[FRONTEND] Added compilation flag to force use of .nc cache modifier (#134)

in DRAM loads. /!\ USE CAREFULLY - THIS CAN BREAK CORRECTNESS IF MISUSED
/!\
This commit is contained in:
Philippe Tillet
2021-07-20 17:58:06 -04:00
committed by Philippe Tillet
parent 2824345065
commit 01276b5153
7 changed files with 27 additions and 18 deletions

View File

@@ -78,11 +78,11 @@ void init_triton_driver(py::module &&m) {
void init_triton_codegen(py::module &&m) {
m.def(
"add_passes_to_emit_bin", [](ir::module &ir, drv::device *dev, int num_warps, int num_stages) {
"add_passes_to_emit_bin", [](ir::module &ir, drv::device *dev, int num_warps, int num_stages, bool force_nc_cache) {
drv::module *mod;
drv::kernel *ker;
size_t shared_mem;
triton::codegen::add_passes_to_emit_bin(ir, dev, num_warps, num_stages, mod, ker, shared_mem);
triton::codegen::add_passes_to_emit_bin(ir, dev, num_warps, num_stages, force_nc_cache, mod, ker, shared_mem);
std::stringstream ss;
ir::print(ir, ss);
return std::make_tuple(mod, ker, shared_mem, ss.str());