print computeMode

cwpearson · Feb 10, 2020 · 036267a · 036267a
1 parent 2508ba8
commit 036267a
Show file tree

Hide file tree

Showing 2 changed files with 20 additions and 5 deletions.
diff --git a/include/stencil/local_domain.cuh b/include/stencil/local_domain.cuh
@@ -1,7 +1,6 @@
 #pragma once
 
 #include <iostream>
-#include <mpi.h>
 
 #include "stencil/cuda_runtime.hpp"
 #include "stencil/dim3.hpp"
@@ -45,13 +44,13 @@ public:
 
   ~LocalDomain() {
 
-    int rank;
-    MPI_Comm_rank(MPI_COMM_WORLD, &rank);
+    // int rank;
+    // MPI_Comm_rank(MPI_COMM_WORLD, &rank);
 
-    std::cerr << "rank=" << rank << " ~LocalDomain(): device=" << dev_ << "\n";
+    // std::cerr << "rank=" << rank << " ~LocalDomain(): device=" << dev_ << "\n";
     CUDA_RUNTIME(cudaSetDevice(dev_));
     for (auto p : currDataPtrs_) {
-      std::cerr << "rank=" << rank << " ~LocalDomain(): cudaFree " << uintptr_t(p) << "\n";
+      // std::cerr << "rank=" << rank << " ~LocalDomain(): cudaFree " << uintptr_t(p) << "\n";
       CUDA_RUNTIME(cudaFree(p));
     }
     CUDA_RUNTIME(cudaFree(devCurrDataPtrs_));

diff --git a/include/stencil/stencil.hpp b/include/stencil/stencil.hpp
@@ -127,6 +127,22 @@ class DistributedDomain {
     CUDA_RUNTIME(cudaGetDeviceCount(&deviceCount));
     std::cerr << "[" << rank_ << "] cudaGetDeviceCount= " << deviceCount << "\n";
 
+  /*
+cudaComputeModeDefault = 0
+    Default compute mode (Multiple threads can use cudaSetDevice() with this device) 
+cudaComputeModeExclusive = 1
+    Compute-exclusive-thread mode (Only one thread in one process will be able to use cudaSetDevice() with this device) 
+cudaComputeModeProhibited = 2
+    Compute-prohibited mode (No threads can use cudaSetDevice() with this device) 
+cudaComputeModeExclusiveProcess = 3
+    Compute-exclusive-process mode (Many threads in one process will be able to use cudaSetDevice() with this device) 
+  */
+    cudaDeviceProp prop;
+    for (int i = 0; i < deviceCount; ++i) {
+      CUDA_RUNTIME(cudaGetDeviceProperties(&prop, i));
+      std::cerr << "[" << rank_ << "] cudaDeviceProp.computeMode=" << prop.computeMode << "\n";
+    }
+
     // Determine GPUs this DistributedDomain is reposible for
     if (gpus_.empty()) {
       // if fewer colocated ranks than GPUs, round-robin GPUs to ranks