Documented what GMX_CPU_ACCELERATION really means

Discussion arose in I0848f2e570daef881368a45d4a429a3f80fde81b

Change-Id: Ib99ab86ff7d8acf913121decb00756f47327f046

diff --git a/CMakeLists.txt b/CMakeLists.txt
index ebfefb3586dfdfabbb08b6de3a50f4c340aade45..0a96f0287d20abdc301ca07bda0a513a60f30ac0 100644 (file)
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -661,6 +661,14 @@ endif(NOT GMX_SYSTEM_XDR)
 include(gmxTestAVXMaskload)
 
 # Process nonbonded accelerated kernels settings
+#
+# Note that for the backward-compatible x86 SIMD architectures, the
+# GMX_CPU_ACCELERATION determines the maximum level of the instruction
+# set used (e.g. GMX_CPU_ACCLERATION=SSE4.1 implies
+# SSE2). Accordingly, there are a set of CMake variables
+# GMX_<arch>_<feature-set> that are exported to the C code to specify
+# CPU features that should be used. This means that the logic for
+# requiring such backward compatibility is all located here.
 string(TOUPPER ${GMX_CPU_ACCELERATION} GMX_CPU_ACCELERATION)
 if(${GMX_CPU_ACCELERATION} STREQUAL "NONE")
     # nothing to do