Import latest changes (#137)

* fix shared Dirichlet-Neumann edges for scalars * add support for oogs min/max * fix kernel launch issues for some kernels if threadBlockCount is zero * fix oogs buffer size * fix swapped bdf/ext in scalar rhs * do not use extrapolated velocity in convection term for scalars
Nek5000 · Aug 13, 2020 · d9c871f · d9c871f
1 parent 9a1a70b
commit d9c871f
Show file tree

Hide file tree

Showing 25 changed files with 687 additions and 696 deletions.
diff --git a/3rd_party/gslib/ogs/ogs.hpp b/3rd_party/gslib/ogs/ogs.hpp
@@ -247,8 +247,9 @@ typedef struct {
   occa::memory o_scatterOffsets, o_gatherOffsets;
   occa::memory o_scatterIds, o_gatherIds;
 
-  occa::kernel packBufDoubleKernel, unpackBufDoubleKernel;
-  occa::kernel packBufFloatKernel, unpackBufFloatKernel;
+  occa::kernel packBufDoubleKernel, packBufFloatKernel;
+  occa::kernel unpackBufDoubleAddKernel, unpackBufDoubleMinKernel, unpackBufDoubleMaxKernel;
+  occa::kernel unpackBufFloatAddKernel;
 
   oogs_mode mode;
 

diff --git a/3rd_party/gslib/ogs/okl/oogs.okl b/3rd_party/gslib/ogs/okl/oogs.okl
@@ -1,9 +1,9 @@
 @kernel void packBuf_float(const dlong Nscatter,
-                             const   int Nentries,
-                   @restrict const  dlong *  scatterStarts,
-                   @restrict const  dlong *  scatterIds,
-                   @restrict const  float *  q,
-                   @restrict float *  scatterq){
+                           const   int Nentries,
+                           @restrict const  dlong *  scatterStarts,
+                           @restrict const  dlong *  scatterIds,
+                           @restrict const  float *  q,
+                           @restrict float *  scatterq){
 
   for(dlong s=0;s<Nscatter*Nentries;++s;@tile(256,@outer,@inner)){
 
@@ -22,11 +22,11 @@
 }
 
 @kernel void packBuf_double(const dlong Nscatter,
-                             const   int Nentries,
-                   @restrict const  dlong *  scatterStarts,
-                   @restrict const  dlong *  scatterIds,
-                   @restrict const  double *  q,
-                   @restrict double *  scatterq){
+                            const   int Nentries,
+                            @restrict const  dlong *  scatterStarts,
+                            @restrict const  dlong *  scatterIds,
+                            @restrict const  double *  q,
+                            @restrict double *  scatterq){
 
   for(dlong s=0;s<Nscatter*Nentries;++s;@tile(256,@outer,@inner)){
 
@@ -44,12 +44,12 @@
   }
 }
 
-@kernel void unpackBuf_float(const dlong Ngather,
-                             const  int      Nentries,
-		   @restrict const  dlong *  gatherStarts,
-		   @restrict const  dlong *  gatherIds,
-		   @restrict const  float *  q,
-		   @restrict float *  gatherq){
+@kernel void unpackBuf_floatAdd(const dlong Ngather,
+                                const  int      Nentries,
+		                @restrict const  dlong *  gatherStarts,
+		                @restrict const  dlong *  gatherIds,
+		                @restrict const  float *  q,
+		                @restrict float *  gatherq){
 
   for(dlong g=0;g<Ngather*Nentries;++g;@tile(256,@outer,@inner)){
 
@@ -69,13 +69,13 @@
   }
 }
 
-@kernel void unpackBuf_double(const dlong Ngather,
-                 const  int      Nentries,
-       @restrict const  dlong *  gatherStarts,
-       @restrict const  dlong *  gatherIds,
-       @restrict const  double *  q,
-       @restrict double *  gatherq){
-
+@kernel void unpackBuf_doubleAdd(const dlong Ngather,
+                                 const  int      Nentries,
+                                 @restrict const  dlong *  gatherStarts,
+                                 @restrict const  dlong *  gatherIds,
+                                 @restrict const  double *  q,
+                                 @restrict double *  gatherq){
+                          
   for(dlong g=0;g<Ngather*Nentries;++g;@tile(256,@outer,@inner)){
 
     const dlong gid = g%Ngather;
@@ -94,4 +94,54 @@
   }
 }
 
+@kernel void unpackBuf_doubleMin(const dlong Ngather,
+                                 const  int      Nentries,
+                                 @restrict const  dlong *  gatherStarts,
+                                 @restrict const  dlong *  gatherIds,
+                                 @restrict const  double *  q,
+                                 @restrict double *  gatherq){
+
+  for(dlong g=0;g<Ngather*Nentries;++g;@tile(256,@outer,@inner)){
+
+    const dlong gid = g%Ngather;
+    const int k = g/Ngather;
+    const dlong start = gatherStarts[gid];
+    const dlong end = gatherStarts[gid+1];
+
+    const dlong startId = gatherIds[start];
+    double gq = q[startId*Nentries+k];
+    for(dlong n=start;n<end;++n){
+      const dlong id = gatherIds[n];
+      gq = (q[id*Nentries+k] < gq) ? q[id*Nentries+k] : gq;
+    }
+
+    //contiguously packed
+    gatherq[g] = gq;
+  }
+}
 
+@kernel void unpackBuf_doubleMax(const dlong Ngather,
+                                 const  int      Nentries,
+                                 @restrict const  dlong *  gatherStarts,
+                                 @restrict const  dlong *  gatherIds,
+                                 @restrict const  double *  q,
+                                 @restrict double *  gatherq){
+
+  for(dlong g=0;g<Ngather*Nentries;++g;@tile(256,@outer,@inner)){
+
+    const dlong gid = g%Ngather;
+    const int k = g/Ngather;
+    const dlong start = gatherStarts[gid];
+    const dlong end = gatherStarts[gid+1];
+
+    const dlong startId = gatherIds[start];
+    double gq = q[startId*Nentries+k];
+    for(dlong n=start;n<end;++n){
+      const dlong id = gatherIds[n];
+      gq = (q[id*Nentries+k] > gq) ? q[id*Nentries+k] : gq;
+    }
+
+    //contiguously packed
+    gatherq[g] = gq;
+  }
+}