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

Improvements w/ Auto-Tuning and standard benchmarks (#57)

Browse Source 
[PYTHON] Bug-fixes in the auto-tuning module and improvement of the existing API for it
This commit is contained in:
Philippe Tillet
2021-02-03 13:37:21 -08:00
committed by Philippe Tillet
parent ad005d49ac
commit 6fb4800f57
12 changed files with 215 additions and 149 deletions
Download Patch File Download Diff File Expand all files Collapse all files

48
lib/runtime/function.cc
View File

@@ -224,7 +224,7 @@ void kernel::operator()(void *args, size_t args_size, driver::stream *stream, co
for(size_t i = 0; i < 3; i++)
grid[i] = (i < _grid.size()) ? _grid[i] : 1;
// enqueue
stream->enqueue(&*ker_, grid, {opt.num_warps * 32, 1, 1}, args, args_size);
stream->enqueue(&*ker_, grid, {(size_t)opt.num_warps * 32, 1, 1}, args, args_size);
}
std::string kernel::get_asm(asm_mode_t mode) {
@@ -282,35 +282,35 @@ void function::do_loop_nest(std::vector<size_t> const & ranges,
return;
values[i--] = 0;
}
i = D - 1;
i = D - 1; options_t opt;
}
}
void function::init_kernels(const std::string& src, const options_space_t& opts, driver::device *device) {
// all ranges
std::vector<size_t> ranges;
ranges.push_back(opts.num_warps.size());
for(const auto& x: opts.defines)
ranges.push_back(x.second.size());
// functor for source with given option
void function::init_kernels(const std::string& src, const options_t& opt,
const autotune_vals_t& confs, driver::device *device) {
// list of all possible configs
// just augment `opt` with each define of `confs`
// and override warp count
size_t num_opts = std::max(confs.size(), (size_t)1);
std::vector<options_t> opts(num_opts, opt);
for(size_t i = 0; i < confs.size(); i++){
opts[i].defines.insert(confs[i].first.begin(), confs[i].first.end());
opts[i].num_warps = confs[i].second;
}
// compile all possible configs
// compilation errors (e.g., too much shared mem)
// will populate `err`
std::vector<std::pair<options_t, std::string>> err;
auto do_make = [&](std::vector<size_t> params) {
// compilation options
unsigned i = 0;
options_t opt;
opt.num_warps = opts.num_warps[params[i++]];
for(auto D: opts.defines)
opt.defines[D.first] = D.second[params[i++]];
// compile
for(const options_t& opt: opts) {
try{
kernels_.push_back({opt, std::make_shared<kernel>(src, opt, device)});
}catch(const exception::base& e){
err.push_back({opt, e.what()});
}
};
// multi-threaded compilation
do_loop_nest(ranges, do_make);
}
// throw an exception if `err` is not empty
if(kernels_.empty()){
std::ostringstream dbg;
dbg << "Auto-Tuner could not find any valid configuration:" << std::endl;
@@ -357,9 +357,11 @@ kernel* function::autotune(void* args, size_t args_size, const grid_fn_ty& grid_
return it->second;
}
function::function(const std::string& src, const options_space_t& opt,
driver::device *device, const std::vector<std::string>& autotune_key) {
init_kernels(src, opt, device);
function::function(const std::string& src, const options_t &opt, driver::device *device,
const autotune_vals_t& autotune_vals, const std::vector<std::string>& autotune_key) {
// pre-compile all kernels
init_kernels(src, opt, autotune_vals, device);
// find indices of autotune keys
auto arg_names = kernels_.at(0).second->get_arg_names();
for(const std::string& name: autotune_key){
auto it = std::find(arg_names.begin(), arg_names.end(), name);