Merge branch 'development' into gpu-merge-test

kaiserpeng · Oct 15, 2018 · b026bc7 · b026bc7
2 parents 5ab12ea + d0ab3db
commit b026bc7
Show file tree

Hide file tree

Showing 2 changed files with 8 additions and 4 deletions.
diff --git a/Src/Base/AMReX_BaseFab.H b/Src/Base/AMReX_BaseFab.H
@@ -2168,7 +2168,7 @@ template <class T>
 IntVect
 BaseFab<T>::minIndex (const Box& subbox, int comp) const
 {
-    int imin, jmin, kmin;
+    AMREX_D_DECL(int imin, jmin, kmin);
     T min_val = std::numeric_limits<T>::max();
 
     const auto dp0 = this->stridedPtr(subbox,comp);
@@ -2180,7 +2180,7 @@ BaseFab<T>::minIndex (const Box& subbox, int comp) const
             for (int i = 0; i < len[0]; ++i) {
                 if (dp[i] < min_val) {
                     min_val = dp[i];
-                    imin = i;  jmin = j;  kmin = k;
+                    AMREX_D_TERM(imin = i;,  jmin = j;,  kmin = k;)
                 }
             }
         }
@@ -2202,7 +2202,7 @@ IntVect
 BaseFab<T>::maxIndex (const Box& subbox,
                       int        comp) const
 {
-    int imax, jmax, kmax;
+    AMREX_D_DECL(int imax, jmax, kmax);
     T max_val = std::numeric_limits<T>::lowest();
 
     const auto dp0 = this->stridedPtr(subbox,comp);
@@ -2214,7 +2214,7 @@ BaseFab<T>::maxIndex (const Box& subbox,
             for (int i = 0; i < len[0]; ++i) {
                 if (dp[i] > max_val) {
                     max_val = dp[i];
-                    imax = i;  jmax = j;  kmax = k;
+                    AMREX_D_TERM(imax = i;, jmax = j;, kmax = k;)
                 }
             }
         }

diff --git a/Tools/F_scripts/write_cuda_headers.py b/Tools/F_scripts/write_cuda_headers.py
@@ -574,6 +574,10 @@ def convert_cxx(outdir, cxx_files, cpp, defines):
                            "#endif\n".format(func_name))
                 hout.write("    cuda_{}<<<{}numBlocks, {}numThreads, 0, amrex::Device::cudaStream()>>>\n    ({});\n".format(func_name, func_name, func_name, args))
 
+                # Catch errors in the launch configuration.
+
+                hout.write("    CudaAPICheck(cudaGetLastError());\n")
+
                 if 'AMREX_DEBUG' in defines:
                     hout.write("CudaAPICheck(cudaDeviceSynchronize());\n")