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#Td = #sparse_tensor.encoding<{ map = (d0) -> (d0 : dense) }> 5 6#Tddd = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : dense, d1 : dense, d2 : dense) }> 7#Tdds = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : dense, d1 : dense, d2 : compressed) }> 8#Tdsd = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : dense, d1 : compressed, d2 : dense) }> 9#Tdss = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : dense, d1 : compressed, d2 : compressed) }> 10#Tsdd = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : compressed, d1 : dense, d2 : dense) }> 11#Tsds = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : compressed, d1 : dense, d2 : compressed) }> 12#Tssd = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : compressed, d1 : compressed, d2 : dense) }> 13#Tsss = #sparse_tensor.encoding<{ map = (d0, d1, d2) -> (d0 : compressed, d1 : compressed, d2 : compressed) }> 14 15#trait3 = { 16 indexing_maps = [ 17 affine_map<(i,j,k) -> (i,j,k)>, // A 18 affine_map<(i,j,k) -> (i,j,k)>, // B 19 affine_map<(i,j,k) -> (i,j,k)> // X (out) 20 ], 21 iterator_types = ["parallel", "parallel", "parallel"], 22 doc = "X(i,j,k) = A(i,j,k) OP B(i,j,k)" 23} 24 25// CHECK-LABEL: func @add_ddd( 26// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 27// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 28// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 29// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 30// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 31// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 32// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 33// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 0 : index 34// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 1 : index 35// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 36// CHECK-DAG: %[[VAL_9:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 37// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 38// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_11]] : memref<32x16x8xf32>) 39// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_6]] to %[[VAL_3]] step %[[VAL_7]] { 40// CHECK: %[[VAL_14:.*]] = arith.muli %[[VAL_12]], %[[VAL_4]] : index 41// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_6]] to %[[VAL_4]] step %[[VAL_7]] { 42// CHECK: %[[VAL_15:.*]] = arith.addi %[[VAL_13]], %[[VAL_14]] : index 43// CHECK: %[[VAL_17:.*]] = arith.muli %[[VAL_15]], %[[VAL_5]] : index 44// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_6]] to %[[VAL_5]] step %[[VAL_7]] { 45// CHECK: %[[VAL_18:.*]] = arith.addi %[[VAL_16]], %[[VAL_17]] : index 46// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_18]]] : memref<?xf32> 47// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 48// CHECK: %[[VAL_21:.*]] = arith.addf %[[VAL_19]], %[[VAL_20]] : f32 49// CHECK: memref.store %[[VAL_21]], %[[VAL_11]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 50// CHECK: } 51// CHECK: } 52// CHECK: } 53// CHECK: %[[VAL_22:.*]] = bufferization.to_tensor %[[VAL_11]] : memref<32x16x8xf32> 54// CHECK: return %[[VAL_22]] : tensor<32x16x8xf32> 55// CHECK: } 56func.func @add_ddd(%arga: tensor<32x16x8xf32, #Tddd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 57 %0 = linalg.generic #trait3 58 ins(%arga, %argb: tensor<32x16x8xf32, #Tddd>, tensor<32x16x8xf32>) 59 outs(%argx: tensor<32x16x8xf32>) { 60 ^bb(%a: f32, %b: f32, %x: f32): 61 %0 = arith.addf %a, %b : f32 62 linalg.yield %0 : f32 63 } -> tensor<32x16x8xf32> 64 return %0 : tensor<32x16x8xf32> 65} 66 67// CHECK-LABEL: func @mul_ddd( 68// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 69// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 70// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 71// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 72// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 73// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 74// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 75// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 0 : index 76// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 1 : index 77// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 78// CHECK-DAG: %[[VAL_9:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 79// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 80// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_11]] : memref<32x16x8xf32>) 81// CHECK: scf.for %[[VAL_12:.*]] = %[[VAL_6]] to %[[VAL_3]] step %[[VAL_7]] { 82// CHECK: %[[VAL_14:.*]] = arith.muli %[[VAL_12]], %[[VAL_4]] : index 83// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_6]] to %[[VAL_4]] step %[[VAL_7]] { 84// CHECK: %[[VAL_15:.*]] = arith.addi %[[VAL_13]], %[[VAL_14]] : index 85// CHECK: %[[VAL_17:.*]] = arith.muli %[[VAL_15]], %[[VAL_5]] : index 86// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_6]] to %[[VAL_5]] step %[[VAL_7]] { 87// CHECK: %[[VAL_18:.*]] = arith.addi %[[VAL_16]], %[[VAL_17]] : index 88// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_18]]] : memref<?xf32> 89// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 90// CHECK: %[[VAL_21:.*]] = arith.mulf %[[VAL_19]], %[[VAL_20]] : f32 91// CHECK: memref.store %[[VAL_21]], %[[VAL_11]]{{\[}}%[[VAL_12]], %[[VAL_13]], %[[VAL_16]]] : memref<32x16x8xf32> 92// CHECK: } 93// CHECK: } 94// CHECK: } 95// CHECK: %[[VAL_22:.*]] = bufferization.to_tensor %[[VAL_11]] : memref<32x16x8xf32> 96// CHECK: return %[[VAL_22]] : tensor<32x16x8xf32> 97// CHECK: } 98func.func @mul_ddd(%arga: tensor<32x16x8xf32, #Tddd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 99 %0 = linalg.generic #trait3 100 ins(%arga, %argb: tensor<32x16x8xf32, #Tddd>, tensor<32x16x8xf32>) 101 outs(%argx: tensor<32x16x8xf32>) { 102 ^bb(%a: f32, %b: f32, %x: f32): 103 %0 = arith.mulf %a, %b : f32 104 linalg.yield %0 : f32 105 } -> tensor<32x16x8xf32> 106 return %0 : tensor<32x16x8xf32> 107} 108 109// CHECK-LABEL: func @add_dds( 110// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 111// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 112// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 113// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 114// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 115// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 116// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 117// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 118// CHECK-DAG: %[[VAL_8:.*]] = arith.constant true 119// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 120// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 121// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 122// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 123// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 124// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 125// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_15]] : memref<32x16x8xf32>) 126// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_9]] { 127// CHECK: %[[VAL_18:.*]] = arith.muli %[[VAL_16]], %[[VAL_5]] : index 128// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_9]] { 129// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_17]], %[[VAL_18]] : index 130// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_19]]] : memref<?xindex> 131// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_19]], %[[VAL_9]] : index 132// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xindex> 133// CHECK: %[[VAL_23:.*]]:2 = scf.while (%[[VAL_24:.*]] = %[[VAL_20]], %[[VAL_25:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 134// CHECK: %[[VAL_26:.*]] = arith.cmpi ult, %[[VAL_24]], %[[VAL_22]] : index 135// CHECK: scf.condition(%[[VAL_26]]) %[[VAL_24]], %[[VAL_25]] : index, index 136// CHECK: } do { 137// CHECK: ^bb0(%[[VAL_27:.*]]: index, %[[VAL_28:.*]]: index): 138// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_27]]] : memref<?xindex> 139// CHECK: %[[VAL_30:.*]] = arith.cmpi eq, %[[VAL_29]], %[[VAL_28]] : index 140// CHECK: scf.if %[[VAL_30]] { 141// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_27]]] : memref<?xf32> 142// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 143// CHECK: %[[VAL_33:.*]] = arith.addf %[[VAL_31]], %[[VAL_32]] : f32 144// CHECK: memref.store %[[VAL_33]], %[[VAL_15]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 145// CHECK: } else { 146// CHECK: scf.if %[[VAL_8]] { 147// CHECK: %[[VAL_34:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 148// CHECK: memref.store %[[VAL_34]], %[[VAL_15]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_28]]] : memref<32x16x8xf32> 149// CHECK: } else { 150// CHECK: } 151// CHECK: } 152// CHECK: %[[VAL_35:.*]] = arith.cmpi eq, %[[VAL_29]], %[[VAL_28]] : index 153// CHECK: %[[VAL_36:.*]] = arith.addi %[[VAL_27]], %[[VAL_9]] : index 154// CHECK: %[[VAL_37:.*]] = arith.select %[[VAL_35]], %[[VAL_36]], %[[VAL_27]] : index 155// CHECK: %[[VAL_38:.*]] = arith.addi %[[VAL_28]], %[[VAL_9]] : index 156// CHECK: scf.yield %[[VAL_37]], %[[VAL_38]] : index, index 157// CHECK: } 158// CHECK: scf.for %[[VAL_39:.*]] = %[[VAL_40:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 159// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_39]]] : memref<32x16x8xf32> 160// CHECK: memref.store %[[VAL_41]], %[[VAL_15]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_39]]] : memref<32x16x8xf32> 161// CHECK: } 162// CHECK: } 163// CHECK: } 164// CHECK: %[[VAL_42:.*]] = bufferization.to_tensor %[[VAL_15]] : memref<32x16x8xf32> 165// CHECK: return %[[VAL_42]] : tensor<32x16x8xf32> 166// CHECK: } 167func.func @add_dds(%arga: tensor<32x16x8xf32, #Tdds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 168 %0 = linalg.generic #trait3 169 ins(%arga, %argb: tensor<32x16x8xf32, #Tdds>, tensor<32x16x8xf32>) 170 outs(%argx: tensor<32x16x8xf32>) { 171 ^bb(%a: f32, %b: f32, %x: f32): 172 %0 = arith.addf %a, %b : f32 173 linalg.yield %0 : f32 174 } -> tensor<32x16x8xf32> 175 return %0 : tensor<32x16x8xf32> 176} 177 178// CHECK-LABEL: func @mul_dds( 179// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 180// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 181// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 182// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 183// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 184// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 185// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 0 : index 186// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 1 : index 187// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 188// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 189// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 190// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 191// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 192// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_13]] : memref<32x16x8xf32>) 193// CHECK: scf.for %[[VAL_14:.*]] = %[[VAL_6]] to %[[VAL_4]] step %[[VAL_7]] { 194// CHECK: %[[VAL_16:.*]] = arith.muli %[[VAL_14]], %[[VAL_5]] : index 195// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_6]] to %[[VAL_5]] step %[[VAL_7]] { 196// CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_15]], %[[VAL_16]] : index 197// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_17]]] : memref<?xindex> 198// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_17]], %[[VAL_7]] : index 199// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex> 200// CHECK: scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_7]] { 201// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex> 202// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xf32> 203// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_14]], %[[VAL_15]], %[[VAL_22]]] : memref<32x16x8xf32> 204// CHECK: %[[VAL_25:.*]] = arith.mulf %[[VAL_23]], %[[VAL_24]] : f32 205// CHECK: memref.store %[[VAL_25]], %[[VAL_13]]{{\[}}%[[VAL_14]], %[[VAL_15]], %[[VAL_22]]] : memref<32x16x8xf32> 206// CHECK: } 207// CHECK: } 208// CHECK: } 209// CHECK: %[[VAL_26:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<32x16x8xf32> 210// CHECK: return %[[VAL_26]] : tensor<32x16x8xf32> 211// CHECK: } 212func.func @mul_dds(%arga: tensor<32x16x8xf32, #Tdds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 213 %0 = linalg.generic #trait3 214 ins(%arga, %argb: tensor<32x16x8xf32, #Tdds>, tensor<32x16x8xf32>) 215 outs(%argx: tensor<32x16x8xf32>) { 216 ^bb(%a: f32, %b: f32, %x: f32): 217 %0 = arith.mulf %a, %b : f32 218 linalg.yield %0 : f32 219 } -> tensor<32x16x8xf32> 220 return %0 : tensor<32x16x8xf32> 221} 222 223// CHECK-LABEL: func @add_dsd( 224// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 225// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 226// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 227// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 228// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 229// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 230// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 231// CHECK-DAG: %[[VAL_6:.*]] = arith.constant true 232// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 233// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 234// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 235// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 236// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 237// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 238// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 239// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 240// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_7]] to %[[VAL_3]] step %[[VAL_8]] { 241// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_15]]] : memref<?xindex> 242// CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_15]], %[[VAL_8]] : index 243// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_17]]] : memref<?xindex> 244// CHECK: %[[VAL_19:.*]]:2 = scf.while (%[[VAL_20:.*]] = %[[VAL_16]], %[[VAL_21:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 245// CHECK: %[[VAL_22:.*]] = arith.cmpi ult, %[[VAL_20]], %[[VAL_18]] : index 246// CHECK: scf.condition(%[[VAL_22]]) %[[VAL_20]], %[[VAL_21]] : index, index 247// CHECK: } do { 248// CHECK: ^bb0(%[[VAL_23:.*]]: index, %[[VAL_24:.*]]: index): 249// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_23]]] : memref<?xindex> 250// CHECK: %[[VAL_26:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 251// CHECK: scf.if %[[VAL_26]] { 252// CHECK: %[[VAL_28:.*]] = arith.muli %[[VAL_23]], %[[VAL_5]] : index 253// CHECK: scf.for %[[VAL_27:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 254// CHECK: %[[VAL_29:.*]] = arith.addi %[[VAL_27]], %[[VAL_28]] : index 255// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_29]]] : memref<?xf32> 256// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_27]]] : memref<32x16x8xf32> 257// CHECK: %[[VAL_32:.*]] = arith.addf %[[VAL_30]], %[[VAL_31]] : f32 258// CHECK: memref.store %[[VAL_32]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_27]]] : memref<32x16x8xf32> 259// CHECK: } 260// CHECK: } else { 261// CHECK: scf.if %[[VAL_6]] { 262// CHECK: scf.for %[[VAL_33:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 263// CHECK: %[[VAL_34:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_33]]] : memref<32x16x8xf32> 264// CHECK: memref.store %[[VAL_34]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_24]], %[[VAL_33]]] : memref<32x16x8xf32> 265// CHECK: } 266// CHECK: } else { 267// CHECK: } 268// CHECK: } 269// CHECK: %[[VAL_35:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 270// CHECK: %[[VAL_36:.*]] = arith.addi %[[VAL_23]], %[[VAL_8]] : index 271// CHECK: %[[VAL_37:.*]] = arith.select %[[VAL_35]], %[[VAL_36]], %[[VAL_23]] : index 272// CHECK: %[[VAL_38:.*]] = arith.addi %[[VAL_24]], %[[VAL_8]] : index 273// CHECK: scf.yield %[[VAL_37]], %[[VAL_38]] : index, index 274// CHECK: } 275// CHECK: scf.for %[[VAL_39:.*]] = %[[VAL_40:.*]]#1 to %[[VAL_4]] step %[[VAL_8]] { 276// CHECK: scf.for %[[VAL_41:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 277// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_39]], %[[VAL_41]]] : memref<32x16x8xf32> 278// CHECK: memref.store %[[VAL_42]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_39]], %[[VAL_41]]] : memref<32x16x8xf32> 279// CHECK: } 280// CHECK: } 281// CHECK: } 282// CHECK: %[[VAL_43:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 283// CHECK: return %[[VAL_43]] : tensor<32x16x8xf32> 284// CHECK: } 285func.func @add_dsd(%arga: tensor<32x16x8xf32, #Tdsd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 286 %0 = linalg.generic #trait3 287 ins(%arga, %argb: tensor<32x16x8xf32, #Tdsd>, tensor<32x16x8xf32>) 288 outs(%argx: tensor<32x16x8xf32>) { 289 ^bb(%a: f32, %b: f32, %x: f32): 290 %0 = arith.addf %a, %b : f32 291 linalg.yield %0 : f32 292 } -> tensor<32x16x8xf32> 293 return %0 : tensor<32x16x8xf32> 294} 295 296// CHECK-LABEL: func @mul_dsd( 297// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 298// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 299// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 300// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 301// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 302// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 8 : index 303// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 304// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 305// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 306// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 307// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 308// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 309// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 310// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_12]] : memref<32x16x8xf32>) 311// CHECK: scf.for %[[VAL_13:.*]] = %[[VAL_5]] to %[[VAL_3]] step %[[VAL_6]] { 312// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_13]]] : memref<?xindex> 313// CHECK: %[[VAL_15:.*]] = arith.addi %[[VAL_13]], %[[VAL_6]] : index 314// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_15]]] : memref<?xindex> 315// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_14]] to %[[VAL_16]] step %[[VAL_6]] { 316// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_17]]] : memref<?xindex> 317// CHECK: %[[VAL_20:.*]] = arith.muli %[[VAL_17]], %[[VAL_4]] : index 318// CHECK: scf.for %[[VAL_19:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 319// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_19]], %[[VAL_20]] : index 320// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xf32> 321// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_13]], %[[VAL_18]], %[[VAL_19]]] : memref<32x16x8xf32> 322// CHECK: %[[VAL_24:.*]] = arith.mulf %[[VAL_22]], %[[VAL_23]] : f32 323// CHECK: memref.store %[[VAL_24]], %[[VAL_12]]{{\[}}%[[VAL_13]], %[[VAL_18]], %[[VAL_19]]] : memref<32x16x8xf32> 324// CHECK: } 325// CHECK: } 326// CHECK: } 327// CHECK: %[[VAL_25:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<32x16x8xf32> 328// CHECK: return %[[VAL_25]] : tensor<32x16x8xf32> 329// CHECK: } 330func.func @mul_dsd(%arga: tensor<32x16x8xf32, #Tdsd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 331 %0 = linalg.generic #trait3 332 ins(%arga, %argb: tensor<32x16x8xf32, #Tdsd>, tensor<32x16x8xf32>) 333 outs(%argx: tensor<32x16x8xf32>) { 334 ^bb(%a: f32, %b: f32, %x: f32): 335 %0 = arith.mulf %a, %b : f32 336 linalg.yield %0 : f32 337 } -> tensor<32x16x8xf32> 338 return %0 : tensor<32x16x8xf32> 339} 340 341// CHECK-LABEL: func @add_dss( 342// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 343// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 344// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 345// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 346// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 347// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 348// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 349// CHECK-DAG: %[[VAL_7:.*]] = arith.constant true 350// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 0 : index 351// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 352// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 353// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 354// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 355// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 356// CHECK-DAG: %[[VAL_14:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 357// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 358// CHECK-DAG: %[[VAL_17:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 359// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_17]] : memref<32x16x8xf32>) 360// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_8]] to %[[VAL_4]] step %[[VAL_9]] { 361// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_18]]] : memref<?xindex> 362// CHECK: %[[VAL_20:.*]] = arith.addi %[[VAL_18]], %[[VAL_9]] : index 363// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_20]]] : memref<?xindex> 364// CHECK: %[[VAL_22:.*]]:2 = scf.while (%[[VAL_23:.*]] = %[[VAL_19]], %[[VAL_24:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 365// CHECK: %[[VAL_25:.*]] = arith.cmpi ult, %[[VAL_23]], %[[VAL_21]] : index 366// CHECK: scf.condition(%[[VAL_25]]) %[[VAL_23]], %[[VAL_24]] : index, index 367// CHECK: } do { 368// CHECK: ^bb0(%[[VAL_26:.*]]: index, %[[VAL_27:.*]]: index): 369// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_26]]] : memref<?xindex> 370// CHECK: %[[VAL_29:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 371// CHECK: scf.if %[[VAL_29]] { 372// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_26]]] : memref<?xindex> 373// CHECK: %[[VAL_31:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 374// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_31]]] : memref<?xindex> 375// CHECK: %[[VAL_33:.*]]:2 = scf.while (%[[VAL_34:.*]] = %[[VAL_30]], %[[VAL_35:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 376// CHECK: %[[VAL_36:.*]] = arith.cmpi ult, %[[VAL_34]], %[[VAL_32]] : index 377// CHECK: scf.condition(%[[VAL_36]]) %[[VAL_34]], %[[VAL_35]] : index, index 378// CHECK: } do { 379// CHECK: ^bb0(%[[VAL_37:.*]]: index, %[[VAL_38:.*]]: index): 380// CHECK: %[[VAL_39:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_37]]] : memref<?xindex> 381// CHECK: %[[VAL_40:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 382// CHECK: scf.if %[[VAL_40]] { 383// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_37]]] : memref<?xf32> 384// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 385// CHECK: %[[VAL_43:.*]] = arith.addf %[[VAL_41]], %[[VAL_42]] : f32 386// CHECK: memref.store %[[VAL_43]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 387// CHECK: } else { 388// CHECK: scf.if %[[VAL_7]] { 389// CHECK: %[[VAL_44:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 390// CHECK: memref.store %[[VAL_44]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_38]]] : memref<32x16x8xf32> 391// CHECK: } else { 392// CHECK: } 393// CHECK: } 394// CHECK: %[[VAL_45:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 395// CHECK: %[[VAL_46:.*]] = arith.addi %[[VAL_37]], %[[VAL_9]] : index 396// CHECK: %[[VAL_47:.*]] = arith.select %[[VAL_45]], %[[VAL_46]], %[[VAL_37]] : index 397// CHECK: %[[VAL_48:.*]] = arith.addi %[[VAL_38]], %[[VAL_9]] : index 398// CHECK: scf.yield %[[VAL_47]], %[[VAL_48]] : index, index 399// CHECK: } 400// CHECK: scf.for %[[VAL_49:.*]] = %[[VAL_50:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 401// CHECK: %[[VAL_51:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_49]]] : memref<32x16x8xf32> 402// CHECK: memref.store %[[VAL_51]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_49]]] : memref<32x16x8xf32> 403// CHECK: } 404// CHECK: } else { 405// CHECK: scf.if %[[VAL_7]] { 406// CHECK: scf.for %[[VAL_52:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 407// CHECK: %[[VAL_53:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_52]]] : memref<32x16x8xf32> 408// CHECK: memref.store %[[VAL_53]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_27]], %[[VAL_52]]] : memref<32x16x8xf32> 409// CHECK: } 410// CHECK: } else { 411// CHECK: } 412// CHECK: } 413// CHECK: %[[VAL_54:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 414// CHECK: %[[VAL_55:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 415// CHECK: %[[VAL_56:.*]] = arith.select %[[VAL_54]], %[[VAL_55]], %[[VAL_26]] : index 416// CHECK: %[[VAL_57:.*]] = arith.addi %[[VAL_27]], %[[VAL_9]] : index 417// CHECK: scf.yield %[[VAL_56]], %[[VAL_57]] : index, index 418// CHECK: } 419// CHECK: scf.for %[[VAL_58:.*]] = %[[VAL_59:.*]]#1 to %[[VAL_5]] step %[[VAL_9]] { 420// CHECK: scf.for %[[VAL_60:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 421// CHECK: %[[VAL_61:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_18]], %[[VAL_58]], %[[VAL_60]]] : memref<32x16x8xf32> 422// CHECK: memref.store %[[VAL_61]], %[[VAL_17]]{{\[}}%[[VAL_18]], %[[VAL_58]], %[[VAL_60]]] : memref<32x16x8xf32> 423// CHECK: } 424// CHECK: } 425// CHECK: } 426// CHECK: %[[VAL_62:.*]] = bufferization.to_tensor %[[VAL_17]] : memref<32x16x8xf32> 427// CHECK: return %[[VAL_62]] : tensor<32x16x8xf32> 428// CHECK: } 429func.func @add_dss(%arga: tensor<32x16x8xf32, #Tdss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 430 %0 = linalg.generic #trait3 431 ins(%arga, %argb: tensor<32x16x8xf32, #Tdss>, tensor<32x16x8xf32>) 432 outs(%argx: tensor<32x16x8xf32>) { 433 ^bb(%a: f32, %b: f32, %x: f32): 434 %0 = arith.addf %a, %b : f32 435 linalg.yield %0 : f32 436 } -> tensor<32x16x8xf32> 437 return %0 : tensor<32x16x8xf32> 438} 439 440// CHECK-LABEL: func @mul_dss( 441// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 442// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 443// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 444// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 445// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 446// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 447// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 448// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 449// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 450// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 451// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 452// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 453// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 454// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 455// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 456// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 457// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_15]]] : memref<?xindex> 458// CHECK: %[[VAL_17:.*]] = arith.addi %[[VAL_15]], %[[VAL_6]] : index 459// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_17]]] : memref<?xindex> 460// CHECK: scf.for %[[VAL_19:.*]] = %[[VAL_16]] to %[[VAL_18]] step %[[VAL_6]] { 461// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex> 462// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_19]]] : memref<?xindex> 463// CHECK: %[[VAL_22:.*]] = arith.addi %[[VAL_19]], %[[VAL_6]] : index 464// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_22]]] : memref<?xindex> 465// CHECK: scf.for %[[VAL_24:.*]] = %[[VAL_21]] to %[[VAL_23]] step %[[VAL_6]] { 466// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_24]]] : memref<?xindex> 467// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_24]]] : memref<?xf32> 468// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_15]], %[[VAL_20]], %[[VAL_25]]] : memref<32x16x8xf32> 469// CHECK: %[[VAL_28:.*]] = arith.mulf %[[VAL_26]], %[[VAL_27]] : f32 470// CHECK: memref.store %[[VAL_28]], %[[VAL_14]]{{\[}}%[[VAL_15]], %[[VAL_20]], %[[VAL_25]]] : memref<32x16x8xf32> 471// CHECK: } 472// CHECK: } 473// CHECK: } 474// CHECK: %[[VAL_29:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 475// CHECK: return %[[VAL_29]] : tensor<32x16x8xf32> 476// CHECK: } 477func.func @mul_dss(%arga: tensor<32x16x8xf32, #Tdss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 478 %0 = linalg.generic #trait3 479 ins(%arga, %argb: tensor<32x16x8xf32, #Tdss>, tensor<32x16x8xf32>) 480 outs(%argx: tensor<32x16x8xf32>) { 481 ^bb(%a: f32, %b: f32, %x: f32): 482 %0 = arith.mulf %a, %b : f32 483 linalg.yield %0 : f32 484 } -> tensor<32x16x8xf32> 485 return %0 : tensor<32x16x8xf32> 486} 487 488// CHECK-LABEL: func @add_sdd( 489// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 490// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 491// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 492// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 493// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 494// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 495// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 496// CHECK-DAG: %[[VAL_6:.*]] = arith.constant true 497// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 498// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 499// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 500// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 501// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 502// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 503// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 504// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 505// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_7]]] : memref<?xindex> 506// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_8]]] : memref<?xindex> 507// CHECK: %[[VAL_17:.*]]:2 = scf.while (%[[VAL_18:.*]] = %[[VAL_15]], %[[VAL_19:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 508// CHECK: %[[VAL_20:.*]] = arith.cmpi ult, %[[VAL_18]], %[[VAL_16]] : index 509// CHECK: scf.condition(%[[VAL_20]]) %[[VAL_18]], %[[VAL_19]] : index, index 510// CHECK: } do { 511// CHECK: ^bb0(%[[VAL_21:.*]]: index, %[[VAL_22:.*]]: index): 512// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_21]]] : memref<?xindex> 513// CHECK: %[[VAL_24:.*]] = arith.cmpi eq, %[[VAL_23]], %[[VAL_22]] : index 514// CHECK: scf.if %[[VAL_24]] { 515// CHECK: %[[VAL_26:.*]] = arith.muli %[[VAL_21]], %[[VAL_4]] : index 516// CHECK: scf.for %[[VAL_25:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 517// CHECK: %[[VAL_27:.*]] = arith.addi %[[VAL_25]], %[[VAL_26]] : index 518// CHECK: %[[VAL_29:.*]] = arith.muli %[[VAL_27]], %[[VAL_5]] : index 519// CHECK: scf.for %[[VAL_28:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 520// CHECK: %[[VAL_30:.*]] = arith.addi %[[VAL_28]], %[[VAL_29]] : index 521// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_30]]] : memref<?xf32> 522// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_22]], %[[VAL_25]], %[[VAL_28]]] : memref<32x16x8xf32> 523// CHECK: %[[VAL_33:.*]] = arith.addf %[[VAL_31]], %[[VAL_32]] : f32 524// CHECK: memref.store %[[VAL_33]], %[[VAL_14]]{{\[}}%[[VAL_22]], %[[VAL_25]], %[[VAL_28]]] : memref<32x16x8xf32> 525// CHECK: } 526// CHECK: } 527// CHECK: } else { 528// CHECK: scf.if %[[VAL_6]] { 529// CHECK: scf.for %[[VAL_34:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 530// CHECK: scf.for %[[VAL_35:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 531// CHECK: %[[VAL_36:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_22]], %[[VAL_34]], %[[VAL_35]]] : memref<32x16x8xf32> 532// CHECK: memref.store %[[VAL_36]], %[[VAL_14]]{{\[}}%[[VAL_22]], %[[VAL_34]], %[[VAL_35]]] : memref<32x16x8xf32> 533// CHECK: } 534// CHECK: } 535// CHECK: } else { 536// CHECK: } 537// CHECK: } 538// CHECK: %[[VAL_37:.*]] = arith.cmpi eq, %[[VAL_23]], %[[VAL_22]] : index 539// CHECK: %[[VAL_38:.*]] = arith.addi %[[VAL_21]], %[[VAL_8]] : index 540// CHECK: %[[VAL_39:.*]] = arith.select %[[VAL_37]], %[[VAL_38]], %[[VAL_21]] : index 541// CHECK: %[[VAL_40:.*]] = arith.addi %[[VAL_22]], %[[VAL_8]] : index 542// CHECK: scf.yield %[[VAL_39]], %[[VAL_40]] : index, index 543// CHECK: } 544// CHECK: scf.for %[[VAL_41:.*]] = %[[VAL_42:.*]]#1 to %[[VAL_3]] step %[[VAL_8]] { 545// CHECK: scf.for %[[VAL_43:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 546// CHECK: scf.for %[[VAL_44:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 547// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_41]], %[[VAL_43]], %[[VAL_44]]] : memref<32x16x8xf32> 548// CHECK: memref.store %[[VAL_45]], %[[VAL_14]]{{\[}}%[[VAL_41]], %[[VAL_43]], %[[VAL_44]]] : memref<32x16x8xf32> 549// CHECK: } 550// CHECK: } 551// CHECK: } 552// CHECK: %[[VAL_46:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 553// CHECK: return %[[VAL_46]] : tensor<32x16x8xf32> 554// CHECK: } 555func.func @add_sdd(%arga: tensor<32x16x8xf32, #Tsdd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 556 %0 = linalg.generic #trait3 557 ins(%arga, %argb: tensor<32x16x8xf32, #Tsdd>, tensor<32x16x8xf32>) 558 outs(%argx: tensor<32x16x8xf32>) { 559 ^bb(%a: f32, %b: f32, %x: f32): 560 %0 = arith.addf %a, %b : f32 561 linalg.yield %0 : f32 562 } -> tensor<32x16x8xf32> 563 return %0 : tensor<32x16x8xf32> 564} 565 566// CHECK-LABEL: func @mul_sdd( 567// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 568// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 569// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 570// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 571// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 16 : index 572// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 8 : index 573// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 574// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 575// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 576// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 577// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 578// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 579// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 580// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_12]] : memref<32x16x8xf32>) 581// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_5]]] : memref<?xindex> 582// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_6]]] : memref<?xindex> 583// CHECK: scf.for %[[VAL_15:.*]] = %[[VAL_13]] to %[[VAL_14]] step %[[VAL_6]] { 584// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_15]]] : memref<?xindex> 585// CHECK: %[[VAL_18:.*]] = arith.muli %[[VAL_15]], %[[VAL_3]] : index 586// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_5]] to %[[VAL_3]] step %[[VAL_6]] { 587// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_17]], %[[VAL_18]] : index 588// CHECK: %[[VAL_21:.*]] = arith.muli %[[VAL_19]], %[[VAL_4]] : index 589// CHECK: scf.for %[[VAL_20:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 590// CHECK: %[[VAL_22:.*]] = arith.addi %[[VAL_20]], %[[VAL_21]] : index 591// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_22]]] : memref<?xf32> 592// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_20]]] : memref<32x16x8xf32> 593// CHECK: %[[VAL_25:.*]] = arith.mulf %[[VAL_23]], %[[VAL_24]] : f32 594// CHECK: memref.store %[[VAL_25]], %[[VAL_12]]{{\[}}%[[VAL_16]], %[[VAL_17]], %[[VAL_20]]] : memref<32x16x8xf32> 595// CHECK: } 596// CHECK: } 597// CHECK: } 598// CHECK: %[[VAL_26:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<32x16x8xf32> 599// CHECK: return %[[VAL_26]] : tensor<32x16x8xf32> 600// CHECK: } 601func.func @mul_sdd(%arga: tensor<32x16x8xf32, #Tsdd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 602 %0 = linalg.generic #trait3 603 ins(%arga, %argb: tensor<32x16x8xf32, #Tsdd>, tensor<32x16x8xf32>) 604 outs(%argx: tensor<32x16x8xf32>) { 605 ^bb(%a: f32, %b: f32, %x: f32): 606 %0 = arith.mulf %a, %b : f32 607 linalg.yield %0 : f32 608 } -> tensor<32x16x8xf32> 609 return %0 : tensor<32x16x8xf32> 610} 611 612// CHECK-LABEL: func @add_sds( 613// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 614// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 615// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 616// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 617// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 618// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 619// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 620// CHECK-DAG: %[[VAL_7:.*]] = arith.constant true 621// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 0 : index 622// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 623// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 624// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 625// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 626// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 627// CHECK-DAG: %[[VAL_14:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 628// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 629// CHECK-DAG: %[[VAL_17:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 630// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_17]] : memref<32x16x8xf32>) 631// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_8]]] : memref<?xindex> 632// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_9]]] : memref<?xindex> 633// CHECK: %[[VAL_20:.*]]:2 = scf.while (%[[VAL_21:.*]] = %[[VAL_18]], %[[VAL_22:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 634// CHECK: %[[VAL_23:.*]] = arith.cmpi ult, %[[VAL_21]], %[[VAL_19]] : index 635// CHECK: scf.condition(%[[VAL_23]]) %[[VAL_21]], %[[VAL_22]] : index, index 636// CHECK: } do { 637// CHECK: ^bb0(%[[VAL_24:.*]]: index, %[[VAL_25:.*]]: index): 638// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_24]]] : memref<?xindex> 639// CHECK: %[[VAL_27:.*]] = arith.cmpi eq, %[[VAL_26]], %[[VAL_25]] : index 640// CHECK: scf.if %[[VAL_27]] { 641// CHECK: %[[VAL_29:.*]] = arith.muli %[[VAL_24]], %[[VAL_5]] : index 642// CHECK: scf.for %[[VAL_28:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 643// CHECK: %[[VAL_30:.*]] = arith.addi %[[VAL_28]], %[[VAL_29]] : index 644// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_30]]] : memref<?xindex> 645// CHECK: %[[VAL_32:.*]] = arith.addi %[[VAL_30]], %[[VAL_9]] : index 646// CHECK: %[[VAL_33:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_32]]] : memref<?xindex> 647// CHECK: %[[VAL_34:.*]]:2 = scf.while (%[[VAL_35:.*]] = %[[VAL_31]], %[[VAL_36:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 648// CHECK: %[[VAL_37:.*]] = arith.cmpi ult, %[[VAL_35]], %[[VAL_33]] : index 649// CHECK: scf.condition(%[[VAL_37]]) %[[VAL_35]], %[[VAL_36]] : index, index 650// CHECK: } do { 651// CHECK: ^bb0(%[[VAL_38:.*]]: index, %[[VAL_39:.*]]: index): 652// CHECK: %[[VAL_40:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_38]]] : memref<?xindex> 653// CHECK: %[[VAL_41:.*]] = arith.cmpi eq, %[[VAL_40]], %[[VAL_39]] : index 654// CHECK: scf.if %[[VAL_41]] { 655// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_38]]] : memref<?xf32> 656// CHECK: %[[VAL_43:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 657// CHECK: %[[VAL_44:.*]] = arith.addf %[[VAL_42]], %[[VAL_43]] : f32 658// CHECK: memref.store %[[VAL_44]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 659// CHECK: } else { 660// CHECK: scf.if %[[VAL_7]] { 661// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 662// CHECK: memref.store %[[VAL_45]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_39]]] : memref<32x16x8xf32> 663// CHECK: } else { 664// CHECK: } 665// CHECK: } 666// CHECK: %[[VAL_46:.*]] = arith.cmpi eq, %[[VAL_40]], %[[VAL_39]] : index 667// CHECK: %[[VAL_47:.*]] = arith.addi %[[VAL_38]], %[[VAL_9]] : index 668// CHECK: %[[VAL_48:.*]] = arith.select %[[VAL_46]], %[[VAL_47]], %[[VAL_38]] : index 669// CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_39]], %[[VAL_9]] : index 670// CHECK: scf.yield %[[VAL_48]], %[[VAL_49]] : index, index 671// CHECK: } 672// CHECK: scf.for %[[VAL_50:.*]] = %[[VAL_51:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 673// CHECK: %[[VAL_52:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_50]]] : memref<32x16x8xf32> 674// CHECK: memref.store %[[VAL_52]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_28]], %[[VAL_50]]] : memref<32x16x8xf32> 675// CHECK: } 676// CHECK: } 677// CHECK: } else { 678// CHECK: scf.if %[[VAL_7]] { 679// CHECK: scf.for %[[VAL_53:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 680// CHECK: scf.for %[[VAL_54:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 681// CHECK: %[[VAL_55:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_25]], %[[VAL_53]], %[[VAL_54]]] : memref<32x16x8xf32> 682// CHECK: memref.store %[[VAL_55]], %[[VAL_17]]{{\[}}%[[VAL_25]], %[[VAL_53]], %[[VAL_54]]] : memref<32x16x8xf32> 683// CHECK: } 684// CHECK: } 685// CHECK: } else { 686// CHECK: } 687// CHECK: } 688// CHECK: %[[VAL_56:.*]] = arith.cmpi eq, %[[VAL_26]], %[[VAL_25]] : index 689// CHECK: %[[VAL_57:.*]] = arith.addi %[[VAL_24]], %[[VAL_9]] : index 690// CHECK: %[[VAL_58:.*]] = arith.select %[[VAL_56]], %[[VAL_57]], %[[VAL_24]] : index 691// CHECK: %[[VAL_59:.*]] = arith.addi %[[VAL_25]], %[[VAL_9]] : index 692// CHECK: scf.yield %[[VAL_58]], %[[VAL_59]] : index, index 693// CHECK: } 694// CHECK: scf.for %[[VAL_60:.*]] = %[[VAL_61:.*]]#1 to %[[VAL_4]] step %[[VAL_9]] { 695// CHECK: scf.for %[[VAL_62:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 696// CHECK: scf.for %[[VAL_63:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 697// CHECK: %[[VAL_64:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_60]], %[[VAL_62]], %[[VAL_63]]] : memref<32x16x8xf32> 698// CHECK: memref.store %[[VAL_64]], %[[VAL_17]]{{\[}}%[[VAL_60]], %[[VAL_62]], %[[VAL_63]]] : memref<32x16x8xf32> 699// CHECK: } 700// CHECK: } 701// CHECK: } 702// CHECK: %[[VAL_65:.*]] = bufferization.to_tensor %[[VAL_17]] : memref<32x16x8xf32> 703// CHECK: return %[[VAL_65]] : tensor<32x16x8xf32> 704// CHECK: } 705func.func @add_sds(%arga: tensor<32x16x8xf32, #Tsds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 706 %0 = linalg.generic #trait3 707 ins(%arga, %argb: tensor<32x16x8xf32, #Tsds>, tensor<32x16x8xf32>) 708 outs(%argx: tensor<32x16x8xf32>) { 709 ^bb(%a: f32, %b: f32, %x: f32): 710 %0 = arith.addf %a, %b : f32 711 linalg.yield %0 : f32 712 } -> tensor<32x16x8xf32> 713 return %0 : tensor<32x16x8xf32> 714} 715 716// CHECK-LABEL: func @mul_sds( 717// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 718// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 719// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 720// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 721// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 722// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 723// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 724// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 725// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 726// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 727// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 728// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 729// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 730// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 731// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_14]] : memref<32x16x8xf32>) 732// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_5]]] : memref<?xindex> 733// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_6]]] : memref<?xindex> 734// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_15]] to %[[VAL_16]] step %[[VAL_6]] { 735// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_17]]] : memref<?xindex> 736// CHECK: %[[VAL_20:.*]] = arith.muli %[[VAL_17]], %[[VAL_4]] : index 737// CHECK: scf.for %[[VAL_19:.*]] = %[[VAL_5]] to %[[VAL_4]] step %[[VAL_6]] { 738// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_19]], %[[VAL_20]] : index 739// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex> 740// CHECK: %[[VAL_23:.*]] = arith.addi %[[VAL_21]], %[[VAL_6]] : index 741// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_23]]] : memref<?xindex> 742// CHECK: scf.for %[[VAL_25:.*]] = %[[VAL_22]] to %[[VAL_24]] step %[[VAL_6]] { 743// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_25]]] : memref<?xindex> 744// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_25]]] : memref<?xf32> 745// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_18]], %[[VAL_19]], %[[VAL_26]]] : memref<32x16x8xf32> 746// CHECK: %[[VAL_29:.*]] = arith.mulf %[[VAL_27]], %[[VAL_28]] : f32 747// CHECK: memref.store %[[VAL_29]], %[[VAL_14]]{{\[}}%[[VAL_18]], %[[VAL_19]], %[[VAL_26]]] : memref<32x16x8xf32> 748// CHECK: } 749// CHECK: } 750// CHECK: } 751// CHECK: %[[VAL_30:.*]] = bufferization.to_tensor %[[VAL_14]] : memref<32x16x8xf32> 752// CHECK: return %[[VAL_30]] : tensor<32x16x8xf32> 753// CHECK: } 754func.func @mul_sds(%arga: tensor<32x16x8xf32, #Tsds>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 755 %0 = linalg.generic #trait3 756 ins(%arga, %argb: tensor<32x16x8xf32, #Tsds>, tensor<32x16x8xf32>) 757 outs(%argx: tensor<32x16x8xf32>) { 758 ^bb(%a: f32, %b: f32, %x: f32): 759 %0 = arith.mulf %a, %b : f32 760 linalg.yield %0 : f32 761 } -> tensor<32x16x8xf32> 762 return %0 : tensor<32x16x8xf32> 763} 764 765// CHECK-LABEL: func @add_ssd( 766// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 767// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 768// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 769// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 770// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 32 : index 771// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 16 : index 772// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 8 : index 773// CHECK-DAG: %[[VAL_6:.*]] = arith.constant true 774// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 775// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 776// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 777// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 778// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 779// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 780// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 781// CHECK-DAG: %[[VAL_14:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 782// CHECK-DAG: %[[VAL_16:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 783// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_16]] : memref<32x16x8xf32>) 784// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_7]]] : memref<?xindex> 785// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_8]]] : memref<?xindex> 786// CHECK: %[[VAL_19:.*]]:2 = scf.while (%[[VAL_20:.*]] = %[[VAL_17]], %[[VAL_21:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 787// CHECK: %[[VAL_22:.*]] = arith.cmpi ult, %[[VAL_20]], %[[VAL_18]] : index 788// CHECK: scf.condition(%[[VAL_22]]) %[[VAL_20]], %[[VAL_21]] : index, index 789// CHECK: } do { 790// CHECK: ^bb0(%[[VAL_23:.*]]: index, %[[VAL_24:.*]]: index): 791// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_23]]] : memref<?xindex> 792// CHECK: %[[VAL_26:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 793// CHECK: scf.if %[[VAL_26]] { 794// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_23]]] : memref<?xindex> 795// CHECK: %[[VAL_28:.*]] = arith.addi %[[VAL_23]], %[[VAL_8]] : index 796// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_28]]] : memref<?xindex> 797// CHECK: %[[VAL_30:.*]]:2 = scf.while (%[[VAL_31:.*]] = %[[VAL_27]], %[[VAL_32:.*]] = %[[VAL_7]]) : (index, index) -> (index, index) { 798// CHECK: %[[VAL_33:.*]] = arith.cmpi ult, %[[VAL_31]], %[[VAL_29]] : index 799// CHECK: scf.condition(%[[VAL_33]]) %[[VAL_31]], %[[VAL_32]] : index, index 800// CHECK: } do { 801// CHECK: ^bb0(%[[VAL_34:.*]]: index, %[[VAL_35:.*]]: index): 802// CHECK: %[[VAL_36:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_34]]] : memref<?xindex> 803// CHECK: %[[VAL_37:.*]] = arith.cmpi eq, %[[VAL_36]], %[[VAL_35]] : index 804// CHECK: scf.if %[[VAL_37]] { 805// CHECK: %[[VAL_39:.*]] = arith.muli %[[VAL_34]], %[[VAL_5]] : index 806// CHECK: scf.for %[[VAL_38:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 807// CHECK: %[[VAL_40:.*]] = arith.addi %[[VAL_38]], %[[VAL_39]] : index 808// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_40]]] : memref<?xf32> 809// CHECK: %[[VAL_42:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_38]]] : memref<32x16x8xf32> 810// CHECK: %[[VAL_43:.*]] = arith.addf %[[VAL_41]], %[[VAL_42]] : f32 811// CHECK: memref.store %[[VAL_43]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_38]]] : memref<32x16x8xf32> 812// CHECK: } 813// CHECK: } else { 814// CHECK: scf.if %[[VAL_6]] { 815// CHECK: scf.for %[[VAL_44:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 816// CHECK: %[[VAL_45:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_44]]] : memref<32x16x8xf32> 817// CHECK: memref.store %[[VAL_45]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_35]], %[[VAL_44]]] : memref<32x16x8xf32> 818// CHECK: } 819// CHECK: } else { 820// CHECK: } 821// CHECK: } 822// CHECK: %[[VAL_46:.*]] = arith.cmpi eq, %[[VAL_36]], %[[VAL_35]] : index 823// CHECK: %[[VAL_47:.*]] = arith.addi %[[VAL_34]], %[[VAL_8]] : index 824// CHECK: %[[VAL_48:.*]] = arith.select %[[VAL_46]], %[[VAL_47]], %[[VAL_34]] : index 825// CHECK: %[[VAL_49:.*]] = arith.addi %[[VAL_35]], %[[VAL_8]] : index 826// CHECK: scf.yield %[[VAL_48]], %[[VAL_49]] : index, index 827// CHECK: } 828// CHECK: scf.for %[[VAL_50:.*]] = %[[VAL_51:.*]]#1 to %[[VAL_4]] step %[[VAL_8]] { 829// CHECK: scf.for %[[VAL_52:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 830// CHECK: %[[VAL_53:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_50]], %[[VAL_52]]] : memref<32x16x8xf32> 831// CHECK: memref.store %[[VAL_53]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_50]], %[[VAL_52]]] : memref<32x16x8xf32> 832// CHECK: } 833// CHECK: } 834// CHECK: } else { 835// CHECK: scf.if %[[VAL_6]] { 836// CHECK: scf.for %[[VAL_54:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 837// CHECK: scf.for %[[VAL_55:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 838// CHECK: %[[VAL_56:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_24]], %[[VAL_54]], %[[VAL_55]]] : memref<32x16x8xf32> 839// CHECK: memref.store %[[VAL_56]], %[[VAL_16]]{{\[}}%[[VAL_24]], %[[VAL_54]], %[[VAL_55]]] : memref<32x16x8xf32> 840// CHECK: } 841// CHECK: } 842// CHECK: } else { 843// CHECK: } 844// CHECK: } 845// CHECK: %[[VAL_57:.*]] = arith.cmpi eq, %[[VAL_25]], %[[VAL_24]] : index 846// CHECK: %[[VAL_58:.*]] = arith.addi %[[VAL_23]], %[[VAL_8]] : index 847// CHECK: %[[VAL_59:.*]] = arith.select %[[VAL_57]], %[[VAL_58]], %[[VAL_23]] : index 848// CHECK: %[[VAL_60:.*]] = arith.addi %[[VAL_24]], %[[VAL_8]] : index 849// CHECK: scf.yield %[[VAL_59]], %[[VAL_60]] : index, index 850// CHECK: } 851// CHECK: scf.for %[[VAL_61:.*]] = %[[VAL_62:.*]]#1 to %[[VAL_3]] step %[[VAL_8]] { 852// CHECK: scf.for %[[VAL_63:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 853// CHECK: scf.for %[[VAL_64:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 854// CHECK: %[[VAL_65:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_61]], %[[VAL_63]], %[[VAL_64]]] : memref<32x16x8xf32> 855// CHECK: memref.store %[[VAL_65]], %[[VAL_16]]{{\[}}%[[VAL_61]], %[[VAL_63]], %[[VAL_64]]] : memref<32x16x8xf32> 856// CHECK: } 857// CHECK: } 858// CHECK: } 859// CHECK: %[[VAL_66:.*]] = bufferization.to_tensor %[[VAL_16]] : memref<32x16x8xf32> 860// CHECK: return %[[VAL_66]] : tensor<32x16x8xf32> 861// CHECK: } 862func.func @add_ssd(%arga: tensor<32x16x8xf32, #Tssd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 863 %0 = linalg.generic #trait3 864 ins(%arga, %argb: tensor<32x16x8xf32, #Tssd>, tensor<32x16x8xf32>) 865 outs(%argx: tensor<32x16x8xf32>) { 866 ^bb(%a: f32, %b: f32, %x: f32): 867 %0 = arith.addf %a, %b : f32 868 linalg.yield %0 : f32 869 } -> tensor<32x16x8xf32> 870 return %0 : tensor<32x16x8xf32> 871} 872 873// CHECK-LABEL: func @mul_ssd( 874// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 875// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 876// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 877// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 878// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 8 : index 879// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0 : index 880// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 1 : index 881// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 882// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 883// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 884// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 885// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 886// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 887// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 888// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_13]] : memref<32x16x8xf32>) 889// CHECK: %[[VAL_14:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_4]]] : memref<?xindex> 890// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_5]]] : memref<?xindex> 891// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_14]] to %[[VAL_15]] step %[[VAL_5]] { 892// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_16]]] : memref<?xindex> 893// CHECK: %[[VAL_18:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_16]]] : memref<?xindex> 894// CHECK: %[[VAL_19:.*]] = arith.addi %[[VAL_16]], %[[VAL_5]] : index 895// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_19]]] : memref<?xindex> 896// CHECK: scf.for %[[VAL_21:.*]] = %[[VAL_18]] to %[[VAL_20]] step %[[VAL_5]] { 897// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_21]]] : memref<?xindex> 898// CHECK: %[[VAL_24:.*]] = arith.muli %[[VAL_21]], %[[VAL_3]] : index 899// CHECK: scf.for %[[VAL_23:.*]] = %[[VAL_4]] to %[[VAL_3]] step %[[VAL_5]] { 900// CHECK: %[[VAL_25:.*]] = arith.addi %[[VAL_23]], %[[VAL_24]] : index 901// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_25]]] : memref<?xf32> 902// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_17]], %[[VAL_22]], %[[VAL_23]]] : memref<32x16x8xf32> 903// CHECK: %[[VAL_28:.*]] = arith.mulf %[[VAL_26]], %[[VAL_27]] : f32 904// CHECK: memref.store %[[VAL_28]], %[[VAL_13]]{{\[}}%[[VAL_17]], %[[VAL_22]], %[[VAL_23]]] : memref<32x16x8xf32> 905// CHECK: } 906// CHECK: } 907// CHECK: } 908// CHECK: %[[VAL_29:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<32x16x8xf32> 909// CHECK: return %[[VAL_29]] : tensor<32x16x8xf32> 910// CHECK: } 911func.func @mul_ssd(%arga: tensor<32x16x8xf32, #Tssd>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 912 %0 = linalg.generic #trait3 913 ins(%arga, %argb: tensor<32x16x8xf32, #Tssd>, tensor<32x16x8xf32>) 914 outs(%argx: tensor<32x16x8xf32>) { 915 ^bb(%a: f32, %b: f32, %x: f32): 916 %0 = arith.mulf %a, %b : f32 917 linalg.yield %0 : f32 918 } -> tensor<32x16x8xf32> 919 return %0 : tensor<32x16x8xf32> 920} 921 922// CHECK-LABEL: func @add_sss( 923// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 924// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 925// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 926// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 927// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 32 : index 928// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 16 : index 929// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 8 : index 930// CHECK-DAG: %[[VAL_7:.*]] = arith.constant true 931// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 0 : index 932// CHECK-DAG: %[[VAL_9:.*]] = arith.constant 1 : index 933// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 934// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 935// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 936// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 937// CHECK-DAG: %[[VAL_14:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 938// CHECK-DAG: %[[VAL_15:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 939// CHECK-DAG: %[[VAL_16:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 940// CHECK-DAG: %[[VAL_17:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 941// CHECK-DAG: %[[VAL_19:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 942// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_19]] : memref<32x16x8xf32>) 943// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_8]]] : memref<?xindex> 944// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_9]]] : memref<?xindex> 945// CHECK: %[[VAL_22:.*]]:2 = scf.while (%[[VAL_23:.*]] = %[[VAL_20]], %[[VAL_24:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 946// CHECK: %[[VAL_25:.*]] = arith.cmpi ult, %[[VAL_23]], %[[VAL_21]] : index 947// CHECK: scf.condition(%[[VAL_25]]) %[[VAL_23]], %[[VAL_24]] : index, index 948// CHECK: } do { 949// CHECK: ^bb0(%[[VAL_26:.*]]: index, %[[VAL_27:.*]]: index): 950// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_26]]] : memref<?xindex> 951// CHECK: %[[VAL_29:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 952// CHECK: scf.if %[[VAL_29]] { 953// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_26]]] : memref<?xindex> 954// CHECK: %[[VAL_31:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 955// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_31]]] : memref<?xindex> 956// CHECK: %[[VAL_33:.*]]:2 = scf.while (%[[VAL_34:.*]] = %[[VAL_30]], %[[VAL_35:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 957// CHECK: %[[VAL_36:.*]] = arith.cmpi ult, %[[VAL_34]], %[[VAL_32]] : index 958// CHECK: scf.condition(%[[VAL_36]]) %[[VAL_34]], %[[VAL_35]] : index, index 959// CHECK: } do { 960// CHECK: ^bb0(%[[VAL_37:.*]]: index, %[[VAL_38:.*]]: index): 961// CHECK: %[[VAL_39:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_37]]] : memref<?xindex> 962// CHECK: %[[VAL_40:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 963// CHECK: scf.if %[[VAL_40]] { 964// CHECK: %[[VAL_41:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_37]]] : memref<?xindex> 965// CHECK: %[[VAL_42:.*]] = arith.addi %[[VAL_37]], %[[VAL_9]] : index 966// CHECK: %[[VAL_43:.*]] = memref.load %[[VAL_14]]{{\[}}%[[VAL_42]]] : memref<?xindex> 967// CHECK: %[[VAL_44:.*]]:2 = scf.while (%[[VAL_45:.*]] = %[[VAL_41]], %[[VAL_46:.*]] = %[[VAL_8]]) : (index, index) -> (index, index) { 968// CHECK: %[[VAL_47:.*]] = arith.cmpi ult, %[[VAL_45]], %[[VAL_43]] : index 969// CHECK: scf.condition(%[[VAL_47]]) %[[VAL_45]], %[[VAL_46]] : index, index 970// CHECK: } do { 971// CHECK: ^bb0(%[[VAL_48:.*]]: index, %[[VAL_49:.*]]: index): 972// CHECK: %[[VAL_50:.*]] = memref.load %[[VAL_15]]{{\[}}%[[VAL_48]]] : memref<?xindex> 973// CHECK: %[[VAL_51:.*]] = arith.cmpi eq, %[[VAL_50]], %[[VAL_49]] : index 974// CHECK: scf.if %[[VAL_51]] { 975// CHECK: %[[VAL_52:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_48]]] : memref<?xf32> 976// CHECK: %[[VAL_53:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 977// CHECK: %[[VAL_54:.*]] = arith.addf %[[VAL_52]], %[[VAL_53]] : f32 978// CHECK: memref.store %[[VAL_54]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 979// CHECK: } else { 980// CHECK: scf.if %[[VAL_7]] { 981// CHECK: %[[VAL_55:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 982// CHECK: memref.store %[[VAL_55]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_49]]] : memref<32x16x8xf32> 983// CHECK: } else { 984// CHECK: } 985// CHECK: } 986// CHECK: %[[VAL_56:.*]] = arith.cmpi eq, %[[VAL_50]], %[[VAL_49]] : index 987// CHECK: %[[VAL_57:.*]] = arith.addi %[[VAL_48]], %[[VAL_9]] : index 988// CHECK: %[[VAL_58:.*]] = arith.select %[[VAL_56]], %[[VAL_57]], %[[VAL_48]] : index 989// CHECK: %[[VAL_59:.*]] = arith.addi %[[VAL_49]], %[[VAL_9]] : index 990// CHECK: scf.yield %[[VAL_58]], %[[VAL_59]] : index, index 991// CHECK: } 992// CHECK: scf.for %[[VAL_60:.*]] = %[[VAL_61:.*]]#1 to %[[VAL_6]] step %[[VAL_9]] { 993// CHECK: %[[VAL_62:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_60]]] : memref<32x16x8xf32> 994// CHECK: memref.store %[[VAL_62]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_60]]] : memref<32x16x8xf32> 995// CHECK: } 996// CHECK: } else { 997// CHECK: scf.if %[[VAL_7]] { 998// CHECK: scf.for %[[VAL_63:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 999// CHECK: %[[VAL_64:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_63]]] : memref<32x16x8xf32> 1000// CHECK: memref.store %[[VAL_64]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_38]], %[[VAL_63]]] : memref<32x16x8xf32> 1001// CHECK: } 1002// CHECK: } else { 1003// CHECK: } 1004// CHECK: } 1005// CHECK: %[[VAL_65:.*]] = arith.cmpi eq, %[[VAL_39]], %[[VAL_38]] : index 1006// CHECK: %[[VAL_66:.*]] = arith.addi %[[VAL_37]], %[[VAL_9]] : index 1007// CHECK: %[[VAL_67:.*]] = arith.select %[[VAL_65]], %[[VAL_66]], %[[VAL_37]] : index 1008// CHECK: %[[VAL_68:.*]] = arith.addi %[[VAL_38]], %[[VAL_9]] : index 1009// CHECK: scf.yield %[[VAL_67]], %[[VAL_68]] : index, index 1010// CHECK: } 1011// CHECK: scf.for %[[VAL_69:.*]] = %[[VAL_70:.*]]#1 to %[[VAL_5]] step %[[VAL_9]] { 1012// CHECK: scf.for %[[VAL_71:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 1013// CHECK: %[[VAL_72:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_69]], %[[VAL_71]]] : memref<32x16x8xf32> 1014// CHECK: memref.store %[[VAL_72]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_69]], %[[VAL_71]]] : memref<32x16x8xf32> 1015// CHECK: } 1016// CHECK: } 1017// CHECK: } else { 1018// CHECK: scf.if %[[VAL_7]] { 1019// CHECK: scf.for %[[VAL_73:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 1020// CHECK: scf.for %[[VAL_74:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 1021// CHECK: %[[VAL_75:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_27]], %[[VAL_73]], %[[VAL_74]]] : memref<32x16x8xf32> 1022// CHECK: memref.store %[[VAL_75]], %[[VAL_19]]{{\[}}%[[VAL_27]], %[[VAL_73]], %[[VAL_74]]] : memref<32x16x8xf32> 1023// CHECK: } 1024// CHECK: } 1025// CHECK: } else { 1026// CHECK: } 1027// CHECK: } 1028// CHECK: %[[VAL_76:.*]] = arith.cmpi eq, %[[VAL_28]], %[[VAL_27]] : index 1029// CHECK: %[[VAL_77:.*]] = arith.addi %[[VAL_26]], %[[VAL_9]] : index 1030// CHECK: %[[VAL_78:.*]] = arith.select %[[VAL_76]], %[[VAL_77]], %[[VAL_26]] : index 1031// CHECK: %[[VAL_79:.*]] = arith.addi %[[VAL_27]], %[[VAL_9]] : index 1032// CHECK: scf.yield %[[VAL_78]], %[[VAL_79]] : index, index 1033// CHECK: } 1034// CHECK: scf.for %[[VAL_80:.*]] = %[[VAL_81:.*]]#1 to %[[VAL_4]] step %[[VAL_9]] { 1035// CHECK: scf.for %[[VAL_82:.*]] = %[[VAL_8]] to %[[VAL_5]] step %[[VAL_9]] { 1036// CHECK: scf.for %[[VAL_83:.*]] = %[[VAL_8]] to %[[VAL_6]] step %[[VAL_9]] { 1037// CHECK: %[[VAL_84:.*]] = memref.load %[[VAL_17]]{{\[}}%[[VAL_80]], %[[VAL_82]], %[[VAL_83]]] : memref<32x16x8xf32> 1038// CHECK: memref.store %[[VAL_84]], %[[VAL_19]]{{\[}}%[[VAL_80]], %[[VAL_82]], %[[VAL_83]]] : memref<32x16x8xf32> 1039// CHECK: } 1040// CHECK: } 1041// CHECK: } 1042// CHECK: %[[VAL_85:.*]] = bufferization.to_tensor %[[VAL_19]] : memref<32x16x8xf32> 1043// CHECK: return %[[VAL_85]] : tensor<32x16x8xf32> 1044// CHECK: } 1045func.func @add_sss(%arga: tensor<32x16x8xf32, #Tsss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 1046 %0 = linalg.generic #trait3 1047 ins(%arga, %argb: tensor<32x16x8xf32, #Tsss>, tensor<32x16x8xf32>) 1048 outs(%argx: tensor<32x16x8xf32>) { 1049 ^bb(%a: f32, %b: f32, %x: f32): 1050 %0 = arith.addf %a, %b : f32 1051 linalg.yield %0 : f32 1052 } -> tensor<32x16x8xf32> 1053 return %0 : tensor<32x16x8xf32> 1054} 1055 1056// CHECK-LABEL: func @mul_sss( 1057// CHECK-SAME: %[[VAL_0:.*]]: tensor<32x16x8xf32, #sparse{{[0-9]*}}>, 1058// CHECK-SAME: %[[VAL_1:.*]]: tensor<32x16x8xf32>, 1059// CHECK-SAME: %[[VAL_2:.*]]: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 1060// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 1061// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 0 : index 1062// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 1 : index 1063// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1064// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 0 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1065// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1066// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 1 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1067// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1068// CHECK-DAG: %[[VAL_11:.*]] = sparse_tensor.coordinates %[[VAL_0]] {level = 2 : index} : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1069// CHECK-DAG: %[[VAL_12:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<32x16x8xf32, #sparse{{[0-9]*}}> to memref<?xf32> 1070// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 1071// CHECK-DAG: %[[VAL_15:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<32x16x8xf32> to memref<32x16x8xf32> 1072// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_15]] : memref<32x16x8xf32>) 1073// CHECK: %[[VAL_16:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_4]]] : memref<?xindex> 1074// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_5]]] : memref<?xindex> 1075// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_16]] to %[[VAL_17]] step %[[VAL_5]] { 1076// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_18]]] : memref<?xindex> 1077// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_18]]] : memref<?xindex> 1078// CHECK: %[[VAL_21:.*]] = arith.addi %[[VAL_18]], %[[VAL_5]] : index 1079// CHECK: %[[VAL_22:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_21]]] : memref<?xindex> 1080// CHECK: scf.for %[[VAL_23:.*]] = %[[VAL_20]] to %[[VAL_22]] step %[[VAL_5]] { 1081// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_23]]] : memref<?xindex> 1082// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_23]]] : memref<?xindex> 1083// CHECK: %[[VAL_26:.*]] = arith.addi %[[VAL_23]], %[[VAL_5]] : index 1084// CHECK: %[[VAL_27:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_26]]] : memref<?xindex> 1085// CHECK: scf.for %[[VAL_28:.*]] = %[[VAL_25]] to %[[VAL_27]] step %[[VAL_5]] { 1086// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_28]]] : memref<?xindex> 1087// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_28]]] : memref<?xf32> 1088// CHECK: %[[VAL_31:.*]] = memref.load %[[VAL_13]]{{\[}}%[[VAL_19]], %[[VAL_24]], %[[VAL_29]]] : memref<32x16x8xf32> 1089// CHECK: %[[VAL_32:.*]] = arith.mulf %[[VAL_30]], %[[VAL_31]] : f32 1090// CHECK: memref.store %[[VAL_32]], %[[VAL_15]]{{\[}}%[[VAL_19]], %[[VAL_24]], %[[VAL_29]]] : memref<32x16x8xf32> 1091// CHECK: } 1092// CHECK: } 1093// CHECK: } 1094// CHECK: %[[VAL_33:.*]] = bufferization.to_tensor %[[VAL_15]] : memref<32x16x8xf32> 1095// CHECK: return %[[VAL_33]] : tensor<32x16x8xf32> 1096// CHECK: } 1097func.func @mul_sss(%arga: tensor<32x16x8xf32, #Tsss>, %argb: tensor<32x16x8xf32>, %argx: tensor<32x16x8xf32>) -> tensor<32x16x8xf32> { 1098 %0 = linalg.generic #trait3 1099 ins(%arga, %argb: tensor<32x16x8xf32, #Tsss>, tensor<32x16x8xf32>) 1100 outs(%argx: tensor<32x16x8xf32>) { 1101 ^bb(%a: f32, %b: f32, %x: f32): 1102 %0 = arith.mulf %a, %b : f32 1103 linalg.yield %0 : f32 1104 } -> tensor<32x16x8xf32> 1105 return %0 : tensor<32x16x8xf32> 1106} 1107 1108#trait_kernel_3d = { 1109 indexing_maps = [ 1110 affine_map<(i,j,k,l) -> (i,k,l)>, // B 1111 affine_map<(i,j,k,l) -> (k,j)>, // C 1112 affine_map<(i,j,k,l) -> (l,j)>, // D 1113 affine_map<(i,j,k,l) -> (i,j)> // A (out) 1114 ], 1115 iterator_types = ["parallel", "parallel", "reduction", "reduction"], 1116 doc = "A(i,j) += SUM_k,l B(i,k,l) * C(k,j) * D(l,j)" 1117} 1118 1119// CHECK-LABEL: func @kernel_3d( 1120// CHECK-SAME: %[[VAL_0:.*0]]: tensor<?x?xf32>, 1121// CHECK-SAME: %[[VAL_1:.*1]]: tensor<?x?x?xf32, #sparse{{[0-9]*}}>, 1122// CHECK-SAME: %[[VAL_2:.*2]]: tensor<?x?xf32>, 1123// CHECK-SAME: %[[VAL_3:.*3]]: tensor<?x?xf32>) -> tensor<?x?xf32> { 1124// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 1125// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 1 : index 1126// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_1]] {level = 2 : index} : tensor<?x?x?xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1127// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.coordinates %[[VAL_1]] {level = 2 : index} : tensor<?x?x?xf32, #sparse{{[0-9]*}}> to memref<?xindex> 1128// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<?x?x?xf32, #sparse{{[0-9]*}}> to memref<?xf32> 1129// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.lvl %[[VAL_1]], %[[VAL_6]] : tensor<?x?x?xf32, #sparse{{[0-9]*}}> 1130// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<?x?xf32> to memref<?x?xf32> 1131// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_3]] : tensor<?x?xf32> to memref<?x?xf32> 1132// CHECK-DAG: %[[VAL_13:.*]] = sparse_tensor.lvl %[[VAL_1]], %[[VAL_5]] : tensor<?x?x?xf32, #sparse{{[0-9]*}}> 1133// CHECK-DAG: %[[VAL_14:.*]] = tensor.dim %[[VAL_2]], %[[VAL_6]] : tensor<?x?xf32> 1134// CHECK-DAG: %[[VAL_16:.*]] = bufferization.to_memref %[[VAL_0]] : tensor<?x?xf32> to memref<?x?xf32> 1135// CHECK: scf.for %[[VAL_17:.*]] = %[[VAL_5]] to %[[VAL_13]] step %[[VAL_6]] { 1136// CHECK: %[[VAL_19:.*]] = arith.muli %[[VAL_17]], %[[VAL_10]] : index 1137// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_5]] to %[[VAL_10]] step %[[VAL_6]] { 1138// CHECK: %[[VAL_20:.*]] = arith.addi %[[VAL_18]], %[[VAL_19]] : index 1139// CHECK: %[[VAL_21:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_20]]] : memref<?xindex> 1140// CHECK: %[[VAL_22:.*]] = arith.addi %[[VAL_20]], %[[VAL_6]] : index 1141// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_22]]] : memref<?xindex> 1142// CHECK: scf.for %[[VAL_24:.*]] = %[[VAL_21]] to %[[VAL_23]] step %[[VAL_6]] { 1143// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_24]]] : memref<?xindex> 1144// CHECK: %[[VAL_26:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_24]]] : memref<?xf32> 1145// CHECK: scf.for %[[VAL_27:.*]] = %[[VAL_5]] to %[[VAL_14]] step %[[VAL_6]] { 1146// CHECK: %[[VAL_28:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_18]], %[[VAL_27]]] : memref<?x?xf32> 1147// CHECK: %[[VAL_29:.*]] = arith.mulf %[[VAL_26]], %[[VAL_28]] : f32 1148// CHECK: %[[VAL_30:.*]] = memref.load %[[VAL_12]]{{\[}}%[[VAL_25]], %[[VAL_27]]] : memref<?x?xf32> 1149// CHECK: %[[VAL_31:.*]] = arith.mulf %[[VAL_29]], %[[VAL_30]] : f32 1150// CHECK: %[[VAL_32:.*]] = memref.load %[[VAL_16]]{{\[}}%[[VAL_17]], %[[VAL_27]]] : memref<?x?xf32> 1151// CHECK: %[[VAL_33:.*]] = arith.addf %[[VAL_31]], %[[VAL_32]] : f32 1152// CHECK: memref.store %[[VAL_33]], %[[VAL_16]]{{\[}}%[[VAL_17]], %[[VAL_27]]] : memref<?x?xf32> 1153// CHECK: } 1154// CHECK: } 1155// CHECK: } 1156// CHECK: } 1157// CHECK: %[[VAL_34:.*]] = bufferization.to_tensor %[[VAL_16]] : memref<?x?xf32> 1158// CHECK: return %[[VAL_34]] : tensor<?x?xf32> 1159// CHECK: } 1160func.func @kernel_3d(%arga: tensor<?x?xf32>, 1161 %argb: tensor<?x?x?xf32, #Tdds>, 1162 %argc: tensor<?x?xf32>, 1163 %argd: tensor<?x?xf32>) -> tensor<?x?xf32> { 1164 %0 = linalg.generic #trait_kernel_3d 1165 ins(%argb, %argc, %argd: tensor<?x?x?xf32, #Tdds>, tensor<?x?xf32>, tensor<?x?xf32>) 1166 outs(%arga: tensor<?x?xf32>) { 1167 ^bb(%b: f32, %c: f32, %d: f32, %a: f32): 1168 %0 = arith.mulf %b, %c : f32 1169 %1 = arith.mulf %0, %d : f32 1170 %2 = arith.addf %1, %a : f32 1171 linalg.yield %2 : f32 1172 } -> tensor<?x?xf32> 1173 return %0 : tensor<?x?xf32> 1174} 1175 1176#trait_sum_reduction = { 1177 indexing_maps = [ 1178 affine_map<(i,j,k) -> (i,j,k)>, // A 1179 affine_map<(i,j,k) -> ()> // x (scalar out) 1180 ], 1181 iterator_types = ["reduction", "reduction", "reduction"], 1182 doc = "x += SUM_ijk A(i,j,k)" 1183} 1184 1185// CHECK-LABEL: func @sum_reduction( 1186// CHECK-SAME: %[[VAL_0:.*]]: tensor<10x20x30xf32, #sparse{{[0-9]*}}>, 1187// CHECK-SAME: %[[VAL_1:.*]]: tensor<f32>) -> tensor<f32> { 1188// CHECK-DAG: %[[VAL_2:.*]] = arith.constant 0 : index 1189// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index 1190// CHECK-DAG: %[[VAL_5:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 0 : index} : tensor<10x20x30xf32, #sparse{{[0-9]*}}> 1191// CHECK-DAG: %[[VAL_6:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 1 : index} : tensor<10x20x30xf32, #sparse{{[0-9]*}}> 1192// CHECK-DAG: %[[VAL_7:.*]] = sparse_tensor.positions %[[VAL_0]] {level = 2 : index} : tensor<10x20x30xf32, #sparse{{[0-9]*}}> 1193// CHECK-DAG: %[[VAL_8:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10x20x30xf32, #sparse{{[0-9]*}}> 1194// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<f32> to memref<f32> 1195// CHECK: %[[VAL_11:.*]] = memref.load %[[VAL_10]][] : memref<f32> 1196// CHECK: %[[VAL_12:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_2]]] : memref<?xindex> 1197// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_5]]{{\[}}%[[VAL_3]]] : memref<?xindex> 1198// CHECK: %[[VAL_14:.*]] = scf.for %[[VAL_15:.*]] = %[[VAL_12]] to %[[VAL_13]] step %[[VAL_3]] iter_args(%[[VAL_16:.*]] = %[[VAL_11]]) -> (f32) { 1199// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_15]]] : memref<?xindex> 1200// CHECK: %[[VAL_18:.*]] = arith.addi %[[VAL_15]], %[[VAL_3]] : index 1201// CHECK: %[[VAL_19:.*]] = memref.load %[[VAL_6]]{{\[}}%[[VAL_18]]] : memref<?xindex> 1202// CHECK: %[[VAL_20:.*]] = scf.for %[[VAL_21:.*]] = %[[VAL_17]] to %[[VAL_19]] step %[[VAL_3]] iter_args(%[[VAL_22:.*]] = %[[VAL_16]]) -> (f32) { 1203// CHECK: %[[VAL_23:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_21]]] : memref<?xindex> 1204// CHECK: %[[VAL_24:.*]] = arith.addi %[[VAL_21]], %[[VAL_3]] : index 1205// CHECK: %[[VAL_25:.*]] = memref.load %[[VAL_7]]{{\[}}%[[VAL_24]]] : memref<?xindex> 1206// CHECK: %[[VAL_26:.*]] = scf.for %[[VAL_27:.*]] = %[[VAL_23]] to %[[VAL_25]] step %[[VAL_3]] iter_args(%[[VAL_28:.*]] = %[[VAL_22]]) -> (f32) { 1207// CHECK: %[[VAL_29:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_27]]] : memref<?xf32> 1208// CHECK: %[[VAL_30:.*]] = arith.addf %[[VAL_28]], %[[VAL_29]] : f32 1209// CHECK: scf.yield %[[VAL_30]] : f32 1210// CHECK: } 1211// CHECK: scf.yield %[[VAL_26]] : f32 1212// CHECK: } 1213// CHECK: scf.yield %[[VAL_20]] : f32 1214// CHECK: } 1215// CHECK: memref.store %[[VAL_14]], %[[VAL_10]][] : memref<f32> 1216// CHECK: %[[VAL_34:.*]] = bufferization.to_tensor %[[VAL_10]] : memref<f32> 1217// CHECK: return %[[VAL_34]] : tensor<f32> 1218// CHECK: } 1219func.func @sum_reduction(%arga: tensor<10x20x30xf32, #Tsss>, %argx: tensor<f32>) -> tensor<f32> { 1220 %0 = linalg.generic #trait_sum_reduction 1221 ins(%arga: tensor<10x20x30xf32, #Tsss>) 1222 outs(%argx: tensor<f32>) { 1223 ^bb(%a: f32, %x: f32): 1224 %0 = arith.addf %x, %a : f32 1225 linalg.yield %0 : f32 1226 } -> tensor<f32> 1227 return %0 : tensor<f32> 1228} 1229 1230#trait_sum_reduction_inv = { 1231 indexing_maps = [ 1232 affine_map<(i,j,k) -> (i,j,k)>, // A 1233 affine_map<(i,j,k) -> (i)>, // b 1234 affine_map<(i,j,k) -> ()> // x (scalar out) 1235 ], 1236 iterator_types = ["reduction", "reduction", "reduction"], 1237 doc = "x += SUM_i A(i,j,k) * b(i)" 1238} 1239 1240// CHECK-LABEL: func @sum_reduction_inv( 1241// CHECK-SAME: %[[VAL_0:.*]]: tensor<?x?x?xf32>, 1242// CHECK-SAME: %[[VAL_1:.*]]: tensor<?xf32, #sparse{{[0-9]*}}> 1243// CHECK-SAME: %[[VAL_2:.*]]: tensor<f32>) -> tensor<f32> { 1244// CHECK-DAG: %[[VAL_3:.*]] = arith.constant 1 : index 1245// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 2 : index 1246// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 0 : index 1247// CHECK-DAG: %[[VAL_6:.*]] = tensor.dim %[[VAL_0]], %[[VAL_3]] : tensor<?x?x?xf32> 1248// CHECK-DAG: %[[VAL_7:.*]] = tensor.dim %[[VAL_0]], %[[VAL_4]] : tensor<?x?x?xf32> 1249// CHECK-DAG: %[[VAL_8:.*]] = bufferization.to_memref %[[VAL_0]] : tensor<?x?x?xf32> to memref<?x?x?xf32> 1250// CHECK-DAG: %[[VAL_9:.*]] = tensor.dim %[[VAL_0]], %[[VAL_5]] : tensor<?x?x?xf32> 1251// CHECK-DAG: %[[VAL_10:.*]] = sparse_tensor.values %[[VAL_1]] : tensor<?xf32, #sparse{{[0-9]*}}> 1252// CHECK-DAG: %[[VAL_12:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<f32> to memref<f32> 1253// CHECK: %[[VAL_13:.*]] = memref.load %[[VAL_12]][] : memref<f32> 1254// CHECK: %[[VAL_14:.*]] = scf.for %[[VAL_15:.*]] = %[[VAL_5]] to %[[VAL_9]] step %[[VAL_3]] iter_args(%[[VAL_16:.*]] = %[[VAL_13]]) -> (f32) { 1255// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_15]]] : memref<?xf32> 1256// CHECK: %[[VAL_18:.*]] = scf.for %[[VAL_19:.*]] = %[[VAL_5]] to %[[VAL_6]] step %[[VAL_3]] iter_args(%[[VAL_20:.*]] = %[[VAL_16]]) -> (f32) { 1257// CHECK: %[[VAL_21:.*]] = scf.for %[[VAL_22:.*]] = %[[VAL_5]] to %[[VAL_7]] step %[[VAL_3]] iter_args(%[[VAL_23:.*]] = %[[VAL_20]]) -> (f32) { 1258// CHECK: %[[VAL_24:.*]] = memref.load %[[VAL_8]]{{\[}}%[[VAL_15]], %[[VAL_19]], %[[VAL_22]]] : memref<?x?x?xf32> 1259// CHECK: %[[VAL_25:.*]] = arith.mulf %[[VAL_24]], %[[VAL_17]] : f32 1260// CHECK: %[[VAL_26:.*]] = arith.addf %[[VAL_23]], %[[VAL_25]] : f32 1261// CHECK: scf.yield %[[VAL_26]] : f32 1262// CHECK: } 1263// CHECK: scf.yield %[[VAL_21]] : f32 1264// CHECK: } 1265// CHECK: scf.yield %[[VAL_18]] : f32 1266// CHECK: } 1267// CHECK: memref.store %[[VAL_14]], %[[VAL_12]][] : memref<f32> 1268// CHECK: %[[VAL_30:.*]] = bufferization.to_tensor %[[VAL_12]] : memref<f32> 1269// CHECK: return %[[VAL_30]] : tensor<f32> 1270// CHECK: } 1271func.func @sum_reduction_inv(%arga: tensor<?x?x?xf32>, 1272 %argb: tensor<?xf32, #Td>, 1273 %argx: tensor<f32>) -> tensor<f32> { 1274 %0 = linalg.generic #trait_sum_reduction_inv 1275 ins(%arga, %argb: tensor<?x?x?xf32>, tensor<?xf32, #Td>) 1276 outs(%argx: tensor<f32>) { 1277 ^bb(%a: f32, %b: f32, %x: f32): 1278 %0 = arith.mulf %a, %b : f32 1279 %1 = arith.addf %x, %0 : f32 1280 linalg.yield %1 : f32 1281 } -> tensor<f32> 1282 return %0 : tensor<f32> 1283} 1284 1285#trait_invariants = { 1286 indexing_maps = [ 1287 affine_map<(i,j,k) -> (i)>, // a 1288 affine_map<(i,j,k) -> (j)>, // b 1289 affine_map<(i,j,k) -> (k)>, // c 1290 affine_map<(i,j,k) -> (i,j,k)> // X (out) 1291 ], 1292 iterator_types = ["parallel", "parallel", "parallel"], 1293 doc = "X(i,j,k) = a(i) * b(j) * c(k)" 1294} 1295 1296// CHECK-LABEL: func @invariants( 1297// CHECK-SAME: %[[VAL_0:.*]]: tensor<10xf32, #sparse{{[0-9]*}}>, 1298// CHECK-SAME: %[[VAL_1:.*]]: tensor<20xf32>, 1299// CHECK-SAME: %[[VAL_2:.*]]: tensor<30xf32>, 1300// CHECK-SAME: %[[VAL_3:.*]]: tensor<10x20x30xf32>) -> tensor<10x20x30xf32> { 1301// CHECK-DAG: %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32 1302// CHECK-DAG: %[[VAL_4:.*]] = arith.constant 10 : index 1303// CHECK-DAG: %[[VAL_5:.*]] = arith.constant 20 : index 1304// CHECK-DAG: %[[VAL_6:.*]] = arith.constant 30 : index 1305// CHECK-DAG: %[[VAL_7:.*]] = arith.constant 0 : index 1306// CHECK-DAG: %[[VAL_8:.*]] = arith.constant 1 : index 1307// CHECK-DAG: %[[VAL_9:.*]] = sparse_tensor.values %[[VAL_0]] : tensor<10xf32, #sparse{{[0-9]*}}> to memref<?xf32> 1308// CHECK-DAG: %[[VAL_10:.*]] = bufferization.to_memref %[[VAL_1]] : tensor<20xf32> to memref<20xf32> 1309// CHECK-DAG: %[[VAL_11:.*]] = bufferization.to_memref %[[VAL_2]] : tensor<30xf32> to memref<30xf32> 1310// CHECK-DAG: %[[VAL_13:.*]] = bufferization.to_memref %[[VAL_3]] : tensor<10x20x30xf32> to memref<10x20x30xf32> 1311// CHECK-DAG: linalg.fill ins(%[[ZERO]] : f32) outs(%[[VAL_13]] : memref<10x20x30xf32>) 1312// CHECK: scf.for %[[VAL_14:.*]] = %[[VAL_7]] to %[[VAL_4]] step %[[VAL_8]] { 1313// CHECK: %[[VAL_15:.*]] = memref.load %[[VAL_9]]{{\[}}%[[VAL_14]]] : memref<?xf32> 1314// CHECK: scf.for %[[VAL_16:.*]] = %[[VAL_7]] to %[[VAL_5]] step %[[VAL_8]] { 1315// CHECK: %[[VAL_17:.*]] = memref.load %[[VAL_10]]{{\[}}%[[VAL_16]]] : memref<20xf32> 1316// CHECK: scf.for %[[VAL_18:.*]] = %[[VAL_7]] to %[[VAL_6]] step %[[VAL_8]] { 1317// CHECK: %[[VAL_19:.*]] = arith.mulf %[[VAL_15]], %[[VAL_17]] : f32 1318// CHECK: %[[VAL_20:.*]] = memref.load %[[VAL_11]]{{\[}}%[[VAL_18]]] : memref<30xf32> 1319// CHECK: %[[VAL_21:.*]] = arith.mulf %[[VAL_19]], %[[VAL_20]] : f32 1320// CHECK: memref.store %[[VAL_21]], %[[VAL_13]]{{\[}}%[[VAL_14]], %[[VAL_16]], %[[VAL_18]]] : memref<10x20x30xf32> 1321// CHECK: } 1322// CHECK: } 1323// CHECK: } 1324// CHECK: %[[VAL_22:.*]] = bufferization.to_tensor %[[VAL_13]] : memref<10x20x30xf32> 1325// CHECK: return %[[VAL_22]] : tensor<10x20x30xf32> 1326// CHECK: } 1327func.func @invariants(%arga: tensor<10xf32, #Td>, 1328 %argb: tensor<20xf32>, 1329 %argc: tensor<30xf32>, 1330 %argx: tensor<10x20x30xf32>) -> tensor<10x20x30xf32> { 1331 %0 = linalg.generic #trait_invariants 1332 ins(%arga, %argb, %argc : tensor<10xf32, #Td>, tensor<20xf32>, tensor<30xf32>) 1333 outs(%argx: tensor<10x20x30xf32>) { 1334 ^bb(%a: f32, %b: f32, %c: f32, %x: f32): 1335 %0 = arith.mulf %a, %b : f32 1336 %1 = arith.mulf %0, %c : f32 1337 linalg.yield %1 : f32 1338 } -> tensor<10x20x30xf32> 1339 return %0 : tensor<10x20x30xf32> 1340} 1341