From 17601375868c6473dbd5b4f162260a48bd160e1b Mon Sep 17 00:00:00 2001
From: Charles Kawczynski <kawczynski.charles@gmail.com>
Date: Mon, 11 Mar 2024 09:33:53 -0400
Subject: [PATCH] Improve CPU performance implementation

---
 test/utils.jl | 16 ++++++++++++++--
 1 file changed, 14 insertions(+), 2 deletions(-)

diff --git a/test/utils.jl b/test/utils.jl
index a31f4eb..b44af9c 100644
--- a/test/utils.jl
+++ b/test/utils.jl
@@ -45,13 +45,25 @@ if !hasmethod(Base.copyto!, Tuple{<:FusedMultiBroadcast})
     end
 end
 
+# This is better than the baseline.
 function copyto_cpu!(pairs::T, ei::EI) where {T, EI}
-    @inbounds @simd ivdep for i in ei
-        MBF.rcopyto_at!(pairs, i)
+    for (dest, bc) in pairs
+        @inbounds @simd ivdep for i in ei
+            dest[i] = bc[i]
+        end
     end
     return nothing
 end
 
+# This should, in theory be better, but it seems like inlining is
+# failing somewhere.
+# function copyto_cpu!(pairs::T, ei::EI) where {T, EI}
+#     @inbounds @simd ivdep for i in ei
+#         MBF.rcopyto_at!(pairs, i)
+#     end
+#     return nothing
+# end
+
 import CUDA
 import Adapt
 function copyto_cuda!(pairs::Tuple) # (Pair(dest1, bc1),Pair(dest2, bc2),...)