[PATCH] PPC assembly implementation of SHA1
authorPaul Mackerras <paulus@samba.org>
Sat, 23 Apr 2005 06:08:43 +0000 (23:08 -0700)
committerLinus Torvalds <torvalds@ppc970.osdl.org>
Sat, 23 Apr 2005 06:08:43 +0000 (23:08 -0700)
Here is a SHA1 implementation with the core written in PPC assembly.
On my 2GHz G5, it does 218MB/s, compared to 135MB/s for the openssl
version or 45MB/s for the mozilla version.

Signed-off-by: Paul Mackerras <paulus@samba.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Makefile
ppc/sha1.c [new file with mode: 0644]
ppc/sha1.h [new file with mode: 0644]
ppc/sha1ppc.S [new file with mode: 0644]

index 9573a11..57e7023 100644 (file)
--- a/Makefile
+++ b/Makefile
@@ -34,9 +34,14 @@ ifdef MOZILLA_SHA1
   SHA1_HEADER="mozilla-sha1/sha1.h"
   LIB_OBJS += mozilla-sha1/sha1.o
 else
+ifdef PPC_SHA1
+  SHA1_HEADER="ppc/sha1.h"
+  LIB_OBJS += ppc/sha1.o ppc/sha1ppc.o
+else
   SHA1_HEADER=<openssl/sha.h>
   LIBS += -lssl
 endif
+endif
 
 CFLAGS += '-DSHA1_HEADER=$(SHA1_HEADER)'
 
@@ -77,7 +82,7 @@ unpack-file.o: $(LIB_H)
 write-tree.o: $(LIB_H)
 
 clean:
-       rm -f *.o mozilla-sha1/*.o $(PROG) $(LIB_FILE)
+       rm -f *.o mozilla-sha1/*.o ppc/*.o $(PROG) $(LIB_FILE)
 
 backup: clean
        cd .. ; tar czvf dircache.tar.gz dir-cache
diff --git a/ppc/sha1.c b/ppc/sha1.c
new file mode 100644 (file)
index 0000000..5ba4fc5
--- /dev/null
@@ -0,0 +1,72 @@
+/*
+ * SHA-1 implementation.
+ *
+ * Copyright (C) 2005 Paul Mackerras <paulus@samba.org>
+ *
+ * This version assumes we are running on a big-endian machine.
+ * It calls an external sha1_core() to process blocks of 64 bytes.
+ */
+#include <stdio.h>
+#include <string.h>
+#include "sha1.h"
+
+extern void sha1_core(uint32_t *hash, const unsigned char *p,
+                     unsigned int nblocks);
+
+int SHA1_Init(SHA_CTX *c)
+{
+       c->hash[0] = 0x67452301;
+       c->hash[1] = 0xEFCDAB89;
+       c->hash[2] = 0x98BADCFE;
+       c->hash[3] = 0x10325476;
+       c->hash[4] = 0xC3D2E1F0;
+       c->len = 0;
+       c->cnt = 0;
+       return 0;
+}
+
+int SHA1_Update(SHA_CTX *c, const void *ptr, unsigned long n)
+{
+       unsigned long nb;
+       const unsigned char *p = ptr;
+
+       c->len += n << 3;
+       while (n != 0) {
+               if (c->cnt || n < 64) {
+                       nb = 64 - c->cnt;
+                       if (nb > n)
+                               nb = n;
+                       memcpy(&c->buf.b[c->cnt], p, nb);
+                       if ((c->cnt += nb) == 64) {
+                               sha1_core(c->hash, c->buf.b, 1);
+                               c->cnt = 0;
+                       }
+               } else {
+                       nb = n >> 6;
+                       sha1_core(c->hash, p, nb);
+                       nb <<= 6;
+               }
+               n -= nb;
+               p += nb;
+       }
+       return 0;
+}      
+
+int SHA1_Final(unsigned char *hash, SHA_CTX *c)
+{
+       unsigned int cnt = c->cnt;
+
+       c->buf.b[cnt++] = 0x80;
+       if (cnt > 56) {
+               if (cnt < 64)
+                       memset(&c->buf.b[cnt], 0, 64 - cnt);
+               sha1_core(c->hash, c->buf.b, 1);
+               cnt = 0;
+       }
+       if (cnt < 56)
+               memset(&c->buf.b[cnt], 0, 56 - cnt);
+       c->buf.l[7] = c->len;
+       sha1_core(c->hash, c->buf.b, 1);
+       memcpy(hash, c->hash, 20);
+       return 0;
+}
diff --git a/ppc/sha1.h b/ppc/sha1.h
new file mode 100644 (file)
index 0000000..c3c51aa
--- /dev/null
@@ -0,0 +1,20 @@
+/*
+ * SHA-1 implementation.
+ *
+ * Copyright (C) 2005 Paul Mackerras <paulus@samba.org>
+ */
+#include <stdint.h>
+
+typedef struct sha_context {
+       uint32_t hash[5];
+       uint32_t cnt;
+       uint64_t len;
+       union {
+               unsigned char b[64];
+               uint64_t l[8];
+       } buf;
+} SHA_CTX;
+
+int SHA1_Init(SHA_CTX *c);
+int SHA1_Update(SHA_CTX *c, const void *p, unsigned long n);
+int SHA1_Final(unsigned char *hash, SHA_CTX *c);
diff --git a/ppc/sha1ppc.S b/ppc/sha1ppc.S
new file mode 100644 (file)
index 0000000..e85611a
--- /dev/null
@@ -0,0 +1,185 @@
+/*
+ * SHA-1 implementation for PowerPC.
+ *
+ * Copyright (C) 2005 Paul Mackerras <paulus@samba.org>
+ */
+#define FS     80
+
+/*
+ * We roll the registers for T, A, B, C, D, E around on each
+ * iteration; T on iteration t is A on iteration t+1, and so on.
+ * We use registers 7 - 12 for this.
+ */
+#define RT(t)  ((((t)+5)%6)+7)
+#define RA(t)  ((((t)+4)%6)+7)
+#define RB(t)  ((((t)+3)%6)+7)
+#define RC(t)  ((((t)+2)%6)+7)
+#define RD(t)  ((((t)+1)%6)+7)
+#define RE(t)  ((((t)+0)%6)+7)
+
+/* We use registers 16 - 31 for the W values */
+#define W(t)   (((t)%16)+16)
+
+#define STEPD0(t)                              \
+       and     %r6,RB(t),RC(t);                \
+       andc    %r0,RD(t),RB(t);                \
+       rotlwi  RT(t),RA(t),5;                  \
+       rotlwi  RB(t),RB(t),30;                 \
+       or      %r6,%r6,%r0;                    \
+       add     %r0,RE(t),%r15;                 \
+       add     RT(t),RT(t),%r6;                \
+       add     %r0,%r0,W(t);                   \
+       add     RT(t),RT(t),%r0
+
+#define STEPD1(t)                              \
+       xor     %r6,RB(t),RC(t);                \
+       rotlwi  RT(t),RA(t),5;                  \
+       rotlwi  RB(t),RB(t),30;                 \
+       xor     %r6,%r6,RD(t);                  \
+       add     %r0,RE(t),%r15;                 \
+       add     RT(t),RT(t),%r6;                \
+       add     %r0,%r0,W(t);                   \
+       add     RT(t),RT(t),%r0
+
+#define STEPD2(t)                              \
+       and     %r6,RB(t),RC(t);                \
+       and     %r0,RB(t),RD(t);                \
+       rotlwi  RT(t),RA(t),5;                  \
+       rotlwi  RB(t),RB(t),30;                 \
+       or      %r6,%r6,%r0;                    \
+       and     %r0,RC(t),RD(t);                \
+       or      %r6,%r6,%r0;                    \
+       add     %r0,RE(t),%r15;                 \
+       add     RT(t),RT(t),%r6;                \
+       add     %r0,%r0,W(t);                   \
+       add     RT(t),RT(t),%r0
+
+#define LOADW(t)                               \
+       lwz     W(t),(t)*4(%r4)
+
+#define UPDATEW(t)                             \
+       xor     %r0,W((t)-3),W((t)-8);          \
+       xor     W(t),W((t)-16),W((t)-14);       \
+       xor     W(t),W(t),%r0;                  \
+       rotlwi  W(t),W(t),1
+
+#define STEP0LD4(t)                            \
+       STEPD0(t);   LOADW((t)+4);              \
+       STEPD0((t)+1); LOADW((t)+5);            \
+       STEPD0((t)+2); LOADW((t)+6);            \
+       STEPD0((t)+3); LOADW((t)+7)
+
+#define STEPUP4(t, fn)                         \
+       STEP##fn(t);   UPDATEW((t)+4);          \
+       STEP##fn((t)+1); UPDATEW((t)+5);        \
+       STEP##fn((t)+2); UPDATEW((t)+6);        \
+       STEP##fn((t)+3); UPDATEW((t)+7)
+
+#define STEPUP20(t, fn)                                \
+       STEPUP4(t, fn);                         \
+       STEPUP4((t)+4, fn);                     \
+       STEPUP4((t)+8, fn);                     \
+       STEPUP4((t)+12, fn);                    \
+       STEPUP4((t)+16, fn)
+
+       .globl  sha1_core
+sha1_core:
+       stwu    %r1,-FS(%r1)
+       stw     %r15,FS-68(%r1)
+       stw     %r16,FS-64(%r1)
+       stw     %r17,FS-60(%r1)
+       stw     %r18,FS-56(%r1)
+       stw     %r19,FS-52(%r1)
+       stw     %r20,FS-48(%r1)
+       stw     %r21,FS-44(%r1)
+       stw     %r22,FS-40(%r1)
+       stw     %r23,FS-36(%r1)
+       stw     %r24,FS-32(%r1)
+       stw     %r25,FS-28(%r1)
+       stw     %r26,FS-24(%r1)
+       stw     %r27,FS-20(%r1)
+       stw     %r28,FS-16(%r1)
+       stw     %r29,FS-12(%r1)
+       stw     %r30,FS-8(%r1)
+       stw     %r31,FS-4(%r1)
+
+       /* Load up A - E */
+       lwz     RA(0),0(%r3)    /* A */
+       lwz     RB(0),4(%r3)    /* B */
+       lwz     RC(0),8(%r3)    /* C */
+       lwz     RD(0),12(%r3)   /* D */
+       lwz     RE(0),16(%r3)   /* E */
+
+       mtctr   %r5
+
+1:     LOADW(0)
+       LOADW(1)
+       LOADW(2)
+       LOADW(3)
+
+       lis     %r15,0x5a82     /* K0-19 */
+       ori     %r15,%r15,0x7999
+       STEP0LD4(0)
+       STEP0LD4(4)
+       STEP0LD4(8)
+       STEPUP4(12, D0)
+       STEPUP4(16, D0)
+
+       lis     %r15,0x6ed9     /* K20-39 */
+       ori     %r15,%r15,0xeba1
+       STEPUP20(20, D1)
+
+       lis     %r15,0x8f1b     /* K40-59 */
+       ori     %r15,%r15,0xbcdc
+       STEPUP20(40, D2)
+
+       lis     %r15,0xca62     /* K60-79 */
+       ori     %r15,%r15,0xc1d6
+       STEPUP4(60, D1)
+       STEPUP4(64, D1)
+       STEPUP4(68, D1)
+       STEPUP4(72, D1)
+       STEPD1(76)
+       STEPD1(77)
+       STEPD1(78)
+       STEPD1(79)
+
+       lwz     %r20,16(%r3)
+       lwz     %r19,12(%r3)
+       lwz     %r18,8(%r3)
+       lwz     %r17,4(%r3)
+       lwz     %r16,0(%r3)
+       add     %r20,RE(80),%r20
+       add     RD(0),RD(80),%r19
+       add     RC(0),RC(80),%r18
+       add     RB(0),RB(80),%r17
+       add     RA(0),RA(80),%r16
+       mr      RE(0),%r20
+       stw     RA(0),0(%r3)
+       stw     RB(0),4(%r3)
+       stw     RC(0),8(%r3)
+       stw     RD(0),12(%r3)
+       stw     RE(0),16(%r3)
+
+       addi    %r4,%r4,64
+       bdnz    1b
+
+       lwz     %r15,FS-68(%r1)
+       lwz     %r16,FS-64(%r1)
+       lwz     %r17,FS-60(%r1)
+       lwz     %r18,FS-56(%r1)
+       lwz     %r19,FS-52(%r1)
+       lwz     %r20,FS-48(%r1)
+       lwz     %r21,FS-44(%r1)
+       lwz     %r22,FS-40(%r1)
+       lwz     %r23,FS-36(%r1)
+       lwz     %r24,FS-32(%r1)
+       lwz     %r25,FS-28(%r1)
+       lwz     %r26,FS-24(%r1)
+       lwz     %r27,FS-20(%r1)
+       lwz     %r28,FS-16(%r1)
+       lwz     %r29,FS-12(%r1)
+       lwz     %r30,FS-8(%r1)
+       lwz     %r31,FS-4(%r1)
+       addi    %r1,%r1,FS
+       blr