This lets sdlMAME/sdlMESS compile on Fedora GCC 5.1.1 with LTO enabled with line:

make -jX SSE2=1 LTO=1

(pick X wisely as Xx4 link threads will run to speed up lto)

Super Mario 64 runs at 88% on old I7-920 in a VM (from 55%).

Windows Mingw-64 5.1.1 LTO is not possible for now :(

Why do we need to say 'SSE2=1' since MAME now requires a CPU with that for a minimum?

(nw)

scripts/genie.lua

@@ -800,7 +800,7 @@ if _OPTIONS["OPTIMIZE"] then
 		}
 	end
 	if _OPTIONS["LTO"]=="1" then
--- -flto=4 -> 4 threads
+-- -flto=4 -> 4 threads, reduce if you are low on memory (less than 8G)
 		buildoptions {
 			"-flto=4",
 		}
@@ -826,7 +826,9 @@ end
 
 if _OPTIONS["SSE2"]=="1" then
 	buildoptions {
+		"-msse",
 		"-msse2",
+		"-msse3"
 	}
 end