Home | History | Annotate | Download | only in doc
      1 namespace Eigen {
      2 
      3 /** \eigenManualPage TopicUnalignedArrayAssert Explanation of the assertion on unaligned arrays
      4 
      5 Hello! You are seeing this webpage because your program terminated on an assertion failure like this one:
      6 <pre>
      7 my_program: path/to/eigen/Eigen/src/Core/DenseStorage.h:44:
      8 Eigen::internal::matrix_array<T, Size, MatrixOptions, Align>::internal::matrix_array()
      9 [with T = double, int Size = 2, int MatrixOptions = 2, bool Align = true]:
     10 Assertion `(reinterpret_cast<size_t>(array) & (sizemask)) == 0 && "this assertion
     11 is explained here: http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html
     12 **** READ THIS WEB PAGE !!! ****"' failed.
     13 </pre>
     14 
     15 There are 4 known causes for this issue. Please read on to understand them and learn how to fix them.
     16 
     17 \eigenAutoToc
     18 
     19 \section where Where in my own code is the cause of the problem?
     20 
     21 First of all, you need to find out where in your own code this assertion was triggered from. At first glance, the error message doesn't look helpful, as it refers to a file inside Eigen! However, since your program crashed, if you can reproduce the crash, you can get a backtrace using any debugger. For example, if you're using GCC, you can use the GDB debugger as follows:
     22 \code
     23 $ gdb ./my_program          # Start GDB on your program
     24 > run                       # Start running your program
     25 ...                         # Now reproduce the crash!
     26 > bt                        # Obtain the backtrace
     27 \endcode
     28 Now that you know precisely where in your own code the problem is happening, read on to understand what you need to change.
     29 
     30 \section c1 Cause 1: Structures having Eigen objects as members
     31 
     32 If you have code like this,
     33 
     34 \code
     35 class Foo
     36 {
     37   //...
     38   Eigen::Vector2d v;
     39   //...
     40 };
     41 //...
     42 Foo *foo = new Foo;
     43 \endcode
     44 
     45 then you need to read this separate page: \ref TopicStructHavingEigenMembers "Structures Having Eigen Members".
     46 
     47 Note that here, Eigen::Vector2d is only used as an example, more generally the issue arises for all \ref TopicFixedSizeVectorizable "fixed-size vectorizable Eigen types".
     48 
     49 \section c2 Cause 2: STL Containers or manual memory allocation
     50 
     51 If you use STL Containers such as std::vector, std::map, ..., with %Eigen objects, or with classes containing %Eigen objects, like this,
     52 
     53 \code
     54 std::vector<Eigen::Matrix2f> my_vector;
     55 struct my_class { ... Eigen::Matrix2f m; ... };
     56 std::map<int, my_class> my_map;
     57 \endcode
     58 
     59 then you need to read this separate page: \ref TopicStlContainers "Using STL Containers with Eigen".
     60 
     61 Note that here, Eigen::Matrix2f is only used as an example, more generally the issue arises for all \ref TopicFixedSizeVectorizable "fixed-size vectorizable Eigen types" and \ref TopicStructHavingEigenMembers "structures having such Eigen objects as member".
     62 
     63 The same issue will be exhibited by any classes/functions by-passing operator new to allocate memory, that is, by performing custom memory allocation followed by calls to the placement new operator. This is for instance typically the case of \c std::make_shared or \c std::allocate_shared for which is the solution is to use an \ref aligned_allocator "aligned allocator" as detailed in the \ref TopicStlContainers "solution for STL containers".
     64 
     65 \section c3 Cause 3: Passing Eigen objects by value
     66 
     67 If some function in your code is getting an Eigen object passed by value, like this,
     68 
     69 \code
     70 void func(Eigen::Vector4d v);
     71 \endcode
     72 
     73 then you need to read this separate page: \ref TopicPassingByValue "Passing Eigen objects by value to functions".
     74 
     75 Note that here, Eigen::Vector4d is only used as an example, more generally the issue arises for all \ref TopicFixedSizeVectorizable "fixed-size vectorizable Eigen types".
     76 
     77 \section c4 Cause 4: Compiler making a wrong assumption on stack alignment (for instance GCC on Windows)
     78 
     79 This is a must-read for people using GCC on Windows (like MinGW or TDM-GCC). If you have this assertion failure in an innocent function declaring a local variable like this:
     80 
     81 \code
     82 void foo()
     83 {
     84   Eigen::Quaternionf q;
     85   //...
     86 }
     87 \endcode
     88 
     89 then you need to read this separate page: \ref TopicWrongStackAlignment "Compiler making a wrong assumption on stack alignment".
     90 
     91 Note that here, Eigen::Quaternionf is only used as an example, more generally the issue arises for all \ref TopicFixedSizeVectorizable "fixed-size vectorizable Eigen types".
     92 
     93 \section explanation General explanation of this assertion
     94 
     95 \ref TopicFixedSizeVectorizable "fixed-size vectorizable Eigen objects" must absolutely be created at 16-byte-aligned locations, otherwise SIMD instructions addressing them will crash.
     96 
     97 Eigen normally takes care of these alignment issues for you, by setting an alignment attribute on them and by overloading their "operator new".
     98 
     99 However there are a few corner cases where these alignment settings get overridden: they are the possible causes for this assertion.
    100 
    101 \section getrid I don't care about optimal vectorization, how do I get rid of that stuff?
    102 
    103 Three possibilities:
    104 <ul>
    105   <li>Use the \c DontAlign option to Matrix, Array, Quaternion, etc. objects that gives you trouble. This way Eigen won't try to align them, and thus won"t assume any special alignment. On the down side, you will pay the cost of unaligned loads/stores for them, but on modern CPUs, the overhead is either null or marginal. See \link StructHavingEigenMembers_othersolutions here \endlink for an example.</li>
    106   <li>Define \link TopicPreprocessorDirectivesPerformance EIGEN_DONT_ALIGN_STATICALLY \endlink. That disables all 16-byte (and above) static alignment code, while keeping 16-byte (or above) heap alignment. This has the effect of
    107       vectorizing fixed-size objects (like Matrix4d) through unaligned stores (as controlled by \link TopicPreprocessorDirectivesPerformance EIGEN_UNALIGNED_VECTORIZE \endlink), while keeping unchanged the vectorization of dynamic-size objects
    108       (like MatrixXd). But do note that this breaks ABI compatibility with the default behavior of static alignment.</li>
    109   <li>Or define both \link TopicPreprocessorDirectivesPerformance  EIGEN_DONT_VECTORIZE \endlink and EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT. This keeps the
    110       16-byte alignment code and thus preserves ABI compatibility, but completely disables vectorization.</li>
    111 </ul>
    112 
    113 If you want to know why defining EIGEN_DONT_VECTORIZE does not by itself disable 16-byte alignment and the assertion, here's the explanation:
    114 
    115 It doesn't disable the assertion, because otherwise code that runs fine without vectorization would suddenly crash when enabling vectorization.
    116 It doesn't disable 16-byte alignment, because that would mean that vectorized and non-vectorized code are not mutually ABI-compatible. This ABI compatibility is very important, even for people who develop only an in-house application, as for instance one may want to have in the same application a vectorized path and a non-vectorized path.
    117 
    118 */
    119 
    120 }
    121