Fix bug in fixed_problem_computation (#403)

The issue was the problem check that some constraint values were going -inf and nan. This was caught by exceptions but it didn't crash the process as it is caught. We need more graceful handling of exceptions: #264

Authors:
  - Akif ÇÖRDÜK (https://github.com/akifcorduk)

Approvers:
  - Alice Boucher (https://github.com/aliceb-nv)
  - Ramakrishnap (https://github.com/rgsl888prabhu)

URL: #403

Author: akifcorduk

cpp/CMakeLists.txt

@@ -85,9 +85,8 @@ if(CMAKE_COMPILER_IS_GNUCXX)
 endif(CMAKE_COMPILER_IS_GNUCXX)
 
 # To use sanitizer with cuda runtime, one must follow a few steps:
-# 1. Find libasan.so with "gcc -print-file-name=libasan.so"
-# 2. Run the binary with env var set: LD_PRELOAD=$PATH_TO_LIBASAN ASAN_OPTIONS='protect_shadow_gap=0:replace_intrin=0'
-# 3. (Optional) To run with a debugger (gdb or cuda-gdb) use the additional ASAN option alloc_dealloc_mismatch=0
+# 1. Run the binary with env var set: LD_PRELOAD="$(gcc -print-file-name=libasan.so)" ASAN_OPTIONS='protect_shadow_gap=0:replace_intrin=0'
+# 2. (Optional) To run with a debugger (gdb or cuda-gdb) use the additional ASAN option alloc_dealloc_mismatch=0
 if(BUILD_SANITIZER)
   list(APPEND CUOPT_CXX_FLAGS -fsanitize=address,undefined -fno-omit-frame-pointer -g -Wno-error=maybe-uninitialized)
   add_link_options(-fsanitize=address,undefined)

cpp/src/mip/problem/problem.cu

@@ -1278,6 +1278,7 @@ void problem_t<i_t, f_t>::compute_integer_fixed_problem()
     return;
   }
   rmm::device_uvector<f_t> assignment(n_variables, handle_ptr->get_stream());
+  thrust::fill(handle_ptr->get_thrust_policy(), assignment.begin(), assignment.end(), 0.);
   integer_fixed_problem = std::make_shared<problem_t<i_t, f_t>>(get_problem_after_fixing_vars(
     assignment, integer_indices, integer_fixed_variable_map, handle_ptr));
   integer_fixed_problem->check_problem_representation(true);