brintos

brintos / llvm-project-archived public Read only

0
0
Text · 17.2 KiB · 745a82f Raw
352 lines · plain
1// RUN: mlir-opt  -transform-interpreter -split-input-file --cse %s | FileCheck %s2 3func.func @simple_matmul(%arg0 : tensor<?x?xf32>, %arg1 : tensor<?x?xf32>,4    %arg2 : tensor<?x?xf32>) -> tensor<?x?xf32> {5  %0 = linalg.matmul6    ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)7      outs(%arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>8  return %0 : tensor<?x?xf32>9}10 11module attributes {transform.with_named_sequence} {12  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {13    %matmul = transform.structured.match ops{["linalg.matmul"]} in %arg114      : (!transform.any_op) -> !transform.any_op15    %a, %b = transform.test.tile_using_forall %matmul [10, 20] mapping = [#gpu.block<y>, #gpu.block<x>]16      : (!transform.any_op) -> (!transform.any_op, !transform.any_op)17    transform.yield18  }19}20//  CHECK-DAG: #[[MAP0:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 10)>21//  CHECK-DAG: #[[MAP1:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 20)>22//      CHECK: func.func @simple_matmul(23// CHECK-SAME:     %[[ARG0:[a-zA-Z0-9]+]]: tensor<?x?xf32>24// CHECK-SAME:     %[[ARG1:[a-zA-Z0-9]+]]: tensor<?x?xf32>25// CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]+]]: tensor<?x?xf32>26//  CHECK-DAG:   %[[C0:.+]] = arith.constant 0 : index27//  CHECK-DAG:   %[[C1:.+]] = arith.constant 1 : index28//  CHECK-DAG:   %[[M:.+]] = tensor.dim %[[ARG0]], %[[C0]]29//  CHECK-DAG:   %[[K:.+]] = tensor.dim %[[ARG0]], %[[C1]]30//  CHECK-DAG:   %[[N:.+]] = tensor.dim %[[ARG1]], %[[C1]]31//      CHECK:   %[[RESULT:.+]] = scf.forall (%[[IV0:[a-zA-Z0-9]+]], %[[IV1:[a-zA-Z0-9]+]]) =32// CHECK-SAME:       (0, 0) to (%[[M]], %[[N]]) step (10, 20) shared_outs(%[[INIT:.+]] = %[[ARG2]])33//      CHECK:     %[[TS_Y:.+]] = affine.min #[[MAP0]](%[[IV0]])[%[[M]]]34//      CHECK:     %[[TS_X:.+]] = affine.min #[[MAP1]](%[[IV1]])[%[[N]]]35//      CHECK:     %[[LHS_TILE:.+]] = tensor.extract_slice %[[ARG0]]36// CHECK-SAME:         [%[[IV0]], 0] [%[[TS_Y]], %[[K]]] [1, 1]37//      CHECK:     %[[RHS_TILE:.+]] = tensor.extract_slice %[[ARG1]]38// CHECK-SAME:         [0, %[[IV1]]] [%[[K]], %[[TS_X]]] [1, 1]39//      CHECK:     %[[INIT_TILE:.+]] = tensor.extract_slice %[[INIT]]40// CHECK-SAME:         [%[[IV0]], %[[IV1]]] [%[[TS_Y]], %[[TS_X]]] [1, 1]41//      CHECK:     %[[GEMM_TILE:.+]] = linalg.matmul42// CHECK-SAME:         ins(%[[LHS_TILE]], %[[RHS_TILE]] :43// CHECK-SAME:         outs(%[[INIT_TILE]] :44//      CHECK:     scf.forall.in_parallel {45//      CHECK:       tensor.parallel_insert_slice %[[GEMM_TILE]] into %[[INIT]]46// CHECK-SAME:           [%[[IV0]], %[[IV1]]] [%[[TS_Y]], %[[TS_X]]] [1, 1]47//      CHECK:       mapping = [#gpu.block<y>, #gpu.block<x>]48//      CHECK:   return %[[RESULT]]49 50// -----51 52func.func @simple_matmul_memref(%arg0 : memref<?x?xf32>, %arg1 : memref<?x?xf32>,53    %arg2 : memref<?x?xf32>) {54  linalg.matmul ins(%arg0, %arg1 : memref<?x?xf32>, memref<?x?xf32>)55      outs(%arg2 : memref<?x?xf32>)56  return57}58 59module attributes {transform.with_named_sequence} {60  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {61    %matmul = transform.structured.match ops{["linalg.matmul"]} in %arg162      : (!transform.any_op) -> !transform.any_op63    %a, %b = transform.test.tile_using_forall %matmul [10, 20]64      : (!transform.any_op) -> (!transform.any_op, !transform.any_op)65    transform.yield66  }67}68//  CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 10)>69//  CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 20)>70//      CHECK-LABEL: func.func @simple_matmul_memref(71// CHECK-SAME:     %[[ARG0:[a-zA-Z0-9]+]]: memref<?x?xf32>72// CHECK-SAME:     %[[ARG1:[a-zA-Z0-9]+]]: memref<?x?xf32>73// CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]+]]: memref<?x?xf32>74//  CHECK-DAG:   %[[C0:.+]] = arith.constant 0 : index75//  CHECK-DAG:   %[[C1:.+]] = arith.constant 1 : index76//  CHECK-DAG:   %[[M:.+]] = memref.dim %[[ARG0]], %[[C0]]77//  CHECK-DAG:   %[[K:.+]] = memref.dim %[[ARG0]], %[[C1]]78//  CHECK-DAG:   %[[N:.+]] = memref.dim %[[ARG1]], %[[C1]]79//      CHECK:   scf.forall (%[[IV0:[a-zA-Z0-9]+]], %[[IV1:[a-zA-Z0-9]+]]) = (0, 0) to (%[[M]], %[[N]]) step (10, 20) {80//  CHECK-DAG:     %[[TS_M:.+]] = affine.min #[[$MAP0]](%[[IV0]])[%[[M]]]81//  CHECK-DAG:     %[[TS_N:.+]] = affine.min #[[$MAP1]](%[[IV1]])[%[[N]]]82//  CHECK-DAG:     %[[LHS_TILE:.+]] = memref.subview %[[ARG0]]83// CHECK-SAME:         [%[[IV0]], 0] [%[[TS_M]], %[[K]]] [1, 1]84//  CHECK-DAG:     %[[RHS_TILE:.+]] = memref.subview %[[ARG1]]85// CHECK-SAME:         [0, %[[IV1]]] [%[[K]], %[[TS_N]]] [1, 1]86//  CHECK-DAG:     %[[OUT_TILE:.+]] = memref.subview %[[ARG2]]87// CHECK-SAME:         [%[[IV0]], %[[IV1]]] [%[[TS_M]], %[[TS_N]]] [1, 1]88//      CHECK:     linalg.matmul89// CHECK-SAME:             ins(%[[LHS_TILE]], %[[RHS_TILE]] :90// CHECK-SAME:             outs(%[[OUT_TILE]] :91 92// -----93 94#map0 = affine_map<(d0, d1, d2) -> (d0, d1, d2)>95#map1 = affine_map<(d0, d1, d2) -> (d0, d2, d1)>96#map2 = affine_map<(d0, d1, d2) -> (d2, d0, d1)>97func.func @multi_result(%arg0 : tensor<128x200x300xf32>) -> (tensor<128x300x200xf32>, tensor<300x128x200xf32>) {98  %init0 = tensor.empty() : tensor<128x300x200xf32>99  %init1 = tensor.empty() : tensor<300x128x200xf32>100  %0:2 = linalg.generic {101      indexing_maps = [#map0, #map1, #map2],102      iterator_types = ["parallel", "parallel", "parallel"]}103      ins(%arg0 : tensor<128x200x300xf32>)104      outs(%init0, %init1 : tensor<128x300x200xf32>, tensor<300x128x200xf32>) {105    ^bb0(%b0 : f32, %b1 : f32, %b2 : f32):106      linalg.yield %b0, %b0 : f32, f32107    } -> (tensor<128x300x200xf32>, tensor<300x128x200xf32>)108  return %0#0, %0#1 : tensor<128x300x200xf32>, tensor<300x128x200xf32>109}110 111module attributes {transform.with_named_sequence} {112  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {113    %generic = transform.structured.match ops{["linalg.generic"]} in %arg1114      : (!transform.any_op) -> !transform.any_op115    %a, %b = transform.test.tile_using_forall %generic [10, 0, 20]116      : (!transform.any_op) -> (!transform.any_op, !transform.any_op)117    transform.yield118  }119}120//  CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0) -> (-d0 + 128, 10)>121//      CHECK-LABEL: func.func @multi_result(122// CHECK-SAME:     %[[ARG0:[a-zA-Z0-9]+]]: tensor<128x200x300xf32>)123//  CHECK-DAG:   %[[INIT0:.+]] = tensor.empty()124//  CHECK-DAG:   %[[INIT1:.+]] = tensor.empty()125//      CHECK:   %[[OUTER:[a-zA-Z0-9]+]]:2 = scf.forall (%[[IV0:[a-zA-Z0-9]+]], %[[IV1:[a-zA-Z0-9]+]]) = (0, 0) to (128, 300) step (10, 20)126// CHECK-SAME:       shared_outs(%[[ARG1:[a-zA-Z0-9]+]] = %[[INIT0]], %[[ARG2:[a-zA-Z0-9]+]] = %[[INIT1]])127//      CHECK:     %[[TS_Y:.+]] = affine.min #[[$MAP0]](%[[IV0]])128//      CHECK:     %[[ARG_TILE:.+]] = tensor.extract_slice %[[ARG0]]129// CHECK-SAME:         [%[[IV0]], 0, %[[IV1]]] [%[[TS_Y]], 200, 20] [1, 1, 1]130//  CHECK-DAG:     %[[INIT0_TILE:.+]] = tensor.extract_slice %[[ARG1]]131// CHECK-SAME:         [%[[IV0]], %[[IV1]], 0] [%[[TS_Y]], 20, 200] [1, 1, 1]132//  CHECK-DAG:     %[[INIT1_TILE:.+]] = tensor.extract_slice %[[ARG2]]133// CHECK-SAME:         [%[[IV1]], %[[IV0]], 0] [20, %[[TS_Y]], 200] [1, 1, 1]134//      CHECK:     %[[RESULT_TILE:.+]]:2 = linalg.generic135// CHECK-SAME:         ins(%[[ARG_TILE]] :136// CHECK-SAME:         outs(%[[INIT0_TILE]], %[[INIT1_TILE]] :137//      CHECK:     scf.forall.in_parallel {138//  CHECK-DAG:       tensor.parallel_insert_slice %[[RESULT_TILE]]#0 into %[[ARG1]][%[[IV0]], %[[IV1]], 0] [%[[TS_Y]], 20, 200] [1, 1, 1]139//  CHECK-DAG:       tensor.parallel_insert_slice %[[RESULT_TILE]]#1 into %[[ARG2]][%[[IV1]], %[[IV0]], 0] [20, %[[TS_Y]], 200] [1, 1, 1]140//      CHECK:     }141//      CHECK:   return %[[OUTER]]#0, %[[OUTER]]#1142 143// -----144 145func.func @conv2D(%arg0 : tensor<?x?x?x?xf32>, %arg1 : tensor<?x?x?x?xf32>,146    %arg2 : tensor<?x?x?x?xf32>) -> tensor<?x?x?x?xf32> {147  %0 = linalg.conv_2d_nhwc_hwcf {148      strides = dense<[2, 3]> : tensor<2xi64>,149      dilation = dense<[4, 5]> : tensor<2xi64>}150      ins(%arg0, %arg1 : tensor<?x?x?x?xf32>, tensor<?x?x?x?xf32>)151      outs(%arg2 : tensor<?x?x?x?xf32>) -> tensor<?x?x?x?xf32>152  return %0 : tensor<?x?x?x?xf32>153}154 155module attributes {transform.with_named_sequence} {156  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {157    %conv = transform.structured.match ops{["linalg.conv_2d_nhwc_hwcf"]} in %arg1158      : (!transform.any_op) -> !transform.any_op159    %a, %b = transform.test.tile_using_forall %conv [0, 0, 0, 0, 10, 20, 30]160      : (!transform.any_op) -> (!transform.any_op, !transform.any_op)161    transform.yield162  }163}164//  CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 10)>165//  CHECK-DAG: #[[$MAP1:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 20)>166//  CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 30)>167//  CHECK-DAG: #[[$MAP3:.+]] = affine_map<(d0)[s0] -> (d0 + s0 * 2 - 2)>168//  CHECK-DAG: #[[$MAP4:.+]] = affine_map<(d0)[s0] -> (d0 + s0 * 3 - 3)>169//      CHECK-LABEL: func.func @conv2D(170// CHECK-SAME:     %[[INPUT:[a-zA-Z0-9]+]]: tensor<?x?x?x?xf32>171// CHECK-SAME:     %[[FILTER:[a-zA-Z0-9]+]]: tensor<?x?x?x?xf32>172// CHECK-SAME:     %[[INIT:[a-zA-Z0-9]+]]: tensor<?x?x?x?xf32>173//  CHECK-DAG:   %[[C0:.+]] = arith.constant 0 : index174//  CHECK-DAG:   %[[C1:.+]] = arith.constant 1 : index175//  CHECK-DAG:   %[[C2:.+]] = arith.constant 2 : index176//  CHECK-DAG:   %[[C3:.+]] = arith.constant 3 : index177//  CHECK-DAG:   %[[N:.+]] = tensor.dim %[[INPUT]], %[[C0]]178//  CHECK-DAG:   %[[C:.+]] = tensor.dim %[[INPUT]], %[[C3]]179//  CHECK-DAG:   %[[P:.+]] = tensor.dim %[[FILTER]], %[[C0]]180//  CHECK-DAG:   %[[Q:.+]] = tensor.dim %[[FILTER]], %[[C1]]181//  CHECK-DAG:   %[[F:.+]] = tensor.dim %[[FILTER]], %[[C3]]182//  CHECK-DAG:   %[[R:.+]] = tensor.dim %[[INIT]], %[[C1]]183//  CHECK-DAG:   %[[S:.+]] = tensor.dim %[[INIT]], %[[C2]]184//      CHECK:   %[[RESULT:.+]] = scf.forall (%[[IV0:[a-zA-Z0-9]+]], %[[IV1:[a-zA-Z0-9]+]], %[[IV2:[a-zA-Z0-9]+]]) =185// CHECK-SAME:       (0, 0, 0) to (%[[P]], %[[Q]], %[[C]]) step (10, 20, 30) shared_outs(%[[INIT0:.+]] = %[[INIT]])186//  CHECK-DAG:     %[[TS_P:.+]] = affine.min #[[$MAP0]](%[[IV0]])[%[[P]]]187//  CHECK-DAG:     %[[TS_Q:.+]] = affine.min #[[$MAP1]](%[[IV1]])[%[[Q]]]188//  CHECK-DAG:     %[[TS_C:.+]] = affine.min #[[$MAP2]](%[[IV2]])[%[[C]]]189//  CHECK-DAG:     %[[TS_H:.+]] = affine.apply #[[$MAP3]](%[[TS_P]])[%[[R]]]190//  CHECK-DAG:     %[[TS_W:.+]] = affine.apply #[[$MAP4]](%[[TS_Q]])[%[[S]]]191//  CHECK-DAG:     %[[INPUT_TILE:.+]] = tensor.extract_slice %[[INPUT]]192// CHECK-SAME:         [0, %[[IV0]], %[[IV1]], %[[IV2]]] [%[[N]], %[[TS_H]], %[[TS_W]], %[[TS_C]]]193//  CHECK-DAG:     %[[FILTER_TILE:.+]] = tensor.extract_slice %[[FILTER]]194// CHECK-SAME:         [%[[IV0]], %[[IV1]], %[[IV2]], 0] [%[[TS_P]], %[[TS_Q]], %[[TS_C]], %[[F]]]195//  CHECK-DAG:     %[[INIT_TILE:.+]] = tensor.extract_slice %[[INIT0]]196// CHECK-SAME:         [0, 0, 0, 0] [%[[N]], %[[R]], %[[S]], %[[F]]]197//      CHECK:     %[[CONV_TILE:.+]] = linalg.conv_2d_nhwc_hwcf198// CHECK-SAME:         dilation = dense<[4, 5]> : tensor<2xi64>, strides = dense<[2, 3]> : tensor<2xi64>199// CHECK-SAME:         ins(%[[INPUT_TILE]], %[[FILTER_TILE]] :200// CHECK-SAME:         outs(%[[INIT_TILE]] :201//      CHECK:     scf.forall.in_parallel202//      CHECK:       tensor.parallel_insert_slice %[[CONV_TILE]] into %[[INIT0]]203// CHECK-SAME:           [0, 0, 0, 0] [%[[N]], %[[R]], %[[S]], %[[F]]] [1, 1, 1, 1]204//      CHECK:   return %[[RESULT]]205 206// -----207 208// CHECK: #[[$MAP_ADD:.+]] = affine_map<(d0)[s0] -> (d0 + s0)>209 210func.func @indexed_semantics(%arg0: tensor<?x?xf32>, %arg1: tensor<?x?xf32>) -> tensor<?x?xf32> {211  // Check that we correctly amend "linalg.index" results.212 213  %0 = linalg.generic {214    indexing_maps = [affine_map<(d0, d1) -> (d0, d1)>,215                     affine_map<(d0, d1) -> (d0, d1)>],216    iterator_types = ["parallel", "parallel"]}217    ins(%arg0: tensor<?x?xf32>)218    outs(%arg1: tensor<?x?xf32>) {219  ^bb0(%arg2: f32, %arg3: f32):220    %1 = linalg.index 0 : index221    %2 = linalg.index 1 : index222    %3 = arith.addi %1, %2 : index223    %4 = arith.index_cast %3 : index to i64224    %5 = arith.uitofp %4 : i64 to f32225    %6 = arith.addf %5, %arg2 : f32226    linalg.yield %6 : f32227  } -> (tensor<?x?xf32>)228  return %0 : tensor<?x?xf32>229}230 231module attributes {transform.with_named_sequence} {232  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {233    %generic = transform.structured.match ops{["linalg.generic"]} in %arg1234      : (!transform.any_op) -> !transform.any_op235    %a, %b = transform.test.tile_using_forall %generic [10, 20]236      : (!transform.any_op) -> (!transform.any_op, !transform.any_op)237    transform.yield238  }239}240 241// CHECK-LABEL: @indexed_semantics242//       CHECK: scf.forall (%[[I0:.+]], %[[I1:.+]]) =243//       CHECK:   %[[INDEX0:.+]] = linalg.index 0244//       CHECK:   %[[INDEX0_AMENDED:.+]] = affine.apply #[[$MAP_ADD]](%[[I0]])[%[[INDEX0]]]245//       CHECK:   %[[INDEX1:.+]] = linalg.index 1246//       CHECK:   %[[INDEX1_AMENDED:.+]] = affine.apply #[[$MAP_ADD]](%[[I1]])[%[[INDEX1]]]247//       CHECK:   arith.addi %[[INDEX0_AMENDED]], %[[INDEX1_AMENDED]]248 249// -----250 251func.func @interchange_matmul(%arg0 : tensor<?x?xf32>, %arg1 : tensor<?x?xf32>,252    %arg2 : tensor<?x?xf32>) -> tensor<?x?xf32> {253  %0 = linalg.matmul ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>)254      outs(%arg2 : tensor<?x?xf32>) -> tensor<?x?xf32>255  return %0 : tensor<?x?xf32>256}257 258module attributes {transform.with_named_sequence} {259  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {260    %matmul = transform.structured.match ops{["linalg.matmul"]} in %arg1261      : (!transform.any_op) -> !transform.any_op262    %a, %b = transform.test.tile_using_forall %matmul [10, 20] interchange = [1, 0] mapping = [#gpu.block<y>, #gpu.block<x>]263      : (!transform.any_op) -> (!transform.any_op, !transform.any_op)264    transform.yield265  }266}267//  CHECK-DAG: #[[$MAP0:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 20)>268//  CHECK-DAG: #[[$MAP2:.+]] = affine_map<(d0)[s0] -> (-d0 + s0, 10)>269//      CHECK-LABEL: func.func @interchange_matmul(270// CHECK-SAME:     %[[ARG0:[a-zA-Z0-9]+]]: tensor<?x?xf32>271// CHECK-SAME:     %[[ARG1:[a-zA-Z0-9]+]]: tensor<?x?xf32>272// CHECK-SAME:     %[[ARG2:[a-zA-Z0-9]+]]: tensor<?x?xf32>273//  CHECK-DAG:   %[[C0:.+]] = arith.constant 0 : index274//  CHECK-DAG:   %[[C1:.+]] = arith.constant 1 : index275//  CHECK-DAG:   %[[M:.+]] = tensor.dim %[[ARG0]], %[[C0]]276//  CHECK-DAG:   %[[K:.+]] = tensor.dim %[[ARG0]], %[[C1]]277//  CHECK-DAG:   %[[N:.+]] = tensor.dim %[[ARG1]], %[[C1]]278//      CHECK:   %[[OUTER:[a-zA-Z0-9]+]] = scf.forall (%[[IV0:[a-zA-Z0-9]+]], %[[IV1:[a-zA-Z0-9]+]]279// CHECK-SAME:       (0, 0) to (%[[N]], %[[M]]) step (20, 10)280// CHECK-SAME:       shared_outs(%[[INIT0:.+]] = %[[ARG2]])281//  CHECK-DAG:     %[[TS_N:.+]] = affine.min #[[$MAP0]](%[[IV0]])[%[[N]]]282//  CHECK-DAG:     %[[TS_M:.+]] = affine.min #[[$MAP2]](%[[IV1]])[%[[M]]]283//  CHECK-DAG:     %[[LHS_TILE:.+]] = tensor.extract_slice %[[ARG0]]284// CHECK-SAME:         [%[[IV1]], 0] [%[[TS_M]], %[[K]]] [1, 1]285//  CHECK-DAG:     %[[RHS_TILE:.+]] = tensor.extract_slice %[[ARG1]]286// CHECK-SAME:         [0, %[[IV0]]] [%[[K]], %[[TS_N]]] [1, 1]287//  CHECK-DAG:     %[[INIT_TILE:.+]] = tensor.extract_slice %[[INIT0]]288// CHECK-SAME:         [%[[IV1]], %[[IV0]]] [%[[TS_M]], %[[TS_N]]] [1, 1]289//      CHECK:     %[[GEMM_TILE:.+]] = linalg.matmul290// CHECK-SAME:         ins(%[[LHS_TILE]], %[[RHS_TILE]] :291// CHECK-SAME:         outs(%[[INIT_TILE]] :292//      CHECK:     scf.forall.in_parallel {293//      CHECK:       tensor.parallel_insert_slice %[[GEMM_TILE]] into %[[INIT0]]294// CHECK-SAME:           [%[[IV1]], %[[IV0]]] [%[[TS_M]], %[[TS_N]]] [1, 1]295//      CHECK:     } {mapping = [#gpu.block<y>, #gpu.block<x>]}296//      CHECK:   return %[[OUTER]]297 298// -----299 300func.func @check_scalar_operation(%arg0 : tensor<f32>) -> tensor<f32> {301  %init = tensor.empty() : tensor<f32>302  %0 = linalg.generic {303      indexing_maps = [affine_map<() -> ()>, affine_map<() -> ()>],304      iterator_types = []}      305      ins(%arg0 : tensor<f32>) outs(%init : tensor<f32>){306    ^bb0(%b0 : f32, %b1 : f32):307      %1 = arith.mulf %b0, %b0 : f32308      linalg.yield %1 : f32309  } -> tensor<f32>310  return %0 : tensor<f32>311}312 313module attributes {transform.with_named_sequence} {314  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {315    %generic = transform.structured.match ops{["linalg.generic"]} in %arg1316      : (!transform.any_op) -> !transform.any_op317    %a = transform.test.tile_using_forall %generic []318      : (!transform.any_op) -> (!transform.any_op)319    transform.yield320  }321}322// CHECK-LABEL: func @check_scalar_operation323//   CHECK-NOT:   scf.for324//       CHECK:   linalg.generic325 326// -----327 328func.func @check_scalar_memref_operation(%arg0 : memref<f32>, %arg1 : memref<f32>){329  linalg.generic {330      indexing_maps = [affine_map<() -> ()>, affine_map<() -> ()>],331      iterator_types = []}      332      ins(%arg0 : memref<f32>) outs(%arg1 : memref<f32>){333    ^bb0(%b0 : f32, %b1 : f32):334      %1 = arith.mulf %b0, %b0 : f32335      linalg.yield %1 : f32336  }337  return338}339 340module attributes {transform.with_named_sequence} {341  transform.named_sequence @__transform_main(%arg1 : !transform.any_op {transform.readonly}) {342    %generic = transform.structured.match ops{["linalg.generic"]} in %arg1343      : (!transform.any_op) -> !transform.any_op344    %a = transform.test.tile_using_forall %generic []345      : (!transform.any_op) -> (!transform.any_op)346    transform.yield347  }348}349// CHECK-LABEL: func @check_scalar_memref_operation350//   CHECK-NOT:   scf.for351//       CHECK:   linalg.generic352