Optimizations for 32bit machines by using a 64bit word type and 32bit digit.

Unfortunately we cannot do the same for 64bit since we don't have an 128bit type in kernel.

4 files changed, 16 insertions, 7 deletions

diff --git a/libtommath/bn_mp_div.c b/libtommath/bn_mp_div.c
index aee9c94324d..723c8b503dd 100644
--- a/libtommath/bn_mp_div.c
+++ b/libtommath/bn_mp_div.c
@@ -195,7 +195,7 @@ int mp_div (mp_int * a, mp_int * b, mp_int * c, mp_int * d)
       mp_word tmp;
       tmp = ((mp_word) x.dp[i]) << ((mp_word) DIGIT_BIT);
       tmp |= ((mp_word) x.dp[i - 1]);
-      tmp /= ((mp_word) y.dp[t]);
+      tmp = word_div_int(tmp, (int)y.dp[t]);
       if (tmp > (mp_word) MP_MASK)
         tmp = MP_MASK;
       q.dp[i - t - 1] = (mp_digit) (tmp & (mp_word) (MP_MASK));
diff --git a/libtommath/bn_mp_div_3.c b/libtommath/bn_mp_div_3.c
index 3c60269ecea..608e759e366 100644
--- a/libtommath/bn_mp_div_3.c
+++ b/libtommath/bn_mp_div_3.c
@@ -25,7 +25,7 @@ mp_div_3 (mp_int * a, mp_int *c, mp_digit * d)
   int      res, ix;
   
   /* b = 2**DIGIT_BIT / 3 */
-  b = (((mp_word)1) << ((mp_word)DIGIT_BIT)) / ((mp_word)3);
+  b = word_div_int (((mp_word)1) << DIGIT_BIT, 3);
 
   if ((res = mp_init_size(&q, a->used)) != MP_OKAY) {
      return res;
diff --git a/libtommath/bn_mp_div_d.c b/libtommath/bn_mp_div_d.c
index 6a26d4f0cf6..63efc2cfa69 100644
--- a/libtommath/bn_mp_div_d.c
+++ b/libtommath/bn_mp_div_d.c
@@ -87,7 +87,7 @@ int mp_div_d (mp_int * a, mp_digit b, mp_int * c, mp_digit * d)
      w = (w << ((mp_word)DIGIT_BIT)) | ((mp_word)a->dp[ix]);
      
      if (w >= b) {
-        t = (mp_digit)(w / b);
+        t = (mp_digit)(word_div_int(w, b));
         w -= ((mp_word)t) * ((mp_word)b);
       } else {
         t = 0;
diff --git a/libtommath/tommath.h b/libtommath/tommath.h
index 6653f5550f5..3fa7ae89853 100644
--- a/libtommath/tommath.h
+++ b/libtommath/tommath.h
@@ -20,6 +20,7 @@
 #include <linux/slab.h>
 #include <linux/random.h>
 #include <linux/ctype.h>
+#include <linux/math64.h>
 
 #define CHAR_BIT sizeof(uint8_t)*8
 
@@ -65,19 +66,22 @@ extern "C" {
  * [any size beyond that is ok provided it doesn't overflow the data type]
  */
  
-/* FIXME: This can be improved, but might require to use a 64bit division
- * on 32bit machines and an 128bit on 64.
+/* FIXME: This can be improved, but requires to use 128bit division
+ * on 64bit machines, which is not available in kernel now.
  */
-#if BITS_PER_LONG <= 32
+#if BITS_PER_LONG < 32
 
    typedef uint16_t      mp_digit;
    typedef uint32_t      mp_word;
 # define DIGIT_BIT          15
 
-#elif BITS_PER_LONG == 64
+#elif BITS_PER_LONG <= 64
 
    typedef uint32_t      mp_digit;
    typedef uint64_t      mp_word;
+
+# define word_div_int(x,y) div_u64((x),(y))
+
 # define DIGIT_BIT          31
 
 #endif
@@ -92,6 +96,11 @@ extern "C" {
 
 #endif
 
+#ifndef word_div_int
+# define word_div_int(x,y) ((x)/(y))
+#endif
+
+
 /* define heap macros */
 #ifndef XMALLOC
 # define XMALLOC(x) kmalloc(x, GFP_KERNEL)