Changes from 2993 to limit parallelism more conservatively

ROCm · Apr 23, 2024 · cba5184 · cba5184
1 parent fad9b78
commit cba5184
Showing 1 changed file with 4 additions and 4 deletions.
diff --git a/src/propagate_constant.cpp b/src/propagate_constant.cpp
@@ -84,12 +84,12 @@ void propagate_constant::apply(module& m) const
     // Compute literals in parallel
     std::vector<instruction_ref> const_instrs_vec{const_instrs.begin(), const_instrs.end()};
     std::vector<argument> literals(const_instrs_vec.size());
-    std::size_t n = 1;
+    std::size_t grainsize = 1;
 #if !MIGRAPHX_HAS_EXECUTORS
-    n = std::max<std::size_t>(
-        std::ceil(static_cast<double>(1024) / std::thread::hardware_concurrency()), 1);
+    std::size_t n = std::max<std::size_t>(2048 / std::thread::hardware_concurrency(), 1);
+    grainsize     = const_instrs_vec.size() / n;
 #endif
-    simple_par_for(const_instrs_vec.size(), n, [&](const auto i) {
+    simple_par_for(const_instrs_vec.size(), grainsize, [&](const auto i) {
         literals[i] = const_instrs_vec[i]->eval();
     });