From 0586ef150a3f37baba4edac7af452059475610d5 Mon Sep 17 00:00:00 2001
From: Maximilian Gaul <ptfs410h@woody3.rrze.uni-erlangen.de>
Date: Mon, 15 Nov 2021 19:39:09 +0100
Subject: [PATCH] Fix num of threads instead of num of blocks, add logbook
 template

---
 src/force.cu | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/src/force.cu b/src/force.cu
index 72ceb0a..dfd8792 100644
--- a/src/force.cu
+++ b/src/force.cu
@@ -170,8 +170,8 @@ double computeForce(
     checkError( "c_neigh_numneigh malloc", cudaMalloc((void**)&c_neigh_numneigh, sizeof(int) * Nlocal) );
     checkError( "c_neigh_numneigh memcpy", cudaMemcpy(c_neigh_numneigh, neighbor->numneigh, sizeof(int) * Nlocal, cudaMemcpyHostToDevice) );
 
-    const int num_blocks = 1024;
-    const int num_threads_per_block = ceil((float)Nlocal / (float)num_blocks);
+    const int num_threads_per_block = 32; // this should be multiple of 32 as operations are performed at the level of warps
+    const int num_blocks = ceil((float)Nlocal / (float)num_threads_per_block);
 
     double S = getTimeStamp();
     LIKWID_MARKER_START("force");