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, 9 insertions, 0 deletions

diff --git a/src/lib/math/mp/mp_karat.cpp b/src/lib/math/mp/mp_karat.cpp
index b549a05c8..62620f83d 100644
--- a/src/lib/math/mp/mp_karat.cpp
+++ b/src/lib/math/mp/mp_karat.cpp
@@ -237,6 +237,11 @@ void bigint_mul(word z[], size_t z_size, word workspace[],
       {
       bigint_comba_mul8(z, x, y);
       }
+   else if(x_sw <= 9 && x_size >= 9 &&
+           y_sw <= 9 && y_size >= 9 && z_size >= 18)
+      {
+      bigint_comba_mul9(z, x, y);
+      }
    else if(x_sw <= 16 && x_size >= 16 &&
            y_sw <= 16 && y_size >= 16 && z_size >= 32)
       {
@@ -281,6 +286,10 @@ void bigint_sqr(word z[], size_t z_size, word workspace[],
       {
       bigint_comba_sqr8(z, x);
       }
+   else if(x_sw == 9 && x_size >= 9 && z_size >= 18)
+      {
+      bigint_comba_sqr9(z, x);
+      }
    else if(x_sw <= 16 && x_size >= 16 && z_size >= 32)
       {
       bigint_comba_sqr16(z, x);