brintos

brintos / llvm-project-archived public Read only

0
0
Text · 15.3 KiB · 5930a1d Raw
444 lines · plain
1// RUN: mlir-opt %s | FileCheck %s2// Verify the printed output can be parsed.3// RUN: mlir-opt %s | mlir-opt | FileCheck %s4// Verify the generic form can be parsed.5// RUN: mlir-opt -mlir-print-op-generic %s | mlir-opt | FileCheck %s6 7func.func @std_for(%arg0 : index, %arg1 : index, %arg2 : index) {8  scf.for %i0 = %arg0 to %arg1 step %arg2 {9    scf.for %i1 = %arg0 to %arg1 step %arg2 {10      %min_cmp = arith.cmpi slt, %i0, %i1 : index11      %min = arith.select %min_cmp, %i0, %i1 : index12      %max_cmp = arith.cmpi sge, %i0, %i1 : index13      %max = arith.select %max_cmp, %i0, %i1 : index14      scf.for %i2 = %min to %max step %i1 {15      }16    }17  }18  return19}20// CHECK-LABEL: func @std_for(21//  CHECK-NEXT:   scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {22//  CHECK-NEXT:     scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {23//  CHECK-NEXT:       %{{.*}} = arith.cmpi slt, %{{.*}}, %{{.*}} : index24//  CHECK-NEXT:       %{{.*}} = arith.select %{{.*}}, %{{.*}}, %{{.*}} : index25//  CHECK-NEXT:       %{{.*}} = arith.cmpi sge, %{{.*}}, %{{.*}} : index26//  CHECK-NEXT:       %{{.*}} = arith.select %{{.*}}, %{{.*}}, %{{.*}} : index27//  CHECK-NEXT:       scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} {28 29func.func @std_for_i32(%arg0 : i32, %arg1 : i32, %arg2 : i32) {30  scf.for %i0 = %arg0 to %arg1 step %arg2 : i32 {31    scf.for unsigned %i1 = %arg0 to %arg1 step %arg2 : i32 {32    }33  }34  return35}36// CHECK-LABEL: func @std_for_i32(37//  CHECK-NEXT:   scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} : i32 {38//  CHECK-NEXT:     scf.for unsigned %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} : i32 {39 40func.func @scf_for_i64_iter(%arg1: i64, %arg2: i64) {41  %c1_i64 = arith.constant 1 : i6442  %c0_i64 = arith.constant 0 : i6443  %0 = scf.for %arg3 = %arg1 to %arg2 step %c1_i64 iter_args(%arg4 = %c0_i64) -> (i64) : i64 {44    %1 = arith.addi %arg4, %arg3 : i6445    scf.yield %1 : i6446  }47  return48}49// CHECK-LABEL: scf_for_i64_iter50// CHECK: scf.for %{{.*}} = %{{.*}} to %{{.*}} step %{{.*}} iter_args(%{{.*}} = %{{.*}} -> (i64) : i64 {51 52func.func @std_if(%arg0: i1, %arg1: f32) {53  scf.if %arg0 {54    %0 = arith.addf %arg1, %arg1 : f3255  }56  return57}58// CHECK-LABEL: func @std_if(59//  CHECK-NEXT:   scf.if %{{.*}} {60//  CHECK-NEXT:     %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f3261 62func.func @std_if_else(%arg0: i1, %arg1: f32) {63  scf.if %arg0 {64    %0 = arith.addf %arg1, %arg1 : f3265  } else {66    %1 = arith.addf %arg1, %arg1 : f3267  }68  return69}70// CHECK-LABEL: func @std_if_else(71//  CHECK-NEXT:   scf.if %{{.*}} {72//  CHECK-NEXT:     %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f3273//  CHECK-NEXT:   } else {74//  CHECK-NEXT:     %{{.*}} = arith.addf %{{.*}}, %{{.*}} : f3275 76func.func @std_parallel_loop(%arg0 : index, %arg1 : index, %arg2 : index,77                        %arg3 : index, %arg4 : index) {78  %step = arith.constant 1 : index79  scf.parallel (%i0, %i1) = (%arg0, %arg1) to (%arg2, %arg3)80                                          step (%arg4, %step) {81    %min_cmp = arith.cmpi slt, %i0, %i1 : index82    %min = arith.select %min_cmp, %i0, %i1 : index83    %max_cmp = arith.cmpi sge, %i0, %i1 : index84    %max = arith.select %max_cmp, %i0, %i1 : index85    %zero = arith.constant 0.0 : f3286    %int_zero = arith.constant 0 : i3287    %red:2 = scf.parallel (%i2) = (%min) to (%max) step (%i1)88                                      init (%zero, %int_zero) -> (f32, i32) {89      %one = arith.constant 1.0 : f3290      %int_one = arith.constant 1 : i3291      scf.reduce(%one, %int_one : f32, i32)  {92        ^bb0(%lhs : f32, %rhs: f32):93          %res = arith.addf %lhs, %rhs : f3294          scf.reduce.return %res : f3295      }, {96        ^bb0(%lhs : i32, %rhs: i32):97          %res = arith.muli %lhs, %rhs : i3298          scf.reduce.return %res : i3299      }100    }101    scf.reduce102  }103  return104}105// CHECK-LABEL: func @std_parallel_loop(106//  CHECK-SAME: %[[ARG0:[A-Za-z0-9]+]]:107//  CHECK-SAME: %[[ARG1:[A-Za-z0-9]+]]:108//  CHECK-SAME: %[[ARG2:[A-Za-z0-9]+]]:109//  CHECK-SAME: %[[ARG3:[A-Za-z0-9]+]]:110//  CHECK-SAME: %[[ARG4:[A-Za-z0-9]+]]:111//       CHECK:   %[[STEP:.*]] = arith.constant 1 : index112//  CHECK-NEXT:   scf.parallel (%[[I0:.*]], %[[I1:.*]]) = (%[[ARG0]], %[[ARG1]]) to113//       CHECK:   (%[[ARG2]], %[[ARG3]]) step (%[[ARG4]], %[[STEP]]) {114//  CHECK-NEXT:     %[[MIN_CMP:.*]] = arith.cmpi slt, %[[I0]], %[[I1]] : index115//  CHECK-NEXT:     %[[MIN:.*]] = arith.select %[[MIN_CMP]], %[[I0]], %[[I1]] : index116//  CHECK-NEXT:     %[[MAX_CMP:.*]] = arith.cmpi sge, %[[I0]], %[[I1]] : index117//  CHECK-NEXT:     %[[MAX:.*]] = arith.select %[[MAX_CMP]], %[[I0]], %[[I1]] : index118//  CHECK-NEXT:     %[[ZERO:.*]] = arith.constant 0.000000e+00 : f32119//  CHECK-NEXT:     %[[INT_ZERO:.*]] = arith.constant 0 : i32120//  CHECK-NEXT:     scf.parallel (%{{.*}}) = (%[[MIN]]) to (%[[MAX]])121//  CHECK-SAME:          step (%[[I1]])122//  CHECK-SAME:          init (%[[ZERO]], %[[INT_ZERO]]) -> (f32, i32) {123//  CHECK-NEXT:       %[[ONE:.*]] = arith.constant 1.000000e+00 : f32124//  CHECK-NEXT:       %[[INT_ONE:.*]] = arith.constant 1 : i32125//  CHECK-NEXT:       scf.reduce(%[[ONE]], %[[INT_ONE]] : f32, i32) {126//  CHECK-NEXT:       ^bb0(%[[LHS:.*]]: f32, %[[RHS:.*]]: f32):127//  CHECK-NEXT:         %[[RES:.*]] = arith.addf %[[LHS]], %[[RHS]] : f32128//  CHECK-NEXT:         scf.reduce.return %[[RES]] : f32129//  CHECK-NEXT:       }, {130//  CHECK-NEXT:       ^bb0(%[[LHS:.*]]: i32, %[[RHS:.*]]: i32):131//  CHECK-NEXT:         %[[RES:.*]] = arith.muli %[[LHS]], %[[RHS]] : i32132//  CHECK-NEXT:         scf.reduce.return %[[RES]] : i32133//  CHECK-NEXT:       }134//  CHECK-NEXT:     }135//  CHECK-NEXT:     scf.reduce136 137func.func @parallel_explicit_yield(138    %arg0: index, %arg1: index, %arg2: index) {139  scf.parallel (%i0) = (%arg0) to (%arg1) step (%arg2) {140    scf.reduce141  }142  return143}144 145// CHECK-LABEL: func @parallel_explicit_yield(146//  CHECK-SAME: %[[ARG0:[A-Za-z0-9]+]]:147//  CHECK-SAME: %[[ARG1:[A-Za-z0-9]+]]:148//  CHECK-SAME: %[[ARG2:[A-Za-z0-9]+]]:149//  CHECK-NEXT: scf.parallel (%{{.*}}) = (%[[ARG0]]) to (%[[ARG1]]) step (%[[ARG2]])150//  CHECK-NEXT: scf.reduce151//  CHECK-NEXT: }152//  CHECK-NEXT: return153//  CHECK-NEXT: }154 155func.func @std_if_yield(%arg0: i1, %arg1: f32)156{157  %x, %y = scf.if %arg0 -> (f32, f32) {158    %0 = arith.addf %arg1, %arg1 : f32159    %1 = arith.subf %arg1, %arg1 : f32160    scf.yield %0, %1 : f32, f32161  } else {162    %0 = arith.subf %arg1, %arg1 : f32163    %1 = arith.addf %arg1, %arg1 : f32164    scf.yield %0, %1 : f32, f32165  }166  return167}168// CHECK-LABEL: func @std_if_yield(169//  CHECK-SAME: %[[ARG0:[A-Za-z0-9]+]]:170//  CHECK-SAME: %[[ARG1:[A-Za-z0-9]+]]:171//  CHECK-NEXT: %{{.*}}:2 = scf.if %[[ARG0]] -> (f32, f32) {172//  CHECK-NEXT: %[[T1:.*]] = arith.addf %[[ARG1]], %[[ARG1]]173//  CHECK-NEXT: %[[T2:.*]] = arith.subf %[[ARG1]], %[[ARG1]]174//  CHECK-NEXT: scf.yield %[[T1]], %[[T2]] : f32, f32175//  CHECK-NEXT: } else {176//  CHECK-NEXT: %[[T3:.*]] = arith.subf %[[ARG1]], %[[ARG1]]177//  CHECK-NEXT: %[[T4:.*]] = arith.addf %[[ARG1]], %[[ARG1]]178//  CHECK-NEXT: scf.yield %[[T3]], %[[T4]] : f32, f32179//  CHECK-NEXT: }180 181func.func @std_for_yield(%arg0 : index, %arg1 : index, %arg2 : index) {182  %s0 = arith.constant 0.0 : f32183  %result = scf.for %i0 = %arg0 to %arg1 step %arg2 iter_args(%si = %s0) -> (f32) {184    %sn = arith.addf %si, %si : f32185    scf.yield %sn : f32186  }187  return188}189// CHECK-LABEL: func @std_for_yield(190// CHECK-SAME: %[[ARG0:[A-Za-z0-9]+]]:191// CHECK-SAME: %[[ARG1:[A-Za-z0-9]+]]:192// CHECK-SAME: %[[ARG2:[A-Za-z0-9]+]]:193// CHECK-NEXT: %[[INIT:.*]] = arith.constant194// CHECK-NEXT: %{{.*}} = scf.for %{{.*}} = %[[ARG0]] to %[[ARG1]] step %[[ARG2]]195// CHECK-SAME: iter_args(%[[ITER:.*]] = %[[INIT]]) -> (f32) {196// CHECK-NEXT: %[[NEXT:.*]] = arith.addf %[[ITER]], %[[ITER]] : f32197// CHECK-NEXT: scf.yield %[[NEXT]] : f32198// CHECK-NEXT: }199 200 201func.func @std_for_yield_multi(%arg0 : index, %arg1 : index, %arg2 : index) {202  %s0 = arith.constant 0.0 : f32203  %t0 = arith.constant 1 : i32204  %u0 = arith.constant 1.0 : f32205  %result1:3 = scf.for %i0 = %arg0 to %arg1 step %arg2 iter_args(%si = %s0, %ti = %t0, %ui = %u0) -> (f32, i32, f32) {206    %sn = arith.addf %si, %si : f32207    %tn = arith.addi %ti, %ti : i32208    %un = arith.subf %ui, %ui : f32209    scf.yield %sn, %tn, %un : f32, i32, f32210  }211  return212}213// CHECK-LABEL: func @std_for_yield_multi(214// CHECK-SAME: %[[ARG0:[A-Za-z0-9]+]]:215// CHECK-SAME: %[[ARG1:[A-Za-z0-9]+]]:216// CHECK-SAME: %[[ARG2:[A-Za-z0-9]+]]:217// CHECK-NEXT: %[[INIT1:.*]] = arith.constant218// CHECK-NEXT: %[[INIT2:.*]] = arith.constant219// CHECK-NEXT: %[[INIT3:.*]] = arith.constant220// CHECK-NEXT: %{{.*}}:3 = scf.for %{{.*}} = %[[ARG0]] to %[[ARG1]] step %[[ARG2]]221// CHECK-SAME: iter_args(%[[ITER1:.*]] = %[[INIT1]], %[[ITER2:.*]] = %[[INIT2]], %[[ITER3:.*]] = %[[INIT3]]) -> (f32, i32, f32) {222// CHECK-NEXT: %[[NEXT1:.*]] = arith.addf %[[ITER1]], %[[ITER1]] : f32223// CHECK-NEXT: %[[NEXT2:.*]] = arith.addi %[[ITER2]], %[[ITER2]] : i32224// CHECK-NEXT: %[[NEXT3:.*]] = arith.subf %[[ITER3]], %[[ITER3]] : f32225// CHECK-NEXT: scf.yield %[[NEXT1]], %[[NEXT2]], %[[NEXT3]] : f32, i32, f32226 227 228func.func @conditional_reduce(%buffer: memref<1024xf32>, %lb: index, %ub: index, %step: index) -> (f32) {229  %sum_0 = arith.constant 0.0 : f32230  %c0 = arith.constant 0.0 : f32231  %sum = scf.for %iv = %lb to %ub step %step iter_args(%sum_iter = %sum_0) -> (f32) {232	  %t = memref.load %buffer[%iv] : memref<1024xf32>233	  %cond = arith.cmpf ugt, %t, %c0 : f32234	  %sum_next = scf.if %cond -> (f32) {235	    %new_sum = arith.addf %sum_iter, %t : f32236      scf.yield %new_sum : f32237	  } else {238  		scf.yield %sum_iter : f32239	  }240    scf.yield %sum_next : f32241  }242  return %sum : f32243}244// CHECK-LABEL: func @conditional_reduce(245//  CHECK-SAME: %[[ARG0:[A-Za-z0-9]+]]246//  CHECK-SAME: %[[ARG1:[A-Za-z0-9]+]]247//  CHECK-SAME: %[[ARG2:[A-Za-z0-9]+]]248//  CHECK-SAME: %[[ARG3:[A-Za-z0-9]+]]249//  CHECK-NEXT: %[[INIT:.*]] = arith.constant250//  CHECK-NEXT: %[[ZERO:.*]] = arith.constant251//  CHECK-NEXT: %[[RESULT:.*]] = scf.for %[[IV:.*]] = %[[ARG1]] to %[[ARG2]] step %[[ARG3]]252//  CHECK-SAME: iter_args(%[[ITER:.*]] = %[[INIT]]) -> (f32) {253//  CHECK-NEXT: %[[T:.*]] = memref.load %[[ARG0]][%[[IV]]]254//  CHECK-NEXT: %[[COND:.*]] = arith.cmpf ugt, %[[T]], %[[ZERO]]255//  CHECK-NEXT: %[[IFRES:.*]] = scf.if %[[COND]] -> (f32) {256//  CHECK-NEXT: %[[THENRES:.*]] = arith.addf %[[ITER]], %[[T]]257//  CHECK-NEXT: scf.yield %[[THENRES]] : f32258//  CHECK-NEXT: } else {259//  CHECK-NEXT: scf.yield %[[ITER]] : f32260//  CHECK-NEXT: }261//  CHECK-NEXT: scf.yield %[[IFRES]] : f32262//  CHECK-NEXT: }263//  CHECK-NEXT: return %[[RESULT]]264 265// CHECK-LABEL: @while266func.func @while() {267  %0 = "test.get_some_value"() : () -> i32268  %1 = "test.get_some_value"() : () -> f32269 270  // CHECK: = scf.while (%{{.*}} = %{{.*}}, %{{.*}} = %{{.*}}) : (i32, f32) -> (i64, f64) {271  %2:2 = scf.while (%arg0 = %0, %arg1 = %1) : (i32, f32) -> (i64, f64) {272    %3:2 = "test.some_operation"(%arg0, %arg1) : (i32, f32) -> (i64, f64)273    %4 = "test.some_condition"(%arg0, %arg1) : (i32, f32) -> i1274    // CHECK: scf.condition(%{{.*}}) %{{.*}}, %{{.*}} : i64, f64275    scf.condition(%4) %3#0, %3#1 : i64, f64276  // CHECK: } do {277  } do {278  // CHECK: ^{{.*}}(%{{.*}}: i64, %{{.*}}: f64):279  ^bb0(%arg2: i64, %arg3: f64):280    %5:2 = "test.some_operation"(%arg2, %arg3): (i64, f64) -> (i32, f32)281    // CHECK: scf.yield %{{.*}}, %{{.*}} : i32, f32282    scf.yield %5#0, %5#1 : i32, f32283  // CHECK: attributes {foo = "bar"}284  } attributes {foo="bar"}285  return286}287 288// CHECK-LABEL: @infinite_while289func.func @infinite_while() {290  %true = arith.constant true291 292  // CHECK: scf.while  : () -> () {293  scf.while : () -> () {294    // CHECK: scf.condition(%{{.*}})295    scf.condition(%true)296  // CHECK: } do {297  } do {298    // CHECK: scf.yield299    scf.yield300  }301  return302}303 304// CHECK-LABEL: func @execute_region305func.func @execute_region() -> i64 {306  // CHECK:      scf.execute_region -> i64 {307  // CHECK-NEXT:   arith.constant308  // CHECK-NEXT:   scf.yield309  // CHECK-NEXT: }310  %res = scf.execute_region -> i64 {311    %c1 = arith.constant 1 : i64312    scf.yield %c1 : i64313  }314 315  // CHECK:      scf.execute_region -> (i64, i64) {316  %res2:2 = scf.execute_region -> (i64, i64) {317    %c1 = arith.constant 1 : i64318    scf.yield %c1, %c1 : i64, i64319  }320 321  // CHECK:       scf.execute_region {322  // CHECK-NEXT:    cf.br ^bb1323  // CHECK-NEXT:  ^bb1:324  // CHECK-NEXT:    scf.yield325  // CHECK-NEXT:  }326  "scf.execute_region"() ({327  ^bb0:328    cf.br ^bb1329  ^bb1:330    scf.yield331  }) : () -> ()332  return %res : i64333}334 335// CHECK-LABEL: func.func @normalized_forall336func.func @normalized_forall(%in: tensor<100xf32>, %out: tensor<100xf32>) {337  %c1 = arith.constant 1 : index338  %num_threads = arith.constant 100 : index339 340  //      CHECK:    scf.forall341  // CHECK-NEXT:  tensor.extract_slice342  // CHECK-NEXT:  scf.forall.in_parallel343  // CHECK-NEXT:  tensor.parallel_insert_slice344  // CHECK-NEXT:  }345  // CHECK-NEXT:  }346  // CHECK-NEXT:  return347  %result = scf.forall (%thread_idx) in (%num_threads) shared_outs(%o = %out) -> tensor<100xf32> {348      %1 = tensor.extract_slice %in[%thread_idx][1][1] : tensor<100xf32> to tensor<1xf32>349      scf.forall.in_parallel {350        tensor.parallel_insert_slice %1 into %o[%thread_idx][1][1] :351          tensor<1xf32> into tensor<100xf32>352      }353  }354  return355}356 357// CHECK-LABEL: func.func @explicit_loop_bounds_forall358func.func @explicit_loop_bounds_forall(%in: tensor<100xf32>,359    %out: tensor<100xf32>) {360  %c0 = arith.constant 0 : index361  %c1 = arith.constant 1 : index362  %num_threads = arith.constant 100 : index363 364  //      CHECK:    scf.forall365  // CHECK-NEXT:  tensor.extract_slice366  // CHECK-NEXT:  scf.forall.in_parallel367  // CHECK-NEXT:  tensor.parallel_insert_slice368  // CHECK-NEXT:  }369  // CHECK-NEXT:  }370  // CHECK-NEXT:  return371  %result = scf.forall (%thread_idx) =  (%c0) to (%num_threads) step (%c1) shared_outs(%o = %out) -> tensor<100xf32> {372      %1 = tensor.extract_slice %in[%thread_idx][1][1] : tensor<100xf32> to tensor<1xf32>373      scf.forall.in_parallel {374        tensor.parallel_insert_slice %1 into %o[%thread_idx][1][1] :375          tensor<1xf32> into tensor<100xf32>376      }377  }378  return379}380 381// CHECK-LABEL: func.func @normalized_forall_elide_terminator382func.func @normalized_forall_elide_terminator() -> () {383  %num_threads = arith.constant 100 : index384 385  //      CHECK:    scf.forall386  // CHECK-NEXT:  } {mapping = [#gpu.thread<x>]}387  // CHECK-NEXT:  return388  scf.forall (%thread_idx) in (%num_threads) {389    scf.forall.in_parallel {390    }391  } {mapping = [#gpu.thread<x>]}392  return393 394}395 396// CHECK-LABEL: func.func @explicit_loop_bounds_forall_elide_terminator397func.func @explicit_loop_bounds_forall_elide_terminator() -> () {398  %c0 = arith.constant 0 : index399  %c1 = arith.constant 1 : index400  %num_threads = arith.constant 100 : index401 402  //      CHECK:    scf.forall403  // CHECK-NEXT:  } {mapping = [#gpu.thread<x>]}404  // CHECK-NEXT:  return405  scf.forall (%thread_idx) = (%c0) to (%num_threads) step (%c1) {406    scf.forall.in_parallel {407    }408  } {mapping = [#gpu.thread<x>]}409  return410}411 412// CHECK-LABEL: @switch413func.func @switch(%arg0: index) -> i32 {414  // CHECK: %{{.*}} = scf.index_switch %arg0 -> i32415  %0 = scf.index_switch %arg0 -> i32416  // CHECK-NEXT: case 2 {417  case 2 {418    // CHECK-NEXT: arith.constant419    %c10_i32 = arith.constant 10 : i32420    // CHECK-NEXT: scf.yield %{{.*}} : i32421    scf.yield %c10_i32 : i32422    // CHECK-NEXT: }423  }424  // CHECK-NEXT: case 5 {425  case 5 {426    %c20_i32 = arith.constant 20 : i32427    scf.yield %c20_i32 : i32428  }429  // CHECK: default {430  default {431    %c30_i32 = arith.constant 30 : i32432    scf.yield %c30_i32 : i32433  }434 435  // CHECK: scf.index_switch %arg0436  scf.index_switch %arg0437  // CHECK-NEXT: default {438  default {439    scf.yield440  }441 442  return %0 : i32443}444