Improved PowerPC support. Proper ./config support for ppc targets,

especially for AIX. But most important BIGNUM assembler implementation
submitted by IBM.

Submitted by: Peter Waltenberg <pwalten@au1.ibm.com>
Reviewed by: appro

5 files changed, 2178 insertions, 66 deletions

diff --git a/Configure b/Configure
index 84b03a6626..72b76bbf38 100755
--- a/Configure
+++ b/Configure
@@ -390,7 +390,8 @@ my %table=(
 "linux-aout",	"gcc:-DL_ENDIAN -DTERMIO -O3 -fomit-frame-pointer -m486 -Wall::(unknown):::BN_LLONG ${x86_gcc_des} ${x86_gcc_opts}:${x86_out_asm}",
 "linux-mipsel",   "gcc:-DL_ENDIAN -DTERMIO -O3 -fomit-frame-pointer -Wall::-D_REENTRANT::-ldl:BN_LLONG RC2_CHAR RC4_INDEX DES_INT DES_UNROLL DES_RISC2::::::::::dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
 "linux-mips",   "gcc:-DB_ENDIAN -DTERMIO -O3 -fomit-frame-pointer -Wall::-D_REENTRANT::-ldl:BN_LLONG RC2_CHAR RC4_INDEX DES_INT DES_UNROLL DES_RISC2::::::::::dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
-"linux-ppc",    "gcc:-DB_ENDIAN -DTERMIO -O3 -fomit-frame-pointer -Wall::-D_REENTRANT::-ldl:BN_LLONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL::::::::::dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
+"linux-ppc",    "gcc:-DB_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:BN_LLONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL:asm/linux_ppc32.o:::::::::dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
+"linux-ppc64",  "gcc:-m64 -DB_ENDIAN -DTERMIO -O3 -Wall::-D_REENTRANT::-ldl:SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL:asm/linux_ppc64.o:::::::::dlfcn:linux-shared:-fPIC:-m64:.so..\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
 "linux-m68k",   "gcc:-DB_ENDIAN -DTERMIO -O2 -fomit-frame-pointer -Wall::-D_REENTRANT:::BN_LLONG::",
 "linux-s390",	"gcc:-DB_ENDIAN -DTERMIO -DNO_ASM -O3 -fomit-frame-pointer -Wall::-D_REENTRANT::-ldl:BN_LLONG::::::::::dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
 "linux-s390x",	"gcc:-DB_ENDIAN -DTERMIO -DNO_ASM -O3 -fomit-frame-pointer -Wall::-D_REENTRANT::-ldl:SIXTY_FOUR_BIT_LONG::::::::::dlfcn:linux-shared:-fPIC::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)",
@@ -446,11 +447,12 @@ my %table=(
 
 
 # IBM's AIX.
-"aix-cc",   "cc:-O -DB_ENDIAN -qmaxmem=16384::(unknown):AIX::BN_LLONG RC4_CHAR:::",
-"aix-gcc",  "gcc:-O3 -DB_ENDIAN::(unknown):AIX::BN_LLONG RC4_CHAR:::",
-"aix43-cc",   "cc:-O -DAIX -DB_ENDIAN -qmaxmem=16384::(unknown):::BN_LLONG RC4_CHAR::::::::::dlfcn:aix-shared:::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)::",
-"aix43-gcc",  "gcc:-O1 -DAIX -DB_ENDIAN::(unknown):::BN_LLONG RC4_CHAR::::::::::dlfcn:",
-"aix64-cc",   "cc:-O -DAIX -DB_ENDIAN -qmaxmem=16384 -q64::(unknown):::SIXTY_FOUR_BIT_LONG RC4_CHAR::::::::::dlfcn:aix-shared::-q64:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)::-X 64",
+"aix-gcc",  "gcc:-O -DB_ENDIAN::(unknown):AIX::BN_LLONG RC4_CHAR:asm/aix_ppc32.o:::::::::dlfcn:",
+"aix43-cc", "cc:-O -DB_ENDIAN -qmaxmem=16384::(unknown):::BN_LLONG RC4_CHAR:asm/aix_ppc32.o:::::::::dlfcn:aix-shared:::.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)::",
+# Below targets assume AIX 5. Idea is to effectively disregard $OBJECT_MODE
+# at build time. $OBJECT_MODE is respected at ./config stage!
+"aix-cc",   "cc:-q32 -O -DB_ENDIAN -qmaxmem=16384::(unknown):AIX::BN_LLONG RC4_CHAR:asm/aix_ppc32.o:::::::::dlfcn:aix-shared::-q32:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)::-X 32",
+"aix64-cc", "cc:-q64 -O -DB_ENDIAN -qmaxmem=16384::(unknown):AIX::SIXTY_FOUR_BIT_LONG RC4_CHAR:asm/aix_ppc64.o:::::::::dlfcn:aix-shared::-q64:.so.\$(SHLIB_MAJOR).\$(SHLIB_MINOR)::-X 64",
 
 #
 # Cray T90 and similar (SDSC)
@@ -553,7 +555,7 @@ my %table=(
 
 ##### MacOS X (a.k.a. Rhapsody or Darwin) setup
 "rhapsody-ppc-cc","cc:-O3 -DB_ENDIAN::(unknown):MACOSX_RHAPSODY::BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR:::",
-"darwin-ppc-cc","cc:-O3 -fomit-frame-pointer -fno-common -DB_ENDIAN::-D_REENTRANT:MACOSX::BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR:::::::::::darwin-shared:-fPIC::.\$(SHLIB_MAJOR).\$(SHLIB_MINOR).dylib",
+"darwin-ppc-cc","cc:-O3 -fno-common -DB_ENDIAN::-D_REENTRANT:MACOSX::BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR:asm/osx_ppc32.o::::::::::darwin-shared:-fPIC::.\$(SHLIB_MAJOR).\$(SHLIB_MINOR).dylib",
 "darwin-i386-cc","cc:-O3 -fomit-frame-pointer -fno-common -DB_ENDIAN::-D_REENTRANT:MACOSX::BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR:::::::::::darwin-shared:-fPIC::.\$(SHLIB_MAJOR).\$(SHLIB_MINOR).dylib",
 
 ##### A/UX
diff --git a/TABLE b/TABLE
index 1500669dec..e9895cb181 100644
--- a/TABLE
+++ b/TABLE
@@ -1027,13 +1027,13 @@ $arflags      =
 
 *** aix-cc
 $cc           = cc
-$cflags       = -O -DB_ENDIAN -qmaxmem=16384
+$cflags       = -q32 -O -DB_ENDIAN -qmaxmem=16384
 $unistd       = 
 $thread_cflag = (unknown)
 $sys_id       = AIX
 $lflags       = 
 $bn_ops       = BN_LLONG RC4_CHAR
-$bn_obj       = 
+$bn_obj       = asm/aix_ppc32.o
 $des_obj      = 
 $bf_obj       = 
 $md5_obj      = 
@@ -1042,23 +1042,23 @@ $cast_obj     =
 $rc4_obj      = 
 $rmd160_obj   = 
 $rc5_obj      = 
-$dso_scheme   = 
-$shared_target= 
+$dso_scheme   = dlfcn
+$shared_target= aix-shared
 $shared_cflag = 
-$shared_ldflag = 
-$shared_extension = 
+$shared_ldflag = -q32
+$shared_extension = .so.$(SHLIB_MAJOR).$(SHLIB_MINOR)
 $ranlib       = 
-$arflags      = 
+$arflags      = -X 32
 
 *** aix-gcc
 $cc           = gcc
-$cflags       = -O3 -DB_ENDIAN
+$cflags       = -O -DB_ENDIAN
 $unistd       = 
 $thread_cflag = (unknown)
 $sys_id       = AIX
 $lflags       = 
 $bn_ops       = BN_LLONG RC4_CHAR
-$bn_obj       = 
+$bn_obj       = asm/aix_ppc32.o
 $des_obj      = 
 $bf_obj       = 
 $md5_obj      = 
@@ -1067,7 +1067,7 @@ $cast_obj     =
 $rc4_obj      = 
 $rmd160_obj   = 
 $rc5_obj      = 
-$dso_scheme   = 
+$dso_scheme   = dlfcn
 $shared_target= 
 $shared_cflag = 
 $shared_ldflag = 
@@ -1077,13 +1077,13 @@ $arflags      =
 
 *** aix43-cc
 $cc           = cc
-$cflags       = -O -DAIX -DB_ENDIAN -qmaxmem=16384
+$cflags       = -O -DB_ENDIAN -qmaxmem=16384
 $unistd       = 
 $thread_cflag = (unknown)
 $sys_id       = 
 $lflags       = 
 $bn_ops       = BN_LLONG RC4_CHAR
-$bn_obj       = 
+$bn_obj       = asm/aix_ppc32.o
 $des_obj      = 
 $bf_obj       = 
 $md5_obj      = 
@@ -1100,40 +1100,15 @@ $shared_extension = .so.$(SHLIB_MAJOR).$(SHLIB_MINOR)
 $ranlib       = 
 $arflags      = 
 
-*** aix43-gcc
-$cc           = gcc
-$cflags       = -O1 -DAIX -DB_ENDIAN
-$unistd       = 
-$thread_cflag = (unknown)
-$sys_id       = 
-$lflags       = 
-$bn_ops       = BN_LLONG RC4_CHAR
-$bn_obj       = 
-$des_obj      = 
-$bf_obj       = 
-$md5_obj      = 
-$sha1_obj     = 
-$cast_obj     = 
-$rc4_obj      = 
-$rmd160_obj   = 
-$rc5_obj      = 
-$dso_scheme   = dlfcn
-$shared_target= 
-$shared_cflag = 
-$shared_ldflag = 
-$shared_extension = 
-$ranlib       = 
-$arflags      = 
-
 *** aix64-cc
 $cc           = cc
-$cflags       = -O -DAIX -DB_ENDIAN -qmaxmem=16384 -q64
+$cflags       = -q64 -O -DB_ENDIAN -qmaxmem=16384
 $unistd       = 
 $thread_cflag = (unknown)
-$sys_id       = 
+$sys_id       = AIX
 $lflags       = 
 $bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHAR
-$bn_obj       = 
+$bn_obj       = asm/aix_ppc64.o
 $des_obj      = 
 $bf_obj       = 
 $md5_obj      = 
@@ -1452,13 +1427,13 @@ $arflags      =
 
 *** darwin-ppc-cc
 $cc           = cc
-$cflags       = -O3 -fomit-frame-pointer -fno-common -DB_ENDIAN
+$cflags       = -O3 -fno-common -DB_ENDIAN
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $sys_id       = MACOSX
 $lflags       = 
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_UNROLL BF_PTR
-$bn_obj       = 
+$bn_obj       = asm/osx_ppc32.o
 $des_obj      = 
 $bf_obj       = 
 $md5_obj      = 
@@ -1702,7 +1677,7 @@ $arflags      =
 
 *** debug-levitte-linux-elf
 $cc           = gcc
-$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_CTX_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
+$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $sys_id       = 
@@ -1727,7 +1702,7 @@ $arflags      =
 
 *** debug-levitte-linux-elf-extreme
 $cc           = gcc
-$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_CTX_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -W -Wundef -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
+$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -W -Wundef -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $sys_id       = 
@@ -1752,7 +1727,7 @@ $arflags      =
 
 *** debug-levitte-linux-noasm
 $cc           = gcc
-$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_CTX_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DOPENSSL_NO_ASM -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
+$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DOPENSSL_NO_ASM -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $sys_id       = 
@@ -1777,7 +1752,7 @@ $arflags      =
 
 *** debug-levitte-linux-noasm-extreme
 $cc           = gcc
-$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_CTX_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DOPENSSL_NO_ASM -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -W -Wundef -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
+$cflags       = -DLEVITTE_DEBUG -DREF_CHECK -DCONF_DEBUG -DBN_DEBUG -DBN_DEBUG_RAND -DCRYPTO_MDEBUG -DENGINE_CONF_DEBUG -DOPENSSL_NO_ASM -DL_ENDIAN -DTERMIO -D_POSIX_SOURCE -DPEDANTIC -ggdb -g3 -mcpu=i486 -pedantic -ansi -Wall -W -Wundef -Wshadow -Wcast-align -Wstrict-prototypes -Wmissing-prototypes -Wno-long-long -Wundef -Wconversion -pipe
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $sys_id       = 
@@ -3127,7 +3102,7 @@ $arflags      =
 
 *** linux-ia32-icc
 $cc           = icc
-$cflags       = -DL_ENDIAN -DTERMIO -O2
+$cflags       = -DL_ENDIAN -DTERMIO -O2 -no_cpprt
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $sys_id       = 
@@ -3352,13 +3327,13 @@ $arflags      =
 
 *** linux-ppc
 $cc           = gcc
-$cflags       = -DB_ENDIAN -DTERMIO -O3 -fomit-frame-pointer -Wall
+$cflags       = -DB_ENDIAN -DTERMIO -O3 -Wall
 $unistd       = 
 $thread_cflag = -D_REENTRANT
 $sys_id       = 
 $lflags       = -ldl
 $bn_ops       = BN_LLONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL
-$bn_obj       = 
+$bn_obj       = asm/linux_ppc32.o
 $des_obj      = 
 $bf_obj       = 
 $md5_obj      = 
@@ -3375,6 +3350,31 @@ $shared_extension = .so.$(SHLIB_MAJOR).$(SHLIB_MINOR)
 $ranlib       = 
 $arflags      = 
 
+*** linux-ppc64
+$cc           = gcc
+$cflags       = -m64 -DB_ENDIAN -DTERMIO -O3 -Wall
+$unistd       = 
+$thread_cflag = -D_REENTRANT
+$sys_id       = 
+$lflags       = -ldl
+$bn_ops       = SIXTY_FOUR_BIT_LONG RC4_CHAR RC4_CHUNK DES_RISC1 DES_UNROLL
+$bn_obj       = asm/linux_ppc64.o
+$des_obj      = 
+$bf_obj       = 
+$md5_obj      = 
+$sha1_obj     = 
+$cast_obj     = 
+$rc4_obj      = 
+$rmd160_obj   = 
+$rc5_obj      = 
+$dso_scheme   = dlfcn
+$shared_target= linux-shared
+$shared_cflag = -fPIC
+$shared_ldflag = -m64
+$shared_extension = .so..$(SHLIB_MAJOR).$(SHLIB_MINOR)
+$ranlib       = 
+$arflags      = 
+
 *** linux-ppro
 $cc           = gcc
 $cflags       = -DL_ENDIAN -DTERMIO -O3 -fomit-frame-pointer -mcpu=pentiumpro -Wall
diff --git a/config b/config
index a0dc457c55..46102cebde 100755
--- a/config
+++ b/config
@@ -110,12 +110,8 @@ case "${SYSTEM}:${RELEASE}:${VERSION}:${MACHINE}" in
 	echo "m68k-apple-aux3"; exit 0
 	;;
 
-    AIX:[3456789]:4:*)
-	echo "${MACHINE}-ibm-aix43"; exit 0
-	;;
-
-    AIX:*:[56789]:*)
-	echo "${MACHINE}-ibm-aix43"; exit 0
+    AIX:*:[5-9]:*)
+	echo "${MACHINE}-ibm-aix5"; exit 0
 	;;
 
     AIX:*)
@@ -537,10 +533,7 @@ EOF
 	${CC} -o dummy dummy.c && OUT=`./dummy ${MACHINE}`
 	rm dummy dummy.c
 	;;
-  ppc64-*-linux2)
-	#Use the standard target for PPC architecture until we create a
-	#special one for the 64bit architecture.
-	OUT="linux-ppc" ;;
+  ppc64-*-linux2) OUT="linux-ppc64" ;;
   ppc-*-linux2) OUT="linux-ppc" ;;
   m68k-*-linux*) OUT="linux-m68k" ;;
   ia64-*-linux?) OUT="linux-ia64" ;;
@@ -719,6 +712,35 @@ EOF
 	fi
 	options="$options -D_REENTRANT" ;;
   *-hpux)	OUT="hpux-parisc-$CC" ;;
+  *-aix5)
+	KERNEL_BITS=`(getconf KERNEL_BITMODE) 2>/dev/null`
+	KERNEL_BITS=${KERNEL_BITS:-32}
+	if [ $KERNEL_BITS -eq 64 ]; then
+	    # we default to 64-bit because PKI performance is >3x better...
+	    OBJECT_MODE=${OBJECT_MODE:-$KERNEL_BITS}
+	else
+	    OBJECT_MODE=32
+	fi
+	OUT="aix-cc"
+	if [ "$CC" = "cc" -a $OBJECT_MODE -eq 64 ]; then
+	    OUT="aix64-cc"
+	    echo "WARNING! If you wish to build 32-bit kit, then you have to"
+	    echo "         invoke './Configure aix-cc' *manually*."
+	    if [ "$TEST" = "false" ]; then
+		echo "         You have ~5 seconds to press Ctrl-C to abort."
+		(stty -icanon min 0 time 50; read waste) < /dev/tty
+	    fi
+	elif [ "$CC" = "gcc ]; then
+	    OUT="aix-gcc"
+	fi
+	;;
+  *-aix)
+	if [ "$CC" = "gcc" ]; then
+	    OUT="aix-gcc"
+	else
+	    OUT="aix43-cc"
+	fi
+	;;
   # these are all covered by the catchall below
   # *-aix) OUT="aix-$CC" ;;
   # *-dgux) OUT="dgux" ;;
diff --git a/crypto/bn/Makefile.ssl b/crypto/bn/Makefile.ssl
index 70fec63620..490d6d7dae 100644
--- a/crypto/bn/Makefile.ssl
+++ b/crypto/bn/Makefile.ssl
@@ -129,6 +129,13 @@ asm/pa-risc2W.o: asm/pa-risc2W.s
 asm/pa-risc2.o: asm/pa-risc2.s
 	/usr/ccs/bin/as -o asm/pa-risc2.o asm/pa-risc2.s
 
+# ppc - AIX, Linux, MacOS X...
+asm/linux_ppc32.s: asm/ppc.pl;	$(PERL) $< $@
+asm/linux_ppc64.s: asm/ppc.pl;	$(PERL) $< $@
+asm/aix_ppc32.s: asm/ppc.pl;	$(PERL) $< $@
+asm/aix_ppc64.s: asm/ppc.pl;	$(PERL) $< $@
+asm/osx_ppc32.s: asm/ppc.pl;	$(PERL) $< $@
+
 files:
 	$(PERL) $(TOP)/util/files.pl Makefile.ssl >> $(TOP)/MINFO
 
diff --git a/crypto/bn/asm/ppc.pl b/crypto/bn/asm/ppc.pl
new file mode 100644
index 0000000000..307c7ccb35
--- /dev/null
+++ b/crypto/bn/asm/ppc.pl
@@ -0,0 +1,2081 @@
+#!/usr/bin/env perl
+#
+# Implemented as a Perl wrapper as we want to support several different
+# architectures with single file. We pick up the target based on the
+# file name we are asked to generate.
+#
+# It should be noted though that this perl code is nothing like
+# <openssl>/crypto/perlasm/x86*. In this case perl is used pretty much
+# as pre-processor to cover for platform differences in name decoration,
+# linker tables, 32-/64-bit instruction sets...
+#
+# As you might know there're several PowerPC ABI in use. Most notably
+# Linux and AIX use different 32-bit ABIs. Good news are that these ABIs
+# are similar enough to implement leaf(!) functions, which would be ABI
+# neutral. And that's what you find here: ABI neutral leaf functions.
+# In case you wonder what that is...
+#
+#       AIX performance
+#
+#	MEASUREMENTS WITH cc ON a 200 MhZ PowerPC 604e.
+#
+#	The following is the performance of 32-bit compiler
+#	generated code:
+#
+#	OpenSSL 0.9.6c 21 dec 2001
+#	built on: Tue Jun 11 11:06:51 EDT 2002
+#	options:bn(64,32) ...
+#compiler: cc -DTHREADS  -DAIX -DB_ENDIAN -DBN_LLONG -O3
+#                  sign    verify    sign/s verify/s
+#rsa  512 bits   0.0098s   0.0009s    102.0   1170.6
+#rsa 1024 bits   0.0507s   0.0026s     19.7    387.5
+#rsa 2048 bits   0.3036s   0.0085s      3.3    117.1
+#rsa 4096 bits   2.0040s   0.0299s      0.5     33.4
+#dsa  512 bits   0.0087s   0.0106s    114.3     94.5
+#dsa 1024 bits   0.0256s   0.0313s     39.0     32.0	
+#
+#	Same bechmark with this assembler code:
+#
+#rsa  512 bits   0.0056s   0.0005s    178.6   2049.2
+#rsa 1024 bits   0.0283s   0.0015s     35.3    674.1
+#rsa 2048 bits   0.1744s   0.0050s      5.7    201.2
+#rsa 4096 bits   1.1644s   0.0179s      0.9     55.7
+#dsa  512 bits   0.0052s   0.0062s    191.6    162.0
+#dsa 1024 bits   0.0149s   0.0180s     67.0     55.5
+#
+#	Number of operations increases by at almost 75%
+#
+#	Here are performance numbers for 64-bit compiler
+#	generated code:
+#
+#	OpenSSL 0.9.6g [engine] 9 Aug 2002
+#	built on: Fri Apr 18 16:59:20 EDT 2003
+#	options:bn(64,64) ...
+#	compiler: cc -DTHREADS -D_REENTRANT -q64 -DB_ENDIAN -O3
+#                  sign    verify    sign/s verify/s
+#rsa  512 bits   0.0028s   0.0003s    357.1   3844.4
+#rsa 1024 bits   0.0148s   0.0008s     67.5   1239.7
+#rsa 2048 bits   0.0963s   0.0028s     10.4    353.0
+#rsa 4096 bits   0.6538s   0.0102s      1.5     98.1
+#dsa  512 bits   0.0026s   0.0032s    382.5    313.7
+#dsa 1024 bits   0.0081s   0.0099s    122.8    100.6
+#
+#	Same benchmark with this assembler code:
+#
+#rsa  512 bits   0.0020s   0.0002s    510.4   6273.7
+#rsa 1024 bits   0.0088s   0.0005s    114.1   2128.3
+#rsa 2048 bits   0.0540s   0.0016s     18.5    622.5
+#rsa 4096 bits   0.3700s   0.0058s      2.7    171.0
+#dsa  512 bits   0.0016s   0.0020s    610.7    507.1
+#dsa 1024 bits   0.0047s   0.0058s    212.5    173.2
+#	
+#	Again, performance increases by at about 75%
+#
+#       Mac OS X, Apple G5 1.8GHz (Note this is 32 bit code)
+#       OpenSSL 0.9.7c 30 Sep 2003
+#
+#       Original code.
+#
+#rsa  512 bits   0.0011s   0.0001s    906.1  11012.5
+#rsa 1024 bits   0.0060s   0.0003s    166.6   3363.1
+#rsa 2048 bits   0.0370s   0.0010s     27.1    982.4
+#rsa 4096 bits   0.2426s   0.0036s      4.1    280.4
+#dsa  512 bits   0.0010s   0.0012s   1038.1    841.5
+#dsa 1024 bits   0.0030s   0.0037s    329.6    269.7
+#dsa 2048 bits   0.0101s   0.0127s     98.9     78.6
+#
+#       Same benchmark with this assembler code:
+#
+#rsa  512 bits   0.0007s   0.0001s   1416.2  16645.9
+#rsa 1024 bits   0.0036s   0.0002s    274.4   5380.6
+#rsa 2048 bits   0.0222s   0.0006s     45.1   1589.5
+#rsa 4096 bits   0.1469s   0.0022s      6.8    449.6
+#dsa  512 bits   0.0006s   0.0007s   1664.2   1376.2
+#dsa 1024 bits   0.0018s   0.0023s    545.0    442.2
+#dsa 2048 bits   0.0061s   0.0075s    163.5    132.8
+#
+#        Performance increase of ~60%
+#
+#	If you have comments or suggestions to improve code send
+#	me a note at schari@us.ibm.com
+#
+
+$opf = shift;
+
+if ($opf =~ /32\.s/) {
+	$BITS=	32;
+	$BNSZ=	$BITS/8;
+	$ISA=	"\"ppc\"";
+
+	$LD=	"lwz";		# load
+	$LDU=	"lwzu";		# load and update
+	$ST=	"stw";		# store
+	$STU=	"stwu";		# store and update
+	$UMULL=	"mullw";	# unsigned multiply low
+	$UMULH=	"mulhwu";	# unsigned multiply high
+	$UDIV=	"divwu";	# unsigned divide
+	$UCMPI=	"cmplwi";	# unsigned compare with immediate
+	$UCMP=	"cmplw";	# unsigned compare
+	$COUNTZ="cntlzw";	# count leading zeros
+	$SHL=	"slw";		# shift left
+	$SHR=	"srw";		# unsigned shift right
+	$SHRI=	"srwi";		# unsigned shift right by immediate	
+	$SHLI=	"slwi";		# shift left by immediate
+	$CLRU=	"clrlwi";	# clear upper bits
+	$INSR=	"insrwi";	# insert right
+	$ROTL=	"rotlwi";	# rotate left by immediate
+} elsif ($opf =~ /64\.s/) {
+	$BITS=	64;
+	$BNSZ=	$BITS/8;
+	$ISA=	"\"ppc64\"";
+
+	# same as above, but 64-bit mnemonics...
+	$LD=	"ld";		# load
+	$LDU=	"ldu";		# load and update
+	$ST=	"std";		# store
+	$STU=	"stdu";		# store and update
+	$UMULL=	"mulld";	# unsigned multiply low
+	$UMULH=	"mulhdu";	# unsigned multiply high
+	$UDIV=	"divdu";	# unsigned divide
+	$UCMPI=	"cmpldi";	# unsigned compare with immediate
+	$UCMP=	"cmpld";	# unsigned compare
+	$COUNTZ="cntlzd";	# count leading zeros
+	$SHL=	"sld";		# shift left
+	$SHR=	"srd";		# unsigned shift right
+	$SHRI=	"srdi";		# unsigned shift right by immediate	
+	$SHLI=	"sldi";		# shift left by immediate
+	$CLRU=	"clrldi";	# clear upper bits
+	$INSR=	"insrdi";	# insert right 
+	$ROTL=	"rotldi";	# rotate left by immediate
+} else { die "nonsense $opf"; }
+
+( defined shift || open STDOUT,">$opf" ) || die "can't open $opf: $!";
+
+# function entry points from the AIX code
+#
+# There are other, more elegant, ways to handle this. We (IBM) chose
+# this approach as it plays well with scripts we run to 'namespace'
+# OpenSSL .i.e. we add a prefix to all the public symbols so we can
+# co-exist in the same process with other implementations of OpenSSL.
+# 'cleverer' ways of doing these substitutions tend to hide data we
+# need to be obvious.
+#
+my @items = ("bn_sqr_comba4",
+	     "bn_sqr_comba8",
+	     "bn_mul_comba4",
+	     "bn_mul_comba8",
+	     "bn_sub_words",
+	     "bn_add_words",
+	     "bn_div_words",
+	     "bn_sqr_words",
+	     "bn_mul_words",
+	     "bn_mul_add_words");
+
+if    ($opf =~ /linux/)	{  do_linux();	}
+elsif ($opf =~ /aix/)	{  do_aix();	}
+elsif ($opf =~ /osx/)	{  do_osx();	}
+else			{  do_bsd();	}
+
+sub do_linux {
+    $d=&data();
+
+    if ($BITS==64) {
+      foreach $t (@items) {
+        $d =~ s/\.$t:/\
+\t.section\t".opd","aw"\
+\t.align\t3\
+\t.globl\t$t\
+$t:\
+\t.quad\t.$t,.TOC.\@tocbase,0\
+\t.size\t$t,24\
+\t.previous\n\
+\t.type\t.$t,\@function\
+\t.globl\t.$t\
+.$t:/g;
+      }
+    }
+    else {
+      foreach $t (@items) {
+        $d=~s/\.$t/$t/g;
+      }
+    }
+    # hide internal labels to avoid pollution of name table...
+    $d=~s/Lppcasm_/.Lppcasm_/gm;
+    print $d;
+}
+
+sub do_aix {
+    # AIX assembler is smart enough to please the linker without
+    # making us do something special...
+    print &data();
+}
+
+# MacOSX 32 bit
+sub do_osx {
+    $d=&data();
+    # Change the bn symbol prefix from '.' to '_'
+    foreach $t (@items) {
+      $d=~s/\.$t/_$t/g;
+    }
+    # Change .machine to something OS X asm will accept
+    $d=~s/\.machine.*/.text/g;
+    $d=~s/\#/;/g; # change comment from '#' to ';'
+    print $d;
+}
+
+# BSD (Untested)
+sub do_bsd {
+    $d=&data();
+    foreach $t (@items) {
+      $d=~s/\.$t/_$t/g;
+    }
+    print $d;
+}
+
+sub data {
+	local($data)=<<EOF;
+#--------------------------------------------------------------------
+#
+#
+#
+#
+#	File:		ppc32.s
+#
+#	Created by:	Suresh Chari
+#			IBM Thomas J. Watson Research Library
+#			Hawthorne, NY
+#
+#
+#	Description:	Optimized assembly routines for OpenSSL crypto
+#			on the 32 bitPowerPC platform.
+#
+#
+#	Version History
+#
+#	2. Fixed bn_add,bn_sub and bn_div_words, added comments,
+#	   cleaned up code. Also made a single version which can
+#	   be used for both the AIX and Linux compilers. See NOTE
+#	   below.
+#				12/05/03		Suresh Chari
+#			(with lots of help from)        Andy Polyakov
+##	
+#	1. Initial version	10/20/02		Suresh Chari
+#
+#
+#	The following file works for the xlc,cc
+#	and gcc compilers.
+#
+#	NOTE:	To get the file to link correctly with the gcc compiler
+#	        you have to change the names of the routines and remove
+#		the first .(dot) character. This should automatically
+#		be done in the build process.
+#
+#	Hand optimized assembly code for the following routines
+#	
+#	bn_sqr_comba4
+#	bn_sqr_comba8
+#	bn_mul_comba4
+#	bn_mul_comba8
+#	bn_sub_words
+#	bn_add_words
+#	bn_div_words
+#	bn_sqr_words
+#	bn_mul_words
+#	bn_mul_add_words
+#
+#	NOTE:	It is possible to optimize this code more for
+#	specific PowerPC or Power architectures. On the Northstar
+#	architecture the optimizations in this file do
+#	 NOT provide much improvement.
+#
+#	If you have comments or suggestions to improve code send
+#	me a note at schari\@us.ibm.com
+#
+#--------------------------------------------------------------------------
+#
+#	Defines to be used in the assembly code.
+#	
+.set r0,0	# we use it as storage for value of 0
+.set SP,1	# preserved
+.set RTOC,2	# preserved 
+.set r3,3	# 1st argument/return value
+.set r4,4	# 2nd argument/volatile register
+.set r5,5	# 3rd argument/volatile register
+.set r6,6	# ...
+.set r7,7
+.set r8,8
+.set r9,9
+.set r10,10
+.set r11,11
+.set r12,12
+.set r13,13	# not used, nor any other "below" it...
+
+.set BO_IF_NOT,4
+.set BO_IF,12
+.set BO_dCTR_NZERO,16
+.set BO_dCTR_ZERO,18
+.set BO_ALWAYS,20
+.set CR0_LT,0;
+.set CR0_GT,1;
+.set CR0_EQ,2
+.set CR1_FX,4;
+.set CR1_FEX,5;
+.set CR1_VX,6
+.set LR,8
+
+#	Declare function names to be global
+#	NOTE:	For gcc these names MUST be changed to remove
+#	        the first . i.e. for example change ".bn_sqr_comba4"
+#		to "bn_sqr_comba4". This should be automatically done
+#		in the build.
+	
+	.globl	.bn_sqr_comba4
+	.globl	.bn_sqr_comba8
+	.globl	.bn_mul_comba4
+	.globl	.bn_mul_comba8
+	.globl	.bn_sub_words
+	.globl	.bn_add_words
+	.globl	.bn_div_words
+	.globl	.bn_sqr_words
+	.globl	.bn_mul_words
+	.globl	.bn_mul_add_words
+	
+# .text section
+	
+	.machine	$ISA
+
+#
+#	NOTE:	The following label name should be changed to
+#		"bn_sqr_comba4" i.e. remove the first dot
+#		for the gcc compiler. This should be automatically
+#		done in the build
+#
+
+.align	4
+.bn_sqr_comba4:
+#
+# Optimized version of bn_sqr_comba4.
+#
+# void bn_sqr_comba4(BN_ULONG *r, BN_ULONG *a)
+# r3 contains r
+# r4 contains a
+#
+# Freely use registers r5,r6,r7,r8,r9,r10,r11 as follows:	
+# 
+# r5,r6 are the two BN_ULONGs being multiplied.
+# r7,r8 are the results of the 32x32 giving 64 bit multiply.
+# r9,r10, r11 are the equivalents of c1,c2, c3.
+# Here's the assembly
+#
+#
+	xor		r0,r0,r0		# set r0 = 0. Used in the addze
+						# instructions below
+	
+						#sqr_add_c(a,0,c1,c2,c3)
+	$LD		r5,`0*$BNSZ`(r4)		
+	$UMULL		r9,r5,r5		
+	$UMULH		r10,r5,r5		#in first iteration. No need
+						#to add since c1=c2=c3=0.
+						# Note c3(r11) is NOT set to 0
+						# but will be.
+
+	$ST		r9,`0*$BNSZ`(r3)	# r[0]=c1;
+						# sqr_add_c2(a,1,0,c2,c3,c1);
+	$LD		r6,`1*$BNSZ`(r4)		
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+					
+	addc		r7,r7,r7		# compute (r7,r8)=2*(r7,r8)
+	adde		r8,r8,r8
+	addze		r9,r0			# catch carry if any.
+						# r9= r0(=0) and carry 
+	
+	addc		r10,r7,r10		# now add to temp result.
+	addze		r11,r8                  # r8 added to r11 which is 0 
+	addze		r9,r9
+	
+	$ST		r10,`1*$BNSZ`(r3)	#r[1]=c2; 
+						#sqr_add_c(a,1,c3,c1,c2)
+	$UMULL		r7,r6,r6
+	$UMULH		r8,r6,r6
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r0
+						#sqr_add_c2(a,2,0,c3,c1,c2)
+	$LD		r6,`2*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r7,r7,r7
+	adde		r8,r8,r8
+	addze		r10,r10
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+	$ST		r11,`2*$BNSZ`(r3)	#r[2]=c3 
+						#sqr_add_c2(a,3,0,c1,c2,c3);
+	$LD		r6,`3*$BNSZ`(r4)		
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	addc		r7,r7,r7
+	adde		r8,r8,r8
+	addze		r11,r0
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+						#sqr_add_c2(a,2,1,c1,c2,c3);
+	$LD		r5,`1*$BNSZ`(r4)
+	$LD		r6,`2*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r7,r7,r7
+	adde		r8,r8,r8
+	addze		r11,r11
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+	$ST		r9,`3*$BNSZ`(r3)	#r[3]=c1
+						#sqr_add_c(a,2,c2,c3,c1);
+	$UMULL		r7,r6,r6
+	$UMULH		r8,r6,r6
+	addc		r10,r7,r10
+	adde		r11,r8,r11
+	addze		r9,r0
+						#sqr_add_c2(a,3,1,c2,c3,c1);
+	$LD		r6,`3*$BNSZ`(r4)		
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	addc		r7,r7,r7
+	adde		r8,r8,r8
+	addze		r9,r9
+	
+	addc		r10,r7,r10
+	adde		r11,r8,r11
+	addze		r9,r9
+	$ST		r10,`4*$BNSZ`(r3)	#r[4]=c2
+						#sqr_add_c2(a,3,2,c3,c1,c2);
+	$LD		r5,`2*$BNSZ`(r4)		
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	addc		r7,r7,r7
+	adde		r8,r8,r8
+	addze		r10,r0
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+	$ST		r11,`5*$BNSZ`(r3)	#r[5] = c3
+						#sqr_add_c(a,3,c1,c2,c3);
+	$UMULL		r7,r6,r6		
+	$UMULH		r8,r6,r6
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+
+	$ST		r9,`6*$BNSZ`(r3)	#r[6]=c1
+	$ST		r10,`7*$BNSZ`(r3)	#r[7]=c2
+	bclr	BO_ALWAYS,CR0_LT
+	.long	0x00000000
+
+#
+#	NOTE:	The following label name should be changed to
+#		"bn_sqr_comba8" i.e. remove the first dot
+#		for the gcc compiler. This should be automatically
+#		done in the build
+#
+	
+.align	4
+.bn_sqr_comba8:
+#
+# This is an optimized version of the bn_sqr_comba8 routine.
+# Tightly uses the adde instruction
+#
+#
+# void bn_sqr_comba8(BN_ULONG *r, BN_ULONG *a)
+# r3 contains r
+# r4 contains a
+#
+# Freely use registers r5,r6,r7,r8,r9,r10,r11 as follows:	
+# 
+# r5,r6 are the two BN_ULONGs being multiplied.
+# r7,r8 are the results of the 32x32 giving 64 bit multiply.
+# r9,r10, r11 are the equivalents of c1,c2, c3.
+#
+# Possible optimization of loading all 8 longs of a into registers
+# doesnt provide any speedup
+# 
+
+	xor		r0,r0,r0		#set r0 = 0.Used in addze
+						#instructions below.
+
+						#sqr_add_c(a,0,c1,c2,c3);
+	$LD		r5,`0*$BNSZ`(r4)
+	$UMULL		r9,r5,r5		#1st iteration:	no carries.
+	$UMULH		r10,r5,r5
+	$ST		r9,`0*$BNSZ`(r3)	# r[0]=c1;
+						#sqr_add_c2(a,1,0,c2,c3,c1);
+	$LD		r6,`1*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6	
+	
+	addc		r10,r7,r10		#add the two register number
+	adde		r11,r8,r0 		# (r8,r7) to the three register
+	addze		r9,r0			# number (r9,r11,r10).NOTE:r0=0
+	
+	addc		r10,r7,r10		#add the two register number
+	adde		r11,r8,r11 		# (r8,r7) to the three register
+	addze		r9,r9			# number (r9,r11,r10).
+	
+	$ST		r10,`1*$BNSZ`(r3)	# r[1]=c2
+				
+						#sqr_add_c(a,1,c3,c1,c2);
+	$UMULL		r7,r6,r6
+	$UMULH		r8,r6,r6
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r0
+						#sqr_add_c2(a,2,0,c3,c1,c2);
+	$LD		r6,`2*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+	
+	$ST		r11,`2*$BNSZ`(r3)	#r[2]=c3
+						#sqr_add_c2(a,3,0,c1,c2,c3);
+	$LD		r6,`3*$BNSZ`(r4)	#r6 = a[3]. r5 is already a[0].
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r0
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+						#sqr_add_c2(a,2,1,c1,c2,c3);
+	$LD		r5,`1*$BNSZ`(r4)
+	$LD		r6,`2*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+	
+	$ST		r9,`3*$BNSZ`(r3)	#r[3]=c1;
+						#sqr_add_c(a,2,c2,c3,c1);
+	$UMULL		r7,r6,r6
+	$UMULH		r8,r6,r6
+	
+	addc		r10,r7,r10
+	adde		r11,r8,r11
+	addze		r9,r0
+						#sqr_add_c2(a,3,1,c2,c3,c1);
+	$LD		r6,`3*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r10,r7,r10
+	adde		r11,r8,r11
+	addze		r9,r9
+	
+	addc		r10,r7,r10
+	adde		r11,r8,r11
+	addze		r9,r9
+						#sqr_add_c2(a,4,0,c2,c3,c1);
+	$LD		r5,`0*$BNSZ`(r4)
+	$LD		r6,`4*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r10,r7,r10
+	adde		r11,r8,r11
+	addze		r9,r9
+	
+	addc		r10,r7,r10
+	adde		r11,r8,r11
+	addze		r9,r9
+	$ST		r10,`4*$BNSZ`(r3)	#r[4]=c2;
+						#sqr_add_c2(a,5,0,c3,c1,c2);
+	$LD		r6,`5*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r0
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+						#sqr_add_c2(a,4,1,c3,c1,c2);
+	$LD		r5,`1*$BNSZ`(r4)
+	$LD		r6,`4*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+						#sqr_add_c2(a,3,2,c3,c1,c2);
+	$LD		r5,`2*$BNSZ`(r4)
+	$LD		r6,`3*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+	
+	addc		r11,r7,r11
+	adde		r9,r8,r9
+	addze		r10,r10
+	$ST		r11,`5*$BNSZ`(r3)	#r[5]=c3;
+						#sqr_add_c(a,3,c1,c2,c3);
+	$UMULL		r7,r6,r6
+	$UMULH		r8,r6,r6
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r0
+						#sqr_add_c2(a,4,2,c1,c2,c3);
+	$LD		r6,`4*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+						#sqr_add_c2(a,5,1,c1,c2,c3);
+	$LD		r5,`1*$BNSZ`(r4)
+	$LD		r6,`5*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+	
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+						#sqr_add_c2(a,6,0,c1,c2,c3);
+	$LD		r5,`0*$BNSZ`(r4)
+	$LD		r6,`6*$BNSZ`(r4)
+	$UMULL		r7,r5,r6
+	$UMULH		r8,r5,r6
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11
+	addc		r9,r7,r9
+	adde		r10,r8,r10
+	addze		r11,r11<