v3.9.6

algo/bmw/bmw512-4way.c

@@ -0,0 +1,59 @@
+#include "bmw512-gate.h"
+
+#ifdef BMW512_4WAY
+
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+//#include "sph_keccak.h"
+#include "bmw-hash-4way.h"
+
+void bmw512hash_4way(void *state, const void *input)
+{
+    bmw512_4way_context ctx;
+    bmw512_4way_init( &ctx );
+    bmw512_4way( &ctx, input, 80 );
+    bmw512_4way_close( &ctx, state );
+}
+
+int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
+                          uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t vdata[24*4] __attribute__ ((aligned (64)));
+   uint32_t hash[16*4] __attribute__ ((aligned (32)));
+   uint32_t lane_hash[8] __attribute__ ((aligned (32)));
+   uint32_t *hash7 = &(hash[25]);   // 3*8+1
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+   uint32_t n = pdata[19];
+   const uint32_t first_nonce = pdata[19];
+   __m256i  *noncev = (__m256i*)vdata + 9;   // aligned
+//   const uint32_t Htarg = ptarget[7];
+    int thr_id = mythr->id;  // thr_id arg is deprecated
+
+   mm256_bswap32_intrlv80_4x64( vdata, pdata );
+   do {
+       *noncev = mm256_intrlv_blend_32( mm256_bswap_32(
+                _mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ) ), *noncev );
+	
+      bmw512hash_4way( hash, vdata );
+
+      for ( int lane = 0; lane < 4; lane++ )
+      if ( ( ( hash7[ lane<<1 ] & 0xFFFFFF00 ) == 0 ) )
+      {
+          extr_lane_4x64( lane_hash, hash, lane, 256 );
+          if ( fulltest( lane_hash, ptarget ) )
+          {
+              pdata[19] = n + lane;
+              submit_lane_solution( work, lane_hash, mythr, lane );
+          }
+      }
+      n += 4;
+
+   } while ( (n < max_nonce-4) && !work_restart[thr_id].restart);
+
+   *hashes_done = n - first_nonce + 1;
+   return 0;
+}
+
+#endif

algo/bmw/bmw512-gate.c

@@ -0,0 +1,20 @@
+#include "bmw512-gate.h"
+
+int64_t bmw512_get_max64() { return 0x7ffffLL; }
+
+bool register_bmw512_algo( algo_gate_t* gate )
+{
+  gate->optimizations = AVX2_OPT;
+  gate->set_target      = (void*)&alt_set_target;
+  gate->get_max64       = (void*)&bmw512_get_max64;
+#if defined (BMW512_4WAY)
+  gate->scanhash  = (void*)&scanhash_bmw512_4way;
+  gate->hash      = (void*)&bmw512hash_4way;
+#else
+  gate->scanhash        = (void*)&scanhash_bmw512;
+  gate->hash            = (void*)&bmw512hash;
+#endif
+  return true;
+};
+
+

algo/bmw/bmw512-gate.h

@@ -0,0 +1,23 @@
+#ifndef BMW512_GATE_H__
+#define BMW512_GATE_H__
+
+#include "algo-gate-api.h"
+#include <stdint.h>
+
+#if defined(__AVX2__)
+  #define BMW512_4WAY 1
+#endif
+
+#if defined(BMW512_4WAY)
+
+void bmw512hash_4way( void *state, const void *input );
+int scanhash_bmw512_4way( struct work *work, uint32_t max_nonce,
+                         uint64_t *hashes_done, struct thr_info *mythr );
+
+#endif
+
+void bmw512hash( void *state, const void *input );
+int scanhash_bmw512( struct work *work, uint32_t max_nonce,
+                    uint64_t *hashes_done, struct thr_info *mythr );
+
+#endif

algo/bmw/bmw512.c

@@ -0,0 +1,53 @@
+#include "algo-gate-api.h"
+
+#include <stdlib.h>
+#include <string.h>
+#include <stdint.h>
+
+#include "sph_bmw.h"
+
+void bmw512hash(void *state, const void *input)
+{
+    sph_bmw512_context ctx;
+    uint32_t hash[32];	
+   
+    sph_bmw512_init( &ctx );
+    sph_bmw512( &ctx,input, 80 );
+    sph_bmw512_close( &ctx, hash );
+
+    memcpy( state, hash, 32 );
+}
+
+int scanhash_bmw512( struct work *work, uint32_t max_nonce,
+                     uint64_t *hashes_done, struct thr_info *mythr )
+{
+   uint32_t *pdata = work->data;
+   uint32_t *ptarget = work->target;
+	uint32_t n = pdata[19] - 1;
+	const uint32_t first_nonce = pdata[19];
+	//const uint32_t Htarg = ptarget[7];
+   int thr_id = mythr->id;  // thr_id arg is deprecated
+
+	uint32_t _ALIGN(32) hash64[8];
+	uint32_t endiandata[32];
+
+   for (int i=0; i < 19; i++) 
+           be32enc(&endiandata[i], pdata[i]);
+
+	do {
+	
+		pdata[19] = ++n;
+		be32enc(&endiandata[19], n); 
+		bmw512hash(hash64, endiandata);
+        if (((hash64[7]&0xFFFFFF00)==0) && 
+				fulltest(hash64, ptarget)) {
+            *hashes_done = n - first_nonce + 1;
+			return true;
+		}
+	} while (n < max_nonce && !work_restart[thr_id].restart);
+	
+	*hashes_done = n - first_nonce + 1;
+	pdata[19] = n;
+	return 0;
+}
+