Skip to content

[Feat] Allow dangling producer in wasp pipeline planning (#1263) #1647

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
LeiWang1999 merged 3 commits into from
Jan 9, 2026
Merged

[Feat] Allow dangling producer in wasp pipeline planning (#1263) #1647

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
96f2983
[Fix] Handle cases where producers lack corresponding consumers in sy…
kurisu6912 Jan 9, 2026
2baca64
add test for pipeline without consumer
kurisu6912 Jan 9, 2026
aa91188
fix lint error
kurisu6912 Jan 9, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 17 additions & 1 deletion src/transform/warp_specialized_rewriter.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -751,7 +751,23 @@ class WSCodeEmitter : public StmtMutator {
: parity_;
block_stmt.push_back(makeParityWait(acquire_barrier_id, parity));
}
ICHECK(!map.release[i].empty());
// It is possible that a producer does not participate in any
// producer-consumer dependency that requires synchronization.
// In that case, there will be no associated release pattern.
// We should still emit the (optionally guarded) statement without
// inserting any mbarrier for it instead of failing.
if (map.release[i].empty()) {
LOG(WARNING) << "Producer doesn't have corresponding consumer: "
<< seq_transformed[i];
block_stmt.push_back(seq_transformed[i]);
new_body.push_back(
MakeGroupBlock(block_stmt.size() == 1
? block_stmt[0]
// NOLINTNEXTLINE(performance-move-const-arg)
: SeqStmt(std::move(block_stmt)),
annotations));
continue;
}
for (size_t j = 0; j < map.release[i].size(); j++) {
int pattern_idx = map.release[i][j];
PrimExpr release_barrier_id =
Expand Down
32 changes: 32 additions & 0 deletions testing/python/issue/test_tilelang_issue_1263.py
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,32 @@
import tilelang.testing
import tilelang.language as T


def test_issue_1263_pipeline_no_consumer():
@tilelang.jit()
def test_kernel(M, N):
dtype = "bfloat16"

@T.prim_func
def fwd_main(
KV: T.Tensor((M, N), dtype),
ids: T.Tensor((4,), "int32"),
ids2: T.Tensor((4,), "int32"),
):
with T.Kernel(4, threads=1):
A = T.alloc_shared([N], dtype)
B = T.alloc_shared([N], dtype)

for i in T.Pipelined(4, num_stages=1):
id = ids[i]
id2 = ids2[id]
T.copy(KV[id2, :], A)
T.clear(B)

return fwd_main

test_kernel(1024, 128)


if __name__ == "__main__":
tilelang.testing.main()
Loading