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

[CODEGEN] Bugfix in membar pass (#124)

Browse Source 
Membar pass on top of master is buggy with asynchronous copy. For example, it doesn't wait for asynchronous copies to complete before recoalescing accumulator in GEMM, which leads to undefined behavior when the program doesn't enter the loop. This PR proposes
This commit is contained in:
Philippe Tillet
2021-06-14 12:22:48 -04:00
committed by Philippe Tillet
parent b7b05a560e
commit 5a51f3e529
6 changed files with 51 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
lib/codegen/pass.cc
View File

@@ -92,7 +92,9 @@ void add_passes_to_emit_bin(ir::module &ir, driver::device *dev, int num_warps,
liveness.run(ir);
allocation.run(ir);
prefetch_s.run(ir);
// ir::print(ir, std::cout);
barriers.run(ir);
// ir::print(ir, std::cout);
// ir::print(ir, std::cout);
isel.visit(ir, *llvm);
mod = driver::module::create(dev, std::move(llvm));