rottenwheel/monero
1
0
Fork 0
mirror of https://codeberg.org/anoncontributorxmr/monero.git synced 2024-11-25 17:02:26 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

crypto: plug CNv4 JIT into cn_slow_hash

Browse source 
Enabled by setting the MONERO_USE_CNV4_JIT env var to 1
This commit is contained in:
moneromooo-monero 2019-02-16 17:47:08 +00:00
parent 78ab59ea44
commit 24d281c324
No known key found for this signature in database
GPG key ID: 686F07454D6CEFC3
3 changed files with 102 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
CMakeLists.txt
View file

@ -44,6 +44,8 @@ message(STATUS "CMake version ${CMAKE_VERSION}")
project(monero)
enable_language(C ASM)
function (die msg)
if (NOT WIN32)
string(ASCII 27 Esc)

9
src/crypto/CMakeLists.txt
View file

@ -45,6 +45,8 @@ set(crypto_sources
random.c
skein.c
slow-hash.c
CryptonightR_JIT.c
CryptonightR_template.S
tree-hash.c)
set(crypto_headers)
@ -66,7 +68,9 @@ set(crypto_private_headers
oaes_lib.h
random.h
skein.h
skein_port.h)
skein_port.h
CryptonightR_JIT.h
CryptonightR_template.h)
monero_private_headers(cncrypto
${crypto_private_headers})
@ -101,4 +105,5 @@ if (ANDROID OR IOS)
endif()
endif()
# cheat because cmake and ccache hate each other
set_property(SOURCE CryptonightR_template.S PROPERTY LANGUAGE C)

94
src/crypto/slow-hash.c
View file

@ -40,6 +40,10 @@
#include "oaes_lib.h"
#include "variant2_int_sqrt.h"
#include "variant4_random_math.h"
#include "CryptonightR_JIT.h"
#include <errno.h>
#include <string.h>
#define MEMORY (1 << 21) // 2MB scratchpad
#define ITER (1 << 20)
@ -51,6 +55,16 @@
extern void aesb_single_round(const uint8_t *in, uint8_t *out, const uint8_t *expandedKey);
extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *expandedKey);
static void local_abort(const char *msg)
{
fprintf(stderr, "%s\n", msg);
#ifdef NDEBUG
_exit(1);
#else
abort();
#endif
}
#define VARIANT1_1(p) \
do if (variant == 1) \
{ \
@ -253,11 +267,18 @@ extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *ex
#define VARIANT4_RANDOM_MATH_INIT() \
v4_reg r[9]; \
struct V4_Instruction code[NUM_INSTRUCTIONS_MAX + 1]; \
int jit = use_v4_jit(); \
do if (variant >= 4) \
{ \
for (int i = 0; i < 4; ++i) \
V4_REG_LOAD(r + i, (uint8_t*)(state.hs.w + 12) + sizeof(v4_reg) * i); \
v4_random_math_init(code, height); \
if (jit) \
{ \
int ret = v4_generate_JIT_code(code, hp_jitfunc, 4096); \
if (ret < 0) \
local_abort("Error generating CryptonightR code"); \
} \
} while (0)
#define VARIANT4_RANDOM_MATH(a, b, r, _b, _b1) \
@ -279,7 +300,10 @@ extern void aesb_pseudo_round(const uint8_t *in, uint8_t *out, const uint8_t *ex
V4_REG_LOAD(r + 7, _b1); \
V4_REG_LOAD(r + 8, (uint64_t*)(_b1) + 1); \
\
v4_random_math(code, r); \
if (jit) \
(*hp_jitfunc)(r); \
else \
v4_random_math(code, r); \
\
memcpy(t, a, sizeof(uint64_t) * 2); \
\
@ -409,6 +433,9 @@ union cn_slow_hash_state
THREADV uint8_t *hp_state = NULL;
THREADV int hp_allocated = 0;
THREADV v4_random_math_JIT_func hp_jitfunc = NULL;
THREADV uint8_t *hp_jitfunc_memory = NULL;
THREADV int hp_jitfunc_allocated = 0;
#if defined(_MSC_VER)
#define cpuid(info,x) __cpuidex(info,x,0)
@ -467,6 +494,30 @@ STATIC INLINE int force_software_aes(void)
return use;
}
STATIC INLINE int use_v4_jit(void)
{
#if defined(__x86_64__)
static int use = -1;
if (use != -1)
return use;
const char *env = getenv("MONERO_USE_CNV4_JIT");
if (!env) {
use = 0;
}
else if (!strcmp(env, "0") || !strcmp(env, "no")) {
use = 0;
}
else {
use = 1;
}
return use;
#else
return 0;
#endif
}
STATIC INLINE int check_aes_hw(void)
{
int cpuid_results[4];
@ -718,6 +769,33 @@ void slow_hash_allocate_state(void)
hp_allocated = 0;
hp_state = (uint8_t *) malloc(MEMORY);
}
#if defined(_MSC_VER) || defined(__MINGW32__)
hp_jitfunc_memory = (uint8_t *) VirtualAlloc(hp_jitfunc_memory, 4096 + 4095,
MEM_COMMIT | MEM_RESERVE, PAGE_EXECUTE_READWRITE);
#else
#if defined(__APPLE__) || defined(__FreeBSD__) || defined(__OpenBSD__) || \
defined(__DragonFly__) || defined(__NetBSD__)
hp_jitfunc_memory = mmap(0, 4096 + 4095, PROT_READ | PROT_WRITE | PROT_EXEC,
MAP_PRIVATE | MAP_ANON, 0, 0);
#else
hp_jitfunc_memory = mmap(0, 4096 + 4095, PROT_READ | PROT_WRITE | PROT_EXEC,
MAP_PRIVATE | MAP_ANONYMOUS, 0, 0);
#endif
if(hp_jitfunc_memory == MAP_FAILED)
hp_jitfunc_memory = NULL;
#endif
hp_jitfunc_allocated = 1;
if (hp_jitfunc_memory == NULL)
{
hp_jitfunc_allocated = 0;
hp_jitfunc_memory = malloc(4096 + 4095);
}
hp_jitfunc = (v4_random_math_JIT_func)((size_t)(hp_jitfunc_memory + 4095) & ~4095);
#if !(defined(_MSC_VER) || defined(__MINGW32__))
mprotect(hp_jitfunc, 4096, PROT_READ | PROT_WRITE | PROT_EXEC);
#endif
}
/**
@ -740,8 +818,22 @@ void slow_hash_free_state(void)
#endif
}
if(!hp_jitfunc_allocated)
free(hp_jitfunc_memory);
else
{
#if defined(_MSC_VER) || defined(__MINGW32__)
VirtualFree(hp_jitfunc_memory, 0, MEM_RELEASE);
#else
munmap(hp_jitfunc_memory, 4096 + 4095);
#endif
}
hp_state = NULL;
hp_allocated = 0;
hp_jitfunc = NULL;
hp_jitfunc_memory = NULL;
hp_jitfunc_allocated = 0;
}
/**