Warn about using CUDA+Threads+march=native

diff --git a/doc/external-libs/cuda.html b/doc/external-libs/cuda.html
index 68222e1d312f344d58e5fa06e2bf54e366294f3a..fda03f4c9b7b0d5209fef8c37fa071aad754ae7b 100644 (file)
--- a/doc/external-libs/cuda.html
+++ b/doc/external-libs/cuda.html
@@ -72,5 +72,19 @@
       will lead to segmentation faults in MPI calls.
     </p>
 
+    <p>
+      Using CUDA in combination with architecture-specific C++ compiler flags
+      like <code>-march=native</code> is known to be fragile and there might be
+      compatibility issues with other libraries, e.g. using CUDA 10.1 with
+      <code>-DDEAL_II_WITH_THREADS=ON</code> and
+      <code>-DDEAL_II_CXX_FLAGS=-march=native</code> results in compile time
+      errors like:
+      <pre>
+
+        /usr/lib/gcc/x86_64-linux-gnu/7/include/avx512fintrin.h(11265): error: identifier "__builtin_ia32_scalefsd_round" is undefined
+        /usr/lib/gcc/x86_64-linux-gnu/7/include/avx512fintrin.h(11274): error: identifier "__builtin_ia32_scalefss_round" is undefined
+      </pre>
+    </p>
+
   </body>
 </html>