393 lines · plain
1// RUN: mlir-opt %s -transform-interpreter -split-input-file | FileCheck %s2 3#map1 = affine_map<(d0) -> (d0 + 2)>4#map2 = affine_map<(d0) -> (d0 + 4)>5#map3 = affine_map<(d0) -> (d0 + 3)>6 7func.func @matmul_f32(%A: memref<?xi8>, %M: index, %N: index, %K: index) {8 %c4 = arith.constant 4 : index9 %c3 = arith.constant 3 : index10 %c2 = arith.constant 2 : index11 %c0 = arith.constant 0 : index12 %c1 = arith.constant 1 : index13 %3 = memref.view %A[%c0][%M, %K] : memref<?xi8> to memref<?x?xf32>14 %4 = memref.view %A[%c0][%K, %N] : memref<?xi8> to memref<?x?xf32>15 %5 = memref.view %A[%c0][%M, %N] : memref<?xi8> to memref<?x?xf32>16 %6 = memref.dim %3, %c0 : memref<?x?xf32>17 %7 = memref.dim %3, %c1 : memref<?x?xf32>18 %8 = memref.dim %4, %c1 : memref<?x?xf32>19 scf.for %arg4 = %c0 to %6 step %c2 {20 scf.for %arg5 = %c0 to %8 step %c3 {21 scf.for %arg6 = %c0 to %7 step %c4 {22 %11 = memref.subview %3[%arg4, %arg6][%c2, %c4][1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>23 %14 = memref.subview %4[%arg6, %arg5][%c4, %c3][1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>24 %17 = memref.subview %5[%arg4, %arg5][%c2, %c3][1, 1] : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>25 linalg.matmul26 ins(%11, %14: memref<?x?xf32, strided<[?, 1], offset: ?>>,27 memref<?x?xf32, strided<[?, 1], offset: ?>>)28 outs(%17: memref<?x?xf32, strided<[?, 1], offset: ?>>)29 }30 }31 }32 return33}34 35// CHECK-LABEL: func @matmul_f32(%{{.*}}: memref<?xi8>, %{{.*}}: index, %{{.*}}: index, %{{.*}}: index) {36// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {37// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {38// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {39// CHECK: %[[vA:.*]] = memref.subview {{.*}} : memref<?x?xf32>40// CHECK: %[[vB:.*]] = memref.subview {{.*}} : memref<?x?xf32>41// CHECK: %[[vC:.*]] = memref.subview {{.*}} : memref<?x?xf32>42///43// CHECK: %[[tmpA:.*]] = memref.alloca() : memref<32xi8>44// CHECK: %[[fullA:.*]] = memref.view %[[tmpA]][{{.*}}][{{.*}}] : memref<32xi8> to memref<?x?xf32>45// CHECK: %[[partialA:.*]] = memref.subview %[[fullA]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>46///47// CHECK: %[[tmpB:.*]] = memref.alloca() : memref<48xi8>48// CHECK: %[[fullB:.*]] = memref.view %[[tmpB]][{{.*}}][{{.*}}] : memref<48xi8> to memref<?x?xf32>49// CHECK: %[[partialB:.*]] = memref.subview %[[fullB]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>50///51// CHECK: %[[tmpC:.*]] = memref.alloca() : memref<24xi8>52// CHECK: %[[fullC:.*]] = memref.view %[[tmpC]][{{.*}}][{{.*}}] : memref<24xi8> to memref<?x?xf32>53// CHECK: %[[partialC:.*]] = memref.subview %[[fullC]]{{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1]>>54 55// CHECK: linalg.copy ins(%[[vA]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialA]] : memref<?x?xf32, strided<[?, 1]>>)56// CHECK: linalg.copy ins(%[[vB]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialB]] : memref<?x?xf32, strided<[?, 1]>>)57// CHECK: linalg.copy ins(%[[vC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>) outs(%[[partialC]] : memref<?x?xf32, strided<[?, 1]>>)58//59// CHECK: linalg.matmul ins(%[[partialA]], %[[partialB]]{{.*}} outs(%[[partialC]]60//61// CHECK: linalg.copy ins(%[[partialC]] : memref<?x?xf32, strided<[?, 1]>>) outs(%[[vC]] : memref<?x?xf32, strided<[?, 1], offset: ?>>)62//63// CHECK-NOT: memref.dealloc %[[tmpA]] : memref<32xi8>64// CHECK-NOT: memref.dealloc %[[tmpB]] : memref<48xi8>65// CHECK-NOT: memref.dealloc %[[tmpC]] : memref<24xi8>66 67module attributes {transform.with_named_sequence} {68 transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {69 %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!transform.any_op) -> !transform.any_op70 %1 = transform.structured.promote %0 { use_alloca } : (!transform.any_op) -> !transform.any_op71 transform.yield72 }73}74 75// -----76 77func.func @matmul_f64(%A: memref<?xi8>, %M: index, %N: index, %K: index) {78 %c4 = arith.constant 4 : index79 %c3 = arith.constant 3 : index80 %c2 = arith.constant 2 : index81 %c0 = arith.constant 0 : index82 %c1 = arith.constant 1 : index83 %3 = memref.view %A[%c0][%M, %K] : memref<?xi8> to memref<?x?xf64>84 %4 = memref.view %A[%c0][%K, %N] : memref<?xi8> to memref<?x?xf64>85 %5 = memref.view %A[%c0][%M, %N] : memref<?xi8> to memref<?x?xf64>86 %6 = memref.dim %3, %c0 : memref<?x?xf64>87 %7 = memref.dim %3, %c1 : memref<?x?xf64>88 %8 = memref.dim %4, %c1 : memref<?x?xf64>89 scf.for %arg4 = %c0 to %6 step %c2 {90 scf.for %arg5 = %c0 to %8 step %c3 {91 scf.for %arg6 = %c0 to %7 step %c4 {92 %11 = memref.subview %3[%arg4, %arg6][%c2, %c4][1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1], offset: ?>>93 %14 = memref.subview %4[%arg6, %arg5][%c4, %c3][1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1], offset: ?>>94 %17 = memref.subview %5[%arg4, %arg5][%c2, %c3][1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1], offset: ?>>95 linalg.matmul96 ins(%11, %14: memref<?x?xf64, strided<[?, 1], offset: ?>>,97 memref<?x?xf64, strided<[?, 1], offset: ?>>)98 outs(%17: memref<?x?xf64, strided<[?, 1], offset: ?>>)99 }100 }101 }102 return103}104 105// CHECK-LABEL: func @matmul_f64(%{{.*}}: memref<?xi8>, %{{.*}}: index, %{{.*}}: index, %{{.*}}: index) {106// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {107// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {108// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {109// CHECK: %[[vA_f64:.*]] = memref.subview {{.*}} : memref<?x?xf64>110// CHECK: %[[vB_f64:.*]] = memref.subview {{.*}} : memref<?x?xf64>111// CHECK: %[[vC_f64:.*]] = memref.subview {{.*}} : memref<?x?xf64>112///113// CHECK: %[[tmpA_f64:.*]] = memref.alloc() : memref<64xi8>114// CHECK: %[[fullA_f64:.*]] = memref.view %[[tmpA_f64]][{{.*}}][{{.*}}] : memref<64xi8> to memref<?x?xf64>115// CHECK: %[[partialA_f64:.*]] = memref.subview %[[fullA_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1]>>116///117// CHECK: %[[tmpB_f64:.*]] = memref.alloc() : memref<96xi8>118// CHECK: %[[fullB_f64:.*]] = memref.view %[[tmpB_f64]][{{.*}}][{{.*}}] : memref<96xi8> to memref<?x?xf64>119// CHECK: %[[partialB_f64:.*]] = memref.subview %[[fullB_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1]>>120///121// CHECK: %[[tmpC_f64:.*]] = memref.alloc() : memref<48xi8>122// CHECK: %[[fullC_f64:.*]] = memref.view %[[tmpC_f64]][{{.*}}][{{.*}}] : memref<48xi8> to memref<?x?xf64>123// CHECK: %[[partialC_f64:.*]] = memref.subview %[[fullC_f64]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<?x?xf64> to memref<?x?xf64, strided<[?, 1]>>124 125// CHECK: linalg.copy ins(%[[vA_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialA_f64]] : memref<?x?xf64, strided<[?, 1]>>)126// CHECK: linalg.copy ins(%[[vB_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialB_f64]] : memref<?x?xf64, strided<[?, 1]>>)127// CHECK: linalg.copy ins(%[[vC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>) outs(%[[partialC_f64]] : memref<?x?xf64, strided<[?, 1]>>)128//129// CHECK: linalg.matmul ins(%[[partialA_f64]], %[[partialB_f64]]{{.*}} outs(%[[partialC_f64]]130//131// CHECK: linalg.copy ins(%[[partialC_f64]] : memref<?x?xf64, strided<[?, 1]>>) outs(%[[vC_f64]] : memref<?x?xf64, strided<[?, 1], offset: ?>>)132//133// CHECK: memref.dealloc %[[tmpA_f64]] : memref<64xi8>134// CHECK: memref.dealloc %[[tmpB_f64]] : memref<96xi8>135// CHECK: memref.dealloc %[[tmpC_f64]] : memref<48xi8>136 137module attributes {transform.with_named_sequence} {138 transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {139 %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!transform.any_op) -> !transform.any_op140 %1 = transform.structured.promote %0 : (!transform.any_op) -> !transform.any_op141 transform.yield142 }143}144 145// -----146func.func @gemm_shared(%a : memref<?x?xf32>, %b : memref<?x?xf32>, %c : memref<?x?xf32>)147{148 linalg.matmul ins(%a, %b: memref<?x?xf32>, memref<?x?xf32>)149 outs(%c: memref<?x?xf32>)150 return151}152 153// CHECK: func @gemm_shared154// CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]+]]: memref<?x?xf32>155// CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]+]]: memref<?x?xf32>156// CHECK-SAME: %[[ARG2:[a-zA-Z0-9_]+]]: memref<?x?xf32>157// CHECK: %[[alloc_A:.*]] = memref.alloc() : memref<16x16xf32, #gpu.address_space<workgroup>>158// CHECK: %[[alloc_B:.*]] = memref.alloc() : memref<16x16xf32, #gpu.address_space<workgroup>>159// CHECK-DAG: %[[C16:.*]] = arith.constant 16160// CHECK-DAG: %[[C0:.*]] = arith.constant 0161// CHECK-DAG: %[[C1:.*]] = arith.constant 1162// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {163// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {164// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {165// CHECK: %[[subview_A:.*]] = memref.subview {{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>166// CHECK: %[[subview_B:.*]] = memref.subview {{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>167// CHECK: %[[subview_C:.*]] = memref.subview {{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>168 169// CHECK: %[[shared_A:.*]] = memref.subview %[[alloc_B]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<16x16xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<workgroup>>170// CHECK: %[[shared_B:.*]] = memref.subview %[[alloc_A]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<16x16xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<workgroup>>171 172// CHECK-NEXT: gpu.barrier173// CHECK-NEXT: memref.copy %[[subview_A]], %[[shared_A]] : memref<?x?xf32, strided<[?, 1], offset: ?>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<workgroup>>174// CHECK-NEXT: gpu.barrier175 176// CHECK-NEXT: gpu.barrier177// CHECK-NEXT: memref.copy %[[subview_B]], %[[shared_B]] : memref<?x?xf32, strided<[?, 1], offset: ?>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<workgroup>>178// CHECK-NEXT: gpu.barrier179 180// CHECK: linalg.matmul ins(%[[shared_A]], %[[shared_B]]{{.*}} outs(%[[subview_C]]181 182 183module attributes {transform.with_named_sequence} {184 transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {185 %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!transform.any_op) -> !transform.any_op186 %1, %loops:3 = transform.structured.tile_using_for %0 tile_sizes [16, 16, 16] : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op)187 %2 = transform.structured.promote %1 { operands_to_promote = [0, 1], mapping = [#gpu.memory_space<workgroup>] } : (!transform.any_op) -> !transform.any_op188 transform.yield189 }190}191 192 193// -----194 195func.func @gemm_private(%a : memref<?x?xf32>, %b : memref<?x?xf32>, %c : memref<?x?xf32>)196{197 linalg.matmul ins(%a, %b: memref<?x?xf32>, memref<?x?xf32>)198 outs(%c: memref<?x?xf32>)199 return200}201 202// CHECK: func @gemm_private203// CHECK-SAME: %[[ARG0:[a-zA-Z0-9_]+]]: memref<?x?xf32>204// CHECK-SAME: %[[ARG1:[a-zA-Z0-9_]+]]: memref<?x?xf32>205// CHECK-SAME: %[[ARG2:[a-zA-Z0-9_]+]]: memref<?x?xf32>206// CHECK: %[[alloc_A:.*]] = memref.alloca() : memref<16x16xf32, #gpu.address_space<private>>207// CHECK: %[[alloc_B:.*]] = memref.alloca() : memref<16x16xf32, #gpu.address_space<private>>208// CHECK-DAG: %[[C16:.*]] = arith.constant 16209// CHECK-DAG: %[[C0:.*]] = arith.constant 0210// CHECK-DAG: %[[C1:.*]] = arith.constant 1211// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {212// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {213// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {214// CHECK: %[[subview_A:.*]] = memref.subview {{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>215// CHECK: %[[subview_B:.*]] = memref.subview {{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>216// CHECK: %[[subview_C:.*]] = memref.subview {{.*}} : memref<?x?xf32> to memref<?x?xf32, strided<[?, 1], offset: ?>>217 218// CHECK: %[[private_A:.*]] = memref.subview %[[alloc_B]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<16x16xf32, #gpu.address_space<private>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<private>>219// CHECK: %[[private_B:.*]] = memref.subview %[[alloc_A]][0, 0] [%{{.*}}, %{{.*}}] [1, 1] : memref<16x16xf32, #gpu.address_space<private>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<private>>220 221// CHECK-NEXT: memref.copy %[[subview_A]], %[[private_A]] : memref<?x?xf32, strided<[?, 1], offset: ?>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<private>>222// CHECK-NEXT: memref.copy %[[subview_B]], %[[private_B]] : memref<?x?xf32, strided<[?, 1], offset: ?>> to memref<?x?xf32, strided<[16, 1]>, #gpu.address_space<private>>223 224// CHECK: linalg.matmul ins(%[[private_A]], %[[private_B]]{{.*}} outs(%[[subview_C]]225 226 227module attributes {transform.with_named_sequence} {228 transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {229 %0 = transform.structured.match ops{["linalg.matmul"]} in %arg1 : (!transform.any_op) -> !transform.any_op230 %1, %loops:3 = transform.structured.tile_using_for %0 tile_sizes [16, 16, 16] : (!transform.any_op) -> (!transform.any_op, !transform.any_op, !transform.any_op, !transform.any_op)231 %2 = transform.structured.promote %1 { operands_to_promote = [0, 1], mapping = [#gpu.memory_space<private>] } : (!transform.any_op) -> !transform.any_op232 transform.yield233 }234}235 236 237// -----238 239#map6 = affine_map<(d0, d1, d2) -> (d0, d2)>240#map7 = affine_map<(d0, d1, d2) -> (d1, d2)>241#map8 = affine_map<(d0, d1, d2) -> (d0, d1)>242 243// CHECK: promote_rank_reducing_subviews(%[[arg0:.+]]: memref<{{.*}}>, %[[arg1:.+]]: memref<{{.*}}>, %[[arg2:.+]]: memref<{{.*}}>, %[[lb1:.+]]: index, %[[lb2:.+]]: index, %[[lb3:.+]]: index, %[[lb4:.+]]: index, %[[lb5:.+]]: index, %[[lb6:.+]]: index, %[[ub1:.+]]: index, %[[ub2:.+]]: index244func.func @promote_rank_reducing_subviews(%arg0: memref<?x?x?x64xf32, strided<[?, ?, ?, ?], offset: ?>>, %arg1: memref<128x3x3x64xf32, strided<[?, ?, ?, ?], offset: ?>>, %arg2: memref<?x?x?x128xf32>,245 %arg3: index, %arg4: index, %arg5: index, %arg6: index, %arg7: index, %arg8: index, %ub1: index, %ub2: index) {246 %13 = memref.subview %arg0[%arg3, 0, %arg4, %arg8] [1, 1, %ub1, 32] [1, 1, 1, 1] : memref<?x?x?x64xf32, strided<[?, ?, ?, ?], offset: ?>> to memref<?x32xf32, strided<[?, ?], offset: ?>>247 %14 = memref.subview %arg1[0, %arg6, %arg7, %arg8] [128, 1, 1, 32] [1, 1, 1, 1] : memref<128x3x3x64xf32, strided<[?, ?, ?, ?], offset: ?>> to memref<128x32xf32, strided<[?, ?], offset: ?>>248 %9 = memref.subview %arg2[%arg3, %arg4, %arg5, 0] [1, 1, %ub2, 128] [1, 1, 1, 1] : memref<?x?x?x128xf32> to memref<?x128xf32, strided<[128, 1], offset: ?>>249 250 // CHECK: %[[a_alloc:.+]] = memref.alloc251 // CHECK: %[[a_view:.+]] = memref.view %[[a_alloc]]{{.*}}252 // CHECK: %[[a_pro_subview:.+]] = memref.subview %[[a_view]][0, 0] [%[[ub1]], {{.+}}] [1, 1]253 254 // CHECK: memref.alloc255 // CHECK: %[[b_view:.+]] = memref.view256 // CHECK: %[[b_pro_subview:.+]] = memref.subview %[[b_view]]257 258 // CHECK: memref.alloc259 // CHECK: %[[c_view:.+]] = memref.view260 // CHECK: %[[c_pro_subview:.+]] = memref.subview %[[c_view]]261 262 // CHECK-COUNT-3: linalg.copy263 // CHECK: linalg.generic264 // CHECK-SAME: ins(%[[a_pro_subview]], %[[b_pro_subview]]265 // CHECK-SAME: outs(%[[c_pro_subview]]266 267 linalg.generic {indexing_maps = [#map6, #map7, #map8], iterator_types = ["parallel", "parallel", "reduction"]} ins(%13, %14 : memref<?x32xf32, strided<[?, ?], offset: ?>>, memref<128x32xf32, strided<[?, ?], offset: ?>>) outs(%9 : memref<?x128xf32, strided<[128, 1], offset: ?>>) {268 ^bb0(%arg9: f32, %arg10: f32, %arg11: f32):269 %15 = arith.mulf %arg9, %arg10 : f32270 %16 = arith.addf %arg11, %15 : f32271 linalg.yield %16 : f32272 }273 274 return275}276 277module attributes {transform.with_named_sequence} {278 transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {279 %0 = transform.structured.match interface{LinalgOp} in %arg1 : (!transform.any_op) -> !transform.any_op280 %1 = transform.structured.promote %0 : (!transform.any_op) -> !transform.any_op281 transform.yield282 }283}284 285// -----286 287#map = affine_map<(d0, d1) -> (d0, d1)>288 289 // CHECK-LABEL: func.func @linalg_generic_update_all_function_inputs_outputs(290 // CHECK-SAME: %[[VAL_0:.*]]: memref<8x4xf32, 1>,291 // CHECK-SAME: %[[VAL_1:.*]]: memref<8x4xf32, 1>) -> memref<8x4xf32, 1> {292func.func @linalg_generic_update_all_function_inputs_outputs(%arg0: memref<8x4xf32, 1>, %arg1: memref<8x4xf32, 1>) -> memref<8x4xf32, 1> {293 // CHECK: %[[VAL_2:.*]] = memref.alloc() {alignment = 64 : i64} : memref<8x4xf32, 1>294 // CHECK: %[[VAL_3:.*]] = memref.subview %[[VAL_0]][0, 0] [4, 3] [1, 1] : memref<8x4xf32, 1> to memref<4x3xf32, strided<[4, 1]>, 1>295 // CHECK: %[[VAL_4:.*]] = memref.subview %[[VAL_1]][0, 0] [4, 3] [1, 1] : memref<8x4xf32, 1> to memref<4x3xf32, strided<[4, 1]>, 1>296 // CHECK: %[[VAL_5:.*]] = memref.subview %[[VAL_2]][0, 0] [4, 3] [1, 1] : memref<8x4xf32, 1> to memref<4x3xf32, strided<[4, 1]>, 1>297 298 %alloc = memref.alloc() {alignment = 64 : i64} : memref<8x4xf32, 1>299 %subview = memref.subview %arg0[0, 0] [4, 3] [1, 1] : memref<8x4xf32, 1> to memref<4x3xf32, strided<[4, 1]>, 1>300 %subview_0 = memref.subview %arg1[0, 0] [4, 3] [1, 1] : memref<8x4xf32, 1> to memref<4x3xf32, strided<[4, 1]>, 1>301 %subview_1 = memref.subview %alloc[0, 0] [4, 3] [1, 1] : memref<8x4xf32, 1> to memref<4x3xf32, strided<[4, 1]>, 1>302 303 // CHECK: %[[VAL_6:.*]] = arith.constant 0 : index304 // CHECK: %[[VAL_7:.*]] = arith.constant 4 : index305 // CHECK: %[[VAL_8:.*]] = arith.constant 1 : index306 // CHECK: %[[VAL_9:.*]] = arith.constant 0 : index307 // CHECK: %[[VAL_10:.*]] = arith.constant 3 : index308 // CHECK: %[[VAL_11:.*]] = arith.constant 1 : index309 // CHECK: %[[VAL_12:.*]] = arith.constant 4 : index310 // CHECK: %[[VAL_13:.*]] = arith.constant 0 : index311 // CHECK: %[[VAL_14:.*]] = arith.constant 4 : index312 // CHECK: %[[VAL_15:.*]] = arith.constant 3 : index313 // CHECK: %[[VAL_16:.*]] = arith.constant 1 : index314 // CHECK: %[[VAL_17:.*]] = arith.constant 3 : index315 // CHECK: %[[VAL_18:.*]] = arith.constant 0 : index316 // CHECK: %[[VAL_19:.*]] = arith.constant 1 : index317 // CHECK: %[[VAL_20:.*]] = arith.constant 4 : index318 // CHECK: %[[VAL_21:.*]] = arith.constant 12 : index319 // CHECK: %[[VAL_22:.*]] = memref.alloc() : memref<48xi8, #gpu.address_space<workgroup>>320 // CHECK: %[[VAL_23:.*]] = memref.view %[[VAL_22]]{{\[}}%[[VAL_18]]]{{\[}}%[[VAL_12]], %[[VAL_15]]] : memref<48xi8, #gpu.address_space<workgroup>> to memref<?x?xf32, #gpu.address_space<workgroup>>321 // CHECK: %[[VAL_24:.*]] = memref.subview %[[VAL_23]][0, 0] {{\[}}%[[VAL_14]], %[[VAL_17]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>322 // CHECK: %[[VAL_25:.*]] = arith.constant 0 : index323 // CHECK: %[[VAL_26:.*]] = arith.constant 4 : index324 // CHECK: %[[VAL_27:.*]] = arith.constant 1 : index325 // CHECK: %[[VAL_28:.*]] = arith.constant 0 : index326 // CHECK: %[[VAL_29:.*]] = arith.constant 3 : index327 // CHECK: %[[VAL_30:.*]] = arith.constant 1 : index328 // CHECK: %[[VAL_31:.*]] = arith.constant 4 : index329 // CHECK: %[[VAL_32:.*]] = arith.constant 0 : index330 // CHECK: %[[VAL_33:.*]] = arith.constant 4 : index331 // CHECK: %[[VAL_34:.*]] = arith.constant 3 : index332 // CHECK: %[[VAL_35:.*]] = arith.constant 1 : index333 // CHECK: %[[VAL_36:.*]] = arith.constant 3 : index334 // CHECK: %[[VAL_37:.*]] = arith.constant 0 : index335 // CHECK: %[[VAL_38:.*]] = arith.constant 1 : index336 // CHECK: %[[VAL_39:.*]] = arith.constant 4 : index337 // CHECK: %[[VAL_40:.*]] = arith.constant 12 : index338 // CHECK: %[[VAL_41:.*]] = memref.alloc() : memref<48xi8, #gpu.address_space<workgroup>>339 // CHECK: %[[VAL_42:.*]] = memref.view %[[VAL_41]]{{\[}}%[[VAL_37]]]{{\[}}%[[VAL_31]], %[[VAL_34]]] : memref<48xi8, #gpu.address_space<workgroup>> to memref<?x?xf32, #gpu.address_space<workgroup>>340 // CHECK: %[[VAL_43:.*]] = memref.subview %[[VAL_42]][0, 0] {{\[}}%[[VAL_33]], %[[VAL_36]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>341 // CHECK: %[[VAL_44:.*]] = arith.constant 0 : index342 // CHECK: %[[VAL_45:.*]] = arith.constant 4 : index343 // CHECK: %[[VAL_46:.*]] = arith.constant 1 : index344 // CHECK: %[[VAL_47:.*]] = arith.constant 0 : index345 // CHECK: %[[VAL_48:.*]] = arith.constant 3 : index346 // CHECK: %[[VAL_49:.*]] = arith.constant 1 : index347 // CHECK: %[[VAL_50:.*]] = arith.constant 4 : index348 // CHECK: %[[VAL_51:.*]] = arith.constant 0 : index349 // CHECK: %[[VAL_52:.*]] = arith.constant 4 : index350 // CHECK: %[[VAL_53:.*]] = arith.constant 3 : index351 // CHECK: %[[VAL_54:.*]] = arith.constant 1 : index352 // CHECK: %[[VAL_55:.*]] = arith.constant 3 : index353 // CHECK: %[[VAL_56:.*]] = arith.constant 0 : index354 // CHECK: %[[VAL_57:.*]] = arith.constant 1 : index355 // CHECK: %[[VAL_58:.*]] = arith.constant 4 : index356 // CHECK: %[[VAL_59:.*]] = arith.constant 12 : index357 // CHECK: %[[VAL_60:.*]] = memref.alloc() : memref<48xi8, #gpu.address_space<workgroup>>358 // CHECK: %[[VAL_61:.*]] = memref.view %[[VAL_60]]{{\[}}%[[VAL_56]]]{{\[}}%[[VAL_50]], %[[VAL_53]]] : memref<48xi8, #gpu.address_space<workgroup>> to memref<?x?xf32, #gpu.address_space<workgroup>>359 // CHECK: %[[VAL_62:.*]] = memref.subview %[[VAL_61]][0, 0] {{\[}}%[[VAL_52]], %[[VAL_55]]] [1, 1] : memref<?x?xf32, #gpu.address_space<workgroup>> to memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>360// CHECK: linalg.copy ins(%[[VAL_3]] : memref<4x3xf32, strided<[4, 1]>, 1>) outs(%[[VAL_24]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>)361// CHECK: linalg.copy ins(%[[VAL_4]] : memref<4x3xf32, strided<[4, 1]>, 1>) outs(%[[VAL_43]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>)362 // CHECK: linalg.generic {doc = "", indexing_maps = [#map, #map, #map], iterator_types = ["parallel", "parallel"], library_call = ""} ins(%[[VAL_24]], %[[VAL_43]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>, memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>) outs(%[[VAL_62]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>) {363 // CHECK: ^bb0(%[[VAL_63:.*]]: f32, %[[VAL_64:.*]]: f32, %[[VAL_65:.*]]: f32):364 // CHECK: %[[VAL_66:.*]] = arith.addf %[[VAL_63]], %[[VAL_64]] : f32365 // CHECK: linalg.yield %[[VAL_66]] : f32366 // CHECK: }367 368 369 linalg.generic {doc = "", indexing_maps = [#map, #map, #map], iterator_types = ["parallel", "parallel"], library_call = ""} ins(%subview, %subview_0 : memref<4x3xf32, strided<[4, 1]>, 1>, memref<4x3xf32, strided<[4, 1]>, 1>) outs(%subview_1 : memref<4x3xf32, strided<[4, 1]>, 1>) {370 ^bb0(%in: f32, %in_1: f32, %out: f32):371 %1 = arith.addf %in, %in_1 : f32372 linalg.yield %1 : f32373 }374 375 // CHECK: linalg.copy ins(%[[VAL_62]] : memref<?x?xf32, strided<[?, 1]>, #gpu.address_space<workgroup>>) outs(%[[VAL_5]] : memref<4x3xf32, strided<[4, 1]>, 1>)376 // CHECK: memref.dealloc %[[VAL_22]] : memref<48xi8, #gpu.address_space<workgroup>>377 // CHECK: memref.dealloc %[[VAL_41]] : memref<48xi8, #gpu.address_space<workgroup>>378 // CHECK: memref.dealloc %[[VAL_60]] : memref<48xi8, #gpu.address_space<workgroup>>379 // CHECK: return %[[VAL_2]] : memref<8x4xf32, 1>380 // CHECK: }381 382 return %alloc : memref<8x4xf32, 1>383}384 385 386module attributes {transform.with_named_sequence} {387 transform.named_sequence @__transform_main(%arg1: !transform.any_op {transform.readonly}) {388 %0 = transform.structured.match ops{["linalg.generic"]} in %arg1 : (!transform.any_op) -> !transform.any_op389 %1 = transform.structured.promote %0 { memory_space = #gpu.address_space<workgroup> } : (!transform.any_op) -> !transform.any_op390 transform.yield391 }392}393