx[2], [&](){
Fix64N16::truncate(&fl2, &fout2, 16);
fout2.reveal_to_one(0, nullptr);
});
}
);
for (auto &t: _t) {
t.join();
}
EXPECT_EQ((p->data()[0] >> 16), 6);
}
#endif
} // namespace aby3
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册