I implemented this when troubleshooting performance problem on SPARC Solaris.

As there is an apparent interest for optimization for footprint, I figured
that this can eventually become useful.

crypto/sha/sha_locl.h

@@ -168,6 +168,8 @@ int HASH_INIT (SHA_CTX *c)
 #define F_40_59(b,c,d)	(((b) & (c)) | (((b)|(c)) & (d))) 
 #define	F_60_79(b,c,d)	F_20_39(b,c,d)
 
+#ifndef OPENSSL_SMALL_FOOTPRINT
+
 #define BODY_00_15(i,a,b,c,d,e,f,xi) \
 	(f)=xi+(e)+K_00_19+ROTATE((a),5)+F_00_19((b),(c),(d)); \
 	(b)=ROTATE((b),30);
@@ -470,3 +472,127 @@ void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, int num)
 		}
 	}
 #endif
+
+#else	/* OPENSSL_SMALL_FOOTPRINT */
+
+#define BODY_00_15(xi)		 do {	\
+	T=E+K_00_19+F_00_19(B,C,D);	\
+	E=D, D=C, C=ROTATE(B,30), B=A;	\
+	A=ROTATE(A,5)+T+xi;	    } while(0)
+
+#define BODY_16_19(xa,xb,xc,xd)	 do {	\
+	Xupdate(T,xa,xa,xb,xc,xd);	\
+	T+=E+K_00_19+F_00_19(B,C,D);	\
+	E=D, D=C, C=ROTATE(B,30), B=A;	\
+	A=ROTATE(A,5)+T;	    } while(0)
+
+#define BODY_20_39(xa,xb,xc,xd)	 do {	\
+	Xupdate(T,xa,xa,xb,xc,xd);	\
+	T+=E+K_20_39+F_20_39(B,C,D);	\
+	E=D, D=C, C=ROTATE(B,30), B=A;	\
+	A=ROTATE(A,5)+T;	    } while(0)
+
+#define BODY_40_59(xa,xb,xc,xd)	 do {	\
+	Xupdate(T,xa,xa,xb,xc,xd);	\
+	T+=E+K_40_59+F_40_59(B,C,D);	\
+	E=D, D=C, C=ROTATE(B,30), B=A;	\
+	A=ROTATE(A,5)+T;	    } while(0)
+
+#define BODY_60_79(xa,xb,xc,xd)	 do {	\
+	Xupdate(T,xa,xa,xb,xc,xd);	\
+	T=E+K_60_79+F_60_79(B,C,D);	\
+	E=D, D=C, C=ROTATE(B,30), B=A;	\
+	A=ROTATE(A,5)+T+xa;	    } while(0)
+
+#ifndef DONT_IMPLEMENT_BLOCK_HOST_ORDER
+void HASH_BLOCK_HOST_ORDER (SHA_CTX *c, const void *d, int num)
+	{
+	const SHA_LONG *W=d;
+	register unsigned MD32_REG_T A,B,C,D,E,T;
+	int i;
+	SHA_LONG	X[16];
+
+	A=c->h0;
+	B=c->h1;
+	C=c->h2;
+	D=c->h3;
+	E=c->h4;
+
+	for (;;)
+		{
+	for (i=0;i<16;i++)
+	{ X[i]=W[i]; BODY_00_15(X[i]); }
+	for (i=0;i<4;i++)
+	{ BODY_16_19(X[i],       X[i+2],      X[i+8],     X[(i+13)&15]); }
+	for (;i<24;i++)
+	{ BODY_20_39(X[i&15],    X[(i+2)&15], X[(i+8)&15],X[(i+13)&15]); }
+	for (i=0;i<20;i++)
+	{ BODY_40_59(X[(i+8)&15],X[(i+10)&15],X[i&15],    X[(i+5)&15]);  }
+	for (i=4;i<24;i++)
+	{ BODY_60_79(X[(i+8)&15],X[(i+10)&15],X[i&15],    X[(i+5)&15]);  }
+	
+	c->h0=(c->h0+A)&0xffffffffL; 
+	c->h1=(c->h1+B)&0xffffffffL;
+	c->h2=(c->h2+C)&0xffffffffL;
+	c->h3=(c->h3+D)&0xffffffffL;
+	c->h4=(c->h4+E)&0xffffffffL;
+
+	if (--num <= 0) break;
+
+	A=c->h0;
+	B=c->h1;
+	C=c->h2;
+	D=c->h3;
+	E=c->h4;
+
+	W+=SHA_LBLOCK;
+		}
+	}
+#endif
+
+#ifndef DONT_IMPLEMENT_BLOCK_DATA_ORDER
+void HASH_BLOCK_DATA_ORDER (SHA_CTX *c, const void *p, int num)
+	{
+	const unsigned char *data=p;
+	register unsigned MD32_REG_T A,B,C,D,E,T,l;
+	int i;
+	SHA_LONG	X[16];
+
+	A=c->h0;
+	B=c->h1;
+	C=c->h2;
+	D=c->h3;
+	E=c->h4;
+
+	for (;;)
+		{
+	for (i=0;i<16;i++)
+	{ HOST_c2l(data,l); X[i]=l; BODY_00_15(X[i]); }
+	for (i=0;i<4;i++)
+	{ BODY_16_19(X[i],       X[i+2],      X[i+8],     X[(i+13)&15]); }
+	for (;i<24;i++)
+	{ BODY_20_39(X[i&15],    X[(i+2)&15], X[(i+8)&15],X[(i+13)&15]); }
+	for (i=0;i<20;i++)
+	{ BODY_40_59(X[(i+8)&15],X[(i+10)&15],X[i&15],    X[(i+5)&15]);  }
+	for (i=4;i<24;i++)
+	{ BODY_60_79(X[(i+8)&15],X[(i+10)&15],X[i&15],    X[(i+5)&15]);  }
+
+	c->h0=(c->h0+A)&0xffffffffL; 
+	c->h1=(c->h1+B)&0xffffffffL;
+	c->h2=(c->h2+C)&0xffffffffL;
+	c->h3=(c->h3+D)&0xffffffffL;
+	c->h4=(c->h4+E)&0xffffffffL;
+
+	if (--num <= 0) break;
+
+	A=c->h0;
+	B=c->h1;
+	C=c->h2;
+	D=c->h3;
+	E=c->h4;
+
+		}
+	}
+#endif
+
+#endif