From c23c694affb76b79cdf71003249ce4ca3868b46e Mon Sep 17 00:00:00 2001
From: noraabiakar <nora.abiakar@gmail.com>
Date: Fri, 18 Jan 2019 14:42:24 +0100
Subject: [PATCH] Optimize vectorized compound_indexed_add (#673)
* Optimize "none" index_constraint specialization of compound_indexed_add, so that it only reads/writes each distinct memory index once per vector.
Related to issue #637.
---
include/arbor/simd/simd.hpp | 10 ++++++++--
1 file changed, 8 insertions(+), 2 deletions(-)
diff --git a/include/arbor/simd/simd.hpp b/include/arbor/simd/simd.hpp
index 4a319216..535b8233 100644
--- a/include/arbor/simd/simd.hpp
+++ b/include/arbor/simd/simd.hpp
@@ -220,9 +220,15 @@ namespace simd_detail {
scalar_type a[width];
Impl::copy_to(s, a);
- for (unsigned i = 0; i<width; ++i) {
- p[o[i]] += a[i];
+ scalar_type temp = 0;
+ for (unsigned i = 0; i<width-1; ++i) {
+ temp += a[i];
+ if (o[i] != o[i+1]) {
+ p[o[i]] += temp;
+ temp = 0;
+ }
}
+ p[o[width-1]] = temp;
}
break;
case index_constraint::independent:
--
GitLab