fzorb/REDACTED-rig
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Split config generator to separated files.

Browse source
This commit is contained in:
XMRig 2019-09-06 19:38:22 +07:00
parent 9dc2525ce1
commit 62f086f607
10 changed files with 322 additions and 140 deletions
Download patch file Download diff file Expand all files Collapse all files

43
src/backend/opencl/OclGenerator.h Normal file
View file

@ -0,0 +1,43 @@
/* XMRig
* Copyright 2010 Jeff Garzik <jgarzik@pobox.com>
* Copyright 2012-2014 pooler <pooler@litecoinpool.org>
* Copyright 2014 Lucas Jones <https://github.com/lucasjones>
* Copyright 2014-2016 Wolf9466 <https://github.com/OhGodAPet>
* Copyright 2016 Jay D Dee <jayddee246@gmail.com>
* Copyright 2017-2018 XMR-Stak <https://github.com/fireice-uk>, <https://github.com/psychocrypt>
* Copyright 2018-2019 SChernykh <https://github.com/SChernykh>
* Copyright 2016-2019 XMRig <https://github.com/xmrig>, <support@xmrig.com>
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#ifndef XMRIG_OCLGENERATOR_H
#define XMRIG_OCLGENERATOR_H
namespace xmrig {
class Algorithm;
class OclDevice;
class OclThreads;
using ocl_gen_config_fun = bool (*)(const OclDevice &device, const Algorithm &algorithm, OclThreads &threads);
} // namespace xmrig
#endif /* XMRIG_OCLGENERATOR_H */

3
src/backend/opencl/OclThread.h
View file

@ -40,12 +40,13 @@ class OclThread
{
public:
OclThread() = delete;
OclThread(uint32_t index, uint32_t intensity, uint32_t worksize, uint32_t stridedIndex, uint32_t memChunk, uint32_t threads, const Algorithm &algorithm) :
OclThread(uint32_t index, uint32_t intensity, uint32_t worksize, uint32_t stridedIndex, uint32_t memChunk, uint32_t threads, uint32_t unrollFactor, const Algorithm &algorithm) :
m_algorithm(algorithm),
m_threads(threads, -1),
m_index(index),
m_memChunk(memChunk),
m_stridedIndex(stridedIndex),
m_unrollFactor(unrollFactor),
m_worksize(worksize)
{
setIntensity(intensity);

115
src/backend/opencl/generators/ocl_generic_cn_generator.cpp Normal file
View file

@ -0,0 +1,115 @@
/* XMRig
* Copyright 2010 Jeff Garzik <jgarzik@pobox.com>
* Copyright 2012-2014 pooler <pooler@litecoinpool.org>
* Copyright 2014 Lucas Jones <https://github.com/lucasjones>
* Copyright 2014-2016 Wolf9466 <https://github.com/OhGodAPet>
* Copyright 2016 Jay D Dee <jayddee246@gmail.com>
* Copyright 2017-2018 XMR-Stak <https://github.com/fireice-uk>, <https://github.com/psychocrypt>
* Copyright 2018 Lee Clagett <https://github.com/vtnerd>
* Copyright 2018-2019 SChernykh <https://github.com/SChernykh>
* Copyright 2016-2019 XMRig <https://github.com/xmrig>, <support@xmrig.com>
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "backend/opencl/OclThreads.h"
#include "backend/opencl/wrappers/OclDevice.h"
#include "crypto/cn/CnAlgo.h"
#include "crypto/common/Algorithm.h"
namespace xmrig {
constexpr const size_t oneMiB = 1024u * 1024u;
static inline uint32_t getMaxThreads(const OclDevice &device, const Algorithm &algorithm)
{
if (device.vendorId() == OCL_VENDOR_NVIDIA && (device.name().contains("P100") || device.name().contains("V100"))) {
return 40000u;
}
return ((algorithm.l3() <= oneMiB) ? 2u : 1u) * 1000u;
}
static inline uint32_t getPossibleIntensity(const OclDevice &device, const Algorithm &algorithm)
{
const uint32_t maxThreads = getMaxThreads(device, algorithm);
const size_t minFreeMem = (maxThreads == 40000u ? 512u : 128u) * oneMiB;
const size_t availableMem = device.freeMem() - minFreeMem;
const size_t perThread = algorithm.l3() + 224u;
const auto maxIntensity = static_cast<uint32_t>(availableMem / perThread);
return std::min<uint32_t>(maxThreads, maxIntensity);
}
static uint32_t getIntensity(const OclDevice &device, const Algorithm &algorithm)
{
if (device.type() == OclDevice::Raven) {
return 0;
}
const uint32_t maxIntensity = getPossibleIntensity(device, algorithm);
uint32_t intensity = (maxIntensity / (8 * device.computeUnits())) * device.computeUnits() * 8;
if (intensity == 0) {
return 0;
}
if (device.vendorId() == OCL_VENDOR_AMD && (device.type() == OclDevice::Lexa || device.type() == OclDevice::Baffin || device.computeUnits() <= 16)) {
intensity /= 2;
if (algorithm.family() == Algorithm::CN_HEAVY) {
intensity /= 2;
}
}
return intensity;
}
static uint32_t getStridedIndex(const OclDevice &device, const Algorithm &algorithm)
{
if (device.vendorId() != OCL_VENDOR_AMD) {
return 0;
}
return CnAlgo<>::base(algorithm) == Algorithm::CN_2 ? 2 : 1;
}
bool ocl_generic_cn_generator(const OclDevice &device, const Algorithm &algorithm, OclThreads &threads)
{
if (!algorithm.isCN()) {
return false;
}
const uint32_t intensity = getIntensity(device, algorithm);
if (intensity == 0) {
return false;
}
const uint32_t threadCount = ((device.globalMem() - intensity * 2 * algorithm.l3()) > 128 * oneMiB) ? 2 : 1;
threads.add(OclThread(device.index(), intensity, 8, getStridedIndex(device, algorithm), 2, threadCount, 8, algorithm));
return true;
}
} // namespace xmrig

132
src/backend/opencl/generators/ocl_vega_cn_generator.cpp Normal file
View file

@ -0,0 +1,132 @@
/* XMRig
* Copyright 2010 Jeff Garzik <jgarzik@pobox.com>
* Copyright 2012-2014 pooler <pooler@litecoinpool.org>
* Copyright 2014 Lucas Jones <https://github.com/lucasjones>
* Copyright 2014-2016 Wolf9466 <https://github.com/OhGodAPet>
* Copyright 2016 Jay D Dee <jayddee246@gmail.com>
* Copyright 2017-2018 XMR-Stak <https://github.com/fireice-uk>, <https://github.com/psychocrypt>
* Copyright 2018 Lee Clagett <https://github.com/vtnerd>
* Copyright 2018-2019 SChernykh <https://github.com/SChernykh>
* Copyright 2016-2019 XMRig <https://github.com/xmrig>, <support@xmrig.com>
*
* This program is free software: you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation, either version 3 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "backend/opencl/OclThreads.h"
#include "backend/opencl/wrappers/OclDevice.h"
#include "crypto/cn/CnAlgo.h"
#include "crypto/common/Algorithm.h"
namespace xmrig {
constexpr const size_t oneMiB = 1024u * 1024u;
static inline bool isMatch(const OclDevice &device, const Algorithm &algorithm)
{
return algorithm.isCN() &&
device.vendorId() == OCL_VENDOR_AMD &&
(device.type() == OclDevice::Vega_10 || device.type() == OclDevice::Vega_20);
}
static inline uint32_t getMaxThreads(const OclDevice &device, const Algorithm &algorithm)
{
const uint32_t ratio = (algorithm.l3() <= oneMiB) ? 2u : 1u;
if (device.type() == OclDevice::Vega_10) {
if (device.computeUnits() == 56 && algorithm.family() == Algorithm::CN && CnAlgo<>::base(algorithm) == Algorithm::CN_2) {
return 1792u;
}
}
return ratio * 2024u;
}
static inline uint32_t getPossibleIntensity(const OclDevice &device, const Algorithm &algorithm)
{
const uint32_t maxThreads = getMaxThreads(device, algorithm);
const size_t availableMem = device.freeMem() - (128u * oneMiB);
const size_t perThread = algorithm.l3() + 224u;
const auto maxIntensity = static_cast<uint32_t>(availableMem / perThread);
return std::min<uint32_t>(maxThreads, maxIntensity);
}
static inline uint32_t getIntensity(const OclDevice &device, const Algorithm &algorithm)
{
const uint32_t maxIntensity = getPossibleIntensity(device, algorithm);
if (device.type() == OclDevice::Vega_10) {
if (algorithm.family() == Algorithm::CN_HEAVY && device.computeUnits() == 64 && maxIntensity > 976) {
return 976;
}
}
return maxIntensity / device.computeUnits() * device.computeUnits();
}
static inline uint32_t getWorksize(const Algorithm &algorithm)
{
if (algorithm.family() == Algorithm::CN_PICO) {
return 64;
}
if (CnAlgo<>::base(algorithm) == Algorithm::CN_2) {
return 16;
}
return 8;
}
static uint32_t getStridedIndex(const Algorithm &algorithm)
{
return CnAlgo<>::base(algorithm) == Algorithm::CN_2 ? 2 : 1;
}
static inline uint32_t getMemChunk(const Algorithm &algorithm)
{
return CnAlgo<>::base(algorithm) == Algorithm::CN_2 ? 1 : 2;
}
bool ocl_vega_cn_generator(const OclDevice &device, const Algorithm &algorithm, OclThreads &threads)
{
if (!isMatch(device, algorithm)) {
return false;
}
const uint32_t intensity = getIntensity(device, algorithm);
if (intensity == 0) {
return false;
}
const uint32_t worksize = getWorksize(algorithm);
const uint32_t memChunk = getMemChunk(algorithm);
threads.add(OclThread(device.index(), intensity, worksize, getStridedIndex(algorithm), memChunk, 2, 8, algorithm));
return true;
}
} // namespace xmrig

3
src/backend/opencl/opencl.cmake
View file

@ -5,6 +5,8 @@ if (WITH_OPENCL)
set(HEADERS_BACKEND_OPENCL
src/backend/opencl/cl/OclSource.h
src/backend/opencl/generators/ocl_generic_cn_generator.cpp
src/backend/opencl/generators/ocl_vega_cn_generator.cpp
src/backend/opencl/interfaces/IOclRunner.h
src/backend/opencl/kernels/Cn0Kernel.h
src/backend/opencl/kernels/Cn1Kernel.h
@ -13,6 +15,7 @@ if (WITH_OPENCL)
src/backend/opencl/OclBackend.h
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

135
src/backend/opencl/wrappers/OclDevice.cpp
View file

@ -23,11 +23,10 @@
*/
#include <algorithm>
#include "backend/opencl/OclThreads.h"
#include "backend/opencl/wrappers/OclDevice.h"
#include "backend/opencl/OclGenerator.h"
#include "backend/opencl/OclThreads.h"
#include "backend/opencl/wrappers/OclLib.h"
#include "base/io/log/Log.h"
#include "crypto/cn/CnAlgo.h"
@ -35,6 +34,9 @@
#include "rapidjson/document.h"
#include <algorithm>
typedef union
{
struct { cl_uint type; cl_uint data[5]; } raw;
@ -45,7 +47,14 @@ typedef union
namespace xmrig {
constexpr const size_t oneMiB = 1024u * 1024u;
extern bool ocl_vega_cn_generator(const OclDevice &device, const Algorithm &algorithm, OclThreads &threads);
extern bool ocl_generic_cn_generator(const OclDevice &device, const Algorithm &algorithm, OclThreads &threads);
ocl_gen_config_fun generators[] = {
ocl_vega_cn_generator,
ocl_generic_cn_generator
};
static OclVendor getVendorId(const String &vendor)
@ -176,119 +185,9 @@ uint32_t xmrig::OclDevice::clock() const
void xmrig::OclDevice::generate(const Algorithm &algorithm, OclThreads &threads) const
{
uint32_t intensity = getIntensity(algorithm);
if (intensity == 0) {
return;
}
const uint32_t worksize = getWorksize(algorithm);
const uint32_t stridedIndex = getStridedIndex(algorithm);
const uint32_t memChunk = getMemChunk(algorithm);
const uint32_t threadCount = ((globalMem() - intensity * 2 * algorithm.l3()) > 128 * oneMiB) ? 2 : 1;
threads.add(OclThread(index(), intensity, worksize, stridedIndex, memChunk, threadCount, algorithm));
}
uint32_t xmrig::OclDevice::getIntensity(const Algorithm &algorithm) const
{
if(m_type == Raven) {
return 0;
}
const uint32_t maxIntensity = getPossibleIntensity(algorithm);
if (m_type == Vega_10) {
if (algorithm.family() == Algorithm::CN_HEAVY && m_computeUnits && maxIntensity > 976) {
return 976;
}
return maxIntensity / m_computeUnits * m_computeUnits;
}
uint32_t intensity = (maxIntensity / (8 * m_computeUnits)) * m_computeUnits * 8;
if (intensity == 0) {
return 0;
}
if (m_vendorId == OCL_VENDOR_AMD && (m_type == Lexa || m_type == Baffin || m_computeUnits <= 16)) {
intensity /= 2;
if (algorithm.family() == Algorithm::CN_HEAVY) {
intensity /= 2;
for (auto fn : generators) {
if (fn(*this, algorithm, threads)) {
return;
}
}
return intensity;
}
uint32_t xmrig::OclDevice::getMaxThreads(const Algorithm &algorithm) const
{
if (m_vendorId == OCL_VENDOR_NVIDIA && (m_name.contains("P100") || m_name.contains("V100"))) {
return 40000u;
}
const uint32_t ratio = (algorithm.l3() <= oneMiB) ? 2u : 1u;
if (m_type == Vega_10 || m_type == Vega_20) {
if (computeUnits() == 56 && isCNv2(algorithm)) {
return 1792u;
}
return ratio * 2024u;
}
return ratio * 1000u;
}
uint32_t xmrig::OclDevice::getMemChunk(const Algorithm &algorithm) const
{
if ((m_type == Vega_10 || m_type == Vega_20) && (algorithm.family() == Algorithm::CN_PICO || isCNv2(algorithm))) {
return 1;
}
return 2;
}
uint32_t xmrig::OclDevice::getPossibleIntensity(const Algorithm &algorithm) const
{
const uint32_t maxThreads = getMaxThreads(algorithm);
const size_t minFreeMem = (maxThreads == 40000u ? 512u : 128u) * oneMiB;
const size_t availableMem = freeMem() - minFreeMem;
const size_t perThread = algorithm.l3() + 224u;
const auto maxIntensity = static_cast<uint32_t>(availableMem / perThread);
return std::min<uint32_t>(maxThreads, maxIntensity);
}
uint32_t xmrig::OclDevice::getStridedIndex(const Algorithm &algorithm) const
{
if (m_vendorId == OCL_VENDOR_NVIDIA) {
return 0;
}
if (algorithm.family() == Algorithm::CN_PICO || isCNv2(algorithm)) {
return 2;
}
return 1;
}
uint32_t xmrig::OclDevice::getWorksize(const Algorithm &algorithm) const
{
if (m_type == Vega_10 || m_type == Vega_20) {
if (algorithm.family() == Algorithm::CN_PICO) {
return 64;
}
if (isCNv2(algorithm)) {
return 16;
}
}
return 8;
}

8
src/backend/opencl/wrappers/OclDevice.h
View file

@ -76,17 +76,11 @@ public:
inline const String &name() const { return m_name; }
inline const String &vendor() const { return m_vendor; }
inline OclVendor vendorId() const { return m_vendorId; }
inline Type type() const { return m_type; }
inline uint32_t computeUnits() const { return m_computeUnits; }
inline uint32_t index() const { return m_index; }
private:
uint32_t getIntensity(const Algorithm &algorithm) const;
uint32_t getMaxThreads(const Algorithm &algorithm) const;
uint32_t getMemChunk(const Algorithm &algorithm) const;
uint32_t getPossibleIntensity(const Algorithm &algorithm) const;
uint32_t getStridedIndex(const Algorithm &algorithm) const;
uint32_t getWorksize(const Algorithm &algorithm) const;
bool m_topology = false;
cl_device_id m_id = nullptr;
cl_platform_id m_platform = nullptr;