[mlir][sparse] fix bug in permuting data structure

[bixia]

mlir/test/Dialect/SparseTensor/sparse_perm.mlir

45

Shall we name this "sparse_static_dimensions" and the next test "sparse_dynamic_dimensions" or something similar?

[aartbik]

mlir/test/Dialect/SparseTensor/sparse_perm.mlir

45

Renamed.

[aartbik]

renamed test methods

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

for (unsigned d = 0, rank = map.getNumResults(); d < rank; d++) {
             dynShape, genIntType(rewriter, enc.getIndexBitWidth()));
         Value dim = rewriter.create<ConstantIndexOp>(loc, d);
         // Generate sparse primitives to obtains pointer and indices.
         codegen.pointers[tensor][idx] =
             rewriter.create<ToPointersOp>(loc, ptrTp, t->get(), dim);
         codegen.indices[tensor][idx] =
             rewriter.create<ToIndicesOp>(loc, indTp, t->get(), dim);
       }
-      // Find lower and upper bound in current dimension.
+      // Find lower and upper bound in current dimension. Note that a
+      // permuted encoding queries static type dimensions accordingly,
+      // but queries dynamic type dimensions in the generated order.
       Value up;
-      if (shape[d] == MemRefType::kDynamicSize) {
+      unsigned p = perm(enc, d);
+      if (shape[p] == MemRefType::kDynamicSize) {
         up = rewriter.create<tensor::DimOp>(loc, t->get(), d);
         args.push_back(up);
       } else {
-        up = rewriter.create<ConstantIndexOp>(loc, shape[d]);
+        up = rewriter.create<ConstantIndexOp>(loc, shape[p]);
       }
+      assert(codegen.highs[tensor][idx] == nullptr);
       codegen.sizes[idx] = codegen.highs[tensor][idx] = up;
     }
     // Perform the required bufferization. Dense inputs materialize
     // from the input tensors. Dense outputs need special handling.
     // Sparse inputs use sparse primitives to obtain the values.
     // We also accept in-place all-dense annotated "sparse" outputs.
     Type elementType = getElementTypeOrSelf(t->get().getType());
     if (!enc) {

mlir/test/Dialect/SparseTensor/sparse_perm.mlir

+// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py
+// RUN: mlir-opt %s -sparsification | FileCheck %s
+
+#X = #sparse_tensor.encoding<{
+ dimLevelType = [ "dense", "dense", "dense" ],
+ dimOrdering = affine_map<(i,j,k) -> (k,i,j)>
+}>
+
+#trait = {
+  indexing_maps = [
+    affine_map<(i,j,k) -> (k,i,j)>,  // A (in)
+    affine_map<(i,j,k) -> (i,j,k)>   // X (out)
+  ],
+  iterator_types = ["parallel", "parallel", "parallel"]
+}
+
+// CHECK-LABEL:   builtin.func @sparse_static_dims(
+// CHECK-SAME:                          %[[VAL_0:.*]]: tensor<10x20x30xf32, #sparse_tensor.encoding<{{{.*}}}>>,
+// CHECK-SAME:                          %[[VAL_1:.*]]: tensor<20x30x10xf32>) -> tensor<20x30x10xf32> {
+// CHECK:           %[[VAL_2:.*]] = constant 20 : index
+// CHECK:           %[[VAL_3:.*]] = constant 30 : index
+// CHECK:           %[[VAL_4:.*]] = constant 10 : index
+// CHECK:           %[[VAL_5:.*]] = constant 0 : index
+// CHECK:           %[[VAL_6:.*]] = constant 1 : index
+// CHECK:           %[[VAL_7:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10x20x30xf32, #sparse_tensor.encoding<{{{.*}}}>>
+// CHECK:           %[[VAL_8:.*]] = memref.buffer_cast %[[VAL_1]] : memref<20x30x10xf32>
+// CHECK:           %[[VAL_9:.*]] = memref.alloc() : memref<20x30x10xf32>
+// CHECK:           memref.copy %[[VAL_8]], %[[VAL_9]] : memref<20x30x10xf32> to memref<20x30x10xf32>
+// CHECK:           scf.for %[[VAL_10:.*]] = %[[VAL_5]] to %[[VAL_3]] step %[[VAL_6]] {
+// CHECK:             scf.for %[[VAL_11:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] {
+// CHECK:               %[[VAL_12:.*]] = muli %[[VAL_10]], %[[VAL_4]] : index
+// CHECK:               %[[VAL_13:.*]] = addi %[[VAL_12]], %[[VAL_11]] : index
+// CHECK:               scf.for %[[VAL_14:.*]] = %[[VAL_5]] to %[[VAL_2]] step %[[VAL_6]] {
+// CHECK:                 %[[VAL_15:.*]] = muli %[[VAL_13]], %[[VAL_2]] : index
+// CHECK:                 %[[VAL_16:.*]] = addi %[[VAL_15]], %[[VAL_14]] : index
+// CHECK:                 %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref<?xf32>
+// CHECK:                 memref.store %[[VAL_17]], %[[VAL_9]]{{\[}}%[[VAL_14]], %[[VAL_10]], %[[VAL_11]]] : memref<20x30x10xf32>
+// CHECK:               }
+// CHECK:             }
+// CHECK:           }
+// CHECK:           %[[VAL_18:.*]] = memref.tensor_load %[[VAL_9]] : memref<20x30x10xf32>
+// CHECK:           return %[[VAL_18]] : tensor<20x30x10xf32>
+// CHECK:         }
+func @sparse_static_dims(%arga: tensor<10x20x30xf32, #X>,
+                         %argx: tensor<20x30x10xf32>) -> tensor<20x30x10xf32> {
+  %0 = linalg.generic #trait
+    ins(%arga: tensor<10x20x30xf32, #X>)
+    outs(%argx: tensor<20x30x10xf32>) {
+      ^bb(%a : f32, %x: f32):
+        linalg.yield %a : f32
+  } -> tensor<20x30x10xf32>
+  return %0 : tensor<20x30x10xf32>
+}
+
+// CHECK-LABEL:   builtin.func @sparse_dynamic_dims(
+// CHECK-SAME:                          %[[VAL_0:.*]]: tensor<?x?x?xf32, #sparse_tensor.encoding<{{{.*}}}>>,
+// CHECK-SAME:                          %[[VAL_1:.*]]: tensor<?x?x?xf32>) -> tensor<?x?x?xf32> {
+// CHECK:           %[[VAL_2:.*]] = constant 2 : index
+// CHECK:           %[[VAL_3:.*]] = constant 0 : index
+// CHECK:           %[[VAL_4:.*]] = constant 1 : index
+// CHECK:           %[[VAL_5:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<?x?x?xf32, #sparse_tensor.encoding<{{{.*}}}>>
+// CHECK:           %[[VAL_6:.*]] = tensor.dim %[[VAL_1]], %[[VAL_3]] : tensor<?x?x?xf32>
+// CHECK:           %[[VAL_7:.*]] = tensor.dim %[[VAL_1]], %[[VAL_4]] : tensor<?x?x?xf32>
+// CHECK:           %[[VAL_8:.*]] = tensor.dim %[[VAL_1]], %[[VAL_2]] : tensor<?x?x?xf32>
+// CHECK:           %[[VAL_9:.*]] = memref.buffer_cast %[[VAL_1]] : memref<?x?x?xf32>
+// CHECK:           %[[VAL_10:.*]] = memref.alloc(%[[VAL_6]], %[[VAL_7]], %[[VAL_8]]) : memref<?x?x?xf32>
+// CHECK:           memref.copy %[[VAL_9]], %[[VAL_10]] : memref<?x?x?xf32> to memref<?x?x?xf32>
+// CHECK:           scf.for %[[VAL_11:.*]] = %[[VAL_3]] to %[[VAL_7]] step %[[VAL_4]] {
+// CHECK:             scf.for %[[VAL_12:.*]] = %[[VAL_3]] to %[[VAL_8]] step %[[VAL_4]] {
+// CHECK:               %[[VAL_13:.*]] = muli %[[VAL_8]], %[[VAL_11]] : index
+// CHECK:               %[[VAL_14:.*]] = addi %[[VAL_13]], %[[VAL_12]] : index
+// CHECK:               scf.for %[[VAL_15:.*]] = %[[VAL_3]] to %[[VAL_6]] step %[[VAL_4]] {
+// CHECK:                 %[[VAL_16:.*]] = muli %[[VAL_6]], %[[VAL_14]] : index
+// CHECK:                 %[[VAL_17:.*]] = addi %[[VAL_16]], %[[VAL_15]] : index
+// CHECK:                 %[[VAL_18:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_17]]] : memref<?xf32>
+// CHECK:                 memref.store %[[VAL_18]], %[[VAL_10]]{{\[}}%[[VAL_15]], %[[VAL_11]], %[[VAL_12]]] : memref<?x?x?xf32>
+// CHECK:               }
+// CHECK:             }
+// CHECK:           }
+// CHECK:           %[[VAL_19:.*]] = memref.tensor_load %[[VAL_10]] : memref<?x?x?xf32>
+// CHECK:           return %[[VAL_19]] : tensor<?x?x?xf32>
+// CHECK:         }
+func @sparse_dynamic_dims(%arga: tensor<?x?x?xf32, #X>,
+                          %argx: tensor<?x?x?xf32>) -> tensor<?x?x?xf32> {
+  %0 = linalg.generic #trait
+    ins(%arga: tensor<?x?x?xf32, #X>)
+    outs(%argx: tensor<?x?x?xf32>) {
+      ^bb(%a : f32, %x: f32):
+        linalg.yield %a : f32
+  } -> tensor<?x?x?xf32>
+  return %0 : tensor<?x?x?xf32>
+}