Split SHA1Transform() back into smaller pieces when compiling on landisk,

to defeat gcc optimization bugs. Similar to what had been done on sparc64
in the past.
ok millert@ deraadt@

src/lib/libc/hash/sha1.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: sha1.c,v 1.20 2005/08/08 08:05:35 espie Exp $	*/
+/*	$OpenBSD: sha1.c,v 1.21 2008/07/29 19:32:50 miod Exp $	*/
 
 /*
  * SHA-1 in C
@@ -42,6 +42,64 @@
 #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_int8_t c[64];
+	u_int32_t l[16];
+} CHAR64LONG16;
+
+#ifdef __sh__
+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.
  */
@@ -50,10 +108,6 @@ SHA1Transform(u_int32_t state[5], const u_int8_t buffer[SHA1_BLOCK_LENGTH])
 {
 	u_int32_t a, b, c, d, e;
 	u_int8_t workspace[SHA1_BLOCK_LENGTH];
-	typedef union {
-		u_int8_t c[64];
-		u_int32_t l[16];
-	} CHAR64LONG16;
 	CHAR64LONG16 *block = (CHAR64LONG16 *)workspace;
 
 	(void)memcpy(block, buffer, SHA1_BLOCK_LENGTH);
@@ -65,6 +119,12 @@ SHA1Transform(u_int32_t state[5], const u_int8_t buffer[SHA1_BLOCK_LENGTH])
 	d = state[3];
 	e = state[4];
 
+#ifdef __sh__
+	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);
@@ -86,6 +146,7 @@ SHA1Transform(u_int32_t state[5], const u_int8_t buffer[SHA1_BLOCK_LENGTH])
 	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;