Skip to content

[Gluon][Layout][AMD] Fix layoutToGluon() about handling PaddedSharedEncodingAttr's CGA-layout #10392

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
antiagainst merged 1 commit into from
May 28, 2026
Merged

[Gluon][Layout][AMD] Fix layoutToGluon() about handling PaddedSharedEncodingAttr's CGA-layout #10392

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
12 changes: 9 additions & 3 deletions python/src/gluon_ir.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -286,10 +286,16 @@ py::object layoutToGluon(Attribute layout) {
auto kOffset = mlir::StringAttr::get(ctx, "offset");
auto kBlock = mlir::StringAttr::get(ctx, "block");
const auto &ll = paddedShared.getLinearComponent();
auto outDims = llvm::to_vector(ll.getOutDimNames());

auto ofstBases = ll.getBases().lookup(kOffset);
auto ofstLL = triton::LinearLayout({{kOffset, ofstBases}}, outDims);
auto blkLL = divideLeft(ll, ofstLL);
assert(blkLL.has_value());
auto blkBases = blkLL->getBases().lookup(kBlock);
auto shape = toStdVector(ll.getOutDimSizes());
return layouts.PaddedSharedLayout(intervalPaddingPairs,
ll.getBases().lookup(kOffset),
ll.getBases().lookup(kBlock), shape);
return layouts.PaddedSharedLayout(intervalPaddingPairs, ofstBases, blkBases,
shape);
} else if (auto partitioned =
dyn_cast<ttg::PartitionedSharedEncodingAttr>(layout)) {
py::object partitionLayout =
Expand Down
20 changes: 20 additions & 0 deletions python/test/gluon/test_frontend.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,7 @@
HIP_TARGET_GFX1250 = GPUTarget("hip", "gfx1250", 32)

ALL_TARGETS = [AMPERE_TARGET, HOPPER_TARGET, BLACKWELL_TARGET, HIP_TARGET_RDNA4]
ALL_MULTICTA_TARGETS = [HOPPER_TARGET, BLACKWELL_TARGET, HIP_TARGET_GFX1250]


def anonymize_ir(ir):
Expand Down Expand Up @@ -3350,6 +3351,18 @@ def infer_layout_for_padded_shared_kernel():
ttgl.static_assert(reshaped.type.layout == ref_layout)


@gluon.jit
def test_convert_padded_shared_with_multicta_kernel():
shape: ttgl.constexpr = [512, 128]
initial_order: ttgl.constexpr = [0, 1]
layout: ttgl.constexpr = ttgl.PaddedSharedLayout.with_identity_for(interval_padding_pairs=[[256, 16]],
cga_layout=[[0, 1]], shape=shape,
order=initial_order)
smem = ttgl.allocate_shared_memory(ttgl.int32, shape, layout)
reshaped = smem.permute((0, 1))
ttgl.static_assert(reshaped.layout.cga_layout[0] == ttgl.constexpr([0, 1]))


@pytest.mark.parametrize("target", ALL_TARGETS)
def test_infer_layout_for_padded_shared(target):
# This test is used to test the conversion to gluon object PaddedSharedLayout from PaddedSharedEncodingAttr.
Expand All @@ -3370,6 +3383,13 @@ def test_infer_layout_for_padded_shared(target):
""")


@pytest.mark.parametrize("target", ALL_MULTICTA_TARGETS)
def test_convert_padded_shared_with_multicta(target):
# It is to make sure layoutToGluon() handle CGA layout correctly when
# converting a PaddedSharedEncodingAttr to a PaddedSharedLayout object.
run_parser(test_convert_padded_shared_with_multicta_kernel, *make_args(num_ctas=2), target=target)


@filecheck_test
@gluon.jit
def test_layout_zeros():
Expand Down
Loading