fixing returns 700 and then takes an exit path that causes to return 3. @obilaniu

Author: dendisuhubdy

src/gpuarray_blas_cuda_cublas.c

@@ -835,6 +835,9 @@ static int sgemmBatch(cb_order order, cb_transpose transA, cb_transpose transB,
       return ctx->err->code;
     }
 
+	// added cuda_wait
+	GA_CUDA_EXIT_ON_ERROR(ctx, cuda_wait(Ta, CUDA_WAIT_ALL));
+
     err = cublasSgemmBatched(h->h,
                              convT(transA), convT(transB),
                              M, N, K, &alpha,
@@ -962,6 +965,9 @@ static int dgemmBatch(cb_order order, cb_transpose transA, cb_transpose transB,
       return ctx->err->code;
     }
 
+	// added cuda_wait
+	GA_CUDA_EXIT_ON_ERROR(ctx, cuda_wait(Ta, CUDA_WAIT_ALL));
+
     err = cublasDgemmBatched(h->h,
                              convT(transA), convT(transB),
                              M, N, K, &alpha,