popov/cpuminer-opt-gpu
1
0
Fork 0
mirror of https://github.com/JayDDee/cpuminer-opt.git synced 2025-09-17 23:44:27 +00:00
Code Issues Packages Projects Releases Wiki Activity

v3.10.5

Browse Source
This commit is contained in:
Jay D Dee
2019-12-21 13:19:29 -05:00
parent a17ff6f189
commit c65b0ff7a6
72 changed files with 9090 additions and 1336 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1081
algo/quark/hmq1725-4way.c
View File

File diff suppressed because it is too large Load Diff

7
algo/quark/hmq1725-gate.c
View File

@@ -2,7 +2,10 @@
bool register_hmq1725_algo( algo_gate_t* gate )
{
#if defined(HMQ1725_4WAY)
#if defined(HMQ1725_8WAY)
gate->scanhash = (void*)&scanhash_hmq1725_8way;
gate->hash = (void*)&hmq1725_8way_hash;
#elif defined(HMQ1725_4WAY)
gate->scanhash = (void*)&scanhash_hmq1725_4way;
gate->hash = (void*)&hmq1725_4way_hash;
#else
@@ -10,7 +13,7 @@ bool register_hmq1725_algo( algo_gate_t* gate )
gate->scanhash = (void*)&scanhash_hmq1725;
gate->hash = (void*)&hmq1725hash;
#endif
gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT;
gate->optimizations = SSE2_OPT | AES_OPT | AVX2_OPT | AVX512_OPT;
opt_target_factor = 65536.0;
return true;
};

14
algo/quark/hmq1725-gate.h
View File

@@ -4,13 +4,21 @@
#include "algo-gate-api.h"
#include <stdint.h>
#if defined(__AVX2__) && defined(__AES__)
// #define HMQ1725_4WAY 1
#if defined(__AVX512F__) && defined(__AVX512VL__) && defined(__AVX512DQ__) && defined(__AVX512BW__)
#define HMQ1725_8WAY 1
#elif defined(__AVX2__) && defined(__AES__)
#define HMQ1725_4WAY 1
#endif
bool register_hmq1725_algo( algo_gate_t* gate );
#if defined(HMQ1725_4WAY)
#if defined(HMQ1725_8WAY)
void hmq1725_8way_hash( void *state, const void *input );
int scanhash_hmq1725_8way( struct work *work, uint32_t max_nonce,
uint64_t *hashes_done, struct thr_info *mythr );
#elif defined(HMQ1725_4WAY)
void hmq1725_4way_hash( void *state, const void *input );
int scanhash_hmq1725_4way( struct work *work, uint32_t max_nonce,

6
algo/quark/hmq1725.c
View File

@@ -333,6 +333,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
if (((hash64[7]&0xFFFFFFFF)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
work_set_target_ratio( work, hash64 );
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
@@ -346,6 +347,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
if (((hash64[7]&0xFFFFFFF0)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
work_set_target_ratio( work, hash64 );
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
@@ -359,6 +361,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
if (((hash64[7]&0xFFFFFF00)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
work_set_target_ratio( work, hash64 );
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
@@ -372,6 +375,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
if (((hash64[7]&0xFFFFF000)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
work_set_target_ratio( work, hash64 );
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
@@ -386,6 +390,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
if (((hash64[7]&0xFFFF0000)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
work_set_target_ratio( work, hash64 );
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
@@ -399,6 +404,7 @@ int scanhash_hmq1725( struct work *work, uint32_t max_nonce,
hmq1725hash(hash64, endiandata);
if (fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
work_set_target_ratio( work, hash64 );
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);

14
algo/quark/quark-4way.c
View File

@@ -63,20 +63,6 @@ void quark_8way_hash( void *state, const void *input )
bmw512_8way_update( &ctx.bmw, vhash, 64 );
bmw512_8way_close( &ctx.bmw, vhash );
// AVX 512 cmpeq returns a bit mask instead of a vector mask.
// This should simplify things but the logic doesn't seem to be working.
// The problem appears to be related to the test to skip a hash if it isn't
// to be used. Skipping the test for all 8 way hashes seems to have
// fixed it. The hash selection blending works if the hash is produced
// but the hash wasn't being produced when it should.
// Both decisions are based on the same data, the __mmask8. It works
// as a blend mask but not in a logical comparison, maybe the type is the
// problem. Maybe a cast to int or movm is needed to make it work.
// It's now moot because the hash can only be skipped 1 in 256 iterations
// when hashing parallel 8 ways.
// The performance impact of the workaround should be negligible.
// It's a problem for another day.
vh_mask = _mm512_cmpeq_epi64_mask( _mm512_and_si512( vh[0], bit3_mask ),
zero );