Replies: 12 comments 6 replies
-
|
You would have to change Suppose that the function fg can be expressed as the sum of functions that can be evaluated on different threads. Then you should be able to tape them on separate threads and express the derivatives as a sum of derivatives that are evaluated on different threads. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the reply! So, you mean that there is no switch to turn on the parallel computing for the ipopt solve. What do you mean "Suppose that the function fg can be expressed as the sum of functions"? Isn't fg a vector of constraints? I could not find the derivative calculations in the file you linked (solve_callback.hpp). Is there any other files that need to be modified accordingly? |
Beta Was this translation helpful? Give feedback.
-
|
For example, suppose that the objective function is the log likelihood for independent data samples. Then the object might be huge sum where the only difference between elements of the sum is the data value. You might spit the sum up into different threads. After the threads complete, you would then sum the result for each thread. |
Beta Was this translation helpful? Give feedback.
-
|
I got you. However, I am not trying to solve one problem using multithread but trying to run several cppad ipopt models on different thread at the same time. For example, I used code below to run two problems at the same time where printf("CppAD : Hello World Demo!\n");
#pragma omp parallel for
for (int i = 0; i < 2; i++) {
get_started(i);
}
printf("CppAD : End Demo!\n");And this is the console output: CppAD : Hello World Demo!
cppad-20210000.8 error from a known source:
return_memory: attempt to return memory not in use
v_ptr = 0x7fa5180491c8
thread = 0
capacity = 128
See CPPAD_TRACE_THREAD & CPPAD_TRACE_CAPACITY in
# include <cppad/utility/thread_alloc.hpp>
Error detected by false result for
false
at line 975 in the file
/usr/include/cppad/utility/thread_alloc.hpp
solver: /usr/include/cppad/utility/error_handler.hpp:206: static void CppAD::ErrorHandler::Default(bool, int, const char*, const char*, const char*): Assertion `false' failed.
AbortedWithout the openmp line of code, it can run without error. Could you tell what wrong is here? Thanks a lot!! |
Beta Was this translation helpful? Give feedback.
-
|
Have you looked at the following documentaiton ? For openmp, you may find the following helpful: |
Beta Was this translation helpful? Give feedback.
-
|
I am trying to parallelize IPOPT optimization tasks using OpenMP and CppAD. I have set up the parallel environment according to the CppAD OpenMP example, but when I replace the worker function with an IPOPT model, I get the following error: The problem seems to occur when I use an IPOPT model in the worker function. If I replace the IPOPT model with a simple CppAD model, the parallelization works fine. So it seems that the issue is specifically with IPOPT when used in a parallel context. Here's the function I'm using as my worker function (replacing the worker function from the CppAD OpenMP example): Many thanks! |
Beta Was this translation helpful? Give feedback.
-
|
@Mingze-Li128 It looks like this is a short example. Would you please include the other code so that I can try and reproduce the error. |
Beta Was this translation helpful? Give feedback.
-
|
example.txt |
Beta Was this translation helpful? Give feedback.
-
|
I replaced the worker function in with the code above and got the following error when I tried to compile it: I could try to just fix this, but I want to be sure I have the exact same code as you. What should I do ? |
Beta Was this translation helpful? Give feedback.
-
|
It is the code I am using. Have you included the FG_eval definition in the namespace as: |
Beta Was this translation helpful? Give feedback.
-
|
OK, I put the and the definition of FG_eval in before the problem_specific definition. I also added to the target link libraries in Now the program compiles and links, but I get the following error when I try to run When I run in the debugger I get |
Beta Was this translation helpful? Give feedback.
-
|
It looks to me like we have some work to do to get this example to work. I am busy with other things right now. Perhaps you can strip out the CppAD stuff and see if you can get a multi-threading ipopt example to work. Then we can work on adding the CppAD stuff to it. P.S. |
Beta Was this translation helpful? Give feedback.
-
|
Have there been any updates on this bug? Based on your previous recommendation, I tested a multi-threaded Ipopt example, and it functioned correctly. However, upon examining solve_callback.hpp, I noticed the presence of a Dependent function. According to your documentation at https://coin-or.github.io/CppAD/doc/multi_thread.htm, this function might not be thread-safe. Could this be the cause of the bug I previously mentioned? If so, does this imply that cppad::ipopt cannot be utilized in a multi-threaded environment? Are there any potential workarounds or solutions to this issue? |
Beta Was this translation helpful? Give feedback.
-
|
By adding labels to your code, I found that the bug happens in while creating num_arg_dyn in get_dyn_previous.hpp: Then I found there is a value in num_arg_dyn: Is the CPPAD_ASSERT_FIRST_CALL_NOT_PARALLEL means we cannot use multi thread? The bug is: but we already define it outside of the parallel section as: |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the heads up. I am going to check all of the FIRST_CALL_NOT_PARALLEL cases and make sure they are called by |
Beta Was this translation helpful? Give feedback.
-
|
I fixed ad_parallel so that it initializes all the routines with FIRST_CALL_NOT_PARALLEL and has better documentation; see the changes to parallel_ad in I tested the case above with the new master and got the same error:
|
Beta Was this translation helpful? Give feedback.
-
|
Thanks for the update! I have tried the current master branch. However, the example.cpp now failed to /usr/bin/ld: CMakeFiles/solver.dir/example.cpp.o: in function `void CppAD::put_check_for_nan<double>(CppAD::vector<double> const&, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >&)':
example.cpp:(.text._ZN5CppAD17put_check_for_nanIdEEvRKNS_6vectorIT_EERNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE[_ZN5CppAD17put_check_for_nanIdEEvRKNS_6vectorIT_EERNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEEE]+0x68): undefined reference to `CppAD::local::temp_file[abi:cxx11]()'What's your insight on fixing this error? And here's the Ipopt code with OpenMP: #include "IpIpoptApplication.hpp"
#include "IpSolveStatistics.hpp"
#include <omp.h>
#include <cassert>
using namespace Ipopt;
class HS071_NLP : public TNLP
{
public:
// Default constructor and destructor
HS071_NLP() {}
virtual ~HS071_NLP() {}
// Methods to return some problem specific constants
virtual bool get_nlp_info(Index& n, Index& m, Index& nnz_jac_g,
Index& nnz_h_lag, IndexStyleEnum& index_style);
virtual bool get_bounds_info(Index n, Number* x_l, Number* x_u,
Index m, Number* g_l, Number* g_u);
virtual bool get_starting_point(Index n, bool init_x, Number* x,
bool init_z, Number* z_L, Number* z_U,
Index m, bool init_lambda, Number* lambda);
virtual bool eval_f(Index n, const Number* x, bool new_x, Number& obj_value);
virtual bool eval_grad_f(Index n, const Number* x, bool new_x, Number* grad_f);
virtual bool eval_g(Index n, const Number* x, bool new_x, Index m, Number* g);
virtual bool eval_jac_g(Index n, const Number* x, bool new_x,
Index m, Index nele_jac, Index* iRow, Index* jCol, Number* values);
virtual bool eval_h(Index n, const Number* x, bool new_x,
Number obj_factor, Index m, const Number* lambda,
bool new_lambda, Index nele_hess,
Index* iRow, Index* jCol, Number* values);
virtual void finalize_solution(SolverReturn status,
Index n, const Number* x, const Number* z_L, const Number* z_U,
Index m, const Number* g, const Number* lambda,
Number obj_value, const IpoptData* ip_data,
IpoptCalculatedQuantities* ip_cq);
};
bool HS071_NLP::get_nlp_info(
Index& n,
Index& m,
Index& nnz_jac_g,
Index& nnz_h_lag,
IndexStyleEnum& index_style
)
{
// The problem described in HS071_NLP.hpp has 4 variables, x[0] through x[3]
n = 4;
// one equality constraint and one inequality constraint
m = 2;
// in this example the jacobian is dense and contains 8 nonzeros
nnz_jac_g = 8;
// the Hessian is also dense and has 16 total nonzeros, but we
// only need the lower left corner (since it is symmetric)
nnz_h_lag = 10;
// use the C style indexing (0-based)
index_style = TNLP::C_STYLE;
return true;
}
bool HS071_NLP::get_bounds_info(
Index n,
Number* x_l,
Number* x_u,
Index m,
Number* g_l,
Number* g_u
)
{
// here, the n and m we gave IPOPT in get_nlp_info are passed back to us.
// If desired, we could assert to make sure they are what we think they are.
assert(n == 4);
assert(m == 2);
// the variables have lower bounds of 1
for (Index i = 0; i < 4; i++)
{
x_l[i] = 1.0;
}
// the variables have upper bounds of 5
for (Index i = 0; i < 4; i++)
{
x_u[i] = 5.0;
}
// the first constraint g1 has a lower bound of 25
g_l[0] = 25;
// the first constraint g1 has NO upper bound, here we set it to 2e19.
// Ipopt interprets any number greater than nlp_upper_bound_inf as
// infinity. The default value of nlp_upper_bound_inf and nlp_lower_bound_inf
// is 1e19 and can be changed through ipopt options.
g_u[0] = 2e19;
// the second constraint g2 is an equality constraint, so we set the
// upper and lower bound to the same value
g_l[1] = g_u[1] = 40.0;
return true;
}
bool HS071_NLP::get_starting_point(
Index n,
bool init_x,
Number* x,
bool init_z,
Number* z_L,
Number* z_U,
Index m,
bool init_lambda,
Number* lambda
)
{
// Here, we assume we only have starting values for x, if you code
// your own NLP, you can provide starting values for the dual variables
// if you wish
assert(init_x == true);
assert(init_z == false);
assert(init_lambda == false);
// initialize to the given starting point
x[0] = 1.0;
x[1] = 5.0;
x[2] = 5.0;
x[3] = 1.0;
return true;
}
bool HS071_NLP::eval_f(
Index n,
const Number* x,
bool new_x,
Number& obj_value
)
{
assert(n == 4);
obj_value = x[0] * x[3] * (x[0] + x[1] + x[2]) + x[2];
return true;
}
bool HS071_NLP::eval_grad_f(
Index n,
const Number* x,
bool new_x,
Number* grad_f
)
{
assert(n == 4);
grad_f[0] = x[0] * x[3] + x[3] * (x[0] + x[1] + x[2]);
grad_f[1] = x[0] * x[3];
grad_f[2] = x[0] * x[3] + 1;
grad_f[3] = x[0] * (x[0] + x[1] + x[2]);
return true;
}
bool HS071_NLP::eval_g(
Index n,
const Number* x,
bool new_x,
Index m,
Number* g
)
{
assert(n == 4);
assert(m == 2);
g[0] = x[0] * x[1] * x[2] * x[3];
g[1] = x[0] * x[0] + x[1] * x[1] + x[2] * x[2] + x[3] * x[3];
return true;
}
bool HS071_NLP::eval_jac_g(
Index n,
const Number* x,
bool new_x,
Index m,
Index nele_jac,
Index* iRow,
Index* jCol,
Number* values
)
{
assert(n == 4);
assert(m == 2);
if (values == NULL)
{
// return the structure of the Jacobian
// this particular Jacobian is dense
iRow[0] = 0;
jCol[0] = 0;
iRow[1] = 0;
jCol[1] = 1;
iRow[2] = 0;
jCol[2] = 2;
iRow[3] = 0;
jCol[3] = 3;
iRow[4] = 1;
jCol[4] = 0;
iRow[5] = 1;
jCol[5] = 1;
iRow[6] = 1;
jCol[6] = 2;
iRow[7] = 1;
jCol[7] = 3;
}
else
{
// return the values of the Jacobian of the constraints
values[0] = x[1] * x[2] * x[3]; // 0,0
values[1] = x[0] * x[2] * x[3]; // 0,1
values[2] = x[0] * x[1] * x[3]; // 0,2
values[3] = x[0] * x[1] * x[2]; // 0,3
values[4] = 2 * x[0]; // 1,0
values[5] = 2 * x[1]; // 1,1
values[6] = 2 * x[2]; // 1,2
values[7] = 2 * x[3]; // 1,3
}
return true;
}
bool HS071_NLP::eval_h(
Index n,
const Number* x,
bool new_x,
Number obj_factor,
Index m,
const Number* lambda,
bool new_lambda,
Index nele_hess,
Index* iRow,
Index* jCol,
Number* values
)
{
assert(n == 4);
assert(m == 2);
if (values == NULL)
{
// return the structure. This is a symmetric matrix, fill the lower left
// triangle only.
// the hessian for this problem is actually dense
Index idx = 0;
for (Index row = 0; row < 4; row++)
{
for (Index col = 0; col <= row; col++)
{
iRow[idx] = row;
jCol[idx] = col;
idx++;
}
}
assert(idx == nele_hess);
}
else
{
// return the values. This is a symmetric matrix, fill the lower left
// triangle only
// fill the objective portion
values[0] = obj_factor * (2 * x[3]); // 0,0
values[1] = obj_factor * (x[3]); // 1,0
values[2] = 0.; // 1,1
values[3] = obj_factor * (x[3]); // 2,0
values[4] = 0.; // 2,1
values[5] = 0.; // 2,2
values[6] = obj_factor * (2 * x[0] + x[1] + x[2]); // 3,0
values[7] = obj_factor * (x[0]); // 3,1
values[8] = obj_factor * (x[0]); // 3,2
values[9] = 0.; // 3,3
// add the portion for the first constraint
values[1] += lambda[0] * (x[2] * x[3]); // 1,0
values[3] += lambda[0] * (x[1] * x[3]); // 2,0
values[4] += lambda[0] * (x[0] * x[3]); // 2,1
values[6] += lambda[0] * (x[1] * x[2]); // 3,0
values[7] += lambda[0] * (x[0] * x[2]); // 3,1
values[8] += lambda[0] * (x[0] * x[1]); // 3,2
// add the portion for the second constraint
values[0] += lambda[1] * 2; // 0,0
values[2] += lambda[1] * 2; // 1,1
values[5] += lambda[1] * 2; // 2,2
values[9] += lambda[1] * 2; // 3,3
}
return true;
}
void HS071_NLP::finalize_solution(
SolverReturn status,
Index n,
const Number* x,
const Number* z_L,
const Number* z_U,
Index m,
const Number* g,
const Number* lambda,
Number obj_value,
const IpoptData* ip_data,
IpoptCalculatedQuantities* ip_cq
)
{
// here is where we would store the solution to variables, or write to a file, etc
// so we could use the solution.
// For this example, we write the solution to the console
std::cout << std::endl << std::endl << "Solution of the primal variables, x" << std::endl;
for (Index i = 0; i < n; i++)
{
std::cout << "x[" << i << "] = " << x[i] << std::endl;
}
std::cout << std::endl << std::endl << "Solution of the bound multipliers, z_L and z_U" << std::endl;
for (Index i = 0; i < n; i++)
{
std::cout << "z_L[" << i << "] = " << z_L[i] << std::endl;
}
for (Index i = 0; i < n; i++)
{
std::cout << "z_U[" << i << "] = " << z_U[i] << std::endl;
}
std::cout << std::endl << std::endl << "Objective value" << std::endl;
std::cout << "f(x*) = " << obj_value << std::endl;
std::cout << std::endl << "Final value of the constraints:" << std::endl;
for (Index i = 0; i < m; i++)
{
std::cout << "g(" << i << ") = " << g[i] << std::endl;
}
}
int main(int, char**)
{
ApplicationReturnStatus status[2];
// Setup for OpenMP parallel region
#pragma omp parallel for
for (int i = 0; i < 2; i++)
{
int thread_id = omp_get_thread_num();
// Create a new instance of your nlp
SmartPtr<TNLP> mynlp = new HS071_NLP();
// Create a new instance of IpoptApplication
SmartPtr<IpoptApplication> app = IpoptApplicationFactory();
app->Options()->SetNumericValue("tol", 3.82e-6);
app->Options()->SetStringValue("mu_strategy", "adaptive");
app->Options()->SetStringValue("output_file", "ipopt.out");
// The following overwrites the default name (ipopt.opt) of the options file
// app->Options()->SetStringValue("option_file_name", "hs071.opt");
// Initialize the IpoptApplication and process the options
status[thread_id] = app->Initialize();
if (status[thread_id] != Solve_Succeeded)
{
std::cout << "Thread " << thread_id << ": *** Error during initialization!" << std::endl;
}
else
{
// Ask Ipopt to solve the problem
status[thread_id] = app->OptimizeTNLP(mynlp);
}
}
// Check the results and print the outcome
for (int i = 0; i < 2; i++)
{
if (status[i] == Solve_Succeeded)
{
std::cout << "Model " << i << ": *** The problem solved!" << std::endl;
}
else
{
std::cout << "Model " << i << ": *** The problem FAILED!" << std::endl;
}
}
return 0; // This is a simple example, so we just return 0.
// In a more sophisticated setting, you might want to handle the statuses more appropriately.
} |
Beta Was this translation helpful? Give feedback.
-
|
I had to change two of the includes above as follows: I then compiled the program with debugging. Here is the script I used: This script used the version of ipopt retrieved by I got some Ipopt trace output mixed in with the following error messages: The error messages were not consistent, in fact one time the script ran without detecting an error. It looks to me like there is something special that needs to be done to use MUMPS with openmp. I think you are using a different solver, so perhaps it does not have this problem. Perhaps you can reproduce the error message I got above ? |
Beta Was this translation helpful? Give feedback.
-
I noticed a page in the CPPAD guide introducing how to use cppad in openmp environment (https://cppad.readthedocs.io/en/latest/simple_ad_openmp.cpp.html). I wonder if this multi threading technique can also be applied in the ipopt solve example, to accelerate the solving speed? How can I do that?
Thanks!
Beta Was this translation helpful? Give feedback.
All reactions