$NetBSD: patch-ab,v 1.7 2005/05/21 15:30:06 wiz Exp $

--- sha1.cpp.orig	2005-05-01 17:08:48.000000000 +0200
+++ sha1.cpp
@@ -44,17 +44,71 @@ A million repetitions of "a"
 #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];
+    uint32 l[16];
+} CHAR64LONG16;
+
+#ifdef __sparc_v9__
+void do_R01(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *e, CHAR64LONG16 *);
+void do_R2(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *e, CHAR64LONG16 *);
+void do_R3(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *e, CHAR64LONG16 *);
+void do_R4(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *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)
+
+void
+do_R01(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *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);
+}
+
+void
+do_R2(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *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);
+}
+
+void
+do_R3(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *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);
+}
+
+void
+do_R4(uint32 *a, uint32 *b, uint32 *c, uint32 *d, uint32 *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. */
 
 void SHA1Transform(uint32 state[5], unsigned char buffer[64])
 {
   uint32 a, b, c, d, e;
-  typedef union {
-    unsigned char c[64];
-    uint32 l[16];
-} CHAR64LONG16;
-CHAR64LONG16* block;
+  CHAR64LONG16 *block;
 #ifdef SHA1HANDSOFF
 static unsigned char workspace[64];
     block = (CHAR64LONG16*)workspace;
@@ -101,6 +155,12 @@ static unsigned char workspace[64];
     c = state[2];
     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);
@@ -122,6 +182,7 @@ static unsigned char workspace[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;
     state[1] += b;