Interface nicehash djeZo equihash solver (squashed)

Todo:
    - send block height via stratum protocol (encoded in jobid?)
    - remove equi/blake2 cpu algorithm to use common one

the extranonce imcompatibility is related to the solver nonce data,
offsets may be reversed in nheqminer, to check...

The solver was adapted for SM 3.0+ support (no perf changes)

Note: The solver was not improved on purpose, to be able compare
the two miners performances (nheqminer 0.5c the last open sourced, and ccminer)

Signed-off-by: Tanguy Pruvot <tanguy.pruvot@gmail.com>

stratum: code cleanup, move equi fns in equi folder

Makefile.am

@@ -21,6 +21,8 @@ ccminer_SOURCES	= elist.h miner.h compat.h \
 			  crc32.c hefty1.c \
 			  ccminer.cpp pools.cpp util.cpp bench.cpp bignum.cpp \
 			  api.cpp hashlog.cpp nvml.cpp stats.cpp sysinfos.cpp cuda.cpp \
+			  equi/equi-stratum.cpp equi/equi.cpp equi/blake2/blake2bx.cpp \
+			  equi/equihash.cpp equi/cuda_equi.cu \
 			  heavy/heavy.cu \
 			  heavy/cuda_blake512.cu heavy/cuda_blake512.h \
 			  heavy/cuda_combine.cu heavy/cuda_combine.h \
@@ -97,6 +99,10 @@ ccminer_LDFLAGS += -L/usr/local/llvm/lib
 ccminer_LDADD += -lomp
 endif
 
+#ccminer_CPPFLAGS += -DUSE_LIBSODIUM
+#ccminer_LDFLAGS += -Lequi/lib
+#ccminer_LDADD += -lsodium
+ccminer_LDADD += -lcuda
 
 nvcc_ARCH  = -gencode=arch=compute_50,code=\"sm_50,compute_50\"
 

algos.h

@@ -16,6 +16,7 @@ enum sha_algos {
 	ALGO_DEEP,
 	ALGO_DECRED,
 	ALGO_DMD_GR,
+	ALGO_EQUIHASH,
 	ALGO_FRESH,
 	ALGO_FUGUE256,		/* Fugue256 */
 	ALGO_GROESTL,
@@ -78,6 +79,7 @@ static const char *algo_names[] = {
 	"deep",
 	"decred",
 	"dmd-gr",
+	"equihash",
 	"fresh",
 	"fugue256",
 	"groestl",
@@ -149,6 +151,8 @@ static inline int algo_to_int(char* arg)
 			i = ALGO_C11;
 		else if (!strcasecmp("diamond", arg))
 			i = ALGO_DMD_GR;
+		else if (!strcasecmp("equi", arg))
+			i = ALGO_EQUIHASH;
 		else if (!strcasecmp("doom", arg))
 			i = ALGO_LUFFA;
 		else if (!strcasecmp("hmq17", arg))

api.cpp

@@ -1341,6 +1341,7 @@ void api_set_throughput(int thr_id, uint32_t throughput)
 	if (thr_id < MAX_GPUS && thr_info) {
 		struct cgpu_info *cgpu = &thr_info[thr_id].gpu;
 		cgpu->intensity = throughput2intensity(throughput);
+		if (cgpu->throughput != throughput) cgpu->throughput = throughput;
 	}
 	// to display in bench results
 	if (opt_benchmark)

ccminer.cpp

@@ -44,6 +44,7 @@
 #include "algos.h"
 #include "sia/sia-rpc.h"
 #include "crypto/xmr-rpc.h"
+#include "equi/equihash.h"
 
 #include <cuda_runtime.h>
 
@@ -243,6 +244,7 @@ Options:\n\
 			c11/flax    X11 variant\n\
 			decred      Decred Blake256\n\
 			deep        Deepcoin\n\
+			equihash    Zcash Equihash\n\
 			dmd-gr      Diamond-Groestl\n\
 			fresh       Freshcoin (shavite 80)\n\
 			fugue256    Fuguecoin\n\
@@ -558,6 +560,14 @@ void get_currentalgo(char* buf, int sz)
 	snprintf(buf, sz, "%s", algo_names[opt_algo]);
 }
 
+void format_hashrate(double hashrate, char *output)
+{
+	if (opt_algo == ALGO_EQUIHASH)
+		format_hashrate_unit(hashrate, output, "Sol/s");
+	else
+		format_hashrate_unit(hashrate, output, "H/s");
+}
+
 /**
  * Exit app
  */
@@ -634,6 +644,10 @@ static void calc_network_diff(struct work *work)
 	if (opt_algo == ALGO_LBRY) nbits = swab32(work->data[26]);
 	if (opt_algo == ALGO_DECRED) nbits = work->data[29];
 	if (opt_algo == ALGO_SIA) nbits = work->data[11]; // unsure if correct
+	if (opt_algo == ALGO_EQUIHASH) {
+		net_diff = equi_network_diff(work);
+		return;
+	}
 
 	uint32_t bits = (nbits & 0xffffff);
 	int16_t shift = (swab32(nbits) & 0xff); // 0x1c = 28
@@ -856,6 +870,17 @@ static bool submit_upstream_work(CURL *curl, struct work *work)
 		return true;
 	}
 
+	if (pool->type & POOL_STRATUM && stratum.is_equihash) {
+		struct work submit_work;
+		memcpy(&submit_work, work, sizeof(struct work));
+		//if (!hashlog_already_submittted(submit_work.job_id, submit_work.nonces[idnonce])) {
+			if (equi_stratum_submit(pool, &submit_work))
+				hashlog_remember_submit(&submit_work, submit_work.nonces[idnonce]);
+			stratum.job.shares_count++;
+		//}
+		return true;
+	}
+
 	/* discard if a newer block was received */
 	stale_work = work->height && work->height < g_work.height;
 	if (have_stratum && !stale_work && opt_algo != ALGO_ZR5 && opt_algo != ALGO_SCRYPT_JANE) {
@@ -1508,6 +1533,7 @@ static bool stratum_gen_work(struct stratum_ctx *sctx, struct work *work)
 	/* Generate merkle root */
 	switch (opt_algo) {
 		case ALGO_DECRED:
+		case ALGO_EQUIHASH:
 		case ALGO_SIA:
 			// getwork over stratum, no merkle to generate
 			break;
@@ -1566,6 +1592,13 @@ static bool stratum_gen_work(struct stratum_ctx *sctx, struct work *work)
 		memcpy(&work->data[44], &sctx->job.coinbase[sctx->job.coinbase_size-4], 4);
 		sctx->job.height = work->data[32];
 		//applog_hex(work->data, 180);
+	} else if (opt_algo == ALGO_EQUIHASH) {
+		memcpy(&work->data[9], sctx->job.coinbase, 32+32); // merkle [9..16] + reserved
+		work->data[25] = le32dec(sctx->job.ntime);
+		work->data[26] = le32dec(sctx->job.nbits);
+		memcpy(&work->data[27], sctx->xnonce1, sctx->xnonce1_size & 0x1F); // pool extranonce
+		work->data[35] = 0x80;
+		//applog_hex(work->data, 140);
 	} else if (opt_algo == ALGO_LBRY) {
 		for (i = 0; i < 8; i++)
 			work->data[9 + i] = be32dec((uint32_t *)merkle_root + i);
@@ -1618,7 +1651,7 @@ static bool stratum_gen_work(struct stratum_ctx *sctx, struct work *work)
 
 	pthread_mutex_unlock(&stratum_work_lock);
 
-	if (opt_debug && opt_algo != ALGO_DECRED && opt_algo != ALGO_SIA) {
+	if (opt_debug && opt_algo != ALGO_DECRED && opt_algo != ALGO_EQUIHASH && opt_algo != ALGO_SIA) {
 		uint32_t utm = work->data[17];
 		if (opt_algo != ALGO_ZR5) utm = swab32(utm);
 		char *tm = atime2str(utm - sctx->srvtime_diff);
@@ -1656,6 +1689,9 @@ static bool stratum_gen_work(struct stratum_ctx *sctx, struct work *work)
 		case ALGO_LYRA2:
 			work_set_target(work, sctx->job.diff / (128.0 * opt_difficulty));
 			break;
+		case ALGO_EQUIHASH:
+			equi_work_set_target(work, sctx->job.diff / opt_difficulty);
+			break;
 		default:
 			work_set_target(work, sctx->job.diff / opt_difficulty);
 	}
@@ -1830,6 +1866,9 @@ static void *miner_thread(void *userdata)
 		} else if (opt_algo == ALGO_CRYPTOLIGHT || opt_algo == ALGO_CRYPTONIGHT) {
 			nonceptr = (uint32_t*) (((char*)work.data) + 39);
 			wcmplen = 39;
+		} else if (opt_algo == ALGO_EQUIHASH) {
+			nonceptr = &work.data[EQNONCE_OFFSET]; // 27 is pool extranonce (256bits nonce space)
+			wcmplen = 4+32+32;
 		}
 
 		if (have_stratum) {
@@ -1962,6 +2001,10 @@ static void *miner_thread(void *userdata)
 			nonceptr[1] += 1;
 			nonceptr[2] |= thr_id;
 
+		} else if (opt_algo == ALGO_EQUIHASH) {
+			nonceptr[1]++;
+			nonceptr[1] |= thr_id << 24;
+			//applog_hex(&work.data[27], 32);
 		} else if (opt_algo == ALGO_WILDKECCAK) {
 			//nonceptr[1] += 1;
 		} else if (opt_algo == ALGO_SIA) {
@@ -2264,6 +2307,9 @@ static void *miner_thread(void *userdata)
 		case ALGO_DEEP:
 			rc = scanhash_deep(thr_id, &work, max_nonce, &hashes_done);
 			break;
+		case ALGO_EQUIHASH:
+			rc = scanhash_equihash(thr_id, &work, max_nonce, &hashes_done);
+			break;
 		case ALGO_FRESH:
 			rc = scanhash_fresh(thr_id, &work, max_nonce, &hashes_done);
 			break;
@@ -2431,7 +2477,7 @@ static void *miner_thread(void *userdata)
 				work.nonces[1] = nonceptr[2];
 		}
 
-		if (stratum.rpc2 && rc == -EBUSY || work_restart[thr_id].restart) {
+		if (stratum.rpc2 && (rc == -EBUSY || work_restart[thr_id].restart)) {
 			// bbr scratchpad download or stale result
 			sleep(1);
 			if (!thr_id) pools[cur_pooln].wait_time += 1;
@@ -2489,7 +2535,7 @@ static void *miner_thread(void *userdata)
 		}
 
 		// only required to debug purpose
-		if (opt_debug && check_dups && opt_algo != ALGO_DECRED && opt_algo != ALGO_SIA)
+		if (opt_debug && check_dups && opt_algo != ALGO_DECRED && opt_algo != ALGO_EQUIHASH && opt_algo != ALGO_SIA)
 			hashlog_remember_scan_range(&work);
 
 		/* output */
@@ -3854,6 +3900,10 @@ int main(int argc, char *argv[])
 		allow_mininginfo = false;
 	}
 
+	if (opt_algo == ALGO_EQUIHASH) {
+		opt_extranonce = false; // disable subscribe
+	}
+
 	if (opt_algo == ALGO_CRYPTONIGHT || opt_algo == ALGO_CRYPTOLIGHT) {
 		rpc2_init();
 		if (!opt_quiet) applog(LOG_INFO, "Using JSON-RPC 2.0");

ccminer.vcxproj

@@ -156,7 +156,7 @@
       <PtxAsOptionV>true</PtxAsOptionV>
       <Keep>true</Keep>
       <CodeGeneration>compute_50,sm_50;compute_52,sm_52;compute_30,sm_30;compute_20,sm_21</CodeGeneration>
-      <AdditionalOptions>--ptxas-options="-O2" %(AdditionalOptions)</AdditionalOptions>
+      <AdditionalOptions>--ptxas-options="-O2" --Wno-deprecated-gpu-targets %(AdditionalOptions)</AdditionalOptions>
       <Optimization>O2</Optimization>
     </CudaCompile>
     <CudaLink>
@@ -199,9 +199,10 @@
       <PtxAsOptionV>true</PtxAsOptionV>
       <Keep>true</Keep>
       <CodeGeneration>compute_52,sm_52;compute_50,sm_50;compute_35,sm_35;compute_30,sm_30;compute_20,sm_21</CodeGeneration>
-      <Include>$(NVTOOLSEXT_PATH)\include;..\..\..\Common\C99</Include>
+      <Include>$(NVTOOLSEXT_PATH)\include</Include>
       <Optimization>O3</Optimization>
       <TargetMachinePlatform>64</TargetMachinePlatform>
+      <AdditionalOptions>--Wno-deprecated-gpu-targets %(AdditionalOptions)</AdditionalOptions>
     </CudaCompile>
     <CudaLink>
       <Optimization>O3</Optimization>
@@ -235,6 +236,12 @@
     <ClCompile Include="crypto\cryptolight-cpu.cpp" />
     <ClCompile Include="crypto\cryptonight-cpu.cpp" />
     <ClCompile Include="crypto\cpu\c_keccak.c" />
+    <ClCompile Include="equi\blake2\blake2bx.cpp">
+      <EnableEnhancedInstructionSet Condition="'$(Platform)'=='Win32'">StreamingSIMDExtensions</EnableEnhancedInstructionSet>
+    </ClCompile>
+    <ClCompile Include="equi\equi-stratum.cpp" />
+    <ClCompile Include="equi\equi.cpp" />
+    <ClCompile Include="equi\equihash.cpp" />
     <ClCompile Include="nvapi.cpp" />
     <ClCompile Include="pools.cpp" />
     <ClCompile Include="util.cpp" />
@@ -254,6 +261,8 @@
     <ClCompile Include="lyra2\Lyra2.c" />
     <ClCompile Include="lyra2\Sponge.c" />
     <ClCompile Include="lyra2\Lyra2Z.c" />
+    <ClInclude Include="equi\eqcuda.hpp" />
+    <ClInclude Include="equi\equihash.h" />
     <ClInclude Include="neoscrypt\neoscrypt.h" />
     <ClCompile Include="neoscrypt\neoscrypt.cpp" />
     <ClCompile Include="neoscrypt\neoscrypt-cpu.c" />
@@ -283,6 +292,11 @@
     <CudaCompile Include="crypto\wildkeccak.cu">
       <MaxRegCount>128</MaxRegCount>
     </CudaCompile>
+    <CudaCompile Include="equi\cuda_equi.cu">
+      <CodeGeneration>compute_52,sm_52;compute_50,sm_50;compute_30,sm_30</CodeGeneration>
+      <AdditionalOptions> -Xptxas -dlcm=ca -Xptxas -dscm=cs %(AdditionalOptions)</AdditionalOptions>
+      <MaxRegCount>0</MaxRegCount>
+    </CudaCompile>
     <CudaCompile Include="neoscrypt\cuda_neoscrypt.cu">
       <MaxRegCount>160</MaxRegCount>
     </CudaCompile>
@@ -552,7 +566,7 @@
     </CudaCompile>
     <CudaCompile Include="x15\cuda_x14_shabal512.cu" />
     <CudaCompile Include="x15\cuda_x15_whirlpool.cu" />
-    <CudaCompile Include="x17\hmq17.cu"/>
+    <CudaCompile Include="x17\hmq17.cu" />
     <CudaCompile Include="x15\x15.cu" />
     <CudaCompile Include="x15\whirlpool.cu" />
     <CudaCompile Include="x17\x17.cu">

ccminer.vcxproj.filters

@@ -103,6 +103,9 @@
     <Filter Include="Source Files\CUDA\xmr">
       <UniqueIdentifier>{0f9aec5e-5409-488f-992a-2c108590d1ac}</UniqueIdentifier>
     </Filter>
+    <Filter Include="Source Files\equi">
+      <UniqueIdentifier>{031afae7-2a78-4e32-9738-4b589b6f7ff3}</UniqueIdentifier>
+    </Filter>
   </ItemGroup>
   <ItemGroup>
     <ClCompile Include="compat\jansson\dump.c">
@@ -321,6 +324,18 @@
     <ClCompile Include="crypto\wildkeccak-cpu.cpp">
       <Filter>Source Files\crypto\bbr</Filter>
     </ClCompile>
+    <ClCompile Include="equi\equi.cpp">
+      <Filter>Source Files\equi</Filter>
+    </ClCompile>
+    <ClCompile Include="equi\equihash.cpp">
+      <Filter>Source Files\equi</Filter>
+    </ClCompile>
+    <ClCompile Include="equi\equi-stratum.cpp">
+      <Filter>Source Files\equi</Filter>
+    </ClCompile>
+    <ClCompile Include="equi\blake2\blake2bx.cpp">
+      <Filter>Source Files\equi</Filter>
+    </ClCompile>
   </ItemGroup>
   <ItemGroup>
     <ClInclude Include="algos.h">
@@ -563,6 +578,12 @@
     <ClInclude Include="crypto\cryptonight.h">
       <Filter>Source Files\CUDA\xmr</Filter>
     </ClInclude>
+    <ClInclude Include="equi\eqcuda.hpp">
+      <Filter>Source Files\equi</Filter>
+    </ClInclude>
+    <ClInclude Include="equi\equihash.h">
+      <Filter>Source Files\equi</Filter>
+    </ClInclude>
   </ItemGroup>
   <ItemGroup>
     <CudaCompile Include="cuda.cpp">
@@ -895,6 +916,9 @@
     <CudaCompile Include="crypto\wildkeccak.cu">
       <Filter>Source Files\CUDA</Filter>
     </CudaCompile>
+    <CudaCompile Include="equi\cuda_equi.cu">
+      <Filter>Source Files\equi</Filter>
+    </CudaCompile>
   </ItemGroup>
   <ItemGroup>
     <Image Include="res\ccminer.ico">
@@ -911,4 +935,4 @@
       <Filter>Ressources</Filter>
     </Text>
   </ItemGroup>
-</Project>
+</Project>

compat/ccminer-config.h

@@ -164,7 +164,7 @@
 #define PACKAGE_URL "http://github.com/tpruvot/ccminer"
 
 /* Define to the version of this package. */
-#define PACKAGE_VERSION "2.0"
+#define PACKAGE_VERSION "2.1"
 
 /* If using the C implementation of alloca, define if you know the
    direction of stack growth for your system; otherwise it will be

configure.ac

@@ -1,4 +1,4 @@
-AC_INIT([ccminer], [2.0], [], [ccminer], [http://github.com/tpruvot/ccminer])
+AC_INIT([ccminer], [2.1], [], [ccminer], [http://github.com/tpruvot/ccminer])
 
 AC_PREREQ([2.59c])
 AC_CANONICAL_SYSTEM