Move the rounds into separate functions on sparc64 so gcc's optimizer

doesn't blow up.  This is a hack but is better than compiling sha1.c
with -O0 on sparc64.  From NetBSD (mrg).
deraadt@ OK

src/lib/libc/hash/Makefile.inc

@@ -1,4 +1,4 @@
-#	$OpenBSD: Makefile.inc,v 1.12 2002/01/16 19:29:13 marc Exp $
+#	$OpenBSD: Makefile.inc,v 1.13 2003/01/08 19:53:59 millert Exp $
 
 # hash functions
 .PATH: ${LIBCSRCDIR}/hash
@@ -11,16 +11,3 @@ MLINKS+=sha1.3 SHA1Transform.3
 MLINKS+=rmd160.3 RMD160Init.3 rmd160.3 RMD160Update.3 rmd160.3 RMD160Final.3
 MLINKS+=rmd160.3 RMD160End.3  rmd160.3 RMD160File.3   rmd160.3 RMD160Data.3
 MLINKS+=rmd160.3 RMD160Transform.3
-
-.if (${MACHINE_ARCH} == "sparc64")
-sha1.o:
-	${CC} ${CFLAGS} ${CPPFLAGS} -O0 -c ${.IMPSRC}
-
-sha1.po:
-	${CC} ${CFLAGS} ${CPPFLAGS} -O0 -c ${.IMPSRC} -o $@
-sha1.so:
-	${CC} ${CFLAGS} ${CPPFLAGS} -O0 ${PICFLAG} -DPIC -c ${.IMPSRC} -o $@
-sha1.go:
-	${CC} ${CFLAGS} ${CPPFLAGS} -O0 -g -c ${.IMPSRC} -o $@
-
-.endif

src/lib/libc/hash/sha1.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: sha1.c,v 1.10 2002/12/23 04:33:31 millert Exp $	*/
+/*	$OpenBSD: sha1.c,v 1.11 2003/01/08 19:53:59 millert Exp $	*/
 
 /*
  * SHA-1 in C
@@ -15,7 +15,7 @@
  */
 
 #if defined(LIBC_SCCS) && !defined(lint)
-static char rcsid[] = "$OpenBSD: sha1.c,v 1.10 2002/12/23 04:33:31 millert Exp $";
+static char rcsid[] = "$OpenBSD: sha1.c,v 1.11 2003/01/08 19:53:59 millert Exp $";
 #endif /* LIBC_SCCS and not lint */
 
 #define SHA1HANDSOFF		/* Copies data before messing with it. */
@@ -48,6 +48,63 @@ static char rcsid[] = "$OpenBSD: sha1.c,v 1.10 2002/12/23 04:33:31 millert Exp $
 #define R3(v,w,x,y,z,i) z+=(((w|x)&y)|(w&x))+blk(i)+0x8F1BBCDC+rol(v,5);w=rol(w,30);
 #define R4(v,w,x,y,z,i) z+=(w^x^y)+blk(i)+0xCA62C1D6+rol(v,5);w=rol(w,30);
 
+typedef union {
+    u_char c[64];
+    u_int l[16];
+} CHAR64LONG16;
+
+#ifdef __sparc_v9__
+static void do_R01(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *);
+static void do_R2(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *);
+static void do_R3(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *);
+static void do_R4(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *);
+
+#define nR0(v,w,x,y,z,i) R0(*v,*w,*x,*y,*z,i)
+#define nR1(v,w,x,y,z,i) R1(*v,*w,*x,*y,*z,i)
+#define nR2(v,w,x,y,z,i) R2(*v,*w,*x,*y,*z,i)
+#define nR3(v,w,x,y,z,i) R3(*v,*w,*x,*y,*z,i)
+#define nR4(v,w,x,y,z,i) R4(*v,*w,*x,*y,*z,i)
+
+static void
+do_R01(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *block)
+{
+    nR0(a,b,c,d,e, 0); nR0(e,a,b,c,d, 1); nR0(d,e,a,b,c, 2); nR0(c,d,e,a,b, 3);
+    nR0(b,c,d,e,a, 4); nR0(a,b,c,d,e, 5); nR0(e,a,b,c,d, 6); nR0(d,e,a,b,c, 7);
+    nR0(c,d,e,a,b, 8); nR0(b,c,d,e,a, 9); nR0(a,b,c,d,e,10); nR0(e,a,b,c,d,11);
+    nR0(d,e,a,b,c,12); nR0(c,d,e,a,b,13); nR0(b,c,d,e,a,14); nR0(a,b,c,d,e,15);
+    nR1(e,a,b,c,d,16); nR1(d,e,a,b,c,17); nR1(c,d,e,a,b,18); nR1(b,c,d,e,a,19);
+}
+
+static void
+do_R2(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *block)
+{
+    nR2(a,b,c,d,e,20); nR2(e,a,b,c,d,21); nR2(d,e,a,b,c,22); nR2(c,d,e,a,b,23);
+    nR2(b,c,d,e,a,24); nR2(a,b,c,d,e,25); nR2(e,a,b,c,d,26); nR2(d,e,a,b,c,27);
+    nR2(c,d,e,a,b,28); nR2(b,c,d,e,a,29); nR2(a,b,c,d,e,30); nR2(e,a,b,c,d,31);
+    nR2(d,e,a,b,c,32); nR2(c,d,e,a,b,33); nR2(b,c,d,e,a,34); nR2(a,b,c,d,e,35);
+    nR2(e,a,b,c,d,36); nR2(d,e,a,b,c,37); nR2(c,d,e,a,b,38); nR2(b,c,d,e,a,39);
+}
+
+static void
+do_R3(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *block)
+{
+    nR3(a,b,c,d,e,40); nR3(e,a,b,c,d,41); nR3(d,e,a,b,c,42); nR3(c,d,e,a,b,43);
+    nR3(b,c,d,e,a,44); nR3(a,b,c,d,e,45); nR3(e,a,b,c,d,46); nR3(d,e,a,b,c,47);
+    nR3(c,d,e,a,b,48); nR3(b,c,d,e,a,49); nR3(a,b,c,d,e,50); nR3(e,a,b,c,d,51);
+    nR3(d,e,a,b,c,52); nR3(c,d,e,a,b,53); nR3(b,c,d,e,a,54); nR3(a,b,c,d,e,55);
+    nR3(e,a,b,c,d,56); nR3(d,e,a,b,c,57); nR3(c,d,e,a,b,58); nR3(b,c,d,e,a,59);
+}
+
+static void
+do_R4(u_int32_t *a, u_int32_t *b, u_int32_t *c, u_int32_t *d, u_int32_t *e, CHAR64LONG16 *block)
+{
+    nR4(a,b,c,d,e,60); nR4(e,a,b,c,d,61); nR4(d,e,a,b,c,62); nR4(c,d,e,a,b,63);
+    nR4(b,c,d,e,a,64); nR4(a,b,c,d,e,65); nR4(e,a,b,c,d,66); nR4(d,e,a,b,c,67);
+    nR4(c,d,e,a,b,68); nR4(b,c,d,e,a,69); nR4(a,b,c,d,e,70); nR4(e,a,b,c,d,71);
+    nR4(d,e,a,b,c,72); nR4(c,d,e,a,b,73); nR4(b,c,d,e,a,74); nR4(a,b,c,d,e,75);
+    nR4(e,a,b,c,d,76); nR4(d,e,a,b,c,77); nR4(c,d,e,a,b,78); nR4(b,c,d,e,a,79);
+}
+#endif
 
 /*
  * Hash a single 512-bit block. This is the core of the algorithm.
@@ -56,15 +113,11 @@ void
 SHA1Transform(u_int32_t state[5], const u_char buffer[64])
 {
     u_int32_t a, b, c, d, e;
-    typedef union {
-	u_char c[64];
-	u_int l[16];
-    } CHAR64LONG16;
     CHAR64LONG16 *block;
 
 #ifdef SHA1HANDSOFF
-    static u_char workspace[64];
-    block = (CHAR64LONG16 *)workspace;
+    static CHAR64LONG16 workspace;
+    block = &workspace;
     (void)memcpy(block, buffer, 64);
 #else
     block = (CHAR64LONG16 *)buffer;
@@ -77,6 +130,12 @@ SHA1Transform(u_int32_t state[5], const u_char buffer[64])
     d = state[3];
     e = state[4];
 
+#ifdef __sparc_v9__
+    do_R01(&a, &b, &c, &d, &e, block);
+    do_R2(&a, &b, &c, &d, &e, block);
+    do_R3(&a, &b, &c, &d, &e, block);
+    do_R4(&a, &b, &c, &d, &e, block);
+#else
     /* 4 rounds of 20 operations each. Loop unrolled. */
     R0(a,b,c,d,e, 0); R0(e,a,b,c,d, 1); R0(d,e,a,b,c, 2); R0(c,d,e,a,b, 3);
     R0(b,c,d,e,a, 4); R0(a,b,c,d,e, 5); R0(e,a,b,c,d, 6); R0(d,e,a,b,c, 7);
@@ -98,6 +157,7 @@ SHA1Transform(u_int32_t state[5], const u_char buffer[64])
     R4(c,d,e,a,b,68); R4(b,c,d,e,a,69); R4(a,b,c,d,e,70); R4(e,a,b,c,d,71);
     R4(d,e,a,b,c,72); R4(c,d,e,a,b,73); R4(b,c,d,e,a,74); R4(a,b,c,d,e,75);
     R4(e,a,b,c,d,76); R4(d,e,a,b,c,77); R4(c,d,e,a,b,78); R4(b,c,d,e,a,79);
+#endif
 
     /* Add the working vars back into context.state[] */
     state[0] += a;