lukechilds
/
ethminer
mirror of https://github.com/lukechilds/ethminer.git
1
0
Fork 0
Code Issues Projects Releases Wiki Activity
Browse Source

cleanup of chunks and msperbatch host code

cl-refactor
Genoil 9 years ago
parent
22e3b7f6fc
commit
322534d26e
5 changed files with 8 additions and 88 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. 13
      ethminer/MinerAux.h
  2. 63
      libethash-cl/ethash_cl_miner.cpp
  3. 15
      libethash-cl/ethash_cl_miner.h
  4. 2
      libethcore/EthashGPUMiner.cpp
  5. 1
      libethcore/EthashGPUMiner.h

13
ethminer/MinerAux.h
View File

@ -211,16 +211,6 @@ public:
m_extraGPUMemory = 1000000 * stol(argv[++i]); m_extraGPUMemory = 1000000 * stol(argv[++i]);
#endif #endif
#if ETH_ETHASHCL || !ETH_TRUE #if ETH_ETHASHCL || !ETH_TRUE
else if (arg == "--cl-ms-per-batch" && i + 1 < argc)
try
{
m_msPerBatch = stol(argv[++i]);
}
catch (...)
{
cerr << "Bad " << arg << " option: " << argv[i] << endl;
BOOST_THROW_EXCEPTION(BadArgument());
}
else if (arg == "--allow-opencl-cpu") else if (arg == "--allow-opencl-cpu")
m_clAllowCPU = true; m_clAllowCPU = true;
#endif #endif
@ -447,7 +437,6 @@ public:
if (!EthashGPUMiner::configureGPU( if (!EthashGPUMiner::configureGPU(
m_localWorkSize, m_localWorkSize,
m_globalWorkSizeMultiplier, m_globalWorkSizeMultiplier,
m_msPerBatch,
m_openclPlatform, m_openclPlatform,
m_openclDevice, m_openclDevice,
m_clAllowCPU, m_clAllowCPU,
@ -540,7 +529,6 @@ public:
<< " --cl-extragpu-mem Set the memory (in MB) you believe your GPU requires for stuff other than mining. Windows rendering e.t.c.." << endl << " --cl-extragpu-mem Set the memory (in MB) you believe your GPU requires for stuff other than mining. Windows rendering e.t.c.." << endl
<< " --cl-local-work Set the OpenCL local work size. Default is " << toString(ethash_cl_miner::c_defaultLocalWorkSize) << endl << " --cl-local-work Set the OpenCL local work size. Default is " << toString(ethash_cl_miner::c_defaultLocalWorkSize) << endl
<< " --cl-global-work Set the OpenCL global work size as a multiple of the local work size. Default is " << toString(ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier) << " * " << toString(ethash_cl_miner::c_defaultLocalWorkSize) << endl << " --cl-global-work Set the OpenCL global work size as a multiple of the local work size. Default is " << toString(ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier) << " * " << toString(ethash_cl_miner::c_defaultLocalWorkSize) << endl
<< " --cl-ms-per-batch Set the OpenCL target milliseconds per batch (global workgroup size). Default is " << toString(ethash_cl_miner::c_defaultMSPerBatch) << ". If 0 is given then no autoadjustment of global work size will happen" << endl
#endif #endif
#if ETH_ETHASHCUDA || !ETH_TRUE #if ETH_ETHASHCUDA || !ETH_TRUE
<< " --cuda-extragpu-mem Set the memory (in MB) you believe your GPU requires for stuff other than mining. Windows rendering e.t.c.." << endl << " --cuda-extragpu-mem Set the memory (in MB) you believe your GPU requires for stuff other than mining. Windows rendering e.t.c.." << endl
@ -914,7 +902,6 @@ private:
unsigned m_globalWorkSizeMultiplier = ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier; unsigned m_globalWorkSizeMultiplier = ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier;
unsigned m_localWorkSize = ethash_cl_miner::c_defaultLocalWorkSize; unsigned m_localWorkSize = ethash_cl_miner::c_defaultLocalWorkSize;
#endif #endif
unsigned m_msPerBatch = ethash_cl_miner::c_defaultMSPerBatch;
#endif #endif
#if ETH_ETHASHCUDA || !ETH_TRUE #if ETH_ETHASHCUDA || !ETH_TRUE
unsigned m_globalWorkSizeMultiplier = ethash_cuda_miner::c_defaultGridSize; unsigned m_globalWorkSizeMultiplier = ethash_cuda_miner::c_defaultGridSize;

63
libethash-cl/ethash_cl_miner.cpp
View File

@ -55,7 +55,6 @@ using namespace std;
unsigned const ethash_cl_miner::c_defaultLocalWorkSize = 64; unsigned const ethash_cl_miner::c_defaultLocalWorkSize = 64;
unsigned const ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier = 4096; // * CL_DEFAULT_LOCAL_WORK_SIZE unsigned const ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier = 4096; // * CL_DEFAULT_LOCAL_WORK_SIZE
unsigned const ethash_cl_miner::c_defaultMSPerBatch = 0;
// TODO: If at any point we can use libdevcore in here then we should switch to using a LogChannel // TODO: If at any point we can use libdevcore in here then we should switch to using a LogChannel
#if defined(_WIN32) #if defined(_WIN32)
@ -184,7 +183,6 @@ bool ethash_cl_miner::configureGPU(
unsigned _platformId, unsigned _platformId,
unsigned _localWorkSize, unsigned _localWorkSize,
unsigned _globalWorkSize, unsigned _globalWorkSize,
unsigned _msPerBatch,
bool _allowCPU, bool _allowCPU,
unsigned _extraGPUMemory, unsigned _extraGPUMemory,
uint64_t _currentBlock uint64_t _currentBlock
@ -192,7 +190,6 @@ bool ethash_cl_miner::configureGPU(
{ {
s_workgroupSize = _localWorkSize; s_workgroupSize = _localWorkSize;
s_initialGlobalWorkSize = _globalWorkSize; s_initialGlobalWorkSize = _globalWorkSize;
s_msPerBatch = _msPerBatch;
s_allowCPU = _allowCPU; s_allowCPU = _allowCPU;
s_extraRequiredGPUMem = _extraGPUMemory; s_extraRequiredGPUMem = _extraGPUMemory;
// by default let's only consider the DAG of the first epoch // by default let's only consider the DAG of the first epoch
@ -223,7 +220,6 @@ bool ethash_cl_miner::configureGPU(
bool ethash_cl_miner::s_allowCPU = false; bool ethash_cl_miner::s_allowCPU = false;
unsigned ethash_cl_miner::s_extraRequiredGPUMem; unsigned ethash_cl_miner::s_extraRequiredGPUMem;
unsigned ethash_cl_miner::s_msPerBatch = ethash_cl_miner::c_defaultMSPerBatch;
unsigned ethash_cl_miner::s_workgroupSize = ethash_cl_miner::c_defaultLocalWorkSize; unsigned ethash_cl_miner::s_workgroupSize = ethash_cl_miner::c_defaultLocalWorkSize;
unsigned ethash_cl_miner::s_initialGlobalWorkSize = ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier * ethash_cl_miner::c_defaultLocalWorkSize; unsigned ethash_cl_miner::s_initialGlobalWorkSize = ethash_cl_miner::c_defaultGlobalWorkSizeMultiplier * ethash_cl_miner::c_defaultLocalWorkSize;
@ -362,10 +358,6 @@ bool ethash_cl_miner::init(
m_globalWorkSize = s_initialGlobalWorkSize; m_globalWorkSize = s_initialGlobalWorkSize;
if (m_globalWorkSize % s_workgroupSize != 0) if (m_globalWorkSize % s_workgroupSize != 0)
m_globalWorkSize = ((m_globalWorkSize / s_workgroupSize) + 1) * s_workgroupSize; m_globalWorkSize = ((m_globalWorkSize / s_workgroupSize) + 1) * s_workgroupSize;
// remember the device's address bits
m_deviceBits = device.getInfo<CL_DEVICE_ADDRESS_BITS>();
// make sure first step of global work size adjustment is large enough
m_stepWorkSizeAdjust = pow(2, m_deviceBits / 2 + 1);
// patch source code // patch source code
// note: ETHASH_CL_MINER_KERNEL is simply ethash_cl_miner_kernel.cl compiled // note: ETHASH_CL_MINER_KERNEL is simply ethash_cl_miner_kernel.cl compiled
@ -397,13 +389,12 @@ bool ethash_cl_miner::init(
// create buffer for dag // create buffer for dag
try try
{ {
m_dagChunksCount = 1;
ETHCL_LOG("Creating one big buffer for the DAG"); ETHCL_LOG("Creating one big buffer for the DAG");
m_dagChunks.push_back(cl::Buffer(m_context, CL_MEM_READ_ONLY, _dagSize)); m_dag = cl::Buffer(m_context, CL_MEM_READ_ONLY, _dagSize);
ETHCL_LOG("Loading single big chunk kernels"); ETHCL_LOG("Loading single big chunk kernels");
m_searchKernel = cl::Kernel(program, "ethash_search"); m_searchKernel = cl::Kernel(program, "ethash_search");
ETHCL_LOG("Mapping one big chunk."); ETHCL_LOG("Mapping one big chunk.");
m_queue.enqueueWriteBuffer(m_dagChunks[0], CL_TRUE, 0, _dagSize, _dag); m_queue.enqueueWriteBuffer(m_dag, CL_TRUE, 0, _dagSize, _dag);
} }
catch (cl::Error const& err) catch (cl::Error const& err)
{ {
@ -456,13 +447,12 @@ void ethash_cl_miner::search(uint8_t const* header, uint64_t target, search_hook
#endif #endif
m_queue.finish(); m_queue.finish();
unsigned argPos = 2;
m_searchKernel.setArg(1, m_header); m_searchKernel.setArg(1, m_header);
for (unsigned i = 0; i < m_dagChunksCount; ++i, ++argPos) m_searchKernel.setArg(2, m_dag );
m_searchKernel.setArg(argPos, m_dagChunks[i]);
// pass these to stop the compiler unrolling the loops // pass these to stop the compiler unrolling the loops
m_searchKernel.setArg(argPos + 1, target); m_searchKernel.setArg(4, target);
m_searchKernel.setArg(argPos + 2, ~0u); m_searchKernel.setArg(5, ~0u);
unsigned buf = 0; unsigned buf = 0;
random_device engine; random_device engine;
@ -472,10 +462,7 @@ void ethash_cl_miner::search(uint8_t const* header, uint64_t target, search_hook
auto t = chrono::high_resolution_clock::now(); auto t = chrono::high_resolution_clock::now();
// supply output buffer to kernel // supply output buffer to kernel
m_searchKernel.setArg(0, m_searchBuffer[buf]); m_searchKernel.setArg(0, m_searchBuffer[buf]);
if (m_dagChunksCount == 1)
m_searchKernel.setArg(3, start_nonce); m_searchKernel.setArg(3, start_nonce);
else
m_searchKernel.setArg(6, start_nonce);
// execute it! // execute it!
m_queue.enqueueNDRangeKernel(m_searchKernel, cl::NullRange, m_globalWorkSize, s_workgroupSize); m_queue.enqueueNDRangeKernel(m_searchKernel, cl::NullRange, m_globalWorkSize, s_workgroupSize);
@ -508,44 +495,6 @@ void ethash_cl_miner::search(uint8_t const* header, uint64_t target, search_hook
pending.pop(); pending.pop();
} }
// adjust global work size depending on last search time
if (s_msPerBatch)
{
// Global work size must be:
// - less than or equal to 2 ^ DEVICE_BITS - 1
// - divisible by lobal work size (workgroup size)
auto d = chrono::duration_cast<chrono::milliseconds>(chrono::high_resolution_clock::now() - t);
if (d != chrono::milliseconds(0)) // if duration is zero, we did not get in the actual searh/or search not finished
{
if (d > chrono::milliseconds(s_msPerBatch * 10 / 9))
{
// Divide the step by 2 when adjustment way change
if (m_wayWorkSizeAdjust > -1)
m_stepWorkSizeAdjust = max<unsigned>(1, m_stepWorkSizeAdjust / 2);
m_wayWorkSizeAdjust = -1;
// cerr << "m_stepWorkSizeAdjust: " << m_stepWorkSizeAdjust << ", m_wayWorkSizeAdjust: " << m_wayWorkSizeAdjust << endl;
// cerr << "Batch of " << m_globalWorkSize << " took " << chrono::duration_cast<chrono::milliseconds>(d).count() << " ms, >> " << s_msPerBatch << " ms." << endl;
m_globalWorkSize = max<unsigned>(128, m_globalWorkSize - m_stepWorkSizeAdjust);
// cerr << "New global work size" << m_globalWorkSize << endl;
}
else if (d < chrono::milliseconds(s_msPerBatch * 9 / 10))
{
// Divide the step by 2 when adjustment way change
if (m_wayWorkSizeAdjust < 1)
m_stepWorkSizeAdjust = max<unsigned>(1, m_stepWorkSizeAdjust / 2);
m_wayWorkSizeAdjust = 1;
// cerr << "m_stepWorkSizeAdjust: " << m_stepWorkSizeAdjust << ", m_wayWorkSizeAdjust: " << m_wayWorkSizeAdjust << endl;
// cerr << "Batch of " << m_globalWorkSize << " took " << chrono::duration_cast<chrono::milliseconds>(d).count() << " ms, << " << s_msPerBatch << " ms." << endl;
m_globalWorkSize = min<unsigned>(pow(2, m_deviceBits) - 1, m_globalWorkSize + m_stepWorkSizeAdjust);
// Global work size should never be less than the workgroup size
m_globalWorkSize = max<unsigned>(s_workgroupSize, m_globalWorkSize);
// cerr << "New global work size" << m_globalWorkSize << endl;
}
}
}
} }
// not safe to return until this is ready // not safe to return until this is ready

15
libethash-cl/ethash_cl_miner.h
View File

@ -46,7 +46,6 @@ public:
unsigned _platformId, unsigned _platformId,
unsigned _localWorkSize, unsigned _localWorkSize,
unsigned _globalWorkSize, unsigned _globalWorkSize,
unsigned _msPerBatch,
bool _allowCPU, bool _allowCPU,
unsigned _extraGPUMemory, unsigned _extraGPUMemory,
uint64_t _currentBlock uint64_t _currentBlock
@ -61,16 +60,11 @@ public:
void finish(); void finish();
void search(uint8_t const* _header, uint64_t _target, search_hook& _hook); void search(uint8_t const* _header, uint64_t _target, search_hook& _hook);
void hash_chunk(uint8_t* _ret, uint8_t const* _header, uint64_t _nonce, unsigned _count);
void search_chunk(uint8_t const*_header, uint64_t _target, search_hook& _hook);
/* -- default values -- */ /* -- default values -- */
/// Default value of the local work size. Also known as workgroup size. /// Default value of the local work size. Also known as workgroup size.
static unsigned const c_defaultLocalWorkSize; static unsigned const c_defaultLocalWorkSize;
/// Default value of the global work size as a multiplier of the local work size /// Default value of the global work size as a multiplier of the local work size
static unsigned const c_defaultGlobalWorkSizeMultiplier; static unsigned const c_defaultGlobalWorkSizeMultiplier;
/// Default value of the milliseconds per global work size (per batch)
static unsigned const c_defaultMSPerBatch;
private: private:
@ -80,18 +74,11 @@ private:
cl::Context m_context; cl::Context m_context;
cl::CommandQueue m_queue; cl::CommandQueue m_queue;
cl::Kernel m_searchKernel; cl::Kernel m_searchKernel;
unsigned int m_dagChunksCount; cl::Buffer m_dag;
std::vector<cl::Buffer> m_dagChunks;
cl::Buffer m_header; cl::Buffer m_header;
cl::Buffer m_searchBuffer[c_bufferCount]; cl::Buffer m_searchBuffer[c_bufferCount];
unsigned m_globalWorkSize; unsigned m_globalWorkSize;
bool m_openclOnePointOne; bool m_openclOnePointOne;
unsigned m_deviceBits;
/// The step used in the work size adjustment
unsigned int m_stepWorkSizeAdjust;
/// The Work Size way of adjustment, > 0 when previously increased, < 0 when previously decreased
int m_wayWorkSizeAdjust = 0;
/// The local work size for the search /// The local work size for the search
static unsigned s_workgroupSize; static unsigned s_workgroupSize;

2
libethcore/EthashGPUMiner.cpp
View File

@ -202,7 +202,6 @@ void EthashGPUMiner::listDevices()
bool EthashGPUMiner::configureGPU( bool EthashGPUMiner::configureGPU(
unsigned _localWorkSize, unsigned _localWorkSize,
unsigned _globalWorkSizeMultiplier, unsigned _globalWorkSizeMultiplier,
unsigned _msPerBatch,
unsigned _platformId, unsigned _platformId,
unsigned _deviceId, unsigned _deviceId,
bool _allowCPU, bool _allowCPU,
@ -223,7 +222,6 @@ bool EthashGPUMiner::configureGPU(
_platformId, _platformId,
_localWorkSize, _localWorkSize,
_globalWorkSizeMultiplier * _localWorkSize, _globalWorkSizeMultiplier * _localWorkSize,
_msPerBatch,
_allowCPU, _allowCPU,
_extraGPUMemory, _extraGPUMemory,
_currentBlock) _currentBlock)

1
libethcore/EthashGPUMiner.h
View File

@ -48,7 +48,6 @@ public:
static bool configureGPU( static bool configureGPU(
unsigned _localWorkSize, unsigned _localWorkSize,
unsigned _globalWorkSizeMultiplier, unsigned _globalWorkSizeMultiplier,
unsigned _msPerBatch,
unsigned _platformId, unsigned _platformId,
unsigned _deviceId, unsigned _deviceId,
bool _allowCPU, bool _allowCPU,

Write Preview
Loading…
Cancel
Save