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).

1 files changed, 18 insertions, 0 deletions

diff --git a/src/lib/math/ec_gfp/curve_gfp.h b/src/lib/math/ec_gfp/curve_gfp.h
index 67fb3d2cf..59639d537 100644
--- a/src/lib/math/ec_gfp/curve_gfp.h
+++ b/src/lib/math/ec_gfp/curve_gfp.h
@@ -24,6 +24,8 @@ class CurveGFp_Repr
       virtual const BigInt& get_a() const = 0;
       virtual const BigInt& get_b() const = 0;
 
+      virtual size_t get_p_words() const = 0;
+
       /*
       * Returns to_curve_rep(get_a())
       */
@@ -43,6 +45,10 @@ class CurveGFp_Repr
 
       virtual void curve_sqr(BigInt& z, const BigInt& x,
                              secure_vector<word>& ws) const = 0;
+
+      virtual void normalize(BigInt& x,
+                             secure_vector<word>& ws,
+                             size_t bound) const;
    };
 
 /**
@@ -109,6 +115,8 @@ class BOTAN_DLL CurveGFp
          return xt;
          }
 
+      // TODO: from_rep taking && ref
+
       void mul(BigInt& z, const BigInt& x, const BigInt& y, secure_vector<word>& ws) const
          {
          m_repr->curve_mul(z, x, y, ws);
@@ -133,6 +141,16 @@ class BOTAN_DLL CurveGFp
          return z;
          }
 
+      /**
+      * Adjust x to be in [0,p)
+      * @param bound if greater than zero, assume that no more than bound
+      *        additions or subtractions are required to move x into range.
+      */
+      void normalize(BigInt& x, secure_vector<word>& ws, size_t bound = 0) const
+         {
+         m_repr->normalize(x, ws, bound);
+         }
+
       void swap(CurveGFp& other)
          {
          std::swap(m_repr, other.m_repr);