fixes to allow more consistent build tests

caffe2/core/blob_gpu_test.cc

@@ -15,6 +15,7 @@ TYPED_TEST_CASE(TensorGPUTest, TensorTypes);
 TYPED_TEST_CASE(TensorGPUDeathTest, TensorTypes);
 
 TYPED_TEST(TensorGPUTest, TensorInitializedEmpty) {
+  if (!caffe2::HasCudaGPU()) return;
   Tensor<TypeParam, CUDAContext> tensor;
   EXPECT_EQ(tensor.ndim(), 0);
   vector<int> dims(3);
@@ -31,6 +32,7 @@ TYPED_TEST(TensorGPUTest, TensorInitializedEmpty) {
 }
 
 TYPED_TEST(TensorGPUTest, TensorInitializedNonEmpty) {
+  if (!HasCudaGPU()) return;
   vector<int> dims(3);
   dims[0] = 2;
   dims[1] = 3;
@@ -57,6 +59,7 @@ TYPED_TEST(TensorGPUTest, TensorInitializedNonEmpty) {
 }
 
 TYPED_TEST(TensorGPUTest, TensorShareData) {
+  if (!HasCudaGPU()) return;
   vector<int> dims(3);
   dims[0] = 2;
   dims[1] = 3;
@@ -71,6 +74,7 @@ TYPED_TEST(TensorGPUTest, TensorShareData) {
 }
 
 TYPED_TEST(TensorGPUTest, TensorShareDataCanUseDifferentShapes) {
+  if (!HasCudaGPU()) return;
   vector<int> dims(3);
   dims[0] = 2;
   dims[1] = 3;
@@ -89,6 +93,7 @@ TYPED_TEST(TensorGPUTest, TensorShareDataCanUseDifferentShapes) {
 }
 
 TYPED_TEST(TensorGPUTest, NoLongerSharesAfterReshape) {
+  if (!HasCudaGPU()) return;
   vector<int> dims(3);
   dims[0] = 2;
   dims[1] = 3;
@@ -108,6 +113,7 @@ TYPED_TEST(TensorGPUTest, NoLongerSharesAfterReshape) {
 
 
 TYPED_TEST(TensorGPUDeathTest, CannotAccessDataWhenEmpty) {
+  if (!HasCudaGPU()) return;
   ::testing::FLAGS_gtest_death_test_style = "threadsafe";
   Tensor<TypeParam, CUDAContext> tensor;
   EXPECT_EQ(tensor.ndim(), 0);

caffe2/core/common_gpu.cc

@@ -5,6 +5,21 @@
 
 namespace caffe2 {
 
+bool HasCudaGPU() {
+  int count;
+  auto err = cudaGetDeviceCount(&count);
+  if (err == cudaErrorNoDevice || err == cudaErrorInsufficientDriver) {
+    return false;
+  }
+  // cudaGetDeviceCount() should only return the above two errors. If
+  // there are other kinds of errors, maybe you have called some other
+  // cuda functions before HasCudaGPU().
+  CHECK(err == cudaSuccess)
+      << "Unexpected error from cudaGetDeviceCount(). Did you run some "
+         "cuda functions before calling HasCudaGPU()?";
+  return true;
+}
+
 namespace {
 int gDefaultGPUID = 0;
 }
@@ -137,6 +152,11 @@ const char* curandGetErrorString(curandStatus_t error) {
 }
 
 bool Caffe2EnableCudaPeerAccess() {
+  // If the current run does not have any cuda devices, do nothing.
+  if (!HasCudaGPU()) {
+    LOG(INFO) << "No cuda gpu present. Skipping.";
+    return true;
+  }
   int device_count;
   CUDA_CHECK(cudaGetDeviceCount(&device_count));
   int init_device;

caffe2/core/common_gpu.h

@@ -14,6 +14,10 @@
 
 namespace caffe2 {
 
+// Check if the current running session has a cuda gpu present. Note that this
+// is different from having caffe2 built with cuda - it is possible that
+// caffe2 is built with cuda but there is no cuda hardware available.
+bool HasCudaGPU();
 // Sets and gets the default GPU id. If the function is not called, we will use
 // GPU 0 ast he default gpu id. If there is an operator that says it runs on the
 // GPU but did not specify which GPU, this default gpuid is going to be used.

caffe2/core/cuda_memorypool_test.cc

@@ -16,6 +16,7 @@ class MemoryPoolTest : public ::testing::Test {
   // should define it if you need to initialize the varaibles.
   // Otherwise, this can be skipped.
   void SetUp() override {
+    if (!HasCudaGPU()) return;
     int device_count_;
     CUDA_CHECK(cudaGetDeviceCount(&device_count_));
     // If we test with the memory pool, initialize the memory pool.
@@ -29,6 +30,7 @@ class MemoryPoolTest : public ::testing::Test {
   }
 
   void TearDown() override {
+    if (!HasCudaGPU()) return;
     if (UsePoolOrNot::value) {
       CHECK(CudaMemoryPool::FinalizeMemoryPool());
     }
@@ -47,6 +49,7 @@ TYPED_TEST(MemoryPoolTest, InitializeAndFinalizeWorks) {
 }
 
 TYPED_TEST(MemoryPoolTest, AllocateAndDeallocate) {
+  if (!HasCudaGPU()) return;
   const int nbytes = 1048576;
   for (int i = 0; i < this->device_count_; ++i) {
     LOG(INFO) << "Device " << i << " of " << this->device_count_;

caffe2/operators/BREW

@@ -94,7 +94,6 @@ cc_test(
   deps = [
       ":core_ops",
       ":core_ops_gpu",
-      ":core_ops_cudnn",
       "//data/mnist:mnist_minidb",
       "//gtest:caffe2_gtest_main",
   ]

pycaffe2/caffe2_python.cc

@@ -450,8 +450,11 @@ PyObject* FeedBlob(PyObject* self, PyObject* args) {
 // Here are functions that are purely GPU-based functions to be filled.
 
 PyObject* NumberOfGPUs(PyObject* self, PyObject* args) {
-  int num_devices;
-  if (cudaGetDeviceCount(&num_devices) != cudaSuccess) {
+  int num_devices = 0;
+  auto err = cudaGetDeviceCount(&num_devices);
+  if (err == cudaErrorNoDevice || err == cudaErrorInsufficientDriver) {
+    return Py_BuildValue("i", 0);
+  } else if (err != cudaSuccess) {
     PyErr_SetString(PyExc_RuntimeError, "Runtime CUDA error.");
     return NULL;
   }

pycaffe2/muji_test.py

@@ -64,5 +64,7 @@ class TestMuji(unittest.TestCase):
 if __name__ == '__main__':
   if not workspace.has_gpu_support:
     print 'No GPU support. skipping muji test.'
+  elif workspace.NumberOfGPUs() == 0:
+    print 'No GPU device. Skipping gpu test.'
   else:
     unittest.main()

pycaffe2/workspace_gpu_test.py

@@ -36,6 +36,8 @@ class TestWorkspaceGPU(unittest.TestCase):
 
 if __name__ == '__main__':
   if not workspace.has_gpu_support:
-    print 'No GPU support. skipping gpu test.'
+    print 'No GPU support. Skipping gpu test.'
+  elif workspace.NumberOfGPUs() == 0:
+    print 'No GPU device. Skipping gpu test.'
   else:
     unittest.main()