From b0403a8ee4efce599657801bd4921374b82b2733 Mon Sep 17 00:00:00 2001 From: johnmccutchan Date: Thu, 14 Feb 2008 22:15:25 +0000 Subject: [PATCH] Add notes about future optimization possibilities. --- Extras/BulletMultiThreaded/SpuRaycastTask/SpuRaycastTask.cpp | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/Extras/BulletMultiThreaded/SpuRaycastTask/SpuRaycastTask.cpp b/Extras/BulletMultiThreaded/SpuRaycastTask/SpuRaycastTask.cpp index 0f71fb5b6..7f6567ea5 100644 --- a/Extras/BulletMultiThreaded/SpuRaycastTask/SpuRaycastTask.cpp +++ b/Extras/BulletMultiThreaded/SpuRaycastTask/SpuRaycastTask.cpp @@ -10,7 +10,10 @@ /* Future optimization strategies: 1. BBOX prune before loading shape data -2. When doing bvh tree traversal do it once for entire batch of rays. +2. Could reduce number of dmas for ray output data to a single read and write. + By sharing the temporary work unit output structures across objects. +3. The reason SpuRaycastNodeCallback1 is slower is because the triangle data isn't + being cached across calls. Fix that by doing the final ray pruning inside the callback. */ /* Future work: