valarray-inst.cc (gslice::_Indexer::_Indexer): Don't flip lengths and strides.

        * src/valarray-inst.cc (gslice::_Indexer::_Indexer): Don't flip
        lengths and strides.
        (__gslice_to_index): Document.

From-SVN: r39583

1 files changed, 73 insertions, 49 deletions

diff --git a/libstdc++-v3/src/valarray-inst.cc b/libstdc++-v3/src/valarray-inst.cc
index 0587484..9950fba 100644
--- a/libstdc++-v3/src/valarray-inst.cc
+++ b/libstdc++-v3/src/valarray-inst.cc
@@ -2,59 +2,83 @@
 
 namespace std
 {
-    // Some explicit instanciations.
-    template void
-    __valarray_fill(size_t* __restrict__, size_t, const size_t&);
+  // Some explicit instanciations.
+  template void
+     __valarray_fill(size_t* __restrict__, size_t, const size_t&);
+  
+  template void
+     __valarray_copy(const size_t* __restrict__, size_t, size_t* __restrict__);
+  
+  template valarray<size_t>::valarray(size_t);
+  template valarray<size_t>::valarray(const valarray<size_t>&);
+  template valarray<size_t>::~valarray();
+  template size_t valarray<size_t>::size() const;
+  template size_t& valarray<size_t>::operator[](size_t);
 
-    template void
-    __valarray_copy(const size_t* __restrict__, size_t, size_t* __restrict__);
 
-    template valarray<size_t>::valarray(size_t);
-    template valarray<size_t>::valarray(const valarray<size_t>&);
-    template valarray<size_t>::~valarray();
-    template size_t valarray<size_t>::size() const;
-    template size_t& valarray<size_t>::operator[](size_t);
+  inline size_t
+  __valarray_product(const valarray<size_t>& __a)
+  {
+    typedef const size_t* __restrict__ _Tp;
+    const size_t __n = __a.size();
+    // XXX: This ugly cast is necessary because
+    //      valarray::operator[]() const return a VALUE!
+    //      Try to get the committee to correct that gross error.
+    valarray<size_t>& __t = const_cast<valarray<size_t>&>(__a);
+    return __valarray_product(&__t[0], &__t[0] + __n);
+  }
+  
+  // Map a gslice, described by its multidimensional LENGTHS
+  // and corresponding STRIDES, to a linear array of INDEXES
+  // for the purpose of indexing a flat, one-dimensional array
+  // representation of a gslice_array.
+  void
+  __gslice_to_index(size_t __o, const valarray<size_t>& __l,
+                    const valarray<size_t>& __s, valarray<size_t>& __i)
+  {
+    // There are as much as dimensions as there are strides.
+    size_t __n = __l.size();
 
+    // Get a buffer to hold current multi-index as we go through
+    // the gslice for the purpose of computing its linear-image.
+    size_t* const __t = static_cast<size_t*>
+      (__builtin_alloca(__n * sizeof (size_t)));
+    __valarray_fill(__t, __n, size_t(0));
 
-    inline size_t
-    __valarray_product(const valarray<size_t>& __a)
-    {
-        typedef const size_t* __restrict__ _Tp;
-        const size_t __n = __a.size();
-        // XXX: This ugly cast is necessary because
-        //      valarray::operator[]() const return a VALUE!
-        //      Try to get the committee to correct that gross error.
-        valarray<size_t>& __t = const_cast<valarray<size_t>&>(__a);
-        return __valarray_product(&__t[0], &__t[0] + __n);
-    }
+    // Note that this should match the product of all numbers appearing
+    // in __l which describes the multidimensional sizes of the
+    // the generalized slice.
+    const size_t __z = __i.size();
     
-    void __gslice_to_index(size_t __o, const valarray<size_t>& __l,
-                           const valarray<size_t>& __s,
-                           valarray<size_t>& __i)
-    {
-        size_t __n = __l.size();
-        size_t* const __t = static_cast<size_t*>
-            (__builtin_alloca(__n*sizeof(size_t)));
-        __valarray_fill(__t, __n, size_t(0));
-        const size_t __z = __i.size();
-        __valarray_fill(&__i[0], __z, __o);
-        for (size_t __j=0; __j<__z; ++__j) {
-            for (size_t __k=0; __k<__n; ++__k)
-                __i[__j] += __s[__k]*__t[__k];
-            ++__t[__n-1];
-            for (size_t __k=__n-1; __k; --__k) {
-                if (__t[__k] >= __l[__k]) {
-                    __t[__k] = 0;
-                    ++__t[__k-1];
-                }
-            }
-        }
-    }
-    
-    gslice::_Indexer::_Indexer(size_t __o, const valarray<size_t>& __s,
-                               const valarray<size_t>& __l)
-            : _M_count(1), _M_start(__o), _M_size(__s), _M_stride(__l),
-              _M_index(__l.size() ? __valarray_product(__l) : 0)
-    { __gslice_to_index(__o, __l, __s, _M_index); }
+    for (size_t __j = 0; __j < __z; ++__j)
+      {
+        // Compute the linear-index image of (t_0, ... t_{n-1}).
+        // Normaly, we should use inner_product<>(), but we do it the
+        // the hard way here to avoid link-time can of worms.
+        size_t __a = __o;
+        for (size_t __k = 0; __k < __n; ++__k)
+          __a += __s[__k] * __t[__k];
+
+        __i[__j] = __a;
 
+        // Process the next multi-index.  The loop ought to be
+        // backward since we're making a lexicagraphical visit.
+        ++__t[__n-1];
+        for (size_t __k=__n-1; __k; --__k)
+          {
+            if (__t[__k] >= __l[__k])
+              {
+                __t[__k] = 0;
+                ++__t[__k-1];
+              }
+          }
+      }
+  }
+  
+  gslice::_Indexer::_Indexer(size_t __o, const valarray<size_t>& __l,
+                             const valarray<size_t>& __s)
+      : _M_count(1), _M_start(__o), _M_size(__l), _M_stride(__s),
+        _M_index(__l.size() == 0 ? 0 : __valarray_product(__l))
+  { __gslice_to_index(__o, __l, __s, _M_index); }
+  
 }