Ecosystem/xmrig
Ecosystem/xmrig
1
0
Fork 0
mirror of https://github.com/xmrig/xmrig.git synced 2025-01-09 12:29:24 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge pull request #2751 from SChernykh/dev

Browse source 
VAES crash fixes
This commit is contained in:
xmrig 2021-11-30 09:49:37 +07:00 committed by GitHub
commit e673d541c1
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 42 additions and 29 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
src/backend/cpu/platform/BasicCpuInfo.cpp
View file

@ -140,7 +140,7 @@ static inline bool has_osxsave() { return has_feature(PROCESSOR_INFO,
static inline bool has_aes_ni() { return has_feature(PROCESSOR_INFO, ECX_Reg, 1 << 25); }
static inline bool has_avx() { return has_feature(PROCESSOR_INFO, ECX_Reg, 1 << 28) && has_osxsave() && has_xcr_avx(); }
static inline bool has_avx2() { return has_feature(EXTENDED_FEATURES, EBX_Reg, 1 << 5) && has_osxsave() && has_xcr_avx(); }
static inline bool has_vaes() { return has_feature(EXTENDED_FEATURES, ECX_Reg, 1 << 9); }
static inline bool has_vaes() { return has_feature(EXTENDED_FEATURES, ECX_Reg, 1 << 9) && has_osxsave() && has_xcr_avx(); }
static inline bool has_avx512f() { return has_feature(EXTENDED_FEATURES, EBX_Reg, 1 << 16) && has_osxsave() && has_xcr_avx512(); }
static inline bool has_bmi2() { return has_feature(EXTENDED_FEATURES, EBX_Reg, 1 << 8); }
static inline bool has_pdpe1gb() { return has_feature(PROCESSOR_EXT_INFO, EDX_Reg, 1 << 26); }

40
src/crypto/cn/CryptoNight_x86_vaes.cpp
View file

@ -179,16 +179,16 @@ NOINLINE void cn_explode_scratchpad_vaes(cryptonight_ctx* ctx)
if (props.half_mem() && !ctx->first_half) {
const __m256i* p = reinterpret_cast<const __m256i*>(ctx->save_state);
xin01 = _mm256_load_si256(p + 0);
xin23 = _mm256_load_si256(p + 1);
xin45 = _mm256_load_si256(p + 2);
xin67 = _mm256_load_si256(p + 3);
xin01 = _mm256_loadu_si256(p + 0);
xin23 = _mm256_loadu_si256(p + 1);
xin45 = _mm256_loadu_si256(p + 2);
xin67 = _mm256_loadu_si256(p + 3);
}
else {
xin01 = _mm256_load_si256(reinterpret_cast<const __m256i*>(input + 4));
xin23 = _mm256_load_si256(reinterpret_cast<const __m256i*>(input + 6));
xin45 = _mm256_load_si256(reinterpret_cast<const __m256i*>(input + 8));
xin67 = _mm256_load_si256(reinterpret_cast<const __m256i*>(input + 10));
xin01 = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(input + 4));
xin23 = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(input + 6));
xin45 = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(input + 8));
xin67 = _mm256_loadu_si256(reinterpret_cast<const __m256i*>(input + 10));
}
constexpr int output_increment = 64 / sizeof(__m256i);
@ -228,10 +228,10 @@ NOINLINE void cn_explode_scratchpad_vaes(cryptonight_ctx* ctx)
if (props.half_mem() && ctx->first_half) {
__m256i* p = reinterpret_cast<__m256i*>(ctx->save_state);
_mm256_store_si256(p + 0, xin01);
_mm256_store_si256(p + 1, xin23);
_mm256_store_si256(p + 2, xin45);
_mm256_store_si256(p + 3, xin67);
_mm256_storeu_si256(p + 0, xin01);
_mm256_storeu_si256(p + 1, xin23);
_mm256_storeu_si256(p + 2, xin45);
_mm256_storeu_si256(p + 3, xin67);
}
_mm256_zeroupper();
@ -347,10 +347,10 @@ NOINLINE void cn_implode_scratchpad_vaes(cryptonight_ctx* ctx)
vaes_genkey(reinterpret_cast<__m128i*>(output) + 2, &k0, &k1, &k2, &k3, &k4, &k5, &k6, &k7, &k8, &k9);
xout01 = _mm256_load_si256(output + 2);
xout23 = _mm256_load_si256(output + 3);
xout45 = _mm256_load_si256(output + 4);
xout67 = _mm256_load_si256(output + 5);
xout01 = _mm256_loadu_si256(output + 2);
xout23 = _mm256_loadu_si256(output + 3);
xout45 = _mm256_loadu_si256(output + 4);
xout67 = _mm256_loadu_si256(output + 5);
const __m256i* input_begin = input;
for (size_t part = 0; part < (props.half_mem() ? 2 : 1); ++part) {
@ -390,10 +390,10 @@ NOINLINE void cn_implode_scratchpad_vaes(cryptonight_ctx* ctx)
}
}
_mm256_store_si256(output + 2, xout01);
_mm256_store_si256(output + 3, xout23);
_mm256_store_si256(output + 4, xout45);
_mm256_store_si256(output + 5, xout67);
_mm256_storeu_si256(output + 2, xout01);
_mm256_storeu_si256(output + 3, xout23);
_mm256_storeu_si256(output + 4, xout45);
_mm256_storeu_si256(output + 5, xout67);
_mm256_zeroupper();
}

29
src/crypto/ghostrider/ghostrider.cpp
View file

@ -61,6 +61,10 @@
# include <intrin.h>
#endif
#ifdef XMRIG_OS_WIN
# include <Windows.h>
#endif
#define CORE_HASH(i, x) static void h##i(const uint8_t* data, size_t size, uint8_t* output) \
{ \
sph_##x##_context ctx; \
@ -328,7 +332,16 @@ void benchmark()
LOG_VERBOSE("%24s | N | Hashrate", "Algorithm");
LOG_VERBOSE("-------------------------|-----|-------------");
# ifdef XMRIG_OS_WIN
LARGE_INTEGER timer_freq;
QueryPerformanceFrequency(&timer_freq);
auto measure_time = []() { LARGE_INTEGER t; QueryPerformanceCounter(&t); return t.QuadPart; };
auto delta_time = [&timer_freq](LONGLONG t1, LONGLONG t2) { return static_cast<double>(t2 - t1) / timer_freq.QuadPart; };
# else
using namespace std::chrono;
auto measure_time = []() { return high_resolution_clock::now(); };
auto delta_time = [](const high_resolution_clock::time_point& t1, const high_resolution_clock::time_point& t2) { return duration_cast<nanoseconds>(t2 - t1).count() / 1e9; };
# endif
for (uint32_t algo = 0; algo < 6; ++algo) {
for (uint64_t step : { 1, 2, 4}) {
@ -339,20 +352,20 @@ void benchmark()
auto f = CnHash::fn(cn_hash[algo], av[step], Assembly::AUTO);
const high_resolution_clock::time_point start_time = high_resolution_clock::now();
auto start_time = measure_time();
double min_dt = 1e10;
for (uint32_t iter = 0;; ++iter) {
const high_resolution_clock::time_point t1 = high_resolution_clock::now();
auto t1 = measure_time();
// Stop after 15 milliseconds, but only if at least 10 iterations were done
if ((iter >= 10) && (duration_cast<milliseconds>(t1 - start_time).count() >= 15)) {
if ((iter >= 10) && (delta_time(start_time, t1) >= 0.015)) {
break;
}
f(buf, sizeof(buf), hash, ctx, 0);
const double dt = duration_cast<nanoseconds>(high_resolution_clock::now() - t1).count() / 1e9;
const double dt = delta_time(t1, measure_time());
if (dt < min_dt) {
min_dt = dt;
}
@ -388,14 +401,14 @@ void benchmark()
auto f = CnHash::fn(cn_hash[algo], av[step], Assembly::AUTO);
const high_resolution_clock::time_point start_time = high_resolution_clock::now();
auto start_time = measure_time();
double min_dt = 1e10;
for (uint32_t iter = 0;; ++iter) {
const high_resolution_clock::time_point t1 = high_resolution_clock::now();
auto t1 = measure_time();
// Stop after 30 milliseconds, but only if at least 10 iterations were done
if ((iter >= 10) && (duration_cast<milliseconds>(t1 - start_time).count() >= 30)) {
if ((iter >= 10) && (delta_time(start_time, t1) >= 0.03)) {
break;
}
@ -403,7 +416,7 @@ void benchmark()
f(buf, sizeof(buf), hash, ctx, 0);
helper->wait();
const double dt = duration_cast<nanoseconds>(high_resolution_clock::now() - t1).count() / 1e9;
const double dt = delta_time(t1, measure_time());
if (dt < min_dt) {
min_dt = dt;
}