Add specialized reduction for P-521 along with 9x9 Comba routines.

Roughly 35-50% faster on my laptop (depending on if mlock is enabled,
the overhead in that allocator is becoming much more of a hotspot).

2 files changed, 8 insertions, 8 deletions

diff --git a/src/lib/math/bigint/bigint.cpp b/src/lib/math/bigint/bigint.cpp
index 059b019e4..90a319c5a 100644
--- a/src/lib/math/bigint/bigint.cpp
+++ b/src/lib/math/bigint/bigint.cpp
@@ -10,6 +10,7 @@
 #include <botan/get_byte.h>
 #include <botan/parsing.h>
 #include <botan/internal/rounding.h>
+#include <botan/internal/bit_ops.h>
 
 namespace Botan {
 
@@ -208,7 +209,6 @@ void BigInt::clear_bit(size_t n)
 void BigInt::mask_bits(size_t n)
    {
    if(n == 0) { clear(); return; }
-   if(n >= bits()) return;
 
    const size_t top_word = n / MP_WORD_BITS;
    const word mask = (static_cast<word>(1) << (n % MP_WORD_BITS)) - 1;
@@ -237,13 +237,8 @@ size_t BigInt::bits() const
    if(words == 0)
       return 0;
 
-   size_t full_words = words - 1, top_bits = MP_WORD_BITS;
-   word top_word = word_at(full_words), mask = MP_WORD_TOP_BIT;
-
-   while(top_bits && ((top_word & mask) == 0))
-      { mask >>= 1; top_bits--; }
-
-   return (full_words * MP_WORD_BITS + top_bits);
+   const size_t full_words = words - 1;
+   return (full_words * MP_WORD_BITS + high_bit(word_at(full_words)));
    }
 
 /*
diff --git a/src/lib/math/bigint/bigint.h b/src/lib/math/bigint/bigint.h
index 0d9b43357..2205c7e83 100644
--- a/src/lib/math/bigint/bigint.h
+++ b/src/lib/math/bigint/bigint.h
@@ -120,6 +120,11 @@ class BOTAN_DLL BigInt
         std::swap(m_signedness, other.m_signedness);
         }
 
+     void swap_reg(secure_vector<word>& reg)
+        {
+        m_reg.swap(reg);
+        }
+
      /**
      * += operator
      * @param y the BigInt to add to this