1 namespace Eigen { 2 3 /** \page TopicInsideEigenExample What happens inside Eigen, on a simple example 4 5 \b Table \b of \b contents 6 - \ref WhyInteresting 7 - \ref ConstructingVectors 8 - \ref ConstructionOfSumXpr 9 - \ref Assignment 10 \n 11 12 <hr> 13 14 15 Consider the following example program: 16 17 \code 18 #include<Eigen/Core> 19 20 int main() 21 { 22 int size = 50; 23 // VectorXf is a vector of floats, with dynamic size. 24 Eigen::VectorXf u(size), v(size), w(size); 25 u = v + w; 26 } 27 \endcode 28 29 The goal of this page is to understand how Eigen compiles it, assuming that SSE2 vectorization is enabled (GCC option -msse2). 30 31 \section WhyInteresting Why it's interesting 32 33 Maybe you think, that the above example program is so simple, that compiling it shouldn't involve anything interesting. So before starting, let us explain what is nontrivial in compiling it correctly -- that is, producing optimized code -- so that the complexity of Eigen, that we'll explain here, is really useful. 34 35 Look at the line of code 36 \code 37 u = v + w; // (*) 38 \endcode 39 40 The first important thing about compiling it, is that the arrays should be traversed only once, like 41 \code 42 for(int i = 0; i < size; i++) u[i] = v[i] + w[i]; 43 \endcode 44 The problem is that if we make a naive C++ library where the VectorXf class has an operator+ returning a VectorXf, then the line of code (*) will amount to: 45 \code 46 VectorXf tmp = v + w; 47 VectorXf u = tmp; 48 \endcode 49 Obviously, the introduction of the temporary \a tmp here is useless. It has a very bad effect on performance, first because the creation of \a tmp requires a dynamic memory allocation in this context, and second as there are now two for loops: 50 \code 51 for(int i = 0; i < size; i++) tmp[i] = v[i] + w[i]; 52 for(int i = 0; i < size; i++) u[i] = tmp[i]; 53 \endcode 54 Traversing the arrays twice instead of once is terrible for performance, as it means that we do many redundant memory accesses. 55 56 The second important thing about compiling the above program, is to make correct use of SSE2 instructions. Notice that Eigen also supports AltiVec and that all the discussion that we make here applies also to AltiVec. 57 58 SSE2, like AltiVec, is a set of instructions allowing to perform computations on packets of 128 bits at once. Since a float is 32 bits, this means that SSE2 instructions can handle 4 floats at once. This means that, if correctly used, they can make our computation go up to 4x faster. 59 60 However, in the above program, we have chosen size=50, so our vectors consist of 50 float's, and 50 is not a multiple of 4. This means that we cannot hope to do all of that computation using SSE2 instructions. The second best thing, to which we should aim, is to handle the 48 first coefficients with SSE2 instructions, since 48 is the biggest multiple of 4 below 50, and then handle separately, without SSE2, the 49th and 50th coefficients. Something like this: 61 62 \code 63 for(int i = 0; i < 4*(size/4); i+=4) u.packet(i) = v.packet(i) + w.packet(i); 64 for(int i = 4*(size/4); i < size; i++) u[i] = v[i] + w[i]; 65 \endcode 66 67 So let us look line by line at our example program, and let's follow Eigen as it compiles it. 68 69 \section ConstructingVectors Constructing vectors 70 71 Let's analyze the first line: 72 73 \code 74 Eigen::VectorXf u(size), v(size), w(size); 75 \endcode 76 77 First of all, VectorXf is the following typedef: 78 \code 79 typedef Matrix<float, Dynamic, 1> VectorXf; 80 \endcode 81 82 The class template Matrix is declared in src/Core/util/ForwardDeclarations.h with 6 template parameters, but the last 3 are automatically determined by the first 3. So you don't need to worry about them for now. Here, Matrix\<float, Dynamic, 1\> means a matrix of floats, with a dynamic number of rows and 1 column. 83 84 The Matrix class inherits a base class, MatrixBase. Don't worry about it, for now it suffices to say that MatrixBase is what unifies matrices/vectors and all the expressions types -- more on that below. 85 86 When we do 87 \code 88 Eigen::VectorXf u(size); 89 \endcode 90 the constructor that is called is Matrix::Matrix(int), in src/Core/Matrix.h. Besides some assertions, all it does is to construct the \a m_storage member, which is of type DenseStorage\<float, Dynamic, Dynamic, 1\>. 91 92 You may wonder, isn't it overengineering to have the storage in a separate class? The reason is that the Matrix class template covers all kinds of matrices and vector: both fixed-size and dynamic-size. The storage method is not the same in these two cases. For fixed-size, the matrix coefficients are stored as a plain member array. For dynamic-size, the coefficients will be stored as a pointer to a dynamically-allocated array. Because of this, we need to abstract storage away from the Matrix class. That's DenseStorage. 93 94 Let's look at this constructor, in src/Core/DenseStorage.h. You can see that there are many partial template specializations of DenseStorages here, treating separately the cases where dimensions are Dynamic or fixed at compile-time. The partial specialization that we are looking at is: 95 \code 96 template<typename T, int _Cols> class DenseStorage<T, Dynamic, Dynamic, _Cols> 97 \endcode 98 99 Here, the constructor called is DenseStorage::DenseStorage(int size, int rows, int columns) 100 with size=50, rows=50, columns=1. 101 102 Here is this constructor: 103 \code 104 inline DenseStorage(int size, int rows, int) : m_data(internal::aligned_new<T>(size)), m_rows(rows) {} 105 \endcode 106 107 Here, the \a m_data member is the actual array of coefficients of the matrix. As you see, it is dynamically allocated. Rather than calling new[] or malloc(), as you can see, we have our own internal::aligned_new defined in src/Core/util/Memory.h. What it does is that if vectorization is enabled, then it uses a platform-specific call to allocate a 128-bit-aligned array, as that is very useful for vectorization with both SSE2 and AltiVec. If vectorization is disabled, it amounts to the standard new[]. 108 109 As you can see, the constructor also sets the \a m_rows member to \a size. Notice that there is no \a m_columns member: indeed, in this partial specialization of DenseStorage, we know the number of columns at compile-time, since the _Cols template parameter is different from Dynamic. Namely, in our case, _Cols is 1, which is to say that our vector is just a matrix with 1 column. Hence, there is no need to store the number of columns as a runtime variable. 110 111 When you call VectorXf::data() to get the pointer to the array of coefficients, it returns DenseStorage::data() which returns the \a m_data member. 112 113 When you call VectorXf::size() to get the size of the vector, this is actually a method in the base class MatrixBase. It determines that the vector is a column-vector, since ColsAtCompileTime==1 (this comes from the template parameters in the typedef VectorXf). It deduces that the size is the number of rows, so it returns VectorXf::rows(), which returns DenseStorage::rows(), which returns the \a m_rows member, which was set to \a size by the constructor. 114 115 \section ConstructionOfSumXpr Construction of the sum expression 116 117 Now that our vectors are constructed, let's move on to the next line: 118 119 \code 120 u = v + w; 121 \endcode 122 123 The executive summary is that operator+ returns a "sum of vectors" expression, but doesn't actually perform the computation. It is the operator=, whose call occurs thereafter, that does the computation. 124 125 Let us now see what Eigen does when it sees this: 126 127 \code 128 v + w 129 \endcode 130 131 Here, v and w are of type VectorXf, which is a typedef for a specialization of Matrix (as we explained above), which is a subclass of MatrixBase. So what is being called is 132 133 \code 134 MatrixBase::operator+(const MatrixBase&) 135 \endcode 136 137 The return type of this operator is 138 \code 139 CwiseBinaryOp<internal::scalar_sum_op<float>, VectorXf, VectorXf> 140 \endcode 141 The CwiseBinaryOp class is our first encounter with an expression template. As we said, the operator+ doesn't by itself perform any computation, it just returns an abstract "sum of vectors" expression. Since there are also "difference of vectors" and "coefficient-wise product of vectors" expressions, we unify them all as "coefficient-wise binary operations", which we abbreviate as "CwiseBinaryOp". "Coefficient-wise" means that the operations is performed coefficient by coefficient. "binary" means that there are two operands -- we are adding two vectors with one another. 142 143 Now you might ask, what if we did something like 144 145 \code 146 v + w + u; 147 \endcode 148 149 The first v + w would return a CwiseBinaryOp as above, so in order for this to compile, we'd need to define an operator+ also in the class CwiseBinaryOp... at this point it starts looking like a nightmare: are we going to have to define all operators in each of the expression classes (as you guessed, CwiseBinaryOp is only one of many) ? This looks like a dead end! 150 151 The solution is that CwiseBinaryOp itself, as well as Matrix and all the other expression types, is a subclass of MatrixBase. So it is enough to define once and for all the operators in class MatrixBase. 152 153 Since MatrixBase is the common base class of different subclasses, the aspects that depend on the subclass must be abstracted from MatrixBase. This is called polymorphism. 154 155 The classical approach to polymorphism in C++ is by means of virtual functions. This is dynamic polymorphism. Here we don't want dynamic polymorphism because the whole design of Eigen is based around the assumption that all the complexity, all the abstraction, gets resolved at compile-time. This is crucial: if the abstraction can't get resolved at compile-time, Eigen's compile-time optimization mechanisms become useless, not to mention that if that abstraction has to be resolved at runtime it'll incur an overhead by itself. 156 157 Here, what we want is to have a single class MatrixBase as the base of many subclasses, in such a way that each MatrixBase object (be it a matrix, or vector, or any kind of expression) knows at compile-time (as opposed to run-time) of which particular subclass it is an object (i.e. whether it is a matrix, or an expression, and what kind of expression). 158 159 The solution is the <a href="http://en.wikipedia.org/wiki/Curiously_Recurring_Template_Pattern">Curiously Recurring Template Pattern</a>. Let's do the break now. Hopefully you can read this wikipedia page during the break if needed, but it won't be allowed during the exam. 160 161 In short, MatrixBase takes a template parameter \a Derived. Whenever we define a subclass Subclass, we actually make Subclass inherit MatrixBase\<Subclass\>. The point is that different subclasses inherit different MatrixBase types. Thanks to this, whenever we have an object of a subclass, and we call on it some MatrixBase method, we still remember even from inside the MatrixBase method which particular subclass we're talking about. 162 163 This means that we can put almost all the methods and operators in the base class MatrixBase, and have only the bare minimum in the subclasses. If you look at the subclasses in Eigen, like for instance the CwiseBinaryOp class, they have very few methods. There are coeff() and sometimes coeffRef() methods for access to the coefficients, there are rows() and cols() methods returning the number of rows and columns, but there isn't much more than that. All the meat is in MatrixBase, so it only needs to be coded once for all kinds of expressions, matrices, and vectors. 164 165 So let's end this digression and come back to the piece of code from our example program that we were currently analyzing, 166 167 \code 168 v + w 169 \endcode 170 171 Now that MatrixBase is a good friend, let's write fully the prototype of the operator+ that gets called here (this code is from src/Core/MatrixBase.h): 172 173 \code 174 template<typename Derived> 175 class MatrixBase 176 { 177 // ... 178 179 template<typename OtherDerived> 180 const CwiseBinaryOp<internal::scalar_sum_op<typename internal::traits<Derived>::Scalar>, Derived, OtherDerived> 181 operator+(const MatrixBase<OtherDerived> &other) const; 182 183 // ... 184 }; 185 \endcode 186 187 Here of course, \a Derived and \a OtherDerived are VectorXf. 188 189 As we said, CwiseBinaryOp is also used for other operations such as substration, so it takes another template parameter determining the operation that will be applied to coefficients. This template parameter is a functor, that is, a class in which we have an operator() so it behaves like a function. Here, the functor used is internal::scalar_sum_op. It is defined in src/Core/Functors.h. 190 191 Let us now explain the internal::traits here. The internal::scalar_sum_op class takes one template parameter: the type of the numbers to handle. Here of course we want to pass the scalar type (a.k.a. numeric type) of VectorXf, which is \c float. How do we determine which is the scalar type of \a Derived ? Throughout Eigen, all matrix and expression types define a typedef \a Scalar which gives its scalar type. For example, VectorXf::Scalar is a typedef for \c float. So here, if life was easy, we could find the numeric type of \a Derived as just 192 \code 193 typename Derived::Scalar 194 \endcode 195 Unfortunately, we can't do that here, as the compiler would complain that the type Derived hasn't yet been defined. So we use a workaround: in src/Core/util/ForwardDeclarations.h, we declared (not defined!) all our subclasses, like Matrix, and we also declared the following class template: 196 \code 197 template<typename T> struct internal::traits; 198 \endcode 199 In src/Core/Matrix.h, right \em before the definition of class Matrix, we define a partial specialization of internal::traits for T=Matrix\<any template parameters\>. In this specialization of internal::traits, we define the Scalar typedef. So when we actually define Matrix, it is legal to refer to "typename internal::traits\<Matrix\>::Scalar". 200 201 Anyway, we have declared our operator+. In our case, where \a Derived and \a OtherDerived are VectorXf, the above declaration amounts to: 202 \code 203 class MatrixBase<VectorXf> 204 { 205 // ... 206 207 const CwiseBinaryOp<internal::scalar_sum_op<float>, VectorXf, VectorXf> 208 operator+(const MatrixBase<VectorXf> &other) const; 209 210 // ... 211 }; 212 \endcode 213 214 Let's now jump to src/Core/CwiseBinaryOp.h to see how it is defined. As you can see there, all it does is to return a CwiseBinaryOp object, and this object is just storing references to the left-hand-side and right-hand-side expressions -- here, these are the vectors \a v and \a w. Well, the CwiseBinaryOp object is also storing an instance of the (empty) functor class, but you shouldn't worry about it as that is a minor implementation detail. 215 216 Thus, the operator+ hasn't performed any actual computation. To summarize, the operation \a v + \a w just returned an object of type CwiseBinaryOp which did nothing else than just storing references to \a v and \a w. 217 218 \section Assignment The assignment 219 220 At this point, the expression \a v + \a w has finished evaluating, so, in the process of compiling the line of code 221 \code 222 u = v + w; 223 \endcode 224 we now enter the operator=. 225 226 What operator= is being called here? The vector u is an object of class VectorXf, i.e. Matrix. In src/Core/Matrix.h, inside the definition of class Matrix, we see this: 227 \code 228 template<typename OtherDerived> 229 inline Matrix& operator=(const MatrixBase<OtherDerived>& other) 230 { 231 eigen_assert(m_storage.data()!=0 && "you cannot use operator= with a non initialized matrix (instead use set()"); 232 return Base::operator=(other.derived()); 233 } 234 \endcode 235 Here, Base is a typedef for MatrixBase\<Matrix\>. So, what is being called is the operator= of MatrixBase. Let's see its prototype in src/Core/MatrixBase.h: 236 \code 237 template<typename OtherDerived> 238 Derived& operator=(const MatrixBase<OtherDerived>& other); 239 \endcode 240 Here, \a Derived is VectorXf (since u is a VectorXf) and \a OtherDerived is CwiseBinaryOp. More specifically, as explained in the previous section, \a OtherDerived is: 241 \code 242 CwiseBinaryOp<internal::scalar_sum_op<float>, VectorXf, VectorXf> 243 \endcode 244 So the full prototype of the operator= being called is: 245 \code 246 VectorXf& MatrixBase<VectorXf>::operator=(const MatrixBase<CwiseBinaryOp<internal::scalar_sum_op<float>, VectorXf, VectorXf> > & other); 247 \endcode 248 This operator= literally reads "copying a sum of two VectorXf's into another VectorXf". 249 250 Let's now look at the implementation of this operator=. It resides in the file src/Core/Assign.h. 251 252 What we can see there is: 253 \code 254 template<typename Derived> 255 template<typename OtherDerived> 256 inline Derived& MatrixBase<Derived> 257 ::operator=(const MatrixBase<OtherDerived>& other) 258 { 259 return internal::assign_selector<Derived,OtherDerived>::run(derived(), other.derived()); 260 } 261 \endcode 262 263 OK so our next task is to understand internal::assign_selector :) 264 265 Here is its declaration (all that is still in the same file src/Core/Assign.h) 266 \code 267 template<typename Derived, typename OtherDerived, 268 bool EvalBeforeAssigning = int(OtherDerived::Flags) & EvalBeforeAssigningBit, 269 bool NeedToTranspose = Derived::IsVectorAtCompileTime 270 && OtherDerived::IsVectorAtCompileTime 271 && int(Derived::RowsAtCompileTime) == int(OtherDerived::ColsAtCompileTime) 272 && int(Derived::ColsAtCompileTime) == int(OtherDerived::RowsAtCompileTime) 273 && int(Derived::SizeAtCompileTime) != 1> 274 struct internal::assign_selector; 275 \endcode 276 277 So internal::assign_selector takes 4 template parameters, but the 2 last ones are automatically determined by the 2 first ones. 278 279 EvalBeforeAssigning is here to enforce the EvalBeforeAssigningBit. As explained <a href="TopicLazyEvaluation.html">here</a>, certain expressions have this flag which makes them automatically evaluate into temporaries before assigning them to another expression. This is the case of the Product expression, in order to avoid strange aliasing effects when doing "m = m * m;" However, of course here our CwiseBinaryOp expression doesn't have the EvalBeforeAssigningBit: we said since the beginning that we didn't want a temporary to be introduced here. So if you go to src/Core/CwiseBinaryOp.h, you'll see that the Flags in internal::traits\<CwiseBinaryOp\> don't include the EvalBeforeAssigningBit. The Flags member of CwiseBinaryOp is then imported from the internal::traits by the EIGEN_GENERIC_PUBLIC_INTERFACE macro. Anyway, here the template parameter EvalBeforeAssigning has the value \c false. 280 281 NeedToTranspose is here for the case where the user wants to copy a row-vector into a column-vector. We allow this as a special exception to the general rule that in assignments we require the dimesions to match. Anyway, here both the left-hand and right-hand sides are column vectors, in the sense that ColsAtCompileTime is equal to 1. So NeedToTranspose is \c false too. 282 283 So, here we are in the partial specialization: 284 \code 285 internal::assign_selector<Derived, OtherDerived, false, false> 286 \endcode 287 288 Here's how it is defined: 289 \code 290 template<typename Derived, typename OtherDerived> 291 struct internal::assign_selector<Derived,OtherDerived,false,false> { 292 static Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); } 293 }; 294 \endcode 295 296 OK so now our next job is to understand how lazyAssign works :) 297 298 \code 299 template<typename Derived> 300 template<typename OtherDerived> 301 inline Derived& MatrixBase<Derived> 302 ::lazyAssign(const MatrixBase<OtherDerived>& other) 303 { 304 EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived) 305 eigen_assert(rows() == other.rows() && cols() == other.cols()); 306 internal::assign_impl<Derived, OtherDerived>::run(derived(),other.derived()); 307 return derived(); 308 } 309 \endcode 310 311 What do we see here? Some assertions, and then the only interesting line is: 312 \code 313 internal::assign_impl<Derived, OtherDerived>::run(derived(),other.derived()); 314 \endcode 315 316 OK so now we want to know what is inside internal::assign_impl. 317 318 Here is its declaration: 319 \code 320 template<typename Derived1, typename Derived2, 321 int Vectorization = internal::assign_traits<Derived1, Derived2>::Vectorization, 322 int Unrolling = internal::assign_traits<Derived1, Derived2>::Unrolling> 323 struct internal::assign_impl; 324 \endcode 325 Again, internal::assign_selector takes 4 template parameters, but the 2 last ones are automatically determined by the 2 first ones. 326 327 These two parameters \a Vectorization and \a Unrolling are determined by a helper class internal::assign_traits. Its job is to determine which vectorization strategy to use (that is \a Vectorization) and which unrolling strategy to use (that is \a Unrolling). 328 329 We'll not enter into the details of how these strategies are chosen (this is in the implementation of internal::assign_traits at the top of the same file). Let's just say that here \a Vectorization has the value \a LinearVectorization, and \a Unrolling has the value \a NoUnrolling (the latter is obvious since our vectors have dynamic size so there's no way to unroll the loop at compile-time). 330 331 So the partial specialization of internal::assign_impl that we're looking at is: 332 \code 333 internal::assign_impl<Derived1, Derived2, LinearVectorization, NoUnrolling> 334 \endcode 335 336 Here is how it's defined: 337 \code 338 template<typename Derived1, typename Derived2> 339 struct internal::assign_impl<Derived1, Derived2, LinearVectorization, NoUnrolling> 340 { 341 static void run(Derived1 &dst, const Derived2 &src) 342 { 343 const int size = dst.size(); 344 const int packetSize = internal::packet_traits<typename Derived1::Scalar>::size; 345 const int alignedStart = internal::assign_traits<Derived1,Derived2>::DstIsAligned ? 0 346 : internal::first_aligned(&dst.coeffRef(0), size); 347 const int alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize; 348 349 for(int index = 0; index < alignedStart; index++) 350 dst.copyCoeff(index, src); 351 352 for(int index = alignedStart; index < alignedEnd; index += packetSize) 353 { 354 dst.template copyPacket<Derived2, Aligned, internal::assign_traits<Derived1,Derived2>::SrcAlignment>(index, src); 355 } 356 357 for(int index = alignedEnd; index < size; index++) 358 dst.copyCoeff(index, src); 359 } 360 }; 361 \endcode 362 363 Here's how it works. \a LinearVectorization means that the left-hand and right-hand side expression can be accessed linearly i.e. you can refer to their coefficients by one integer \a index, as opposed to having to refer to its coefficients by two integers \a row, \a column. 364 365 As we said at the beginning, vectorization works with blocks of 4 floats. Here, \a PacketSize is 4. 366 367 There are two potential problems that we need to deal with: 368 \li first, vectorization works much better if the packets are 128-bit-aligned. This is especially important for write access. So when writing to the coefficients of \a dst, we want to group these coefficients by packets of 4 such that each of these packets is 128-bit-aligned. In general, this requires to skip a few coefficients at the beginning of \a dst. This is the purpose of \a alignedStart. We then copy these first few coefficients one by one, not by packets. However, in our case, the \a dst expression is a VectorXf and remember that in the construction of the vectors we allocated aligned arrays. Thanks to \a DstIsAligned, Eigen remembers that without having to do any runtime check, so \a alignedStart is zero and this part is avoided altogether. 369 \li second, the number of coefficients to copy is not in general a multiple of \a packetSize. Here, there are 50 coefficients to copy and \a packetSize is 4. So we'll have to copy the last 2 coefficients one by one, not by packets. Here, \a alignedEnd is 48. 370 371 Now come the actual loops. 372 373 First, the vectorized part: the 48 first coefficients out of 50 will be copied by packets of 4: 374 \code 375 for(int index = alignedStart; index < alignedEnd; index += packetSize) 376 { 377 dst.template copyPacket<Derived2, Aligned, internal::assign_traits<Derived1,Derived2>::SrcAlignment>(index, src); 378 } 379 \endcode 380 381 What is copyPacket? It is defined in src/Core/Coeffs.h: 382 \code 383 template<typename Derived> 384 template<typename OtherDerived, int StoreMode, int LoadMode> 385 inline void MatrixBase<Derived>::copyPacket(int index, const MatrixBase<OtherDerived>& other) 386 { 387 eigen_internal_assert(index >= 0 && index < size()); 388 derived().template writePacket<StoreMode>(index, 389 other.derived().template packet<LoadMode>(index)); 390 } 391 \endcode 392 393 OK, what are writePacket() and packet() here? 394 395 First, writePacket() here is a method on the left-hand side VectorXf. So we go to src/Core/Matrix.h to look at its definition: 396 \code 397 template<int StoreMode> 398 inline void writePacket(int index, const PacketScalar& x) 399 { 400 internal::pstoret<Scalar, PacketScalar, StoreMode>(m_storage.data() + index, x); 401 } 402 \endcode 403 Here, \a StoreMode is \a #Aligned, indicating that we are doing a 128-bit-aligned write access, \a PacketScalar is a type representing a "SSE packet of 4 floats" and internal::pstoret is a function writing such a packet in memory. Their definitions are architecture-specific, we find them in src/Core/arch/SSE/PacketMath.h: 404 405 The line in src/Core/arch/SSE/PacketMath.h that determines the PacketScalar type (via a typedef in Matrix.h) is: 406 \code 407 template<> struct internal::packet_traits<float> { typedef __m128 type; enum {size=4}; }; 408 \endcode 409 Here, __m128 is a SSE-specific type. Notice that the enum \a size here is what was used to define \a packetSize above. 410 411 And here is the implementation of internal::pstoret: 412 \code 413 template<> inline void internal::pstore(float* to, const __m128& from) { _mm_store_ps(to, from); } 414 \endcode 415 Here, __mm_store_ps is a SSE-specific intrinsic function, representing a single SSE instruction. The difference between internal::pstore and internal::pstoret is that internal::pstoret is a dispatcher handling both the aligned and unaligned cases, you find its definition in src/Core/GenericPacketMath.h: 416 \code 417 template<typename Scalar, typename Packet, int LoadMode> 418 inline void internal::pstoret(Scalar* to, const Packet& from) 419 { 420 if(LoadMode == Aligned) 421 internal::pstore(to, from); 422 else 423 internal::pstoreu(to, from); 424 } 425 \endcode 426 427 OK, that explains how writePacket() works. Now let's look into the packet() call. Remember that we are analyzing this line of code inside copyPacket(): 428 \code 429 derived().template writePacket<StoreMode>(index, 430 other.derived().template packet<LoadMode>(index)); 431 \endcode 432 433 Here, \a other is our sum expression \a v + \a w. The .derived() is just casting from MatrixBase to the subclass which here is CwiseBinaryOp. So let's go to src/Core/CwiseBinaryOp.h: 434 \code 435 class CwiseBinaryOp 436 { 437 // ... 438 template<int LoadMode> 439 inline PacketScalar packet(int index) const 440 { 441 return m_functor.packetOp(m_lhs.template packet<LoadMode>(index), m_rhs.template packet<LoadMode>(index)); 442 } 443 }; 444 \endcode 445 Here, \a m_lhs is the vector \a v, and \a m_rhs is the vector \a w. So the packet() function here is Matrix::packet(). The template parameter \a LoadMode is \a #Aligned. So we're looking at 446 \code 447 class Matrix 448 { 449 // ... 450 template<int LoadMode> 451 inline PacketScalar packet(int index) const 452 { 453 return internal::ploadt<Scalar, LoadMode>(m_storage.data() + index); 454 } 455 }; 456 \endcode 457 We let you look up the definition of internal::ploadt in GenericPacketMath.h and the internal::pload in src/Core/arch/SSE/PacketMath.h. It is very similar to the above for internal::pstore. 458 459 Let's go back to CwiseBinaryOp::packet(). Once the packets from the vectors \a v and \a w have been returned, what does this function do? It calls m_functor.packetOp() on them. What is m_functor? Here we must remember what particular template specialization of CwiseBinaryOp we're dealing with: 460 \code 461 CwiseBinaryOp<internal::scalar_sum_op<float>, VectorXf, VectorXf> 462 \endcode 463 So m_functor is an object of the empty class internal::scalar_sum_op<float>. As we mentioned above, don't worry about why we constructed an object of this empty class at all -- it's an implementation detail, the point is that some other functors need to store member data. 464 465 Anyway, internal::scalar_sum_op is defined in src/Core/Functors.h: 466 \code 467 template<typename Scalar> struct internal::scalar_sum_op EIGEN_EMPTY_STRUCT { 468 inline const Scalar operator() (const Scalar& a, const Scalar& b) const { return a + b; } 469 template<typename PacketScalar> 470 inline const PacketScalar packetOp(const PacketScalar& a, const PacketScalar& b) const 471 { return internal::padd(a,b); } 472 }; 473 \endcode 474 As you can see, all what packetOp() does is to call internal::padd on the two packets. Here is the definition of internal::padd from src/Core/arch/SSE/PacketMath.h: 475 \code 476 template<> inline __m128 internal::padd(const __m128& a, const __m128& b) { return _mm_add_ps(a,b); } 477 \endcode 478 Here, _mm_add_ps is a SSE-specific intrinsic function, representing a single SSE instruction. 479 480 To summarize, the loop 481 \code 482 for(int index = alignedStart; index < alignedEnd; index += packetSize) 483 { 484 dst.template copyPacket<Derived2, Aligned, internal::assign_traits<Derived1,Derived2>::SrcAlignment>(index, src); 485 } 486 \endcode 487 has been compiled to the following code: for \a index going from 0 to the 11 ( = 48/4 - 1), read the i-th packet (of 4 floats) from the vector v and the i-th packet from the vector w using two __mm_load_ps SSE instructions, then add them together using a __mm_add_ps instruction, then store the result using a __mm_store_ps instruction. 488 489 There remains the second loop handling the last few (here, the last 2) coefficients: 490 \code 491 for(int index = alignedEnd; index < size; index++) 492 dst.copyCoeff(index, src); 493 \endcode 494 However, it works just like the one we just explained, it is just simpler because there is no SSE vectorization involved here. copyPacket() becomes copyCoeff(), packet() becomes coeff(), writePacket() becomes coeffRef(). If you followed us this far, you can probably understand this part by yourself. 495 496 We see that all the C++ abstraction of Eigen goes away during compilation and that we indeed are precisely controlling which assembly instructions we emit. Such is the beauty of C++! Since we have such precise control over the emitted assembly instructions, but such complex logic to choose the right instructions, we can say that Eigen really behaves like an optimizing compiler. If you prefer, you could say that Eigen behaves like a script for the compiler. In a sense, C++ template metaprogramming is scripting the compiler -- and it's been shown that this scripting language is Turing-complete. See <a href="http://en.wikipedia.org/wiki/Template_metaprogramming"> Wikipedia</a>. 497 498 */ 499 500 } 501