Ecosystem/xmrig
Ecosystem/xmrig
1
0
Fork 0
mirror of https://github.com/xmrig/xmrig.git synced 2024-11-18 10:01:06 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'evo' into beta

Browse source
This commit is contained in:
XMRig 2019-10-02 10:32:34 +07:00
commit 923d47fff0
32 changed files with 294 additions and 187 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

10
CHANGELOG.md
View file

@ -1,3 +1,9 @@
# v4.2.1-beta
- [#1150](https://github.com/xmrig/xmrig/issues/1150) Fixed build on FreeBSD.
- [#1175](https://github.com/xmrig/xmrig/issues/1175) Fixed support for systems where total count of NUMA nodes not equal usable nodes count.
- [#1199](https://github.com/xmrig/xmrig/issues/1199) Fixed excessive memory allocation for OpenCL threads with low intensity.
- [#1212](https://github.com/xmrig/xmrig/issues/1212) Fixed low RandomX performance after fast algorithm switching.
# v4.2.0-beta
- [#1202](https://github.com/xmrig/xmrig/issues/1202) Fixed algorithm verification in donate strategy.
- Added per pool option `coin` with single possible value `monero` for pools without algorithm negotiation, for upcoming Monero fork.
@ -23,6 +29,10 @@
- [#268](https://github.com/xmrig/xmrig-amd/pull/268) [#270](https://github.com/xmrig/xmrig-amd/pull/270) [#271](https://github.com/xmrig/xmrig-amd/pull/271) [#273](https://github.com/xmrig/xmrig-amd/pull/273) [#274](https://github.com/xmrig/xmrig-amd/pull/274) [#1171](https://github.com/xmrig/xmrig/pull/1171) Added RandomX support for OpenCL, thanks [@SChernykh](https://github.com/SChernykh).
- Algorithm `cn/wow` removed, as no longer alive.
# v3.2.0
- Added per pool option `coin` with single possible value `monero` for pools without algorithm negotiation, for upcoming Monero fork.
- [#1183](https://github.com/xmrig/xmrig/issues/1183) Fixed compatibility with systemd.
# v3.1.3
- [#1180](https://github.com/xmrig/xmrig/issues/1180) Fixed possible duplicated shares after algorithm switching.
- Fixed wrong config file permissions after write (only gcc builds on recent Windows 10 affected).

4
README.md
View file

@ -1,5 +1,7 @@
# XMRig
**:warning: [Monero will change PoW algorithm to RandomX on November 30.](https://github.com/xmrig/xmrig/issues/1204)**
[![Github All Releases](https://img.shields.io/github/downloads/xmrig/xmrig/total.svg)](https://github.com/xmrig/xmrig/releases)
[![GitHub release](https://img.shields.io/github/release/xmrig/xmrig/all.svg)](https://github.com/xmrig/xmrig/releases)
[![GitHub Release Date](https://img.shields.io/github/release-date-pre/xmrig/xmrig.svg)](https://github.com/xmrig/xmrig/releases)
@ -26,7 +28,7 @@ XMRig is a high performance RandomX and CryptoNight CPU miner, with official sup
* Clone with `git clone https://github.com/xmrig/xmrig.git` :hammer: [Build instructions](https://github.com/xmrig/xmrig/wiki/Build).
## Usage
Preferend way to configure miner is [JSON config file](src/config.json) as more flexible and human frendly, command line interface not cover all features, for example mining profiles for different algorithms. Most impotant options can be changed in runtime without miner restart by editing config or via API.
The preferred way to configure the miner is the [JSON config file](src/config.json) as it is more flexible and human friendly. The command line interface does not cover all features, such as mining profiles for different algorithms. Important options can be changed during runtime without miner restart by editing the config file or executing API calls.
### Options
```

48
doc/ALGORITHMS.md
View file

@ -1,34 +1,20 @@
# Algorithms
Since version 3 mining [algorithm](#algorithm-names) should specified for each pool separately (`algo` option), earlier versions was use one global `algo` option and per pool `variant` option (this option was removed in v3). If your pool support [mining algorithm negotiation](https://github.com/xmrig/xmrig-proxy/issues/168) you may not specify this option at all.
#### Example
```json
{
"pools": [
{
"url": "...",
"algo": "cn/r",
...
}
],
...
}
```
Algorithm can be defined in 3 ways:
#### Pools with mining algorithm negotiation support.
1. By pool, using algorithm negotiation, in this case no need specify algorithm on miner side.
2. Per pool `coin` option, currently only usable value for this option is `monero`.
3. Per pool `algo` option.
* [www.hashvault.pro](https://www.hashvault.pro/)
* [moneroocean.stream](https://moneroocean.stream)
## Algorithm names
Option `coin` useful for pools without algorithm negotiation support or daemon to allow automatically switch algorithm in next hard fork.
## Algorithm names
| Name | Memory | Version | Notes |
|------|--------|---------|-------|
| `rx/0` | 2 MB | 3.2.0+ | RandomX (Monero). |
| `argon2/chukwa` | 512 KB | 3.1.0+ | Argon2id (Chukwa). |
| `argon2/wrkz` | 256 KB | 3.1.0+ | Argon2id (WRKZ) |
| `rx/test` | 2 MB | 3.0.0+ | RandomX (reference configuration). |
| `rx/0` | 2 MB | 3.0.0+ | RandomX (reference configuration), reserved for future use. |
| `rx/wow` | 1 MB | 3.0.0+ | RandomWOW. |
| `rx/loki` | 2 MB | 3.0.0+ | RandomXL. |
| `cn/fast` | 2 MB | 3.0.0+ | CryptoNight variant 1 with half iterations. |
@ -50,3 +36,21 @@ Since version 3 mining [algorithm](#algorithm-names) should specified for each p
| `cn-lite/1` | 1 MB | 2.5.0+ | CryptoNight-Lite variant 1. |
| `cn-lite/0` | 1 MB | 0.8.0+ | CryptoNight-Lite variant 0. |
| `cn/0` | 2 MB | 0.5.0+ | CryptoNight (original). |
## Migration to v3
Since version 3 mining [algorithm](#algorithm-names) should specified for each pool separately (`algo` option), earlier versions was use one global `algo` option and per pool `variant` option (this option was removed in v3). If your pool support [mining algorithm negotiation](https://github.com/xmrig/xmrig-proxy/issues/168) you may not specify this option at all.
#### Example
```json
{
"pools": [
{
"url": "...",
"algo": "cn/r",
"coin": null
...
}
],
...
}
```

6
src/3rdparty/argon2/CMakeLists.txt vendored
View file

@ -22,7 +22,7 @@ set(ARGON2_X86_64_SOURCES arch/x86_64/lib/argon2-arch.c arch/x86_64/lib/cpu-flag
if (CMAKE_C_COMPILER_ID MATCHES MSVC)
function(add_feature_impl FEATURE MSVC_FLAG DEF)
add_library(argon2-${FEATURE} STATIC arch/x86_64/lib/argon2-${FEATURE}.c)
target_include_directories(argon2-${FEATURE} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
target_include_directories(argon2-${FEATURE} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../../)
target_include_directories(argon2-${FEATURE} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/lib)
set_target_properties(argon2-${FEATURE} PROPERTIES POSITION_INDEPENDENT_CODE True)
@ -38,7 +38,7 @@ if (CMAKE_C_COMPILER_ID MATCHES MSVC)
elseif (NOT XMRIG_ARM AND CMAKE_SIZEOF_VOID_P EQUAL 8)
function(add_feature_impl FEATURE GCC_FLAG DEF)
add_library(argon2-${FEATURE} STATIC arch/x86_64/lib/argon2-${FEATURE}.c)
target_include_directories(argon2-${FEATURE} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
target_include_directories(argon2-${FEATURE} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../../)
target_include_directories(argon2-${FEATURE} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/lib)
set_target_properties(argon2-${FEATURE} PROPERTIES POSITION_INDEPENDENT_CODE True)
@ -84,5 +84,5 @@ endif()
add_library(argon2 STATIC ${ARGON2_SOURCES})
target_link_libraries(argon2 ${ARGON2_LIBS})
target_include_directories(argon2 PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/include)
target_include_directories(argon2 PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../../)
target_include_directories(argon2 PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/lib)

2
src/3rdparty/argon2/lib/argon2.c vendored
View file

@ -15,7 +15,7 @@
#include <stdlib.h>
#include <stdio.h>
#include "argon2.h"
#include "3rdparty/argon2.h"
#include "encoding.h"
#include "core.h"

2
src/3rdparty/argon2/lib/core.h vendored
View file

@ -14,7 +14,7 @@
#ifndef ARGON2_CORE_H
#define ARGON2_CORE_H
#include "argon2.h"
#include "3rdparty/argon2.h"
#if defined(_MSC_VER)
#define ALIGN(n) __declspec(align(16))

2
src/3rdparty/argon2/lib/encoding.h vendored
View file

@ -1,6 +1,6 @@
#ifndef ENCODING_H
#define ENCODING_H
#include "argon2.h"
#include "3rdparty/argon2.h"
#define ARGON2_MAX_DECODED_LANES UINT32_C(255)
#define ARGON2_MIN_DECODED_SALT_LEN UINT32_C(8)

2
src/3rdparty/argon2/lib/impl-select.c vendored
View file

@ -3,7 +3,7 @@
#include "impl-select.h"
#include "argon2.h"
#include "3rdparty/argon2.h"
#define BENCH_SAMPLES 1024
#define BENCH_MEM_BLOCKS 512

1
src/backend/common/Workers.cpp
View file

@ -100,6 +100,7 @@ void xmrig::Workers<T>::start(const std::vector<T> &data)
}
d_ptr->hashrate = new Hashrate(m_workers.size());
Nonce::touch(T::backend());
for (Thread<T> *worker : m_workers) {
worker->start(Workers<T>::onReady);

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

@ -177,7 +177,7 @@ xmrig::HwlocCpuInfo::HwlocCpuInfo()
m_threads = countByType(m_topology, HWLOC_OBJ_PU);
m_cores = countByType(m_topology, HWLOC_OBJ_CORE);
m_nodes = std::max<size_t>(countByType(m_topology, HWLOC_OBJ_NUMANODE), 1);
m_nodes = std::max(hwloc_bitmap_weight(hwloc_topology_get_complete_nodeset(m_topology)), 1);
m_packages = countByType(m_topology, HWLOC_OBJ_PACKAGE);
if (m_nodes > 1) {

11
src/backend/opencl/OclBackend.cpp
View file

@ -35,6 +35,7 @@
#include "backend/opencl/OclConfig.h"
#include "backend/opencl/OclLaunchData.h"
#include "backend/opencl/OclWorker.h"
#include "backend/opencl/runners/tools/OclSharedState.h"
#include "backend/opencl/wrappers/OclContext.h"
#include "backend/opencl/wrappers/OclLib.h"
#include "base/io/log/Log.h"
@ -164,7 +165,7 @@ public:
}
inline void start()
inline void start(const Job &job)
{
LOG_INFO("%s use profile " BLUE_BG(WHITE_BOLD_S " %s ") WHITE_BOLD_S " (" CYAN_BOLD("%zu") WHITE_BOLD(" threads)") " scratchpad " CYAN_BOLD("%zu KB"),
tag,
@ -194,6 +195,8 @@ public:
i++;
}
OclSharedState::start(threads, job);
status.start(threads.size());
workers.start(threads);
}
@ -329,7 +332,7 @@ void xmrig::OclBackend::setJob(const Job &job)
return stop();
}
if (!d_ptr->context.init(d_ptr->devices, threads, job)) {
if (!d_ptr->context.init(d_ptr->devices, threads)) {
LOG_WARN("%s " RED_BOLD("disabled") YELLOW(" (OpenCL context unavailable)"), tag);
return stop();
@ -338,7 +341,7 @@ void xmrig::OclBackend::setJob(const Job &job)
stop();
d_ptr->threads = std::move(threads);
d_ptr->start();
d_ptr->start(job);
}
@ -371,6 +374,8 @@ void xmrig::OclBackend::stop()
d_ptr->workers.stop();
d_ptr->threads.clear();
OclSharedState::release();
LOG_INFO("%s" YELLOW(" stopped") BLACK_BOLD(" (%" PRIu64 " ms)"), tag, Chrono::steadyMSecs() - ts);
}

23
src/backend/opencl/OclConfig.cpp
View file

@ -193,32 +193,13 @@ std::vector<xmrig::OclLaunchData> xmrig::OclConfig::get(const Miner *miner, cons
continue;
}
# ifdef XMRIG_ALGO_RANDOMX
auto dataset = algorithm.family() == Algorithm::RANDOM_X ? std::make_shared<OclRxDataset>() : nullptr;
# endif
if (thread.threads().size() > 1) {
auto interleave = std::make_shared<OclInterleave>(thread.threads().size());
for (int64_t affinity : thread.threads()) {
OclLaunchData data(miner, algorithm, *this, platform, thread, devices[thread.index()], affinity);
data.interleave = interleave;
# ifdef XMRIG_ALGO_RANDOMX
data.dataset = dataset;
# endif
out.emplace_back(std::move(data));
out.emplace_back(miner, algorithm, *this, platform, thread, devices[thread.index()], affinity);
}
}
else {
OclLaunchData data(miner, algorithm, *this, platform, thread, devices[thread.index()], thread.threads().front());
# ifdef XMRIG_ALGO_RANDOMX
data.dataset = dataset;
# endif
out.emplace_back(std::move(data));
out.emplace_back(miner, algorithm, *this, platform, thread, devices[thread.index()], thread.threads().front());
}
}

12
src/backend/opencl/OclLaunchData.h
View file

@ -27,19 +27,14 @@
#define XMRIG_OCLLAUNCHDATA_H
#include "backend/opencl/OclInterleave.h"
#include "backend/opencl/OclThread.h"
#include "backend/opencl/runners/tools/OclSharedData.h"
#include "backend/opencl/wrappers/OclDevice.h"
#include "backend/opencl/wrappers/OclPlatform.h"
#include "crypto/common/Algorithm.h"
#include "crypto/common/Nonce.h"
#ifdef XMRIG_ALGO_RANDOMX
# include "backend/opencl/runners/tools/OclRxDataset.h"
#endif
using cl_context = struct _cl_context *;
@ -72,11 +67,6 @@ public:
const OclDevice device;
const OclPlatform platform;
const OclThread thread;
OclInterleavePtr interleave;
# ifdef XMRIG_ALGO_RANDOMX
OclRxDatasetPtr dataset;
# endif
};

20
src/backend/opencl/OclWorker.cpp
View file

@ -28,6 +28,8 @@
#include "backend/common/Tags.h"
#include "backend/opencl/runners/OclCnRunner.h"
#include "backend/opencl/runners/tools/OclSharedData.h"
#include "backend/opencl/runners/tools/OclSharedState.h"
#include "base/io/log/Log.h"
#include "base/tools/Chrono.h"
#include "core/Miner.h"
@ -75,7 +77,7 @@ xmrig::OclWorker::OclWorker(size_t id, const OclLaunchData &data) :
m_algorithm(data.algorithm),
m_miner(data.miner),
m_intensity(data.thread.intensity()),
m_interleave(data.interleave)
m_sharedData(OclSharedState::get(data.device.index()))
{
switch (m_algorithm.family()) {
case Algorithm::RANDOM_X:
@ -149,9 +151,7 @@ void xmrig::OclWorker::start()
while (Nonce::sequence(Nonce::OPENCL) > 0) {
if (!isReady()) {
if (m_interleave) {
m_interleave->setResumeCounter(0);
}
m_sharedData.setResumeCounter(0);
do {
std::this_thread::sleep_for(std::chrono::milliseconds(200));
@ -162,9 +162,7 @@ void xmrig::OclWorker::start()
break;
}
if (m_interleave) {
m_interleave->resumeDelay(m_id);
}
m_sharedData.resumeDelay(m_id);
if (!consumeJob()) {
return;
@ -172,9 +170,7 @@ void xmrig::OclWorker::start()
}
while (!Nonce::isOutdated(Nonce::OPENCL, m_job.sequence())) {
if (m_interleave) {
m_interleave->adjustDelay(m_id);
}
m_sharedData.adjustDelay(m_id);
const uint64_t t = Chrono::steadyMSecs();
@ -233,9 +229,7 @@ void xmrig::OclWorker::storeStats(uint64_t t)
m_count += m_intensity;
if (m_interleave) {
m_interleave->setRunTime(Chrono::steadyMSecs() - t);
}
m_sharedData.setRunTime(Chrono::steadyMSecs() - t);
Worker::storeStats();
}

2
src/backend/opencl/OclWorker.h
View file

@ -67,7 +67,7 @@ private:
const Miner *m_miner;
const uint32_t m_intensity;
IOclRunner *m_runner = nullptr;
OclInterleavePtr m_interleave;
OclSharedData &m_sharedData;
WorkerJob<1> m_job;
};

8
src/backend/opencl/opencl.cmake
View file

@ -14,7 +14,6 @@ if (WITH_OPENCL)
src/backend/opencl/OclCache.h
src/backend/opencl/OclConfig.h
src/backend/opencl/OclGenerator.h
src/backend/opencl/OclInterleave.h
src/backend/opencl/OclLaunchData.h
src/backend/opencl/OclThread.h
src/backend/opencl/OclThreads.h
@ -22,6 +21,8 @@ if (WITH_OPENCL)
src/backend/opencl/runners/OclBaseRunner.h
src/backend/opencl/runners/OclCnRunner.h
src/backend/opencl/runners/tools/OclCnR.h
src/backend/opencl/runners/tools/OclSharedData.h
src/backend/opencl/runners/tools/OclSharedState.h
src/backend/opencl/wrappers/OclContext.h
src/backend/opencl/wrappers/OclDevice.h
src/backend/opencl/wrappers/OclError.h
@ -42,7 +43,6 @@ if (WITH_OPENCL)
src/backend/opencl/OclBackend.cpp
src/backend/opencl/OclCache.cpp
src/backend/opencl/OclConfig.cpp
src/backend/opencl/OclInterleave.cpp
src/backend/opencl/OclLaunchData.cpp
src/backend/opencl/OclThread.cpp
src/backend/opencl/OclThreads.cpp
@ -50,6 +50,8 @@ if (WITH_OPENCL)
src/backend/opencl/runners/OclBaseRunner.cpp
src/backend/opencl/runners/OclCnRunner.cpp
src/backend/opencl/runners/tools/OclCnR.cpp
src/backend/opencl/runners/tools/OclSharedData.cpp
src/backend/opencl/runners/tools/OclSharedState.cpp
src/backend/opencl/wrappers/OclContext.cpp
src/backend/opencl/wrappers/OclDevice.cpp
src/backend/opencl/wrappers/OclError.cpp
@ -78,7 +80,6 @@ if (WITH_OPENCL)
src/backend/opencl/runners/OclRxBaseRunner.h
src/backend/opencl/runners/OclRxJitRunner.h
src/backend/opencl/runners/OclRxVmRunner.h
src/backend/opencl/runners/tools/OclRxDataset.h
)
list(APPEND SOURCES_BACKEND_OPENCL
@ -95,7 +96,6 @@ if (WITH_OPENCL)
src/backend/opencl/runners/OclRxBaseRunner.cpp
src/backend/opencl/runners/OclRxJitRunner.cpp
src/backend/opencl/runners/OclRxVmRunner.cpp
src/backend/opencl/runners/tools/OclRxDataset.cpp
)
endif()

17
src/backend/opencl/runners/OclBaseRunner.cpp
View file

@ -23,10 +23,11 @@
*/
#include "backend/opencl/runners/OclBaseRunner.h"
#include "backend/opencl/cl/OclSource.h"
#include "backend/opencl/OclCache.h"
#include "backend/opencl/OclLaunchData.h"
#include "backend/opencl/runners/OclBaseRunner.h"
#include "backend/opencl/runners/tools/OclSharedState.h"
#include "backend/opencl/wrappers/OclError.h"
#include "backend/opencl/wrappers/OclLib.h"
#include "base/io/log/Log.h"
@ -34,6 +35,9 @@
#include "crypto/common/VirtualMemory.h"
constexpr size_t oneGiB = 1024 * 1024 * 1024;
xmrig::OclBaseRunner::OclBaseRunner(size_t id, const OclLaunchData &data) :
m_algorithm(data.algorithm),
m_ctx(data.ctx),
@ -93,16 +97,17 @@ void xmrig::OclBaseRunner::build()
void xmrig::OclBaseRunner::init()
{
m_queue = OclLib::createCommandQueue(m_ctx, data().device.id());
m_queue = OclLib::createCommandQueue(m_ctx, data().device.id());
constexpr size_t oneGiB = 1024 * 1024 * 1024;
size_t size = bufferSize();
size_t size = align(bufferSize());
if (size < oneGiB && data().device.vendorId() == OCL_VENDOR_AMD && data().device.freeMemSize() >= oneGiB) {
size = oneGiB;
m_buffer = OclSharedState::get(data().device.index()).createBuffer(m_ctx, size, m_offset);
}
else {
m_buffer = OclLib::createBuffer(m_ctx, CL_MEM_READ_WRITE, size);
}
m_buffer = OclLib::createBuffer(m_ctx, CL_MEM_READ_WRITE, size);
m_input = createSubBuffer(CL_MEM_READ_ONLY | CL_MEM_HOST_WRITE_ONLY, Job::kMaxBlobSize);
m_output = createSubBuffer(CL_MEM_READ_WRITE, sizeof(cl_uint) * 0x100);
}

5
src/backend/opencl/runners/OclRxBaseRunner.cpp
View file

@ -30,6 +30,7 @@
#include "backend/opencl/kernels/rx/FindSharesKernel.h"
#include "backend/opencl/kernels/rx/HashAesKernel.h"
#include "backend/opencl/OclLaunchData.h"
#include "backend/opencl/runners/tools/OclSharedState.h"
#include "backend/opencl/wrappers/OclLib.h"
#include "base/net/stratum/Job.h"
#include "crypto/rx/Rx.h"
@ -75,6 +76,7 @@ xmrig::OclRxBaseRunner::~OclRxBaseRunner()
OclLib::release(m_hashes);
OclLib::release(m_rounding);
OclLib::release(m_scratchpads);
OclLib::release(m_dataset);
}
@ -120,7 +122,7 @@ void xmrig::OclRxBaseRunner::set(const Job &job, uint8_t *blob)
m_seed = job.seed();
auto dataset = Rx::dataset(job, 0);
enqueueWriteBuffer(data().dataset->get(), CL_TRUE, 0, dataset->size(), dataset->raw());
enqueueWriteBuffer(m_dataset, CL_TRUE, 0, dataset->size(), dataset->raw());
}
if (job.size() < Job::kMaxBlobSize) {
@ -177,4 +179,5 @@ void xmrig::OclRxBaseRunner::init()
m_hashes = createSubBuffer(CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS, 64 * m_intensity);
m_entropy = createSubBuffer(CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS, (128 + 2560) * m_intensity);
m_rounding = createSubBuffer(CL_MEM_READ_WRITE | CL_MEM_HOST_NO_ACCESS, sizeof(uint32_t) * m_intensity);
m_dataset = OclSharedState::get(data().device.index()).dataset();
}

1
src/backend/opencl/runners/OclRxBaseRunner.h
View file

@ -62,6 +62,7 @@ protected:
Blake2bHashRegistersKernel *m_blake2b_hash_registers_64 = nullptr;
Blake2bInitialHashKernel *m_blake2b_initial_hash = nullptr;
Buffer m_seed;
cl_mem m_dataset = nullptr;
cl_mem m_entropy = nullptr;
cl_mem m_hashes = nullptr;
cl_mem m_rounding = nullptr;

2
src/backend/opencl/runners/OclRxJitRunner.cpp
View file

@ -74,7 +74,7 @@ void xmrig::OclRxJitRunner::build()
}
m_randomx_run = new RxRunKernel(m_asmProgram);
m_randomx_run->setArgs(data().dataset->get(), m_scratchpads, m_registers, m_rounding, m_programs, m_intensity, m_algorithm);
m_randomx_run->setArgs(m_dataset, m_scratchpads, m_registers, m_rounding, m_programs, m_intensity, m_algorithm);
}

2
src/backend/opencl/runners/OclRxVmRunner.cpp
View file

@ -70,7 +70,7 @@ void xmrig::OclRxVmRunner::build()
m_init_vm->setArgs(m_entropy, m_vm_states, m_rounding);
m_execute_vm = new ExecuteVmKernel(m_program);
m_execute_vm->setArgs(m_vm_states, m_rounding, m_scratchpads, data().dataset->get(), m_intensity);
m_execute_vm->setArgs(m_vm_states, m_rounding, m_scratchpads, m_dataset, m_intensity);
}

86
src/backend/opencl/OclInterleave.cpp → src/backend/opencl/runners/tools/OclSharedData.cpp
View file

@ -23,17 +23,44 @@
*/
#include "backend/opencl/OclInterleave.h"
#include "backend/opencl/runners/tools/OclSharedData.h"
#include "backend/opencl/wrappers/OclLib.h"
#include "base/io/log/Log.h"
#include "base/tools/Chrono.h"
#include "crypto/rx/Rx.h"
#include "crypto/rx/RxDataset.h"
#include <algorithm>
#include <cinttypes>
#include <stdexcept>
#include <thread>
uint64_t xmrig::OclInterleave::adjustDelay(size_t id)
constexpr size_t oneGiB = 1024 * 1024 * 1024;
cl_mem xmrig::OclSharedData::createBuffer(cl_context context, size_t size, size_t &offset)
{
std::lock_guard<std::mutex> lock(m_mutex);
offset += size * m_offset++;
size = std::max(size * m_threads, oneGiB);
if (!m_buffer) {
m_buffer = OclLib::createBuffer(context, CL_MEM_READ_WRITE, size);
}
return OclLib::retain(m_buffer);
}
uint64_t xmrig::OclSharedData::adjustDelay(size_t id)
{
if (m_threads < 2) {
return 0;
}
const uint64_t t0 = Chrono::steadyMSecs();
uint64_t delay = 0;
@ -69,8 +96,12 @@ uint64_t xmrig::OclInterleave::adjustDelay(size_t id)
}
uint64_t xmrig::OclInterleave::resumeDelay(size_t id)
uint64_t xmrig::OclSharedData::resumeDelay(size_t id)
{
if (m_threads < 2) {
return 0;
}
uint64_t delay = 0;
{
@ -99,14 +130,28 @@ uint64_t xmrig::OclInterleave::resumeDelay(size_t id)
}
void xmrig::OclInterleave::setResumeCounter(uint32_t value)
void xmrig::OclSharedData::release()
{
OclLib::release(m_buffer);
# ifdef XMRIG_ALGO_RANDOMX
OclLib::release(m_dataset);
# endif
}
void xmrig::OclSharedData::setResumeCounter(uint32_t value)
{
if (m_threads < 2) {
return;
}
std::lock_guard<std::mutex> lock(m_mutex);
m_resumeCounter = value;
}
void xmrig::OclInterleave::setRunTime(uint64_t time)
void xmrig::OclSharedData::setRunTime(uint64_t time)
{
// averagingBias = 1.0 - only the last delta time is taken into account
// averagingBias = 0.5 - the last delta time has the same weight as all the previous ones combined
@ -116,3 +161,34 @@ void xmrig::OclInterleave::setRunTime(uint64_t time)
std::lock_guard<std::mutex> lock(m_mutex);
m_averageRunTime = m_averageRunTime * (1.0 - averagingBias) + time * averagingBias;
}
#ifdef XMRIG_ALGO_RANDOMX
cl_mem xmrig::OclSharedData::dataset() const
{
if (!m_dataset) {
throw std::runtime_error("RandomX dataset is not available");
}
return OclLib::retain(m_dataset);
}
void xmrig::OclSharedData::createDataset(cl_context ctx, const Job &job, bool host)
{
if (m_dataset) {
return;
}
cl_int ret;
if (host) {
auto dataset = Rx::dataset(job, 0);
m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR, RxDataset::maxSize(), dataset->raw(), &ret);
}
else {
m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY, RxDataset::maxSize(), nullptr, &ret);
}
}
#endif

40
src/backend/opencl/OclInterleave.h → src/backend/opencl/runners/tools/OclSharedData.h
View file

@ -22,42 +22,62 @@
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef XMRIG_OCLINTERLEAVE_H
#define XMRIG_OCLINTERLEAVE_H
#ifndef XMRIG_OCLSHAREDDATA_H
#define XMRIG_OCLSHAREDDATA_H
#include <memory>
#include <mutex>
using cl_context = struct _cl_context *;
using cl_mem = struct _cl_mem *;
namespace xmrig {
class OclInterleave
class Job;
class OclSharedData
{
public:
OclInterleave() = delete;
inline OclInterleave(size_t threads) : m_threads(threads) {}
OclSharedData() = default;
cl_mem createBuffer(cl_context context, size_t size, size_t &offset);
uint64_t adjustDelay(size_t id);
uint64_t resumeDelay(size_t id);
void release();
void setResumeCounter(uint32_t value);
void setRunTime(uint64_t time);
inline size_t threads() const { return m_threads; }
inline OclSharedData &operator++() { ++m_threads; return *this; }
# ifdef XMRIG_ALGO_RANDOMX
cl_mem dataset() const;
void createDataset(cl_context ctx, const Job &job, bool host);
# endif
private:
const size_t m_threads;
cl_mem m_buffer = nullptr;
double m_averageRunTime = 0.0;
double m_threshold = 0.95;
size_t m_offset = 0;
size_t m_threads = 0;
std::mutex m_mutex;
uint32_t m_resumeCounter = 0;
uint64_t m_timestamp = 0;
# ifdef XMRIG_ALGO_RANDOMX
cl_mem m_dataset = nullptr;
# endif
};
using OclInterleavePtr = std::shared_ptr<OclInterleave>;
} /* namespace xmrig */
#endif /* XMRIG_OCLINTERLEAVE_H */
#endif /* XMRIG_OCLSHAREDDATA_H */

62
src/backend/opencl/runners/tools/OclRxDataset.cpp → src/backend/opencl/runners/tools/OclSharedState.cpp
View file

@ -23,32 +23,52 @@
*/
#include "backend/opencl/runners/tools/OclRxDataset.h"
#include "backend/opencl/wrappers/OclLib.h"
#include "crypto/rx/Rx.h"
#include "crypto/rx/RxDataset.h"
#include "backend/opencl/runners/tools/OclSharedState.h"
#include "backend/opencl/runners/tools/OclSharedData.h"
void xmrig::OclRxDataset::createBuffer(cl_context ctx, const Job &job, bool host)
#include <cassert>
#include <map>
namespace xmrig {
static std::map<uint32_t, OclSharedData> map;
} // namespace xmrig
xmrig::OclSharedData &xmrig::OclSharedState::get(uint32_t index)
{
if (m_dataset) {
return;
}
cl_int ret;
if (host) {
auto dataset = Rx::dataset(job, 0);
m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY | CL_MEM_USE_HOST_PTR, RxDataset::maxSize(), dataset->raw(), &ret);
}
else {
m_dataset = OclLib::createBuffer(ctx, CL_MEM_READ_ONLY, RxDataset::maxSize(), nullptr, &ret);
}
return map[index];
}
xmrig::OclRxDataset::~OclRxDataset()
void xmrig::OclSharedState::release()
{
OclLib::release(m_dataset);
for (auto &kv : map) {
kv.second.release();
}
map.clear();
}
void xmrig::OclSharedState::start(const std::vector<OclLaunchData> &threads, const Job &job)
{
assert(map.empty());
for (const auto &data : threads) {
auto &sharedData = map[data.device.index()];
++sharedData;
# ifdef XMRIG_ALGO_RANDOMX
if (data.algorithm.family() == Algorithm::RANDOM_X) {
sharedData.createDataset(data.ctx, job, data.thread.isDatasetHost());
}
# endif
}
}

37
src/backend/opencl/runners/tools/OclRxDataset.h → src/backend/opencl/runners/tools/OclSharedState.h
View file

@ -22,47 +22,26 @@
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef XMRIG_OCLRXDATASET_H
#define XMRIG_OCLRXDATASET_H
#ifndef XMRIG_OCLSHAREDSTATE_H
#define XMRIG_OCLSHAREDSTATE_H
#include "base/tools/Object.h"
#include <memory>
using cl_context = struct _cl_context *;
using cl_mem = struct _cl_mem *;
#include "backend/opencl/OclLaunchData.h"
namespace xmrig {
class Job;
class OclRxDataset
class OclSharedState
{
public:
XMRIG_DISABLE_COPY_MOVE(OclRxDataset)
OclRxDataset() = default;
~OclRxDataset();
inline cl_mem get() const { return m_dataset; }
void createBuffer(cl_context ctx,const Job &job, bool host);
private:
cl_mem m_dataset = nullptr;
static OclSharedData &get(uint32_t index);
static void release();
static void start(const std::vector<OclLaunchData> &threads, const Job &job);
};
using OclRxDatasetPtr = std::shared_ptr<OclRxDataset>;
} /* namespace xmrig */
#endif /* XMRIG_OCLINTERLEAVE_H */
#endif /* XMRIG_OCLSHAREDSTATE_H */

9
src/backend/opencl/wrappers/OclContext.cpp
View file

@ -24,6 +24,7 @@
#include "backend/opencl/wrappers/OclContext.h"
#include "backend/opencl/runners/tools/OclSharedState.h"
#include "backend/opencl/wrappers/OclLib.h"
@ -42,7 +43,7 @@ xmrig::OclContext::~OclContext()
}
bool xmrig::OclContext::init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads, const Job &job)
bool xmrig::OclContext::init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads)
{
if (!m_ctx) {
std::vector<cl_device_id> ids(devices.size());
@ -59,12 +60,6 @@ bool xmrig::OclContext::init(const std::vector<OclDevice> &devices, std::vector<
for (OclLaunchData &data : threads) {
data.ctx = m_ctx;
# ifdef XMRIG_ALGO_RANDOMX
if (data.algorithm.family() == Algorithm::RANDOM_X) {
data.dataset->createBuffer(m_ctx, job, data.thread.isDatasetHost());
}
# endif
}
return true;

2
src/backend/opencl/wrappers/OclContext.h
View file

@ -49,7 +49,7 @@ public:
OclContext(const OclDevice &device);
~OclContext();
bool init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads, const Job &job);
bool init(const std::vector<OclDevice> &devices, std::vector<OclLaunchData> &threads);
inline bool isValid() const { return m_ctx != nullptr; }
inline cl_context ctx() const { return m_ctx; }

16
src/backend/opencl/wrappers/OclLib.cpp
View file

@ -73,6 +73,7 @@ static const char *kReleaseDevice = "clReleaseDevice";
static const char *kReleaseKernel = "clReleaseKernel";
static const char *kReleaseMemObject = "clReleaseMemObject";
static const char *kReleaseProgram = "clReleaseProgram";
static const char *kRetainMemObject = "clRetainMemObject";
static const char *kRetainProgram = "clRetainProgram";
static const char *kSetKernelArg = "clSetKernelArg";
static const char *kSetMemObjectDestructorCallback = "clSetMemObjectDestructorCallback";
@ -106,6 +107,7 @@ typedef cl_int (CL_API_CALL *releaseDevice_t)(cl_device_id device);
typedef cl_int (CL_API_CALL *releaseKernel_t)(cl_kernel);
typedef cl_int (CL_API_CALL *releaseMemObject_t)(cl_mem);
typedef cl_int (CL_API_CALL *releaseProgram_t)(cl_program);
typedef cl_int (CL_API_CALL *retainMemObject_t)(cl_mem);
typedef cl_int (CL_API_CALL *retainProgram_t)(cl_program);
typedef cl_int (CL_API_CALL *setKernelArg_t)(cl_kernel, cl_uint, size_t, const void *);
typedef cl_int (CL_API_CALL *setMemObjectDestructorCallback_t)(cl_mem, void (CL_CALLBACK *)(cl_mem, void *), void *);
@ -148,6 +150,7 @@ static releaseDevice_t pReleaseDevice = nu
static releaseKernel_t pReleaseKernel = nullptr;
static releaseMemObject_t pReleaseMemObject = nullptr;
static releaseProgram_t pReleaseProgram = nullptr;
static retainMemObject_t pRetainMemObject = nullptr;
static retainProgram_t pRetainProgram = nullptr;
static setKernelArg_t pSetKernelArg = nullptr;
static setMemObjectDestructorCallback_t pSetMemObjectDestructorCallback = nullptr;
@ -239,6 +242,7 @@ bool xmrig::OclLib::load()
DLSYM(SetMemObjectDestructorCallback);
DLSYM(CreateSubBuffer);
DLSYM(RetainProgram);
DLSYM(RetainMemObject);
# if defined(CL_VERSION_2_0)
uv_dlsym(&oclLib, kCreateCommandQueueWithProperties, reinterpret_cast<void**>(&pCreateCommandQueueWithProperties));
@ -670,6 +674,18 @@ cl_mem xmrig::OclLib::createSubBuffer(cl_mem buffer, cl_mem_flags flags, size_t
}
cl_mem xmrig::OclLib::retain(cl_mem memobj) noexcept
{
assert(pRetainMemObject != nullptr);
if (memobj != nullptr) {
pRetainMemObject(memobj);
}
return memobj;
}
cl_program xmrig::OclLib::createProgramWithBinary(cl_context context, cl_uint num_devices, const cl_device_id *device_list, const size_t *lengths, const unsigned char **binaries, cl_int *binary_status, cl_int *errcode_ret) noexcept
{
assert(pCreateProgramWithBinary != nullptr);

1
src/backend/opencl/wrappers/OclLib.h
View file

@ -79,6 +79,7 @@ public:
static cl_mem createBuffer(cl_context context, cl_mem_flags flags, size_t size, void *host_ptr, cl_int *errcode_ret) noexcept;
static cl_mem createSubBuffer(cl_mem buffer, cl_mem_flags flags, size_t offset, size_t size, cl_int *errcode_ret) noexcept;
static cl_mem createSubBuffer(cl_mem buffer, cl_mem_flags flags, size_t offset, size_t size);
static cl_mem retain(cl_mem memobj) noexcept;
static cl_program createProgramWithBinary(cl_context context, cl_uint num_devices, const cl_device_id *device_list, const size_t *lengths, const unsigned char **binaries, cl_int *binary_status, cl_int *errcode_ret) noexcept;
static cl_program createProgramWithSource(cl_context context, cl_uint count, const char **strings, const size_t *lengths, cl_int *errcode_ret) noexcept;
static cl_program retain(cl_program program) noexcept;

40
src/crypto/rx/Rx.cpp
View file

@ -47,6 +47,7 @@
#endif
#include <atomic>
#include <map>
#include <mutex>
#include <thread>
@ -96,12 +97,14 @@ class RxPrivate
public:
XMRIG_DISABLE_COPY_MOVE(RxPrivate)
inline RxPrivate()
inline RxPrivate() :
m_counter(0),
m_last(0)
{
m_async = new uv_async_t;
m_async->data = this;
uv_async_init(uv_default_loop(), m_async, RxPrivate::onReady);
uv_async_init(uv_default_loop(), m_async, [](uv_async_t *) { d_ptr->onReady(); });
# ifdef XMRIG_FEATURE_HWLOC
if (Cpu::info()->nodes() > 1) {
@ -130,11 +133,12 @@ public:
inline bool isNUMA() const { return m_numa; }
inline bool isReady(const Job &job) const { return m_ready == count() && m_algorithm == job.algorithm() && m_seed == job.seed(); }
inline const Algorithm &algorithm() const { return m_algorithm; }
inline const Buffer &seed() const { return m_seed; }
inline size_t count() const { return isNUMA() ? datasets.size() : 1; }
inline void asyncSend() { m_ready++; if (m_ready == count()) { uv_async_send(m_async); } }
inline uint64_t counter() { return m_counter.load(std::memory_order_relaxed); }
inline void asyncSend(uint64_t counter) { m_ready++; if (m_ready == count()) { m_last = counter; uv_async_send(m_async); } }
static void allocate(uint32_t nodeId)
{
@ -176,14 +180,14 @@ public:
}
static void initDataset(uint32_t nodeId, uint32_t threads)
static void initDataset(uint32_t nodeId, uint32_t threads, uint64_t counter)
{
std::lock_guard<std::mutex> lock(mutex);
const uint64_t ts = Chrono::steadyMSecs();
d_ptr->getOrAllocate(nodeId)->init(d_ptr->seed(), threads);
d_ptr->asyncSend();
d_ptr->asyncSend(counter);
LOG_INFO("%s" CYAN_BOLD("#%u") GREEN(" init done ") CYAN_BOLD("%zu/%zu") BLACK_BOLD(" (%" PRIu64 " ms)"), tag, nodeId, d_ptr->m_ready, d_ptr->count(), Chrono::steadyMSecs() - ts);
}
@ -222,32 +226,31 @@ public:
m_hugePages = hugePages;
m_listener = listener;
m_seed = job.seed();
}
inline bool isReady(const Job &job)
{
return m_ready == count() && m_algorithm == job.algorithm() && m_seed == job.seed();
++m_counter;
}
std::map<uint32_t, RxDataset *> datasets;
private:
static void onReady(uv_async_t *)
inline void onReady()
{
if (d_ptr->m_listener) {
d_ptr->m_listener->onDatasetReady();
if (m_listener && counter() == m_last.load(std::memory_order_relaxed)) {
m_listener->onDatasetReady();
}
}
Algorithm m_algorithm;
bool m_hugePages = true;
bool m_numa = true;
Buffer m_seed;
IRxListener *m_listener = nullptr;
size_t m_ready = 0;
uv_async_t *m_async;
std::atomic<uint64_t> m_counter;
std::atomic<uint64_t> m_last;
uv_async_t *m_async = nullptr;
};
@ -269,6 +272,7 @@ bool xmrig::Rx::init(const Job &job, int initThreads, bool hugePages, bool numa,
d_ptr->setState(job, hugePages, numa, listener);
const uint32_t threads = initThreads < 1 ? static_cast<uint32_t>(Cpu::info()->threads()) : static_cast<uint32_t>(initThreads);
const String buf = Buffer::toHex(job.seed().data(), 8);
const uint64_t counter = d_ptr->counter();
LOG_INFO("%s" MAGENTA_BOLD("init dataset%s") " algo " WHITE_BOLD("%s (") CYAN_BOLD("%u") WHITE_BOLD(" threads)") BLACK_BOLD(" seed %s..."),
tag,
@ -281,14 +285,14 @@ bool xmrig::Rx::init(const Job &job, int initThreads, bool hugePages, bool numa,
# ifdef XMRIG_FEATURE_HWLOC
if (d_ptr->isNUMA()) {
for (auto const &item : d_ptr->datasets) {
std::thread thread(RxPrivate::initDataset, item.first, threads);
std::thread thread(RxPrivate::initDataset, item.first, threads, counter);
thread.detach();
}
}
else
# endif
{
std::thread thread(RxPrivate::initDataset, 0, threads);
std::thread thread(RxPrivate::initDataset, 0, threads, counter);
thread.detach();
}
@ -311,7 +315,7 @@ xmrig::RxDataset *xmrig::Rx::dataset(const Job &job, uint32_t nodeId)
return nullptr;
}
return d_ptr->datasets.at(d_ptr->isNUMA() ? nodeId : 0);
return d_ptr->datasets.at(d_ptr->isNUMA() ? (d_ptr->datasets.count(nodeId) ? nodeId : HwlocCpuInfo::nodeIndexes().front()) : 0);
}

2
src/crypto/rx/RxVm.h
View file

@ -63,4 +63,4 @@ private:
} /* namespace xmrig */
#endif /* XMRIG_RX_CACHE_H */
#endif /* XMRIG_RX_VM_H */

4
src/version.h
View file

@ -28,14 +28,14 @@
#define APP_ID "xmrig"
#define APP_NAME "XMRig"
#define APP_DESC "XMRig miner"
#define APP_VERSION "4.2.0-beta"
#define APP_VERSION "4.3.0-evo"
#define APP_DOMAIN "xmrig.com"
#define APP_SITE "www.xmrig.com"
#define APP_COPYRIGHT "Copyright (C) 2016-2019 xmrig.com"
#define APP_KIND "miner"
#define APP_VER_MAJOR 4
#define APP_VER_MINOR 2
#define APP_VER_MINOR 3
#define APP_VER_PATCH 0
#ifdef _MSC_VER