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

removing C++11 interface

Browse Source
This commit is contained in:
Philippe Tillet
2015-02-08 23:19:38 -05:00
parent 85fb438806
commit a6d7671831
21 changed files with 423 additions and 956 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
python/autotune/pysrc/autotune.py
View File

@@ -143,7 +143,7 @@ def do_tuning(args):
A = atd.empty(sizes, datatype, context=context)
C = atd.empty(sizes, datatype, context=context)
return execute(A + C, sizes, Template, parameters, fname)
tune(execution_handler, 100, 5000, 2, (),'log', 'log')
tune(execution_handler, 64, 5000, 2, (),'log', 'log')
#Row-wise dot
if operation=='gemv':
for A_trans in args.gemv_layouts:
@@ -152,7 +152,7 @@ def do_tuning(args):
x = atd.empty(sizes[1], datatype, context=context)
LHS = A if A_trans=='N' else A.T
return execute(atd.dot(LHS, x), sizes, Template[A_trans], parameters, fname)
tune(execution_handler, 100, 5000, 2, (A_trans,),'log', 'log')
tune(execution_handler, 64, 6000, 2, (A_trans,),'log', 'log')
#Matrix Product
if operation=='gemm':
for L in args.gemm_layouts:
@@ -194,7 +194,7 @@ class ArgumentsHandler:
full_parser = tune_subparsers.add_parser('full', help = 'Tune each operation for randomly chosen sizes')
full_parser.add_argument("--build-model", default=True, type=bool)
full_parser.add_argument("--sample-size", default=30, type=int)
full_parser.add_argument("--sample-size", default=60, type=int)
args = parser.parse_args()
self.__dict__ = args.__dict__.copy()

11
python/autotune/pysrc/misc_tools.py
View File

@@ -218,11 +218,16 @@ def benchmark(template, symbolic):
raise ValueError("Template has too low occupancy")
else:
queue.models[template, atd.float32] = atd.model(template, queue)
x = atd.array(symbolic)
atd.synchronize(symbolic.context)
x, events, cache = atd.flush(symbolic)
atd.synchronize(symbolic.context)
return 1e-9*sum([e.end - e.start for e in events])
timings = []
current_time = 0
while current_time < 1e-3:
x, events, cache = atd.flush(symbolic)
atd.synchronize(symbolic.context)
timings.append(1e-9*sum([e.end - e.start for e in events]))
current_time = current_time + timings[-1]
return np.median(timings)
def sanitize_string(string, keep_chars = ['_']):

2
python/pyatidlas/external/boost/include/boost/functional/hash/extensions.hpp vendored
View File

@@ -193,7 +193,7 @@ namespace boost
#if !defined(BOOST_NO_CXX11_SMART_PTR)
template <typename T>
inline std::size_t hash_value(std::shared_ptr<T> const& x) {
inline std::size_t hash_value(tools::shared_ptr<T> const& x) {
return boost::hash_value(x.get());
}

2
python/pyatidlas/external/boost/include/boost/get_pointer.hpp vendored
View File

@@ -36,7 +36,7 @@ template<class T> T * get_pointer( std::unique_ptr<T> const& p )
return p.get();
}
template<class T> T * get_pointer( std::shared_ptr<T> const& p )
template<class T> T * get_pointer( tools::shared_ptr<T> const& p )
{
return p.get();
}

9
python/pyatidlas/src/_atidlas.cpp
View File

@@ -315,15 +315,14 @@ namespace detail
boost::shared_ptr<cl::Context> make_context(cl::Device const & dev)
{ return boost::shared_ptr<cl::Context>(new cl::Context(std::vector<cl::Device>(1, dev))); }
bp::tuple flush(atd::array_expression const & expression, unsigned int queue_id, bp::list dependencies, int label, std::string const & program_name, bool force_recompile)
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<cl::Event> events;
atd::operation_cache cache;
std::vector<cl::Event> cdependencies = to_vector<cl::Event>(dependencies);
boost::shared_ptr<atd::array> parray(new atd::array(atd::control(expression, atd::execution_options_type(queue_id, &events, &cache, &cdependencies),
atd::dispatcher_options_type(label), atd::compilation_options_type(program_name, force_recompile))));
return bp::make_tuple(*parray, to_list(events.begin(), events.end()), cache);
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()), cache);
}
}
@@ -402,7 +401,7 @@ void export_cl()
bp::def("synchronize", &atd::cl_ext::synchronize);
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("label")=-1, bp::arg("program_name")="", bp::arg("recompile") = false));
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::cl_ext::queue_properties)