Permalink
Browse files

veltor update, 10x faster :p

From Alexis work, sib hash rate 200% also..
1 parent 3eba451 commit 36aedbb48e2af906560e9c76c652361796ba9062 @tpruvot committed Nov 3, 2016
Showing with 971 additions and 1,071 deletions.
  1. +3 −3 ccminer.cpp
  2. +2 −2 neoscrypt/cuda_vectors.h
  3. +911 −1,013 x11/cuda_streebog.cu
  4. +10 −9 x11/sib.cu
  5. +45 −44 x11/veltor.cu
View
@@ -246,16 +246,16 @@ Options:\n\
skein Skein SHA2 (Skeincoin)\n\
skein2 Double Skein (Woodcoin)\n\
s3 S3 (1Coin)\n\
+ vanilla Blake256-8 (VNL)\n\
veltor Thorsriddle streebog\n\
+ whirlcoin Old Whirlcoin (Whirlpool algo)\n\
+ whirlpool Whirlpool algo\n\
x11evo Permuted x11 (Revolver)\n\
x11 X11 (DarkCoin)\n\
x13 X13 (MaruCoin)\n\
x14 X14\n\
x15 X15\n\
x17 X17\n\
- vanilla Blake256-8 (VNL)\n\
- whirlcoin Old Whirlcoin (Whirlpool algo)\n\
- whirlpool Whirlpool algo\n\
zr5 ZR5 (ZiftrCoin)\n\
-d, --devices Comma separated list of CUDA devices to use.\n\
Device IDs start counting from 0! Alternatively takes\n\
@@ -482,7 +482,7 @@ static __forceinline__ __device__ uint32_t rotateR(uint32_t vec4, uint32_t shift
// require a uint32_t[9] ret array
// note: djm neoscrypt implementation is near the limits of gpu capabilities
// and weird behaviors can happen when tuning device functions code...
-__device__ void shift256R(uint32_t* ret, const uint8 &vec4, uint32_t shift)
+__device__ static void shift256R(uint32_t* ret, const uint8 &vec4, uint32_t shift)
{
uint8_t *v = (uint8_t*) &vec4.s0;
uint8_t *r = (uint8_t*) ret;
@@ -496,7 +496,7 @@ __device__ void shift256R(uint32_t* ret, const uint8 &vec4, uint32_t shift)
#else
// same for SM 3.5+, really faster ?
-__device__ void shift256R(uint32_t* ret, const uint8 &vec4, uint32_t shift)
+__device__ static void shift256R(uint32_t* ret, const uint8 &vec4, uint32_t shift)
{
uint32_t truc = 0, truc2 = cuda_swab32(vec4.s7), truc3 = 0;
asm("shf.r.clamp.b32 %0, %1, %2, %3;" : "=r"(truc) : "r"(truc3), "r"(truc2), "r"(shift));
Oops, something went wrong.

0 comments on commit 36aedbb

Please sign in to comment.