Make sha512-armv4.pl byte-order neutral.

[openssl.git] / crypto / sha / sha512.c

diff --git a/crypto/sha/sha512.c b/crypto/sha/sha512.c
index 9c06133daba5a5228f6455dc36e5d03f20186f43..654b59a125a8942fbb40778ace35eda70df95142 100644 (file)
--- a/crypto/sha/sha512.c
+++ b/crypto/sha/sha512.c
@@ -61,6 +61,19 @@ const char SHA512_version[]="SHA-512" OPENSSL_VERSION_PTEXT;
 
 int SHA384_Init (SHA512_CTX *c)
        {
+#if defined(SHA512_ASM) && (defined(__arm__) || defined(__arm))
+       /* maintain dword order required by assembler module */
+       unsigned int *h = (unsigned int *)c->h;
+
+       h[0]  = 0xcbbb9d5d; h[1]  = 0xc1059ed8;
+       h[2]  = 0x629a292a; h[3]  = 0x367cd507;
+       h[4]  = 0x9159015a; h[5]  = 0x3070dd17;
+       h[6]  = 0x152fecd8; h[7]  = 0xf70e5939;
+       h[8]  = 0x67332667; h[9]  = 0xffc00b31;
+       h[10] = 0x8eb44a87; h[11] = 0x68581511;
+       h[12] = 0xdb0c2e0d; h[13] = 0x64f98fa7;
+       h[14] = 0x47b5481d; h[15] = 0xbefa4fa4;
+#else
        c->h[0]=U64(0xcbbb9d5dc1059ed8);
        c->h[1]=U64(0x629a292a367cd507);
        c->h[2]=U64(0x9159015a3070dd17);
@@ -69,6 +82,7 @@ int SHA384_Init (SHA512_CTX *c)
        c->h[5]=U64(0x8eb44a8768581511);
        c->h[6]=U64(0xdb0c2e0d64f98fa7);
        c->h[7]=U64(0x47b5481dbefa4fa4);
+#endif
         c->Nl=0;        c->Nh=0;
         c->num=0;       c->md_len=SHA384_DIGEST_LENGTH;
         return 1;
@@ -76,6 +90,19 @@ int SHA384_Init (SHA512_CTX *c)
 
 int SHA512_Init (SHA512_CTX *c)
        {
+#if defined(SHA512_ASM) && (defined(__arm__) || defined(__arm))
+       /* maintain dword order required by assembler module */
+       unsigned int *h = (unsigned int *)c->h;
+
+       h[0]  = 0x6a09e667; h[1]  = 0xf3bcc908;
+       h[2]  = 0xbb67ae85; h[3]  = 0x84caa73b;
+       h[4]  = 0x3c6ef372; h[5]  = 0xfe94f82b;
+       h[6]  = 0xa54ff53a; h[7]  = 0x5f1d36f1;
+       h[8]  = 0x510e527f; h[9]  = 0xade682d1;
+       h[10] = 0x9b05688c; h[11] = 0x2b3e6c1f;
+       h[12] = 0x1f83d9ab; h[13] = 0xfb41bd6b;
+       h[14] = 0x5be0cd19; h[15] = 0x137e2179;
+#else
        c->h[0]=U64(0x6a09e667f3bcc908);
        c->h[1]=U64(0xbb67ae8584caa73b);
        c->h[2]=U64(0x3c6ef372fe94f82b);
@@ -84,6 +111,7 @@ int SHA512_Init (SHA512_CTX *c)
        c->h[5]=U64(0x9b05688c2b3e6c1f);
        c->h[6]=U64(0x1f83d9abfb41bd6b);
        c->h[7]=U64(0x5be0cd19137e2179);
+#endif
         c->Nl=0;        c->Nh=0;
         c->num=0;       c->md_len=SHA512_DIGEST_LENGTH;
         return 1;
@@ -132,6 +160,24 @@ int SHA512_Final (unsigned char *md, SHA512_CTX *c)
 
        if (md==0) return 0;
 
+#if defined(SHA512_ASM) && (defined(__arm__) || defined(__arm))
+       /* recall assembler dword order... */
+       n = c->md_len;
+       if (n == SHA384_DIGEST_LENGTH || n == SHA512_DIGEST_LENGTH)
+               {
+               unsigned int *h = (unsigned int *)c->h, t;
+
+               for (n/=4;n;n--)
+                       {
+                       t = *(h++);
+                       *(md++) = (unsigned char)(t>>24);
+                       *(md++) = (unsigned char)(t>>16);
+                       *(md++) = (unsigned char)(t>>8);
+                       *(md++) = (unsigned char)(t);
+                       }
+               }
+       else    return 0;
+#else
        switch (c->md_len)
                {
                /* Let compiler decide if it's appropriate to unroll... */
@@ -168,7 +214,7 @@ int SHA512_Final (unsigned char *md, SHA512_CTX *c)
                /* ... as well as make sure md_len is not abused. */
                default:        return 0;
                }
-
+#endif
        return 1;
        }
 
@@ -390,64 +436,57 @@ static const SHA_LONG64 K512[80] = {
 
 
 #if defined(__i386) || defined(__i386__) || defined(_M_IX86)
-
-#if defined(OPENSSL_IA32_SSE2) && !defined(OPENSSL_NO_ASM) && !defined(I386_ONLY)
-#define        GO_FOR_SSE2(ctx,in,num)         do {            \
-       void    sha512_block_sse2(void *,const void *,size_t);  \
-       if (!(OPENSSL_ia32cap_P & (1<<26))) break;      \
-       sha512_block_sse2(ctx->h,in,num); return;       \
-                                       } while (0)
-#endif
-
+/*
+ * This code should give better results on 32-bit CPU with less than
+ * ~24 registers, both size and performance wise...
+ */
 static void sha512_block_data_order (SHA512_CTX *ctx, const void *in, size_t num)
        {
        const SHA_LONG64 *W=in;
-       SHA_LONG64      T1;
+       SHA_LONG64      A,E,T;
        SHA_LONG64      X[9+80],*F;
        int i;
 
-#ifdef GO_FOR_SSE2
-       GO_FOR_SSE2(ctx,in,num);
-#endif
-
                        while (num--) {
 
-       F = X+80;
-       F[0] = ctx->h[0];       F[1] = ctx->h[1];
+       F    = X+80;
+       A    = ctx->h[0];       F[1] = ctx->h[1];
        F[2] = ctx->h[2];       F[3] = ctx->h[3];
-       F[4] = ctx->h[4];       F[5] = ctx->h[5];
+       E    = ctx->h[4];       F[5] = ctx->h[5];
        F[6] = ctx->h[6];       F[7] = ctx->h[7];
 
        for (i=0;i<16;i++,F--)
                {
 #ifdef B_ENDIAN
-               T1 = W[i];
+               T = W[i];
 #else
-               T1 = PULL64(W[i]);
+               T = PULL64(W[i]);
 #endif
-               F[8]  = T1;
-               T1   += F[7] + Sigma1(F[4]) + Ch(F[4],F[5],F[6]) + K512[i];
-               F[3] += T1;
-               T1   += Sigma0(F[0]) + Maj(F[0],F[1],F[2]);
-               F[-1] = T1;
+               F[0] = A;
+               F[4] = E;
+               F[8] = T;
+               T   += F[7] + Sigma1(E) + Ch(E,F[5],F[6]) + K512[i];
+               E    = F[3] + T;
+               A    = T + Sigma0(A) + Maj(A,F[1],F[2]);
                }
 
        for (;i<80;i++,F--)
                {
-               T1    = sigma0(F[8+16-1]);
-               T1   += sigma1(F[8+16-14]);
-
-               T1   += F[8+16] + F[8+16-9];
-               F[8]  = T1;
-               T1   += F[7] + Sigma1(F[4]) + Ch(F[4],F[5],F[6]) + K512[i];
-               F[3] += T1;
-               T1   += Sigma0(F[0]) + Maj(F[0],F[1],F[2]);
-               F[-1] = T1;
+               T    = sigma0(F[8+16-1]);
+               T   += sigma1(F[8+16-14]);
+               T   += F[8+16] + F[8+16-9];
+
+               F[0] = A;
+               F[4] = E;
+               F[8] = T;
+               T   += F[7] + Sigma1(E) + Ch(E,F[5],F[6]) + K512[i];
+               E    = F[3] + T;
+               A    = T + Sigma0(A) + Maj(A,F[1],F[2]);
                }
 
-       ctx->h[0] += F[0];      ctx->h[1] += F[1];
+       ctx->h[0] += A;         ctx->h[1] += F[1];
        ctx->h[2] += F[2];      ctx->h[3] += F[3];
-       ctx->h[4] += F[4];      ctx->h[5] += F[5];
+       ctx->h[4] += E;         ctx->h[5] += F[5];
        ctx->h[6] += F[6];      ctx->h[7] += F[7];
 
                        W+=SHA_LBLOCK;