Skip to content

Commit

Permalink
v23.13
Browse files Browse the repository at this point in the history
  • Loading branch information
@JayDDee
JayDDee committed Nov 21, 2023
1 parent fc696db commit 045b42b
Show file tree
Hide file tree
Showing 21 changed files with 573 additions and 214 deletions.
1 change: 1 addition & 0 deletions Makefile.am
View file
Original file line number Diff line number Diff line change
Expand Up @@ -250,6 +250,7 @@ cpuminer_SOURCES = \
algo/x16/x16rt.c \
algo/x16/x16rt-4way.c \
algo/x16/hex.c \
algo/x16/x20r.c \
algo/x16/x21s-4way.c \
algo/x16/x21s.c \
algo/x16/minotaur.c \
Expand Down
5 changes: 5 additions & 0 deletions RELEASE_NOTES
View file
Original file line number Diff line number Diff line change
Expand Up @@ -75,6 +75,11 @@ If not what makes it happen or not happen?
Change Log
----------

v23.13

Added x20r algo.
Eliminated redundant hash order calculations for x16r family.

v23.12

Several bugs fixes and speed improvements for x16r family for all CPU architectures.
Expand Down
1 change: 1 addition & 0 deletions algo-gate-api.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -368,6 +368,7 @@ bool register_algo_gate( int algo, algo_gate_t *gate )
case ALGO_X16RT_VEIL: rc = register_x16rt_veil_algo ( gate ); break;
case ALGO_X16S: rc = register_x16s_algo ( gate ); break;
case ALGO_X17: rc = register_x17_algo ( gate ); break;
case ALGO_X20R: rc = register_x20r_algo ( gate ); break;
case ALGO_X21S: rc = register_x21s_algo ( gate ); break;
case ALGO_X22I: rc = register_x22i_algo ( gate ); break;
case ALGO_X25X: rc = register_x25x_algo ( gate ); break;
Expand Down
106 changes: 51 additions & 55 deletions algo/x16/x16r-4way.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -19,12 +19,12 @@
// Perform midstate prehash of hash functions with block size <= 72 bytes,
// 76 bytes for hash functions that operate on 32 bit data.

void x16r_8way_prehash( void *vdata, void *pdata )
void x16r_8way_prehash( void *vdata, void *pdata, const char *hash_order )
{
uint32_t vdata2[20*8] __attribute__ ((aligned (64)));
uint32_t edata[20] __attribute__ ((aligned (64)));

const char elem = x16r_hash_order[0];
const char elem = hash_order[0];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';

switch ( algo )
Expand Down Expand Up @@ -110,7 +110,8 @@ void x16r_8way_prehash( void *vdata, void *pdata )
// Called by wrapper hash function to optionally continue hashing and
// convert to final hash.

int x16r_8way_hash_generic( void* output, const void* input, int thrid )
int x16r_8way_hash_generic( void* output, const void* input, int thrid,
const char *hash_order, const int func_count )
{
uint32_t vhash[20*8] __attribute__ ((aligned (128)));
uint32_t hash0[20] __attribute__ ((aligned (16)));
Expand All @@ -136,9 +137,9 @@ int x16r_8way_hash_generic( void* output, const void* input, int thrid )
dintrlv_8x64( hash0, hash1, hash2, hash3, hash4, hash5, hash6, hash7,
input, 640 );

for ( int i = 0; i < 16; i++ )
for ( int i = 0; i < func_count; i++ )
{
const char elem = x16r_hash_order[i];
const char elem = hash_order[i];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';

switch ( algo )
Expand Down Expand Up @@ -474,7 +475,8 @@ int x16r_8way_hash_generic( void* output, const void* input, int thrid )
int x16r_8way_hash( void* output, const void* input, int thrid )
{
uint8_t hash[64*8] __attribute__ ((aligned (128)));
if ( !x16r_8way_hash_generic( hash, input, thrid ) )
if ( !x16r_8way_hash_generic( hash, input, thrid, x16r_hash_order,
X16R_HASH_FUNC_COUNT ) )
return 0;

memcpy( output, hash, 32 );
Expand All @@ -495,7 +497,6 @@ int scanhash_x16r_8way( struct work *work, uint32_t max_nonce,
{
uint32_t hash[16*8] __attribute__ ((aligned (128)));
uint32_t vdata[20*8] __attribute__ ((aligned (64)));
uint32_t bedata1[2];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
Expand All @@ -508,21 +509,18 @@ int scanhash_x16r_8way( struct work *work, uint32_t max_nonce,

if ( bench ) ptarget[7] = 0x0cff;

bedata1[0] = bswap_32( pdata[1] );
bedata1[1] = bswap_32( pdata[2] );

static __thread uint32_t s_ntime = UINT32_MAX;
const uint32_t ntime = bswap_32( pdata[17] );
if ( s_ntime != ntime )
static __thread uint32_t saved_height = UINT32_MAX;
if ( work->height != saved_height )
{
x16_r_s_getAlgoString( (const uint8_t*)bedata1, x16r_hash_order );
s_ntime = ntime;

if ( opt_debug && !thr_id )
applog( LOG_INFO, "Hash order %s Ntime %08x", x16r_hash_order, ntime );
vdata[1] = bswap_32( pdata[1] );
vdata[2] = bswap_32( pdata[2] );
saved_height = work->height;
x16_r_s_getAlgoString( (const uint8_t*)(&vdata[1]), x16r_hash_order );
if ( !opt_quiet && !thr_id )
applog( LOG_INFO, "hash order %s", x16r_hash_order );
}

x16r_8way_prehash( vdata, pdata );
x16r_8way_prehash( vdata, pdata, x16r_hash_order );
*noncev = mm512_intrlv_blend_32( _mm512_set_epi32(
n+7, 0, n+6, 0, n+5, 0, n+4, 0,
n+3, 0, n+2, 0, n+1, 0, n, 0 ), *noncev );
Expand All @@ -546,12 +544,12 @@ int scanhash_x16r_8way( struct work *work, uint32_t max_nonce,

#elif defined (X16R_4WAY)

void x16r_4way_prehash( void *vdata, void *pdata )
void x16r_4way_prehash( void *vdata, void *pdata, const char *hash_order )
{
uint32_t vdata2[20*4] __attribute__ ((aligned (64)));
uint32_t edata[20] __attribute__ ((aligned (64)));

const char elem = x16r_hash_order[0];
const char elem = hash_order[0];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';

switch ( algo )
Expand Down Expand Up @@ -627,7 +625,8 @@ void x16r_4way_prehash( void *vdata, void *pdata )
}
}

int x16r_4way_hash_generic( void* output, const void* input, int thrid )
int x16r_4way_hash_generic( void* output, const void* input, int thrid,
const char *hash_order, const int func_count )
{
uint32_t vhash[20*4] __attribute__ ((aligned (128)));
uint32_t hash0[20] __attribute__ ((aligned (32)));
Expand All @@ -644,9 +643,9 @@ int x16r_4way_hash_generic( void* output, const void* input, int thrid )

dintrlv_4x64( hash0, hash1, hash2, hash3, input, 640 );

for ( int i = 0; i < 16; i++ )
for ( int i = 0; i < func_count; i++ )
{
const char elem = x16r_hash_order[i];
const char elem = hash_order[i];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';

switch ( algo )
Expand Down Expand Up @@ -908,7 +907,8 @@ int x16r_4way_hash_generic( void* output, const void* input, int thrid )
int x16r_4way_hash( void* output, const void* input, int thrid )
{
uint8_t hash[64*4] __attribute__ ((aligned (64)));
if ( !x16r_4way_hash_generic( hash, input, thrid ) )
if ( !x16r_4way_hash_generic( hash, input, thrid, x16r_hash_order,
X16R_HASH_FUNC_COUNT ) )
return 0;

memcpy( output, hash, 32 );
Expand All @@ -924,7 +924,6 @@ int scanhash_x16r_4way( struct work *work, uint32_t max_nonce,
{
uint32_t hash[16*4] __attribute__ ((aligned (64)));
uint32_t vdata[20*4] __attribute__ ((aligned (64)));
uint32_t bedata1[2];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
Expand All @@ -937,20 +936,18 @@ int scanhash_x16r_4way( struct work *work, uint32_t max_nonce,

if ( bench ) ptarget[7] = 0x0cff;

bedata1[0] = bswap_32( pdata[1] );
bedata1[1] = bswap_32( pdata[2] );

static __thread uint32_t s_ntime = UINT32_MAX;
const uint32_t ntime = bswap_32( pdata[17] );
if ( s_ntime != ntime )
static __thread uint32_t saved_height = UINT32_MAX;
if ( work->height != saved_height )
{
x16_r_s_getAlgoString( (const uint8_t*)bedata1, x16r_hash_order );
s_ntime = ntime;
if ( opt_debug && !thr_id )
applog( LOG_INFO, "Hash order %s Ntime %08x", x16r_hash_order, ntime );
vdata[1] = bswap_32( pdata[1] );
vdata[2] = bswap_32( pdata[2] );
saved_height = work->height;
x16_r_s_getAlgoString( (const uint8_t*)(&vdata[1]), x16r_hash_order );
if ( !opt_quiet && !thr_id )
applog( LOG_INFO, "hash order %s", x16r_hash_order );
}

x16r_4way_prehash( vdata, pdata );
x16r_4way_prehash( vdata, pdata, x16r_hash_order );
*noncev = mm256_intrlv_blend_32(
_mm256_set_epi32( n+3, 0, n+2, 0, n+1, 0, n, 0 ), *noncev );
do
Expand All @@ -973,10 +970,10 @@ int scanhash_x16r_4way( struct work *work, uint32_t max_nonce,

#elif defined (X16R_2WAY)

void x16r_2x64_prehash( void *vdata, void *pdata )
void x16r_2x64_prehash( void *vdata, void *pdata, const char *hash_order )
{
uint32_t edata[20] __attribute__ ((aligned (64)));
const char elem = x16r_hash_order[0];
const char elem = hash_order[0];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';

switch ( algo )
Expand Down Expand Up @@ -1051,7 +1048,8 @@ void x16r_2x64_prehash( void *vdata, void *pdata )
}
}

int x16r_2x64_hash_generic( void* output, const void* input, int thrid )
int x16r_2x64_hash_generic( void* output, const void* input, int thrid,
const char *hash_order, const int func_count )
{
uint32_t vhash[20*2] __attribute__ ((aligned (64)));
uint32_t hash0[20] __attribute__ ((aligned (32)));
Expand All @@ -1064,9 +1062,9 @@ int x16r_2x64_hash_generic( void* output, const void* input, int thrid )

dintrlv_2x64( hash0, hash1, input, 640 );

for ( int i = 0; i < 16; i++ )
for ( int i = 0; i < func_count; i++ )
{
const char elem = x16r_hash_order[i];
const char elem = hash_order[i];
const uint8_t algo = elem >= 'A' ? elem - 'A' + 10 : elem - '0';

switch ( algo )
Expand Down Expand Up @@ -1313,7 +1311,8 @@ int x16r_2x64_hash_generic( void* output, const void* input, int thrid )
int x16r_2x64_hash( void* output, const void* input, int thrid )
{
uint8_t hash[64*2] __attribute__ ((aligned (64)));
if ( !x16r_2x64_hash_generic( hash, input, thrid ) )
if ( !x16r_2x64_hash_generic( hash, input, thrid, x16r_hash_order,
X16R_HASH_FUNC_COUNT ) )
return 0;

memcpy( output, hash, 32 );
Expand All @@ -1327,7 +1326,6 @@ int scanhash_x16r_2x64( struct work *work, uint32_t max_nonce,
{
uint32_t hash[16*2] __attribute__ ((aligned (64)));
uint32_t vdata[20*2] __attribute__ ((aligned (64)));
uint32_t bedata1[2];
uint32_t *pdata = work->data;
uint32_t *ptarget = work->target;
const uint32_t first_nonce = pdata[19];
Expand All @@ -1340,20 +1338,18 @@ int scanhash_x16r_2x64( struct work *work, uint32_t max_nonce,

if ( bench ) ptarget[7] = 0x0cff;

bedata1[0] = bswap_32( pdata[1] );
bedata1[1] = bswap_32( pdata[2] );

static __thread uint32_t s_ntime = UINT32_MAX;
const uint32_t ntime = bswap_32( pdata[17] );
if ( s_ntime != ntime )
static __thread uint32_t saved_height = UINT32_MAX;
if ( work->height != saved_height )
{
x16_r_s_getAlgoString( (const uint8_t*)bedata1, x16r_hash_order );
s_ntime = ntime;
if ( opt_debug && !thr_id )
applog( LOG_INFO, "Hash order %s Ntime %08x", x16r_hash_order, ntime );
vdata[1] = bswap_32( pdata[1] );
vdata[2] = bswap_32( pdata[2] );
saved_height = work->height;
x16_r_s_getAlgoString( (const uint8_t*)(&vdata[1]), x16r_hash_order );
if ( !opt_quiet && !thr_id )
applog( LOG_INFO, "hash order %s", x16r_hash_order );
}

x16r_2x64_prehash( vdata, pdata );
x16r_2x64_prehash( vdata, pdata, x16r_hash_order );
*noncev = v128_intrlv_blend_32( v128_set32( n+1, 0, n, 0 ), *noncev );
do
{
Expand Down
Loading

0 comments on commit 045b42b

Please sign in to comment.