Move bn_{mul,sqr}_comba{4,8}() from bn_asm.c to bn_mul.c/bn_sqr.c.

Wrap these in HAVE_BN_{MUL,SQR}_COMBA{4,8} defines. Add these defines to
bn_arch.h where the architecture currently provides its own version.

ok tb@

1 files changed, 150 insertions, 1 deletions

diff --git a/src/lib/libcrypto/bn/bn_mul.c b/src/lib/libcrypto/bn/bn_mul.c
index b7a7f8bcef..3a69ef35da 100644
--- a/src/lib/libcrypto/bn/bn_mul.c
+++ b/src/lib/libcrypto/bn/bn_mul.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: bn_mul.c,v 1.27 2023/01/20 12:16:46 jsing Exp $ */
+/* $OpenBSD: bn_mul.c,v 1.28 2023/01/20 17:31:52 jsing Exp $ */
 /* Copyright (C) 1995-1998 Eric Young (eay@cryptsoft.com)
  * All rights reserved.
  *
@@ -62,8 +62,157 @@
 
 #include <openssl/opensslconf.h>
 
+#include "bn_arch.h"
 #include "bn_local.h"
 
+#ifndef HAVE_BN_MUL_COMBA4
+void
+bn_mul_comba4(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
+{
+        BN_ULONG c1, c2, c3;
+
+        c1 = 0;
+        c2 = 0;
+        c3 = 0;
+        mul_add_c(a[0], b[0], c1, c2, c3);
+        r[0] = c1;
+        c1 = 0;
+        mul_add_c(a[0], b[1], c2, c3, c1);
+        mul_add_c(a[1], b[0], c2, c3, c1);
+        r[1] = c2;
+        c2 = 0;
+        mul_add_c(a[2], b[0], c3, c1, c2);
+        mul_add_c(a[1], b[1], c3, c1, c2);
+        mul_add_c(a[0], b[2], c3, c1, c2);
+        r[2] = c3;
+        c3 = 0;
+        mul_add_c(a[0], b[3], c1, c2, c3);
+        mul_add_c(a[1], b[2], c1, c2, c3);
+        mul_add_c(a[2], b[1], c1, c2, c3);
+        mul_add_c(a[3], b[0], c1, c2, c3);
+        r[3] = c1;
+        c1 = 0;
+        mul_add_c(a[3], b[1], c2, c3, c1);
+        mul_add_c(a[2], b[2], c2, c3, c1);
+        mul_add_c(a[1], b[3], c2, c3, c1);
+        r[4] = c2;
+        c2 = 0;
+        mul_add_c(a[2], b[3], c3, c1, c2);
+        mul_add_c(a[3], b[2], c3, c1, c2);
+        r[5] = c3;
+        c3 = 0;
+        mul_add_c(a[3], b[3], c1, c2, c3);
+        r[6] = c1;
+        r[7] = c2;
+}
+#endif
+
+#ifndef HAVE_BN_MUL_COMBA8
+void
+bn_mul_comba8(BN_ULONG *r, BN_ULONG *a, BN_ULONG *b)
+{
+        BN_ULONG c1, c2, c3;
+
+        c1 = 0;
+        c2 = 0;
+        c3 = 0;
+        mul_add_c(a[0], b[0], c1, c2, c3);
+        r[0] = c1;
+        c1 = 0;
+        mul_add_c(a[0], b[1], c2, c3, c1);
+        mul_add_c(a[1], b[0], c2, c3, c1);
+        r[1] = c2;
+        c2 = 0;
+        mul_add_c(a[2], b[0], c3, c1, c2);
+        mul_add_c(a[1], b[1], c3, c1, c2);
+        mul_add_c(a[0], b[2], c3, c1, c2);
+        r[2] = c3;
+        c3 = 0;
+        mul_add_c(a[0], b[3], c1, c2, c3);
+        mul_add_c(a[1], b[2], c1, c2, c3);
+        mul_add_c(a[2], b[1], c1, c2, c3);
+        mul_add_c(a[3], b[0], c1, c2, c3);
+        r[3] = c1;
+        c1 = 0;
+        mul_add_c(a[4], b[0], c2, c3, c1);
+        mul_add_c(a[3], b[1], c2, c3, c1);
+        mul_add_c(a[2], b[2], c2, c3, c1);
+        mul_add_c(a[1], b[3], c2, c3, c1);
+        mul_add_c(a[0], b[4], c2, c3, c1);
+        r[4] = c2;
+        c2 = 0;
+        mul_add_c(a[0], b[5], c3, c1, c2);
+        mul_add_c(a[1], b[4], c3, c1, c2);
+        mul_add_c(a[2], b[3], c3, c1, c2);
+        mul_add_c(a[3], b[2], c3, c1, c2);
+        mul_add_c(a[4], b[1], c3, c1, c2);
+        mul_add_c(a[5], b[0], c3, c1, c2);
+        r[5] = c3;
+        c3 = 0;
+        mul_add_c(a[6], b[0], c1, c2, c3);
+        mul_add_c(a[5], b[1], c1, c2, c3);
+        mul_add_c(a[4], b[2], c1, c2, c3);
+        mul_add_c(a[3], b[3], c1, c2, c3);
+        mul_add_c(a[2], b[4], c1, c2, c3);
+        mul_add_c(a[1], b[5], c1, c2, c3);
+        mul_add_c(a[0], b[6], c1, c2, c3);
+        r[6] = c1;
+        c1 = 0;
+        mul_add_c(a[0], b[7], c2, c3, c1);
+        mul_add_c(a[1], b[6], c2, c3, c1);
+        mul_add_c(a[2], b[5], c2, c3, c1);
+        mul_add_c(a[3], b[4], c2, c3, c1);
+        mul_add_c(a[4], b[3], c2, c3, c1);
+        mul_add_c(a[5], b[2], c2, c3, c1);
+        mul_add_c(a[6], b[1], c2, c3, c1);
+        mul_add_c(a[7], b[0], c2, c3, c1);
+        r[7] = c2;
+        c2 = 0;
+        mul_add_c(a[7], b[1], c3, c1, c2);
+        mul_add_c(a[6], b[2], c3, c1, c2);
+        mul_add_c(a[5], b[3], c3, c1, c2);
+        mul_add_c(a[4], b[4], c3, c1, c2);
+        mul_add_c(a[3], b[5], c3, c1, c2);
+        mul_add_c(a[2], b[6], c3, c1, c2);
+        mul_add_c(a[1], b[7], c3, c1, c2);
+        r[8] = c3;
+        c3 = 0;
+        mul_add_c(a[2], b[7], c1, c2, c3);
+        mul_add_c(a[3], b[6], c1, c2, c3);
+        mul_add_c(a[4], b[5], c1, c2, c3);
+        mul_add_c(a[5], b[4], c1, c2, c3);
+        mul_add_c(a[6], b[3], c1, c2, c3);
+        mul_add_c(a[7], b[2], c1, c2, c3);
+        r[9] = c1;
+        c1 = 0;
+        mul_add_c(a[7], b[3], c2, c3, c1);
+        mul_add_c(a[6], b[4], c2, c3, c1);
+        mul_add_c(a[5], b[5], c2, c3, c1);
+        mul_add_c(a[4], b[6], c2, c3, c1);
+        mul_add_c(a[3], b[7], c2, c3, c1);
+        r[10] = c2;
+        c2 = 0;
+        mul_add_c(a[4], b[7], c3, c1, c2);
+        mul_add_c(a[5], b[6], c3, c1, c2);
+        mul_add_c(a[6], b[5], c3, c1, c2);
+        mul_add_c(a[7], b[4], c3, c1, c2);
+        r[11] = c3;
+        c3 = 0;
+        mul_add_c(a[7], b[5], c1, c2, c3);
+        mul_add_c(a[6], b[6], c1, c2, c3);
+        mul_add_c(a[5], b[7], c1, c2, c3);
+        r[12] = c1;
+        c1 = 0;
+        mul_add_c(a[6], b[7], c2, c3, c1);
+        mul_add_c(a[7], b[6], c2, c3, c1);
+        r[13] = c2;
+        c2 = 0;
+        mul_add_c(a[7], b[7], c3, c1, c2);
+        r[14] = c3;
+        r[15] = c1;
+}
+#endif
+
 #if defined(OPENSSL_NO_ASM) || !defined(OPENSSL_BN_ASM_PART_WORDS)
 /*
  * Here follows a specialised variant of bn_sub_words(), which has the property