v3.21.3 Unreleased

JayDDee · Mar 13, 2023 · c6bc9d6 · c6bc9d6
1 parent b339450
commit c6bc9d6
Show file tree

Hide file tree

Showing 49 changed files with 1,130 additions and 1,115 deletions.
diff --git a/Makefile.am b/Makefile.am
@@ -264,6 +264,8 @@ cpuminer_SOURCES = \
   algo/x16/x16r-4way.c \
   algo/x16/x16rv2.c \
   algo/x16/x16rv2-4way.c \
+  algo/x16/x16rt.c \
+  algo/x16/x16rt-4way.c \
   algo/x16/hex.c \
   algo/x16/x21s-4way.c \
   algo/x16/x21s.c \

diff --git a/RELEASE_NOTES b/RELEASE_NOTES
@@ -65,7 +65,11 @@ If not what makes it happen or not happen?
 Change Log
 ----------
 
-v3.22.3
+v3.21.3.1 UNRELEASED
+
+Revert to 3.21.2
+
+v3.21.3 CANCELLED
 
 #392 #379 #389 Fixed misaligned address segfault solo mining.
 #392 Fixed stats for myr-gr algo, and a few others, for CPUs without AVX2.
@@ -74,10 +78,10 @@ v3.22.3
      Windows binaries no longer support CPU groups,
      Windows binaries support CPUs with up to 64 threads.
 Midstate prehash is now centralized, done only once instead of by every thread
-for selected algos. 
+for selected algos.
 Small optimizations to serialized vectoring.
 
-v3.22.2
+v3.21.2 
 
 Faster SALSA SIMD shuffle for yespower, yescrypt & scryptn2.
 Fixed a couple of compiler warnings with gcc-12.

diff --git a/algo-gate-api.c b/algo-gate-api.c
@@ -253,7 +253,6 @@ void init_algo_gate( algo_gate_t* gate )
    gate->miner_thread_init       = (void*)&return_true;
    gate->scanhash                = (void*)&scanhash_generic;
    gate->hash                    = (void*)&null_hash;
-   gate->prehash                 = (void*)&return_true;
    gate->get_new_work            = (void*)&std_get_new_work;
    gate->work_decode             = (void*)&std_le_work_decode;
    gate->decode_extra_data       = (void*)&do_nothing;

diff --git a/algo-gate-api.h b/algo-gate-api.h
@@ -119,17 +119,14 @@ typedef struct
 // to be registered with the gate. 
 int ( *scanhash ) ( struct work*, uint32_t, uint64_t*, struct thr_info* );
 
-int ( *hash )     ( void*, const void*, const int );
+int ( *hash )     ( void*, const void*, int );
 
 //optional, safe to use default in most cases
 
 // Called once by each miner thread to allocate thread local buffers and
 // other initialization specific to miner threads.
 bool ( *miner_thread_init )     ( int );
 
-// Perform prehash after receiving new work
-int ( *prehash )                ( struct work* );
-
 // Get thread local copy of blockheader with unique nonce.
 void ( *get_new_work )          ( struct work*, struct work*, int, uint32_t* );
 

diff --git a/algo/blake/blake2s-4way.c b/algo/blake/blake2s-4way.c
@@ -1,50 +1,12 @@
 #include "blake2s-gate.h"
 #include "blake2s-hash-4way.h"
-//#include "sph-blake2s.h"
 #include <string.h>
 #include <stdint.h>
 
 #if defined(BLAKE2S_16WAY)
 
 static __thread blake2s_16way_state blake2s_16w_ctx;
 
-/*
-static blake2s_16way_state blake2s_16w_ctx;
-static uint32_t blake2s_16way_vdata[20*16] __attribute__ ((aligned (64)));
-*/
-/*
-int blake2s_16way_prehash( struct work *work )
-{
-   uint32_t edata[20] __attribute__ ((aligned (64)));
-   blake2s_state ctx;
-   mm128_bswap32_80( edata, work->data );
-   blake2s_init( &ctx, BLAKE2S_OUTBYTES );
-   ctx.buflen = ctx.t[0] = 64;
-   blake2s_compress( &ctx, (const uint8_t*)edata );
-
-   blake2s_16way_init( &blake2s_16w_ctx, BLAKE2S_OUTBYTES );
-   intrlv_16x32( blake2s_16w_ctx.h, ctx.h, ctx.h, ctx.h, ctx.h,
-                                    ctx.h, ctx.h, ctx.h, ctx.h,
-                                    ctx.h, ctx.h, ctx.h, ctx.h,
-                                    ctx.h, ctx.h, ctx.h, ctx.h, 256 );
-   intrlv_16x32( blake2s_16way_vdata, edata, edata, edata, edata,
-                                      edata, edata, edata, edata,
-                                      edata, edata, edata, edata,
-                                      edata, edata, edata, edata, 640 );
-   blake2s_16w_ctx.t[0] = 64;
-   return 1;
-}
-*/
-/*
-int blake2s_16way_prehash( struct work *work )
-{
-   mm512_bswap32_intrlv80_16x32( blake2s_16way_vdata, work->data );
-   blake2s_16way_init( &blake2s_16w_ctx, BLAKE2S_OUTBYTES );
-   blake2s_16way_update( &blake2s_16w_ctx, blake2s_16way_vdata, 64 );
-   return 1;
-}
-*/
-
 void blake2s_16way_hash( void *output, const void *input )
 {
    blake2s_16way_state ctx;
@@ -68,40 +30,10 @@ int scanhash_blake2s_16way( struct work *work, uint32_t max_nonce,
    uint32_t n = first_nonce;
    int thr_id = mythr->id;  
 
-/*   
-//   pthread_rwlock_rdlock( &g_work_lock );
-       memcpy( (__m512i*)vdata +16, (__m512i*)blake2s_16way_vdata +16, 3*4*16 );
-//     casti_m512i( vdata, 16 ) = casti_m512i( blake2s_16way_vdata, 16 );
-//     casti_m512i( vdata, 17 ) = casti_m512i( blake2s_16way_vdata, 17 );
-//     casti_m512i( vdata, 18 ) = casti_m512i( blake2s_16way_vdata, 18 );
-       
-//   pthread_rwlock_unlock( &g_work_lock );
-*/
-/*
-   uint32_t edata[20] __attribute__ ((aligned (64)));
-   blake2s_state ctx;
-   mm128_bswap32_80( edata, pdata );
-   blake2s_init( &ctx, BLAKE2S_OUTBYTES );
-   ctx.buflen = ctx.t[0] = 64;
-   blake2s_compress( &ctx, (const uint8_t*)edata );
-
-   blake2s_16way_init( &blake2s_16w_ctx, BLAKE2S_OUTBYTES );
-   intrlv_16x32( blake2s_16w_ctx.h, ctx.h, ctx.h, ctx.h, ctx.h,
-                                    ctx.h, ctx.h, ctx.h, ctx.h,
-                                    ctx.h, ctx.h, ctx.h, ctx.h,
-                                    ctx.h, ctx.h, ctx.h, ctx.h, 256 );
-   intrlv_16x32( blake2s_16way_blake2s_16way_vdata, edata, edata, edata, edata,
-                                      edata, edata, edata, edata,
-                                      edata, edata, edata, edata,
-                                      edata, edata, edata, edata, 640 );
-   blake2s_16w_ctx.t[0] = 64;
-*/
-
    mm512_bswap32_intrlv80_16x32( vdata, pdata );
    blake2s_16way_init( &blake2s_16w_ctx, BLAKE2S_OUTBYTES );
    blake2s_16way_update( &blake2s_16w_ctx, vdata, 64 );
 
-
    do {
       *noncev = mm512_bswap_32( _mm512_set_epi32(
 	                  n+15, n+14, n+13, n+12, n+11, n+10, n+ 9, n+ 8,
@@ -131,36 +63,6 @@ int scanhash_blake2s_16way( struct work *work, uint32_t max_nonce,
 
 static __thread blake2s_8way_state blake2s_8w_ctx;
 
-/*
-static blake2s_8way_state blake2s_8w_ctx;
-static uint32_t blake2s_8way_vdata[20*8] __attribute__ ((aligned (32)));
-
-int blake2s_8way_prehash( struct work *work )
-{
-   uint32_t edata[20] __attribute__ ((aligned (64)));
-   blake2s_state ctx;
-   mm128_bswap32_80( edata, work->data );
-   blake2s_init( &ctx, BLAKE2S_OUTBYTES );
-   ctx.buflen = ctx.t[0] = 64;
-   blake2s_compress( &ctx, (const uint8_t*)edata );
-
-   blake2s_8way_init( &blake2s_8w_ctx, BLAKE2S_OUTBYTES );
-
-   for ( int i = 0; i < 8; i++ )
-      casti_m256i( blake2s_8w_ctx.h, i ) = _mm256_set1_epi32( ctx.h[i] );
-
-   casti_m256i( blake2s_8way_vdata, 16 ) = _mm256_set1_epi32( edata[16] );
-   casti_m256i( blake2s_8way_vdata, 17 ) = _mm256_set1_epi32( edata[17] );
-   casti_m256i( blake2s_8way_vdata, 18 ) = _mm256_set1_epi32( edata[18] );
-
-//   intrlv_8x32( blake2s_8w_ctx.h, ctx.h, ctx.h, ctx.h, ctx.h,
-//                                  ctx.h, ctx.h, ctx.h, ctx.h, 256 );
-//   intrlv_8x32( blake2s_8way_vdata, edata, edata, edata, edata,
-//                                    edata, edata, edata, edata, 640 );
-   blake2s_8w_ctx.t[0] = 64;
-}
-*/
-
 void blake2s_8way_hash( void *output, const void *input )
 {
    blake2s_8way_state ctx;
@@ -184,41 +86,10 @@ int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
    uint32_t n = first_nonce;
    int thr_id = mythr->id; 
 
-/*   
-//   pthread_rwlock_rdlock( &g_work_lock );
-       memcpy( &vdata[16*8], &blake2s_8way_vdata[16*8], 3*4*8 );
-//   pthread_rwlock_unlock( &g_work_lock );
-*/
-/*
-   uint32_t edata[20] __attribute__ ((aligned (64)));
-   blake2s_state ctx;
-   mm128_bswap32_80( edata, pdata );
-   blake2s_init( &ctx, BLAKE2S_OUTBYTES );
-   ctx.buflen = ctx.t[0] = 64;
-   blake2s_compress( &ctx, (const uint8_t*)edata );
-
-   blake2s_8way_init( &blake2s_8w_ctx, BLAKE2S_OUTBYTES );
-   for ( int i = 0; i < 8; i++ )
-      casti_m256i( blake2s_8w_ctx.h, i ) = _mm256_set1_epi32( ctx.h[i] );
-
-   casti_m256i( vdata, 16 ) = _mm256_set1_epi32( edata[16] );
-   casti_m256i( vdata, 17 ) = _mm256_set1_epi32( edata[17] );
-   casti_m256i( vdata, 18 ) = _mm256_set1_epi32( edata[18] );
-
-
-//  intrlv_8x32( blake2s_8w_ctx.h, ctx.h, ctx.h, ctx.h, ctx.h,
-//                                  ctx.h, ctx.h, ctx.h, ctx.h, 256 );
-//   intrlv_8x32( vdata, edata, edata, edata, edata,
-//                                    edata, edata, edata, edata, 640 );
-
-   blake2s_8w_ctx.t[0] = 64;
-*/
-
    mm256_bswap32_intrlv80_8x32( vdata, pdata );
    blake2s_8way_init( &blake2s_8w_ctx, BLAKE2S_OUTBYTES );
    blake2s_8way_update( &blake2s_8w_ctx, vdata, 64 );
 
-
    do {
       *noncev = mm256_bswap_32( _mm256_set_epi32( n+7, n+6, n+5, n+4,
                                                   n+3, n+2, n+1, n ) );
@@ -246,25 +117,7 @@ int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
 #elif defined(BLAKE2S_4WAY)
 
 static __thread blake2s_4way_state blake2s_4w_ctx;
-/*
-static blake2s_4way_state blake2s_4w_ctx;
-static uint32_t blake2s_4way_vdata[20*4] __attribute__ ((aligned (32)));
-
-int blake2s_4way_prehash( struct work *work )
-{
-   uint32_t edata[20] __attribute__ ((aligned (64)));
-   blake2s_state ctx;
-   mm128_bswap32_80( edata, work->data );
-   blake2s_init( &ctx, BLAKE2S_OUTBYTES );
-   ctx.buflen = ctx.t[0] = 64;
-   blake2s_compress( &ctx, (const uint8_t*)edata );
 
-   blake2s_4way_init( &blake2s_4w_ctx, BLAKE2S_OUTBYTES );
-   intrlv_4x32( blake2s_4w_ctx.h, ctx.h, ctx.h, ctx.h, ctx.h, 256 );
-   intrlv_4x32( blake2s_4way_vdata, edata, edata, edata, edata, 640 );
-   blake2s_4w_ctx.t[0] = 64;
-}
-*/
 void blake2s_4way_hash( void *output, const void *input )
 {
    blake2s_4way_state ctx;
@@ -287,15 +140,11 @@ int scanhash_blake2s_4way( struct work *work, uint32_t max_nonce,
    __m128i  *noncev = (__m128i*)vdata + 19;   // aligned
    uint32_t n = first_nonce;
    int thr_id = mythr->id; 
-/*
-   pthread_rwlock_rdlock( &g_work_lock );
-       memcpy( vdata, blake2s_4way_vdata, sizeof vdata );
-   pthread_rwlock_unlock( &g_work_lock );
-*/
+
    mm128_bswap32_intrlv80_4x32( vdata, pdata );
    blake2s_4way_init( &blake2s_4w_ctx, BLAKE2S_OUTBYTES );
    blake2s_4way_update( &blake2s_4w_ctx, vdata, 64 );
-   
+
    do {
       *noncev = mm128_bswap_32( _mm_set_epi32( n+3, n+2, n+1, n ) );
       pdata[19] = n;

diff --git a/algo/blake/blake2s-gate.c b/algo/blake/blake2s-gate.c
@@ -5,15 +5,13 @@ bool register_blake2s_algo( algo_gate_t* gate )
 #if defined(BLAKE2S_16WAY)
   gate->scanhash  = (void*)&scanhash_blake2s_16way;
   gate->hash      = (void*)&blake2s_16way_hash;
-//  gate->prehash   = (void*)&blake2s_16way_prehash;
 #elif defined(BLAKE2S_8WAY)
+//#if defined(BLAKE2S_8WAY)
   gate->scanhash  = (void*)&scanhash_blake2s_8way;
   gate->hash      = (void*)&blake2s_8way_hash;
-//  gate->prehash   = (void*)&blake2s_8way_prehash;
 #elif defined(BLAKE2S_4WAY)
   gate->scanhash  = (void*)&scanhash_blake2s_4way;
   gate->hash      = (void*)&blake2s_4way_hash;
-//  gate->prehash   = (void*)&blake2s_4way_prehash;
 #else
   gate->scanhash  = (void*)&scanhash_blake2s;
   gate->hash      = (void*)&blake2s_hash;

diff --git a/algo/blake/blake2s-gate.h b/algo/blake/blake2s-gate.h
@@ -23,22 +23,18 @@ bool register_blake2s_algo( algo_gate_t* gate );
 void blake2s_16way_hash( void *state, const void *input );
 int scanhash_blake2s_16way( struct work *work, uint32_t max_nonce,
                          uint64_t *hashes_done, struct thr_info *mythr );
-int blake2s_16way_prehash( struct work * );
 
 #elif defined (BLAKE2S_8WAY)
 
 void blake2s_8way_hash( void *state, const void *input );
 int scanhash_blake2s_8way( struct work *work, uint32_t max_nonce,
                          uint64_t *hashes_done, struct thr_info *mythr );
-int blake2s_8way_prehash( struct work * );
 
 #elif defined (BLAKE2S_4WAY)
 
 void blake2s_4way_hash( void *state, const void *input );
 int scanhash_blake2s_4way( struct work *work, uint32_t max_nonce,
                          uint64_t *hashes_done, struct thr_info *mythr );
-int blake2s_4way_prehash( struct work * );
-
 #else
 
 void blake2s_hash( void *state, const void *input );

diff --git a/algo/blake/blake2s-hash-4way.c b/algo/blake/blake2s-hash-4way.c
@@ -105,8 +105,8 @@ int blake2s_4way_compress( blake2s_4way_state *S, const __m128i* block )
 
 #define G4W( sigma0, sigma1, a, b, c, d ) \
 do { \
-   const uint8_t s0 = sigma0; \
-   const uint8_t s1 = sigma1; \
+   uint8_t s0 = sigma0; \
+   uint8_t s1 = sigma1; \
    a = _mm_add_epi32( _mm_add_epi32( a, b ), m[ s0 ] ); \
    d = mm128_swap32_16( _mm_xor_si128( d, a ) ); \
    c = _mm_add_epi32( c, d ); \
@@ -120,7 +120,7 @@ do { \
 
 #define ROUND4W(r)  \
 do { \
-   const uint8_t *sigma = (const uint8_t*)&blake2s_sigma[r]; \
+   uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
    G4W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
    G4W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
    G4W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \
@@ -317,8 +317,8 @@ do { \
 
 #define G8W( sigma0, sigma1, a, b, c, d) \
 do { \
-   const uint8_t s0 = sigma0; \
-   const uint8_t s1 = sigma1; \
+   uint8_t s0 = sigma0; \
+   uint8_t s1 = sigma1; \
    a = _mm256_add_epi32( _mm256_add_epi32( a, b ), m[ s0 ] ); \
    d = mm256_swap32_16( _mm256_xor_si256( d, a ) ); \
    c = _mm256_add_epi32( c, d ); \
@@ -331,7 +331,7 @@ do { \
 
 #define ROUND8W(r)  \
 do { \
-   const uint8_t *sigma = (const uint8_t*)&blake2s_sigma[r]; \
+   uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
    G8W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
    G8W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
    G8W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \
@@ -529,8 +529,8 @@ int blake2s_16way_compress( blake2s_16way_state *S, const __m512i *block )
 
 #define G16W( sigma0, sigma1, a, b, c, d) \
 do { \
-   const uint8_t s0 = sigma0; \
-   const uint8_t s1 = sigma1; \
+   uint8_t s0 = sigma0; \
+   uint8_t s1 = sigma1; \
    a = _mm512_add_epi32( _mm512_add_epi32( a, b ), m[ s0 ] ); \
    d = mm512_ror_32( _mm512_xor_si512( d, a ), 16 ); \
    c = _mm512_add_epi32( c, d ); \
@@ -543,7 +543,7 @@ do { \
 
 #define ROUND16W(r)  \
 do { \
-   const uint8_t *sigma = (const uint8_t*)&blake2s_sigma[r]; \
+   uint8_t *sigma = (uint8_t*)&blake2s_sigma[r]; \
    G16W( sigma[ 0], sigma[ 1], v[ 0], v[ 4], v[ 8], v[12] ); \
    G16W( sigma[ 2], sigma[ 3], v[ 1], v[ 5], v[ 9], v[13] ); \
    G16W( sigma[ 4], sigma[ 5], v[ 2], v[ 6], v[10], v[14] ); \