1/*
2 * Copyright 2005-2021 The OpenSSL Project Authors. All Rights Reserved.
3 *
4 * Licensed under the Apache License 2.0 (the "License").  You may not use
5 * this file except in compliance with the License.  You can obtain a copy
6 * in the file LICENSE in the source distribution or at
7 * https://www.openssl.org/source/license.html
8 */
9
10#include <stdlib.h>
11#include <openssl/bn.h>
12#include "internal/cryptlib.h"
13#include "crypto/sparc_arch.h"
14#include "bn_local.h"    /* for definition of bn_mul_mont */
15
16int bn_mul_mont(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
17                const BN_ULONG *np, const BN_ULONG *n0, int num)
18{
19    int bn_mul_mont_vis3(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
20                         const BN_ULONG *np, const BN_ULONG *n0, int num);
21    int bn_mul_mont_fpu(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
22                        const BN_ULONG *np, const BN_ULONG *n0, int num);
23    int bn_mul_mont_int(BN_ULONG *rp, const BN_ULONG *ap, const BN_ULONG *bp,
24                        const BN_ULONG *np, const BN_ULONG *n0, int num);
25
26    if (!(num & 1) && num >= 6) {
27        if ((num & 15) == 0 && num <= 64 &&
28            (OPENSSL_sparcv9cap_P[1] & (CFR_MONTMUL | CFR_MONTSQR)) ==
29            (CFR_MONTMUL | CFR_MONTSQR)) {
30            typedef int (*bn_mul_mont_f) (BN_ULONG *rp, const BN_ULONG *ap,
31                                          const BN_ULONG *bp,
32                                          const BN_ULONG *np,
33                                          const BN_ULONG *n0);
34            int bn_mul_mont_t4_8(BN_ULONG *rp, const BN_ULONG *ap,
35                                 const BN_ULONG *bp, const BN_ULONG *np,
36                                 const BN_ULONG *n0);
37            int bn_mul_mont_t4_16(BN_ULONG *rp, const BN_ULONG *ap,
38                                  const BN_ULONG *bp, const BN_ULONG *np,
39                                  const BN_ULONG *n0);
40            int bn_mul_mont_t4_24(BN_ULONG *rp, const BN_ULONG *ap,
41                                  const BN_ULONG *bp, const BN_ULONG *np,
42                                  const BN_ULONG *n0);
43            int bn_mul_mont_t4_32(BN_ULONG *rp, const BN_ULONG *ap,
44                                  const BN_ULONG *bp, const BN_ULONG *np,
45                                  const BN_ULONG *n0);
46            static const bn_mul_mont_f funcs[4] = {
47                bn_mul_mont_t4_8, bn_mul_mont_t4_16,
48                bn_mul_mont_t4_24, bn_mul_mont_t4_32
49            };
50            bn_mul_mont_f worker = funcs[num / 16 - 1];
51
52            if ((*worker) (rp, ap, bp, np, n0))
53                return 1;
54            /* retry once and fall back */
55            if ((*worker) (rp, ap, bp, np, n0))
56                return 1;
57            return bn_mul_mont_vis3(rp, ap, bp, np, n0, num);
58        }
59        if ((OPENSSL_sparcv9cap_P[0] & SPARCV9_VIS3))
60            return bn_mul_mont_vis3(rp, ap, bp, np, n0, num);
61        else if (num >= 8 &&
62                 /*
63                  * bn_mul_mont_fpu doesn't use FMADD, we just use the
64                  * flag to detect when FPU path is preferable in cases
65                  * when current heuristics is unreliable. [it works
66                  * out because FMADD-capable processors where FPU
67                  * code path is undesirable are also VIS3-capable and
68                  * VIS3 code path takes precedence.]
69                  */
70                 ( (OPENSSL_sparcv9cap_P[0] & SPARCV9_FMADD) ||
71                   (OPENSSL_sparcv9cap_P[0] &
72                    (SPARCV9_PREFER_FPU | SPARCV9_VIS1)) ==
73                   (SPARCV9_PREFER_FPU | SPARCV9_VIS1) ))
74            return bn_mul_mont_fpu(rp, ap, bp, np, n0, num);
75    }
76    return bn_mul_mont_int(rp, ap, bp, np, n0, num);
77}
78