shin/NeuralNetworkLib
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

modified SSE code building

Browse Source
This commit is contained in:
Tomáš Černík
2016-02-07 22:54:43 +01:00
parent 5f43fb8cfb
commit bec7a4f3ae
10 changed files with 28 additions and 43 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
include/NeuralNetwork/ActivationFunction/ActivationFunction.h
View File

@@ -18,7 +18,7 @@ namespace ActivationFunction {
* @brief Returns derivation of output, it is slower than version with output as it needs to compute output * @brief Returns derivation of output, it is slower than version with output as it needs to compute output
* @param input is input of function * @param input is input of function
*/ */
inline float derivatedOutput(const float &input) { inline float derivatedOutput(const float &input) const {
return derivatedOutput(input,operator()(input)); return derivatedOutput(input,operator()(input));
}; };
@@ -28,13 +28,13 @@ namespace ActivationFunction {
* @param output is output of function * @param output is output of function
* @see derivatedOutput * @see derivatedOutput
*/ */
virtual float derivatedOutput(const float &input, const float &output) =0; virtual float derivatedOutput(const float &input, const float &output) const=0;
/** /**
* @brief Returns value of output * @brief Returns value of output
* @param x is input of function * @param x is input of function
*/ */
virtual float operator()(const float &x)=0; virtual float operator()(const float &x) const=0;
/** /**
* @brief Function returns clone of object * @brief Function returns clone of object

4
include/NeuralNetwork/ActivationFunction/Heaviside.h
View File

@@ -9,8 +9,8 @@ namespace ActivationFunction {
public: public:
Heaviside(const float &lambdaP=1.0): lambda(lambdaP) {} Heaviside(const float &lambdaP=1.0): lambda(lambdaP) {}
inline virtual float derivatedOutput(const float &,const float &) override { return 1.0; } inline virtual float derivatedOutput(const float &,const float &) const override { return 1.0; }
inline virtual float operator()(const float &x) override { return x>lambda ? 1.0f : 0.0f; }; inline virtual float operator()(const float &x) const override { return x>lambda ? 1.0f : 0.0f; };
virtual ActivationFunction* clone() const override { virtual ActivationFunction* clone() const override {
return new Heaviside(lambda); return new Heaviside(lambda);

4
include/NeuralNetwork/ActivationFunction/HyperbolicTangent.h
View File

@@ -11,9 +11,9 @@ namespace ActivationFunction {
public: public:
HyperbolicTangent(const float& lam=1):lambda(lam) {} HyperbolicTangent(const float& lam=1):lambda(lam) {}
inline virtual float derivatedOutput(const float &,const float &output) override { return lambda*(1-output*output); } inline virtual float derivatedOutput(const float &,const float &output) const override { return lambda*(1-output*output); }
inline virtual float operator()(const float &x) override { return tanh(lambda*x); }; inline virtual float operator()(const float &x) const override { return tanh(lambda*x); };
virtual ActivationFunction* clone() const override { virtual ActivationFunction* clone() const override {
return new HyperbolicTangent(lambda); return new HyperbolicTangent(lambda);
} }

4
include/NeuralNetwork/ActivationFunction/Linear.h
View File

@@ -9,9 +9,9 @@ namespace ActivationFunction {
public: public:
Linear(const float &lambdaP=1.0): lambda(lambdaP) {} Linear(const float &lambdaP=1.0): lambda(lambdaP) {}
inline virtual float derivatedOutput(const float &,const float &) override { return lambda; } inline virtual float derivatedOutput(const float &,const float &) const override { return lambda; }
inline virtual float operator()(const float &x) override { return x*lambda; }; inline virtual float operator()(const float &x) const override { return x*lambda; };
virtual ActivationFunction* clone() const override { virtual ActivationFunction* clone() const override {
return new Linear(lambda); return new Linear(lambda);

7
include/NeuralNetwork/ActivationFunction/Sigmoid.h
View File

@@ -16,10 +16,11 @@ namespace ActivationFunction {
public: public:
Sigmoid(const float lambdaP = -0.5): lambda(lambdaP) {} Sigmoid(const float lambdaP = -0.5): lambda(lambdaP) {}
inline virtual float derivatedOutput(const float &, const float &output) override { return -lambda*output*(1.0f-output); } inline virtual float derivatedOutput(const float &, const float &output) const override { return -lambda*output*(1.0f-output); }
inline virtual float operator()(const float &x) override { return 1.0f / (1.0f +exp(lambda*x) ); }; inline virtual float operator()(const float &x) const override { return 1.0f / (1.0f +exp(lambda*x) ); };
inline virtual __m128 operator()(const __m128 &x) override {
inline virtual __m128 operator()(const __m128 &x) const override {
// exp_ps is extremly slow! // exp_ps is extremly slow!
return _mm_div_ps(_mm_set1_ps(1.0),_mm_add_ps(exp_ps(_mm_mul_ps(_mm_set1_ps(lambda),x)),_mm_set1_ps(1.0))); return _mm_div_ps(_mm_set1_ps(1.0),_mm_add_ps(exp_ps(_mm_mul_ps(_mm_set1_ps(lambda),x)),_mm_set1_ps(1.0)));
} }

4
include/NeuralNetwork/ActivationFunction/StreamingActivationFunction.h
View File

@@ -14,13 +14,13 @@ namespace ActivationFunction {
class StreamingActivationFunction : public ActivationFunction { class StreamingActivationFunction : public ActivationFunction {
public: public:
virtual float operator()(const float &x)=0; virtual float operator()(const float &x) const=0;
/** /**
* @brief Returns value of four outputs * @brief Returns value of four outputs
* @param x is float[4], in every array value can be stored * @param x is float[4], in every array value can be stored
*/ */
virtual __m128 operator()(const __m128 &x)=0; virtual __m128 operator()(const __m128 &x) const=0;
}; };
} }
} }

15
include/NeuralNetwork/BasisFunction/Linear.h
View File

@@ -17,7 +17,9 @@ namespace BasisFunction {
public: public:
Linear() {} Linear() {}
inline virtual float computeStreaming(const std::vector<float>& weights, const std::vector<float>& input) const override { inline virtual float operator()(const std::vector<float>& weights, const std::vector<float>& input) const override {
#ifdef USE_SSE
size_t inputSize=input.size(); size_t inputSize=input.size();
size_t alignedPrev=inputSize-inputSize%4; size_t alignedPrev=inputSize-inputSize%4;
@@ -35,24 +37,23 @@ namespace BasisFunction {
partialSolution.sse=_mm_add_ps(partialSolution.sse,_mm_mul_ps(_mm_load_ss(weightsData+k),_mm_load_ss(inputData+k))); partialSolution.sse=_mm_add_ps(partialSolution.sse,_mm_mul_ps(_mm_load_ss(weightsData+k),_mm_load_ss(inputData+k)));
} }
#ifdef USE_SSE2 //pre-SSE3 solution #ifdef USE_SSE2 //pre-SSE3 solution
partialSolution.sse= _mm_add_ps(_mm_movehl_ps(partialSolution.sse, partialSolution.sse), partialSolution.sse); partialSolution.sse= _mm_add_ps(_mm_movehl_ps(partialSolution.sse, partialSolution.sse), partialSolution.sse);
partialSolution.sse=_mm_add_ss(partialSolution.sse, _mm_shuffle_ps(partialSolution.sse,partialSolution.sse, 1)); partialSolution.sse=_mm_add_ss(partialSolution.sse, _mm_shuffle_ps(partialSolution.sse,partialSolution.sse, 1));
#else #else
partialSolution.sse = _mm_hadd_ps(partialSolution.sse, partialSolution.sse); partialSolution.sse = _mm_hadd_ps(partialSolution.sse, partialSolution.sse);
partialSolution.sse = _mm_hadd_ps(partialSolution.sse, partialSolution.sse); partialSolution.sse = _mm_hadd_ps(partialSolution.sse, partialSolution.sse);
#endif #endif
return partialSolution.f[0]; return partialSolution.f[0];
} #else
inline virtual float compute(const std::vector<float>& weights, const std::vector<float>& input) const override {
register float tmp = 0; register float tmp = 0;
size_t inputSize=input.size(); size_t inputSize=input.size();
for(size_t k=0;k<inputSize;k++) { for(size_t k=0;k<inputSize;k++) {
tmp+=input[k]*weights[k]; tmp+=input[k]*weights[k];
} }
return tmp; return tmp;
#endif
} }
virtual BasisFunction* clone() const override { virtual BasisFunction* clone() const override {

8
include/NeuralNetwork/BasisFunction/StreamingBasisFunction.h
View File

@@ -13,13 +13,7 @@ namespace BasisFunction {
float f[4]; float f[4];
}; };
virtual float operator()(const std::vector<float>& weights, const std::vector<float>& input) const override { virtual float operator()(const std::vector<float>& weights, const std::vector<float>& input) const =0;
return computeStreaming(weights,input);
}
virtual float computeStreaming(const std::vector<float>& weights, const std::vector<float>& input) const =0;
virtual float compute(const std::vector<float>& weights, const std::vector<float>& input) const =0;
}; };
} }
} }

12
tests/basis.cpp
View File

@@ -8,13 +8,11 @@
int main() { int main() {
{ {
NeuralNetwork::BasisFunction::Linear l; NeuralNetwork::BasisFunction::Linear l;
assert(39.0==l.compute({1,2,3,5},{1,2,3,5})); assert(39.0==l({1,2,3,5},{1,2,3,5}));
assert(39.0==l.computeStreaming({1,2,3,5},{1,2,3,5}));
} }
{ {
NeuralNetwork::BasisFunction::Linear l; NeuralNetwork::BasisFunction::Linear l;
assert(88.0==l.computeStreaming({1,2,3,5,7},{1,2,3,5,7})); assert(88.0==l({1,2,3,5,7},{1,2,3,5,7}));
assert(88.0==l.compute({1,2,3,5,7},{1,2,3,5,7}));
} }
{ {
NeuralNetwork::BasisFunction::Linear l; NeuralNetwork::BasisFunction::Linear l;
@@ -22,8 +20,7 @@ int main() {
for(int in=0;in<100;in++) { for(int in=0;in<100;in++) {
w.push_back(2); w.push_back(2);
} }
assert(400.0==l.computeStreaming(w,w)); assert(400.0==l(w,w));
assert(400.0==l.compute(w,w));
} }
{ {
NeuralNetwork::BasisFunction::Linear l; NeuralNetwork::BasisFunction::Linear l;
@@ -31,8 +28,7 @@ int main() {
for(int in=0;in<55;in++) { for(int in=0;in<55;in++) {
w.push_back(2); w.push_back(2);
} }
assert(220.0==l.computeStreaming(w,w)); assert(220.0==l(w,w));
assert(220.0==l.compute(w,w));
} }
{ {
NeuralNetwork::BasisFunction::Product l; NeuralNetwork::BasisFunction::Product l;

7
tests/feedforward.cpp
View File

@@ -3,13 +3,6 @@
#include <cassert> #include <cassert>
#include <iostream> #include <iostream>
void printVec(const std::vector<float> &v) {
for(int i=0;i<v.size();i++) {
std::cout << v[i] << ", ";
}
std::cout<< "\n";
}
int main() { int main() {
{ // XOR problem { // XOR problem
NeuralNetwork::FeedForward::Network n(2); NeuralNetwork::FeedForward::Network n(2);