Skip to content

Commit

Permalink
* Move invariants out of loop (minor performance gain)
Browse files Browse the repository at this point in the history 
* Fix issue that miner could only work with diff 0.0039... or higher (not that that matters much now:)
* quick'n'dirty improve of hash pre-check
  • Loading branch information
@Neisklar
Neisklar committed Aug 19, 2013
1 parent b37f2b5 commit b1af442
Show file tree
Hide file tree
Showing 2 changed files with 163 additions and 56 deletions.
5 changes: 5 additions & 0 deletions cpu-miner.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -1301,6 +1301,11 @@ int main(int argc, char *argv[])
/* parse command line */
parse_cmdline(argc, argv);

if (opt_algo==ALGO_QUARK)
{
init_quarkhash_contexts();
}

pthread_mutex_init(&applog_lock, NULL);
pthread_mutex_init(&stats_lock, NULL);
pthread_mutex_init(&g_work_lock, NULL);
Expand Down
214 changes: 158 additions & 56 deletions quark.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -11,84 +11,119 @@
#include "sph_keccak.h"
#include "sph_skein.h"


/* Move init out of loop, so init once externally, and then use one single memcpy with that bigger memory block */
typedef struct {
sph_blake512_context blake1, blake2;
sph_bmw512_context bmw1, bmw2;
sph_groestl512_context groestl1, groestl2;
sph_skein512_context skein1, skein2;
sph_jh512_context jh1, jh2;
sph_keccak512_context keccak1, keccak2;
} quarkhash_context_holder;

quarkhash_context_holder base_contexts;

void init_quarkhash_contexts()
{
sph_blake512_init(&base_contexts.blake1);
sph_bmw512_init(&base_contexts.bmw1);
sph_groestl512_init(&base_contexts.groestl1);
sph_skein512_init(&base_contexts.skein1);
sph_groestl512_init(&base_contexts.groestl2);
sph_jh512_init(&base_contexts.jh1);
sph_blake512_init(&base_contexts.blake2);
sph_bmw512_init(&base_contexts.bmw2);
sph_keccak512_init(&base_contexts.keccak1);
sph_skein512_init(&base_contexts.skein2);
sph_keccak512_init(&base_contexts.keccak2);
sph_jh512_init(&base_contexts.jh2);
}

static void quarkhash(void *state, const void *input)
{
sph_blake512_context ctx_blake;
sph_bmw512_context ctx_bmw;
sph_groestl512_context ctx_groestl;
sph_jh512_context ctx_jh;
sph_keccak512_context ctx_keccak;
sph_skein512_context ctx_skein;
static unsigned char pblank[1];
// sph_blake512_context ctx_blake;
// sph_bmw512_context ctx_bmw;
// sph_groestl512_context ctx_groestl;
// sph_jh512_context ctx_jh;
// sph_keccak512_context ctx_keccak;
// sph_skein512_context ctx_skein;
// static unsigned char pblank[1];

quarkhash_context_holder ctx;

uint32_t mask = 8;
uint32_t zero = 0;

//these uint512 in the c++ source of the client are backed by an array of uint32
uint32_t hashA[16], hashB[16];


//do one memcopy to get fresh contexts, its faster even with a larger block then issuing 9 memcopies
memcpy(&ctx, &base_contexts, sizeof(base_contexts));


sph_blake512_init(&ctx_blake);
sph_blake512 (&ctx_blake, input, 80);
sph_blake512_close (&ctx_blake, hashA); //0
// sph_blake512_init(&ctx.blake1);
sph_blake512 (&ctx.blake1, input, 80);
sph_blake512_close (&ctx.blake1, hashA); //0

sph_bmw512_init(&ctx_bmw);
sph_bmw512 (&ctx_bmw, hashA, 64); //0
sph_bmw512_close(&ctx_bmw, hashB); //1
// sph_bmw512_init(&ctx.bmw1);
sph_bmw512 (&ctx.bmw1, hashA, 64); //0
sph_bmw512_close(&ctx.bmw1, hashB); //1

if ((hashB[0] & mask) != zero) //1
{
sph_groestl512_init(&ctx_groestl);
sph_groestl512 (&ctx_groestl, hashB, 64); //1
sph_groestl512_close(&ctx_groestl, hashA); //2
// sph_groestl512_init(&ctx.groestl1);
sph_groestl512 (&ctx.groestl1, hashB, 64); //1
sph_groestl512_close(&ctx.groestl1, hashA); //2
}
else
{
sph_skein512_init(&ctx_skein);
sph_skein512 (&ctx_skein, hashB, 64); //1
sph_skein512_close(&ctx_skein, hashA); //2
// sph_skein512_init(&ctx.skein1);
sph_skein512 (&ctx.skein1, hashB, 64); //1
sph_skein512_close(&ctx.skein1, hashA); //2
}

sph_groestl512_init(&ctx_groestl);
sph_groestl512 (&ctx_groestl, hashA, 64); //2
sph_groestl512_close(&ctx_groestl, hashB); //3
// sph_groestl512_init(&ctx.groestl2);
sph_groestl512 (&ctx.groestl2, hashA, 64); //2
sph_groestl512_close(&ctx.groestl2, hashB); //3

sph_jh512_init(&ctx_jh);
sph_jh512 (&ctx_jh, hashB, 64); //3
sph_jh512_close(&ctx_jh, hashA); //4
// sph_jh512_init(&ctx.jh1);
sph_jh512 (&ctx.jh1, hashB, 64); //3
sph_jh512_close(&ctx.jh1, hashA); //4

if ((hashA[0] & mask) != zero) //4
{
sph_blake512_init(&ctx_blake);
sph_blake512 (&ctx_blake, hashA, 64); //
sph_blake512_close(&ctx_blake, hashB); //5
// sph_blake512_init(&ctx.blake2);
sph_blake512 (&ctx.blake2, hashA, 64); //
sph_blake512_close(&ctx.blake2, hashB); //5
}
else
{
sph_bmw512_init(&ctx_bmw);
sph_bmw512 (&ctx_bmw, hashA, 64); //4
sph_bmw512_close(&ctx_bmw, hashB); //5
// sph_bmw512_init(&ctx.bmw2);
sph_bmw512 (&ctx.bmw2, hashA, 64); //4
sph_bmw512_close(&ctx.bmw2, hashB); //5
}

sph_keccak512_init(&ctx_keccak);
sph_keccak512 (&ctx_keccak,hashB, 64); //5
sph_keccak512_close(&ctx_keccak, hashA); //6
// sph_keccak512_init(&ctx.keccak1);
sph_keccak512 (&ctx.keccak1, hashB, 64); //5
sph_keccak512_close(&ctx.keccak1, hashA); //6

sph_skein512_init(&ctx_skein);
sph_skein512 (&ctx_skein, hashA, 64); //6
sph_skein512_close(&ctx_skein, hashB); //7
// sph_skein512_init(&ctx.skein2);
sph_skein512 (&ctx.skein2, hashA, 64); //6
sph_skein512_close(&ctx.skein2, hashB); //7

if ((hashB[0] & mask) != zero) //7
{
sph_keccak512_init(&ctx_keccak);
sph_keccak512 (&ctx_keccak, hashB, 64); //
sph_keccak512_close(&ctx_keccak, hashA); //8
// sph_keccak512_init(&ctx.keccak2);
sph_keccak512 (&ctx.keccak2, hashB, 64); //
sph_keccak512_close(&ctx.keccak2, hashA); //8
}
else
{
sph_jh512_init(&ctx_jh);
sph_jh512 (&ctx_jh, hashB, 64); //7
sph_jh512_close(&ctx_jh, hashA); //8
// sph_jh512_init(&ctx.jh2);
sph_jh512 (&ctx.jh2, hashB, 64); //7
sph_jh512_close(&ctx.jh2, hashA); //8
}

memcpy(state, hashA, 32);
Expand Down Expand Up @@ -129,21 +164,88 @@ int scanhash_quark(int thr_id, uint32_t *pdata, const uint32_t *ptarget,
// applog(LOG_DEBUG, "Thr: %02d, firstN: %08x, maxN: %08x, ToDo: %d", thr_id, first_nonce, max_nonce, max_nonce-first_nonce);
// }

/* I'm to lazy to put the loop in an inline function... so dirty copy'n'paste.... */
/* i know that i could set a variable, but i don't know how the compiler will optimize it, not that then the cpu needs to load the value *everytime* in a register */
if (ptarget[7]==0) {
do {
pdata[19] = ++n;
be32enc(&endiandata[19], n);
quarkhash(hash64, &endiandata);
if (((hash64[7]&0xFFFFFFFF)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
}
else if (ptarget[7]<=0xF)
{
do {
pdata[19] = ++n;
be32enc(&endiandata[19], n);
quarkhash(hash64, &endiandata);
if (((hash64[7]&0xFFFFFFF0)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
}
else if (ptarget[7]<=0xFF)
{
do {
pdata[19] = ++n;
be32enc(&endiandata[19], n);
quarkhash(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);
}
else if (ptarget[7]<=0xFFF)
{
do {
pdata[19] = ++n;
be32enc(&endiandata[19], n);
quarkhash(hash64, &endiandata);
if (((hash64[7]&0xFFFFF000)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);

}
else if (ptarget[7]<=0xFFFF)
{
do {
pdata[19] = ++n;
be32enc(&endiandata[19], n);
quarkhash(hash64, &endiandata);
if (((hash64[7]&0xFFFF0000)==0) &&
fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);

}
else
{
do {
pdata[19] = ++n;
be32enc(&endiandata[19], n);
quarkhash(hash64, &endiandata);
if (fulltest(hash64, ptarget)) {
*hashes_done = n - first_nonce + 1;
return true;
}
} while (n < max_nonce && !work_restart[thr_id].restart);
}



do {

pdata[19] = ++n;
be32enc(&endiandata[19], n);
quarkhash(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;
Expand Down

0 comments on commit b1af442

Please sign in to comment.