v3.5.2

Makefile.am

@@ -122,6 +122,7 @@ cpuminer_SOURCES = \
   algo/skein/skein2.c \
   algo/s3.c \
   algo/tiger/sph_tiger.c \
+  algo/timetravel.c \
   algo/veltor.c \
   algo/whirlpool/whirlpool.c \
   algo/whirlpool/whirlpoolx.c \

RELEASE_NOTES

@@ -1,6 +1,10 @@
 Change Log
 ----------
 
+V3.5.2
+
+Timetrave (machinecoin) added and optimized.
+
 v3.5.1
 
 Bastion 9% faster with AES, benchkmark still not working.

algo-gate-api.c

@@ -187,6 +187,7 @@ bool register_algo_gate( int algo, algo_gate_t *gate )
      case ALGO_SKEIN:       register_skein_algo      ( gate ); break;
      case ALGO_SKEIN2:      register_skein2_algo     ( gate ); break;
      case ALGO_S3:          register_s3_algo         ( gate ); break;
+     case ALGO_TIMETRAVEL:  register_timetravel_algo ( gate ); break;
      case ALGO_VANILLA:     register_vanilla_algo    ( gate ); break;
      case ALGO_VELTOR:      register_veltor_algo     ( gate ); break;
      case ALGO_WHIRLPOOL:   register_whirlpool_algo  ( gate ); break;

algo/groestl/aes_ni/hash-groestl.c

@@ -139,6 +139,8 @@ HashReturn_gr reinit_groestl(hashState_groestl* ctx)
   /* set other variables */
   ctx->buf_ptr = 0;
   ctx->block_counter = 0;
+
+// not used
   ctx->bits_in_last_byte = 0;
 
   return SUCCESS_GR;
@@ -149,9 +151,16 @@ HashReturn_gr reinit_groestl(hashState_groestl* ctx)
 HashReturn_gr update_groestl(hashState_groestl* ctx,
 		  const BitSequence_gr* input,
 	  DataLength_gr databitlen) {
+
   int index = 0;
   int msglen = (int)(databitlen/8);
-  int rem = (int)(databitlen%8);
+  int rem = (int)(databitlen%8);  // not used
+
+// The only data length used is either 64 bytes (512 bits,
+// or 80 bytes (640 bits). The sph version of groestl used a byte
+// size for the data length, so odd bits aren't supported there.
+// No need to support them here either, change the arg to bytes
+// for consistency.     
 
   /* non-integral number of message bytes can only be supplied in the
      last call to this function */
@@ -160,6 +169,8 @@ HashReturn_gr update_groestl(hashState_groestl* ctx,
   /* if the buffer contains data that has not yet been digested, first
      add data to buffer until full */
 
+//// This  code can never run, it is indeed dead. buf_ptr is initialized
+//// to 0 in init_groestl and hasn't been changed yet
 // The following block of code never gets hit when hashing x11 or quark
 // leave it here in case it might be needed.
 //  if (ctx->buf_ptr)
@@ -187,6 +198,10 @@ HashReturn_gr update_groestl(hashState_groestl* ctx,
 
   /* digest bulk of message */
   Transform(ctx, input+index, msglen-index);
+
+// index is always zero here, the following line sets it == msglen
+// meaning the next while test will always fail. it's all part of
+// supporting odd bits.
   index += ((msglen-index)/ctx->statesize)*ctx->statesize;
 
   /* store remaining data in buffer */
@@ -194,7 +209,11 @@ HashReturn_gr update_groestl(hashState_groestl* ctx,
   {
     ctx->buffer[(int)ctx->buf_ptr++] = input[index++];
   }
+// buf_ptr should be msglen now.
 
+//// This code isn't quite dead but but would only run if datalen
+/// is not a multiple of 8. As a result bits_in_last_byte is never
+//// modified from its initial zero.
 // Another block that doesn't get used by x11 or quark
 //  /* if non-integral number of bytes have been supplied, store
 //     remaining bits in last byte, together with information about
@@ -223,8 +242,13 @@ HashReturn_gr final_groestl(hashState_groestl* ctx,
     ctx->buffer[(int)ctx->buf_ptr-1] ^= 0x1<<(7-BILB);
     BILB = 0;
   }
+//This sets the first pad byte 
   else ctx->buffer[(int)ctx->buf_ptr++] = 0x80;
 
+//  buf_ptr is left == msglen after update_groestl, 64 (bytes).
+//  It has now been incrememnted to 65. The test below should fail
+//  with 64 and 80 and require 1 pad block. Why does 64 bit need a pad block?
+//  length padding?
   /* pad with '0'-bits */
   if (ctx->buf_ptr > ctx->statesize-LENGTHFIELDLEN) {
     /* padding requires two blocks */
@@ -235,6 +259,19 @@ HashReturn_gr final_groestl(hashState_groestl* ctx,
     Transform(ctx, ctx->buffer, ctx->statesize);
     ctx->buf_ptr = 0;
   }
+
+// the padding can be vectorized, including the first pad byte above
+// 64 bit: buffer[64..79] = {0x80000000,0,0,0}
+//         buffer[80..95] = {0,0,0,0}
+//         buffer[96..111] = {0,0,0,0}
+//         buffer[112..128 = {0,0,length padding}
+// 80 bit: buffer[64..79] = unchanged
+//         buffer[80..95] = {0x800000000,0,0,0}
+//         buffer[96..111] = {0,0,0,0}
+//         buffer[112..128 = {0,0,length padding}
+
+
+// this will pad up to 120 bytes
   while (ctx->buf_ptr < ctx->statesize-LENGTHFIELDLEN) {
     ctx->buffer[(int)ctx->buf_ptr++] = 0;
   }
@@ -257,6 +294,7 @@ HashReturn_gr final_groestl(hashState_groestl* ctx,
     output[j] = s[i];
   }
 
+// the following is redundant as init_groestl will reset to zero.
   /* zeroise relevant variables and deallocate memory */
   
   for (i = 0; i < ctx->columns; i++) {

algo/groestl/aes_ni/hash-groestl.h

@@ -88,6 +88,8 @@ typedef unsigned char BitSequence_gr;
 typedef unsigned long long DataLength_gr;
 typedef enum { SUCCESS_GR = 0, FAIL_GR = 1, BAD_HASHBITLEN_GR = 2} HashReturn_gr;
 
+// Use area128 overlay for buffer to facilitate fast copying
+
 typedef struct {
   __attribute__ ((aligned (32))) u64 chaining[SIZE/8];      /* actual state */
   __attribute__ ((aligned (32))) BitSequence_gr buffer[SIZE];  /* data buffer */

algo/timetravel.c

@@ -0,0 +1,292 @@
+#include <miner.h>
+#include "algo-gate-api.h"
+
+#include <stdlib.h>
+#include <stdint.h>
+#include <string.h>
+#include <stdio.h>
+
+#include "algo/blake/sph_blake.h"
+#include "algo/bmw/sph_bmw.h"
+#include "algo/jh/sph_jh.h"
+#include "algo/keccak/sph_keccak.h"
+#include "algo/skein/sph_skein.h"
+#include "algo/luffa/sph_luffa.h"
+#include "algo/luffa/sse2/luffa_for_sse2.h"
+#include "algo/cubehash/sse2/cubehash_sse2.h"
+
+#ifdef NO_AES_NI
+  #include "algo/groestl/sph_groestl.h"
+#else
+  #include "algo/groestl/aes_ni/hash-groestl.h"
+#endif
+
+// Machinecoin Genesis Timestamp
+#define HASH_FUNC_BASE_TIMESTAMP 1389040865
+
+#define HASH_FUNC_COUNT 8
+#define HASH_FUNC_COUNT_PERMUTATIONS 40320
+
+//static int permutation[HASH_FUNC_COUNT] = { 0 };
+static __thread uint32_t s_ntime = UINT32_MAX;
+static __thread int permutation[HASH_FUNC_COUNT] = { 0 };
+
+inline void reverse( int *pbegin, int *pend )
+{
+   while ( (pbegin != pend) && (pbegin != --pend) )
+   {
+      swap_vars( *pbegin, *pend );
+      pbegin++;
+   }
+}
+
+static void next_permutation( int *pbegin, int *pend )
+{
+   if ( pbegin == pend )
+	return;
+
+   int *i = pbegin;
+   ++i;
+   if ( i == pend )
+	return;
+
+   i = pend;
+   --i;
+
+   while (1)
+   {
+	int *j = i;
+	--i;
+
+	if ( *i < *j )
+        {
+           int *k = pend;
+
+	   while ( !(*i < *--k) ) /* do nothing */ ;
+
+	   swap_vars( *i, *k );
+	   reverse(j, pend);
+		return; // true
+	}
+
+	if ( i == pbegin )
+        {
+	   reverse(pbegin, pend);
+	   return; // false
+	}
+        // else?
+   }
+}
+
+typedef struct {
+        sph_blake512_context    blake;
+        sph_bmw512_context      bmw;
+        sph_skein512_context    skein;
+        sph_jh512_context       jh;
+        sph_keccak512_context   keccak;
+        sph_luffa512_context    luffa;
+//        hashState_luffa         luffa;
+        cubehashParam           cube;
+// ctx optimization doesn't work for groestl, do it the old way
+//#ifdef NO_AES_NI
+//        sph_groestl512_context  groestl;
+//#else
+//        hashState_groestl       groestl;
+//#endif
+} tt_ctx_holder;
+
+tt_ctx_holder tt_ctx;
+
+void init_tt_ctx()
+{
+        sph_blake512_init( &tt_ctx.blake );
+        sph_bmw512_init( &tt_ctx.bmw );
+        sph_skein512_init( &tt_ctx.skein );
+        sph_jh512_init( &tt_ctx.jh );
+        sph_keccak512_init( &tt_ctx.keccak );
+        sph_luffa512_init( &tt_ctx.luffa );
+//        init_luffa( &tt_ctx.luffa, 512 );
+        cubehashInit( &tt_ctx.cube, 512, 16, 32 );
+//#ifdef NO_AES_NI
+//        sph_groestl512_init( &tt_ctx.groestl );
+//#else
+//        init_groestl( &tt_ctx.groestl );
+//#endif
+};
+
+void timetravel_hash(void *output, const void *input)
+{
+   uint32_t _ALIGN(64) hash[128]; // 16 bytes * HASH_FUNC_COUNT
+   uint32_t *hashA, *hashB;
+   uint32_t dataLen = 64;
+   uint32_t *work_data = (uint32_t *)input;
+   const uint32_t timestamp = work_data[17];
+   tt_ctx_holder ctx;
+   memcpy( &ctx, &tt_ctx, sizeof(tt_ctx) );
+   int i;
+
+// workaround for initializing groestl ctx
+#ifdef NO_AES_NI
+   sph_groestl512_context  ctx_groestl;
+#else
+   hashState_groestl       ctx_groestl;
+#endif
+
+   for ( i = 0; i < HASH_FUNC_COUNT; i++ )
+   {
+        if (i == 0)
+        {
+	   dataLen = 80;
+	   hashA = work_data;
+	}
+        else
+        {
+           dataLen = 64;
+	   hashA = &hash[16 * (i - 1)];
+	}
+	hashB = &hash[16 * i];
+
+	switch ( permutation[i] )
+        {
+	   case 0:
+//		sph_blake512_init( &ctx.blake );
+		sph_blake512( &ctx.blake, hashA, dataLen );
+		sph_blake512_close( &ctx.blake, hashB );
+		break;
+	   case 1:
+//		sph_bmw512_init( &ctx.bmw );
+		sph_bmw512( &ctx.bmw, hashA, dataLen );
+		sph_bmw512_close( &ctx.bmw, hashB );
+		break;
+	   case 2:
+#ifdef NO_AES_NI
+		sph_groestl512_init( &ctx_groestl );
+		sph_groestl512( &ctx_groestl, hashA, dataLen );
+		sph_groestl512_close( &ctx_groestl, hashB );
+#else
+                init_groestl( &ctx_groestl );
+                update_groestl( &ctx_groestl, (char*)hashA, dataLen*8 );
+                final_groestl( &ctx_groestl, (char*)hashB );
+#endif
+		break;
+	   case 3:
+//		sph_skein512_init( &ctx.skein );
+		sph_skein512( &ctx.skein, hashA, dataLen );
+		sph_skein512_close( &ctx.skein, hashB );
+		break;
+	   case 4:
+//		sph_jh512_init( &ctx.jh );
+		sph_jh512( &ctx.jh, hashA, dataLen );
+		sph_jh512_close( &ctx.jh, hashB);
+		break;
+	   case 5:
+//		sph_keccak512_init( &ctx.keccak );
+		sph_keccak512( &ctx.keccak, hashA, dataLen );
+		sph_keccak512_close( &ctx.keccak, hashB );
+		break;
+	   case 6:
+//              sph_luffa512_init( &ctx.luffa );
+                sph_luffa512 ( &ctx.luffa, hashA, dataLen );
+                sph_luffa512_close( &ctx.luffa, hashB );
+//                init_luffa( &ctx.luffa, 512 );
+//                update_luffa( &ctx.luffa, (const BitSequence*)hashA, dataLen*8 );
+//                final_luffa( &ctx.luffa, (BitSequence*)hashB );
+		break;
+	   case 7:
+//                cubehashInit( &ctx.cube, 512, 16, 32 );
+                cubehashUpdate( &ctx.cube, (const byte*) hashA, dataLen );
+                cubehashDigest( &ctx.cube, (byte*)hashB );
+		break;
+	   default:
+		break;
+	   }
+	}
+
+	memcpy(output, &hash[16 * (HASH_FUNC_COUNT - 1)], 32);
+}
+
+int scanhash_timetravel( int thr_id, struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done )
+{
+	uint32_t _ALIGN(64) hash[8];
+	uint32_t _ALIGN(64) endiandata[20];
+	uint32_t *pdata = work->data;
+	uint32_t *ptarget = work->target;
+
+	const uint32_t Htarg = ptarget[7];
+	const uint32_t first_nonce = pdata[19];
+	uint32_t nonce = first_nonce;
+	volatile uint8_t *restart = &(work_restart[thr_id].restart);
+        int i;
+
+	if (opt_benchmark)
+		ptarget[7] = 0x0cff;
+
+	for (int k=0; k < 19; k++)
+		be32enc(&endiandata[k], pdata[k]);
+
+        const uint32_t timestamp = endiandata[17];
+        if ( timestamp != s_ntime )
+        {
+           const int steps = ( timestamp - HASH_FUNC_BASE_TIMESTAMP )
+                         % HASH_FUNC_COUNT_PERMUTATIONS;
+           for ( i = 0; i < HASH_FUNC_COUNT; i++ )
+              permutation[i] = i;
+           for ( i = 0; i < steps; i++ )
+              next_permutation( permutation, permutation + HASH_FUNC_COUNT );
+           s_ntime = timestamp;
+        }
+
+	do {
+		be32enc(&endiandata[19], nonce);
+		timetravel_hash(hash, endiandata);
+
+		if (hash[7] <= Htarg && fulltest(hash, ptarget)) {
+			work_set_target_ratio(work, hash);
+			pdata[19] = nonce;
+			*hashes_done = pdata[19] - first_nonce;
+			return 1;
+		}
+		nonce++;
+
+	} while (nonce < max_nonce && !(*restart));
+
+	pdata[19] = nonce;
+	*hashes_done = pdata[19] - first_nonce + 1;
+	return 0;
+}
+
+void timetravel_set_target( struct work* work, double job_diff )
+{
+ work_set_target( work, job_diff / (256.0 * opt_diff_factor) );
+}
+
+// set_data_endian is a reasonable gate to use, it's called upon receipt
+// of new work (new ntime) and has the right arg to access it.
+void timetravel_calc_perm( struct work *work )
+{
+   // We want to permute algorithms. To get started we
+   // initialize an array with a sorted sequence of unique
+   // integers where every integer represents its own algorithm.
+   int ntime, steps, i;
+   be32enc( &ntime, work->data[ STD_NTIME_INDEX ] );
+   steps = ( ntime - HASH_FUNC_BASE_TIMESTAMP )
+                         % HASH_FUNC_COUNT_PERMUTATIONS;
+   for ( i = 0; i < HASH_FUNC_COUNT; i++ ) 
+        permutation[i] = i;
+   for ( i = 0; i < steps; i++ ) 
+        next_permutation( permutation, permutation + HASH_FUNC_COUNT );
+}
+
+bool register_timetravel_algo( algo_gate_t* gate )
+{
+  gate->optimizations = SSE2_OPT | AES_OPT | AVX_OPT | AVX2_OPT;
+  init_tt_ctx();
+  gate->scanhash   = (void*)&scanhash_timetravel;
+  gate->hash       = (void*)&timetravel_hash;
+  gate->set_target = (void*)&timetravel_set_target;
+  gate->get_max64  = (void*)&get_max64_0xffffLL;
+//  gate->set_work_data_endian = (void*)&timetravel_calc_perm;
+  return true;
+};
+

algo/x11/x11evo.c

@@ -76,12 +76,14 @@ uint32_t getCurrentAlgoSeq(uint32_t current_time, uint32_t base_time)
 	return (current_time - base_time) / (60 * 60 * 24);
 }
 
+/*
 void swap( uint8_t *a, uint8_t *b )
 {
 	uint8_t __tmp = *a;
 	*a = *b;
 	*b = __tmp;
 }
+*/
 
 void initPerm( uint8_t n[], uint8_t count )
 {
@@ -102,10 +104,10 @@ int nextPerm( uint8_t n[], uint32_t count )
 
 	if ( tail > 0 )
             for ( j = count - 1; j>tail && n[j] <= n[tail - 1]; j-- );
-	         swap( &n[tail - 1], &n[j] );
+	         swap_vars( n[tail - 1], n[j] );
 
 	for ( i = tail, j = count - 1; i<j; i++, j-- )
-		swap( &n[i], &n[j] );
+		swap_vars( n[i], n[j] );
 
 	return ( tail != 0 );
 }

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([cpuminer-opt], [3.5.1])
+AC_INIT([cpuminer-opt], [3.5.2])
 
 AC_PREREQ([2.59c])
 AC_CANONICAL_SYSTEM

cpu-miner.c

@@ -741,7 +741,7 @@ static int share_result( int result, struct work *work, const char *reason )
    else
         sres = (result ? "Accepted" : "Rejected" );
 
-   // Contrary to convention 100% means zero rejects, exactly 100%. 
+   // Contrary to rounding convention 100% means zero rejects, exactly 100%. 
    // Rates > 99% and < 100% (rejects>0) display 99.9%.
    if ( result )
    {
@@ -763,11 +763,20 @@ static int share_result( int result, struct work *work, const char *reason )
    scale_hash_for_display ( &hashcount, hc_units );
    scale_hash_for_display ( &hashrate, hr_units );
    if ( hc_units[0] )
+   {
       sprintf(hc, "%.2f", hashcount );
+      if ( hashrate < 10 )
+         // very low hashrate, add digits
+         sprintf(hr, "%.4f", hashrate );
       else
+         sprintf(hr, "%.2f", hashrate );
+   }
+   else
+   {
       // no fractions of a hash
       sprintf(hc, "%.0f", hashcount );
       sprintf(hr, "%.2f", hashrate );
+   }
 
 #if ((defined(_WIN64) || defined(__WINDOWS__)))
    applog( LOG_NOTICE, "%s %lu/%lu (%s%%), %s %sH, %s %sH/s",

miner.h

@@ -120,6 +120,9 @@ static inline uint32_t swab32(uint32_t v)
 #endif
 }
 
+// Swap any two variables of the same type without using a temp
+#define swap_vars(a,b) a^=b; b^=a; a^=b;
+
 #ifdef HAVE_SYS_ENDIAN_H
 #include <sys/endian.h>
 #endif
@@ -510,6 +513,7 @@ enum algos {
         ALGO_SKEIN,       
         ALGO_SKEIN2,      
         ALGO_S3,          
+        ALGO_TIMETRAVEL,
         ALGO_VANILLA,
         ALGO_VELTOR,
         ALGO_WHIRLPOOL,
@@ -567,6 +571,7 @@ static const char *algo_names[] = {
         "skein",
         "skein2",
         "s3",
+        "timetravel",
         "vanilla",
         "veltor",
         "whirlpool",
@@ -677,6 +682,7 @@ Options:\n\
                           shavite3     Shavite3\n\
                           skein        Skein+Sha (Skeincoin)\n\
                           skein2       Double Skein (Woodcoin)\n\
+                          timetravel   Machinecoin\n\
                           vanilla      blake256r8vnl (VCash)\n\
                           veltor\n\
                           whirlpool\n\