Skip to content

debug: log propagation details #5896

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
wujingyue merged 5 commits into from
Feb 3, 2026
+38 −12
Merged

debug: log propagation details #5896

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
379e88d
Print propagation for debugging
wujingyue Jan 29, 2026
0ee7101
Use TransformPropagator's debugging option
wujingyue Jan 30, 2026
e6c6491
Merge branch 'main' into wjy/prop
wujingyue Feb 2, 2026
7e7e5a7
Merge branch 'main' into wjy/prop
wujingyue Feb 3, 2026
080fdea
Update csrc/multidevice/propagation.cpp
wujingyue Feb 3, 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
24 changes: 13 additions & 11 deletions csrc/base.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -281,24 +281,26 @@ SPECIALIZE_PRINTER(VoidStar);
SPECIALIZE_PRINTER(uint32_t);
SPECIALIZE_PRINTER(int64_t);
SPECIALIZE_PRINTER(uint64_t);
SPECIALIZE_PRINTER(DataType);
SPECIALIZE_PRINTER(MemoryType);
SPECIALIZE_PRINTER(UnaryOpType);

SPECIALIZE_PRINTER(BinaryOpType);
SPECIALIZE_PRINTER(TernaryOpType);
SPECIALIZE_PRINTER(LoadStoreOpType);
SPECIALIZE_PRINTER(CircularBufferLoopStage);
SPECIALIZE_PRINTER(tma::TensorMapInterleave);
SPECIALIZE_PRINTER(tma::TensorMapL2Promotion);
SPECIALIZE_PRINTER(tma::TensorMapFloatOOBFill);
SPECIALIZE_PRINTER(DataType);
SPECIALIZE_PRINTER(LoadStoreOpType);
SPECIALIZE_PRINTER(MemoryType);
SPECIALIZE_PRINTER(MmaInputSmemSwizzle);
SPECIALIZE_PRINTER(SwizzleType);
SPECIALIZE_PRINTER(ParallelType);
SPECIALIZE_PRINTER(SwizzleMode);
SPECIALIZE_PRINTER(SwizzleType);
SPECIALIZE_PRINTER(TernaryOpType);
SPECIALIZE_PRINTER(UnaryOpType);
SPECIALIZE_PRINTER(std::optional<bool>);
SPECIALIZE_PRINTER(std::vector<int64_t>);
SPECIALIZE_PRINTER(std::vector<int>);
SPECIALIZE_PRINTER(std::vector<uint32_t>);
SPECIALIZE_PRINTER(std::vector<int64_t>);
SPECIALIZE_PRINTER(std::vector<uint64_t>);
SPECIALIZE_PRINTER(std::optional<bool>);
SPECIALIZE_PRINTER(tma::TensorMapFloatOOBFill);
SPECIALIZE_PRINTER(tma::TensorMapInterleave);
SPECIALIZE_PRINTER(tma::TensorMapL2Promotion);

#undef SPECIALIZE_PRINTER

Expand Down
7 changes: 7 additions & 0 deletions csrc/multidevice/propagation.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,7 @@
#include <unordered_map>
#include <vector>

#include "base.h"
#include "ir/interface_nodes.h"
#include "ir/internal_base_nodes.h"
#include "ir/internal_nodes.h"
Expand Down Expand Up @@ -255,6 +256,12 @@ void shardLoopLike(
TensorView* target,
const std::unordered_set<ParallelType>& selected_parallel_types,
PropagateDirection direction) {
if (isDebugDumpEnabled(DebugDumpOption::TransformPropagator)) {
debug() << "Propagating shardings from " << ref->toString() << " to "
<< target->toString() << " in " << direction << " for "
<< toDelimitedString(selected_parallel_types) << std::endl;
}

std::unordered_set<IterDomain*> device_or_stream_ids;
const std::unordered_map<IterDomain*, IterDomain*> ref2target =
getRef2TargetMap(ref, target, direction);
Expand Down
1 change: 0 additions & 1 deletion csrc/preseg_passes/propagate_shardings.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,7 +13,6 @@
#include "ir/iostream.h"
#include "ir/utils.h"
#include "multidevice/propagation.h"
#include "multidevice/utils.h"
#include "scheduler/utils.h"

namespace nvfuser::preseg_passes {
Expand Down
16 changes: 16 additions & 0 deletions csrc/scheduler/utils.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -43,6 +43,22 @@
#include <type.h>
#include <val_graph_visitor.h>

namespace nvfuser {

std::ostream& operator<<(std::ostream& os, PropagateDirection direction) {
switch (direction) {
case PropagateDirection::kForward:
os << "Forward";
break;
case PropagateDirection::kBackward:
os << "Backward";
break;
}
return os;
}

} // namespace nvfuser

namespace nvfuser::scheduler_utils {

// Minimal PTX code for a no-op kernel, used for occupancy queries
Expand Down
2 changes: 2 additions & 0 deletions csrc/scheduler/utils.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,6 +29,8 @@ class HeuristicDataCache;
//! BoundedDirectionalTransformPropagator.
enum class PropagateDirection { kBackward = 0, kForward };

std::ostream& operator<<(std::ostream& os, PropagateDirection direction);

namespace scheduler_utils {

// Assume any only half of the register file is available to spend on buffers,
Expand Down