Skip to content

[mlir] Add affine.delinearize_index and affine.linearize_index ValueBoundsOpInterfaceImpl #118829

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

Closed
wants to merge 1 commit into from
Closed

[mlir] Add affine.delinearize_index and affine.linearize_index ValueBoundsOpInterfaceImpl #118829

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
65 changes: 65 additions & 0 deletions mlir/lib/Dialect/Affine/IR/ValueBoundsOpInterfaceImpl.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -49,6 +49,67 @@ struct AffineApplyOpInterface
}
};

struct AffineDelinearizeIndexOpInterface
: public ValueBoundsOpInterface::ExternalModel<
AffineDelinearizeIndexOpInterface, AffineDelinearizeIndexOp> {
void populateBoundsForIndexValue(Operation *op, Value value,
ValueBoundsConstraintSet &cstr) const {
auto delinearizeOp = cast<AffineDelinearizeIndexOp>(op);
auto result = cast<OpResult>(value);
int64_t resultIdx = result.getResultNumber();
assert(result.getOwner() == delinearizeOp && "invalid value");

AffineExpr linearIdxExpr = cstr.getExpr(delinearizeOp.getLinearIndex());
SmallVector<OpFoldResult> basis = delinearizeOp.getMixedBasis();
SmallVector<AffineExpr> basisExprs;
AffineExpr modExpr = getAffineConstantExpr(1, op->getContext());
AffineExpr strideExpr = getAffineConstantExpr(1, op->getContext());
for (int i = basis.size() - 1; i >= resultIdx; --i) {
AffineExpr basisExpr = cstr.getExpr(basis[i]);
modExpr = modExpr * basisExpr;
if (i > resultIdx)
Copy link
Contributor

Choose a reason for hiding this comment

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

It feels like an equivalent way to phrase this is to walk up to the index before resultIdx, take the product of those, and then go one step further to get the final mod.

However, I'm not going to insist too much on that

strideExpr = strideExpr * basisExpr;
}
AffineExpr bound = linearIdxExpr;
Copy link
Contributor

Choose a reason for hiding this comment

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

nit: A comment describing the expression would be nice

if (resultIdx > 0)
bound = bound % modExpr;
if (resultIdx < delinearizeOp->getNumResults())
bound = bound.floorDiv(strideExpr);

cstr.bound(value) == bound;
}
};

struct AffineLinearizeIndexOpInterface
: public ValueBoundsOpInterface::ExternalModel<
AffineLinearizeIndexOpInterface, AffineLinearizeIndexOp> {
void populateBoundsForIndexValue(Operation *op, Value value,
ValueBoundsConstraintSet &cstr) const {
auto linearizeOp = cast<AffineLinearizeIndexOp>(op);
assert(value == linearizeOp.getResult() && "invalid value");

SmallVector<OpFoldResult> basis = linearizeOp.getMixedBasis();
SmallVector<AffineExpr> basisExprs = llvm::map_to_vector(
basis, [&](OpFoldResult ofr) { return cstr.getExpr(ofr); });
basisExprs.push_back(getAffineConstantExpr(1, op->getContext()));

SmallVector<OpFoldResult> indices(linearizeOp.getMultiIndex());
SmallVector<AffineExpr> indexExprs = llvm::map_to_vector(
indices, [&](OpFoldResult ofr) { return cstr.getExpr(ofr); });
Copy link
Contributor

Choose a reason for hiding this comment

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

instead of creating the vector and then reversing, you can do llvm::map_to_vector(llvm::reverse(indices), ... here and above.


AffineExpr linearIdxExpr = getAffineConstantExpr(0, op->getContext());
AffineExpr strideExpr = getAffineConstantExpr(1, op->getContext());
std::reverse(indexExprs.begin(), indexExprs.end());
std::reverse(basisExprs.begin(), basisExprs.end());
for (size_t i = 0; i < indexExprs.size(); ++i) {
strideExpr = strideExpr * basisExprs[i];
linearIdxExpr = linearIdxExpr + indexExprs[i] * strideExpr;
}

cstr.bound(value) == linearIdxExpr;
}
};

struct AffineMinOpInterface
: public ValueBoundsOpInterface::ExternalModel<AffineMinOpInterface,
AffineMinOp> {
Expand Down Expand Up @@ -98,6 +159,10 @@ void mlir::affine::registerValueBoundsOpInterfaceExternalModels(
DialectRegistry &registry) {
registry.addExtension(+[](MLIRContext *ctx, AffineDialect *dialect) {
AffineApplyOp::attachInterface<AffineApplyOpInterface>(*ctx);
AffineDelinearizeIndexOp::attachInterface<
AffineDelinearizeIndexOpInterface>(*ctx);
AffineLinearizeIndexOp::attachInterface<AffineLinearizeIndexOpInterface>(
*ctx);
AffineMaxOp::attachInterface<AffineMaxOpInterface>(*ctx);
AffineMinOp::attachInterface<AffineMinOpInterface>(*ctx);
});
Expand Down
37 changes: 37 additions & 0 deletions mlir/test/Dialect/Affine/value-bounds-op-interface-impl.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -155,3 +155,40 @@ func.func @compare_maps(%a: index, %b: index) {
: (index, index, index, index) -> ()
return
}

// -----

func.func @compare_affine_linearize_index(%a: index, %b: index) {
%0 = affine.linearize_index [%a, %b] by (2, 4) : index
%1 = affine.linearize_index [%a, %b] by (4) : index
// expected-remark @below{{true}}
"test.compare"(%0, %a, %b) {rhs_map = affine_map<(a, b) -> (a * 4 + b)>}
: (index, index, index) -> ()
// expected-remark @below{{true}}
"test.compare"(%1, %a, %b) {rhs_map = affine_map<(a, b) -> (a * 4 + b)>}
: (index, index, index) -> ()
return
}

// -----

// CHECK: #[[$MAP:.*]] = affine_map<()[s0] -> (s0 floordiv 4)>
// CHECK: #[[$MAP1:.*]] = affine_map<()[s0] -> (s0 mod 4)>

// CHECK-LABEL: func @affine_delinearize_index(
// CHECK-SAME: %[[a:.*]]: index
func.func @affine_delinearize_index(%a: index) -> (index, index, index, index) {
%0:2 = affine.delinearize_index %a into (2, 4) : index, index
%1:2 = affine.delinearize_index %a into (4) : index, index

// CHECK: %[[BOUND0:.+]] = affine.apply #[[$MAP]]()[%[[a]]]
%2 = "test.reify_bound"(%0#0) {type = "EQ"} : (index) -> (index)
// CHECK: %[[BOUND1:.+]] = affine.apply #[[$MAP1]]()[%[[a]]]
%3 = "test.reify_bound"(%0#1) {type = "EQ"} : (index) -> (index)
// CHECK: %[[BOUND2:.+]] = affine.apply #[[$MAP]]()[%[[a]]]
%4 = "test.reify_bound"(%0#0) {type = "EQ"} : (index) -> (index)
// CHECK: %[[BOUND3:.+]] = affine.apply #[[$MAP1]]()[%[[a]]]
%5 = "test.reify_bound"(%0#1) {type = "EQ"} : (index) -> (index)
Copy link
Contributor

Choose a reason for hiding this comment

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

This is testing the bounds of %0 twice, did you mean %1 here?

// CHECK: return %[[BOUND0]], %[[BOUND1]], %[[BOUND2]], %[[BOUND3]]
return %2, %3, %4, %5: index, index, index, index
}
Loading