Update Sphinx documentation, commit 33d7f6e [skip ci].

_sources/changelog.rst

@@ -10,6 +10,8 @@ New
 - Add thermonets: neural, differentiable, high-performance
   models for the Earth's thermosphere density
   (`#425 <https://github.com/bluescarni/heyoka/pull/425>`__).
+- Add support for variational ODE systems and Taylor map computation
+  (`#422 <https://github.com/bluescarni/heyoka/pull/422>`__).
 - Introduce the ability to represent unknown functions
   and their derivatives in the expression system
   (`#416 <https://github.com/bluescarni/heyoka/pull/416>`__).

_sources/tut_batch_mode.rst

@@ -18,13 +18,14 @@ batch mode can lead to an increase in floating-point throughput up to a factor o
 It is important to emphasise that batch mode does not reduce
 the CPU time required to integrate a system of ODEs. Rather, as a fine-grained
 form of data parallelism, batch mode allows to integrate multiple ODE systems in parallel
-at no additional cost, and it is thus most useful when the need arise
+at (almost) no additional cost, and it is thus most useful when the need arise
 to integrate the same ODE system with different initial conditions and parameters.
 
-Although batch mode can in principle be used with all floating-point types supported
+Although batch mode can in principle be used with all the fundamental C++ floating-point types supported
 by heyoka, in practice at this time no CPU provides SIMD instructions for extended-precision
 datatypes. Thus, here we will consider the application of batch mode only to
-standard ``double`` precision computations.
+standard ``double`` precision computations (keeping in mind that batch mode is also supported
+for single-precision ``float`` computations).
 
 The value of the batch size :math:`n` can be freely chosen by the user. In order
 to achieve optimal performance, however, :math:`n` should match the SIMD width of the

changelog.html

@@ -594,6 +594,8 @@ <h3>New<a class="headerlink" href="#new" title="Link to this heading">#</a></h3>
 <li><p>Add thermonets: neural, differentiable, high-performance
 models for the Earth’s thermosphere density
 (<a class="reference external" href="https://github.com/bluescarni/heyoka/pull/425">#425</a>).</p></li>
+<li><p>Add support for variational ODE systems and Taylor map computation
+(<a class="reference external" href="https://github.com/bluescarni/heyoka/pull/422">#422</a>).</p></li>
 <li><p>Introduce the ability to represent unknown functions
 and their derivatives in the expression system
 (<a class="reference external" href="https://github.com/bluescarni/heyoka/pull/416">#416</a>).</p></li>

tut_batch_mode.html

@@ -434,12 +434,13 @@ <h2> Contents </h2>
 <p>It is important to emphasise that batch mode does not reduce
 the CPU time required to integrate a system of ODEs. Rather, as a fine-grained
 form of data parallelism, batch mode allows to integrate multiple ODE systems in parallel
-at no additional cost, and it is thus most useful when the need arise
+at (almost) no additional cost, and it is thus most useful when the need arise
 to integrate the same ODE system with different initial conditions and parameters.</p>
-<p>Although batch mode can in principle be used with all floating-point types supported
+<p>Although batch mode can in principle be used with all the fundamental C++ floating-point types supported
 by heyoka, in practice at this time no CPU provides SIMD instructions for extended-precision
 datatypes. Thus, here we will consider the application of batch mode only to
-standard <code class="docutils literal notranslate"><span class="pre">double</span></code> precision computations.</p>
+standard <code class="docutils literal notranslate"><span class="pre">double</span></code> precision computations (keeping in mind that batch mode is also supported
+for single-precision <code class="docutils literal notranslate"><span class="pre">float</span></code> computations).</p>
 <p>The value of the batch size <span class="math notranslate nohighlight">\(n\)</span> can be freely chosen by the user. In order
 to achieve optimal performance, however, <span class="math notranslate nohighlight">\(n\)</span> should match the SIMD width of the
 processor in use. Because at this time the most widespread SIMD instruction set is