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

[CODEGEN] Add cache modifier to tl.load (#351)

Browse Source 
* Add cache modifier to tl.load
* Add comment to cache_modifier
* Remove force_nc_cache
* Update test
This commit is contained in:
daadaada
2021-10-18 13:14:04 +08:00
committed by GitHub
parent 90ded16c32
commit 858dec8372
16 changed files with 119 additions and 63 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
lib/codegen/pass.cc
View File

@@ -25,7 +25,7 @@ namespace codegen {
// TODO:
// There should be a proper pass manager there!
std::unique_ptr<llvm::Module> add_passes_to_emit_bin(ir::module &ir, llvm::LLVMContext& ctx, codegen::target* target,
int cc, int num_warps, int num_stages, bool force_nc_cache, int& shared_static) {
int cc, int num_warps, int num_stages, int& shared_static) {
// generate llvm code
std::string name = ir.get_function_list()[0]->get_name();
std::unique_ptr<llvm::Module> llvm(new llvm::Module(name, ctx));
@@ -46,7 +46,7 @@ std::unique_ptr<llvm::Module> add_passes_to_emit_bin(ir::module &ir, llvm::LLVMC
codegen::transform::coalesce coalesce(&align, &layouts);
codegen::transform::prefetch prefetch_s(target);
codegen::transform::membar barriers(&liveness, &layouts, &allocation, &prefetch_s, target);
codegen::generator isel(&axes, &layouts, &align, &allocation, &swizzle, target, num_warps, force_nc_cache);
codegen::generator isel(&axes, &layouts, &align, &allocation, &swizzle, target, num_warps);
// run passes
dce.run(ir);
peephole.run(ir);