support multiple architectures (e.g. SSE, AVX) in the same executable… (

#118) * support multiple architectures (e.g. SSE, AVX) in the same executable via a macro * fix CI tests * fix makefile * define RTNEURAL_NAMESPACE (if not already) as a safety net * RTNEURAL_NAMESPACE is configurable in cmake
jatinchowdhury18 · Nov 23, 2023 · 1e7b570 · 1e7b570
1 parent 37fde44
commit 1e7b570
Show file tree

Hide file tree

Showing 61 changed files with 114 additions and 105 deletions.
diff --git a/RTNeural/CMakeLists.txt b/RTNeural/CMakeLists.txt
@@ -46,3 +46,8 @@ target_include_directories(RTNeural
     INTERFACE
         ..
 )
+set(RTNEURAL_NAMESPACE "RTNeural" CACHE STRING "Namespace to use for RTNeural code")
+target_compile_definitions(RTNeural
+    PUBLIC
+        RTNEURAL_NAMESPACE=${RTNEURAL_NAMESPACE}
+)
diff --git a/RTNeural/Layer.h b/RTNeural/Layer.h
@@ -4,7 +4,7 @@
 #include <cstddef>
 #include <string>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /** Virtual base class for a generic neural network layer. */
@@ -34,6 +34,6 @@ class Layer
     const int out_size;
 };
 
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #endif // LAYER_H_INCLUDED
diff --git a/RTNeural/Model.h b/RTNeural/Model.h
@@ -19,7 +19,7 @@
 #include "lstm/lstm.h"
 #include "lstm/lstm.tpp"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /**
@@ -112,6 +112,6 @@ class Model
     std::vector<vec_type> outs;
 };
 
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #endif // MODEL_H_INCLUDED
diff --git a/RTNeural/ModelT.h b/RTNeural/ModelT.h
@@ -2,7 +2,7 @@
 
 #include "model_loader.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 #ifndef DOXYGEN
@@ -575,4 +575,4 @@ class ModelT2D
     static constexpr size_t n_layers = sizeof...(Layers);
 };
 #endif // RTNEURAL_USE_XSIMD
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
diff --git a/RTNeural/RTNeural.cpp b/RTNeural/RTNeural.cpp
@@ -1,7 +1,7 @@
 #include "RTNeural.h"
 
 // forward declare some template classes
-template class RTNeural::Model<float>;
-template class RTNeural::Model<double>;
-template class RTNeural::Layer<float>;
-template class RTNeural::Layer<double>;
+template class RTNEURAL_NAMESPACE::Model<float>;
+template class RTNEURAL_NAMESPACE::Model<double>;
+template class RTNEURAL_NAMESPACE::Layer<float>;
+template class RTNEURAL_NAMESPACE::Layer<double>;
diff --git a/RTNeural/RTNeural.h b/RTNeural/RTNeural.h
@@ -5,6 +5,10 @@
 // C++ STL includes
 #include <limits>
 
+#ifndef RTNEURAL_NAMESPACE
+#define RTNEURAL_NAMESPACE RTNeural
+#endif
+
 // Handle default RTNeural defines
 #ifndef RTNEURAL_DEFAULT_ALIGNMENT
 #if _MSC_VER

diff --git a/RTNeural/activation/activation.h b/RTNeural/activation/activation.h
@@ -4,7 +4,7 @@
 #include "../Layer.h"
 #include <functional>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /** Base class for activation layers. */
@@ -35,7 +35,7 @@ class Activation : public Layer<T>
     const std::function<T(T)> func;
 };
 
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #if RTNEURAL_USE_EIGEN
 #include "activation_eigen.h"
@@ -48,7 +48,7 @@ class Activation : public Layer<T>
 #include "../maths/maths_stl.h"
 #include <cmath>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /** Dynamic implementation of a tanh activation layer. */
@@ -424,7 +424,7 @@ class PReLUActivationT
     T outs[size];
     T alpha[size];
 };
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #endif // RTNEURAL_USE_EIGEN
 

diff --git a/RTNeural/activation/activation_eigen.h b/RTNeural/activation/activation_eigen.h
@@ -4,7 +4,7 @@
 #include "../common.h"
 #include "../maths/maths_eigen.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /** Dynamic implementation of a tanh activation layer. */
@@ -463,6 +463,6 @@ class PReLUActivationT
     T outs_internal alignas(RTNEURAL_DEFAULT_ALIGNMENT)[out_size];
     v_type alpha;
 };
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #endif // ACTIVATIONEIGEN_H_INCLUDED
diff --git a/RTNeural/activation/activation_xsimd.h b/RTNeural/activation/activation_xsimd.h
@@ -4,7 +4,7 @@
 #include "../common.h"
 #include "../maths/maths_xsimd.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /** Dynamic implementation of a tanh activation layer. */
@@ -430,6 +430,6 @@ class PReLUActivationT
     v_type outs[v_io_size];
     v_type alpha[v_io_size];
 };
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #endif // ACTIVATIONXSIMD_H_INCLUDED
diff --git a/RTNeural/batchnorm/batchnorm.h b/RTNeural/batchnorm/batchnorm.h
@@ -12,7 +12,7 @@
 #include "../common.h"
 #include <vector>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 /** Dynamic batch normalization layer. */
 template <typename T>

diff --git a/RTNeural/batchnorm/batchnorm.tpp b/RTNeural/batchnorm/batchnorm.tpp
@@ -1,6 +1,6 @@
 #include "batchnorm.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 #if !RTNEURAL_USE_EIGEN && !RTNEURAL_USE_XSIMD
 

diff --git a/RTNeural/batchnorm/batchnorm2d.h b/RTNeural/batchnorm/batchnorm2d.h
@@ -10,7 +10,7 @@
 #else
 #include "../Layer.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 /** Dynamic batch normalization layer. */
 template <typename T>

diff --git a/RTNeural/batchnorm/batchnorm2d.tpp b/RTNeural/batchnorm/batchnorm2d.tpp
@@ -2,7 +2,7 @@
 
 #if !RTNEURAL_USE_EIGEN && !RTNEURAL_USE_XSIMD
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 template <typename T>
 BatchNorm2DLayer<T>::BatchNorm2DLayer(int in_num_filters, int in_num_features)

diff --git a/RTNeural/batchnorm/batchnorm2d_eigen.h b/RTNeural/batchnorm/batchnorm2d_eigen.h
@@ -4,7 +4,7 @@
 #include "../Layer.h"
 #include <Eigen/Dense>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 /** Dynamic batch normalization layer. */
 template <typename T>

diff --git a/RTNeural/batchnorm/batchnorm2d_eigen.tpp b/RTNeural/batchnorm/batchnorm2d_eigen.tpp
@@ -1,6 +1,6 @@
 #include "batchnorm2d_eigen.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 template <typename T>
 BatchNorm2DLayer<T>::BatchNorm2DLayer(int in_num_filters, int in_num_features)

diff --git a/RTNeural/batchnorm/batchnorm2d_xsimd.h b/RTNeural/batchnorm/batchnorm2d_xsimd.h
@@ -4,7 +4,7 @@
 #include "../Layer.h"
 #include <xsimd/xsimd.hpp>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 /** Dynamic batch normalization layer. */
 template <typename T>

diff --git a/RTNeural/batchnorm/batchnorm2d_xsimd.tpp b/RTNeural/batchnorm/batchnorm2d_xsimd.tpp
@@ -1,6 +1,6 @@
 #include "batchnorm2d_xsimd.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 template <typename T>
 BatchNorm2DLayer<T>::BatchNorm2DLayer(int in_num_filters, int in_num_features)

diff --git a/RTNeural/batchnorm/batchnorm_eigen.h b/RTNeural/batchnorm/batchnorm_eigen.h
@@ -4,7 +4,7 @@
 #include "../Layer.h"
 #include <Eigen/Dense>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 /** Dynamic batch normalization layer. */
 template <typename T>

diff --git a/RTNeural/batchnorm/batchnorm_eigen.tpp b/RTNeural/batchnorm/batchnorm_eigen.tpp
@@ -1,6 +1,6 @@
 #include "batchnorm_eigen.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 template <typename T>
 BatchNorm1DLayer<T>::BatchNorm1DLayer(int size)

diff --git a/RTNeural/batchnorm/batchnorm_xsimd.h b/RTNeural/batchnorm/batchnorm_xsimd.h
@@ -4,7 +4,7 @@
 #include "../Layer.h"
 #include <xsimd/xsimd.hpp>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 /** Dynamic batch normalization layer. */
 template <typename T>

diff --git a/RTNeural/batchnorm/batchnorm_xsimd.tpp b/RTNeural/batchnorm/batchnorm_xsimd.tpp
@@ -1,6 +1,6 @@
 #include "batchnorm_xsimd.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 template <typename T>
 BatchNorm1DLayer<T>::BatchNorm1DLayer(int size)

diff --git a/RTNeural/common.h b/RTNeural/common.h
@@ -1,6 +1,6 @@
 #pragma once
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /**
@@ -29,26 +29,26 @@ constexpr T ceil_div(T num, T den)
 {
     return (num + den - 1) / den;
 }
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #if RTNEURAL_USE_EIGEN
 #include <Eigen/Dense>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 #if RTNEURAL_DEFAULT_ALIGNMENT == 32
 constexpr auto RTNeuralEigenAlignment = Eigen::Aligned32;
 #else
 constexpr auto RTNeuralEigenAlignment = Eigen::Aligned16;
 #endif
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #elif RTNEURAL_USE_XSIMD
 #include <xsimd/xsimd.hpp>
 
 #include "xsimd-legacy/algorithms/algorithms.hpp"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 template <typename T>
@@ -236,20 +236,20 @@ static inline void elu(const T* in, T* out, int dim, T alpha) noexcept
     for(auto i = vec_size; i < dim; ++i)
         out[i] = in[i] > (T)0 ? in[i] : (alpha * (MathsProvider::exp(in[i]) - (T)1));
 }
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #else // STL backend
 #include <algorithm>
 #include <cmath>
 #include <numeric>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 template <typename T>
 static inline T vMult(const T* arg1, const T* arg2, int dim) noexcept
 {
     return std::inner_product(arg1, arg1 + dim, arg2, (T)0);
 }
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #endif
diff --git a/RTNeural/conv1d/conv1d.h b/RTNeural/conv1d/conv1d.h
@@ -12,7 +12,7 @@
 #include "../common.h"
 #include <vector>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /**
@@ -241,6 +241,6 @@ class Conv1DT
             state_ptrs[k] = (state_ptr + state_size - k * dilation_rate) % state_size;
     }
 };
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 #endif
 #endif // CONV1D_H_INCLUDED
diff --git a/RTNeural/conv1d/conv1d.tpp b/RTNeural/conv1d/conv1d.tpp
@@ -1,6 +1,6 @@
 #include "conv1d.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 #if !RTNEURAL_USE_EIGEN && !RTNEURAL_USE_XSIMD
@@ -166,4 +166,4 @@ void Conv1DT<T, in_sizet, out_sizet, kernel_size, dilation_rate, dynamic_state>:
 
 #endif
 
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
diff --git a/RTNeural/conv1d/conv1d_eigen.h b/RTNeural/conv1d/conv1d_eigen.h
@@ -4,7 +4,7 @@
 #include "../Layer.h"
 #include <Eigen/Dense>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /**
@@ -214,6 +214,6 @@ class Conv1DT
     }
 };
 
-} // RTNeural
+} // RTNEURAL_NAMESPACE
 
 #endif // CONV1DEIGEN_H_INCLUDED
diff --git a/RTNeural/conv1d/conv1d_eigen.tpp b/RTNeural/conv1d/conv1d_eigen.tpp
@@ -1,6 +1,6 @@
 #include "conv1d_eigen.h"
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 template <typename T>
@@ -105,4 +105,4 @@ void Conv1DT<T, in_sizet, out_sizet, kernel_size, dilation_rate, dynamic_state>:
         bias(i) = biasVals[i];
 }
 
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
diff --git a/RTNeural/conv1d/conv1d_xsimd.h b/RTNeural/conv1d/conv1d_xsimd.h
@@ -6,7 +6,7 @@
 #include <numeric>
 #include <vector>
 
-namespace RTNeural
+namespace RTNEURAL_NAMESPACE
 {
 
 /**
@@ -313,6 +313,6 @@ class Conv1DT
             state_ptrs[k] = (state_ptr + state_size - k * dilation_rate) % state_size;
     }
 };
-} // namespace RTNeural
+} // namespace RTNEURAL_NAMESPACE
 
 #endif // CONV1DXSIMD_H_INCLUDED