gaspersic/triton
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Python: Refactored wrapper

Browse Source
This commit is contained in:
Philippe Tillet
2015-04-29 17:48:57 -04:00
parent a67476671d
commit 5ef01f041a
8 changed files with 678 additions and 670 deletions
Download Patch File Download Diff File Expand all files Collapse all files

675
python/pyisaac/src/_isaac.cpp
View File

@@ -1,687 +1,22 @@
#include <list>
#include <functional>
#include <boost/python.hpp>
#include <boost/python/suite/indexing/vector_indexing_suite.hpp>
#include <boost/python/suite/indexing/map_indexing_suite.hpp>
#include <boost/numpy.hpp>
#include <boost/numpy/dtype.hpp>
#include "driver.h"
#include "core.h"
#include "model.h"
#include "isaac/array.h"
#include "isaac/backend/templates/vaxpy.h"
#include "isaac/backend/templates/maxpy.h"
#include "isaac/backend/templates/reduction.h"
#include "isaac/backend/templates/mreduction.h"
#include "isaac/backend/templates/mproduct.h"
#include "isaac/model/model.h"
#define MAP_ENUM(v, ns) .value(#v, ns::v)
namespace bp = boost::python;
namespace atd = isaac;
namespace np = boost::numpy;
namespace detail
{
atd::numeric_type to_atd_dtype(np::dtype const & T)
{
if(T==np::detail::get_int_dtype<8, false>()) return atd::CHAR_TYPE;
else if(T==np::detail::get_int_dtype<8, true>()) return atd::UCHAR_TYPE;
else if(T==np::detail::get_int_dtype<16, false>()) return atd::SHORT_TYPE;
else if(T==np::detail::get_int_dtype<16, true>()) return atd::USHORT_TYPE;
else if(T==np::detail::get_int_dtype<32, false>()) return atd::INT_TYPE;
else if(T==np::detail::get_int_dtype<32, true>()) return atd::UINT_TYPE;
else if(T==np::detail::get_int_dtype<64, false>()) return atd::LONG_TYPE;
else if(T==np::detail::get_int_dtype<64, true>()) return atd::ULONG_TYPE;
// else if(T==np::detail::get_float_dtype<16>()) return atd::HALF_TYPE;
else if(T==np::detail::get_float_dtype<32>()) return atd::FLOAT_TYPE;
else if(T==np::detail::get_float_dtype<64>()) return atd::DOUBLE_TYPE;
else{
PyErr_SetString(PyExc_TypeError, "Unrecognized datatype");
bp::throw_error_already_set();
throw; // suppress warning; throw_error_already_set() never returns but isn't marked noreturn: https://svn.boost.org/trac/boost/ticket/1482
}
}
np::dtype to_np_dtype(atd::numeric_type const & T) throw()
{
if(T==atd::CHAR_TYPE) return np::detail::get_int_dtype<8, false>();
else if(T==atd::UCHAR_TYPE) return np::detail::get_int_dtype<8, true>();
else if(T==atd::SHORT_TYPE) return np::detail::get_int_dtype<16, false>();
else if(T==atd::USHORT_TYPE) return np::detail::get_int_dtype<16, true>();
else if(T==atd::INT_TYPE) return np::detail::get_int_dtype<32, false>();
else if(T==atd::UINT_TYPE) return np::detail::get_int_dtype<32, true>();
else if(T==atd::LONG_TYPE) return np::detail::get_int_dtype<64, false>();
else if(T==atd::ULONG_TYPE) return np::detail::get_int_dtype<64, true>();
// else if(T==atd::HALF_TYPE) return np::detail::get_float_dtype<16>();
else if(T==atd::FLOAT_TYPE) return np::detail::get_float_dtype<32>();
else if(T==atd::DOUBLE_TYPE) return np::detail::get_float_dtype<64>();
else{
PyErr_SetString(PyExc_TypeError, "Unrecognized datatype");
bp::throw_error_already_set();
throw; // suppress warning; throw_error_already_set() never returns but isn't marked noreturn: https://svn.boost.org/trac/boost/ticket/1482
}
}
bp::tuple get_shape(atd::array const & x)
{
return bp::make_tuple(x.shape()[0], x.shape()[1]);
}
//void set_shape(atd::array & x, bp::tuple const & t)
//{
// unsigned int len = bp::len(t);
// atd::int_t size1 = bp::extract<atd::int_t>(t[0]);
// atd::int_t size2 = len<2?1:bp::extract<atd::int_t>(t[1]);
// x.reshape(size1, size2);
//}
//boost::python::dict create_queues(atd::cl_ext::queues_t queues)
//{
// boost::python::dict dictionary;
// for (atd::cl_ext::queues_t::iterator it = queues.begin(); it != queues.end(); ++it) {
// bp::list list;
// for (atd::cl_ext::queues_t::mapped_type::iterator itt = it->second.begin(); itt != it->second.end(); ++itt)
// list.append(*itt);
// dictionary[it->first] = list;
// }
// return dictionary;
//}
template<class T>
struct datatype : public atd::value_scalar
{
datatype(T t) : atd::value_scalar(t){ }
};
template<class T>
unsigned int size(datatype<T> const & dt)
{ return atd::size_of(dt.dtype()) ; }
#define INSTANTIATE(name, clname) \
struct name : public detail::datatype<clname> { name(clname value) : detail::datatype<clname>(value){} };
INSTANTIATE(int8, cl_char)
INSTANTIATE(uint8, cl_uchar)
INSTANTIATE(int16, cl_short)
INSTANTIATE(uint16, cl_ushort)
INSTANTIATE(int32, cl_int)
INSTANTIATE(uint32, cl_uint)
INSTANTIATE(int64, cl_long)
INSTANTIATE(uint64, cl_ulong)
INSTANTIATE(float32, cl_float)
INSTANTIATE(float64, cl_double)
#undef INSTANTIATE
}
void export_core()
{
#define INSTANTIATE(name, clname) \
bp::class_<detail::datatype<clname>, bp::bases<atd::value_scalar> >(#name, bp::init<clname>());\
bp::class_<detail::name, bp::bases<detail::datatype<clname> > >(#name, bp::init<clname>())\
.add_property("size", &detail::size<clname>)\
;
INSTANTIATE(int8, cl_char)
INSTANTIATE(uint8, cl_uchar)
INSTANTIATE(int16, cl_short)
INSTANTIATE(uint16, cl_ushort)
INSTANTIATE(int32, cl_int)
INSTANTIATE(uint32, cl_uint)
INSTANTIATE(int64, cl_long)
INSTANTIATE(uint64, cl_ulong)
INSTANTIATE(float32, cl_float)
INSTANTIATE(float64, cl_double)
#undef INSTANTIATE
bp::enum_<atd::expression_type>("operations")
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(MATRIX_AXPY_TYPE, atd)
MAP_ENUM(REDUCTION_TYPE, atd)
MAP_ENUM(ROW_WISE_REDUCTION_TYPE, atd)
MAP_ENUM(COL_WISE_REDUCTION_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
;
}
namespace detail
{
template<class IT>
bp::list to_list(IT const & begin, IT const & end)
{
bp::list res;
for (IT it = begin; it != end; ++it)
res.append(*it);
return res;
}
template<class T>
std::vector<T> to_vector(bp::list const & list)
{
std::size_t len = bp::len(list);
std::vector<T> res; res.reserve(len);
for(std::size_t i = 0 ; i < len ; ++i)
res.push_back(boost::python::extract<T>(list[i]));
return res;
}
bp::list nv_compute_capability(atd::driver::Device const & device)
{
bp::list res;
std::pair<unsigned int, unsigned int> cc = device.nv_compute_capability();
res.append(cc.first);
res.append(cc.second);
return res;
}
bp::list get_platforms()
{
std::vector<atd::driver::Platform> platforms(atd::driver::Platform::get());
return to_list(platforms.begin(), platforms.end());
}
bp::list get_devices(atd::driver::Platform const & platform)
{
std::vector<atd::driver::Device> devices(platform.devices());
return to_list(devices.begin(), devices.end());
}
atd::numeric_type extract_dtype(bp::object const & odtype)
{
std::string name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="class")
name = bp::extract<std::string>(odtype.attr("__name__"))();
else
name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="int8") return atd::CHAR_TYPE;
else if(name=="uint8") return atd::UCHAR_TYPE;
else if(name=="int16") return atd::SHORT_TYPE;
else if(name=="uint16") return atd::USHORT_TYPE;
else if(name=="int32") return atd::INT_TYPE;
else if(name=="uint32") return atd::UINT_TYPE;
else if(name=="int64") return atd::LONG_TYPE;
else if(name=="uint64") return atd::ULONG_TYPE;
else if(name=="float32") return atd::FLOAT_TYPE;
else if(name=="float64") return atd::DOUBLE_TYPE;
else
{
PyErr_SetString(PyExc_TypeError, "Data type not understood");
bp::throw_error_already_set();
throw;
}
}
atd::expression_type extract_template_type(bp::object const & odtype)
{
std::string name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="class")
name = bp::extract<std::string>(odtype.attr("__name__"))();
else
name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="vaxpy") return atd::VECTOR_AXPY_TYPE;
else if(name=="maxpy") return atd::MATRIX_AXPY_TYPE;
else if(name=="reduction") return atd::REDUCTION_TYPE;
else if(name=="mreduction_rows") return atd::ROW_WISE_REDUCTION_TYPE;
else if(name=="mreduction_cols") return atd::COL_WISE_REDUCTION_TYPE;
else if(name=="mproduct_nn") return atd::MATRIX_PRODUCT_NN_TYPE;
else if(name=="mproduct_tn") return atd::MATRIX_PRODUCT_TN_TYPE;
else if(name=="mproduct_nt") return atd::MATRIX_PRODUCT_NT_TYPE;
else if(name=="mproduct_tt") return atd::MATRIX_PRODUCT_TT_TYPE;
else
{
PyErr_SetString(PyExc_TypeError, "Template type not understood");
bp::throw_error_already_set();
throw;
}
}
struct model_map_indexing
{
static atd::model& get_item(atd::model_map_t& container, bp::tuple i_)
{
atd::expression_type expression = extract_template_type(i_[0]);
atd::numeric_type dtype = extract_dtype(i_[1]);
atd::model_map_t::iterator i = container.find(std::make_pair(expression, dtype));
if (i == container.end())
{
PyErr_SetString(PyExc_KeyError, "Invalid key");
bp::throw_error_already_set();
}
return *i->second;
}
static void set_item(atd::model_map_t& container, bp::tuple i_, atd::model const & v)
{
atd::expression_type expression = extract_template_type(i_[0]);
atd::numeric_type dtype = extract_dtype(i_[1]);
container[std::make_pair(expression, dtype)].reset(new atd::model(v));
}
};
std::string to_string(atd::driver::device_type type)
{
if(type==atd::driver::DEVICE_TYPE_CPU) return "CPU";
if(type==atd::driver::DEVICE_TYPE_GPU) return "GPU";
if(type==atd::driver::DEVICE_TYPE_ACCELERATOR) return "ACCELERATOR";
throw;
}
std::shared_ptr<atd::driver::Context> make_context(atd::driver::Device const & dev)
{ return std::shared_ptr<atd::driver::Context>(new atd::driver::Context(dev)); }
bp::tuple flush(atd::array_expression const & expression, unsigned int queue_id, bp::list dependencies, bool tune, int label, std::string const & program_name, bool force_recompile)
{
std::list<atd::driver::Event> events;
std::vector<atd::driver::Event> cdependencies = to_vector<atd::driver::Event>(dependencies);
std::shared_ptr<atd::array> parray(new atd::array(atd::control(expression,
atd::execution_options_type(queue_id, &events, &cdependencies),
atd::dispatcher_options_type(tune, label),
atd::compilation_options_type(program_name, force_recompile))));
return bp::make_tuple(parray, to_list(events.begin(), events.end()));
}
}
struct state_type{ };
state_type state;
void export_cl()
{
typedef std::vector<atd::driver::CommandQueue> queues_t;
bp::class_<queues_t>("queues")
.def("__len__", &queues_t::size)
.def("__getitem__", &bp::vector_indexing_suite<queues_t>::get_item, bp::return_internal_reference<>())
.def("__setitem__", &bp::vector_indexing_suite<queues_t>::set_item, bp::with_custodian_and_ward<1,2>())
.def("append", &bp::vector_indexing_suite<queues_t>::append)
;
bp::class_<atd::model_map_t>("models")
.def("__getitem__", &detail::model_map_indexing::get_item, bp::return_internal_reference<>())
.def("__setitem__", &detail::model_map_indexing::set_item, bp::with_custodian_and_ward<1,2>())
;
bp::enum_<atd::driver::backend_type>
("backend_type")
.value("OPENCL", atd::driver::OPENCL)
#ifdef ISAAC_WITH_CUDA
.value("CUDA", atd::driver::CUDA)
#endif
;
bp::enum_<atd::driver::device_type>
("device_type")
.value("DEVICE_TYPE_GPU", atd::driver::DEVICE_TYPE_GPU)
.value("DEVICE_TYPE_CPU", atd::driver::DEVICE_TYPE_CPU)
;
bp::class_<atd::driver::Platform>("platform", bp::no_init)
.def("get_devices", &detail::get_devices)
.add_property("name",&atd::driver::Platform::name)
;
bp::enum_<isaac::driver::Device::VENDOR>
("vendor")
.value("AMD", atd::driver::Device::AMD)
.value("INTEL", atd::driver::Device::INTEL)
.value("NVIDIA", atd::driver::Device::NVIDIA)
.value("UNKNOWN", atd::driver::Device::UNKNOWN)
;
bp::class_<atd::driver::Device>("device", bp::no_init)
.add_property("clock_rate", &atd::driver::Device::clock_rate)
.add_property("name", &atd::driver::Device::name)
.add_property("type", &atd::driver::Device::type)
.add_property("platform", &atd::driver::Device::platform)
.add_property("vendor", &atd::driver::Device::vendor)
.add_property("nv_compute_capability", &detail::nv_compute_capability)
;
bp::class_<atd::driver::Context>("context", bp::no_init)
.def("__init__", bp::make_constructor(&detail::make_context))
.add_property("queues", bp::make_function(static_cast<std::vector<atd::driver::CommandQueue> & (*)(const atd::driver::Context&)>( [](const atd::driver::Context & ctx) -> std::vector<atd::driver::CommandQueue> & { return atd::driver::queues[ctx]; }) , bp::return_internal_reference<>()))
.add_property("backend", &atd::driver::Context::backend)
;
bp::class_<atd::driver::CommandQueue>("command_queue", bp::init<atd::driver::Context, atd::driver::Device>())
.def("synchronize", &atd::driver::CommandQueue::synchronize)
.add_property("models", bp::make_function(&atd::get_model_map, bp::return_internal_reference<>()))
.add_property("device", bp::make_function(&atd::driver::CommandQueue::device, bp::return_internal_reference<>()))
;
bp::class_<atd::driver::Event>("event", bp::init<atd::driver::backend_type>())
.add_property("elapsed_time", &atd::driver::Event::elapsed_time)
;
bp::def("device_type_to_string", &detail::to_string);
bp::def("get_platforms", &detail::get_platforms);
bp::def("flush", &detail::flush, (bp::arg("expression"), bp::arg("queue_id") = 0, bp::arg("dependencies")=bp::list(), bp::arg("tune") = false, bp::arg("label")=-1, bp::arg("program_name")="", bp::arg("recompile") = false));
bp::class_<state_type>("state_type")
.def_readwrite("queue_properties",&atd::driver::queues.queue_properties)
;
bp::scope().attr("state") = bp::object(bp::ptr(&state));
bp::scope().attr("CL_QUEUE_PROFILING_ENABLE") = CL_QUEUE_PROFILING_ENABLE;
bp::scope().attr("CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE") = CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE;
}
namespace detail
{
std::shared_ptr<atd::array>
ndarray_to_atdarray(const np::ndarray& array, const atd::driver::Context& ctx)
{
int d = array.get_nd();
if (d > 2) {
PyErr_SetString(PyExc_TypeError, "Only 1-D and 2-D arrays are supported!");
bp::throw_error_already_set();
}
atd::numeric_type dtype = to_atd_dtype(array.get_dtype());
atd::int_t size = (atd::int_t)array.shape(0);
atd::array* v = new atd::array(size, dtype, ctx);
void* data = (void*)array.get_data();
atd::copy(data, *v);
return std::shared_ptr<atd::array>(v);
}
std::shared_ptr<atd::array> create_array(bp::object const & obj, bp::object odtype, atd::driver::Context context)
{
return ndarray_to_atdarray(np::from_object(obj, to_np_dtype(extract_dtype(odtype))), context);
}
std::shared_ptr<atd::array> create_empty_array(bp::object sizes, bp::object odtype, atd::driver::Context context)
{
typedef std::shared_ptr<atd::array> result_type;
std::size_t len;
int size1;
int size2;
try{
len = bp::len(sizes);
size1 = bp::extract<int>(sizes[0])();
size2 = bp::extract<int>(sizes[1])();
}catch(bp::error_already_set const &){
PyErr_Clear();
len = 1;
size1 = bp::extract<int>(sizes)();
}
atd::numeric_type dtype = extract_dtype(odtype);
if(len < 1 || len > 2)
{
PyErr_SetString(PyExc_TypeError, "Only 1-D and 2-D arrays are supported!");
bp::throw_error_already_set();
}
if(len==1)
return result_type(new atd::array(size1, dtype, context));
return result_type(new atd::array(size1, size2, dtype, context));
}
std::string type_name(bp::object const & obj)
{
std::string name = bp::extract<std::string>(obj.attr("__class__").attr("__name__"))();
if(name=="class")
return bp::extract<std::string>(obj.attr("__name__"))();
else
return bp::extract<std::string>(obj.attr("__class__").attr("__name__"))();
}
std::shared_ptr<atd::scalar> construct_scalar(bp::object obj, atd::driver::Context const & context)
{
typedef std::shared_ptr<atd::scalar> result_type;
std::string name = type_name(obj);
if(name=="int") return result_type(new atd::scalar(bp::extract<int>(obj)(), context));
else if(name=="float") return result_type(new atd::scalar(bp::extract<double>(obj)(), context));
else if(name=="long") return result_type(new atd::scalar(bp::extract<long>(obj)(), context));
else if(name=="int") return result_type(new atd::scalar(bp::extract<int>(obj)(), context));
else if(name=="int8") return result_type(new atd::scalar(atd::CHAR_TYPE, context));
else if(name=="uint8") return result_type(new atd::scalar(atd::UCHAR_TYPE, context));
else if(name=="int16") return result_type(new atd::scalar(atd::SHORT_TYPE, context));
else if(name=="uint16") return result_type(new atd::scalar(atd::USHORT_TYPE, context));
else if(name=="int32") return result_type(new atd::scalar(atd::INT_TYPE, context));
else if(name=="uint32") return result_type(new atd::scalar(atd::UINT_TYPE, context));
else if(name=="int64") return result_type(new atd::scalar(atd::LONG_TYPE, context));
else if(name=="uint64") return result_type(new atd::scalar(atd::ULONG_TYPE, context));
else if(name=="float32") return result_type(new atd::scalar(atd::FLOAT_TYPE, context));
else if(name=="float64") return result_type(new atd::scalar(atd::DOUBLE_TYPE, context));
else{
PyErr_SetString(PyExc_TypeError, "Data type not understood");
bp::throw_error_already_set();
throw;
}
}
}
void export_array()
{
#define ADD_SCALAR_HANDLING(OP)\
.def(bp::self OP int())\
.def(bp::self OP long())\
.def(bp::self OP double())\
.def(bp::self OP bp::other<atd::value_scalar>())\
.def(int() OP bp::self)\
.def(long() OP bp::self)\
.def(double() OP bp::self)\
.def(bp::other<atd::value_scalar>() OP bp::self)
#define ADD_ARRAY_OPERATOR(OP)\
.def(bp::self OP bp::self)\
ADD_SCALAR_HANDLING(OP)
bp::class_<atd::expressions_tuple>
("array_expression_container", bp::init<atd::array_expression const &>())
;
bp::class_<atd::array_expression >("array_expression", bp::no_init)
ADD_ARRAY_OPERATOR(+)
ADD_ARRAY_OPERATOR(-)
ADD_ARRAY_OPERATOR(*)
ADD_ARRAY_OPERATOR(/)
ADD_ARRAY_OPERATOR(>)
ADD_ARRAY_OPERATOR(>=)
ADD_ARRAY_OPERATOR(<)
ADD_ARRAY_OPERATOR(<=)
ADD_ARRAY_OPERATOR(==)
ADD_ARRAY_OPERATOR(!=)
.add_property("context", bp::make_function(&atd::array_expression::context, bp::return_internal_reference<>()))
.def(bp::self_ns::abs(bp::self))
// .def(bp::self_ns::pow(bp::self))
;
#undef ADD_ARRAY_OPERATOR
#define ADD_ARRAY_OPERATOR(OP) \
.def(bp::self OP bp::self)\
.def(bp::self OP bp::other<atd::array_expression>())\
.def(bp::other<atd::array_expression>() OP bp::self) \
ADD_SCALAR_HANDLING(OP)
bp::class_<atd::array,
std::shared_ptr<atd::array> >
( "array", bp::no_init)
.def("__init__", bp::make_constructor(detail::create_array, bp::default_call_policies(), (bp::arg("obj"), bp::arg("dtype") = bp::scope().attr("float32"), bp::arg("context")=atd::driver::queues.default_context())))
.def(bp::init<atd::array_expression>())
.add_property("dtype", &atd::array::dtype)
.add_property("context", bp::make_function(&atd::array::context, bp::return_internal_reference<>()))
.add_property("T", &atd::array::T)
.add_property("shape", &detail::get_shape)
ADD_ARRAY_OPERATOR(+)
ADD_ARRAY_OPERATOR(-)
ADD_ARRAY_OPERATOR(*)
ADD_ARRAY_OPERATOR(/)
ADD_ARRAY_OPERATOR(>)
ADD_ARRAY_OPERATOR(>=)
ADD_ARRAY_OPERATOR(<)
ADD_ARRAY_OPERATOR(<=)
ADD_ARRAY_OPERATOR(==)
ADD_ARRAY_OPERATOR(!=)
.def(bp::self_ns::abs(bp::self))
// .def(bp::self_ns::pow(bp::self))
.def(bp::self_ns::str(bp::self_ns::self))
;
bp::class_<atd::scalar, bp::bases<atd::array> >
("scalar", bp::no_init)
.def("__init__", bp::make_constructor(detail::construct_scalar, bp::default_call_policies(), (bp::arg(""), bp::arg("context")=atd::driver::queues.default_context())))
;
//Other numpy-like initializers
bp::def("empty", &detail::create_empty_array, (bp::arg("shape"), bp::arg("dtype") = bp::scope().attr("float32"), bp::arg("context")=atd::driver::queues.default_context()));
//Binary
#define MAP_FUNCTION(name) \
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &, atd::array const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &, atd::array const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &, atd::array_expression const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &, atd::array_expression const &)>(&atd::name));
MAP_FUNCTION(maximum)
MAP_FUNCTION(minimum)
MAP_FUNCTION(pow)
MAP_FUNCTION(dot)
#undef MAP_FUNCTION
//Unary
#define MAP_FUNCTION(name) \
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &)>(&atd::name));
MAP_FUNCTION(abs)
MAP_FUNCTION(acos)
MAP_FUNCTION(asin)
MAP_FUNCTION(atan)
MAP_FUNCTION(ceil)
MAP_FUNCTION(cos)
MAP_FUNCTION(cosh)
MAP_FUNCTION(exp)
MAP_FUNCTION(floor)
MAP_FUNCTION(log)
MAP_FUNCTION(log10)
MAP_FUNCTION(sin)
MAP_FUNCTION(sinh)
MAP_FUNCTION(sqrt)
MAP_FUNCTION(tan)
MAP_FUNCTION(tanh)
#undef MAP_FUNCTION
/*--- Reduction operators----*/
//---------------------------------------
#define MAP_FUNCTION(name) \
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &, atd::int_t)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &, atd::int_t)>(&atd::name));
MAP_FUNCTION(sum)
MAP_FUNCTION(max)
MAP_FUNCTION(min)
MAP_FUNCTION(argmax)
MAP_FUNCTION(argmin)
#undef MAP_FUNCTION
}
void export_scalar()
{
bp::class_<atd::value_scalar>("value_scalar", bp::no_init)
.add_property("dtype", &atd::value_scalar::dtype);
}
namespace detail
{
bp::list input_sizes(isaac::base & temp, atd::expressions_tuple const & tree)
{
std::vector<int> tmp = temp.input_sizes(tree);
return detail::to_list(tmp.begin(), tmp.end());
}
std::shared_ptr<atd::model> construct_model(bp::object dtype, bp::object const & tp, atd::driver::CommandQueue & queue)
{
return std::shared_ptr<atd::model>(new atd::model(extract_template_type(tp), extract_dtype(dtype), (atd::base const &)bp::extract<atd::base>(tp), queue));
}
}
void export_model()
{
bp::class_<isaac::model>("model", bp::no_init)
.def("__init__", bp::make_constructor(detail::construct_model))
.def("execute", &atd::model::execute);
bp::enum_<isaac::fetching_policy_type>
("fetching_policy_type")
.value("FETCH_FROM_LOCAL", atd::FETCH_FROM_LOCAL)
.value("FETCH_FROM_GLOBAL_STRIDED", atd::FETCH_FROM_GLOBAL_STRIDED)
.value("FETCH_FROM_GLOBAL_CONTIGUOUS", atd::FETCH_FROM_GLOBAL_CONTIGUOUS)
;
//Base
{
#define __PROP(name) .def_readonly(#name, &isaac::base::parameters_type::name)
bp::class_<isaac::base, boost::noncopyable>("base", bp::no_init)
.def("lmem_usage", &isaac::base::lmem_usage)
.def("registers_usage", &isaac::base::registers_usage)
.def("is_invalid", &isaac::base::is_invalid)
.def("input_sizes", &detail::input_sizes)
;
#undef __PROP
}
#define WRAP_BASE(name) bp::class_<isaac::base_impl<isaac::name, isaac::name::parameters_type>, bp::bases<isaac::base>, boost::noncopyable>(#name "_base_impl", bp::no_init);
#define WRAP_TEMPLATE(name, basename, ...) bp::class_<isaac::name, bp::bases<isaac::base_impl<isaac::basename, isaac::basename::parameters_type> > >(#name, bp::init<__VA_ARGS__>())\
.add_property("local_size_0", &atd::name::local_size_0)\
.add_property("local_size_1", &atd::name::local_size_1);
#define WRAP_SINGLE_TEMPLATE(name, ...) WRAP_BASE(name) WRAP_TEMPLATE(name, name, __VA_ARGS__)
//Vector AXPY
WRAP_SINGLE_TEMPLATE(vaxpy, uint, uint, uint, isaac::fetching_policy_type)
WRAP_SINGLE_TEMPLATE(maxpy, uint, uint, uint, uint, uint, isaac::fetching_policy_type)
WRAP_SINGLE_TEMPLATE(reduction, uint, uint, uint, isaac::fetching_policy_type)
WRAP_BASE(mreduction)
WRAP_TEMPLATE(mreduction_rows, mreduction, uint, uint, uint, uint, uint, isaac::fetching_policy_type)
WRAP_TEMPLATE(mreduction_cols, mreduction, uint, uint, uint, uint, uint, isaac::fetching_policy_type)
WRAP_BASE(mproduct)
WRAP_TEMPLATE(mproduct_nn, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
WRAP_TEMPLATE(mproduct_tn, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
WRAP_TEMPLATE(mproduct_nt, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
WRAP_TEMPLATE(mproduct_tt, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
}
BOOST_PYTHON_MODULE(_isaac)
{
Py_Initialize();
np::initialize();
boost::numpy::initialize();
// specify that this module is actually a package
bp::object package = bp::scope();
package.attr("__path__") = "_isaac";
export_scalar();
export_core();
export_cl();
export_driver();
export_model();
export_array();
}

131
python/pyisaac/src/common.hpp Normal file
View File

@@ -0,0 +1,131 @@
#ifndef PYTHON_ISAAC_SRC_COMMON_HPP
#define PYTHON_ISAAC_SRC_COMMON_HPP
#include <boost/python.hpp>
#include <boost/numpy.hpp>
#include <boost/numpy/dtype.hpp>
#include "isaac/array.h"
#define MAP_ENUM(v, ns) .value(#v, ns::v)
namespace bp = boost::python;
namespace np = boost::numpy;
namespace atd = isaac;
namespace detail
{
template<class IT>
bp::list to_list(IT const & begin, IT const & end)
{
bp::list res;
for (IT it = begin; it != end; ++it)
res.append(*it);
return res;
}
template<class T>
std::vector<T> to_vector(bp::list const & list)
{
std::size_t len = bp::len(list);
std::vector<T> res; res.reserve(len);
for(std::size_t i = 0 ; i < len ; ++i)
res.push_back(boost::python::extract<T>(list[i]));
return res;
}
inline atd::numeric_type extract_dtype(bp::object const & odtype)
{
std::string name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="class")
name = bp::extract<std::string>(odtype.attr("__name__"))();
else
name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="int8") return atd::CHAR_TYPE;
else if(name=="uint8") return atd::UCHAR_TYPE;
else if(name=="int16") return atd::SHORT_TYPE;
else if(name=="uint16") return atd::USHORT_TYPE;
else if(name=="int32") return atd::INT_TYPE;
else if(name=="uint32") return atd::UINT_TYPE;
else if(name=="int64") return atd::LONG_TYPE;
else if(name=="uint64") return atd::ULONG_TYPE;
else if(name=="float32") return atd::FLOAT_TYPE;
else if(name=="float64") return atd::DOUBLE_TYPE;
else
{
PyErr_SetString(PyExc_TypeError, "Data type not understood");
bp::throw_error_already_set();
throw;
}
}
inline atd::expression_type extract_template_type(bp::object const & odtype)
{
std::string name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="class")
name = bp::extract<std::string>(odtype.attr("__name__"))();
else
name = bp::extract<std::string>(odtype.attr("__class__").attr("__name__"))();
if(name=="vaxpy") return atd::VECTOR_AXPY_TYPE;
else if(name=="maxpy") return atd::MATRIX_AXPY_TYPE;
else if(name=="reduction") return atd::REDUCTION_TYPE;
else if(name=="mreduction_rows") return atd::ROW_WISE_REDUCTION_TYPE;
else if(name=="mreduction_cols") return atd::COL_WISE_REDUCTION_TYPE;
else if(name=="mproduct_nn") return atd::MATRIX_PRODUCT_NN_TYPE;
else if(name=="mproduct_tn") return atd::MATRIX_PRODUCT_TN_TYPE;
else if(name=="mproduct_nt") return atd::MATRIX_PRODUCT_NT_TYPE;
else if(name=="mproduct_tt") return atd::MATRIX_PRODUCT_TT_TYPE;
else
{
PyErr_SetString(PyExc_TypeError, "Template type not understood");
bp::throw_error_already_set();
throw;
}
}
inline atd::numeric_type to_atd_dtype(np::dtype const & T)
{
if(T==np::detail::get_int_dtype<8, false>()) return atd::CHAR_TYPE;
else if(T==np::detail::get_int_dtype<8, true>()) return atd::UCHAR_TYPE;
else if(T==np::detail::get_int_dtype<16, false>()) return atd::SHORT_TYPE;
else if(T==np::detail::get_int_dtype<16, true>()) return atd::USHORT_TYPE;
else if(T==np::detail::get_int_dtype<32, false>()) return atd::INT_TYPE;
else if(T==np::detail::get_int_dtype<32, true>()) return atd::UINT_TYPE;
else if(T==np::detail::get_int_dtype<64, false>()) return atd::LONG_TYPE;
else if(T==np::detail::get_int_dtype<64, true>()) return atd::ULONG_TYPE;
// else if(T==np::detail::get_float_dtype<16>()) return atd::HALF_TYPE;
else if(T==np::detail::get_float_dtype<32>()) return atd::FLOAT_TYPE;
else if(T==np::detail::get_float_dtype<64>()) return atd::DOUBLE_TYPE;
else{
PyErr_SetString(PyExc_TypeError, "Unrecognized datatype");
bp::throw_error_already_set();
throw; // suppress warning; throw_error_already_set() never returns but isn't marked noreturn: https://svn.boost.org/trac/boost/ticket/1482
}
}
inline np::dtype to_np_dtype(atd::numeric_type const & T) throw()
{
if(T==atd::CHAR_TYPE) return np::detail::get_int_dtype<8, false>();
else if(T==atd::UCHAR_TYPE) return np::detail::get_int_dtype<8, true>();
else if(T==atd::SHORT_TYPE) return np::detail::get_int_dtype<16, false>();
else if(T==atd::USHORT_TYPE) return np::detail::get_int_dtype<16, true>();
else if(T==atd::INT_TYPE) return np::detail::get_int_dtype<32, false>();
else if(T==atd::UINT_TYPE) return np::detail::get_int_dtype<32, true>();
else if(T==atd::LONG_TYPE) return np::detail::get_int_dtype<64, false>();
else if(T==atd::ULONG_TYPE) return np::detail::get_int_dtype<64, true>();
// else if(T==atd::HALF_TYPE) return np::detail::get_float_dtype<16>();
else if(T==atd::FLOAT_TYPE) return np::detail::get_float_dtype<32>();
else if(T==atd::DOUBLE_TYPE) return np::detail::get_float_dtype<64>();
else{
PyErr_SetString(PyExc_TypeError, "Unrecognized datatype");
bp::throw_error_already_set();
throw; // suppress warning; throw_error_already_set() never returns but isn't marked noreturn: https://svn.boost.org/trac/boost/ticket/1482
}
}
}
#endif

293
python/pyisaac/src/core.cpp Normal file
View File

@@ -0,0 +1,293 @@
#include "core.h"
#include "common.hpp"
std::shared_ptr<atd::array>
ndarray_to_atdarray(const np::ndarray& array, const atd::driver::Context& ctx)
{
int d = array.get_nd();
if (d > 2) {
PyErr_SetString(PyExc_TypeError, "Only 1-D and 2-D arrays are supported!");
bp::throw_error_already_set();
}
atd::numeric_type dtype = detail::to_atd_dtype(array.get_dtype());
atd::int_t size = (atd::int_t)array.shape(0);
atd::array* v = new atd::array(size, dtype, ctx);
void* data = (void*)array.get_data();
atd::copy(data, *v);
return std::shared_ptr<atd::array>(v);
}
std::shared_ptr<atd::array> create_array(bp::object const & obj, bp::object odtype, atd::driver::Context context)
{
return ndarray_to_atdarray(np::from_object(obj, detail::to_np_dtype(detail::extract_dtype(odtype))), context);
}
std::shared_ptr<atd::array> create_empty_array(bp::object sizes, bp::object odtype, atd::driver::Context context)
{
typedef std::shared_ptr<atd::array> result_type;
std::size_t len;
int size1;
int size2;
try{
len = bp::len(sizes);
size1 = bp::extract<int>(sizes[0])();
size2 = bp::extract<int>(sizes[1])();
}catch(bp::error_already_set const &){
PyErr_Clear();
len = 1;
size1 = bp::extract<int>(sizes)();
}
atd::numeric_type dtype = detail::extract_dtype(odtype);
if(len < 1 || len > 2)
{
PyErr_SetString(PyExc_TypeError, "Only 1-D and 2-D arrays are supported!");
bp::throw_error_already_set();
}
if(len==1)
return result_type(new atd::array(size1, dtype, context));
return result_type(new atd::array(size1, size2, dtype, context));
}
std::string type_name(bp::object const & obj)
{
std::string name = bp::extract<std::string>(obj.attr("__class__").attr("__name__"))();
if(name=="class")
return bp::extract<std::string>(obj.attr("__name__"))();
else
return bp::extract<std::string>(obj.attr("__class__").attr("__name__"))();
}
std::shared_ptr<atd::scalar> construct_scalar(bp::object obj, atd::driver::Context const & context)
{
typedef std::shared_ptr<atd::scalar> result_type;
std::string name = type_name(obj);
if(name=="int") return result_type(new atd::scalar(bp::extract<int>(obj)(), context));
else if(name=="float") return result_type(new atd::scalar(bp::extract<double>(obj)(), context));
else if(name=="long") return result_type(new atd::scalar(bp::extract<long>(obj)(), context));
else if(name=="int") return result_type(new atd::scalar(bp::extract<int>(obj)(), context));
else if(name=="int8") return result_type(new atd::scalar(atd::CHAR_TYPE, context));
else if(name=="uint8") return result_type(new atd::scalar(atd::UCHAR_TYPE, context));
else if(name=="int16") return result_type(new atd::scalar(atd::SHORT_TYPE, context));
else if(name=="uint16") return result_type(new atd::scalar(atd::USHORT_TYPE, context));
else if(name=="int32") return result_type(new atd::scalar(atd::INT_TYPE, context));
else if(name=="uint32") return result_type(new atd::scalar(atd::UINT_TYPE, context));
else if(name=="int64") return result_type(new atd::scalar(atd::LONG_TYPE, context));
else if(name=="uint64") return result_type(new atd::scalar(atd::ULONG_TYPE, context));
else if(name=="float32") return result_type(new atd::scalar(atd::FLOAT_TYPE, context));
else if(name=="float64") return result_type(new atd::scalar(atd::DOUBLE_TYPE, context));
else{
PyErr_SetString(PyExc_TypeError, "Data type not understood");
bp::throw_error_already_set();
throw;
}
}
bp::tuple get_shape(atd::array const & x)
{
return bp::make_tuple(x.shape()[0], x.shape()[1]);
}
template<class T>
struct datatype : public atd::value_scalar
{
datatype(T t) : atd::value_scalar(t){ }
};
template<class T>
unsigned int size(datatype<T> const & dt)
{ return atd::size_of(dt.dtype()) ; }
#define INSTANTIATE(name, clname) \
struct name : public datatype<clname> { name(clname value) : datatype<clname>(value){} };
INSTANTIATE(int8, cl_char)
INSTANTIATE(uint8, cl_uchar)
INSTANTIATE(int16, cl_short)
INSTANTIATE(uint16, cl_ushort)
INSTANTIATE(int32, cl_int)
INSTANTIATE(uint32, cl_uint)
INSTANTIATE(int64, cl_long)
INSTANTIATE(uint64, cl_ulong)
INSTANTIATE(float32, cl_float)
INSTANTIATE(float64, cl_double)
#undef INSTANTIATE
void export_core()
{
/*-------------------
* TYPES
*------------------*/
#define INSTANTIATE(name, clname) \
bp::class_<datatype<clname>, bp::bases<atd::value_scalar> >(#name, bp::init<clname>());\
bp::class_<name, bp::bases<datatype<clname> > >(#name, bp::init<clname>())\
.add_property("size", &size<clname>)\
;
INSTANTIATE(int8, cl_char)
INSTANTIATE(uint8, cl_uchar)
INSTANTIATE(int16, cl_short)
INSTANTIATE(uint16, cl_ushort)
INSTANTIATE(int32, cl_int)
INSTANTIATE(uint32, cl_uint)
INSTANTIATE(int64, cl_long)
INSTANTIATE(uint64, cl_ulong)
INSTANTIATE(float32, cl_float)
INSTANTIATE(float64, cl_double)
#undef INSTANTIATE
bp::enum_<atd::expression_type>("operations")
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(MATRIX_AXPY_TYPE, atd)
MAP_ENUM(REDUCTION_TYPE, atd)
MAP_ENUM(ROW_WISE_REDUCTION_TYPE, atd)
MAP_ENUM(COL_WISE_REDUCTION_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
MAP_ENUM(VECTOR_AXPY_TYPE, atd)
;
/*-------------------
* ARRAY
*------------------*/
#define ADD_SCALAR_HANDLING(OP)\
.def(bp::self OP int())\
.def(bp::self OP long())\
.def(bp::self OP double())\
.def(bp::self OP bp::other<atd::value_scalar>())\
.def(int() OP bp::self)\
.def(long() OP bp::self)\
.def(double() OP bp::self)\
.def(bp::other<atd::value_scalar>() OP bp::self)
#define ADD_ARRAY_OPERATOR(OP)\
.def(bp::self OP bp::self)\
ADD_SCALAR_HANDLING(OP)
bp::class_<atd::expressions_tuple>
("array_expression_container", bp::init<atd::array_expression const &>())
;
bp::class_<atd::array_expression >("array_expression", bp::no_init)
ADD_ARRAY_OPERATOR(+)
ADD_ARRAY_OPERATOR(-)
ADD_ARRAY_OPERATOR(*)
ADD_ARRAY_OPERATOR(/)
ADD_ARRAY_OPERATOR(>)
ADD_ARRAY_OPERATOR(>=)
ADD_ARRAY_OPERATOR(<)
ADD_ARRAY_OPERATOR(<=)
ADD_ARRAY_OPERATOR(==)
ADD_ARRAY_OPERATOR(!=)
.add_property("context", bp::make_function(&atd::array_expression::context, bp::return_internal_reference<>()))
.def(bp::self_ns::abs(bp::self))
// .def(bp::self_ns::pow(bp::self))
;
#undef ADD_ARRAY_OPERATOR
#define ADD_ARRAY_OPERATOR(OP) \
.def(bp::self OP bp::self)\
.def(bp::self OP bp::other<atd::array_expression>())\
.def(bp::other<atd::array_expression>() OP bp::self) \
ADD_SCALAR_HANDLING(OP)
bp::class_<atd::array,
std::shared_ptr<atd::array> >
( "array", bp::no_init)
.def("__init__", bp::make_constructor(create_array, bp::default_call_policies(), (bp::arg("obj"), bp::arg("dtype") = bp::scope().attr("float32"), bp::arg("context")=atd::driver::queues.default_context())))
.def(bp::init<atd::array_expression>())
.add_property("dtype", &atd::array::dtype)
.add_property("context", bp::make_function(&atd::array::context, bp::return_internal_reference<>()))
.add_property("T", &atd::array::T)
.add_property("shape", &get_shape)
ADD_ARRAY_OPERATOR(+)
ADD_ARRAY_OPERATOR(-)
ADD_ARRAY_OPERATOR(*)
ADD_ARRAY_OPERATOR(/)
ADD_ARRAY_OPERATOR(>)
ADD_ARRAY_OPERATOR(>=)
ADD_ARRAY_OPERATOR(<)
ADD_ARRAY_OPERATOR(<=)
ADD_ARRAY_OPERATOR(==)
ADD_ARRAY_OPERATOR(!=)
.def(bp::self_ns::abs(bp::self))
// .def(bp::self_ns::pow(bp::self))
.def(bp::self_ns::str(bp::self_ns::self))
;
bp::class_<atd::scalar, bp::bases<atd::array> >
("scalar", bp::no_init)
.def("__init__", bp::make_constructor(construct_scalar, bp::default_call_policies(), (bp::arg(""), bp::arg("context")=atd::driver::queues.default_context())))
;
//Other numpy-like initializers
bp::def("empty", &create_empty_array, (bp::arg("shape"), bp::arg("dtype") = bp::scope().attr("float32"), bp::arg("context")=atd::driver::queues.default_context()));
//Binary
#define MAP_FUNCTION(name) \
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &, atd::array const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &, atd::array const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &, atd::array_expression const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &, atd::array_expression const &)>(&atd::name));
MAP_FUNCTION(maximum)
MAP_FUNCTION(minimum)
MAP_FUNCTION(pow)
MAP_FUNCTION(dot)
#undef MAP_FUNCTION
//Unary
#define MAP_FUNCTION(name) \
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &)>(&atd::name));
MAP_FUNCTION(abs)
MAP_FUNCTION(acos)
MAP_FUNCTION(asin)
MAP_FUNCTION(atan)
MAP_FUNCTION(ceil)
MAP_FUNCTION(cos)
MAP_FUNCTION(cosh)
MAP_FUNCTION(exp)
MAP_FUNCTION(floor)
MAP_FUNCTION(log)
MAP_FUNCTION(log10)
MAP_FUNCTION(sin)
MAP_FUNCTION(sinh)
MAP_FUNCTION(sqrt)
MAP_FUNCTION(tan)
MAP_FUNCTION(tanh)
#undef MAP_FUNCTION
/*--- Reduction operators----*/
//---------------------------------------
#define MAP_FUNCTION(name) \
bp::def(#name, static_cast<atd::array_expression (*)(atd::array const &, atd::int_t)>(&atd::name));\
bp::def(#name, static_cast<atd::array_expression (*)(atd::array_expression const &, atd::int_t)>(&atd::name));
MAP_FUNCTION(sum)
MAP_FUNCTION(max)
MAP_FUNCTION(min)
MAP_FUNCTION(argmax)
MAP_FUNCTION(argmin)
#undef MAP_FUNCTION
/*-------------------
* SCALAR
*------------------*/
bp::class_<atd::value_scalar>("value_scalar", bp::no_init)
.add_property("dtype", &atd::value_scalar::dtype);
}

6
python/pyisaac/src/core.h Normal file
View File

@@ -0,0 +1,6 @@
#ifndef PYISAAC_SRC_CORE_H
#define PYISAAC_SRC_CORE_H
void export_core();
#endif

165
python/pyisaac/src/driver.cpp Normal file
View File

@@ -0,0 +1,165 @@
#include <boost/python.hpp>
#include <boost/python/suite/indexing/vector_indexing_suite.hpp>
#include <boost/python/suite/indexing/map_indexing_suite.hpp>
#include "isaac/array.h"
#include "isaac/model/model.h"
#include "common.hpp"
#include "driver.h"
bp::list nv_compute_capability(atd::driver::Device const & device)
{
bp::list res;
std::pair<unsigned int, unsigned int> cc = device.nv_compute_capability();
res.append(cc.first);
res.append(cc.second);
return res;
}
bp::list get_platforms()
{
std::vector<atd::driver::Platform> platforms(atd::driver::Platform::get());
return detail::to_list(platforms.begin(), platforms.end());
}
bp::list get_devices(atd::driver::Platform const & platform)
{
std::vector<atd::driver::Device> devices(platform.devices());
return detail::to_list(devices.begin(), devices.end());
}
struct model_map_indexing
{
static atd::model& get_item(atd::model_map_t& container, bp::tuple i_)
{
atd::expression_type expression = detail::extract_template_type(i_[0]);
atd::numeric_type dtype = detail::extract_dtype(i_[1]);
atd::model_map_t::iterator i = container.find(std::make_pair(expression, dtype));
if (i == container.end())
{
PyErr_SetString(PyExc_KeyError, "Invalid key");
bp::throw_error_already_set();
}
return *i->second;
}
static void set_item(atd::model_map_t& container, bp::tuple i_, atd::model const & v)
{
atd::expression_type expression = detail::extract_template_type(i_[0]);
atd::numeric_type dtype = detail::extract_dtype(i_[1]);
container[std::make_pair(expression, dtype)].reset(new atd::model(v));
}
};
std::string to_string(atd::driver::device_type type)
{
if(type==atd::driver::DEVICE_TYPE_CPU) return "CPU";
if(type==atd::driver::DEVICE_TYPE_GPU) return "GPU";
if(type==atd::driver::DEVICE_TYPE_ACCELERATOR) return "ACCELERATOR";
throw;
}
std::shared_ptr<atd::driver::Context> make_context(atd::driver::Device const & dev)
{ return std::shared_ptr<atd::driver::Context>(new atd::driver::Context(dev)); }
bp::tuple flush(atd::array_expression const & expression, unsigned int queue_id, bp::list dependencies, bool tune, int label, std::string const & program_name, bool force_recompile)
{
std::list<atd::driver::Event> events;
std::vector<atd::driver::Event> cdependencies = detail::to_vector<atd::driver::Event>(dependencies);
std::shared_ptr<atd::array> parray(new atd::array(atd::control(expression,
atd::execution_options_type(queue_id, &events, &cdependencies),
atd::dispatcher_options_type(tune, label),
atd::compilation_options_type(program_name, force_recompile))));
return bp::make_tuple(parray, detail::to_list(events.begin(), events.end()));
}
struct state_type{ };
state_type state;
void export_cl()
{
typedef std::vector<atd::driver::CommandQueue> queues_t;
bp::class_<queues_t>("queues")
.def("__len__", &queues_t::size)
.def("__getitem__", &bp::vector_indexing_suite<queues_t>::get_item, bp::return_internal_reference<>())
.def("__setitem__", &bp::vector_indexing_suite<queues_t>::set_item, bp::with_custodian_and_ward<1,2>())
.def("append", &bp::vector_indexing_suite<queues_t>::append)
;
bp::class_<atd::model_map_t>("models")
.def("__getitem__", &model_map_indexing::get_item, bp::return_internal_reference<>())
.def("__setitem__", &model_map_indexing::set_item, bp::with_custodian_and_ward<1,2>())
;
bp::enum_<atd::driver::backend_type>
("backend_type")
.value("OPENCL", atd::driver::OPENCL)
#ifdef ISAAC_WITH_CUDA
.value("CUDA", atd::driver::CUDA)
#endif
;
bp::enum_<atd::driver::device_type>
("device_type")
.value("DEVICE_TYPE_GPU", atd::driver::DEVICE_TYPE_GPU)
.value("DEVICE_TYPE_CPU", atd::driver::DEVICE_TYPE_CPU)
;
bp::class_<atd::driver::Platform>("platform", bp::no_init)
.def("get_devices", &get_devices)
.add_property("name",&atd::driver::Platform::name)
;
bp::enum_<isaac::driver::Device::VENDOR>
("vendor")
.value("AMD", atd::driver::Device::AMD)
.value("INTEL", atd::driver::Device::INTEL)
.value("NVIDIA", atd::driver::Device::NVIDIA)
.value("UNKNOWN", atd::driver::Device::UNKNOWN)
;
bp::class_<atd::driver::Device>("device", bp::no_init)
.add_property("clock_rate", &atd::driver::Device::clock_rate)
.add_property("name", &atd::driver::Device::name)
.add_property("type", &atd::driver::Device::type)
.add_property("platform", &atd::driver::Device::platform)
.add_property("vendor", &atd::driver::Device::vendor)
.add_property("nv_compute_capability", &nv_compute_capability)
;
bp::class_<atd::driver::Context>("context", bp::no_init)
.def("__init__", bp::make_constructor(&make_context))
.add_property("queues", bp::make_function(static_cast<std::vector<atd::driver::CommandQueue> & (*)(const atd::driver::Context&)>( [](const atd::driver::Context & ctx) -> std::vector<atd::driver::CommandQueue> & { return atd::driver::queues[ctx]; }) , bp::return_internal_reference<>()))
.add_property("backend", &atd::driver::Context::backend)
;
bp::class_<atd::driver::CommandQueue>("command_queue", bp::init<atd::driver::Context, atd::driver::Device>())
.def("synchronize", &atd::driver::CommandQueue::synchronize)
.add_property("models", bp::make_function(&atd::get_model_map, bp::return_internal_reference<>()))
.add_property("device", bp::make_function(&atd::driver::CommandQueue::device, bp::return_internal_reference<>()))
;
bp::class_<atd::driver::Event>("event", bp::init<atd::driver::backend_type>())
.add_property("elapsed_time", &atd::driver::Event::elapsed_time)
;
bp::def("device_type_to_string", &to_string);
bp::def("get_platforms", &get_platforms);
bp::def("flush", &flush, (bp::arg("expression"), bp::arg("queue_id") = 0, bp::arg("dependencies")=bp::list(), bp::arg("tune") = false, bp::arg("label")=-1, bp::arg("program_name")="", bp::arg("recompile") = false));
bp::class_<state_type>("state_type")
.def_readwrite("queue_properties",&atd::driver::queues.queue_properties)
;
bp::scope().attr("state") = bp::object(bp::ptr(&state));
bp::scope().attr("CL_QUEUE_PROFILING_ENABLE") = CL_QUEUE_PROFILING_ENABLE;
bp::scope().attr("CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE") = CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE;
}

6
python/pyisaac/src/driver.h Normal file
View File

@@ -0,0 +1,6 @@
#ifndef PYISAAC_SRC_DRIVER_H
#define PYISAAC_SRC_DRIVER_H
void export_driver();
#endif

66
python/pyisaac/src/model.cpp Normal file
View File

@@ -0,0 +1,66 @@
#include "isaac/backend/templates/vaxpy.h"
#include "isaac/backend/templates/maxpy.h"
#include "isaac/backend/templates/reduction.h"
#include "isaac/backend/templates/mreduction.h"
#include "isaac/backend/templates/mproduct.h"
#include "isaac/model/model.h"
#include "model.h"
#include "common.hpp"
bp::list input_sizes(isaac::base & temp, atd::expressions_tuple const & tree)
{
std::vector<int> tmp = temp.input_sizes(tree);
return detail::to_list(tmp.begin(), tmp.end());
}
std::shared_ptr<atd::model> construct_model(bp::object dtype, bp::object const & tp, atd::driver::CommandQueue & queue)
{
return std::shared_ptr<atd::model>(new atd::model(detail::extract_template_type(tp), detail::extract_dtype(dtype), (atd::base const &)bp::extract<atd::base>(tp), queue));
}
void export_model()
{
bp::class_<isaac::model>("model", bp::no_init)
.def("__init__", bp::make_constructor(construct_model))
.def("execute", &atd::model::execute);
bp::enum_<isaac::fetching_policy_type>
("fetching_policy_type")
.value("FETCH_FROM_LOCAL", atd::FETCH_FROM_LOCAL)
.value("FETCH_FROM_GLOBAL_STRIDED", atd::FETCH_FROM_GLOBAL_STRIDED)
.value("FETCH_FROM_GLOBAL_CONTIGUOUS", atd::FETCH_FROM_GLOBAL_CONTIGUOUS)
;
//Base
{
#define __PROP(name) .def_readonly(#name, &isaac::base::parameters_type::name)
bp::class_<isaac::base, boost::noncopyable>("base", bp::no_init)
.def("lmem_usage", &isaac::base::lmem_usage)
.def("registers_usage", &isaac::base::registers_usage)
.def("is_invalid", &isaac::base::is_invalid)
.def("input_sizes", &input_sizes)
;
#undef __PROP
}
#define WRAP_BASE(name) bp::class_<isaac::base_impl<isaac::name, isaac::name::parameters_type>, bp::bases<isaac::base>, boost::noncopyable>(#name "_base_impl", bp::no_init);
#define WRAP_TEMPLATE(name, basename, ...) bp::class_<isaac::name, bp::bases<isaac::base_impl<isaac::basename, isaac::basename::parameters_type> > >(#name, bp::init<__VA_ARGS__>())\
.add_property("local_size_0", &atd::name::local_size_0)\
.add_property("local_size_1", &atd::name::local_size_1);
#define WRAP_SINGLE_TEMPLATE(name, ...) WRAP_BASE(name) WRAP_TEMPLATE(name, name, __VA_ARGS__)
//Vector AXPY
WRAP_SINGLE_TEMPLATE(vaxpy, uint, uint, uint, isaac::fetching_policy_type)
WRAP_SINGLE_TEMPLATE(maxpy, uint, uint, uint, uint, uint, isaac::fetching_policy_type)
WRAP_SINGLE_TEMPLATE(reduction, uint, uint, uint, isaac::fetching_policy_type)
WRAP_BASE(mreduction)
WRAP_TEMPLATE(mreduction_rows, mreduction, uint, uint, uint, uint, uint, isaac::fetching_policy_type)
WRAP_TEMPLATE(mreduction_cols, mreduction, uint, uint, uint, uint, uint, isaac::fetching_policy_type)
WRAP_BASE(mproduct)
WRAP_TEMPLATE(mproduct_nn, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
WRAP_TEMPLATE(mproduct_tn, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
WRAP_TEMPLATE(mproduct_nt, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
WRAP_TEMPLATE(mproduct_tt, mproduct, uint, uint, uint, uint, uint, uint, uint, uint, isaac::fetching_policy_type, isaac::fetching_policy_type, uint, uint)
}

6
python/pyisaac/src/model.h Normal file
View File

@@ -0,0 +1,6 @@
#ifndef PYISAAC_SRC_MODEL_H
#define PYISAAC_SRC_MODEL_H
void export_model();
#endif