v3.9.2.5

RELEASE_NOTES

@@ -38,6 +38,11 @@ supported.
 Change Log
 ----------
 
+v3.9.2.5
+
+Fixed 2 regressions: hodl AES detection, x16r invalid shares with AVX2.
+More restructuring.
+
 v3.9.2.4
 
 Yet another affinity fix. Hopefully the last one.

algo-gate-api.h

@@ -2,8 +2,7 @@
 #include <stdbool.h>
 #include <stdint.h>
 #include "miner.h"
-#include "avxdefs.h"
-#include "interleave.h"
+#include "simd-utils.h"
 
 /////////////////////////////
 ////

algo/blake/blake-hash-4way.h

@@ -45,7 +45,7 @@ extern "C"{
 
 #include <stddef.h>
 #include "algo/sha/sph_types.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #define SPH_SIZE_blake256   256
 

algo/blake/blake2s-hash-4way.h

@@ -16,7 +16,7 @@
 
 #if defined(__SSE4_2__)
 
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #include <stddef.h>
 #include <stdint.h>

algo/bmw/bmw-hash-4way.h

@@ -43,7 +43,7 @@ extern "C"{
 #include <stddef.h>
 
 #include "algo/sha/sph_types.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #define SPH_SIZE_bmw256   256
 

algo/cubehash/cube-hash-2way.h

@@ -4,7 +4,7 @@
 #if defined(__AVX2__)
 
 #include <stdint.h>
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 // 2x128, 2 way parallel SSE2
 

algo/cubehash/cubehash_sse2.c

@@ -13,7 +13,7 @@
 #include <stdbool.h>
 #include <unistd.h>
 #include <memory.h>
-#include "avxdefs.h"
+#include "simd-utils.h"
 #include <stdio.h>
 
 // The result of hashing 10 rounds of initial data which is params and 

algo/groestl/aes_ni/hash-groestl.c

@@ -12,7 +12,7 @@
 #include <memory.h>
 #include "hash-groestl.h"
 #include "miner.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #ifndef NO_AES_NI
 

algo/groestl/aes_ni/hash-groestl256.c

@@ -9,7 +9,7 @@
 #include <memory.h>
 #include "hash-groestl256.h"
 #include "miner.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #ifndef NO_AES_NI
 

algo/hamsi/hamsi-hash-4way.h

@@ -40,7 +40,7 @@
 
 #if defined (__AVX2__)
 
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #ifdef __cplusplus
 extern "C"{

algo/haval/haval-hash-4way.h

@@ -69,7 +69,7 @@ extern "C"{
 
 #include <stddef.h>
 #include "algo/sha/sph_types.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #define SPH_SIZE_haval256_5   256
 

algo/hodl/hodl-gate.c

@@ -156,7 +156,7 @@ int hodl_scanhash( int thr_id, struct work* work, uint32_t max_nonce,
 
 bool register_hodl_algo( algo_gate_t* gate )
 {
-#if defined(__AES__)
+#if !defined(__AES__)
   applog( LOG_ERR, "Only CPUs with AES are supported, use legacy version.");
   return false;
 #endif

algo/jh/jh-hash-4way.h

@@ -44,7 +44,7 @@ extern "C"{
 
 #include <stddef.h>
 #include "algo/sha/sph_types.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #define SPH_SIZE_jh256   256
 

algo/keccak/keccak-hash-4way.h

@@ -44,7 +44,7 @@ extern "C"{
 
 #include <stddef.h>
 #include "algo/sha/sph_types.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #define SPH_SIZE_keccak256   256
 

algo/luffa/luffa-hash-2way.c

@@ -24,7 +24,7 @@
 
 #if defined(__AVX2__)
 
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 #define MASK _mm256_set_epi32( 0UL, 0UL, 0UL, 0xffffffffUL, \
                                0UL, 0UL, 0UL, 0xffffffffUL )

algo/luffa/luffa-hash-2way.h

@@ -24,7 +24,7 @@
 
 #include <immintrin.h>
 #include "algo/sha/sha3-defs.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 
 /* The length of digests*/
 #define DIGEST_BIT_LEN_224 224

algo/luffa/luffa_for_sse2.c

@@ -20,7 +20,7 @@
 
 #include <string.h>
 #include <emmintrin.h>
-#include "avxdefs.h"
+#include "simd-utils.h"
 #include "luffa_for_sse2.h"
 
 #define MULT2(a0,a1) do \

algo/lyra2/allium-4way.c

@@ -44,10 +44,11 @@ void allium_4way_hash( void *state, const void *input )
    blake256_4way( &ctx.blake, input + (64<<2), 16 );
    blake256_4way_close( &ctx.blake, vhash32 );
 
-   mm256_reinterleave_4x64( vhash64, vhash32, 256 );
+   mm256_rintrlv_4x32_4x64( vhash64, vhash32, 256 );
    keccak256_4way( &ctx.keccak, vhash64, 32 );
    keccak256_4way_close( &ctx.keccak, vhash64 );
-   mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+
+   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
 
    LYRA2RE( hash0, 32, hash0, 32, hash0, 32, 1, 8, 8 );
    LYRA2RE( hash1, 32, hash1, 32, hash1, 32, 1, 8, 8 );
@@ -67,34 +68,30 @@ void allium_4way_hash( void *state, const void *input )
    LYRA2RE( hash2, 32, hash2, 32, hash2, 32, 1, 8, 8 );
    LYRA2RE( hash3, 32, hash3, 32, hash3, 32, 1, 8, 8 );
 
-   mm256_interleave_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
+   mm256_intrlv_4x64( vhash64, hash0, hash1, hash2, hash3, 256 );
+
    skein256_4way( &ctx.skein, vhash64, 32 );
    skein256_4way_close( &ctx.skein, vhash64 );
-   mm256_deinterleave_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
 
-   update_and_final_groestl256( &ctx.groestl, hash0, hash0, 256 );
+   mm256_dintrlv_4x64( hash0, hash1, hash2, hash3, vhash64, 256 );
+
+   update_and_final_groestl256( &ctx.groestl, state, hash0, 256 );
    memcpy( &ctx.groestl, &allium_4way_ctx.groestl,
            sizeof(hashState_groestl256) );
-   update_and_final_groestl256( &ctx.groestl, hash1, hash1, 256 );
+   update_and_final_groestl256( &ctx.groestl, state+32, hash1, 256 );
    memcpy( &ctx.groestl, &allium_4way_ctx.groestl,
            sizeof(hashState_groestl256) );
-   update_and_final_groestl256( &ctx.groestl, hash2, hash2, 256 );
+   update_and_final_groestl256( &ctx.groestl, state+64, hash2, 256 );
    memcpy( &ctx.groestl, &allium_4way_ctx.groestl,
            sizeof(hashState_groestl256) );
-   update_and_final_groestl256( &ctx.groestl, hash3, hash3, 256 );
-
-   memcpy( state,    hash0, 32 );
-   memcpy( state+32, hash1, 32 );
-   memcpy( state+64, hash2, 32 );
-   memcpy( state+96, hash3, 32 );
+   update_and_final_groestl256( &ctx.groestl, state+96, hash3, 256 );
 }
 
 int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
                              uint64_t *hashes_done, struct thr_info *mythr )
 {
    uint32_t hash[8*4] __attribute__ ((aligned (64)));
    uint32_t vdata[20*4] __attribute__ ((aligned (64)));
-   uint32_t _ALIGN(64) edata[20];
    uint32_t *pdata = work->data;
    uint32_t *ptarget = work->target;
    const uint32_t first_nonce = pdata[19];
@@ -106,13 +103,7 @@ int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
    if ( opt_benchmark )
       ( (uint32_t*)ptarget )[7] = 0x0000ff;
 
-   casti_m128i( edata, 0 ) = mm128_bswap_32( casti_m128i( pdata, 0 ) );
-   casti_m128i( edata, 1 ) = mm128_bswap_32( casti_m128i( pdata, 1 ) );
-   casti_m128i( edata, 2 ) = mm128_bswap_32( casti_m128i( pdata, 2 ) );
-   casti_m128i( edata, 3 ) = mm128_bswap_32( casti_m128i( pdata, 3 ) );
-   casti_m128i( edata, 4 ) = mm128_bswap_32( casti_m128i( pdata, 4 ) );
-
-   mm128_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
+   mm128_bswap_intrlv80_4x32( vdata, pdata );
    blake256_4way_init( &allium_4way_ctx.blake );
    blake256_4way( &allium_4way_ctx.blake, vdata, 64 );
 
@@ -124,7 +115,7 @@ int scanhash_allium_4way( int thr_id, struct work *work, uint32_t max_nonce,
 
      for ( int lane = 0; lane < 4; lane++ ) if ( (hash+(lane<<3))[7] <= Htarg )
      {
-        if ( fulltest( hash+(lane<<3), ptarget ) )
+        if ( fulltest( hash+(lane<<3), ptarget ) && !opt_benchmark )
         {
            pdata[19] = n + lane;
            submit_solution( work, hash+(lane<<3), mythr, lane );

algo/lyra2/lyra2-gate.h

@@ -5,7 +5,9 @@
 #include <stdint.h>
 #include "lyra2.h"
 
-#if defined(__AVX2__)
+//#if defined(__AVX2__)
+
+#if defined(__SSE2__)
   #define LYRA2REV3_4WAY
 #endif
 

algo/lyra2/lyra2.c

@@ -566,7 +566,7 @@ int LYRA2RE( void *K, uint64_t kLen, const void *pwd, const uint64_t pwdlen,
 
 #if defined(__AVX2__)
    memset_zero_256( (__m256i*)wholeMatrix, i>>5 );
-#elif defined(__SSE4_2__)
+#elif defined(__SSE2__)
    memset_zero_128( (__m128i*)wholeMatrix, i>>4 );   
 #else
    memset( wholeMatrix, 0, i );

algo/lyra2/lyra2h-4way.c

@@ -36,67 +36,53 @@ void lyra2h_4way_hash( void *state, const void *input )
      blake256_4way( &ctx_blake, input + (64*4), 16 );
      blake256_4way_close( &ctx_blake, vhash );
 
-     mm128_deinterleave_4x32( hash0, hash1, hash2, hash3, vhash, 256 );
-
-     LYRA2Z( lyra2h_4way_matrix, hash0, 32, hash0, 32, hash0, 32, 16, 16, 16 );
-     LYRA2Z( lyra2h_4way_matrix, hash1, 32, hash1, 32, hash1, 32, 16, 16, 16 );
-     LYRA2Z( lyra2h_4way_matrix, hash2, 32, hash2, 32, hash2, 32, 16, 16, 16 );
-     LYRA2Z( lyra2h_4way_matrix, hash3, 32, hash3, 32, hash3, 32, 16, 16, 16 );
-
-     memcpy( state,    hash0, 32 );
-     memcpy( state+32, hash1, 32 );
-     memcpy( state+64, hash2, 32 );
-     memcpy( state+96, hash3, 32 );
+     mm128_dintrlv_4x32( hash0, hash1, hash2, hash3, vhash, 256 );
+
+     LYRA2Z( lyra2h_4way_matrix, state, 32, hash0, 32, hash0, 32,
+             16, 16, 16 );
+     LYRA2Z( lyra2h_4way_matrix, state+32, 32, hash1, 32, hash1,
+             32, 16, 16, 16 );
+     LYRA2Z( lyra2h_4way_matrix, state+64, 32, hash2, 32, hash2,
+             32, 16, 16, 16 );
+     LYRA2Z( lyra2h_4way_matrix, state+96, 32, hash3, 32, hash3,
+             32, 16, 16, 16 );
 }
 
 int scanhash_lyra2h_4way( int thr_id, struct work *work, uint32_t max_nonce,
                           uint64_t *hashes_done, struct thr_info *mythr )
 {
    uint32_t hash[8*4] __attribute__ ((aligned (64)));
    uint32_t vdata[20*4] __attribute__ ((aligned (64)));
-   uint32_t _ALIGN(64) edata[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 n = first_nonce;
-   uint32_t *nonces = work->nonces;
-   int num_found = 0;
-   uint32_t *noncep= vdata + 76; // 19*4
+   __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
    /* int */ thr_id = mythr->id;  // thr_id arg is deprecated
 
    if ( opt_benchmark )
       ptarget[7] = 0x0000ff;
 
-   for ( int i=0; i < 20; i++ )
-      be32enc( &edata[i], pdata[i] );
-
-   mm128_interleave_4x32( vdata, edata, edata, edata, edata, 640 );
-
+   mm128_bswap_intrlv80_4x32( vdata, pdata );
    lyra2h_4way_midstate( vdata );
 
    do {
-      be32enc( noncep,   n   );
-      be32enc( noncep+1, n+1 );
-      be32enc( noncep+2, n+2 );
-      be32enc( noncep+3, n+3 );
-
-      be32enc( &edata[19], n );
+     *noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
       lyra2h_4way_hash( hash, vdata );
 
       for ( int i = 0; i < 4; i++ )
-      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget ) )
+      if ( (hash+(i<<3))[7] <= Htarg && fulltest( hash+(i<<3), ptarget )
+           && !opt_benchmark )
       {
           pdata[19] = n+i;         
-          nonces[ num_found++ ] = n+i;
-          work_set_target_ratio( work, hash+(i<<3) );
+          submit_solution( work, hash+(i<<3), mythr, i );
       }
       n += 4;
-   } while ( (num_found == 0) && (n < max_nonce-4)
-                   && !work_restart[thr_id].restart);
+   } while (  (n < max_nonce-4) && !work_restart[thr_id].restart);
 
    *hashes_done = n - first_nonce + 1;
-   return num_found;
+   return 0;
 }
 
 #endif

algo/lyra2/lyra2re.c

@@ -6,7 +6,7 @@
 #include "algo/keccak/sph_keccak.h"
 #include "lyra2.h"
 #include "algo-gate-api.h"
-#include "avxdefs.h"
+#include "simd-utils.h"
 #if defined(__AES__)
   #include "algo/groestl/aes_ni/hash-groestl256.h"
 #endif