diff --git a/frame/base/bli_cpuid.c b/frame/base/bli_cpuid.c
index 7385143eb..690e22304 100644
--- a/frame/base/bli_cpuid.c
+++ b/frame/base/bli_cpuid.c
@@ -45,19 +45,21 @@
   #define __arm__
 #endif
 
-#ifndef BLIS_CONFIGURETIME_CPUID
-  #include "blis.h"
-#else
+#ifdef BLIS_CONFIGURETIME_CPUID
   #define BLIS_EXPORT_BLIS
   #include "bli_system.h"
   #include "bli_type_defs.h"
   #include "bli_cpuid.h"
+#else
+  #include "blis.h"
 #endif
 
 // -----------------------------------------------------------------------------
 
 #if defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86)
 
+#include "cpuid.h"
+
 arch_t bli_cpuid_query_id( void )
 {
 	uint32_t vendor, family, model, features;
diff --git a/frame/base/bli_cpuid.h b/frame/base/bli_cpuid.h
index 3c4d82d1d..88d7aaa82 100644
--- a/frame/base/bli_cpuid.h
+++ b/frame/base/bli_cpuid.h
@@ -125,7 +125,9 @@ static bool_t bli_cpuid_has_features( uint32_t have, uint32_t want )
 
 #if defined(__x86_64__) || defined(_M_X64) || defined(__i386) || defined(_M_IX86)
 
-#include "cpuid.h"
+// cpuid.h is now #included in bli_cpuid.c instead of here. See issue #393
+// // for more information why this move was made.
+// //#include "cpuid.h"
 
 void get_cpu_name( char *cpu_name );
 int  vpu_count( void );