summaryrefslogtreecommitdiffstats
path: root/crypto
diff options
context:
space:
mode:
authorHubert Kario <hkario@redhat.com>2023-02-08 14:13:24 +0100
committerTomas Mraz <tomas@openssl.org>2023-02-11 14:56:27 +0100
commitc062097bf3e305c9c3b90c0941bb58b8cd7baf88 (patch)
tree4eb4f23053b999779482f488147f6fd818040d54 /crypto
parent9b4015eccca93d401f9a7ee2d88244872983dcc6 (diff)
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)
Diffstat (limited to 'crypto')
-rw-r--r--crypto/bn/rsa_sup_mul.c24
1 files changed, 23 insertions, 1 deletions
diff --git a/crypto/bn/rsa_sup_mul.c b/crypto/bn/rsa_sup_mul.c
index 0e0d02e194..3b57161b45 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