Skip to content
Draft
Changes from all commits
Commits
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
17 changes: 17 additions & 0 deletions src/transform/inject_tma_barrier.cc
Original file line number Diff line number Diff line change
Expand Up @@ -504,6 +504,23 @@ class TmaBarrierRewriter : public IRMutatorWithAnalyzer {
return IRMutatorWithAnalyzer::VisitStmt_(op);
}

Stmt VisitStmt_(const EvaluateNode *op) final {
// Check whether this is ptx_arrive_barrier()
if (const CallNode *call = op->value.as<CallNode>()) {
if (call->op.same_as(builtin::ptx_arrive_barrier())) {
// producer arrives → guard with tl_shuffle_elect()
if (has_warp_specialization_ && is_producer_) {
PrimExpr cond = Call(DataType::Bool(), tl_shuffle_elect(),
{IntImm(DataType::Int(32), 128)});
Stmt arrive_stmt = StmtExprMutator::VisitStmt_(op);
return IfThenElse(cond, arrive_stmt, Stmt());
}
}
}

return StmtExprMutator::VisitStmt_(op);
}

PrimExpr VisitExpr_(const CallNode *op) {
if (op->op.same_as(tma_load()) || op->op.same_as(tma_load_im2col())) {
auto call_ref = tvm::ffi::GetRef<Call>(op);
Expand Down
Loading