Skip to content

[mlir][sparse] make sparse compiler more admissible. #90927

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
merged 1 commit into from
May 3, 2024
Merged

[mlir][sparse] make sparse compiler more admissible. #90927

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
6 changes: 0 additions & 6 deletions mlir/lib/Dialect/Linalg/Transforms/ElementwiseOpFusion.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -432,12 +432,6 @@ class FuseElementwiseOps : public OpRewritePattern<GenericOp> {

Operation *producer = opOperand.get().getDefiningOp();

// Do not fuse a sparse-in/dense-out operation, as the
// result is too often not sparsifiable anymore.
if (sparse_tensor::hasAnySparseOperand(producer) &&
!sparse_tensor::hasAnySparseResult(producer))
return failure();

// Find the producer of the operand.
FailureOr<ElementwiseOpFusionResult> fusionResult =
fuseElementwiseOps(rewriter, &opOperand);
Expand Down
54 changes: 29 additions & 25 deletions mlir/lib/Dialect/SparseTensor/Utils/Merger.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1356,50 +1356,54 @@ Merger::buildTensorExp(linalg::GenericOp op, Value v) {
// See buildLattices() for an explanation of rejecting certain
// division and shift operations.
if (def->getNumOperands() == 2) {
const auto [x, xDepSp] = buildTensorExp(op, def->getOperand(0));
const auto [y, yDepSp] = buildTensorExp(op, def->getOperand(1));
bool hasSpDep = xDepSp || yDepSp;
const auto [x, xSpVals] = buildTensorExp(op, def->getOperand(0));
const auto [y, ySpVals] = buildTensorExp(op, def->getOperand(1));
// For a conjunctive operation, it yields a "sparse" result if any operand
// is sparse. For a disjunctive operation, it yields a "sparse" result if
// all operands are sparse.
bool conjSpVals = xSpVals || ySpVals;
bool disjSpVals = xSpVals && ySpVals;
if (x.has_value() && y.has_value()) {
const ExprId e0 = *x;
const ExprId e1 = *y;
if (isa<arith::MulFOp>(def))
return {addExp(TensorExp::Kind::kMulF, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kMulF, e0, e1), conjSpVals};
if (isa<complex::MulOp>(def))
return {addExp(TensorExp::Kind::kMulC, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kMulC, e0, e1), conjSpVals};
if (isa<arith::MulIOp>(def))
return {addExp(TensorExp::Kind::kMulI, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kMulI, e0, e1), conjSpVals};
if (isa<arith::DivFOp>(def) && !maybeZero(e1))
return {addExp(TensorExp::Kind::kDivF, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kDivF, e0, e1), conjSpVals};
if (isa<complex::DivOp>(def) && !maybeZero(e1))
return {addExp(TensorExp::Kind::kDivC, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kDivC, e0, e1), conjSpVals};
if (isa<arith::DivSIOp>(def) && !maybeZero(e1))
return {addExp(TensorExp::Kind::kDivS, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kDivS, e0, e1), conjSpVals};
if (isa<arith::DivUIOp>(def) && !maybeZero(e1))
return {addExp(TensorExp::Kind::kDivU, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kDivU, e0, e1), conjSpVals};
if (isa<arith::AddFOp>(def))
return {addExp(TensorExp::Kind::kAddF, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kAddF, e0, e1), disjSpVals};
if (isa<complex::AddOp>(def))
return {addExp(TensorExp::Kind::kAddC, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kAddC, e0, e1), disjSpVals};
if (isa<arith::AddIOp>(def))
return {addExp(TensorExp::Kind::kAddI, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kAddI, e0, e1), disjSpVals};
if (isa<arith::SubFOp>(def))
return {addExp(TensorExp::Kind::kSubF, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kSubF, e0, e1), disjSpVals};
if (isa<complex::SubOp>(def))
return {addExp(TensorExp::Kind::kSubC, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kSubC, e0, e1), disjSpVals};
if (isa<arith::SubIOp>(def))
return {addExp(TensorExp::Kind::kSubI, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kSubI, e0, e1), disjSpVals};
if (isa<arith::AndIOp>(def))
return {addExp(TensorExp::Kind::kAndI, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kAndI, e0, e1), conjSpVals};
if (isa<arith::OrIOp>(def))
return {addExp(TensorExp::Kind::kOrI, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kOrI, e0, e1), disjSpVals};
if (isa<arith::XOrIOp>(def))
return {addExp(TensorExp::Kind::kXorI, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kXorI, e0, e1), disjSpVals};
if (isa<arith::ShRSIOp>(def) && isInvariant(e1))
return {addExp(TensorExp::Kind::kShrS, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kShrS, e0, e1), conjSpVals};
if (isa<arith::ShRUIOp>(def) && isInvariant(e1))
return {addExp(TensorExp::Kind::kShrU, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kShrU, e0, e1), conjSpVals};
if (isa<arith::ShLIOp>(def) && isInvariant(e1))
return {addExp(TensorExp::Kind::kShlI, e0, e1), hasSpDep};
return {addExp(TensorExp::Kind::kShlI, e0, e1), conjSpVals};
if (auto ci = dyn_cast<arith::CmpIOp>(def)) {
if (ci.getPredicate() == arith::CmpIPredicate::eq &&
ci.getPredicate() == arith::CmpIPredicate::sle &&
Expand All @@ -1413,7 +1417,7 @@ Merger::buildTensorExp(linalg::GenericOp op, Value v) {

auto e = addExp(TensorExp::Kind::kCmpI, e0, e1, nullptr,
ci.getPredicateAttr());
return {e, hasSpDep};
return {e, conjSpVals};
}
if (auto cf = dyn_cast<arith::CmpFOp>(def)) {
if (cf.getPredicate() == arith::CmpFPredicate::OEQ &&
Expand All @@ -1431,15 +1435,15 @@ Merger::buildTensorExp(linalg::GenericOp op, Value v) {
}
auto e = addExp(TensorExp::Kind::kCmpF, e0, e1, nullptr,
cf.getPredicateAttr());
return {e, hasSpDep};
return {e, conjSpVals};
}
if (auto binop = dyn_cast<sparse_tensor::BinaryOp>(def)) {
if (isAdmissibleBranch(binop, binop.getOverlapRegion()) &&
(binop.getLeftIdentity() ||
isAdmissibleBranch(binop, binop.getLeftRegion())) &&
(binop.getRightIdentity() ||
isAdmissibleBranch(binop, binop.getRightRegion())))
return {addExp(TensorExp::Kind::kBinary, e0, e1, def), hasSpDep};
return {addExp(TensorExp::Kind::kBinary, e0, e1, def), conjSpVals};
}
}
}
Expand Down
69 changes: 50 additions & 19 deletions mlir/test/Dialect/SparseTensor/sparse_fusion.mlir
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,4 +1,4 @@
// RUN: mlir-opt %s --linalg-fuse-elementwise-ops | FileCheck %s
// RUN: mlir-opt %s --linalg-fuse-elementwise-ops --sparse-reinterpret-map --sparsification | FileCheck %s

#SV = #sparse_tensor.encoding<{ map = (d0) -> (d0 : compressed) }>

Expand All @@ -11,35 +11,66 @@
doc = "B(i) = OP A(i)"
}

// CHECK-LABEL: func @sparse_fusion
// CHECK: linalg.generic
// CHECK: arith.addf
// CHECK: linalg.generic
// CHECK: math.exp
// CHECK: arith.maximumf
// CHECK-NOT: linalg.generic
// CHECK: return

// CHECK-LABEL: func.func @sparse_fusion(
// CHECK-SAME: %[[VAL_0:.*]]: tensor<100xf64, #sparse>) -> tensor<100xf64> {
// CHECK-DAG: %[[VAL_1:.*]] = arith.constant true
// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 1 : index
// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 0 : index
// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0.000000e+00 : f64
// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 100 : index
// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1.000000e+00 : f64
// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 1.000000e+02 : f64
// CHECK-DAG: %[[VAL_8:.*]] = tensor.empty() : tensor<100xf64>
// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<100xf64, #sparse> to memref<?xindex>
// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<100xf64, #sparse> to memref<?xindex>
// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<100xf64, #sparse> to memref<?xf64>
// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_8]] : memref<100xf64>
// CHECK: linalg.fill ins(%[[VAL_4]] : f64) outs(%[[VAL_12]] : memref<100xf64>)
// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_3]]] : memref<?xindex>
// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_2]]] : memref<?xindex>
// CHECK: %[[VAL_15:.*]]:2 = scf.while (%[[VAL_16:.*]] = %[[VAL_13]], %[[VAL_17:.*]] = %[[VAL_3]]) : (index, index) -> (index, index) {
// CHECK: %[[VAL_18:.*]] = arith.cmpi ult, %[[VAL_16]], %[[VAL_14]] : index
// CHECK: scf.condition(%[[VAL_18]]) %[[VAL_16]], %[[VAL_17]] : index, index
// CHECK: } do {
// CHECK: ^bb0(%[[VAL_19:.*]]: index, %[[VAL_20:.*]]: index):
// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_19]]] : memref<?xindex>
// CHECK: %[[VAL_22:.*]] = arith.cmpi eq, %[[VAL_21]], %[[VAL_20]] : index
// CHECK: scf.if %[[VAL_22]] {
// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_19]]] : memref<?xf64>
// CHECK: %[[VAL_24:.*]] = arith.addf %[[VAL_23]], %[[VAL_6]] : f64
// CHECK: %[[VAL_25:.*]] = math.exp %[[VAL_24]] : f64
// CHECK: %[[VAL_26:.*]] = arith.maximumf %[[VAL_25]], %[[VAL_7]] : f64
// CHECK: memref.store %[[VAL_26]], %[[VAL_12]]{{\[}}%[[VAL_20]]] : memref<100xf64>
// CHECK: } else {
// CHECK: scf.if %[[VAL_1]] {
// CHECK: memref.store %[[VAL_7]], %[[VAL_12]]{{\[}}%[[VAL_20]]] : memref<100xf64>
// CHECK: } else {
// CHECK: }
// CHECK: }
// CHECK: %[[VAL_27:.*]] = arith.cmpi eq, %[[VAL_21]], %[[VAL_20]] : index
// CHECK: %[[VAL_28:.*]] = arith.addi %[[VAL_19]], %[[VAL_2]] : index
// CHECK: %[[VAL_29:.*]] = arith.select %[[VAL_27]], %[[VAL_28]], %[[VAL_19]] : index
// CHECK: %[[VAL_30:.*]] = arith.addi %[[VAL_20]], %[[VAL_2]] : index
// CHECK: scf.yield %[[VAL_29]], %[[VAL_30]] : index, index
// CHECK: }
// CHECK: scf.for %[[VAL_31:.*]] = %[[VAL_32:.*]]#1 to %[[VAL_5]] step %[[VAL_2]] {
// CHECK: memref.store %[[VAL_7]], %[[VAL_12]]{{\[}}%[[VAL_31]]] : memref<100xf64>
// CHECK: }
// CHECK: %[[VAL_33:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<100xf64>
// CHECK: return %[[VAL_33]] : tensor<100xf64>
// CHECK: }
func.func @sparse_fusion(%argA: tensor<100xf64, #SV>) -> tensor<100xf64> {
%c1 = arith.constant 1.0 : f64
%c100 = arith.constant 100.0 : f64

//
// Densifying op.
// Should not be fused with subsequent dense ops.
//
%t0 = tensor.empty() : tensor<100xf64>
%l0 = linalg.generic #trait
ins(%argA: tensor<100xf64, #SV>) outs(%t0: tensor<100xf64>) {
^bb0(%in0: f64, %out0: f64):
%b0 = arith.addf %in0, %c1 : f64
linalg.yield %b0 : f64
} -> tensor<100xf64>


//
// Two following dense ops.
// Should be fused, but not with above.
//
%t1 = tensor.empty() : tensor<100xf64>
%l1 = linalg.generic #trait
ins(%l0: tensor<100xf64>) outs(%t1: tensor<100xf64>) {
Expand Down