1// RUN: mlir-opt %s --sparse-reinterpret-map -sparsification | FileCheck %s 2 3#map = affine_map<(d0, d1, d2) -> (d0, d1, d2)> 4#BCSR = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : batch, d1 : dense, d2 : compressed)}> 5 6// CHECK-LABEL: func.func @main( 7// CHECK-SAME: %[[VAL_0:.*]]: tensor<8x4x2xf32, #sparse{{[0-9]*}}>) -> tensor<8x4x2xf32> { 8// CHECK-DAG: %[[VAL_1:.*]] = arith.constant 1 : index 9// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 0 : index 10// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 0.000000e+00 : f32 11// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 4 : index 12// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 13// CHECK-DAG: %[[VAL_6:.*]] = tensor.empty() : tensor<8x4x2xf32> 14// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<8x4x2xf32, #sparse{{[0-9]*}}> to memref<8x?xindex> 15// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<8x4x2xf32, #sparse{{[0-9]*}}> to memref<8x?xindex> 16// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<8x4x2xf32, #sparse{{[0-9]*}}> to memref<8x?xf32> 17// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_6]] : tensor<8x4x2xf32> 18// CHECK-DAG: linalg.fill ins(%[[VAL_3]] : f32) outs(%[[VAL_10]] : memref<8x4x2xf32>) 19// CHECK: scf.for %[[VAL_11:.*]] = %[[VAL_2]] to %[[VAL_5]] step %[[VAL_1]] { 20// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_2]] to %[[VAL_4]] step %[[VAL_1]] { 21// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_11]], %[[VAL_12]]] : memref<8x?xindex> 22// CHECK: %[[VAL_14:.*]] = arith.addi %[[VAL_12]], %[[VAL_1]] : index 23// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_11]], %[[VAL_14]]] : memref<8x?xindex> 24// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_13]] to %[[VAL_15]] step %[[VAL_1]] { 25// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_11]], %[[VAL_16]]] : memref<8x?xindex> 26// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_11]], %[[VAL_16]]] : memref<8x?xf32> 27// CHECK: %[[VAL_19:.*]] = arith.negf %[[VAL_18]] : f32 28// CHECK: memref.store %[[VAL_19]], %[[VAL_10]]{{\[}}%[[VAL_11]], %[[VAL_12]], %[[VAL_17]]] : memref<8x4x2xf32> 29// CHECK: } 30// CHECK: } 31// CHECK: } 32// CHECK: %[[VAL_20:.*]] = bufferization.to_tensor %[[VAL_10]] : memref<8x4x2xf32> 33// CHECK: return %[[VAL_20]] : tensor<8x4x2xf32> 34// CHECK: } 35func.func @main(%arg0: tensor<8x4x2xf32, #BCSR>) -> tensor<8x4x2xf32> { 36 %0 = tensor.empty() : tensor<8x4x2xf32> 37 %1 = linalg.generic { 38 indexing_maps = [#map, #map], 39 iterator_types = ["parallel", "parallel", "parallel"] 40 } 41 ins(%arg0 : tensor<8x4x2xf32, #BCSR>) 42 outs(%0 : tensor<8x4x2xf32>) { 43 ^bb0(%in: f32, %out: f32): 44 %2 = arith.negf %in : f32 45 linalg.yield %2 : f32 46 } -> tensor<8x4x2xf32> 47 return %1 : tensor<8x4x2xf32> 48} 49