* Enable extra Arm64 optimization on Windows for GHASH, RAND and AES

Reviewed-by: Tom Cosgrove <tom.cosgrove@arm.com>
Reviewed-by: Tomas Mraz <tomas@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/21673)

diff --git a/CHANGES.md b/CHANGES.md
index af5f3cd2aba9b9b4a0632dd14c4ac007d8252903..ef801c2f7362b8c6a7b341dbf83f62e8c860f4a3 100644 (file)
--- a/CHANGES.md
+++ b/CHANGES.md
@@ -25,6 +25,10 @@ OpenSSL 3.2
 
 ### Changes between 3.1 and 3.2 [xx XXX xxxx]
 
+ * Enable extra Arm64 optimization on Windows for GHASH, RAND and AES.
+
+   *Evgeny Karpov*
+
  * Added a function to delete objects from store by URI - OSSL_STORE_delete()
    and the corresponding provider-storemgmt API function
    OSSL_FUNC_store_delete().

diff --git a/crypto/arm_arch.h b/crypto/arm_arch.h
index 43aa6b97c5eed7f6b22d3e98e0b7c64c9d276344..83acbe012654dfa458fab41dfed972d4bfd5d1a7 100644 (file)
--- a/crypto/arm_arch.h
+++ b/crypto/arm_arch.h
@@ -49,6 +49,8 @@
 #   else
 #    error "unsupported ARM architecture"
 #   endif
+#  elif defined(__ARM_ARCH)
+#   define __ARM_ARCH__ __ARM_ARCH
 #  endif
 # endif
 

diff --git a/crypto/modes/gcm128.c b/crypto/modes/gcm128.c
index 677eb21a0bd35ad8dc880b93dca445859f94c1de..77ff5dd06b45a44e8d7fb7664204d54e87ce83cb 100644 (file)
--- a/crypto/modes/gcm128.c
+++ b/crypto/modes/gcm128.c
@@ -369,7 +369,7 @@ void gcm_gmult_4bit_x86(u64 Xi[2], const u128 Htable[16]);
 void gcm_ghash_4bit_x86(u64 Xi[2], const u128 Htable[16], const u8 *inp,
                         size_t len);
 #  endif
-# elif defined(__arm__) || defined(__arm) || defined(__aarch64__)
+# elif defined(__arm__) || defined(__arm) || defined(__aarch64__) || defined(_M_ARM64)
 #  include "arm_arch.h"
 #  if __ARM_MAX_ARCH__>=7
 #   define GHASH_ASM_ARM

diff --git a/crypto/sm3/sm3_local.h b/crypto/sm3/sm3_local.h
index cb5a187a12fe1a8e46ebe9a6003f870588bac961..18e10890eb56ed9bbab33d7d1efe049aca474b8f 100644 (file)
--- a/crypto/sm3/sm3_local.h
+++ b/crypto/sm3/sm3_local.h
@@ -34,7 +34,7 @@
       } while (0)
 
 #if defined(OPENSSL_SM3_ASM)
-# if defined(__aarch64__)
+# if defined(__aarch64__) || defined(_M_ARM64)
 #  include "crypto/arm_arch.h"
 #  define HWSM3_CAPABLE (OPENSSL_armcap_P & ARMV8_SM3)
 void ossl_hwsm3_block_data_order(SM3_CTX *c, const void *p, size_t num);

diff --git a/include/crypto/aes_platform.h b/include/crypto/aes_platform.h
index cbc035926e036268397f1464a987afd3cb9fe031..e641450b15dfc86ace79b4033184a32981bb77c3 100644 (file)
--- a/include/crypto/aes_platform.h
+++ b/include/crypto/aes_platform.h
@@ -90,7 +90,7 @@ void gcm_ghash_p8(u64 Xi[2],const u128 Htable[16],const u8 *inp, size_t len);
 #   endif /* OPENSSL_SYS_AIX || OPENSSL_SYS_MACOSX */
 #  endif /* PPC */
 
-#  if (defined(__arm__) || defined(__arm) || defined(__aarch64__))
+#  if (defined(__arm__) || defined(__arm) || defined(__aarch64__) || defined(_M_ARM64)) 
 #   include "arm_arch.h"
 #   if __ARM_MAX_ARCH__>=7
 #    if defined(BSAES_ASM)
@@ -106,7 +106,7 @@ void gcm_ghash_p8(u64 Xi[2],const u128 Htable[16],const u8 *inp, size_t len);
 #    define HWAES_decrypt aes_v8_decrypt
 #    define HWAES_cbc_encrypt aes_v8_cbc_encrypt
 #    define HWAES_ecb_encrypt aes_v8_ecb_encrypt
-#    if __ARM_MAX_ARCH__>=8 && defined(__aarch64__)
+#    if __ARM_MAX_ARCH__>=8 && (defined(__aarch64__) || defined(_M_ARM64))
 #     define HWAES_xts_encrypt aes_v8_xts_encrypt
 #     define HWAES_xts_decrypt aes_v8_xts_decrypt
 #    endif
@@ -114,7 +114,7 @@ void gcm_ghash_p8(u64 Xi[2],const u128 Htable[16],const u8 *inp, size_t len);
 #    define AES_PMULL_CAPABLE ((OPENSSL_armcap_P & ARMV8_PMULL) && (OPENSSL_armcap_P & ARMV8_AES))
 #    define AES_GCM_ENC_BYTES 512
 #    define AES_GCM_DEC_BYTES 512
-#    if __ARM_MAX_ARCH__>=8 && defined(__aarch64__)
+#    if __ARM_MAX_ARCH__>=8 && (defined(__aarch64__) || defined(_M_ARM64))
 #     define AES_gcm_encrypt armv8_aes_gcm_encrypt
 #     define AES_gcm_decrypt armv8_aes_gcm_decrypt
 #     define AES_GCM_ASM(gctx) ((gctx)->ctr==aes_v8_ctr32_encrypt_blocks && \

diff --git a/include/crypto/sm4_platform.h b/include/crypto/sm4_platform.h
index 8a26885097b38a3f363ff19de1d3ebf4f1ae9155..cc4f514256c2207a42e545ee5e5b689a746694e6 100644 (file)
--- a/include/crypto/sm4_platform.h
+++ b/include/crypto/sm4_platform.h
@@ -12,7 +12,7 @@
 # pragma once
 
 # if defined(OPENSSL_CPUID_OBJ)
-#  if defined(__aarch64__)
+#  if defined(__aarch64__) ||  defined (_M_ARM64)
 #   include "arm_arch.h"
 extern unsigned int OPENSSL_arm_midr;
 static inline int vpsm4_capable(void)