patch bump

CMakeLists.txt

@@ -29,7 +29,7 @@
 ## NOTE: More details: https://arma.sourceforge.net/faq.html
 
 
-cmake_minimum_required(VERSION 2.8.12 FATAL_ERROR)
+cmake_minimum_required(VERSION 2.8.12...3.5 FATAL_ERROR)
 
 if(NOT (${CMAKE_MAJOR_VERSION} LESS 3))
   if(POLICY CMP0025)

docs.html

@@ -4280,7 +4280,7 @@ For form 2:
 <li>
 For form 3:
 <ul>
-<li>apply the given <i>lambda_function</i> to each slice</li>
+<li>apply the given <i>lambda_function</i> to each column vector or row vector</li>
 <li>the function must accept a reference to a <a href="#Col">Col</a> or <a href="#Row">Row</a> object with the same element type as the underlying matrix</li>
 </ul>
 </li>
@@ -8873,6 +8873,10 @@ Conversion of a <i>mat</i> object into <i>colvec</i>, <i>rowvec</i> or <i>std::v
 </li>
 <br>
 <li>
+<b>Caveat:</b> to convert an expression that results in a 1x1 matrix to a pure scalar value, use <a href="#as_scalar">as_scalar()</a>
+</li>
+<br>
+<li>
 Examples:
 <ul>
 <pre>
@@ -20222,6 +20226,7 @@ List of additions and changes for each version:
 <li>Version 12.8:
 <ul>
 <li>faster detection of symmetric expressions by <a href="#pinv">pinv()</a> and <a href="#rank">rank()</a></li>
+<li>faster handling of sparse submatrix views</li>
 <li>expanded <a href="#shift">shift()</a> to handle sparse matrices</li>
 <li>expanded <a href="#conv_to">conv_to</a> for more flexible conversions between sparse and dense matrices</li>
 <li>added <a href="#misc_fns">cbrt()</a></li>

include/armadillo

@@ -19,9 +19,9 @@
 #ifndef ARMA_INCLUDES
 #define ARMA_INCLUDES
 
-// NOTE: functions that are designed to be user accessible are described in the documentation (docs.html).
-// NOTE: all other functions and classes (ie. not explicitly described in the documentation)
-// NOTE: are considered as internal implementation details, and may be changed or removed without notice.
+// WARNING: the documentation (docs.html) describes the public API (functions, classes, constants);
+// WARNING: any functionality which is _not explicitly_ described in the documentation
+// WARNING: is considered as internal implementation detail, and may be changed or removed without notice.
 
 #include "armadillo_bits/config.hpp"
 #include "armadillo_bits/compiler_check.hpp"

include/armadillo_bits/SpSubview_meat.hpp

@@ -44,18 +44,28 @@ SpSubview<eT>::SpSubview(const SpMat<eT>& in_m, const uword in_row1, const uword
   
   m.sync_csc();
   
-  // There must be a O(1) way to do this
-  uword lend     = m.col_ptrs[in_col1 + in_n_cols];
-  uword lend_row = in_row1 + in_n_rows;
-  uword count   = 0;
+  // count the number of non-zeros in the subview
+  uword count = 0;
   
-  for(uword i = m.col_ptrs[in_col1]; i < lend; ++i)
+  if(n_rows == m.n_rows)
     {
-    const uword m_row_indices_i = m.row_indices[i];
+    count = m.col_ptrs[aux_col1 + n_cols] - m.col_ptrs[aux_col1];
+    }
+  else
+    {
+    arma_extra_debug_print("counting non-zeros in sparse subview");
     
-    const bool condition = (m_row_indices_i >= in_row1) && (m_row_indices_i < lend_row);
+    uword lend     = m.col_ptrs[in_col1 + in_n_cols];
+    uword lend_row = in_row1 + in_n_rows;
     
-    count += condition ? uword(1) : uword(0);
+    for(uword i = m.col_ptrs[in_col1]; i < lend; ++i)
+      {
+      const uword m_row_indices_i = m.row_indices[i];
+      
+      const bool condition = (m_row_indices_i >= in_row1) && (m_row_indices_i < lend_row);
+      
+      count += condition ? uword(1) : uword(0);
+      }
     }
   
   access::rw(n_nonzero) = count;

include/armadillo_bits/arma_version.hpp

@@ -23,7 +23,7 @@
 
 #define ARMA_VERSION_MAJOR 12
 #define ARMA_VERSION_MINOR 8
-#define ARMA_VERSION_PATCH 2
+#define ARMA_VERSION_PATCH 4
 #define ARMA_VERSION_NAME  "Cortisol Injector"
 
 

include/armadillo_bits/compiler_setup.hpp

@@ -377,14 +377,11 @@
 
 
 #if defined(__SUNPRO_CC)
-  
   // http://www.oracle.com/technetwork/server-storage/solarisstudio/training/index-jsp-141991.html
   // http://www.oracle.com/technetwork/server-storage/solarisstudio/documentation/cplusplus-faq-355066.html
-  
   #if (__SUNPRO_CC < 0x5140)
     #error "*** newer compiler required ***"
   #endif
-  
 #endif
 
 

include/armadillo_bits/config.hpp

@@ -201,6 +201,11 @@
 
 #if defined(ARMA_EXTRA_DEBUG)
   #undef  ARMA_NO_DEBUG
+  #undef  ARMA_DONT_CHECK_NONFINITE
+  
+  #undef  ARMA_CHECK_NONFINITE
+  #define ARMA_CHECK_NONFINITE
+  
   #undef  ARMA_WARN_LEVEL
   #define ARMA_WARN_LEVEL 3
 #endif

include/armadillo_bits/config.hpp.cmake

@@ -201,6 +201,11 @@
 
 #if defined(ARMA_EXTRA_DEBUG)
   #undef  ARMA_NO_DEBUG
+  #undef  ARMA_DONT_CHECK_NONFINITE
+  
+  #undef  ARMA_CHECK_NONFINITE
+  #define ARMA_CHECK_NONFINITE
+  
   #undef  ARMA_WARN_LEVEL
   #define ARMA_WARN_LEVEL 3
 #endif

include/armadillo_bits/debug.hpp

@@ -579,7 +579,7 @@ arma_assert_same_size(const uword A_n_rows, const uword A_n_cols, const uword B_
 
 
 
-//! stop if given matrices have different sizes
+//! stop if given matrices do not have the same size
 template<typename eT1, typename eT2>
 arma_hot
 inline
@@ -600,7 +600,7 @@ arma_assert_same_size(const Mat<eT1>& A, const Mat<eT2>& B, const char* x)
 
 
 
-//! stop if given proxies have different sizes
+//! stop if given proxies do not have the same size
 template<typename eT1, typename eT2>
 arma_hot
 inline
@@ -804,7 +804,7 @@ arma_assert_same_size(const uword A_n_rows, const uword A_n_cols, const uword A_
 
 
 
-//! stop if given cubes have different sizes
+//! stop if given cubes do not have the same size
 template<typename eT1, typename eT2>
 arma_hot
 inline
@@ -883,7 +883,7 @@ arma_assert_same_size(const subview_cube<eT>& A, const ProxyCube<T1>& B, const c
 
 
 
-//! stop if given cube proxies have different sizes
+//! stop if given cube proxies do not have the same size
 template<typename eT1, typename eT2>
 arma_hot
 inline

include/armadillo_bits/fft_engine_fftw3.hpp

@@ -23,6 +23,13 @@
 
 #if defined(ARMA_USE_FFTW3)
 
+struct fft_engine_fftw3_aux
+  {
+  #if (!defined(ARMA_DONT_USE_STD_MUTEX))
+  static inline std::mutex& get_plan_mutex() { static std::mutex plan_mutex; return plan_mutex; }
+  #endif
+  };
+
 template<typename cx_type, bool inverse>
 class fft_engine_fftw3
   {
@@ -74,7 +81,30 @@ class fft_engine_fftw3
     const int fftw3_flags = fftw3_flag_destroy | fftw3_flag_estimate;
     
     arma_extra_debug_print("fft_engine_fftw3::constructor: generating 1D plan");
-    fftw3_plan = fftw3::plan_dft_1d<cx_type>(N, X_work.memptr(), Y_work.memptr(), fftw3_sign, fftw3_flags);
+    
+    // only fftw3::execute() is thread safe, as per FFTW docs:
+    // https://www.fftw.org/fftw3_doc/Thread-safety.html
+    
+    #if defined(ARMA_USE_OPENMP)
+      {
+      #pragma omp critical (arma_fft_engine_fftw3)
+        {
+        fftw3_plan = fftw3::plan_dft_1d<cx_type>(N, X_work.memptr(), Y_work.memptr(), fftw3_sign, fftw3_flags);
+        }
+      }
+    #elif (!defined(ARMA_DONT_USE_STD_MUTEX))
+      {
+      std::mutex& plan_mutex = fft_engine_fftw3_aux::get_plan_mutex();
+      
+      const std::lock_guard<std::mutex> lock(plan_mutex);
+      
+      fftw3_plan = fftw3::plan_dft_1d<cx_type>(N, X_work.memptr(), Y_work.memptr(), fftw3_sign, fftw3_flags);
+      }
+    #else
+      {
+      fftw3_plan = fftw3::plan_dft_1d<cx_type>(N, X_work.memptr(), Y_work.memptr(), fftw3_sign, fftw3_flags);
+      }
+    #endif
     
     if(fftw3_plan == nullptr)  { arma_stop_runtime_error("fft_engine_fftw3::constructor: failed to create plan"); }
     }

include/armadillo_bits/fn_conv_to.hpp

@@ -30,7 +30,7 @@ class conv_to
   
   template<typename in_eT, typename T1>
   inline static out_eT from(const Base<in_eT, T1>& in, const typename arma_not_cx<in_eT>::result* junk = nullptr);
-
+  
   template<typename in_eT, typename T1>
   inline static out_eT from(const Base<in_eT, T1>& in, const typename arma_cx_only<in_eT>::result* junk = nullptr);
   

include/armadillo_bits/fn_dot.hpp

@@ -289,13 +289,33 @@ dot
   {
   arma_extra_debug_sigprint();
   
+  typedef typename T1::elem_type eT;
+  
+  if(is_SpSubview_col<T2>::value)
+    {
+    // TODO: refactor to use C++17 "if constexpr" to avoid reinterpret_cast shenanigans
+    
+    const SpSubview_col<eT>& yy = reinterpret_cast< const SpSubview_col<eT>& >(y);
+    
+    if(yy.n_rows == yy.m.n_rows)
+      {
+      arma_extra_debug_print("using sparse column vector specialisation");
+      
+      const quasi_unwrap<T1> U(x);
+      
+      arma_debug_assert_same_size(U.M.n_elem, uword(1), yy.n_elem, uword(1), "dot()");
+      
+      yy.m.sync();
+      
+      return dense_sparse_helper::dot(U.M.memptr(), yy.m, yy.aux_col1);
+      }
+    }
+  
   const   Proxy<T1> pa(x);
   const SpProxy<T2> pb(y);
   
   arma_debug_assert_same_size(pa.get_n_rows(), pa.get_n_cols(), pb.get_n_rows(), pb.get_n_cols(), "dot()");
   
-  typedef typename T1::elem_type eT;
-  
   eT result = eT(0);
   
   typename SpProxy<T2>::const_iterator_type it     = pb.begin();

tests2/spsubview.cpp

@@ -1538,3 +1538,27 @@ TEST_CASE("sp_subview_row_iterator_test_2")
 
   REQUIRE( true );
   }
+
+
+
+TEST_CASE("sp_subview_n_nonzero_test")
+  {
+  sp_mat x;
+  x.sprandu(100, 100, 0.3);
+  
+  sp_vec y = x.col(0);
+  
+  REQUIRE( y.n_nonzero == x.col(0).n_nonzero );
+  
+  sp_mat z = x.cols(0, 5);
+  
+  REQUIRE( z.n_nonzero == x.cols(0, 5).n_nonzero );
+  
+  z = x.cols(11, 74);
+  
+  REQUIRE( z.n_nonzero == x.cols(11, 74).n_nonzero );
+  
+  z = x.submat(12, 14, 77, 17);
+  
+  REQUIRE( z.n_nonzero == x.submat(12, 14, 77, 17).n_nonzero );
+  }