Enable optimized crypto operations for x86_64

This adds initial support for assembly crypto acceleration on x86_64 for
ELF (Linux, *BSD, Solaris) and Mach-O (OS-X) systems.

The build method is a little different than OpenSSL and OpenBSD.  All
the .s files are generated ahead of time when the tarball is generated,
so there are no complicated makefile rules at configure/build time. This
also means the builds are faster and perl is not required on the build
system.

Thanks to Wouter Clarie for providing the initial cleanup and patch
that this is based on.

3 files changed, 102 insertions, 8 deletions

diff --git a/crypto/Makefile.am b/crypto/Makefile.am
index 39b143d..91c58db 100644
--- a/crypto/Makefile.am
+++ b/crypto/Makefile.am
@@ -10,7 +10,11 @@ EXTRA_DIST = VERSION
 
 libcrypto_la_LDFLAGS = -version-info @LIBCRYPTO_VERSION@
 libcrypto_la_LIBADD = libcompat.la libcompatnoopt.la
-libcrypto_la_CFLAGS = $(CFLAGS) $(USER_CFLAGS) -DOPENSSL_NO_HW_PADLOCK
+libcrypto_la_CFLAGS = $(CFLAGS) $(USER_CFLAGS)
+libcrypto_la_CFLAGS += -DOPENSSL_NO_HW_PADLOCK
+if OPENSSL_NO_ASM
+libcrypto_la_CFLAGS += -DOPENSSL_NO_ASM
+endif
 
 noinst_LTLIBRARIES = libcompat.la libcompatnoopt.la
 
@@ -103,6 +107,21 @@ noinst_HEADERS += compat/chacha_private.h
 libcrypto_la_SOURCES =
 EXTRA_libcrypto_la_SOURCES =
 
+include Makefile.am.elf-x86_64
+include Makefile.am.macosx-x86_64
+
+if !HOST_ASM_ELF_X86_64
+if !HOST_ASM_MACOSX_X86_64
+libcrypto_la_SOURCES += aes/aes_cbc.c
+libcrypto_la_SOURCES += aes/aes_core.c
+libcrypto_la_SOURCES += camellia/camellia.c
+libcrypto_la_SOURCES += camellia/cmll_cbc.c
+libcrypto_la_SOURCES += rc4/rc4_enc.c
+libcrypto_la_SOURCES += rc4/rc4_skey.c
+libcrypto_la_SOURCES += whrlpool/wp_block.c
+endif
+endif
+
 libcrypto_la_SOURCES += cpt_err.c
 libcrypto_la_SOURCES += cryptlib.c
 libcrypto_la_SOURCES += cversion.c
@@ -118,9 +137,7 @@ noinst_HEADERS += md32_common.h
 noinst_HEADERS += o_time.h
 
 # aes
-libcrypto_la_SOURCES += aes/aes_cbc.c
 libcrypto_la_SOURCES += aes/aes_cfb.c
-libcrypto_la_SOURCES += aes/aes_core.c
 libcrypto_la_SOURCES += aes/aes_ctr.c
 libcrypto_la_SOURCES += aes/aes_ecb.c
 libcrypto_la_SOURCES += aes/aes_ige.c
@@ -284,8 +301,6 @@ libcrypto_la_SOURCES += buffer/buf_str.c
 libcrypto_la_SOURCES += buffer/buffer.c
 
 # camellia
-libcrypto_la_SOURCES += camellia/camellia.c
-libcrypto_la_SOURCES += camellia/cmll_cbc.c
 libcrypto_la_SOURCES += camellia/cmll_cfb.c
 libcrypto_la_SOURCES += camellia/cmll_ctr.c
 libcrypto_la_SOURCES += camellia/cmll_ecb.c
@@ -666,8 +681,6 @@ libcrypto_la_SOURCES += rc2/rc2ofb64.c
 noinst_HEADERS += rc2/rc2_locl.h
 
 # rc4
-libcrypto_la_SOURCES += rc4/rc4_enc.c
-libcrypto_la_SOURCES += rc4/rc4_skey.c
 noinst_HEADERS += rc4/rc4_locl.h
 
 # ripemd
@@ -739,7 +752,6 @@ libcrypto_la_SOURCES += ui/ui_util.c
 noinst_HEADERS += ui/ui_locl.h
 
 # whrlpool
-libcrypto_la_SOURCES += whrlpool/wp_block.c
 libcrypto_la_SOURCES += whrlpool/wp_dgst.c
 noinst_HEADERS += whrlpool/wp_locl.h
 
diff --git a/crypto/Makefile.am.elf-x86_64 b/crypto/Makefile.am.elf-x86_64
new file mode 100644
index 0000000..6257c40
--- /dev/null
+++ b/crypto/Makefile.am.elf-x86_64
@@ -0,0 +1,41 @@
+
+ASM_X86_64_ELF = aes/aes-elf-x86_64.s
+ASM_X86_64_ELF += aes/bsaes-elf-x86_64.s
+ASM_X86_64_ELF += aes/vpaes-elf-x86_64.s
+ASM_X86_64_ELF += aes/aesni-elf-x86_64.s
+ASM_X86_64_ELF += aes/aesni-sha1-elf-x86_64.s
+ASM_X86_64_ELF += bn/modexp512-elf-x86_64.s
+ASM_X86_64_ELF += bn/mont-elf-x86_64.s
+ASM_X86_64_ELF += bn/mont5-elf-x86_64.s
+ASM_X86_64_ELF += bn/gf2m-elf-x86_64.s
+ASM_X86_64_ELF += camellia/cmll-elf-x86_64.s
+ASM_X86_64_ELF += md5/md5-elf-x86_64.s
+ASM_X86_64_ELF += modes/ghash-elf-x86_64.s
+ASM_X86_64_ELF += rc4/rc4-elf-x86_64.s
+ASM_X86_64_ELF += rc4/rc4-md5-elf-x86_64.s
+ASM_X86_64_ELF += sha/sha1-elf-x86_64.s
+ASM_X86_64_ELF += sha/sha256-elf-x86_64.S
+ASM_X86_64_ELF += sha/sha512-elf-x86_64.S
+ASM_X86_64_ELF += whrlpool/wp-elf-x86_64.s
+ASM_X86_64_ELF += cpuid-elf-x86_64.S
+
+EXTRA_DIST += $(ASM_X86_64_ELF)
+
+if HOST_ASM_ELF_X86_64
+libcrypto_la_CFLAGS += -DAES_ASM
+libcrypto_la_CFLAGS += -DBSAES_ASM
+libcrypto_la_CFLAGS += -DVPAES_ASM
+libcrypto_la_CFLAGS += -DOPENSSL_IA32_SSE2
+libcrypto_la_CFLAGS += -DOPENSSL_BN_ASM_MONT
+libcrypto_la_CFLAGS += -DOPENSSL_BN_ASM_MONT5
+libcrypto_la_CFLAGS += -DOPENSSL_BN_ASM_GF2m
+libcrypto_la_CFLAGS += -DMD5_ASM
+libcrypto_la_CFLAGS += -DGHASH_ASM
+libcrypto_la_CFLAGS += -DRSA_ASM
+libcrypto_la_CFLAGS += -DSHA1_ASM
+libcrypto_la_CFLAGS += -DSHA256_ASM
+libcrypto_la_CFLAGS += -DSHA512_ASM
+libcrypto_la_CFLAGS += -DWHIRLPOOL_ASM
+libcrypto_la_CFLAGS += -DOPENSSL_CPUID_OBJ
+libcrypto_la_SOURCES += $(ASM_X86_64_ELF)
+endif
diff --git a/crypto/Makefile.am.macosx-x86_64 b/crypto/Makefile.am.macosx-x86_64
new file mode 100644
index 0000000..e361aae
--- /dev/null
+++ b/crypto/Makefile.am.macosx-x86_64
@@ -0,0 +1,41 @@
+
+ASM_X86_64_MACOSX = aes/aes-macosx-x86_64.s
+ASM_X86_64_MACOSX += aes/bsaes-macosx-x86_64.s
+ASM_X86_64_MACOSX += aes/vpaes-macosx-x86_64.s
+ASM_X86_64_MACOSX += aes/aesni-macosx-x86_64.s
+ASM_X86_64_MACOSX += aes/aesni-sha1-macosx-x86_64.s
+ASM_X86_64_MACOSX += bn/modexp512-macosx-x86_64.s
+ASM_X86_64_MACOSX += bn/mont-macosx-x86_64.s
+ASM_X86_64_MACOSX += bn/mont5-macosx-x86_64.s
+ASM_X86_64_MACOSX += bn/gf2m-macosx-x86_64.s
+ASM_X86_64_MACOSX += camellia/cmll-macosx-x86_64.s
+ASM_X86_64_MACOSX += md5/md5-macosx-x86_64.s
+ASM_X86_64_MACOSX += modes/ghash-macosx-x86_64.s
+ASM_X86_64_MACOSX += rc4/rc4-macosx-x86_64.s
+ASM_X86_64_MACOSX += rc4/rc4-md5-macosx-x86_64.s
+ASM_X86_64_MACOSX += sha/sha1-macosx-x86_64.s
+ASM_X86_64_MACOSX += sha/sha256-macosx-x86_64.S
+ASM_X86_64_MACOSX += sha/sha512-macosx-x86_64.S
+ASM_X86_64_MACOSX += whrlpool/wp-macosx-x86_64.s
+ASM_X86_64_MACOSX += cpuid-macosx-x86_64.S
+
+EXTRA_DIST += $(ASM_X86_64_MACOSX)
+
+if HOST_ASM_MACOSX_X86_64
+libcrypto_la_CFLAGS += -DAES_ASM
+libcrypto_la_CFLAGS += -DBSAES_ASM
+libcrypto_la_CFLAGS += -DVPAES_ASM
+libcrypto_la_CFLAGS += -DOPENSSL_IA32_SSE2
+libcrypto_la_CFLAGS += -DOPENSSL_BN_ASM_MONT
+libcrypto_la_CFLAGS += -DOPENSSL_BN_ASM_MONT5
+libcrypto_la_CFLAGS += -DOPENSSL_BN_ASM_GF2m
+libcrypto_la_CFLAGS += -DMD5_ASM
+libcrypto_la_CFLAGS += -DGHASH_ASM
+libcrypto_la_CFLAGS += -DRSA_ASM
+libcrypto_la_CFLAGS += -DSHA1_ASM
+libcrypto_la_CFLAGS += -DSHA256_ASM
+libcrypto_la_CFLAGS += -DSHA512_ASM
+libcrypto_la_CFLAGS += -DWHIRLPOOL_ASM
+libcrypto_la_CFLAGS += -DOPENSSL_CPUID_OBJ
+libcrypto_la_SOURCES += $(ASM_X86_64_MACOSX)
+endif