From 5835fe293ab693afac377d6341bda0a6b706f48b Mon Sep 17 00:00:00 2001
From: Peter Thoman <peter@metaclassofnil.com>
Date: Fri, 3 May 2013 14:27:02 +0200
Subject: [PATCH] When processing a parallel loop, save some context switching
 by processing one chunk on the calling thread

---
 thread/threadpool.cpp | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/thread/threadpool.cpp b/thread/threadpool.cpp
index 30fbb9af96..dcc733025d 100644
--- a/thread/threadpool.cpp
+++ b/thread/threadpool.cpp
@@ -64,10 +64,13 @@ void ThreadPool::ParallelLoop(function<void(int,int)> loop, int lower, int upper
 		// could do slightly better load balancing for the generic case, 
 		// but doesn't matter since all our loops are power of 2
 		int chunk = range/numThreads; 
-		for(int s=lower, i=0; i<numThreads; s+=chunk, ++i) {
-			workers[i]->Process(bind(loop, s, std::min(s+chunk,upper)));
+		int s = lower;
+		for(int i=0; i<numThreads-1; ++i) {
+			workers[i]->Process(bind(loop, s, s+chunk));
+			s+=chunk;
 		}
-		for(int i=0; i<numThreads; ++i) {
+		loop(s, upper);
+		for(int i=0; i<numThreads-1; ++i) {
 			workers[i]->WaitForCompletion();
 		}
 	} else {