| 1 | //===- ShardingInterfaceImpl.cpp ------------------------------------------===// |
| 2 | // |
| 3 | // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
| 4 | // See https://llvm.org/LICENSE.txt for license information. |
| 5 | // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
| 6 | // |
| 7 | //===----------------------------------------------------------------------===// |
| 8 | |
| 9 | #include "mlir/Dialect/Mesh/Interfaces/ShardingInterface.h" |
| 10 | #include "mlir/Dialect/Mesh/Interfaces/ShardingInterfaceImpl.h" |
| 11 | #include "mlir/Dialect/Tensor/IR/ShardingInterfaceImpl.h" |
| 12 | #include "mlir/Dialect/Tensor/IR/Tensor.h" |
| 13 | #include "mlir/IR/DialectRegistry.h" |
| 14 | |
| 15 | using namespace mlir; |
| 16 | using namespace mlir::tensor; |
| 17 | using namespace mlir::mesh; |
| 18 | |
| 19 | namespace { |
| 20 | |
| 21 | // Sharding of tensor.empty/tensor.splat |
| 22 | template <typename OpTy> |
| 23 | struct CreatorOpShardingInterface |
| 24 | : public ShardingInterface::ExternalModel<CreatorOpShardingInterface<OpTy>, |
| 25 | OpTy> { |
| 26 | SmallVector<utils::IteratorType> getLoopIteratorTypes(Operation *op) const { |
| 27 | auto ndims = mlir::cast<ShapedType>(Val: op->getResult(idx: 0).getType()).getRank(); |
| 28 | return SmallVector<utils::IteratorType>(ndims, |
| 29 | utils::IteratorType::parallel); |
| 30 | } |
| 31 | |
| 32 | SmallVector<AffineMap> getIndexingMaps(Operation *op) const { |
| 33 | MLIRContext *ctx = op->getContext(); |
| 34 | Value val = op->getResult(idx: 0); |
| 35 | auto type = dyn_cast<RankedTensorType>(Val: val.getType()); |
| 36 | if (!type) |
| 37 | return {}; |
| 38 | return SmallVector<AffineMap>( |
| 39 | op->getNumOperands() + op->getNumResults(), |
| 40 | {AffineMap::getMultiDimIdentityMap(numDims: type.getRank(), context: ctx)}); |
| 41 | } |
| 42 | |
| 43 | LogicalResult spmdize(Operation *op, ArrayRef<Value> spmdizedOperands, |
| 44 | ArrayRef<MeshSharding> operandShardings, |
| 45 | ArrayRef<MeshSharding> resultShardings, |
| 46 | IRMapping &spmdizationMap, |
| 47 | SymbolTableCollection &symbolTable, |
| 48 | OpBuilder &builder) const { |
| 49 | assert(resultShardings.size() == 1); |
| 50 | auto resType = cast<RankedTensorType>(Val: op->getResult(idx: 0).getType()); |
| 51 | mlir::mesh::MeshOp mesh; |
| 52 | ShapedType shardType; |
| 53 | if (resType.getRank() > 0) { |
| 54 | mesh = mesh::getMesh(op, meshSymbol: resultShardings[0].getMeshAttr(), symbolTableCollection&: symbolTable); |
| 55 | shardType = |
| 56 | cast<ShapedType>(Val: mesh::shardType(type: resType, mesh, sharding: resultShardings[0])); |
| 57 | } else { |
| 58 | shardType = resType; |
| 59 | } |
| 60 | Operation *newOp = nullptr; |
| 61 | // if the sharding introduces a new dynamic dimension, we take it from |
| 62 | // the dynamic sharding info. For now bail out if it's not |
| 63 | // provided. |
| 64 | if (!shardType.hasStaticShape()) { |
| 65 | assert(op->getResult(0).hasOneUse()); |
| 66 | SmallVector<Value> newOperands; |
| 67 | auto oldType = cast<ShapedType>(Val&: resType); |
| 68 | assert(oldType.getRank() == shardType.getRank()); |
| 69 | int currOldOprndNum = -1; |
| 70 | mesh::ShardShapeOp shapeForDevice; |
| 71 | ValueRange device; |
| 72 | Operation *newSharding = nullptr; |
| 73 | for (auto i = 0; i < oldType.getRank(); ++i) { |
| 74 | if (!oldType.isDynamicDim(idx: i) && shardType.isDynamicDim(idx: i)) { |
| 75 | if (!newSharding) { |
| 76 | newSharding = |
| 77 | builder.create<ShardingOp>(location: op->getLoc(), args: resultShardings[0]); |
| 78 | device = |
| 79 | builder.create<mesh::ProcessMultiIndexOp>(location: op->getLoc(), args&: mesh) |
| 80 | .getResults(); |
| 81 | shapeForDevice = builder.create<mesh::ShardShapeOp>( |
| 82 | location: op->getLoc(), args: oldType.getShape(), args&: spmdizedOperands, |
| 83 | args: newSharding->getResult(idx: 0), args&: device); |
| 84 | } |
| 85 | newOperands.emplace_back(Args: shapeForDevice.getResult()[i]); |
| 86 | } else if (oldType.isDynamicDim(idx: i)) { |
| 87 | assert(shardType.isDynamicDim(i)); |
| 88 | newOperands.emplace_back(Args: spmdizedOperands[++currOldOprndNum]); |
| 89 | } |
| 90 | } |
| 91 | newOp = builder.create<OpTy>(op->getLoc(), shardType, newOperands); |
| 92 | spmdizationMap.map(from: op->getResult(idx: 0), to: newOp->getResult(idx: 0)); |
| 93 | } else { |
| 94 | // `clone` will populate the mapping of old to new results. |
| 95 | newOp = builder.clone(op&: *op, mapper&: spmdizationMap); |
| 96 | } |
| 97 | newOp->getResult(idx: 0).setType(shardType); |
| 98 | |
| 99 | return success(); |
| 100 | } |
| 101 | }; |
| 102 | } // namespace |
| 103 | |
| 104 | void mlir::tensor::registerShardingInterfaceExternalModels( |
| 105 | DialectRegistry ®istry) { |
| 106 | |
| 107 | registry.addExtension(extensionFn: +[](MLIRContext *ctx, TensorDialect *dialect) { |
| 108 | EmptyOp::template attachInterface<CreatorOpShardingInterface<EmptyOp>>( |
| 109 | context&: *ctx); |
| 110 | SplatOp::template attachInterface<CreatorOpShardingInterface<SplatOp>>( |
| 111 | context&: *ctx); |
| 112 | }); |
| 113 | } |
| 114 | |