Add compile-time rotation functions

The problem with asm rol/ror is the compiler can't schedule effectively.
But we only need asm in the case when the rotation is variable, so distinguish
the two cases. If a compile time constant, then static_assert that the rotation
is in the correct range and do the straightforward expression knowing the compiler
will probably do the right thing. Otherwise do a tricky expression that both
GCC and Clang happen to have recognize. Avoid the reduction case; instead
require that the rotation be in range (this reverts 2b37c13dcf).

Remove the asm rotations (making this branch illnamed), because now both Clang
and GCC will create a roll without any extra help.

Remove the reduction/mask by the word size for the variable case. The compiler
can't optimize that it out well, but it's easy to ensure it is valid in the callers,
especially now that the variable input cases are easy to grep for.

1 files changed, 51 insertions, 51 deletions

diff --git a/src/lib/hash/md5/md5.cpp b/src/lib/hash/md5/md5.cpp
index 174443a67..de75ab2d1 100644
--- a/src/lib/hash/md5/md5.cpp
+++ b/src/lib/hash/md5/md5.cpp
@@ -19,41 +19,41 @@ namespace {
 /*
 * MD5 FF Function
 */
-inline void FF(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t msg,
-               uint8_t S, uint32_t magic)
+template<size_t S>
+inline void FF(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t M)
    {
-   A += (D ^ (B & (C ^ D))) + msg + magic;
-   A  = rotate_left(A, S) + B;
+   A += (D ^ (B & (C ^ D))) + M;
+   A  = rotl<S>(A) + B;
    }
 
 /*
 * MD5 GG Function
 */
-inline void GG(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t msg,
-               uint8_t S, uint32_t magic)
+template<size_t S>
+inline void GG(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t M)
    {
-   A += (C ^ (D & (B ^ C))) + msg + magic;
-   A  = rotate_left(A, S) + B;
+   A += (C ^ (D & (B ^ C))) + M;
+   A  = rotl<S>(A) + B;
    }
 
 /*
 * MD5 HH Function
 */
-inline void HH(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t msg,
-               uint8_t S, uint32_t magic)
+template<size_t S>
+inline void HH(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t M)
    {
-   A += (B ^ C ^ D) + msg + magic;
-   A  = rotate_left(A, S) + B;
+   A += (B ^ C ^ D) + M;
+   A  = rotl<S>(A) + B;
    }
 
 /*
 * MD5 II Function
 */
-inline void II(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t msg,
-               uint8_t S, uint32_t magic)
+template<size_t S>
+inline void II(uint32_t& A, uint32_t B, uint32_t C, uint32_t D, uint32_t M)
    {
-   A += (C ^ (B | ~D)) + msg + magic;
-   A  = rotate_left(A, S) + B;
+   A += (C ^ (B | ~D)) + M;
+   A  = rotl<S>(A) + B;
    }
 
 }
@@ -69,41 +69,41 @@ void MD5::compress_n(const uint8_t input[], size_t blocks)
       {
       load_le(m_M.data(), input, m_M.size());
 
-      FF(A,B,C,D,m_M[ 0], 7,0xD76AA478);   FF(D,A,B,C,m_M[ 1],12,0xE8C7B756);
-      FF(C,D,A,B,m_M[ 2],17,0x242070DB);   FF(B,C,D,A,m_M[ 3],22,0xC1BDCEEE);
-      FF(A,B,C,D,m_M[ 4], 7,0xF57C0FAF);   FF(D,A,B,C,m_M[ 5],12,0x4787C62A);
-      FF(C,D,A,B,m_M[ 6],17,0xA8304613);   FF(B,C,D,A,m_M[ 7],22,0xFD469501);
-      FF(A,B,C,D,m_M[ 8], 7,0x698098D8);   FF(D,A,B,C,m_M[ 9],12,0x8B44F7AF);
-      FF(C,D,A,B,m_M[10],17,0xFFFF5BB1);   FF(B,C,D,A,m_M[11],22,0x895CD7BE);
-      FF(A,B,C,D,m_M[12], 7,0x6B901122);   FF(D,A,B,C,m_M[13],12,0xFD987193);
-      FF(C,D,A,B,m_M[14],17,0xA679438E);   FF(B,C,D,A,m_M[15],22,0x49B40821);
-
-      GG(A,B,C,D,m_M[ 1], 5,0xF61E2562);   GG(D,A,B,C,m_M[ 6], 9,0xC040B340);
-      GG(C,D,A,B,m_M[11],14,0x265E5A51);   GG(B,C,D,A,m_M[ 0],20,0xE9B6C7AA);
-      GG(A,B,C,D,m_M[ 5], 5,0xD62F105D);   GG(D,A,B,C,m_M[10], 9,0x02441453);
-      GG(C,D,A,B,m_M[15],14,0xD8A1E681);   GG(B,C,D,A,m_M[ 4],20,0xE7D3FBC8);
-      GG(A,B,C,D,m_M[ 9], 5,0x21E1CDE6);   GG(D,A,B,C,m_M[14], 9,0xC33707D6);
-      GG(C,D,A,B,m_M[ 3],14,0xF4D50D87);   GG(B,C,D,A,m_M[ 8],20,0x455A14ED);
-      GG(A,B,C,D,m_M[13], 5,0xA9E3E905);   GG(D,A,B,C,m_M[ 2], 9,0xFCEFA3F8);
-      GG(C,D,A,B,m_M[ 7],14,0x676F02D9);   GG(B,C,D,A,m_M[12],20,0x8D2A4C8A);
-
-      HH(A,B,C,D,m_M[ 5], 4,0xFFFA3942);   HH(D,A,B,C,m_M[ 8],11,0x8771F681);
-      HH(C,D,A,B,m_M[11],16,0x6D9D6122);   HH(B,C,D,A,m_M[14],23,0xFDE5380C);
-      HH(A,B,C,D,m_M[ 1], 4,0xA4BEEA44);   HH(D,A,B,C,m_M[ 4],11,0x4BDECFA9);
-      HH(C,D,A,B,m_M[ 7],16,0xF6BB4B60);   HH(B,C,D,A,m_M[10],23,0xBEBFBC70);
-      HH(A,B,C,D,m_M[13], 4,0x289B7EC6);   HH(D,A,B,C,m_M[ 0],11,0xEAA127FA);
-      HH(C,D,A,B,m_M[ 3],16,0xD4EF3085);   HH(B,C,D,A,m_M[ 6],23,0x04881D05);
-      HH(A,B,C,D,m_M[ 9], 4,0xD9D4D039);   HH(D,A,B,C,m_M[12],11,0xE6DB99E5);
-      HH(C,D,A,B,m_M[15],16,0x1FA27CF8);   HH(B,C,D,A,m_M[ 2],23,0xC4AC5665);
-
-      II(A,B,C,D,m_M[ 0], 6,0xF4292244);   II(D,A,B,C,m_M[ 7],10,0x432AFF97);
-      II(C,D,A,B,m_M[14],15,0xAB9423A7);   II(B,C,D,A,m_M[ 5],21,0xFC93A039);
-      II(A,B,C,D,m_M[12], 6,0x655B59C3);   II(D,A,B,C,m_M[ 3],10,0x8F0CCC92);
-      II(C,D,A,B,m_M[10],15,0xFFEFF47D);   II(B,C,D,A,m_M[ 1],21,0x85845DD1);
-      II(A,B,C,D,m_M[ 8], 6,0x6FA87E4F);   II(D,A,B,C,m_M[15],10,0xFE2CE6E0);
-      II(C,D,A,B,m_M[ 6],15,0xA3014314);   II(B,C,D,A,m_M[13],21,0x4E0811A1);
-      II(A,B,C,D,m_M[ 4], 6,0xF7537E82);   II(D,A,B,C,m_M[11],10,0xBD3AF235);
-      II(C,D,A,B,m_M[ 2],15,0x2AD7D2BB);   II(B,C,D,A,m_M[ 9],21,0xEB86D391);
+      FF< 7>(A,B,C,D,m_M[ 0]+0xD76AA478);   FF<12>(D,A,B,C,m_M[ 1]+0xE8C7B756);
+      FF<17>(C,D,A,B,m_M[ 2]+0x242070DB);   FF<22>(B,C,D,A,m_M[ 3]+0xC1BDCEEE);
+      FF< 7>(A,B,C,D,m_M[ 4]+0xF57C0FAF);   FF<12>(D,A,B,C,m_M[ 5]+0x4787C62A);
+      FF<17>(C,D,A,B,m_M[ 6]+0xA8304613);   FF<22>(B,C,D,A,m_M[ 7]+0xFD469501);
+      FF< 7>(A,B,C,D,m_M[ 8]+0x698098D8);   FF<12>(D,A,B,C,m_M[ 9]+0x8B44F7AF);
+      FF<17>(C,D,A,B,m_M[10]+0xFFFF5BB1);   FF<22>(B,C,D,A,m_M[11]+0x895CD7BE);
+      FF< 7>(A,B,C,D,m_M[12]+0x6B901122);   FF<12>(D,A,B,C,m_M[13]+0xFD987193);
+      FF<17>(C,D,A,B,m_M[14]+0xA679438E);   FF<22>(B,C,D,A,m_M[15]+0x49B40821);
+
+      GG< 5>(A,B,C,D,m_M[ 1]+0xF61E2562);   GG< 9>(D,A,B,C,m_M[ 6]+0xC040B340);
+      GG<14>(C,D,A,B,m_M[11]+0x265E5A51);   GG<20>(B,C,D,A,m_M[ 0]+0xE9B6C7AA);
+      GG< 5>(A,B,C,D,m_M[ 5]+0xD62F105D);   GG< 9>(D,A,B,C,m_M[10]+0x02441453);
+      GG<14>(C,D,A,B,m_M[15]+0xD8A1E681);   GG<20>(B,C,D,A,m_M[ 4]+0xE7D3FBC8);
+      GG< 5>(A,B,C,D,m_M[ 9]+0x21E1CDE6);   GG< 9>(D,A,B,C,m_M[14]+0xC33707D6);
+      GG<14>(C,D,A,B,m_M[ 3]+0xF4D50D87);   GG<20>(B,C,D,A,m_M[ 8]+0x455A14ED);
+      GG< 5>(A,B,C,D,m_M[13]+0xA9E3E905);   GG< 9>(D,A,B,C,m_M[ 2]+0xFCEFA3F8);
+      GG<14>(C,D,A,B,m_M[ 7]+0x676F02D9);   GG<20>(B,C,D,A,m_M[12]+0x8D2A4C8A);
+
+      HH< 4>(A,B,C,D,m_M[ 5]+0xFFFA3942);   HH<11>(D,A,B,C,m_M[ 8]+0x8771F681);
+      HH<16>(C,D,A,B,m_M[11]+0x6D9D6122);   HH<23>(B,C,D,A,m_M[14]+0xFDE5380C);
+      HH< 4>(A,B,C,D,m_M[ 1]+0xA4BEEA44);   HH<11>(D,A,B,C,m_M[ 4]+0x4BDECFA9);
+      HH<16>(C,D,A,B,m_M[ 7]+0xF6BB4B60);   HH<23>(B,C,D,A,m_M[10]+0xBEBFBC70);
+      HH< 4>(A,B,C,D,m_M[13]+0x289B7EC6);   HH<11>(D,A,B,C,m_M[ 0]+0xEAA127FA);
+      HH<16>(C,D,A,B,m_M[ 3]+0xD4EF3085);   HH<23>(B,C,D,A,m_M[ 6]+0x04881D05);
+      HH< 4>(A,B,C,D,m_M[ 9]+0xD9D4D039);   HH<11>(D,A,B,C,m_M[12]+0xE6DB99E5);
+      HH<16>(C,D,A,B,m_M[15]+0x1FA27CF8);   HH<23>(B,C,D,A,m_M[ 2]+0xC4AC5665);
+
+      II< 6>(A,B,C,D,m_M[ 0]+0xF4292244);   II<10>(D,A,B,C,m_M[ 7]+0x432AFF97);
+      II<15>(C,D,A,B,m_M[14]+0xAB9423A7);   II<21>(B,C,D,A,m_M[ 5]+0xFC93A039);
+      II< 6>(A,B,C,D,m_M[12]+0x655B59C3);   II<10>(D,A,B,C,m_M[ 3]+0x8F0CCC92);
+      II<15>(C,D,A,B,m_M[10]+0xFFEFF47D);   II<21>(B,C,D,A,m_M[ 1]+0x85845DD1);
+      II< 6>(A,B,C,D,m_M[ 8]+0x6FA87E4F);   II<10>(D,A,B,C,m_M[15]+0xFE2CE6E0);
+      II<15>(C,D,A,B,m_M[ 6]+0xA3014314);   II<21>(B,C,D,A,m_M[13]+0x4E0811A1);
+      II< 6>(A,B,C,D,m_M[ 4]+0xF7537E82);   II<10>(D,A,B,C,m_M[11]+0xBD3AF235);
+      II<15>(C,D,A,B,m_M[ 2]+0x2AD7D2BB);   II<21>(B,C,D,A,m_M[ 9]+0xEB86D391);
 
       A = (m_digest[0] += A);
       B = (m_digest[1] += B);