Use bn_sqr_words() from bn_mod_sqr_words().

Use bn_sqr_words() and bn_montgomery_reduce_words(), rather than using
bn_montgomery_multiply_words(). This provides better performance on
architectures that have assembly optimised bn_sqr_words(), such as amd64.

ok tb@

1 files changed, 3 insertions, 5 deletions

diff --git a/src/lib/libcrypto/bn/bn_mod_words.c b/src/lib/libcrypto/bn/bn_mod_words.c
index d9aee8701a..7c07b49fab 100644
--- a/src/lib/libcrypto/bn/bn_mod_words.c
+++ b/src/lib/libcrypto/bn/bn_mod_words.c
@@ -1,4 +1,4 @@
-/*      $OpenBSD: bn_mod_words.c,v 1.3 2025/08/05 15:15:54 jsing Exp $  */
+/*      $OpenBSD: bn_mod_words.c,v 1.4 2025/09/01 15:18:55 jsing Exp $  */
 /*
  * Copyright (c) 2024 Joel Sing <jsing@openbsd.org>
  *
@@ -100,15 +100,13 @@ bn_mod_sqr_words(BN_ULONG *r, const BN_ULONG *a, const BN_ULONG *m,
 {
         if (n == 4) {
                 bn_sqr_comba4(t, a);
-                bn_montgomery_reduce_words(r, t, m, m0, n);
         } else if (n == 6) {
                 bn_sqr_comba6(t, a);
-                bn_montgomery_reduce_words(r, t, m, m0, n);
         } else if (n == 8) {
                 bn_sqr_comba8(t, a);
-                bn_montgomery_reduce_words(r, t, m, m0, n);
         } else {
-                bn_montgomery_multiply_words(r, a, a, m, t, m0, n);
+                bn_sqr_words(t, a, n);
         }
+        bn_montgomery_reduce_words(r, t, m, m0, n);
 }
 #endif