clean up handling of GPU errors (#116)

LLNL · Feb 3, 2021 · 85a74ed · 85a74ed
1 parent 0196941
commit 85a74ed
Show file tree

Hide file tree

Showing 5 changed files with 12 additions and 17 deletions.
diff --git a/include/hydrogen/device/gpu/cuda/CUDAError.hpp b/include/hydrogen/device/gpu/cuda/CUDAError.hpp
@@ -21,9 +21,8 @@
         auto h_check_cuda_error_code__ = cmd;                           \
         H_ASSERT(h_check_cuda_error_code__ == cudaSuccess,              \
                  ::hydrogen::CUDAError,                                 \
-                 (cudaDeviceReset(),                                    \
                   ::hydrogen::cuda::BuildCUDAErrorMessage(              \
-                      #cmd, h_check_cuda_error_code__)));               \
+                      #cmd, h_check_cuda_error_code__));                \
         H_SYNC_CUDA();                                                  \
     } while (false)
 

diff --git a/include/hydrogen/device/gpu/cuda/cuBLASError.hpp b/include/hydrogen/device/gpu/cuda/cuBLASError.hpp
@@ -16,10 +16,9 @@
         auto h_check_cublas_err_code__ = cmd;                           \
         H_ASSERT(h_check_cublas_err_code__ == CUBLAS_STATUS_SUCCESS,    \
                  cuBLASError,                                           \
-                 (cudaDeviceReset(),                                    \
-                  cublas::BuildcuBLASErrorMessage(                      \
-                      #cmd,                                             \
-                      h_check_cublas_err_code__)));                     \
+                 cublas::BuildcuBLASErrorMessage(                       \
+                     #cmd,                                              \
+                     h_check_cublas_err_code__));                       \
         H_SYNC_CUDA();                                                  \
     } while (false)
 

diff --git a/include/hydrogen/device/gpu/cuda/cuSOLVERError.hpp b/include/hydrogen/device/gpu/cuda/cuSOLVERError.hpp
@@ -16,10 +16,9 @@
         auto h_check_cusolver_err_code__ = cmd;                           \
         H_ASSERT(h_check_cusolver_err_code__ == CUSOLVER_STATUS_SUCCESS,  \
                  cuSOLVERError,                                         \
-                 (cudaDeviceReset(),                                    \
-                  cusolver::BuildcuSOLVERErrorMessage(                    \
-                      #cmd,                                             \
-                      h_check_cusolver_err_code__)));                     \
+                 cusolver::BuildcuSOLVERErrorMessage(                   \
+                     #cmd,                                              \
+                     h_check_cusolver_err_code__));                     \
         H_SYNC_CUDA();                                                  \
     } while (false)
 

diff --git a/include/hydrogen/device/gpu/rocm/ROCmError.hpp b/include/hydrogen/device/gpu/rocm/ROCmError.hpp
@@ -21,9 +21,8 @@
         auto h_check_hip_error_code__ = cmd;                            \
         H_ASSERT(h_check_hip_error_code__ == hipSuccess,                \
                  ::hydrogen::HIPError,                                  \
-                 (hipDeviceReset(),                                     \
-                  ::hydrogen::rocm::BuildHipErrorMessage(               \
-                      #cmd, h_check_hip_error_code__)));                \
+                 ::hydrogen::rocm::BuildHipErrorMessage(                \
+                     #cmd, h_check_hip_error_code__));                  \
         H_SYNC_HIP();                                                   \
     } while (false)
 

diff --git a/include/hydrogen/device/gpu/rocm/rocBLASError.hpp b/include/hydrogen/device/gpu/rocm/rocBLASError.hpp
@@ -16,10 +16,9 @@
         auto h_check_rocblas_err_code__ = cmd;                          \
         H_ASSERT(h_check_rocblas_err_code__ == rocblas_status_success,  \
                  rocBLASError,                                          \
-                 (hipDeviceReset(),                                     \
-                  rocblas::BuildrocBLASErrorMessage(                    \
-                      #cmd,                                             \
-                      h_check_rocblas_err_code__)));                    \
+                 rocblas::BuildrocBLASErrorMessage(                     \
+                     #cmd,                                              \
+                     h_check_rocblas_err_code__));                      \
         H_SYNC_HIP();                                                   \
     } while (false)