Reenabled SIMD kernels without cut-offs

In 4.6 there was a bug that caused incorrect results with group scheme
SIMD kernels with systems without cut-offs. This seems to have been
resolved already in 5.0.
Refs #1249.

Change-Id: I86c2b2ae097769472c1748e6e03a673a58a938eb

diff --git a/src/gromacs/mdlib/forcerec.cpp b/src/gromacs/mdlib/forcerec.cpp
index 7cde5b73b295ba129c457201269aab2f70542406..1333573e30fc5aa50c707f43ca03c2bba4f0e516 100644 (file)
--- a/src/gromacs/mdlib/forcerec.cpp
+++ b/src/gromacs/mdlib/forcerec.cpp
@@ -2480,23 +2480,18 @@ void init_forcerec(FILE              *fp,
     fr->AllvsAll_work   = NULL;
     fr->AllvsAll_workgb = NULL;
 
-    /* All-vs-all kernels have not been implemented in 4.6, and
-     * the SIMD group kernels are also buggy in this case. Non-SIMD
-     * group kernels are OK. See Redmine #1249. */
+    /* All-vs-all kernels have not been implemented in 4.6 and later.
+     * See Redmine #1249. */
     if (fr->bAllvsAll)
     {
         fr->bAllvsAll            = FALSE;
-        fr->use_simd_kernels     = FALSE;
         if (fp != NULL)
         {
             fprintf(fp,
                     "\nYour simulation settings would have triggered the efficient all-vs-all\n"
                     "kernels in GROMACS 4.5, but these have not been implemented in GROMACS\n"
-                    "4.6. Also, we can't use the accelerated SIMD kernels here because\n"
-                    "of an unfixed bug. The reference C kernels are correct, though, so\n"
-                    "we are proceeding by disabling all CPU architecture-specific\n"
-                    "(e.g. SSE2/SSE4/AVX) routines. If performance is important, please\n"
-                    "use GROMACS 4.5.7 or try cutoff-scheme = Verlet.\n\n");
+                    "4.6 and 5.x. If performance is important, please use GROMACS 4.5.7\n"
+                    "or try cutoff-scheme = Verlet.\n\n");
         }
     }