1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
|
From 98f9a401c3964c7ff0e6ca048685e28a2a6401d4 Mon Sep 17 00:00:00 2001
From: Hubert Kario <hkario@redhat.com>
Date: Wed, 8 Feb 2023 14:13:24 +0100
Subject: [PATCH] rsa: add msvc intrinsic for non x64 platforms
_umul128() is x86_64 (x64) only, while __umulh() works everywhere, but
doesn't generate optimal code on x64
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Paul Dale <pauli@openssl.org>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/20244)
(cherry picked from commit 075652f224479dad2e64b92e791b296177af8705)
---
crypto/bn/rsa_sup_mul.c | 24 +++++++++++++++++++++++-
1 file changed, 23 insertions(+), 1 deletion(-)
diff --git a/crypto/bn/rsa_sup_mul.c b/crypto/bn/rsa_sup_mul.c
index 0e0d02e1946e..3b57161b4589 100644
--- a/crypto/bn/rsa_sup_mul.c
+++ b/crypto/bn/rsa_sup_mul.c
@@ -110,12 +110,34 @@ static ossl_inline void _mul_limb(limb_t *hi, limb_t *lo, limb_t a, limb_t b)
*lo = (limb_t)t;
}
#elif (BN_BYTES == 8) && (defined _MSC_VER)
-/* https://learn.microsoft.com/en-us/cpp/intrinsics/umul128?view=msvc-170 */
+# if defined(_M_X64)
+/*
+ * on x86_64 (x64) we can use the _umul128 intrinsic to get one `mul`
+ * instruction to get both high and low 64 bits of the multiplication.
+ * https://learn.microsoft.com/en-us/cpp/intrinsics/umul128?view=msvc-140
+ */
+#include <intrin.h>
#pragma intrinsic(_umul128)
static ossl_inline void _mul_limb(limb_t *hi, limb_t *lo, limb_t a, limb_t b)
{
*lo = _umul128(a, b, hi);
}
+# elif defined(_M_ARM64) || defined (_M_IA64)
+/*
+ * We can't use the __umulh() on x86_64 as then msvc generates two `mul`
+ * instructions; so use this more portable intrinsic on platforms that
+ * don't support _umul128 (like aarch64 (ARM64) or ia64)
+ * https://learn.microsoft.com/en-us/cpp/intrinsics/umulh?view=msvc-140
+ */
+#include <intrin.h>
+static ossl_inline void _mul_limb(limb_t *hi, limb_t *lo, limb_t a, limb_t b)
+{
+ *lo = a * b;
+ *hi = __umulh(a, b);
+}
+# else
+# error Only x64, ARM64 and IA64 supported.
+# endif /* defined(_M_X64) */
#else
/*
* if the compiler doesn't have either a 128bit data type nor a "return
|
