Skip to content

[mlir][bufferization] Fix promote-buffers-to-stack crash caused by nested memref #127441

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

Open
wants to merge 2 commits into
base: main
Choose a base branch
from
Open

[mlir][bufferization] Fix promote-buffers-to-stack crash caused by nested memref #127441

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
403cc1e
[mlir][bufferization] Fix `promote-buffers-to-stack` crash cause by n…
CoTinker Feb 17, 2025
0df1d0f
Update promote-buffers-to-stack.mlir
CoTinker Feb 17, 2025
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
7 changes: 6 additions & 1 deletion mlir/lib/Dialect/Bufferization/Transforms/BufferOptimizations.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,6 +89,11 @@ static bool defaultIsSmallAlloc(Value alloc, unsigned maximumSizeInBytes,
auto type = dyn_cast<ShapedType>(alloc.getType());
if (!type || !alloc.getDefiningOp<memref::AllocOp>())
return false;
auto elementType = type.getElementType();
// Not support nested memref type.
if (isa<MemRefType>(elementType))
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This will still crash on UnrankedMemRefType and certain types that implement MemRefElementTypeInterface. I think the right way to fix this is to add a helper function DataLayout::hasKnownTypeSize(Type). Internally, there could be function that returns std::optional<TypeSize>, which is called by both getTypeSizeInBits and hasKnownTypeSize.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Great idea.

return false;

if (!type.hasStaticShape()) {
// Check if the dynamic shape dimension of the alloc is produced by
// `memref.rank`. If this is the case, it is likely to be small.
Expand All @@ -104,7 +109,7 @@ static bool defaultIsSmallAlloc(Value alloc, unsigned maximumSizeInBytes,
return false;
}
unsigned bitwidth = mlir::DataLayout::closest(alloc.getDefiningOp())
.getTypeSizeInBits(type.getElementType());
.getTypeSizeInBits(elementType);
return type.getNumElements() * bitwidth <= maximumSizeInBytes * 8;
}

Expand Down
11 changes: 11 additions & 0 deletions mlir/test/Transforms/promote-buffers-to-stack.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -611,3 +611,14 @@ module attributes { dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<index, 256>>} {
// LOWLIMIT-NEXT: memref.alloc() {alignment = 64 : i64, custom_attr}
// RANK-NEXT: memref.alloca() {alignment = 64 : i64, custom_attr}
// CHECK-NEXT: return

// -----

// Test Case: AllocOp with nested memref type, it is not
// converted. Ensure this case not crash.

// CHECK-LABEL: func @nested_memref
func.func @nested_memref() {
%0 = memref.alloc() : memref<1xmemref<1xf32>>
return
}