disable GIL

benchmark/timings-parallel.py

@@ -2,6 +2,10 @@
 import numpy as np
 import scipy.sparse as spa
 from time import perf_counter_ns
+from concurrent.futures import ThreadPoolExecutor
+
+
+num_threads = proxsuite.proxqp.omp_get_max_threads()
 
 
 def generate_mixed_qp(n, n_eq, n_in, seed=1):
@@ -23,7 +27,7 @@ def generate_mixed_qp(n, n_eq, n_in, seed=1):
     u = A @ v
     l = -1.0e20 * np.ones(m)
 
-    return P.toarray(), q, A[:n_eq, :], u[:n_eq], A[n_in:, :], u[n_in:], l[n_in:]
+    return P.toarray(), q, A[:n_eq, :], u[:n_eq], A[n_in:, :], l[n_in:], u[n_in:]
 
 
 n = 500
@@ -32,36 +36,57 @@ def generate_mixed_qp(n, n_eq, n_in, seed=1):
 
 num_qps = 128
 
+print(f"Problem specs: {n=} {n_eq=} {n_in=}. Generating {num_qps} such problems.")
+problems = [generate_mixed_qp(n, n_eq, n_in, seed=j) for j in range(num_qps)]
+print(f"Generated problems. Solving {num_qps} problems with proxsuite.proxqp.omp_get_max_threads()={num_threads} threads.")
+
 # qps = []
 timings = {}
 qps = proxsuite.proxqp.dense.VectorQP()
 
 tic = perf_counter_ns()
-for j in range(num_qps):
+print("Setting up problem vector")
+for H, g, A, b, C, l, u in problems:
     qp = proxsuite.proxqp.dense.QP(n, n_eq, n_in)
-    H, g, A, b, C, u, l = generate_mixed_qp(n, n_eq, n_in, seed=j)
     qp.init(H, g, A, b, C, l, u)
     qp.settings.eps_abs = 1e-9
     qp.settings.verbose = False
     qp.settings.initial_guess = proxsuite.proxqp.InitialGuess.NO_INITIAL_GUESS
     qps.append(qp)
-timings["problem_data"] = (perf_counter_ns() - tic) * 1e-6
+timings["setup_vector_of_qps"] = (perf_counter_ns() - tic) * 1e-6
 
+print("Solving problem vector in parallel with default thread config")
+tic = perf_counter_ns()
+proxsuite.proxqp.dense.solve_in_parallel(qps=qps)
+timings[f"solve_parallel_heuristics_threads"] = (perf_counter_ns() - tic) * 1e-6
+
+print("Solving problem vector serially")
 tic = perf_counter_ns()
 for qp in qps:
     qp.solve()
 timings["solve_serial"] = (perf_counter_ns() - tic) * 1e-6
 
-num_threads = proxsuite.proxqp.omp_get_max_threads()
+print("Solving problem vector in parallel with various thread configs")
 for j in range(1, num_threads):
     tic = perf_counter_ns()
-    proxsuite.proxqp.dense.solve_in_parallel(j, qps)
+    proxsuite.proxqp.dense.solve_in_parallel(qps=qps, num_threads=j)
     timings[f"solve_parallel_{j}_threads"] = (perf_counter_ns() - tic) * 1e-6
 
+print("Solving each problem serially with dense backend.")
+tic = perf_counter_ns()
+for H, g, A, b, C, l, u in problems:
+    proxsuite.proxqp.dense.solve(H, g, A, b, C, l, u, initial_guess=proxsuite.proxqp.InitialGuess.NO_INITIAL_GUESS, eps_abs=1e-9)
+timings["solve_serial_dense"] = (perf_counter_ns() - tic) * 1e-6
+
+print("Solving each problem in parallel (with a ThreadPoolExecutor) with dense backend.")
+def solve_problem(problem):  # just a little helper function to keep things clean
+    H, g, A, b, C, l, u = problem
+    return proxsuite.proxqp.dense.solve(H, g, A, b, C, l, u, initial_guess=proxsuite.proxqp.InitialGuess.NO_INITIAL_GUESS, eps_abs=1e-9)
 
 tic = perf_counter_ns()
-proxsuite.proxqp.dense.solve_in_parallel(qps=qps)
-timings[f"solve_parallel_heuristics_threads"] = (perf_counter_ns() - tic) * 1e-6
+with ThreadPoolExecutor(max_workers=num_threads) as executor:
+    results = list(executor.map(solve_problem, problems))
+timings["solve_parallel_dense"] = (perf_counter_ns() - tic) * 1e-6
 
 for k, v in timings.items():
-    print(f"{k}: {v}ms")
+    print(f"{k}: {v:.3f}ms")

bindings/python/src/expose-solve.hpp

@@ -76,7 +76,8 @@ solveDenseQp(nanobind::module_ m)
     nanobind::arg("eps_duality_gap_abs") = nanobind::none(),
     nanobind::arg("eps_duality_gap_rel") = nanobind::none(),
     nanobind::arg("primal_infeasibility_solving") = false,
-    nanobind::arg("default_H_eigenvalue_estimate") = 0.);
+    nanobind::arg("default_H_eigenvalue_estimate") = 0.,
+    nanobind::call_guard<nanobind::gil_scoped_release>());
 
   m.def(
     "solve",
@@ -139,7 +140,8 @@ solveDenseQp(nanobind::module_ m)
     nanobind::arg("eps_duality_gap_abs") = nanobind::none(),
     nanobind::arg("eps_duality_gap_rel") = nanobind::none(),
     nanobind::arg("primal_infeasibility_solving") = false,
-    nanobind::arg("default_H_eigenvalue_estimate") = 0.);
+    nanobind::arg("default_H_eigenvalue_estimate") = 0.,
+    nanobind::call_guard<nanobind::gil_scoped_release>());
 }
 
 } // namespace python
@@ -186,7 +188,8 @@ solveSparseQp(nanobind::module_ m)
     nanobind::arg("eps_duality_gap_abs") = nanobind::none(),
     nanobind::arg("eps_duality_gap_rel") = nanobind::none(),
     nanobind::arg("primal_infeasibility_solving") = false,
-    nanobind::arg("default_H_eigenvalue_estimate") = 0.);
+    nanobind::arg("default_H_eigenvalue_estimate") = 0.,
+    nanobind::call_guard<nanobind::gil_scoped_release>());
 }
 
 } // namespace python