[mlir][sparse] codegen for sparse alloc

Reviewed By: Peiming

Differential Revision: https://reviews.llvm.org/D133241

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorCodegen.cpp

@@ -34,6 +34,16 @@ namespace {
 // Helper methods.
 //===----------------------------------------------------------------------===//
 
+/// Reorders stored dimension to original dimension.
+static unsigned toOrig(const SparseTensorEncodingAttr &enc, unsigned i) {
+  auto order = enc.getDimOrdering();
+  if (order) {
+    assert(order.isPermutation());
+    return order.getDimPosition(i);
+  }
+  return i;
+}
+
 /// Reorders original dimension to stored dimension.
 static unsigned toStored(const SparseTensorEncodingAttr &enc, unsigned i) {
   auto order = enc.getDimOrdering();
@@ -87,7 +97,7 @@ static Optional<Type> convertSparseTensorType(Type type) {
     // tensor type.
     switch (enc.getDimLevelType()[r]) {
     case SparseTensorEncodingAttr::DimLevelType::Dense:
-      break;
+      break; // no fields
     case SparseTensorEncodingAttr::DimLevelType::Compressed:
     case SparseTensorEncodingAttr::DimLevelType::CompressedNu:
     case SparseTensorEncodingAttr::DimLevelType::CompressedNo:
@@ -111,7 +121,7 @@ static Optional<Type> convertSparseTensorType(Type type) {
   return TupleType::get(context, fields);
 }
 
-// Returns field index for pointers (d), indices (d) for set field.
+// Returns field index of sparse tensor type for pointers/indices, when set.
 static unsigned getFieldIndex(Type type, unsigned ptrDim, unsigned idxDim) {
   auto enc = getSparseTensorEncoding(type);
   assert(enc);
@@ -161,6 +171,94 @@ static Value createTupleGet(OpBuilder &builder, Location loc, Value tuple,
                                       builder.getIntegerAttr(indexType, field));
 }
 
+/// Creates tuple.
+static Value createTupleMake(OpBuilder &builder, Location loc, Type type,
+                             ValueRange values) {
+  return builder.create<StorageNewOp>(loc, type, values);
+}
+
+/// Create allocation operation.
+static Value createAllocation(OpBuilder &builder, Location loc, Type type,
+                              Value sz) {
+  auto memType = MemRefType::get({ShapedType::kDynamicSize}, type);
+  return builder.create<memref::AllocOp>(loc, memType, sz);
+}
+
+/// Creates allocation tuple for sparse tensor type.
+///
+/// TODO: for efficiency, we will need heuristis to make educated guesses
+///       on the required final sizes; also, we will need an improved
+///       memory allocation scheme with capacity and reallocation
+///
+static Value createAllocTuple(OpBuilder &builder, Location loc, Type type,
+                              ValueRange dynSizes) {
+  auto enc = getSparseTensorEncoding(type);
+  assert(enc);
+  // Construct the basic types.
+  unsigned idxWidth = enc.getIndexBitWidth();
+  unsigned ptrWidth = enc.getPointerBitWidth();
+  RankedTensorType rType = type.cast<RankedTensorType>();
+  Type indexType = builder.getIndexType();
+  Type idxType = idxWidth ? builder.getIntegerType(idxWidth) : indexType;
+  Type ptrType = ptrWidth ? builder.getIntegerType(ptrWidth) : indexType;
+  Type eltType = rType.getElementType();
+  // Build the allocation tuple, using heuristics for pre-allocation.
+  auto shape = rType.getShape();
+  unsigned rank = shape.size();
+  SmallVector<Value, 8> fields;
+  bool allDense = true;
+  Value one = constantIndex(builder, loc, 1);
+  Value linear = one;
+  Value heuristic = one; // FIX, see TODO above
+  // Build original sizes.
+  SmallVector<Value, 8> sizes;
+  for (unsigned r = 0, o = 0; r < rank; r++) {
+    if (ShapedType::isDynamic(shape[r]))
+      sizes.push_back(dynSizes[o++]);
+    else
+      sizes.push_back(constantIndex(builder, loc, shape[r]));
+  }
+  // The dimSizes array.
+  Value dimSizes =
+      builder.create<memref::AllocOp>(loc, MemRefType::get({rank}, indexType));
+  fields.push_back(dimSizes);
+  // Per-dimension storage.
+  for (unsigned r = 0; r < rank; r++) {
+    // Get the original dimension (ro) for the current stored dimension.
+    unsigned ro = toOrig(enc, r);
+    builder.create<memref::StoreOp>(loc, sizes[ro], dimSizes,
+                                    constantIndex(builder, loc, r));
+    linear = builder.create<arith::MulIOp>(loc, linear, sizes[ro]);
+    // Allocate fiels.
+    switch (enc.getDimLevelType()[r]) {
+    case SparseTensorEncodingAttr::DimLevelType::Dense:
+      break; // no fields
+    case SparseTensorEncodingAttr::DimLevelType::Compressed:
+    case SparseTensorEncodingAttr::DimLevelType::CompressedNu:
+    case SparseTensorEncodingAttr::DimLevelType::CompressedNo:
+    case SparseTensorEncodingAttr::DimLevelType::CompressedNuNo:
+      fields.push_back(createAllocation(builder, loc, ptrType, heuristic));
+      fields.push_back(createAllocation(builder, loc, idxType, heuristic));
+      allDense = false;
+      break;
+    case SparseTensorEncodingAttr::DimLevelType::Singleton:
+    case SparseTensorEncodingAttr::DimLevelType::SingletonNu:
+    case SparseTensorEncodingAttr::DimLevelType::SingletonNo:
+    case SparseTensorEncodingAttr::DimLevelType::SingletonNuNo:
+      fields.push_back(createAllocation(builder, loc, idxType, heuristic));
+      allDense = false;
+      break;
+    }
+  }
+  // The values array. For all-dense, the full length is required.
+  // In all other case, we resort to the heuristical initial value.
+  Value valuesSz = allDense ? linear : heuristic;
+  fields.push_back(createAllocation(builder, loc, eltType, valuesSz));
+  // Construct tuple allocation.
+  Type tupleType = *convertSparseTensorType(type);
+  return createTupleMake(builder, loc, tupleType, fields);
+}
+
 /// Returns integral constant, if defined.
 static Optional<int64_t> getConstantInt(Value val) {
   if (auto constantOp = val.getDefiningOp<arith::ConstantOp>())
@@ -233,6 +331,28 @@ public:
   }
 };
 
+/// Sparse codgen rule for the alloc operator.
+class SparseTensorAllocConverter
+    : public OpConversionPattern<bufferization::AllocTensorOp> {
+public:
+  using OpConversionPattern::OpConversionPattern;
+  LogicalResult
+  matchAndRewrite(bufferization::AllocTensorOp op, OpAdaptor adaptor,
+                  ConversionPatternRewriter &rewriter) const override {
+    RankedTensorType resType = op.getType();
+    auto enc = getSparseTensorEncoding(resType);
+    if (!enc)
+      return failure();
+    if (op.getCopy())
+      return rewriter.notifyMatchFailure(op, "tensor copy not implemented");
+    // Construct allocation tuple.
+    Value tuple = createAllocTuple(rewriter, op->getLoc(), resType,
+                                   adaptor.getOperands());
+    rewriter.replaceOp(op, tuple);
+    return success();
+  }
+};
+
 /// Sparse codegen rule for the dealloc operator.
 class SparseTensorDeallocConverter
     : public OpConversionPattern<bufferization::DeallocTensorOp> {
@@ -311,6 +431,22 @@ public:
   }
 };
 
+/// Sparse codegen rule for tensor rematerialization.
+class SparseTensorLoadConverter : public OpConversionPattern<LoadOp> {
+public:
+  using OpConversionPattern::OpConversionPattern;
+  LogicalResult
+  matchAndRewrite(LoadOp op, OpAdaptor adaptor,
+                  ConversionPatternRewriter &rewriter) const override {
+    if (op.getHasInserts()) {
+      // Finalize any pending insertions.
+      // TODO: implement
+    }
+    rewriter.replaceOp(op, adaptor.getOperands());
+    return success();
+  }
+};
+
 } // namespace
 
 //===----------------------------------------------------------------------===//
@@ -331,7 +467,8 @@ mlir::SparseTensorTypeToBufferConverter::SparseTensorTypeToBufferConverter() {
 void mlir::populateSparseTensorCodegenPatterns(TypeConverter &typeConverter,
                                                RewritePatternSet &patterns) {
   patterns.add<SparseReturnConverter, SparseDimOpConverter, SparseCastConverter,
-               SparseTensorDeallocConverter, SparseToPointersConverter,
+               SparseTensorAllocConverter, SparseTensorDeallocConverter,
-               SparseToIndicesConverter, SparseToValuesConverter>(
+               SparseToPointersConverter, SparseToIndicesConverter,
+               SparseToValuesConverter, SparseTensorLoadConverter>(
       typeConverter, patterns.getContext());
 }

mlir/lib/Dialect/SparseTensor/Transforms/SparseTensorPasses.cpp

@@ -156,7 +156,7 @@ struct SparseTensorCodegenPass
     ConversionTarget target(*ctx);
     // Almost everything in the sparse dialect must go!
     target.addIllegalDialect<SparseTensorDialect>();
-    target.addLegalOp<StorageGetOp, StorageSetOp>();
+    target.addLegalOp<StorageGetOp, StorageSetOp, StorageNewOp>();
     // All dynamic rules below accept new function, call, return, and various
     // tensor and bufferization operations as legal output of the rewriting
     // provided that all sparse tensor types have been fully rewritten.
@@ -169,6 +169,10 @@ struct SparseTensorCodegenPass
     target.addDynamicallyLegalOp<func::ReturnOp>([&](func::ReturnOp op) {
       return converter.isLegal(op.getOperandTypes());
     });
+    target.addDynamicallyLegalOp<bufferization::AllocTensorOp>(
+        [&](bufferization::AllocTensorOp op) {
+          return converter.isLegal(op.getType());
+        });
     target.addDynamicallyLegalOp<bufferization::DeallocTensorOp>(
         [&](bufferization::DeallocTensorOp op) {
           return converter.isLegal(op.getTensor().getType());

mlir/test/Dialect/SparseTensor/codegen.mlir

@@ -1,6 +1,6 @@
 // RUN: mlir-opt %s --sparse-tensor-codegen  --canonicalize --cse | FileCheck %s --check-prefix=CHECK-CODEGEN
-// RUN: mlir-opt %s --sparse-tensor-codegen --sparse-tensor-storage-expansion --canonicalize --cse | FileCheck %s --check-prefix=CHECK-STORAGE
+// FIXME:
-
+// R_U_N: mlir-opt %s --sparse-tensor-codegen --sparse-tensor-storage-expansion --canonicalize --cse | FileCheck %s --check-prefix=CHECK-STORAGE
 
 #SparseVector = #sparse_tensor.encoding<{
   dimLevelType = [ "compressed" ],
@@ -26,6 +26,11 @@
   pointerBitWidth = 32
 }>
 
+#CSC = #sparse_tensor.encoding<{
+  dimLevelType = [ "dense", "compressed" ],
+  dimOrdering = affine_map<(i, j) -> (j, i)>
+}>
+
 #DCSR = #sparse_tensor.encoding<{
   dimLevelType = [ "compressed", "compressed" ],
   indexBitWidth = 64,
@@ -45,7 +50,7 @@
 //  CHECK-STORAGE-SAME: %[[A0:.*0]]: memref<1xindex>,
 //  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xi32>,
 //  CHECK-STORAGE-SAME: %[[A2:.*2]]: memref<?xi64>,
-//  CHECK-STORAGE-SAME: %[[A3:.*3]]: memref<?xf64>) 
+//  CHECK-STORAGE-SAME: %[[A3:.*3]]: memref<?xf64>)
 //       CHECK-STORAGE: return %[[A0]], %[[A1]], %[[A2]], %[[A3]] : memref<1xindex>, memref<?xi32>, memref<?xi64>, memref<?xf64>
 func.func @sparse_nop(%arg0: tensor<?xf64, #SparseVector>) -> tensor<?xf64, #SparseVector> {
   return %arg0 : tensor<?xf64, #SparseVector>
@@ -59,7 +64,7 @@ func.func @sparse_nop(%arg0: tensor<?xf64, #SparseVector>) -> tensor<?xf64, #Spa
 //  CHECK-STORAGE-SAME: %[[A0:.*0]]: memref<1xindex>,
 //  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xi32>,
 //  CHECK-STORAGE-SAME: %[[A2:.*2]]: memref<?xi64>,
-//  CHECK-STORAGE-SAME: %[[A3:.*3]]: memref<?xf32>) 
+//  CHECK-STORAGE-SAME: %[[A3:.*3]]: memref<?xf32>)
 //       CHECK-STORAGE: return %[[A0]], %[[A1]], %[[A2]], %[[A3]] : memref<1xindex>, memref<?xi32>, memref<?xi64>, memref<?xf32>
 func.func @sparse_nop_cast(%arg0: tensor<64xf32, #SparseVector>) -> tensor<?xf32, #SparseVector> {
   %0 = tensor.cast %arg0 : tensor<64xf32, #SparseVector> to tensor<?xf32, #SparseVector>
@@ -72,7 +77,7 @@ func.func @sparse_nop_cast(%arg0: tensor<64xf32, #SparseVector>) -> tensor<?xf32
 //
 // CHECK-STORAGE-LABEL: func @sparse_nop_cast_3d(
 //  CHECK-STORAGE-SAME: %[[A0:.*0]]: memref<3xindex>,
-//  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xf32>) 
+//  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xf32>)
 //       CHECK-STORAGE: return %[[A0]], %[[A1]] : memref<3xindex>, memref<?xf32>
 func.func @sparse_nop_cast_3d(%arg0: tensor<10x20x30xf32, #Dense3D>) -> tensor<?x?x?xf32, #Dense3D> {
   %0 = tensor.cast %arg0 : tensor<10x20x30xf32, #Dense3D> to tensor<?x?x?xf32, #Dense3D>
@@ -142,7 +147,7 @@ func.func @sparse_dcsr(%arg0: tensor<?x?xf64, #DCSR>) {
 //
 // CHECK-STORAGE-LABEL: func @sparse_dense_3d(
 //  CHECK-STORAGE-SAME: %[[A0:.*0]]: memref<3xindex>,
-//  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xf64>) 
+//  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xf64>)
 //       CHECK-STORAGE: %[[C:.*]] = arith.constant 20 : index
 //       CHECK-STORAGE: return %[[C]] : index
 func.func @sparse_dense_3d(%arg0: tensor<10x20x30xf64, #Dense3D>) -> index {
@@ -165,7 +170,7 @@ func.func @sparse_dense_3d(%arg0: tensor<10x20x30xf64, #Dense3D>) -> index {
 //
 // CHECK-STORAGE-LABEL: func @sparse_dense_3d_dyn(
 //  CHECK-STORAGE-SAME: %[[A0:.*0]]: memref<3xindex>,
-//  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xf64>) 
+//  CHECK-STORAGE-SAME: %[[A1:.*1]]: memref<?xf64>)
 //       CHECK-STORAGE: %[[C:.*]] = arith.constant 2 : index
 //       CHECK-STORAGE: %[[L:.*]] = memref.load %[[A0]][%[[C]]] : memref<3xindex>
 //       CHECK-STORAGE: return %[[L]] : index
@@ -186,7 +191,7 @@ func.func @sparse_dense_3d_dyn(%arg0: tensor<?x?x?xf64, #Dense3D>) -> index {
 //  CHECK-STORAGE-SAME: %[[A2:.*2]]: memref<?xi64>,
 //  CHECK-STORAGE-SAME: %[[A3:.*3]]: memref<?xi32>,
 //  CHECK-STORAGE-SAME: %[[A4:.*4]]: memref<?xi64>,
-//  CHECK-STORAGE-SAME: %[[A5:.*5]]: memref<?xf64>) 
+//  CHECK-STORAGE-SAME: %[[A5:.*5]]: memref<?xf64>)
 //       CHECK-STORAGE: return %[[A3]] : memref<?xi32>
 func.func @sparse_pointers_dcsr(%arg0: tensor<?x?xf64, #DCSR>) -> memref<?xi32> {
   %c = arith.constant 1 : index
@@ -205,7 +210,7 @@ func.func @sparse_pointers_dcsr(%arg0: tensor<?x?xf64, #DCSR>) -> memref<?xi32>
 //  CHECK-STORAGE-SAME: %[[A2:.*2]]: memref<?xi64>,
 //  CHECK-STORAGE-SAME: %[[A3:.*3]]: memref<?xi32>,
 //  CHECK-STORAGE-SAME: %[[A4:.*4]]: memref<?xi64>,
-//  CHECK-STORAGE-SAME: %[[A5:.*5]]: memref<?xf64>) 
+//  CHECK-STORAGE-SAME: %[[A5:.*5]]: memref<?xf64>)
 //       CHECK-STORAGE: return %[[A4]] : memref<?xi64>
 func.func @sparse_indices_dcsr(%arg0: tensor<?x?xf64, #DCSR>) -> memref<?xi64> {
   %c = arith.constant 1 : index
@@ -224,7 +229,7 @@ func.func @sparse_indices_dcsr(%arg0: tensor<?x?xf64, #DCSR>) -> memref<?xi64> {
 //  CHECK-STORAGE-SAME: %[[A2:.*2]]: memref<?xi64>,
 //  CHECK-STORAGE-SAME: %[[A3:.*3]]: memref<?xi32>,
 //  CHECK-STORAGE-SAME: %[[A4:.*4]]: memref<?xi64>,
-//  CHECK-STORAGE-SAME: %[[A5:.*5]]: memref<?xf64>) 
+//  CHECK-STORAGE-SAME: %[[A5:.*5]]: memref<?xf64>)
 //       CHECK-STORAGE: return %[[A5]] : memref<?xf64>
 func.func @sparse_values_dcsr(%arg0: tensor<?x?xf64, #DCSR>) -> memref<?xf64> {
   %0 = sparse_tensor.values %arg0 : tensor<?x?xf64, #DCSR> to memref<?xf64>
@@ -257,3 +262,46 @@ func.func @sparse_dealloc_csr(%arg0: tensor<?x?xf64, #CSR>) {
   bufferization.dealloc_tensor %arg0 : tensor<?x?xf64, #CSR>
   return
 }
+
+// CHECK-CODEGEN-LABEL: func @sparse_alloc_csc(
+//  CHECK-CODEGEN-SAME: %[[A:.*]]: index)
+//   CHECK-CODEGEN-DAG: %[[C0:.*]] = arith.constant 0 : index
+//   CHECK-CODEGEN-DAG: %[[C1:.*]] = arith.constant 1 : index
+//   CHECK-CODEGEN-DAG: %[[C10:.*]] = arith.constant 10 : index
+//      CHECK-CODEGEN:  %[[T0:.*]] = memref.alloc() : memref<2xindex>
+//      CHECK-CODEGEN:  memref.store %[[A]], %[[T0]][%[[C0]]] : memref<2xindex>
+//      CHECK-CODEGEN:  memref.store %[[C10]], %[[T0]][%[[C1]]] : memref<2xindex>
+//      CHECK-CODEGEN:  %[[T1:.*]] = memref.alloc() : memref<1xindex>
+//      CHECK-CODEGEN:  %[[T2:.*]] = memref.cast %[[T1]] : memref<1xindex> to memref<?xindex>
+//      CHECK-CODEGEN:  %[[T3:.*]] = memref.alloc() : memref<1xindex>
+//      CHECK-CODEGEN:  %[[T4:.*]] = memref.cast %[[T3]] : memref<1xindex> to memref<?xindex>
+//      CHECK-CODEGEN:  %[[T5:.*]] = memref.alloc() : memref<1xf64>
+//      CHECK-CODEGEN:  %[[T6:.*]] = memref.cast %[[T5]] : memref<1xf64> to memref<?xf64>
+//      CHECK-CODEGEN:  %[[T:.*]] = sparse_tensor.storage(%[[T0]], %[[T2]], %[[T4]], %[[T6]])
+//      CHECK-CODEGEN:  return %[[T]] : tuple<memref<2xindex>, memref<?xindex>, memref<?xindex>, memref<?xf64>>
+func.func @sparse_alloc_csc(%arg0: index) -> tensor<10x?xf64, #CSC> {
+  %0 = bufferization.alloc_tensor(%arg0) : tensor<10x?xf64, #CSC>
+  %1 = sparse_tensor.load %0 : tensor<10x?xf64, #CSC>
+  return %1 : tensor<10x?xf64, #CSC>
+}
+
+// CHECK-CODEGEN-LABEL: func @sparse_alloc_3d() -> tuple<memref<3xindex>, memref<?xf64>>
+//   CHECK-CODEGEN-DAG: %[[C0:.*]] = arith.constant 0 : index
+//   CHECK-CODEGEN-DAG: %[[C1:.*]] = arith.constant 1 : index
+//   CHECK-CODEGEN-DAG: %[[C2:.*]] = arith.constant 2 : index
+//   CHECK-CODEGEN-DAG: %[[C10:.*]] = arith.constant 10 : index
+//   CHECK-CODEGEN-DAG: %[[C20:.*]] = arith.constant 20 : index
+//   CHECK-CODEGEN-DAG: %[[C30:.*]] = arith.constant 30 : index
+//       CHECK-CODEGEN: %[[A0:.*]] = memref.alloc() : memref<3xindex>
+//       CHECK-CODEGEN: memref.store %[[C30]], %[[A0]][%[[C0]]] : memref<3xindex>
+//       CHECK-CODEGEN: memref.store %[[C10]], %[[A0]][%[[C1]]] : memref<3xindex>
+//       CHECK-CODEGEN: memref.store %[[C20]], %[[A0]][%[[C2]]] : memref<3xindex>
+//       CHECK-CODEGEN: %[[A:.*]] = memref.alloc() : memref<6000xf64>
+//       CHECK-CODEGEN: %[[A1:.*]] = memref.cast %[[A]] : memref<6000xf64> to memref<?xf64>
+//       CHECK-CODEGEN: %[[T:.*]] = sparse_tensor.storage(%[[A0]], %[[A1]])
+//       CHECK-CODEGEN: return %[[T]] : tuple<memref<3xindex>, memref<?xf64>>
+func.func @sparse_alloc_3d() -> tensor<10x20x30xf64, #Dense3D> {
+  %0 = bufferization.alloc_tensor() : tensor<10x20x30xf64, #Dense3D>
+  %1 = sparse_tensor.load %0 : tensor<10x20x30xf64, #Dense3D>
+  return %1 : tensor<10x20x30xf64, #Dense3D>
+}