From c5f17d45c1866ad7c65bcec66be03212ec5cc3ae Mon Sep 17 00:00:00 2001
From: Andy Polyakov <appro@openssl.org>
Date: Tue, 17 Oct 2006 16:13:18 +0000
Subject: [PATCH] Further synchronizations with md32_common.h update,
 consistent naming for low-level SHA block routines.

---
 crypto/sha/asm/sha1-586.pl      |  4 +-
 crypto/sha/asm/sha1-ia64.pl     | 10 ++--
 crypto/sha/asm/sha1-ppc.pl      |  4 +-
 crypto/sha/asm/sha1-x86_64.pl   |  6 ++-
 crypto/sha/asm/sha512-ia64.pl   |  4 +-
 crypto/sha/asm/sha512-ppc.pl    |  4 +-
 crypto/sha/asm/sha512-x86_64.pl | 29 ++----------
 crypto/sha/sha1s.cpp            | 82 ---------------------------------
 crypto/sha/sha256.c             | 14 +++---
 crypto/sha/sha512.c             | 18 ++++----
 crypto/sha/sha_locl.h           | 30 ++++--------
 11 files changed, 45 insertions(+), 160 deletions(-)
 delete mode 100644 crypto/sha/sha1s.cpp

diff --git a/crypto/sha/asm/sha1-586.pl b/crypto/sha/asm/sha1-586.pl
index 10c7d43641..0b1e1731cb 100644
--- a/crypto/sha/asm/sha1-586.pl
+++ b/crypto/sha/asm/sha1-586.pl
@@ -149,7 +149,7 @@ sub BODY_40_59
 	&add($f,$e);			# f+=ROTATE(a,5)
 	}
 
-&function_begin("sha1_block_asm_data_order",16);
+&function_begin("sha1_block_data_order",16);
 	&mov($tmp1,&wparam(0));	# SHA_CTX *c
 	&mov($T,&wparam(1));	# const void *input
 	&mov($A,&wparam(2));	# size_t num
@@ -214,7 +214,7 @@ sub BODY_40_59
 	&jb(&label("loop"));
 
 	&stack_pop(16);
-&function_end("sha1_block_asm_data_order");
+&function_end("sha1_block_data_order");
 &asciz("SHA1 block transform for x86, CRYPTOGAMS by <appro\@openssl.org>");
 
 &asm_finish();
diff --git a/crypto/sha/asm/sha1-ia64.pl b/crypto/sha/asm/sha1-ia64.pl
index 9e2f73560f..242cadcc52 100644
--- a/crypto/sha/asm/sha1-ia64.pl
+++ b/crypto/sha/asm/sha1-ia64.pl
@@ -239,11 +239,11 @@ tmp3=r11;
 ctx=r32;	// in0
 inp=r33;	// in1
 
-// void sha1_block_asm_data_order(SHA_CTX *c,const void *p,size_t num);
-.global	sha1_block_asm_data_order#
-.proc	sha1_block_asm_data_order#
+// void sha1_block_data_order(SHA_CTX *c,const void *p,size_t num);
+.global	sha1_block_data_order#
+.proc	sha1_block_data_order#
 .align	32
-sha1_block_asm_data_order:
+sha1_block_data_order:
 	.prologue
 { .mmi;	alloc	tmp1=ar.pfs,3,15,0,0
 	$ADDP	tmp0=4,ctx
@@ -306,7 +306,7 @@ $code.=<<___;
 { .mib;	st4	[ctx]=$h4,-16
 	mov	pr=r2,0x1ffff
 	br.ret.sptk.many	b0	};;
-.endp	sha1_block_asm_data_order#
+.endp	sha1_block_data_order#
 ___
 
 print $code;
diff --git a/crypto/sha/asm/sha1-ppc.pl b/crypto/sha/asm/sha1-ppc.pl
index 5c5c7d87ec..fbd051652c 100755
--- a/crypto/sha/asm/sha1-ppc.pl
+++ b/crypto/sha/asm/sha1-ppc.pl
@@ -158,9 +158,9 @@ ___
 $code=<<___;
 .text
 
-.globl	.sha1_block_asm_data_order
+.globl	.sha1_block_data_order
 .align	4
-.sha1_block_asm_data_order:
+.sha1_block_data_order:
 	mflr	r0
 	$STU	$sp,`-($FRAME+64)`($sp)
 	$PUSH	r0,`$FRAME-$SIZE_T*18`($sp)
diff --git a/crypto/sha/asm/sha1-x86_64.pl b/crypto/sha/asm/sha1-x86_64.pl
index 857db2d671..f3cb1347fb 100755
--- a/crypto/sha/asm/sha1-x86_64.pl
+++ b/crypto/sha/asm/sha1-x86_64.pl
@@ -158,6 +158,8 @@ $code.=<<___ if ($i<79);
 	rol	\$30,$b
 	add	$t0,$f
 	rol	\$1,$xi
+___
+$code.=<<___ if ($i<76);
 	mov	$xi,`4*($j%16)`(%rsp)
 ___
 $code.=<<___ if ($i==79);
@@ -200,7 +202,7 @@ ___
 
 $code=".text\n";
 
-&PROLOGUE("sha1_block_asm_data_order");
+&PROLOGUE("sha1_block_data_order");
 $code.=".align	4\n.Lloop:\n";
 for($i=0;$i<20;$i++)	{ &BODY_00_19($i,@V); unshift(@V,pop(@V)); }
 for(;$i<40;$i++)	{ &BODY_20_39($i,@V); unshift(@V,pop(@V)); }
@@ -227,7 +229,7 @@ $code.=<<___;
 	sub	\$1,$num
 	jnz	.Lloop
 ___
-&EPILOGUE("sha1_block_asm_data_order");
+&EPILOGUE("sha1_block_data_order");
 $code.=<<___;
 .asciz	"SHA1 block transform for x86_64, CRYPTOGAMS by <appro\@openssl.org>"
 ___
diff --git a/crypto/sha/asm/sha512-ia64.pl b/crypto/sha/asm/sha512-ia64.pl
index 628e33e426..bba7f6bb15 100755
--- a/crypto/sha/asm/sha512-ia64.pl
+++ b/crypto/sha/asm/sha512-ia64.pl
@@ -71,7 +71,7 @@ if ($output =~ /512.*\.[s|asm]/) {
 	$ADD="add";
 	$SHRU="shr.u";
 	$TABLE="K512";
-	$func="sha512_block";
+	$func="sha512_block_data_order";
 	@Sigma0=(28,34,39);
 	@Sigma1=(14,18,41);
 	@sigma0=(1,  8, 7);
@@ -85,7 +85,7 @@ if ($output =~ /512.*\.[s|asm]/) {
 	$ADD="padd4";
 	$SHRU="pshr4.u";
 	$TABLE="K256";
-	$func="sha256_block";
+	$func="sha256_block_data_order";
 	@Sigma0=( 2,13,22);
 	@Sigma1=( 6,11,25);
 	@sigma0=( 7,18, 3);
diff --git a/crypto/sha/asm/sha512-ppc.pl b/crypto/sha/asm/sha512-ppc.pl
index b866f93eef..877d172215 100755
--- a/crypto/sha/asm/sha512-ppc.pl
+++ b/crypto/sha/asm/sha512-ppc.pl
@@ -41,7 +41,7 @@ if ($output =~ /64/) {
 	die "can't call ../perlasm/ppc-xlate.pl: $!";
 
 if ($output =~ /512/) {
-	$func="sha512_block";
+	$func="sha512_block_data_order";
 	$SZ=8;
 	@Sigma0=(28,34,39);
 	@Sigma1=(14,18,41);
@@ -53,7 +53,7 @@ if ($output =~ /512/) {
 	$ROR="rotrdi";
 	$SHR="srdi";
 } else {
-	$func="sha256_block";
+	$func="sha256_block_data_order";
 	$SZ=4;
 	@Sigma0=( 2,13,22);
 	@Sigma1=( 6,11,25);
diff --git a/crypto/sha/asm/sha512-x86_64.pl b/crypto/sha/asm/sha512-x86_64.pl
index 38157b610c..4cf5c623d6 100755
--- a/crypto/sha/asm/sha512-x86_64.pl
+++ b/crypto/sha/asm/sha512-x86_64.pl
@@ -44,7 +44,7 @@ $output=shift;
 open STDOUT,"| $^X ../perlasm/x86_64-xlate.pl $output";
 
 if ($output =~ /512/) {
-	$func="sha512_block";
+	$func="sha512_block_data_order";
 	$TABLE="K512";
 	$SZ=8;
 	@ROT=($A,$B,$C,$D,$E,$F,$G,$H)=("%rax","%rbx","%rcx","%rdx",
@@ -56,7 +56,7 @@ if ($output =~ /512/) {
 	@sigma1=(19,61, 6);
 	$rounds=80;
 } else {
-	$func="sha256_block";
+	$func="sha256_block_data_order";
 	$TABLE="K256";
 	$SZ=4;
 	@ROT=($A,$B,$C,$D,$E,$F,$G,$H)=("%eax","%ebx","%ecx","%edx",
@@ -77,9 +77,8 @@ $Tbl="%rbp";
 $_ctx="16*$SZ+0*8(%rsp)";
 $_inp="16*$SZ+1*8(%rsp)";
 $_end="16*$SZ+2*8(%rsp)";
-$_ord="16*$SZ+3*8(%rsp)";
-$_rsp="16*$SZ+4*8(%rsp)";
-$framesz="16*$SZ+5*8";
+$_rsp="16*$SZ+3*8(%rsp)";
+$framesz="16*$SZ+4*8";
 
 
 sub ROUND_00_15()
@@ -189,7 +188,6 @@ $func:
 	mov	$ctx,$_ctx		# save ctx, 1st arg
 	mov	$inp,$_inp		# save inp, 2nd arh
 	mov	%rdx,$_end		# save end pointer, "3rd" arg
-	mov	%ecx,$_ord		# save host, 4th arg
 	mov	%rbp,$_rsp		# save copy of %rsp
 
 	.picmeup $Tbl
@@ -209,25 +207,6 @@ $func:
 .Lloop:
 	xor	$round,$round
 ___
-if ($SZ==4) {
-$code.=<<___;
-	cmpl	\$0,$_ord
-	je	.Ldata_order
-.align	16
-.Lhost_order:
-___
-
-	for($i=0;$i<16;$i++) {
-		$code.="	mov	$SZ*$i($inp),$T1\n";
-		&ROUND_00_15($i,@ROT);
-		unshift(@ROT,pop(@ROT));
-	}
-$code.=<<___;
-	jmp	.Lrounds_16_xx
-.align	16
-.Ldata_order:
-___
-} # 256
 	for($i=0;$i<16;$i++) {
 		$code.="	mov	$SZ*$i($inp),$T1\n";
 		$code.="	bswap	$T1\n";
diff --git a/crypto/sha/sha1s.cpp b/crypto/sha/sha1s.cpp
deleted file mode 100644
index af23d1e0f2..0000000000
--- a/crypto/sha/sha1s.cpp
+++ /dev/null
@@ -1,82 +0,0 @@
-//
-// gettsc.inl
-//
-// gives access to the Pentium's (secret) cycle counter
-//
-// This software was written by Leonard Janke (janke@unixg.ubc.ca)
-// in 1996-7 and is entered, by him, into the public domain.
-
-#if defined(__WATCOMC__)
-void GetTSC(unsigned long&);
-#pragma aux GetTSC = 0x0f 0x31 "mov [edi], eax" parm [edi] modify [edx eax];
-#elif defined(__GNUC__)
-inline
-void GetTSC(unsigned long& tsc)
-{
-  asm volatile(".byte 15, 49\n\t"
-	       : "=eax" (tsc)
-	       :
-	       : "%edx", "%eax");
-}
-#elif defined(_MSC_VER)
-inline
-void GetTSC(unsigned long& tsc)
-{
-  unsigned long a;
-  __asm _emit 0fh
-  __asm _emit 31h
-  __asm mov a, eax;
-  tsc=a;
-}
-#endif      
-
-#include <stdio.h>
-#include <stdlib.h>
-#include <openssl/sha.h>
-
-#define sha1_block_x86 sha1_block_asm_data_order
-extern "C" {
-void sha1_block_x86(SHA_CTX *ctx, unsigned char *buffer,int num);
-}
-
-void main(int argc,char *argv[])
-	{
-	unsigned char buffer[64*256];
-	SHA_CTX ctx;
-	unsigned long s1,s2,e1,e2;
-	unsigned char k[16];
-	unsigned long data[2];
-	unsigned char iv[8];
-	int i,num=0,numm;
-	int j=0;
-
-	if (argc >= 2)
-		num=atoi(argv[1]);
-
-	if (num == 0) num=16;
-	if (num > 250) num=16;
-	numm=num+2;
-#if 0
-	num*=64;
-	numm*=64;
-#endif
-
-	for (j=0; j<6; j++)
-		{
-		for (i=0; i<10; i++) /**/
-			{
-			sha1_block_x86(&ctx,buffer,numm);
-			GetTSC(s1);
-			sha1_block_x86(&ctx,buffer,numm);
-			GetTSC(e1);
-			GetTSC(s2);
-			sha1_block_x86(&ctx,buffer,num);
-			GetTSC(e2);
-			sha1_block_x86(&ctx,buffer,num);
-			}
-
-		printf("sha1 (%d bytes) %d %d (%.2f)\n",num*64,
-			e1-s1,e2-s2,(double)((e1-s1)-(e2-s2))/2);
-		}
-	}
-
diff --git a/crypto/sha/sha256.c b/crypto/sha/sha256.c
index f83e5afe71..9b69ff38d2 100644
--- a/crypto/sha/sha256.c
+++ b/crypto/sha/sha256.c
@@ -107,13 +107,14 @@ int SHA224_Final (unsigned char *md, SHA256_CTX *c)
 #define	HASH_TRANSFORM		SHA256_Transform
 #define	HASH_FINAL		SHA256_Final
 #define	HASH_BLOCK_DATA_ORDER	sha256_block_data_order
+#ifndef SHA256_ASM
+static
+#endif
 void sha256_block_data_order (SHA256_CTX *ctx, const void *in, size_t num);
 
 #include "md32_common.h"
 
-#ifdef SHA256_ASM
-void sha256_block (SHA256_CTX *ctx, const void *in, size_t num, int host);
-#else
+#ifndef SHA256_ASM
 static const SHA_LONG K256[64] = {
 	0x428a2f98UL,0x71374491UL,0xb5c0fbcfUL,0xe9b5dba5UL,
 	0x3956c25bUL,0x59f111f1UL,0x923f82a4UL,0xab1c5ed5UL,
@@ -147,7 +148,7 @@ static const SHA_LONG K256[64] = {
 
 #ifdef OPENSSL_SMALL_FOOTPRINT
 
-static void sha256_block (SHA256_CTX *ctx, const void *in, size_t num, int host)
+static void sha256_block_data_order (SHA256_CTX *ctx, const void *in, size_t num)
 	{
 	unsigned MD32_REG_T a,b,c,d,e,f,g,h,s0,s1,T1,T2;
 	SHA_LONG	X[16],l;
@@ -199,7 +200,7 @@ static void sha256_block (SHA256_CTX *ctx, const void *in, size_t num, int host)
 	T1 = X[(i)&0x0f] += s0 + s1 + X[(i+9)&0x0f];	\
 	ROUND_00_15(i,a,b,c,d,e,f,g,h);		} while (0)
 
-static void sha256_block (SHA256_CTX *ctx, const void *in, size_t num, int host)
+static void sha256_block_data_order (SHA256_CTX *ctx, const void *in, size_t num)
 	{
 	unsigned MD32_REG_T a,b,c,d,e,f,g,h,s0,s1,T1;
 	SHA_LONG	X[16];
@@ -278,7 +279,4 @@ static void sha256_block (SHA256_CTX *ctx, const void *in, size_t num, int host)
 #endif
 #endif /* SHA256_ASM */
 
-void HASH_BLOCK_DATA_ORDER (SHA256_CTX *ctx, const void *in, size_t num)
-{   sha256_block (ctx,in,num,0);   }
-
 #endif /* OPENSSL_NO_SHA256 */
diff --git a/crypto/sha/sha512.c b/crypto/sha/sha512.c
index dba4c49c0b..e7a7e5dea8 100644
--- a/crypto/sha/sha512.c
+++ b/crypto/sha/sha512.c
@@ -89,7 +89,7 @@ int SHA512_Init (SHA512_CTX *c)
 #ifndef SHA512_ASM
 static
 #endif
-void sha512_block (SHA512_CTX *ctx, const void *in, size_t num);
+void sha512_block_data_order (SHA512_CTX *ctx, const void *in, size_t num);
 
 int SHA512_Final (unsigned char *md, SHA512_CTX *c)
 	{
@@ -100,7 +100,7 @@ int SHA512_Final (unsigned char *md, SHA512_CTX *c)
 	n++;
 	if (n > (sizeof(c->u)-16))
 		memset (p+n,0,sizeof(c->u)-n), n=0,
-		sha512_block (c,p,1);
+		sha512_block_data_order (c,p,1);
 
 	memset (p+n,0,sizeof(c->u)-16-n);
 #ifdef	B_ENDIAN
@@ -125,7 +125,7 @@ int SHA512_Final (unsigned char *md, SHA512_CTX *c)
 	p[sizeof(c->u)-16] = (unsigned char)(c->Nh>>56);
 #endif
 
-	sha512_block (c,p,1);
+	sha512_block_data_order (c,p,1);
 
 	if (md==0) return 0;
 
@@ -197,7 +197,7 @@ int SHA512_Update (SHA512_CTX *c, const void *_data, size_t len)
 		else	{
 			memcpy (p+c->num,data,n), c->num = 0;
 			len-=n, data+=n;
-			sha512_block (c,p,1);
+			sha512_block_data_order (c,p,1);
 			}
 		}
 
@@ -207,12 +207,12 @@ int SHA512_Update (SHA512_CTX *c, const void *_data, size_t len)
 		if ((size_t)data%sizeof(c->u.d[0]) != 0)
 			while (len >= sizeof(c->u))
 				memcpy (p,data,sizeof(c->u)),
-				sha512_block (c,p,1),
+				sha512_block_data_order (c,p,1),
 				len  -= sizeof(c->u),
 				data += sizeof(c->u);
 		else
 #endif
-			sha512_block (c,data,len/sizeof(c->u)),
+			sha512_block_data_order (c,data,len/sizeof(c->u)),
 			data += len,
 			len  %= sizeof(c->u),
 			data -= len;
@@ -227,7 +227,7 @@ int SHA384_Update (SHA512_CTX *c, const void *data, size_t len)
 {   return SHA512_Update (c,data,len);   }
 
 void SHA512_Transform (SHA512_CTX *c, const unsigned char *data)
-{   sha512_block (c,data,1);  }
+{   sha512_block_data_order (c,data,1);  }
 
 unsigned char *SHA384(const unsigned char *d, size_t n, unsigned char *md)
 	{
@@ -392,7 +392,7 @@ static const SHA_LONG64 K512[80] = {
 
 #ifdef OPENSSL_SMALL_FOOTPRINT
 
-static void sha512_block (SHA512_CTX *ctx, const void *in, size_t num)
+static void sha512_block_data_order (SHA512_CTX *ctx, const void *in, size_t num)
 	{
 	const SHA_LONG64 *W=in;
 	SHA_LONG64	a,b,c,d,e,f,g,h,s0,s1,T1,T2;
@@ -453,7 +453,7 @@ static void sha512_block (SHA512_CTX *ctx, const void *in, size_t num)
 	T1 = X[(i)&0x0f] += s0 + s1 + X[(i+9)&0x0f];	\
 	ROUND_00_15(i,a,b,c,d,e,f,g,h);		} while (0)
 
-static void sha512_block (SHA512_CTX *ctx, const void *in, size_t num)
+static void sha512_block_data_order (SHA512_CTX *ctx, const void *in, size_t num)
 	{
 	const SHA_LONG64 *W=in;
 	SHA_LONG64	a,b,c,d,e,f,g,h,s0,s1,T1;
diff --git a/crypto/sha/sha_locl.h b/crypto/sha/sha_locl.h
index d482ba12a0..247d4df6f5 100644
--- a/crypto/sha/sha_locl.h
+++ b/crypto/sha/sha_locl.h
@@ -85,7 +85,7 @@
 # define HASH_BLOCK_DATA_ORDER   	sha_block_data_order
 # define Xupdate(a,ix,ia,ib,ic,id)	(ix=(a)=(ia^ib^ic^id))
 
-  void sha_block_data_order (SHA_CTX *c, const void *p,size_t num);
+static void sha_block_data_order (SHA_CTX *c, const void *p,size_t num);
 
 #elif defined(SHA_1)
 
@@ -105,22 +105,10 @@
 					)
 # endif
 
-# ifdef SHA1_ASM
-#  if defined(__i386) || defined(__i386__) || defined(_M_IX86) || defined(__INTEL__) \
-   || defined(__x86_64) || defined(__x86_64__) || defined(_M_AMD64) || defined(_M_X64)
-#   define sha1_block_data_order		sha1_block_asm_data_order
-#   define DONT_IMPLEMENT_BLOCK_DATA_ORDER
-#  elif defined(__ia64) || defined(__ia64__) || defined(_M_IA64)
-#   define sha1_block_data_order		sha1_block_asm_data_order
-#   define DONT_IMPLEMENT_BLOCK_DATA_ORDER
-#  elif defined(_ARCH_PPC) || defined(_ARCH_PPC64) || \
-	defined(__ppc) || defined(__ppc__) || defined(__powerpc) || \
-	defined(__ppc64) || defined(__ppc64__) || defined(__powerpc64)
-#   define sha1_block_data_order		sha1_block_asm_data_order
-#   define DONT_IMPLEMENT_BLOCK_DATA_ORDER
-#  endif
-# endif
-  void sha1_block_data_order (SHA_CTX *c, const void *p,size_t num);
+#ifndef SHA1_ASM
+static
+#endif
+void sha1_block_data_order (SHA_CTX *c, const void *p,size_t num);
 
 #else
 # error "Either SHA_0 or SHA_1 must be defined."
@@ -216,8 +204,8 @@ int HASH_INIT (SHA_CTX *c)
 # define X(i)	XX[i]
 #endif
 
-#ifndef DONT_IMPLEMENT_BLOCK_DATA_ORDER
-void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, size_t num)
+#if !defined(SHA_1) || !defined(SHA1_ASM)
+static void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, size_t num)
 	{
 	const unsigned char *data=p;
 	register unsigned MD32_REG_T A,B,C,D,E,T,l;
@@ -416,8 +404,8 @@ void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, size_t num)
 	E=D, D=C, C=ROTATE(B,30), B=A;	\
 	A=ROTATE(A,5)+T+xa;	    } while(0)
 
-#ifndef DONT_IMPLEMENT_BLOCK_DATA_ORDER
-void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, size_t num)
+#if !defined(SHA_1) || !defined(SHA1_ASM)
+static void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, size_t num)
 	{
 	const unsigned char *data=p;
 	register unsigned MD32_REG_T A,B,C,D,E,T,l;