diff --git a/src/mzscheme/gc2/newgc.c b/src/mzscheme/gc2/newgc.c
index 91e2ae8c36..742fffb4f9 100644
--- a/src/mzscheme/gc2/newgc.c
+++ b/src/mzscheme/gc2/newgc.c
@@ -31,6 +31,7 @@
 #include <stdlib.h>
 #include <stdio.h>
 #include <string.h>
+#include "platforms.h"
 #include "gc2.h"
 #include "gc2_dump.h"
 #include "newgc_internal.h"
@@ -38,30 +39,6 @@
 
 static THREAD_LOCAL NewGC *GC;
 
-#ifdef _WIN32
-# include <windows.h>
-# define bzero(m, s) memset(m, 0, s)
-# define inline _inline
-#endif
-
-#if defined(sparc) || defined(__sparc) || defined(__sparc__)
-/* Required for `double' operations: */
-# define GC_ALIGN_EIGHT
-#endif
-
-/* Even when 8-byte alginment is not required by the processor, it's
-   better for floating-point performance (PowerPC) and may be required
-   for some libraries (VecLib in Mac OS X, including x86).
-
-   Under Windows, Mac OS X, and Linux x86_64, malloc() returns 16-byte
-   aligned data. And, actually, VecLib says that it requires
-   16-byte-aligned data. So, in those cases, GC_ALIGN_SIXTEEN might be
-   better --- but that's a lot more expensive, increasing DrScheme's
-   initial footprint by almost 10%. */
-#ifndef GC_ALIGN_EIGHT
-# define GC_ALIGN_EIGHT
-#endif
-
 #include "msgprint.c"
 
 /*****************************************************************************/
diff --git a/src/mzscheme/gc2/platforms.h b/src/mzscheme/gc2/platforms.h
new file mode 100644
index 0000000000..23d04884a2
--- /dev/null
+++ b/src/mzscheme/gc2/platforms.h
@@ -0,0 +1,23 @@
+#ifdef _WIN32
+# include <windows.h>
+# define bzero(m, s) memset(m, 0, s)
+# define inline _inline
+#endif
+
+#if defined(sparc) || defined(__sparc) || defined(__sparc__)
+/* Required for `double' operations: */
+# define GC_ALIGN_EIGHT
+#endif
+
+/* Even when 8-byte alginment is not required by the processor, it's
+   better for floating-point performance (PowerPC) and may be required
+   for some libraries (VecLib in Mac OS X, including x86).
+
+   Under Windows, Mac OS X, and Linux x86_64, malloc() returns 16-byte
+   aligned data. And, actually, VecLib says that it requires
+   16-byte-aligned data. So, in those cases, GC_ALIGN_SIXTEEN might be
+   better --- but that's a lot more expensive, increasing DrScheme's
+   initial footprint by almost 10%. */
+#ifndef GC_ALIGN_EIGHT
+# define GC_ALIGN_EIGHT
+#endif