GOSTSec
/
ccminer
mirror of https://github.com/GOSTSec/ccminer
1
0
Fork 0
Code Issues Releases Wiki Activity
Browse Source

x13: use tsiv hamsi implementation (+70KH)

2upstream
Tanguy Pruvot 10 years ago
parent
3e8457d76c
commit
71f9003901
3 changed files with 147 additions and 189 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. 3
      Makefile.am
  2. 2
      ccminer.vcxproj
  3. 329
      x13/cuda_x13_hamsi512.cu

3
Makefile.am
Unescape View File

@ -93,6 +93,9 @@ x11/cuda_x11_luffa512.o: x11/cuda_x11_luffa512.cu
x11/cuda_x11_luffa512_Cubehash.o: x11/cuda_x11_luffa512_Cubehash.cu x11/cuda_x11_luffa512_Cubehash.o: x11/cuda_x11_luffa512_Cubehash.cu
$(NVCC) $(nvcc_FLAGS) --maxrregcount=80 -o $@ -c $< $(NVCC) $(nvcc_FLAGS) --maxrregcount=80 -o $@ -c $<
x13/cuda_x13_hamsi512.o: x13/cuda_x13_hamsi512.cu
$(NVCC) $(nvcc_FLAGS) --maxrregcount=72 -o $@ -c $<
x17/cuda_x17_sha512.o: x17/cuda_x17_sha512.cu x17/cuda_x17_sha512.o: x17/cuda_x17_sha512.cu
$(NVCC) $(nvcc_FLAGS) --maxrregcount=80 -o $@ -c $< $(NVCC) $(nvcc_FLAGS) --maxrregcount=80 -o $@ -c $<

2
ccminer.vcxproj
Unescape View File

@ -447,7 +447,7 @@
<CudaCompile Include="x11\x11.cu"> <CudaCompile Include="x11\x11.cu">
</CudaCompile> </CudaCompile>
<CudaCompile Include="x13\cuda_x13_hamsi512.cu"> <CudaCompile Include="x13\cuda_x13_hamsi512.cu">
<MaxRegCount>80</MaxRegCount> <MaxRegCount>72</MaxRegCount>
</CudaCompile> </CudaCompile>
<CudaCompile Include="x13\cuda_x13_fugue512.cu"> <CudaCompile Include="x13\cuda_x13_fugue512.cu">
</CudaCompile> </CudaCompile>

329
x13/cuda_x13_hamsi512.cu
Unescape View File

@ -1,49 +1,21 @@
/* /*
* Quick and dirty addition of Hamsi-512 for X13 * Quick Hamsi-512 for X13
* * by tsiv - 2014
* Built on cbuchner1's implementation, actual hashing code
* heavily based on phm's sgminer
*
*/ */
/* #include <stdio.h>
* X13 kernel implementation. #include <stdint.h>
* #include <memory.h>
* ==========================(LICENSE BEGIN)============================
*
* Copyright (c) 2014 phm
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sublicense, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be
* included in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
* ===========================(LICENSE END)=============================
*
* @author phm <phm@inbox.com>
*/
#include "cuda_helper.h" #include "cuda_helper.h"
// aus heavy.cu typedef unsigned char BitSequence;
extern cudaError_t MyStreamSynchronize(cudaStream_t stream, int situation, int thr_id);
__device__ __constant__ static __constant__ uint32_t d_alpha_n[32];
static const uint32_t d_alpha_n[] = { static __constant__ uint32_t d_alpha_f[32];
static __constant__ uint32_t d_T512[64][16];
static const uint32_t alpha_n[] = {
SPH_C32(0xff00f0f0), SPH_C32(0xccccaaaa), SPH_C32(0xf0f0cccc), SPH_C32(0xff00f0f0), SPH_C32(0xccccaaaa), SPH_C32(0xf0f0cccc),
SPH_C32(0xff00aaaa), SPH_C32(0xccccaaaa), SPH_C32(0xf0f0ff00), SPH_C32(0xff00aaaa), SPH_C32(0xccccaaaa), SPH_C32(0xf0f0ff00),
SPH_C32(0xaaaacccc), SPH_C32(0xf0f0ff00), SPH_C32(0xf0f0cccc), SPH_C32(0xaaaacccc), SPH_C32(0xf0f0ff00), SPH_C32(0xf0f0cccc),
@ -57,8 +29,7 @@ static const uint32_t d_alpha_n[] = {
SPH_C32(0xff00aaaa), SPH_C32(0xccccf0f0) SPH_C32(0xff00aaaa), SPH_C32(0xccccf0f0)
}; };
__device__ __constant__ static const uint32_t alpha_f[] = {
static const uint32_t d_alpha_f[] = {
SPH_C32(0xcaf9639c), SPH_C32(0x0ff0f9c0), SPH_C32(0x639c0ff0), SPH_C32(0xcaf9639c), SPH_C32(0x0ff0f9c0), SPH_C32(0x639c0ff0),
SPH_C32(0xcaf9f9c0), SPH_C32(0x0ff0f9c0), SPH_C32(0x639ccaf9), SPH_C32(0xcaf9f9c0), SPH_C32(0x0ff0f9c0), SPH_C32(0x639ccaf9),
SPH_C32(0xf9c00ff0), SPH_C32(0x639ccaf9), SPH_C32(0x639c0ff0), SPH_C32(0xf9c00ff0), SPH_C32(0x639ccaf9), SPH_C32(0x639c0ff0),
@ -106,73 +77,73 @@ static const uint32_t d_alpha_f[] = {
#define hamsi_s1F mF #define hamsi_s1F mF
#define SBOX(a, b, c, d) { \ #define SBOX(a, b, c, d) { \
uint32_t t; \ uint32_t t; \
t = (a); \ t = (a); \
(a) &= (c); \ (a) &= (c); \
(a) ^= (d); \ (a) ^= (d); \
(c) ^= (b); \ (c) ^= (b); \
(c) ^= (a); \ (c) ^= (a); \
(d) |= t; \ (d) |= t; \
(d) ^= (b); \ (d) ^= (b); \
t ^= (c); \ t ^= (c); \
(b) = (d); \ (b) = (d); \
(d) |= t; \ (d) |= t; \
(d) ^= (a); \ (d) ^= (a); \
(a) &= (b); \ (a) &= (b); \
t ^= (a); \ t ^= (a); \
(b) ^= (d); \ (b) ^= (d); \
(b) ^= t; \ (b) ^= t; \
(a) = (c); \ (a) = (c); \
(c) = (b); \ (c) = (b); \
(b) = (d); \ (b) = (d); \
(d) = SPH_T32(~t); \ (d) = SPH_T32(~t); \
} }
#define HAMSI_L(a, b, c, d) { \ #define HAMSI_L(a, b, c, d) { \
(a) = ROTL32(a, 13); \ (a) = ROTL32(a, 13); \
(c) = ROTL32(c, 3); \ (c) = ROTL32(c, 3); \
(b) ^= (a) ^ (c); \ (b) ^= (a) ^ (c); \
(d) ^= (c) ^ SPH_T32((a) << 3); \ (d) ^= (c) ^ ((a) << 3); \
(b) = ROTL32(b, 1); \ (b) = ROTL32(b, 1); \
(d) = ROTL32(d, 7); \ (d) = ROTL32(d, 7); \
(a) ^= (b) ^ (d); \ (a) ^= (b) ^ (d); \
(c) ^= (d) ^ SPH_T32((b) << 7); \ (c) ^= (d) ^ ((b) << 7); \
(a) = ROTL32(a, 5); \ (a) = ROTL32(a, 5); \
(c) = ROTL32(c, 22); \ (c) = ROTL32(c, 22); \
} }
#define ROUND_BIG(rc, alpha) { \ #define ROUND_BIG(rc, alpha) { \
hamsi_s00 ^= alpha[0x00]; \ hamsi_s00 ^= alpha[0x00]; \
hamsi_s01 ^= alpha[0x01] ^ (uint32_t)(rc); \
hamsi_s02 ^= alpha[0x02]; \
hamsi_s03 ^= alpha[0x03]; \
hamsi_s04 ^= alpha[0x04]; \
hamsi_s05 ^= alpha[0x05]; \
hamsi_s06 ^= alpha[0x06]; \
hamsi_s07 ^= alpha[0x07]; \
hamsi_s08 ^= alpha[0x08]; \ hamsi_s08 ^= alpha[0x08]; \
hamsi_s09 ^= alpha[0x09]; \
hamsi_s0A ^= alpha[0x0A]; \
hamsi_s0B ^= alpha[0x0B]; \
hamsi_s0C ^= alpha[0x0C]; \
hamsi_s0D ^= alpha[0x0D]; \
hamsi_s0E ^= alpha[0x0E]; \
hamsi_s0F ^= alpha[0x0F]; \
hamsi_s10 ^= alpha[0x10]; \ hamsi_s10 ^= alpha[0x10]; \
hamsi_s11 ^= alpha[0x11]; \
hamsi_s12 ^= alpha[0x12]; \
hamsi_s13 ^= alpha[0x13]; \
hamsi_s14 ^= alpha[0x14]; \
hamsi_s15 ^= alpha[0x15]; \
hamsi_s16 ^= alpha[0x16]; \
hamsi_s17 ^= alpha[0x17]; \
hamsi_s18 ^= alpha[0x18]; \ hamsi_s18 ^= alpha[0x18]; \
hamsi_s01 ^= alpha[0x01] ^ (uint32_t)(rc); \
hamsi_s09 ^= alpha[0x09]; \
hamsi_s11 ^= alpha[0x11]; \
hamsi_s19 ^= alpha[0x19]; \ hamsi_s19 ^= alpha[0x19]; \
hamsi_s02 ^= alpha[0x02]; \
hamsi_s0A ^= alpha[0x0A]; \
hamsi_s12 ^= alpha[0x12]; \
hamsi_s1A ^= alpha[0x1A]; \ hamsi_s1A ^= alpha[0x1A]; \
hamsi_s03 ^= alpha[0x03]; \
hamsi_s0B ^= alpha[0x0B]; \
hamsi_s13 ^= alpha[0x13]; \
hamsi_s1B ^= alpha[0x1B]; \ hamsi_s1B ^= alpha[0x1B]; \
hamsi_s04 ^= alpha[0x04]; \
hamsi_s0C ^= alpha[0x0C]; \
hamsi_s14 ^= alpha[0x14]; \
hamsi_s1C ^= alpha[0x1C]; \ hamsi_s1C ^= alpha[0x1C]; \
hamsi_s05 ^= alpha[0x05]; \
hamsi_s0D ^= alpha[0x0D]; \
hamsi_s15 ^= alpha[0x15]; \
hamsi_s1D ^= alpha[0x1D]; \ hamsi_s1D ^= alpha[0x1D]; \
hamsi_s06 ^= alpha[0x06]; \
hamsi_s0E ^= alpha[0x0E]; \
hamsi_s16 ^= alpha[0x16]; \
hamsi_s1E ^= alpha[0x1E]; \ hamsi_s1E ^= alpha[0x1E]; \
hamsi_s07 ^= alpha[0x07]; \
hamsi_s0F ^= alpha[0x0F]; \
hamsi_s17 ^= alpha[0x17]; \
hamsi_s1F ^= alpha[0x1F]; \ hamsi_s1F ^= alpha[0x1F]; \
SBOX(hamsi_s00, hamsi_s08, hamsi_s10, hamsi_s18); \ SBOX(hamsi_s00, hamsi_s08, hamsi_s10, hamsi_s18); \
SBOX(hamsi_s01, hamsi_s09, hamsi_s11, hamsi_s19); \ SBOX(hamsi_s01, hamsi_s09, hamsi_s11, hamsi_s19); \
@ -198,30 +169,16 @@ static const uint32_t d_alpha_f[] = {
#define P_BIG { \ #define P_BIG { \
ROUND_BIG(0, d_alpha_n); \ for( int r = 0; r < 6; r++ ) \
ROUND_BIG(1, d_alpha_n); \ ROUND_BIG(r, d_alpha_n); \
ROUND_BIG(2, d_alpha_n); \
ROUND_BIG(3, d_alpha_n); \
ROUND_BIG(4, d_alpha_n); \
ROUND_BIG(5, d_alpha_n); \
} }
#define PF_BIG { \ #define PF_BIG { \
ROUND_BIG(0, d_alpha_f); \ for( int r = 0; r < 12; r++ ) \
ROUND_BIG(1, d_alpha_f); \ ROUND_BIG(r, d_alpha_f); \
ROUND_BIG(2, d_alpha_f); \
ROUND_BIG(3, d_alpha_f); \
ROUND_BIG(4, d_alpha_f); \
ROUND_BIG(5, d_alpha_f); \
ROUND_BIG(6, d_alpha_f); \
ROUND_BIG(7, d_alpha_f); \
ROUND_BIG(8, d_alpha_f); \
ROUND_BIG(9, d_alpha_f); \
ROUND_BIG(10, d_alpha_f); \
ROUND_BIG(11, d_alpha_f); \
} }
#define T_BIG { \ #define T_BIG { \
/* order is important */ \ /* order is important */ \
cF = (h[0xF] ^= hamsi_s17); \ cF = (h[0xF] ^= hamsi_s17); \
cE = (h[0xE] ^= hamsi_s16); \ cE = (h[0xE] ^= hamsi_s16); \
@ -241,8 +198,8 @@ static const uint32_t d_alpha_f[] = {
c0 = (h[0x0] ^= hamsi_s00); \ c0 = (h[0x0] ^= hamsi_s00); \
} }
__device__ __constant__
static const uint32_t d_T512[64][16] = { static const uint32_t T512[64][16] = {
{ SPH_C32(0xef0b0270), SPH_C32(0x3afd0000), SPH_C32(0x5dae0000), { SPH_C32(0xef0b0270), SPH_C32(0x3afd0000), SPH_C32(0x5dae0000),
SPH_C32(0x69490000), SPH_C32(0x9b0f3c06), SPH_C32(0x4405b5f9), SPH_C32(0x69490000), SPH_C32(0x9b0f3c06), SPH_C32(0x4405b5f9),
SPH_C32(0x66140a51), SPH_C32(0x924f5d0a), SPH_C32(0xc96b0030), SPH_C32(0x66140a51), SPH_C32(0x924f5d0a), SPH_C32(0xc96b0030),
@ -629,53 +586,8 @@ static const uint32_t d_T512[64][16] = {
SPH_C32(0xe7e00a94) } SPH_C32(0xe7e00a94) }
}; };
#define INPUT_BIG { \ __global__
const uint32_t *tp = &d_T512[0][0]; \ void x13_hamsi512_gpu_hash_64(int threads, uint32_t startNounce, uint64_t *g_hash, uint32_t *g_nonceVector)
unsigned u, v; \
m0 = 0; \
m1 = 0; \
m2 = 0; \
m3 = 0; \
m4 = 0; \
m5 = 0; \
m6 = 0; \
m7 = 0; \
m8 = 0; \
m9 = 0; \
mA = 0; \
mB = 0; \
mC = 0; \
mD = 0; \
mE = 0; \
mF = 0; \
for (u = 0; u < 8; u ++) { \
unsigned db = buf(u); \
for (v = 0; v < 8; v ++, db >>= 1) { \
uint32_t dm = SPH_T32(-(uint32_t)(db & 1)); \
m0 ^= dm & *tp ++; \
m1 ^= dm & *tp ++; \
m2 ^= dm & *tp ++; \
m3 ^= dm & *tp ++; \
m4 ^= dm & *tp ++; \
m5 ^= dm & *tp ++; \
m6 ^= dm & *tp ++; \
m7 ^= dm & *tp ++; \
m8 ^= dm & *tp ++; \
m9 ^= dm & *tp ++; \
mA ^= dm & *tp ++; \
mB ^= dm & *tp ++; \
mC ^= dm & *tp ++; \
mD ^= dm & *tp ++; \
mE ^= dm & *tp ++; \
mF ^= dm & *tp ++; \
} \
} \
}
/***************************************************/
// Die Hash-Funktion
__global__ void x13_hamsi512_gpu_hash_64(int threads, uint32_t startNounce, uint64_t *g_hash, uint32_t *g_nonceVector)
{ {
int thread = (blockDim.x * blockIdx.x + threadIdx.x); int thread = (blockDim.x * blockIdx.x + threadIdx.x);
if (thread < threads) if (thread < threads)
@ -690,28 +602,73 @@ __global__ void x13_hamsi512_gpu_hash_64(int threads, uint32_t startNounce, uint
uint32_t c4 = SPH_C32(0x72672031), c5 = SPH_C32(0x302c2062), c6 = SPH_C32(0x75732032), c7 = SPH_C32(0x3434362c); uint32_t c4 = SPH_C32(0x72672031), c5 = SPH_C32(0x302c2062), c6 = SPH_C32(0x75732032), c7 = SPH_C32(0x3434362c);
uint32_t c8 = SPH_C32(0x20422d33), c9 = SPH_C32(0x30303120), cA = SPH_C32(0x4c657576), cB = SPH_C32(0x656e2d48); uint32_t c8 = SPH_C32(0x20422d33), c9 = SPH_C32(0x30303120), cA = SPH_C32(0x4c657576), cB = SPH_C32(0x656e2d48);
uint32_t cC = SPH_C32(0x65766572), cD = SPH_C32(0x6c65652c), cE = SPH_C32(0x2042656c), cF = SPH_C32(0x6769756d); uint32_t cC = SPH_C32(0x65766572), cD = SPH_C32(0x6c65652c), cE = SPH_C32(0x2042656c), cF = SPH_C32(0x6769756d);
uint32_t m0, m1, m2, m3, m4, m5, m6, m7; uint32_t m0, m1, m2, m3, m4, m5, m6, m7, m8, m9, mA, mB, mC, mD, mE, mF;
uint32_t m8, m9, mA, mB, mC, mD, mE, mF;
uint32_t h[16] = { c0, c1, c2, c3, c4, c5, c6, c7, c8, c9, cA, cB, cC, cD, cE, cF }; uint32_t h[16] = { c0, c1, c2, c3, c4, c5, c6, c7, c8, c9, cA, cB, cC, cD, cE, cF };
uint32_t *tp, db, dm;
#define buf(u) (h1[i+u])
#pragma unroll 8
for(int i = 0; i < 64; i += 8) { for(int i = 0; i < 64; i += 8) {
INPUT_BIG;
P_BIG; m0 = 0; m1 = 0; m2 = 0; m3 = 0; m4 = 0; m5 = 0; m6 = 0; m7 = 0;
m8 = 0; m9 = 0; mA = 0; mB = 0; mC = 0; mD = 0; mE = 0; mF = 0;
tp = &d_T512[0][0];
#pragma unroll 2
for (int u = 0; u < 8; u ++) {
db = h1[i+u];
#pragma unroll 2
for (int v = 0; v < 8; v ++, db >>= 1) {
dm = -(uint32_t)(db & 1);
m0 ^= dm & *(tp+ 0); m1 ^= dm & *(tp+ 1);
m2 ^= dm & *(tp+ 2); m3 ^= dm & *(tp+ 3);
m4 ^= dm & *(tp+ 4); m5 ^= dm & *(tp+ 5);
m6 ^= dm & *(tp+ 6); m7 ^= dm & *(tp+ 7);
m8 ^= dm & *(tp+ 8); m9 ^= dm & *(tp+ 9);
mA ^= dm & *(tp+10); mB ^= dm & *(tp+11);
mC ^= dm & *(tp+12); mD ^= dm & *(tp+13);
mE ^= dm & *(tp+14); mF ^= dm & *(tp+15);
tp += 16;
}
}
for( int r = 0; r < 6; r += 2 ) {
ROUND_BIG(r, d_alpha_n);
ROUND_BIG(r+1, d_alpha_n);
}
T_BIG; T_BIG;
} }
#undef buf tp = &d_T512[0][0] + 112;
#define buf(u) (u == 0 ? 0x80 : 0)
INPUT_BIG; m0 = *(tp+ 0); m1 = *(tp+ 1);
P_BIG; m2 = *(tp+ 2); m3 = *(tp+ 3);
m4 = *(tp+ 4); m5 = *(tp+ 5);
m6 = *(tp+ 6); m7 = *(tp+ 7);
m8 = *(tp+ 8); m9 = *(tp+ 9);
mA = *(tp+10); mB = *(tp+11);
mC = *(tp+12); mD = *(tp+13);
mE = *(tp+14); mF = *(tp+15);
for( int r = 0; r < 6; r += 2 ) {
ROUND_BIG(r, d_alpha_n);
ROUND_BIG(r+1, d_alpha_n);
}
T_BIG; T_BIG;
#undef buf tp = &d_T512[0][0] + 784;
#define buf(u) (u == 6 ? 2 : 0)
INPUT_BIG; m0 = *(tp+ 0); m1 = *(tp+ 1);
PF_BIG; m2 = *(tp+ 2); m3 = *(tp+ 3);
m4 = *(tp+ 4); m5 = *(tp+ 5);
m6 = *(tp+ 6); m7 = *(tp+ 7);
m8 = *(tp+ 8); m9 = *(tp+ 9);
mA = *(tp+10); mB = *(tp+11);
mC = *(tp+12); mD = *(tp+13);
mE = *(tp+14); mF = *(tp+15);
for( int r = 0; r < 12; r += 2 ) {
ROUND_BIG(r, d_alpha_f);
ROUND_BIG(r+1, d_alpha_f);
}
T_BIG; T_BIG;
#pragma unroll 16 #pragma unroll 16
@ -720,24 +677,22 @@ __global__ void x13_hamsi512_gpu_hash_64(int threads, uint32_t startNounce, uint
} }
} }
__host__
__host__ void x13_hamsi512_cpu_init(int thr_id, int threads) void x13_hamsi512_cpu_init(int thr_id, int threads)
{ {
cudaMemcpyToSymbol(d_alpha_n, alpha_n, sizeof(uint32_t)*32, 0, cudaMemcpyHostToDevice);
cudaMemcpyToSymbol(d_alpha_f, alpha_f, sizeof(uint32_t)*32, 0, cudaMemcpyHostToDevice);
CUDA_SAFE_CALL(cudaMemcpyToSymbol(d_T512, T512, sizeof(uint32_t)*64*16, 0, cudaMemcpyHostToDevice));
} }
__host__ void x13_hamsi512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order) __host__
void x13_hamsi512_cpu_hash_64(int thr_id, int threads, uint32_t startNounce, uint32_t *d_nonceVector, uint32_t *d_hash, int order)
{ {
const int threadsperblock = 256; const int threadsperblock = 256;
// berechne wie viele Thread Blocks wir brauchen
dim3 grid((threads + threadsperblock-1)/threadsperblock); dim3 grid((threads + threadsperblock-1)/threadsperblock);
dim3 block(threadsperblock); dim3 block(threadsperblock);
// Größe des dynamischen Shared Memory Bereichs x13_hamsi512_gpu_hash_64<<<grid, block>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector);
size_t shared_size = 0;
// fprintf(stderr, "threads=%d, %d blocks, %d threads per block, %d bytes shared\n", threads, grid.x, block.x, shared_size);
x13_hamsi512_gpu_hash_64<<<grid, block, shared_size>>>(threads, startNounce, (uint64_t*)d_hash, d_nonceVector);
MyStreamSynchronize(NULL, order, thr_id); MyStreamSynchronize(NULL, order, thr_id);
} }
Write Preview
Loading…
Cancel
Save