Skip to content

Handle conflicts with existing type during backward remat #9953

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
neildhar merged 1 commit into from
Apr 9, 2026
+29 −9
Merged

Handle conflicts with existing type during backward remat #9953

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
17 changes: 8 additions & 9 deletions lib/Dialect/TritonGPU/Transforms/Utility.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -864,7 +864,6 @@ LogicalResult getConvertBackwardSlice(

auto updateLayout = [&](Value value, Attribute encoding) {
assert((isa<RankedTensorType>(value.getType())));
slice.insert(value);
Attribute &existing = layout[value];
if (existing && existing != encoding)
return failure();
Expand All @@ -876,7 +875,8 @@ LogicalResult getConvertBackwardSlice(
auto [currentValueUse, encoding] = queue.back();
Value currentValue = currentValueUse->get();
queue.pop_back();
if (!isa<RankedTensorType>(currentValue.getType()))
auto currentValueType = dyn_cast<RankedTensorType>(currentValue.getType());
if (!currentValueType)
continue;
// Skip propagating through for op/while op/ws op results for now.
// TODO: enable this based on needs.
Expand All @@ -885,6 +885,11 @@ LogicalResult getConvertBackwardSlice(
return failure();
if (failed(updateLayout(currentValue, encoding)))
return failure();
// If the value already has the desired encoding, we can stop here without
// adding it to the slice.
if (currentValueType.getEncoding() == encoding)
continue;
slice.insert(currentValue);

// If there is already an existing conversion to the target layout, we don't
// need to propagate to the operands.
Expand Down Expand Up @@ -915,6 +920,7 @@ LogicalResult getConvertBackwardSlice(
continue;
if (failed(updateLayout(result, encoding)))
return failure();
slice.insert(result);
}
if (isFreeConvert(definingOp)) {
enqueue(definingOp->getOpOperand(0), encoding);
Expand Down Expand Up @@ -945,13 +951,6 @@ LogicalResult getConvertBackwardSlice(
}
if (!srcEncoding)
return failure();
// If the infered layout matches the original one we don't need to keep
// propagating.
if (auto operandType =
dyn_cast<RankedTensorType>(operand.get().getType())) {
if (srcEncoding == operandType.getEncoding())
continue;
}
enqueue(operand, srcEncoding);
}
continue;
Expand Down
21 changes: 21 additions & 0 deletions test/TritonGPU/combine.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -4190,3 +4190,24 @@ module attributes {"ttg.num-ctas" = 1 : i32, "ttg.num-warps" = 4 : i32, "ttg.thr
tt.return %0#0 : i32
}
}

// -----

#src = #ttg.blocked<{sizePerThread = [1, 2], threadsPerWarp = [1, 1], warpsPerCTA = [1, 1], order = [0, 1]}>
#dst = #ttg.blocked<{sizePerThread = [1, 2, 2], threadsPerWarp = [1, 1, 1], warpsPerCTA = [1, 1, 1], order = [0, 1, 2]}>
#lin = #ttg.linear<{register = [[0, 1, 0]], lane = [], warp = [], block = []}>
module attributes {"ttg.num-warps" = 1 : i32, "ttg.threads-per-warp" = 1 : i32} {
// CHECK-LABEL: @test_existing_layout_conflict
// CHECK: ttg.convert_layout
// CHECK: tt.return
tt.func @test_existing_layout_conflict() -> tensor<1x2x2xi32, #dst> {
%r = tt.make_range {end = 2 : i32, start = 0 : i32} : tensor<2xi32, #ttg.slice<{dim = 0, parent = #src}>>
%v = tt.expand_dims %r {axis = 0 : i32} : tensor<2xi32, #ttg.slice<{dim = 0, parent = #src}>> -> tensor<1x2xi32, #src>
%j = tt.join %v, %v : tensor<1x2xi32, #src> -> tensor<1x2x2xi32, #lin>
%r3 = tt.reshape %v : tensor<1x2xi32, #src> -> tensor<1x2x1xi32, #lin>
%b = tt.broadcast %r3 : tensor<1x2x1xi32, #lin> -> tensor<1x2x2xi32, #lin>
%s = arith.addi %j, %b : tensor<1x2x2xi32, #lin>
%o = ttg.convert_layout %s : tensor<1x2x2xi32, #lin> -> tensor<1x2x2xi32, #dst>
tt.return %o : tensor<1x2x2xi32, #dst>
}
}