1// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py 2// RUN: mlir-opt %s --sparse-reinterpret-map -sparsification | FileCheck %s 3 4#X = #sparse_tensor.encoding<{ 5 map = (d0, d1, d2) -> (d2 : dense, d0 : dense, d1 : dense) 6}> 7 8#trait = { 9 indexing_maps = [ 10 affine_map<(i,j,k) -> (k,i,j)>, // A (in) 11 affine_map<(i,j,k) -> (i,j,k)> // X (out) 12 ], 13 iterator_types = ["parallel", "parallel", "parallel"] 14} 15 16// CHECK-LABEL: func @sparse_static_dims( 17// CHECK-SAME: %[[VAL_0:.*]]: tensor<10x20x30xf32, #sparse{{[0-9]*}}>, 18// CHECK-SAME: %[[VAL_1:.*]]: tensor<20x30x10xf32>) -> tensor<20x30x10xf32> { 19// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 20// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 20 : index 21// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 30 : index 22// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 10 : index 23// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 24// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 25// CHECK: %[[DEMAP:.*]] = sparse_tensor.reinterpret_map %[[VAL_0]] 26// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.values %[[DEMAP]] : tensor<30x10x20xf32, #sparse{{[0-9]*}}> 27// CHECK-DAG: %[[VAL_9:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<20x30x10xf32> to memref<20x30x10xf32> 28// CHECK: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_9]] : memref<20x30x10xf32>) 29// CHECK: scf.for %[[VAL_10:.*]] = %[[VAL_5]] to %[[VAL_3]] step %[[VAL_6]] { 30// CHECK: %[[VAL_12:.*]] = arith.muli %[[VAL_10]], %[[VAL_4]] : index 31// CHECK: scf.for %[[VAL_11:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 32// CHECK: %[[VAL_13:.*]] = arith.addi %[[VAL_11]], %[[VAL_12]] : index 33// CHECK: %[[VAL_15:.*]] = arith.muli %[[VAL_13]], %[[VAL_2]] : index 34// CHECK: scf.for %[[VAL_14:.*]] = %[[VAL_5]] to %[[VAL_2]] step %[[VAL_6]] { 35// CHECK: %[[VAL_16:.*]] = arith.addi %[[VAL_14]], %[[VAL_15]] : index 36// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref<?xf32> 37// CHECK: memref.store %[[VAL_17]], %[[VAL_9]]{{\[}}%[[VAL_14]], %[[VAL_10]], %[[VAL_11]]] : memref<20x30x10xf32> 38// CHECK: } 39// CHECK: } 40// CHECK: } 41// CHECK: %[[VAL_18:.*]] = bufferization.to_tensor %[[VAL_9]] : memref<20x30x10xf32> 42// CHECK: return %[[VAL_18]] : tensor<20x30x10xf32> 43// CHECK: } 44func.func @sparse_static_dims(%arga: tensor<10x20x30xf32, #X>, 45 %argx: tensor<20x30x10xf32>) -> tensor<20x30x10xf32> { 46 %0 = linalg.generic #trait 47 ins(%arga: tensor<10x20x30xf32, #X>) 48 outs(%argx: tensor<20x30x10xf32>) { 49 ^bb(%a : f32, %x: f32): 50 linalg.yield %a : f32 51 } -> tensor<20x30x10xf32> 52 return %0 : tensor<20x30x10xf32> 53} 54 55// CHECK-LABEL: func @sparse_dynamic_dims( 56// CHECK-SAME: %[[VAL_0:.*]]: tensor<?x?x?xf32, #sparse{{[0-9]*}}>, 57// CHECK-SAME: %[[VAL_1:.*]]: tensor<?x?x?xf32>) -> tensor<?x?x?xf32> { 58// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 59// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 2 : index 60// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 0 : index 61// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 1 : index 62// CHECK: %[[DEMAP:.*]] = sparse_tensor.reinterpret_map %[[VAL_0]] 63// CHECK-DAG: %[[VAL_5:.*]] = sparse_tensor.values %[[DEMAP]] : tensor<?x?x?xf32, #sparse{{[0-9]*}}> 64// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.lvl %[[DEMAP]], %[[VAL_2]] : tensor<?x?x?xf32, #sparse{{[0-9]*}}> 65// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.lvl %[[DEMAP]], %[[VAL_3]] : tensor<?x?x?xf32, #sparse{{[0-9]*}}> 66// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.lvl %[[DEMAP]], %[[VAL_4]] : tensor<?x?x?xf32, #sparse{{[0-9]*}}> 67// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<?x?x?xf32> to memref<?x?x?xf32> 68// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_10]] : memref<?x?x?xf32>) 69// CHECK: scf.for %[[VAL_11:.*]] = %[[VAL_3]] to %[[VAL_7]] step %[[VAL_4]] { 70// CHECK: %[[VAL_13:.*]] = arith.muli %[[VAL_11]], %[[VAL_8]] : index 71// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_3]] to %[[VAL_8]] step %[[VAL_4]] { 72// CHECK: %[[VAL_14:.*]] = arith.addi %[[VAL_12]], %[[VAL_13]] : index 73// CHECK: %[[VAL_16:.*]] = arith.muli %[[VAL_14]], %[[VAL_6]] : index 74// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_3]] to %[[VAL_6]] step %[[VAL_4]] { 75// CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_15]], %[[VAL_16]] : index 76// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_17]]] : memref<?xf32> 77// CHECK: memref.store %[[VAL_18]], %[[VAL_10]]{{\[}}%[[VAL_15]], %[[VAL_11]], %[[VAL_12]]] : memref<?x?x?xf32> 78// CHECK: } 79// CHECK: } 80// CHECK: } 81// CHECK: %[[VAL_19:.*]] = bufferization.to_tensor %[[VAL_10]] : memref<?x?x?xf32> 82// CHECK: return %[[VAL_19]] : tensor<?x?x?xf32> 83// CHECK: } 84func.func @sparse_dynamic_dims(%arga: tensor<?x?x?xf32, #X>, 85 %argx: tensor<?x?x?xf32>) -> tensor<?x?x?xf32> { 86 %0 = linalg.generic #trait 87 ins(%arga: tensor<?x?x?xf32, #X>) 88 outs(%argx: tensor<?x?x?xf32>) { 89 ^bb(%a : f32, %x: f32): 90 linalg.yield %a : f32 91 } -> tensor<?x?x?xf32> 92 return %0 : tensor<?x?x?xf32> 93} 94