From c4b9f71357fc454c8709d2c190a1fd5b90bf116e Mon Sep 17 00:00:00 2001
From: Peter Boyle <paboyle@ph.ed.ac.uk>
Date: Tue, 21 May 2024 02:22:32 +0100
Subject: [PATCH] CPU compile ordering is important

---
 Grid/threads/Accelerator.h | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/Grid/threads/Accelerator.h b/Grid/threads/Accelerator.h
index 48062b56..1cb56ddd 100644
--- a/Grid/threads/Accelerator.h
+++ b/Grid/threads/Accelerator.h
@@ -539,12 +539,6 @@ inline void acceleratorCopySynchronise(void) { auto discard=hipStreamSynchronize
 
 #endif
 
-inline void acceleratorCopyDeviceToDevice(void *from,void *to,size_t bytes)
-{
-  acceleratorCopyDeviceToDeviceAsynch(from,to,bytes);
-  acceleratorCopySynchronise();
-}
-
 //////////////////////////////////////////////
 // CPU Target - No accelerator just thread instead
 //////////////////////////////////////////////
@@ -553,7 +547,6 @@ inline void acceleratorCopyDeviceToDevice(void *from,void *to,size_t bytes)
 
 #undef GRID_SIMT
 
-
 inline void acceleratorMem(void)
 {
   /*
@@ -656,6 +649,12 @@ accelerator_inline void acceleratorFence(void)
   return;
 }
 
+inline void acceleratorCopyDeviceToDevice(void *from,void *to,size_t bytes)
+{
+  acceleratorCopyDeviceToDeviceAsynch(from,to,bytes);
+  acceleratorCopySynchronise();
+}
+
 template<class T> void acceleratorPut(T& dev,T&host)
 {
   acceleratorCopyToDevice(&host,&dev,sizeof(T));