There is probably more to say - especially to Fortran and likely also to
C, C++, … but let's add the first OpenMP and AMD GPU entries for GCC 17.
I hope that we can soon add some more goodies, knowing that they are in
the pipeline, but that's at least a start. :-)
Comments & suggestions before I push it to be visible at
https://gcc.gnu.org/gcc-17/changes.html ?
Tobias
gcc-17/changes.html: Update OpenMP + GCN
diff --git a/htdocs/gcc-17/changes.html b/htdocs/gcc-17/changes.html
index 6a604048..8d1fa573 100644
--- a/htdocs/gcc-17/changes.html
+++ b/htdocs/gcc-17/changes.html
@@ -40,6 +40,27 @@ a work-in-progress.</p>
<!-- .................................................................. -->
<h2 id="languages">New Languages and Language specific improvements</h2>
+<h3 id="openmp">OpenMP</h3>
+
+<p>
+ See the
+ <a href="../projects/gomp/">GNU Offloading and Multi-Processing Project (GOMP)</a>
+ page for general information.
+</p>
+
+<ul>
+ <li><strong>OpenMP 5.0:</strong> Multiple clauses mapping the same variable is
+ now supported. Handling used-defined map operation (<code>declare
+ mapper</code>) in C and C++ has been improved.</li>
+ <li><strong>OpenMP 5.1:</strong> The <code>iterator</code> modifier to the
+ <code>map</code>, <code>to</code>, and <code>from</code> clauses has been
+ improved for C and C++ and initial support for Fortran added.</li>
+ <li><strong>TR15:</strong> The <code>omp_get_device_distances</code> API
+ routine has been added than can be used to find the offload devices that
+ are in terms of the NUMA node closed to the thread that called the
+ routine.</li>
+</ul>
+
<!-- <h3 id="ada">Ada</h3> -->
<!-- <h3 id="c-family">C family</h3> -->
@@ -49,7 +70,7 @@ a work-in-progress.</p>
<!-- <h3 id="d">D</h3> -->
-<!-- <h3 id="fortran">Fortran</h3> -->
+<h3 id="fortran">Fortran</h3>
<ul>
<li>The Fortran 2023 <code>c_f_strpointer</code> intrinsic is now
supported. Additionally, the <code>f_c_string</code> intrinsic
@@ -67,6 +88,13 @@ a work-in-progress.</p>
<!-- <h3 id="aarch64">AArch64</h3> -->
+<h3 id="amdgcn">AMD GPU (GCN)</h3>
+
+<ul>
+ <li>For offloading to AMD GPUs: The launch overhead of OpenMP target regions
+ and OpenACC compute regions has been further reduced compared with GCC 16.</li>
+</ul>
+
<!-- <h3 id="arc">ARC</h3> -->
<!-- <h3 id="arm">arm</h3> -->
