From dba6b846964e83b0fc3542084e16e653a837dae0 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tommi=20M=C3=A4klin?= <tommi@maklin.fi>
Date: Sat, 17 Aug 2024 09:27:20 +0300
Subject: [PATCH] Revert "Faster computing of group_counts (no need to save
 memory here)."

This reverts commit 7f353cb046778b821ecc0d1fdee27495c97320e5. The
optimizations did not end up working on large inputs.
---
 include/Likelihood.hpp | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/include/Likelihood.hpp b/include/Likelihood.hpp
index 17c5099..d182a89 100644
--- a/include/Likelihood.hpp
+++ b/include/Likelihood.hpp
@@ -119,16 +119,21 @@ class LL_WOR21 : public Likelihood<T> {
 #endif
 
     // This double loop is currently the slowest part in the input reading
-    std::vector<size_t> group_counts(num_ecs*n_groups, 0);
+    std::vector<bm::sparse_vector<V, bm::bvector<>>> local_counts(n_threads);
 #pragma omp parallel for schedule(static)
     for (size_t i = 0; i < num_ecs; ++i) {
 	for (size_t j = 0; j < n_targets; ++j) {
 	    if (alignment(i, j)) {
-		++group_counts[alignment.get_groups()[j]*num_ecs + i];
+		local_counts[omp_get_thread_num()].inc(alignment.get_groups()[j]*num_ecs + i);
 	    }
 	}
     }
 
+    bm::sparse_vector<V, bm::bvector<>> group_counts = std::move(local_counts[0]);
+    for (size_t i = 1; i < n_threads; ++i) {
+	group_counts.merge(local_counts[i]);
+    }
+
     bool mask_groups = min_hits > 0;
     this->groups_mask = std::vector<bool>(n_groups, !mask_groups);
     std::vector<V> masked_group_sizes;