wolfcrypt/src/aes.c: add static volatile word32 x_volatile to PreFetchTe() and PreFetchSBox(), to move volatile accesses out of the loop iteration, reclaiming some (but not all) of the performance regression from 299e7bd.

Author: douzzer

wolfcrypt/src/aes.c

@@ -1880,7 +1880,8 @@ static WARN_UNUSED_RESULT word32 inv_col_mul(
 static WARN_UNUSED_RESULT WC_INLINE word32 PreFetchTe(void)
 {
 #ifndef WOLFSSL_AES_TOUCH_LINES
-    volatile word32 x = 0;
+    static volatile word32 x_volatile = 0;
+    word32 x = x_volatile;
     int i;
     int j;
 
@@ -1891,6 +1892,8 @@ static WARN_UNUSED_RESULT WC_INLINE word32 PreFetchTe(void)
         }
     }
 
+    x_volatile = x;
+
     return x;
 #else
     return 0;
@@ -1901,13 +1904,16 @@ static WARN_UNUSED_RESULT WC_INLINE word32 PreFetchTe(void)
 static WARN_UNUSED_RESULT WC_INLINE word32 PreFetchSBox(void)
 {
 #ifndef WOLFSSL_AES_TOUCH_LINES
-    volatile word32 x = 0;
+    static volatile word32 x_volatile = 0;
+    word32 x = x_volatile;
     int i;
 
     for (i = 0; i < 256; i += WC_CACHE_LINE_SZ/4) {
         x &= Tsbox[i];
     }
 
+    x_volatile = x;
+
     return x;
 #else
     return 0;