disable lite gpu (#43177)

cmake/external/lite.cmake

@@ -115,7 +115,7 @@ if(NOT LITE_SOURCE_DIR OR NOT LITE_BINARY_DIR)
     set(LITE_BUILD_COMMAND $(MAKE) publish_inference -j)
     set(LITE_OPTIONAL_ARGS
         -DWITH_MKL=ON
-        -DLITE_WITH_CUDA=${WITH_GPU}
+        -DLITE_WITH_CUDA=OFF
         -DWITH_MKLDNN=OFF
         -DLITE_WITH_X86=ON
         -DLITE_WITH_PROFILE=OFF
@@ -124,9 +124,6 @@ if(NOT LITE_SOURCE_DIR OR NOT LITE_BINARY_DIR)
         -DWITH_PYTHON=OFF
         -DWITH_TESTING=OFF
         -DLITE_BUILD_EXTRA=ON
-        -DCUDNN_ROOT=${CUDNN_ROOT}
-        -DLITE_WITH_STATIC_CUDA=OFF
-        -DCUDA_ARCH_NAME=${CUDA_ARCH_NAME}
         -DLITE_WITH_XPU=${LITE_WITH_XPU}
         -DXPU_SDK_URL=${XPU_BASE_URL}
         -DXPU_SDK_ENV=${XPU_SDK_ENV}

paddle/fluid/inference/lite/engine.cc

@@ -12,10 +12,6 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-#define LITE_WITH_CUDA 1
-#endif
-
 #ifdef LITE_SUBGRAPH_WITH_XPU
 #define LITE_WITH_XPU 1
 #endif

paddle/fluid/inference/lite/test_tensor_utils.cc

@@ -152,22 +152,12 @@ TEST(LiteEngineOp, TensorCopyAsync) {
   auto* ctx_cpu =
       platform::DeviceContextPool::Instance().Get(platform::CPUPlace());
   test_tensor_copy(*ctx_cpu);
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-  auto* ctx_gpu =
-      platform::DeviceContextPool::Instance().Get(platform::CUDAPlace(0));
-  test_tensor_copy(*ctx_gpu);
-#endif
 }
 
 TEST(LiteEngineOp, TensorShare) {
   auto* ctx_cpu =
       platform::DeviceContextPool::Instance().Get(platform::CPUPlace());
   test_tensor_share(*ctx_cpu);
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-  auto* ctx_gpu =
-      platform::DeviceContextPool::Instance().Get(platform::CUDAPlace(0));
-  test_tensor_share(*ctx_gpu);
-#endif
 }
 
 }  // namespace utils

paddle/fluid/inference/tests/api/lite_mul_model_test.cc

@@ -120,35 +120,13 @@ TEST(AnalysisPredictor, lite_xpu) {
 }
 #endif
 
-#if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
-TEST(AnalysisPredictor, thread_local_stream) {
-  const size_t thread_num = 5;
-  std::vector<std::thread> threads(thread_num);
-  Barrier barrier(thread_num);
-  for (size_t i = 0; i < threads.size(); ++i) {
-    threads[i] = std::thread([&barrier, i]() {
-      AnalysisConfig config;
-      config.EnableUseGpu(100, 0);
-      config.SetModel(FLAGS_infer_model + "/" + "mul_model");
-      config.EnableGpuMultiStream();
-      test_predictor(config, &barrier);
-      test_predictor_zero_copy(config);
-    });
-  }
-  for (auto& th : threads) {
-    th.join();
-  }
-}
-
 TEST(AnalysisPredictor, lite_engine) {
   AnalysisConfig config;
-  config.EnableUseGpu(100, 0);
   config.SetModel(FLAGS_infer_model + "/" + "mul_model");
   config.EnableLiteEngine(paddle::AnalysisConfig::Precision::kFloat32);
   test_predictor(config);
   test_predictor_zero_copy(config);
 }
-#endif
 
 }  // namespace inference
 }  // namespace paddle

paddle/fluid/inference/tests/api/lite_resnet50_test.cc

@@ -23,10 +23,9 @@ limitations under the License. */
 namespace paddle {
 namespace inference {
 
-TEST(AnalysisPredictor, use_gpu) {
+TEST(AnalysisPredictor, use_cpu) {
   std::string model_dir = FLAGS_infer_model + "/" + "model";
   AnalysisConfig config;
-  config.EnableUseGpu(100, 0);
   config.SetModel(model_dir + "/model", model_dir + "/params");
   config.EnableLiteEngine(paddle::AnalysisConfig::Precision::kFloat32, true);
 
@@ -74,10 +73,9 @@ TEST(AnalysisPredictor, use_gpu) {
 
 namespace paddle_infer {
 
-TEST(Predictor, use_gpu) {
+TEST(Predictor, use_cpu) {
   std::string model_dir = FLAGS_infer_model + "/" + "model";
   Config config;
-  config.EnableUseGpu(100, 0);
   config.SetModel(model_dir + "/model", model_dir + "/params");
   config.EnableLiteEngine(PrecisionType::kFloat32);