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

[codegen] some more optimizations

Browse Source
This commit is contained in:
Philippe Tillet
2019-07-18 19:39:40 -07:00
parent 71594da66f
commit 5215fb0424
12 changed files with 108 additions and 73 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
lib/dnn/base.cpp
View File

@@ -30,7 +30,7 @@ void base::enqueue(driver::stream *stream, std::vector<driver::buffer *> args, b
/* the current template has not already been compiled */
if(m_jit.find(this) == m_jit.end()) {
base* clone = this->clone();
jit = m_jit.emplace(clone, std::unique_ptr<rt::jit>(new rt::jit(ctx))).first->second.get();
jit = m_jit.emplace(clone, std::unique_ptr<rt::jit>(new rt::jit(ctx, 8))).first->second.get();
std::ostringstream oss;
clone->triton_c_src(oss);
std::string src = oss.str();
@@ -51,7 +51,8 @@ void base::enqueue(driver::stream *stream, std::vector<driver::buffer *> args, b
jit->add_module(name_.c_str(), src.c_str(), best.params);
}
else {
jit->add_module(name_.c_str(), src.c_str(), jit->get_valid(name_.c_str(), src.c_str()));
// jit->add_module(name_.c_str(), src.c_str(), jit->get_valid(name_.c_str(), src.c_str()));
jit->add_module(name_.c_str(), src.c_str(), {32, 128, 16, 128, 2, 2, 2, 2, 4, 4, 32, 8, 4, 1});
}
triton::driver::kernel* kernel = jit->get_function(name_.c_str());
clone->init_impl(stream, (triton::driver::cu_module*)kernel->module());