1// NOTE: Assertions have been autogenerated by utils/generate-test-checks.py 2// RUN: mlir-opt %s -sparsification | FileCheck %s 3 4#CSR = #sparse_tensor.encoding<{ 5 dimLevelType = [ "dense", "compressed" ], 6 dimOrdering = affine_map<(i,j) -> (i,j)> 7}> 8 9#DCSR = #sparse_tensor.encoding<{ 10 dimLevelType = [ "compressed", "compressed" ], 11 dimOrdering = affine_map<(i,j) -> (i,j)> 12}> 13 14#SparseTensor = #sparse_tensor.encoding<{ 15 dimLevelType = [ "compressed", "compressed", "compressed" ] 16}> 17 18#trait_scale_inpl = { 19 indexing_maps = [ 20 affine_map<(i,j) -> (i,j)> // X (out) 21 ], 22 iterator_types = ["parallel", "parallel"], 23 doc = "X(i,j) *= 2 or X(i,j) += X(i,j)" 24} 25 26// CHECK-LABEL: func @sparse_simply_dynamic1( 27// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> { 28// CHECK-DAG: %[[VAL_1:.*]] = arith.constant 2.000000e+00 : f32 29// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 0 : index 30// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index 31// CHECK: %[[VAL_4:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_2]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> to memref<?xindex> 32// CHECK: %[[VAL_6:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> to memref<?xindex> 33// CHECK: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> to memref<?xf32> 34// CHECK: %[[VAL_9:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_2]]] : memref<?xindex> 35// CHECK: %[[VAL_10:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_3]]] : memref<?xindex> 36// CHECK: scf.for %[[VAL_11:.*]] = %[[VAL_9]] to %[[VAL_10]] step %[[VAL_3]] { 37// CHECK: %[[VAL_12:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_11]]] : memref<?xindex> 38// CHECK: %[[VAL_13:.*]] = arith.addi %[[VAL_11]], %[[VAL_3]] : index 39// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_13]]] : memref<?xindex> 40// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_12]] to %[[VAL_14]] step %[[VAL_3]] { 41// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_15]]] : memref<?xf32> 42// CHECK: %[[VAL_17:.*]] = arith.mulf %[[VAL_16]], %[[VAL_1]] : f32 43// CHECK: memref.store %[[VAL_17]], %[[VAL_8]]{{\[}}%[[VAL_15]]] : memref<?xf32> 44// CHECK: } 45// CHECK: } 46// CHECK: %[[VAL_18:.*]] = sparse_tensor.load %[[VAL_0]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 47// CHECK: return %[[VAL_18]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 48// CHECK: } 49func.func @sparse_simply_dynamic1(%argx: tensor<32x16xf32, #DCSR> {linalg.inplaceable = true}) -> tensor<32x16xf32, #DCSR> { 50 %c = arith.constant 2.0 : f32 51 %0 = linalg.generic #trait_scale_inpl 52 outs(%argx: tensor<32x16xf32, #DCSR>) { 53 ^bb(%x: f32): 54 %1 = arith.mulf %x, %c : f32 55 linalg.yield %1 : f32 56 } -> tensor<32x16xf32, #DCSR> 57 return %0 : tensor<32x16xf32, #DCSR> 58} 59 60// CHECK-LABEL: func @sparse_simply_dynamic2( 61// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 62// CHECK-DAG: %[[VAL_1:.*]] = arith.constant 0 : index 63// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 1 : index 64// CHECK: %[[VAL_3:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_1]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 65// CHECK: %[[VAL_4:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_2]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 66// CHECK: %[[VAL_5:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 67// CHECK: %[[VAL_6:.*]] = memref.load %[[VAL_3]]{{\[}}%[[VAL_1]]] : memref<?xindex> 68// CHECK: %[[VAL_7:.*]] = memref.load %[[VAL_3]]{{\[}}%[[VAL_2]]] : memref<?xindex> 69// CHECK: scf.for %[[VAL_8:.*]] = %[[VAL_6]] to %[[VAL_7]] step %[[VAL_2]] { 70// CHECK: %[[VAL_9:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_8]]] : memref<?xindex> 71// CHECK: %[[VAL_10:.*]] = arith.addi %[[VAL_8]], %[[VAL_2]] : index 72// CHECK: %[[VAL_11:.*]] = memref.load %[[VAL_4]]{{\[}}%[[VAL_10]]] : memref<?xindex> 73// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_9]] to %[[VAL_11]] step %[[VAL_2]] { 74// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_12]]] : memref<?xf32> 75// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_12]]] : memref<?xf32> 76// CHECK: %[[VAL_15:.*]] = arith.addf %[[VAL_13]], %[[VAL_14]] : f32 77// CHECK: memref.store %[[VAL_15]], %[[VAL_5]]{{\[}}%[[VAL_12]]] : memref<?xf32> 78// CHECK: } 79// CHECK: } 80// CHECK: %[[VAL_16:.*]] = sparse_tensor.load %[[VAL_0]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 81// CHECK: return %[[VAL_16]] : tensor<32x16xf32, #sparse_tensor.encoding<{{.*}}>> 82// CHECK: } 83func.func @sparse_simply_dynamic2(%argx: tensor<32x16xf32, #DCSR> {linalg.inplaceable = true}) -> tensor<32x16xf32, #DCSR> { 84 %0 = linalg.generic #trait_scale_inpl 85 outs(%argx: tensor<32x16xf32, #DCSR>) { 86 ^bb(%x: f32): 87 %1 = arith.addf %x, %x : f32 88 linalg.yield %1 : f32 89 } -> tensor<32x16xf32, #DCSR> 90 return %0 : tensor<32x16xf32, #DCSR> 91} 92 93#trait_scale = { 94 indexing_maps = [ 95 affine_map<(i,j) -> (i,j)>, // A 96 affine_map<(i,j) -> (i,j)> // X (out) 97 ], 98 iterator_types = ["parallel", "parallel"], 99 doc = "X(i,j) = A(i,j) * 2.0" 100} 101 102// CHECK-LABEL: func @sparse_truly_dynamic( 103// CHECK-SAME: %[[VAL_0:.*]]: tensor<10x20xf32, #sparse_tensor.encoding<{{.*}}>> 104// CHECK-DAG: %[[VAL_1:.*]] = arith.constant 2.000000e+00 : f32 105// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 10 : index 106// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 1 : index 107// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 2 : index 108// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 0 : index 109// CHECK: %[[VAL_7:.*]] = bufferization.alloc_tensor() : tensor<10x20xf32, #sparse_tensor.encoding<{{.*}}>> 110// CHECK: %[[VAL_8:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<10x20xf32, #sparse_tensor.encoding<{{.*}}>> 111// CHECK: %[[VAL_9:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_4]] : tensor<10x20xf32, #sparse_tensor.encoding<{{.*}}>> 112// CHECK: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10x20xf32, #sparse_tensor.encoding<{{.*}}>> 113// CHECK: %[[VAL_11:.*]] = memref.alloca(%[[VAL_5]]) : memref<?xindex> 114// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_6]] to %[[VAL_2]] step %[[VAL_4]] { 115// CHECK: memref.store %[[VAL_12]], %[[VAL_11]]{{\[}}%[[VAL_6]]] : memref<?xindex> 116// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_12]]] : memref<?xindex> 117// CHECK: %[[VAL_14:.*]] = arith.addi %[[VAL_12]], %[[VAL_4]] : index 118// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_14]]] : memref<?xindex> 119// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_13]] to %[[VAL_15]] step %[[VAL_4]] { 120// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_16]]] : memref<?xindex> 121// CHECK: memref.store %[[VAL_17]], %[[VAL_11]]{{\[}}%[[VAL_4]]] : memref<?xindex> 122// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_16]]] : memref<?xf32> 123// CHECK: %[[VAL_19:.*]] = arith.mulf %[[VAL_18]], %[[VAL_1]] : f32 124// CHECK: sparse_tensor.lex_insert %[[VAL_7]], %[[VAL_11]], %[[VAL_19]] : tensor<10x20xf32, #sparse_tensor.encoding<{{.*}}>> 125// CHECK: } 126// CHECK: } 127// CHECK: %[[VAL_20:.*]] = sparse_tensor.load %[[VAL_7]] hasInserts : tensor<10x20xf32, #sparse_tensor.encoding<{{.*}}>> 128// CHECK: return %[[VAL_20]] : tensor<10x20xf32, #sparse_tensor.encoding<{ 129// CHECK: } 130func.func @sparse_truly_dynamic(%arga: tensor<10x20xf32, #CSR>) -> tensor<10x20xf32, #DCSR> { 131 %s = arith.constant 2.0 : f32 132 %xm = bufferization.alloc_tensor() : tensor<10x20xf32, #DCSR> 133 %0 = linalg.generic #trait_scale 134 ins(%arga: tensor<10x20xf32, #CSR>) 135 outs(%xm: tensor<10x20xf32, #DCSR>) { 136 ^bb(%a: f32, %x: f32): 137 %1 = arith.mulf %a, %s : f32 138 linalg.yield %1 : f32 139 } -> tensor<10x20xf32, #DCSR> 140 return %0 : tensor<10x20xf32, #DCSR> 141} 142 143#trait_sumred = { 144 indexing_maps = [ 145 affine_map<(i,j,k) -> (i,j,k)>, // A 146 affine_map<(i,j,k) -> (i,j,k)>, // B 147 affine_map<(i,j,k) -> (i,j)> // X (out) 148 ], 149 iterator_types = ["parallel", "parallel", "reduction"], 150 doc = "X(i,j) = SUM_k A(i,j,k) * B(i,j,k)" 151} 152 153// CHECK-LABEL: func @sumred( 154// CHECK-SAME: %[[VAL_0:.*]]: tensor<?x?x?xi32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>, 155// CHECK-SAME: %[[VAL_1:.*]]: tensor<?x?x?xi32, #sparse_tensor.encoding<{ dimLevelType = [ "compressed", "compressed", "compressed" ], pointerBitWidth = 0, indexBitWidth = 0 }>>) 156// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 0 : index 157// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index 158// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 2 : index 159// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : i32 160// CHECK: %[[VAL_6:.*]] = tensor.dim %[[VAL_0]], %[[VAL_2]] : tensor<?x?x?xi32, #{{.*}}>> 161// CHECK: %[[VAL_7:.*]] = tensor.dim %[[VAL_0]], %[[VAL_3]] : tensor<?x?x?xi32, #{{.*}}>> 162// CHECK: %[[VAL_8:.*]] = bufferization.alloc_tensor(%[[VAL_6]], %[[VAL_7]]) : tensor<?x?xi32, #{{.*}}>> 163// CHECK: %[[VAL_9:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_2]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 164// CHECK: %[[VAL_10:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_2]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 165// CHECK: %[[VAL_11:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 166// CHECK: %[[VAL_12:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 167// CHECK: %[[VAL_13:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_4]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 168// CHECK: %[[VAL_14:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_4]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 169// CHECK: %[[VAL_15:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xi32> 170// CHECK: %[[VAL_16:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_2]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 171// CHECK: %[[VAL_17:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_2]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 172// CHECK: %[[VAL_18:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_3]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 173// CHECK: %[[VAL_19:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_3]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 174// CHECK: %[[VAL_20:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_4]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 175// CHECK: %[[VAL_21:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_4]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xindex> 176// CHECK: %[[VAL_22:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<?x?x?xi32, #{{.*}}>> to memref<?xi32> 177// CHECK: %[[VAL_23:.*]] = memref.alloca(%[[VAL_4]]) : memref<?xindex> 178// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_2]]] : memref<?xindex> 179// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_3]]] : memref<?xindex> 180// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_2]]] : memref<?xindex> 181// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_3]]] : memref<?xindex> 182// CHECK: %[[VAL_28:.*]]:2 = scf.while (%[[VAL_29:.*]] = %[[VAL_24]], %[[VAL_30:.*]] = %[[VAL_26]]) : (index, index) -> (index, index) { 183// CHECK: %[[VAL_31:.*]] = arith.cmpi ult, %[[VAL_29]], %[[VAL_25]] : index 184// CHECK: %[[VAL_32:.*]] = arith.cmpi ult, %[[VAL_30]], %[[VAL_27]] : index 185// CHECK: %[[VAL_33:.*]] = arith.andi %[[VAL_31]], %[[VAL_32]] : i1 186// CHECK: scf.condition(%[[VAL_33]]) %[[VAL_29]], %[[VAL_30]] : index, index 187// CHECK: } do { 188// CHECK: ^bb0(%[[VAL_34:.*]]: index, %[[VAL_35:.*]]: index): 189// CHECK: %[[VAL_36:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_34]]] : memref<?xindex> 190// CHECK: %[[VAL_37:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_35]]] : memref<?xindex> 191// CHECK: %[[VAL_38:.*]] = arith.cmpi ult, %[[VAL_37]], %[[VAL_36]] : index 192// CHECK: %[[VAL_39:.*]] = arith.select %[[VAL_38]], %[[VAL_37]], %[[VAL_36]] : index 193// CHECK: memref.store %[[VAL_39]], %[[VAL_23]]{{\[}}%[[VAL_2]]] : memref<?xindex> 194// CHECK: %[[VAL_40:.*]] = arith.cmpi eq, %[[VAL_36]], %[[VAL_39]] : index 195// CHECK: %[[VAL_41:.*]] = arith.cmpi eq, %[[VAL_37]], %[[VAL_39]] : index 196// CHECK: %[[VAL_42:.*]] = arith.andi %[[VAL_40]], %[[VAL_41]] : i1 197// CHECK: scf.if %[[VAL_42]] { 198// CHECK: %[[VAL_43:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_34]]] : memref<?xindex> 199// CHECK: %[[VAL_44:.*]] = arith.addi %[[VAL_34]], %[[VAL_3]] : index 200// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_44]]] : memref<?xindex> 201// CHECK: %[[VAL_46:.*]] = memref.load %[[VAL_18]]{{\[}}%[[VAL_35]]] : memref<?xindex> 202// CHECK: %[[VAL_47:.*]] = arith.addi %[[VAL_35]], %[[VAL_3]] : index 203// CHECK: %[[VAL_48:.*]] = memref.load %[[VAL_18]]{{\[}}%[[VAL_47]]] : memref<?xindex> 204// CHECK: %[[VAL_49:.*]]:2 = scf.while (%[[VAL_50:.*]] = %[[VAL_43]], %[[VAL_51:.*]] = %[[VAL_46]]) : (index, index) -> (index, index) { 205// CHECK: %[[VAL_52:.*]] = arith.cmpi ult, %[[VAL_50]], %[[VAL_45]] : index 206// CHECK: %[[VAL_53:.*]] = arith.cmpi ult, %[[VAL_51]], %[[VAL_48]] : index 207// CHECK: %[[VAL_54:.*]] = arith.andi %[[VAL_52]], %[[VAL_53]] : i1 208// CHECK: scf.condition(%[[VAL_54]]) %[[VAL_50]], %[[VAL_51]] : index, index 209// CHECK: } do { 210// CHECK: ^bb0(%[[VAL_55:.*]]: index, %[[VAL_56:.*]]: index): 211// CHECK: %[[VAL_57:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_55]]] : memref<?xindex> 212// CHECK: %[[VAL_58:.*]] = memref.load %[[VAL_19]]{{\[}}%[[VAL_56]]] : memref<?xindex> 213// CHECK: %[[VAL_59:.*]] = arith.cmpi ult, %[[VAL_58]], %[[VAL_57]] : index 214// CHECK: %[[VAL_60:.*]] = arith.select %[[VAL_59]], %[[VAL_58]], %[[VAL_57]] : index 215// CHECK: memref.store %[[VAL_60]], %[[VAL_23]]{{\[}}%[[VAL_3]]] : memref<?xindex> 216// CHECK: %[[VAL_61:.*]] = arith.cmpi eq, %[[VAL_57]], %[[VAL_60]] : index 217// CHECK: %[[VAL_62:.*]] = arith.cmpi eq, %[[VAL_58]], %[[VAL_60]] : index 218// CHECK: %[[VAL_63:.*]] = arith.andi %[[VAL_61]], %[[VAL_62]] : i1 219// CHECK: scf.if %[[VAL_63]] { 220// CHECK: %[[VAL_64:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_55]]] : memref<?xindex> 221// CHECK: %[[VAL_65:.*]] = arith.addi %[[VAL_55]], %[[VAL_3]] : index 222// CHECK: %[[VAL_66:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_65]]] : memref<?xindex> 223// CHECK: %[[VAL_67:.*]] = memref.load %[[VAL_20]]{{\[}}%[[VAL_56]]] : memref<?xindex> 224// CHECK: %[[VAL_68:.*]] = arith.addi %[[VAL_56]], %[[VAL_3]] : index 225// CHECK: %[[VAL_69:.*]] = memref.load %[[VAL_20]]{{\[}}%[[VAL_68]]] : memref<?xindex> 226// CHECK: %[[VAL_70:.*]]:3 = scf.while (%[[VAL_71:.*]] = %[[VAL_64]], %[[VAL_72:.*]] = %[[VAL_67]], %[[VAL_73:.*]] = %[[VAL_5]]) : (index, index, i32) -> (index, index, i32) { 227// CHECK: %[[VAL_74:.*]] = arith.cmpi ult, %[[VAL_71]], %[[VAL_66]] : index 228// CHECK: %[[VAL_75:.*]] = arith.cmpi ult, %[[VAL_72]], %[[VAL_69]] : index 229// CHECK: %[[VAL_76:.*]] = arith.andi %[[VAL_74]], %[[VAL_75]] : i1 230// CHECK: scf.condition(%[[VAL_76]]) %[[VAL_71]], %[[VAL_72]], %[[VAL_73]] : index, index, i32 231// CHECK: } do { 232// CHECK: ^bb0(%[[VAL_77:.*]]: index, %[[VAL_78:.*]]: index, %[[VAL_79:.*]]: i32): 233// CHECK: %[[VAL_80:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_77]]] : memref<?xindex> 234// CHECK: %[[VAL_81:.*]] = memref.load %[[VAL_21]]{{\[}}%[[VAL_78]]] : memref<?xindex> 235// CHECK: %[[VAL_82:.*]] = arith.cmpi ult, %[[VAL_81]], %[[VAL_80]] : index 236// CHECK: %[[VAL_83:.*]] = arith.select %[[VAL_82]], %[[VAL_81]], %[[VAL_80]] : index 237// CHECK: memref.store %[[VAL_83]], %[[VAL_23]]{{\[}}%[[VAL_4]]] : memref<?xindex> 238// CHECK: %[[VAL_84:.*]] = arith.cmpi eq, %[[VAL_80]], %[[VAL_83]] : index 239// CHECK: %[[VAL_85:.*]] = arith.cmpi eq, %[[VAL_81]], %[[VAL_83]] : index 240// CHECK: %[[VAL_86:.*]] = arith.andi %[[VAL_84]], %[[VAL_85]] : i1 241// CHECK: %[[VAL_87:.*]] = scf.if %[[VAL_86]] -> (i32) { 242// CHECK: %[[VAL_88:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_77]]] : memref<?xi32> 243// CHECK: %[[VAL_89:.*]] = memref.load %[[VAL_22]]{{\[}}%[[VAL_78]]] : memref<?xi32> 244// CHECK: %[[VAL_90:.*]] = arith.muli %[[VAL_88]], %[[VAL_89]] : i32 245// CHECK: %[[VAL_91:.*]] = arith.addi %[[VAL_79]], %[[VAL_90]] : i32 246// CHECK: scf.yield %[[VAL_91]] : i32 247// CHECK: } else { 248// CHECK: scf.yield %[[VAL_79]] : i32 249// CHECK: } 250// CHECK: %[[VAL_92:.*]] = arith.cmpi eq, %[[VAL_80]], %[[VAL_83]] : index 251// CHECK: %[[VAL_93:.*]] = arith.addi %[[VAL_77]], %[[VAL_3]] : index 252// CHECK: %[[VAL_94:.*]] = arith.select %[[VAL_92]], %[[VAL_93]], %[[VAL_77]] : index 253// CHECK: %[[VAL_95:.*]] = arith.cmpi eq, %[[VAL_81]], %[[VAL_83]] : index 254// CHECK: %[[VAL_96:.*]] = arith.addi %[[VAL_78]], %[[VAL_3]] : index 255// CHECK: %[[VAL_97:.*]] = arith.select %[[VAL_95]], %[[VAL_96]], %[[VAL_78]] : index 256// CHECK: scf.yield %[[VAL_94]], %[[VAL_97]], %[[VAL_98:.*]] : index, index, i32 257// CHECK: } 258// CHECK: sparse_tensor.lex_insert %[[VAL_8]], %[[VAL_23]], %[[VAL_99:.*]]#2 : tensor<?x?xi32, #{{.*}}>, memref<?xindex>, i32 259// CHECK: } else { 260// CHECK: } 261// CHECK: %[[VAL_100:.*]] = arith.cmpi eq, %[[VAL_57]], %[[VAL_60]] : index 262// CHECK: %[[VAL_101:.*]] = arith.addi %[[VAL_55]], %[[VAL_3]] : index 263// CHECK: %[[VAL_102:.*]] = arith.select %[[VAL_100]], %[[VAL_101]], %[[VAL_55]] : index 264// CHECK: %[[VAL_103:.*]] = arith.cmpi eq, %[[VAL_58]], %[[VAL_60]] : index 265// CHECK: %[[VAL_104:.*]] = arith.addi %[[VAL_56]], %[[VAL_3]] : index 266// CHECK: %[[VAL_105:.*]] = arith.select %[[VAL_103]], %[[VAL_104]], %[[VAL_56]] : index 267// CHECK: scf.yield %[[VAL_102]], %[[VAL_105]] : index, index 268// CHECK: } 269// CHECK: } else { 270// CHECK: } 271// CHECK: %[[VAL_106:.*]] = arith.cmpi eq, %[[VAL_36]], %[[VAL_39]] : index 272// CHECK: %[[VAL_107:.*]] = arith.addi %[[VAL_34]], %[[VAL_3]] : index 273// CHECK: %[[VAL_108:.*]] = arith.select %[[VAL_106]], %[[VAL_107]], %[[VAL_34]] : index 274// CHECK: %[[VAL_109:.*]] = arith.cmpi eq, %[[VAL_37]], %[[VAL_39]] : index 275// CHECK: %[[VAL_110:.*]] = arith.addi %[[VAL_35]], %[[VAL_3]] : index 276// CHECK: %[[VAL_111:.*]] = arith.select %[[VAL_109]], %[[VAL_110]], %[[VAL_35]] : index 277// CHECK: scf.yield %[[VAL_108]], %[[VAL_111]] : index, index 278// CHECK: } 279// CHECK: %[[VAL_112:.*]] = sparse_tensor.load %[[VAL_8]] hasInserts : tensor<?x?xi32, #{{.*}}> 280// CHECK: return %[[VAL_112]] : tensor<?x?xi32, #{{.*}}> 281// CHECK: } 282func.func @sumred(%arga: tensor<?x?x?xi32, #SparseTensor>, 283 %argb: tensor<?x?x?xi32, #SparseTensor>) -> tensor<?x?xi32, #DCSR> { 284 %c0 = arith.constant 0 : index 285 %c1 = arith.constant 1 : index 286 %d0 = tensor.dim %arga, %c0 : tensor<?x?x?xi32, #SparseTensor> 287 %d1 = tensor.dim %arga, %c1 : tensor<?x?x?xi32, #SparseTensor> 288 %xinit = bufferization.alloc_tensor(%d0, %d1) : tensor<?x?xi32, #DCSR> 289 %0 = linalg.generic #trait_sumred 290 ins(%arga, %argb: tensor<?x?x?xi32, #SparseTensor>, 291 tensor<?x?x?xi32, #SparseTensor>) 292 outs(%xinit: tensor<?x?xi32, #DCSR>) { 293 ^bb(%a: i32, %b: i32, %x: i32): 294 %0 = arith.muli %a, %b : i32 295 %1 = arith.addi %x, %0 : i32 296 linalg.yield %1 : i32 297 } -> tensor<?x?xi32, #DCSR> 298 return %0 : tensor<?x?xi32, #DCSR> 299} 300 301#trait_matmat = { 302 indexing_maps = [ 303 affine_map<(i,j,k) -> (i,k)>, // A 304 affine_map<(i,j,k) -> (k,j)>, // B 305 affine_map<(i,j,k) -> (i,j)> // C (out) 306 ], 307 iterator_types = ["parallel", "parallel", "reduction"], 308 doc = "C(i,j) = SUM_k A(i,k) * B(k,j)" 309} 310 311// CHECK-LABEL: func @matmat( 312// CHECK-SAME: %[[VAL_0:.*]]: tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>>, 313// CHECK-SAME: %[[VAL_1:.*]]: tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> { 314// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 0 : index 315// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index 316// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 2 : index 317// CHECK-DAG: %[[VAL_5:.*]] = arith.constant false 318// CHECK-DAG: %[[VAL_6:.*]] = arith.constant true 319// CHECK: %[[VAL_7:.*]] = tensor.dim %[[VAL_0]], %[[VAL_2]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> 320// CHECK: %[[VAL_8:.*]] = tensor.dim %[[VAL_1]], %[[VAL_3]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> 321// CHECK: %[[VAL_9:.*]] = bufferization.alloc_tensor(%[[VAL_7]], %[[VAL_8]]) : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> 322// CHECK: %[[VAL_10:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_2]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 323// CHECK: %[[VAL_11:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_2]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 324// CHECK: %[[VAL_12:.*]] = sparse_tensor.pointers %[[VAL_0]], %[[VAL_3]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 325// CHECK: %[[VAL_13:.*]] = sparse_tensor.indices %[[VAL_0]], %[[VAL_3]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 326// CHECK: %[[VAL_14:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xf32> 327// CHECK: %[[VAL_15:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_2]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 328// CHECK: %[[VAL_16:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_2]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 329// CHECK: %[[VAL_17:.*]] = sparse_tensor.pointers %[[VAL_1]], %[[VAL_3]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 330// CHECK: %[[VAL_18:.*]] = sparse_tensor.indices %[[VAL_1]], %[[VAL_3]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xindex> 331// CHECK: %[[VAL_19:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xf32> 332// CHECK: %[[VAL_20:.*]] = memref.alloca(%[[VAL_4]]) : memref<?xindex> 333// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_2]]] : memref<?xindex> 334// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_3]]] : memref<?xindex> 335// CHECK: scf.for %[[VAL_23:.*]] = %[[VAL_21]] to %[[VAL_22]] step %[[VAL_3]] { 336// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_23]]] : memref<?xindex> 337// CHECK: memref.store %[[VAL_24]], %[[VAL_20]]{{\[}}%[[VAL_2]]] : memref<?xindex> 338// CHECK: %[[VAL_25:.*]], %[[VAL_26:.*]], %[[VAL_27:.*]], %[[VAL_28:.*]] = sparse_tensor.expand %[[VAL_9]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> to memref<?xf32>, memref<?xi1>, memref<?xindex>, index 339// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_23]]] : memref<?xindex> 340// CHECK: %[[VAL_30:.*]] = arith.addi %[[VAL_23]], %[[VAL_3]] : index 341// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_30]]] : memref<?xindex> 342// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_2]]] : memref<?xindex> 343// CHECK: %[[VAL_33:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_3]]] : memref<?xindex> 344// CHECK: %[[VAL_34:.*]]:3 = scf.while (%[[VAL_35:.*]] = %[[VAL_29]], %[[VAL_36:.*]] = %[[VAL_32]], %[[VAL_37:.*]] = %[[VAL_28]]) : (index, index, index) -> (index, index, index) { 345// CHECK: %[[VAL_38:.*]] = arith.cmpi ult, %[[VAL_35]], %[[VAL_31]] : index 346// CHECK: %[[VAL_39:.*]] = arith.cmpi ult, %[[VAL_36]], %[[VAL_33]] : index 347// CHECK: %[[VAL_40:.*]] = arith.andi %[[VAL_38]], %[[VAL_39]] : i1 348// CHECK: scf.condition(%[[VAL_40]]) %[[VAL_35]], %[[VAL_36]], %[[VAL_37]] : index, index, index 349// CHECK: } do { 350// CHECK: ^bb0(%[[VAL_41:.*]]: index, %[[VAL_42:.*]]: index, %[[VAL_43:.*]]: index): 351// CHECK: %[[VAL_44:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_41]]] : memref<?xindex> 352// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_42]]] : memref<?xindex> 353// CHECK: %[[VAL_46:.*]] = arith.cmpi ult, %[[VAL_45]], %[[VAL_44]] : index 354// CHECK: %[[VAL_47:.*]] = arith.select %[[VAL_46]], %[[VAL_45]], %[[VAL_44]] : index 355// CHECK: %[[VAL_48:.*]] = arith.cmpi eq, %[[VAL_44]], %[[VAL_47]] : index 356// CHECK: %[[VAL_49:.*]] = arith.cmpi eq, %[[VAL_45]], %[[VAL_47]] : index 357// CHECK: %[[VAL_50:.*]] = arith.andi %[[VAL_48]], %[[VAL_49]] : i1 358// CHECK: %[[VAL_51:.*]] = scf.if %[[VAL_50]] -> (index) { 359// CHECK: %[[VAL_52:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_41]]] : memref<?xf32> 360// CHECK: %[[VAL_53:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_42]]] : memref<?xindex> 361// CHECK: %[[VAL_54:.*]] = arith.addi %[[VAL_42]], %[[VAL_3]] : index 362// CHECK: %[[VAL_55:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_54]]] : memref<?xindex> 363// CHECK: %[[VAL_56:.*]] = scf.for %[[VAL_57:.*]] = %[[VAL_53]] to %[[VAL_55]] step %[[VAL_3]] iter_args(%[[VAL_58:.*]] = %[[VAL_43]]) -> (index) { 364// CHECK: %[[VAL_59:.*]] = memref.load %[[VAL_18]]{{\[}}%[[VAL_57]]] : memref<?xindex> 365// CHECK: %[[VAL_60:.*]] = memref.load %[[VAL_25]]{{\[}}%[[VAL_59]]] : memref<?xf32> 366// CHECK: %[[VAL_61:.*]] = memref.load %[[VAL_19]]{{\[}}%[[VAL_57]]] : memref<?xf32> 367// CHECK: %[[VAL_62:.*]] = arith.mulf %[[VAL_52]], %[[VAL_61]] : f32 368// CHECK: %[[VAL_63:.*]] = arith.addf %[[VAL_60]], %[[VAL_62]] : f32 369// CHECK: %[[VAL_64:.*]] = memref.load %[[VAL_26]]{{\[}}%[[VAL_59]]] : memref<?xi1> 370// CHECK: %[[VAL_65:.*]] = arith.cmpi eq, %[[VAL_64]], %[[VAL_5]] : i1 371// CHECK: %[[VAL_66:.*]] = scf.if %[[VAL_65]] -> (index) { 372// CHECK: memref.store %[[VAL_6]], %[[VAL_26]]{{\[}}%[[VAL_59]]] : memref<?xi1> 373// CHECK: memref.store %[[VAL_59]], %[[VAL_27]]{{\[}}%[[VAL_58]]] : memref<?xindex> 374// CHECK: %[[VAL_67:.*]] = arith.addi %[[VAL_58]], %[[VAL_3]] : index 375// CHECK: scf.yield %[[VAL_67]] : index 376// CHECK: } else { 377// CHECK: scf.yield %[[VAL_58]] : index 378// CHECK: } 379// CHECK: memref.store %[[VAL_63]], %[[VAL_25]]{{\[}}%[[VAL_59]]] : memref<?xf32> 380// CHECK: scf.yield %[[VAL_68:.*]] : index 381// CHECK: } 382// CHECK: scf.yield %[[VAL_69:.*]] : index 383// CHECK: } else { 384// CHECK: scf.yield %[[VAL_43]] : index 385// CHECK: } 386// CHECK: %[[VAL_70:.*]] = arith.cmpi eq, %[[VAL_44]], %[[VAL_47]] : index 387// CHECK: %[[VAL_71:.*]] = arith.addi %[[VAL_41]], %[[VAL_3]] : index 388// CHECK: %[[VAL_72:.*]] = arith.select %[[VAL_70]], %[[VAL_71]], %[[VAL_41]] : index 389// CHECK: %[[VAL_73:.*]] = arith.cmpi eq, %[[VAL_45]], %[[VAL_47]] : index 390// CHECK: %[[VAL_74:.*]] = arith.addi %[[VAL_42]], %[[VAL_3]] : index 391// CHECK: %[[VAL_75:.*]] = arith.select %[[VAL_73]], %[[VAL_74]], %[[VAL_42]] : index 392// CHECK: scf.yield %[[VAL_72]], %[[VAL_75]], %[[VAL_76:.*]] : index, index, index 393// CHECK: } 394// CHECK: sparse_tensor.compress %[[VAL_9]], %[[VAL_20]], %[[VAL_25]], %[[VAL_26]], %[[VAL_27]], %[[VAL_77:.*]]#2 : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>>, memref<?xindex>, memref<?xf32>, memref<?xi1>, memref<?xindex>, index 395// CHECK: } 396// CHECK: %[[VAL_78:.*]] = sparse_tensor.load %[[VAL_9]] hasInserts : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> 397// CHECK: return %[[VAL_78]] : tensor<?x?xf32, #sparse_tensor.encoding<{{{.*}}}>> 398// CHECK: } 399func.func @matmat(%arga: tensor<?x?xf32, #DCSR>, 400 %argb: tensor<?x?xf32, #DCSR>) -> tensor<?x?xf32, #DCSR> { 401 %c0 = arith.constant 0 : index 402 %c1 = arith.constant 1 : index 403 %d0 = tensor.dim %arga, %c0 : tensor<?x?xf32, #DCSR> 404 %d1 = tensor.dim %argb, %c1 : tensor<?x?xf32, #DCSR> 405 %cinit = bufferization.alloc_tensor(%d0, %d1) : tensor<?x?xf32, #DCSR> 406 %0 = linalg.generic #trait_matmat 407 ins(%arga, %argb: tensor<?x?xf32, #DCSR>, 408 tensor<?x?xf32, #DCSR>) 409 outs(%cinit: tensor<?x?xf32, #DCSR>) { 410 ^bb(%a: f32, %b: f32, %c: f32): 411 %1 = arith.mulf %a, %b : f32 412 %2 = arith.addf %c, %1 : f32 413 linalg.yield %2 : f32 414 } -> tensor<?x?xf32, #DCSR> 415 return %0 : tensor<?x?xf32, #DCSR> 416} 417