From 039081b80977e2a5de84e1f88f8b4d025b559956 Mon Sep 17 00:00:00 2001
From: Andy Polyakov <appro@openssl.org>
Date: Sun, 13 Oct 2013 19:15:15 +0200
Subject: [PATCH] Initial aarch64 bits.

---
 crypto/bn/bn_lcl.h       |  9 +++++++++
 crypto/md32_common.h     | 18 ++++++++++++++++++
 crypto/modes/modes_lcl.h |  8 ++++++++
 crypto/sha/sha512.c      | 13 +++++++++++++
 4 files changed, 48 insertions(+)

diff --git a/crypto/bn/bn_lcl.h b/crypto/bn/bn_lcl.h
index 64df27aaf9..90787e576e 100644
--- a/crypto/bn/bn_lcl.h
+++ b/crypto/bn/bn_lcl.h
@@ -318,6 +318,15 @@ extern "C" {
 	     : "r"(a), "r"(b));
 #    endif
 #  endif
+# elif defined(__aarch64__) && defined(SIXTY_FOUR_BIT_LONG)
+#  if defined(__GNUC__) && __GNUC__>=2
+#   define BN_UMULT_HIGH(a,b)	({	\
+	register BN_ULONG ret;		\
+	asm ("umulh	%0,%1,%2"	\
+	     : "=r"(ret)		\
+	     : "r"(a), "r"(b));		\
+	ret;			})
+#  endif
 # endif		/* cpu */
 #endif		/* OPENSSL_NO_ASM */
 
diff --git a/crypto/md32_common.h b/crypto/md32_common.h
index d7f7a26e70..147a7a00c3 100644
--- a/crypto/md32_common.h
+++ b/crypto/md32_common.h
@@ -215,6 +215,24 @@
 				   asm ("bswapl %0":"=r"(r):"0"(r));	\
 				   *((unsigned int *)(c))=r; (c)+=4; r;	})
 #   endif
+#  elif defined(__aarch64__)
+#   if defined(__BYTE_ORDER__)
+#    if defined(__ORDER_LITTLE_ENDIAN__) && __BYTE_ORDER__==__ORDER_LITTLE_ENDIAN__
+#     define HOST_c2l(c,l)	({ unsigned int r;		\
+				   asm ("rev	%w0,%w1"	\
+					:"=r"(r)		\
+					:"r"(*((const unsigned int *)(c))));\
+				   (c)+=4; (l)=r;		})
+#     define HOST_l2c(l,c)	({ unsigned int r;		\
+				   asm ("rev	%w0,%w1"	\
+					:"=r"(r)		\
+					:"r"((unsigned int)(l)));\
+				   *((unsigned int *)(c))=r; (c)+=4; r;	})
+#    elif defined(__ORDER_BIG_ENDIAN__) && __BYTE_ORDER__==__ORDER_BIG_ENDIAN__
+#     define HOST_c2l(c,l)	((l)=*((const unsigned int *)(c)), (c)+=4, (l))
+#     define HOST_l2c(l,c)	(*((unsigned int *)(c))=(l), (c)+=4, (l))
+#    endif
+#   endif
 #  endif
 # endif
 #endif
diff --git a/crypto/modes/modes_lcl.h b/crypto/modes/modes_lcl.h
index 9d83e12844..35d4d9fa53 100644
--- a/crypto/modes/modes_lcl.h
+++ b/crypto/modes/modes_lcl.h
@@ -29,6 +29,7 @@ typedef unsigned char u8;
 #if defined(__i386)	|| defined(__i386__)	|| \
     defined(__x86_64)	|| defined(__x86_64__)	|| \
     defined(_M_IX86)	|| defined(_M_AMD64)	|| defined(_M_X64) || \
+    defined(__aarch64__)			|| \
     defined(__s390__)	|| defined(__s390x__)
 # undef STRICT_ALIGNMENT
 #endif
@@ -50,6 +51,13 @@ typedef unsigned char u8;
 #  define BSWAP4(x) ({	u32 ret=(x);			\
 			asm ("bswapl %0"		\
 			: "+r"(ret));	ret;		})
+# elif defined(__aarch64__)
+#  define BSWAP8(x) ({	u64 ret;			\
+			asm ("rev %0,%1"		\
+			: "=r"(ret) : "r"(x)); ret;	})
+#  define BSWAP4(x) ({	u32 ret;			\
+			asm ("rev %w0,%w1"		\
+			: "=r"(ret) : "r"(x)); ret;	})
 # elif (defined(__arm__) || defined(__arm)) && !defined(STRICT_ALIGNMENT)
 #  define BSWAP8(x) ({	u32 lo=(u64)(x)>>32,hi=(x);	\
 			asm ("rev %0,%0; rev %1,%1"	\
diff --git a/crypto/sha/sha512.c b/crypto/sha/sha512.c
index aab1f1e8d4..59b4ace234 100644
--- a/crypto/sha/sha512.c
+++ b/crypto/sha/sha512.c
@@ -56,6 +56,7 @@ const char SHA512_version[]="SHA-512" OPENSSL_VERSION_PTEXT;
 #if defined(__i386) || defined(__i386__) || defined(_M_IX86) || \
     defined(__x86_64) || defined(_M_AMD64) || defined(_M_X64) || \
     defined(__s390__) || defined(__s390x__) || \
+    defined(__aarch64__) || \
     defined(SHA512_ASM)
 #define SHA512_BLOCK_CAN_MANAGE_UNALIGNED_DATA
 #endif
@@ -342,6 +343,18 @@ static const SHA_LONG64 K512[80] = {
 				asm ("rotrdi %0,%1,%2"	\
 				: "=r"(ret)		\
 				: "r"(a),"K"(n)); ret;	})
+#  elif defined(__aarch64__)
+#   define ROTR(a,n)	({ SHA_LONG64 ret;		\
+				asm ("ror %0,%1,%2"	\
+				: "=r"(ret)		\
+				: "r"(a),"I"(n)); ret;	})
+#   if  defined(__BYTE_ORDER__) && defined(__ORDER_LITTLE_ENDIAN__) && \
+	__BYTE_ORDER__==__ORDER_LITTLE_ENDIAN__
+#    define PULL64(x)	({ SHA_LONG64 ret;			\
+				asm ("rev	%0,%1"		\
+				: "=r"(ret)			\
+				: "r"(*((const SHA_LONG64 *)(&(x))))); ret;		})
+#   endif
 #  endif
 # elif defined(_MSC_VER)
 #  if defined(_WIN64)	/* applies to both IA-64 and AMD64 */