GOSTSec
/
ccminer-gostd-lite
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity
Browse Source

quark/blake512 opt. pointed by sp without asm 

indeed, the pragma unroll doesnt always make things faster

asm part... to check later
master
Tanguy Pruvot 10 years ago
parent
438308b3a2
commit
c859041993
6 changed files with 16 additions and 15 deletions
Whitespace
Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL
Unified View
Diff Options
Show Stats Download Patch File Download Diff File
  1. 26
      quark/cuda_quark_blake512.cu
  2. 1
      x11/x11.cu
  3. 1
      x13/x13.cu
  4. 1
      x15/x14.cu
  5. 1
      x15/x15.cu
  6. 1
      x17/x17.cu

26
quark/cuda_quark_blake512.cu
Unescape View File

@ -59,18 +59,19 @@ const uint64_t c_u512[16] =
v[b] = ROTR( v[b] ^ v[c], 11); \ v[b] = ROTR( v[b] ^ v[c], 11); \
} }
__device__ static __device__ __forceinline__
void quark_blake512_compress( uint64_t *h, const uint64_t *block, const uint8_t ((*sigma)[16]), const uint64_t *u512, const int T0) void quark_blake512_compress( uint64_t *h, const uint64_t *block, const uint8_t ((*sigma)[16]), const uint64_t *u512, const int T0)
{ {
uint64_t v[16], m[16], i; uint64_t v[16];
uint64_t m[16];
#pragma unroll 16 #pragma unroll
for( i = 0; i < 16; i++) { for(int i=0; i < 16; i++) {
m[i] = cuda_swab64(block[i]); m[i] = cuda_swab64(block[i]);
} }
#pragma unroll 8 //#pragma unroll 8
for (i = 0; i < 8; i++) for(int i=0; i < 8; i++)
v[i] = h[i]; v[i] = h[i];
v[ 8] = u512[0]; v[ 8] = u512[0];
@ -83,7 +84,7 @@ void quark_blake512_compress( uint64_t *h, const uint64_t *block, const uint8_t
v[15] = u512[7]; v[15] = u512[7];
//#pragma unroll 16 //#pragma unroll 16
for( i = 0; i < 16; ++i ) for(int i=0; i < 16; i++)
{ {
/* column step */ /* column step */
G( 0, 4, 8, 12, 0 ); G( 0, 4, 8, 12, 0 );
@ -97,9 +98,14 @@ void quark_blake512_compress( uint64_t *h, const uint64_t *block, const uint8_t
G( 3, 4, 9, 14, 14 ); G( 3, 4, 9, 14, 14 );
} }
#pragma unroll 16 h[0] ^= v[0] ^ v[8];
for( i = 0; i < 16; ++i ) h[1] ^= v[1] ^ v[9];
h[i % 8] ^= v[i]; h[2] ^= v[2] ^ v[10];
h[3] ^= v[3] ^ v[11];
h[4] ^= v[4] ^ v[12];
h[5] ^= v[5] ^ v[13];
h[6] ^= v[6] ^ v[14];
h[7] ^= v[7] ^ v[15];
} }
// Hash-Padding // Hash-Padding

1
x11/x11.cu
Unescape View File

@ -25,7 +25,6 @@ static uint32_t *d_hash[8];
extern void quark_blake512_cpu_init(int thr_id, int threads); extern void quark_blake512_cpu_init(int thr_id, int threads);
extern void quark_blake512_cpu_setBlock_80(void *pdata); extern void quark_blake512_cpu_setBlock_80(void *pdata);
extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order); extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order);
extern void quark_blake512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
extern void quark_bmw512_cpu_init(int thr_id, int threads); extern void quark_bmw512_cpu_init(int thr_id, int threads);
extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order); extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);

1
x13/x13.cu
Unescape View File

@ -29,7 +29,6 @@ static uint32_t *d_hash[8];
extern void quark_blake512_cpu_init(int thr_id, int threads); extern void quark_blake512_cpu_init(int thr_id, int threads);
extern void quark_blake512_cpu_setBlock_80(void *pdata); extern void quark_blake512_cpu_setBlock_80(void *pdata);
extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order); extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order);
extern void quark_blake512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
extern void quark_bmw512_cpu_init(int thr_id, int threads); extern void quark_bmw512_cpu_init(int thr_id, int threads);
extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order); extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);

1
x15/x14.cu
Unescape View File

@ -32,7 +32,6 @@ static uint32_t *d_hash[8];
extern void quark_blake512_cpu_init(int thr_id, int threads); extern void quark_blake512_cpu_init(int thr_id, int threads);
extern void quark_blake512_cpu_setBlock_80(void *pdata); extern void quark_blake512_cpu_setBlock_80(void *pdata);
extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order); extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order);
extern void quark_blake512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
extern void quark_bmw512_cpu_init(int thr_id, int threads); extern void quark_bmw512_cpu_init(int thr_id, int threads);
extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order); extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);

1
x15/x15.cu
Unescape View File

@ -33,7 +33,6 @@ static uint32_t *d_hash[8];
extern void quark_blake512_cpu_init(int thr_id, int threads); extern void quark_blake512_cpu_init(int thr_id, int threads);
extern void quark_blake512_cpu_setBlock_80(void *pdata); extern void quark_blake512_cpu_setBlock_80(void *pdata);
extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order); extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order);
extern void quark_blake512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
extern void quark_bmw512_cpu_init(int thr_id, int threads); extern void quark_bmw512_cpu_init(int thr_id, int threads);
extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order); extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);

1
x17/x17.cu
Unescape View File

@ -36,7 +36,6 @@ static uint32_t *d_hash[8];
extern void quark_blake512_cpu_init(int thr_id, int threads); extern void quark_blake512_cpu_init(int thr_id, int threads);
extern void quark_blake512_cpu_setBlock_80(void *pdata); extern void quark_blake512_cpu_setBlock_80(void *pdata);
extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order); extern void quark_blake512_cpu_hash_80(int thr_id, int threads, uint32_t startNounce, uint32_t *d_hash, int order);
extern void quark_blake512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);
extern void quark_bmw512_cpu_init(int thr_id, int threads); extern void quark_bmw512_cpu_init(int thr_id, int threads);
extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order); extern void quark_bmw512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order);

Write Preview
Loading…
Cancel
Save