1// RUN: mlir-opt %s -async-parallel-for \ 2// RUN: -async-to-async-runtime \ 3// RUN: -async-runtime-ref-counting \ 4// RUN: -async-runtime-ref-counting-opt \ 5// RUN: -arith-expand \ 6// RUN: -convert-async-to-llvm \ 7// RUN: -convert-scf-to-cf \ 8// RUN: -convert-memref-to-llvm \ 9// RUN: -convert-func-to-llvm \ 10// RUN: -reconcile-unrealized-casts \ 11// RUN: | mlir-cpu-runner \ 12// RUN: -e entry -entry-point-result=void -O0 \ 13// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_runner_utils%shlibext \ 14// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_async_runtime%shlibext\ 15// RUN: | FileCheck %s --dump-input=always 16 17// RUN: mlir-opt %s -async-parallel-for \ 18// RUN: -async-to-async-runtime \ 19// RUN: -async-runtime-policy-based-ref-counting \ 20// RUN: -arith-expand \ 21// RUN: -convert-async-to-llvm \ 22// RUN: -convert-scf-to-cf \ 23// RUN: -convert-memref-to-llvm \ 24// RUN: -convert-func-to-llvm \ 25// RUN: -reconcile-unrealized-casts \ 26// RUN: | mlir-cpu-runner \ 27// RUN: -e entry -entry-point-result=void -O0 \ 28// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_runner_utils%shlibext \ 29// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_async_runtime%shlibext\ 30// RUN: | FileCheck %s --dump-input=always 31 32// RUN: mlir-opt %s -async-parallel-for="async-dispatch=false \ 33// RUN: num-workers=20 \ 34// RUN: min-task-size=1" \ 35// RUN: -async-to-async-runtime \ 36// RUN: -async-runtime-ref-counting \ 37// RUN: -async-runtime-ref-counting-opt \ 38// RUN: -arith-expand \ 39// RUN: -convert-async-to-llvm \ 40// RUN: -convert-scf-to-cf \ 41// RUN: -convert-memref-to-llvm \ 42// RUN: -convert-func-to-llvm \ 43// RUN: -reconcile-unrealized-casts \ 44// RUN: | mlir-cpu-runner \ 45// RUN: -e entry -entry-point-result=void -O0 \ 46// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_runner_utils%shlibext \ 47// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_async_runtime%shlibext\ 48// RUN: | FileCheck %s --dump-input=always 49 50func.func @entry() { 51 %c0 = arith.constant 0.0 : f32 52 %c1 = arith.constant 1 : index 53 %c2 = arith.constant 2 : index 54 %c8 = arith.constant 8 : index 55 56 %lb = arith.constant 0 : index 57 %ub = arith.constant 8 : index 58 59 %A = memref.alloc() : memref<8x8xf32> 60 %U = memref.cast %A : memref<8x8xf32> to memref<*xf32> 61 62 // Initialize memref with zeros because we do load and store to in every test 63 // to verify that we process each element of the iteration space once. 64 scf.parallel (%i, %j) = (%lb, %lb) to (%ub, %ub) step (%c1, %c1) { 65 memref.store %c0, %A[%i, %j] : memref<8x8xf32> 66 } 67 68 // 1. (%i, %i) = (0, 8) to (8, 8) step (1, 1) 69 scf.parallel (%i, %j) = (%lb, %lb) to (%ub, %ub) step (%c1, %c1) { 70 %0 = arith.muli %i, %c8 : index 71 %1 = arith.addi %j, %0 : index 72 %2 = arith.index_cast %1 : index to i32 73 %3 = arith.sitofp %2 : i32 to f32 74 %4 = memref.load %A[%i, %j] : memref<8x8xf32> 75 %5 = arith.addf %3, %4 : f32 76 memref.store %5, %A[%i, %j] : memref<8x8xf32> 77 } 78 79 // CHECK: [0, 1, 2, 3, 4, 5, 6, 7] 80 // CHECK-NEXT: [8, 9, 10, 11, 12, 13, 14, 15] 81 // CHECK-NEXT: [16, 17, 18, 19, 20, 21, 22, 23] 82 // CHECK-NEXT: [24, 25, 26, 27, 28, 29, 30, 31] 83 // CHECK-NEXT: [32, 33, 34, 35, 36, 37, 38, 39] 84 // CHECK-NEXT: [40, 41, 42, 43, 44, 45, 46, 47] 85 // CHECK-NEXT: [48, 49, 50, 51, 52, 53, 54, 55] 86 // CHECK-NEXT: [56, 57, 58, 59, 60, 61, 62, 63] 87 call @printMemrefF32(%U): (memref<*xf32>) -> () 88 89 scf.parallel (%i, %j) = (%lb, %lb) to (%ub, %ub) step (%c1, %c1) { 90 memref.store %c0, %A[%i, %j] : memref<8x8xf32> 91 } 92 93 // 2. (%i, %i) = (0, 8) to (8, 8) step (2, 1) 94 scf.parallel (%i, %j) = (%lb, %lb) to (%ub, %ub) step (%c2, %c1) { 95 %0 = arith.muli %i, %c8 : index 96 %1 = arith.addi %j, %0 : index 97 %2 = arith.index_cast %1 : index to i32 98 %3 = arith.sitofp %2 : i32 to f32 99 %4 = memref.load %A[%i, %j] : memref<8x8xf32> 100 %5 = arith.addf %3, %4 : f32 101 memref.store %5, %A[%i, %j] : memref<8x8xf32> 102 } 103 104 // CHECK: [0, 1, 2, 3, 4, 5, 6, 7] 105 // CHECK-NEXT: [0, 0, 0, 0, 0, 0, 0, 0] 106 // CHECK-NEXT: [16, 17, 18, 19, 20, 21, 22, 23] 107 // CHECK-NEXT: [0, 0, 0, 0, 0, 0, 0, 0] 108 // CHECK-NEXT: [32, 33, 34, 35, 36, 37, 38, 39] 109 // CHECK-NEXT: [0, 0, 0, 0, 0, 0, 0, 0] 110 // CHECK-NEXT: [48, 49, 50, 51, 52, 53, 54, 55] 111 // CHECK-NEXT: [0, 0, 0, 0, 0, 0, 0, 0] 112 call @printMemrefF32(%U): (memref<*xf32>) -> () 113 114 scf.parallel (%i, %j) = (%lb, %lb) to (%ub, %ub) step (%c1, %c1) { 115 memref.store %c0, %A[%i, %j] : memref<8x8xf32> 116 } 117 118 // 3. (%i, %i) = (0, 8) to (8, 8) step (1, 2) 119 scf.parallel (%i, %j) = (%lb, %lb) to (%ub, %ub) step (%c1, %c2) { 120 %0 = arith.muli %i, %c8 : index 121 %1 = arith.addi %j, %0 : index 122 %2 = arith.index_cast %1 : index to i32 123 %3 = arith.sitofp %2 : i32 to f32 124 %4 = memref.load %A[%i, %j] : memref<8x8xf32> 125 %5 = arith.addf %3, %4 : f32 126 memref.store %5, %A[%i, %j] : memref<8x8xf32> 127 } 128 129 // CHECK: [0, 0, 2, 0, 4, 0, 6, 0] 130 // CHECK-NEXT: [8, 0, 10, 0, 12, 0, 14, 0] 131 // CHECK-NEXT: [16, 0, 18, 0, 20, 0, 22, 0] 132 // CHECK-NEXT: [24, 0, 26, 0, 28, 0, 30, 0] 133 // CHECK-NEXT: [32, 0, 34, 0, 36, 0, 38, 0] 134 // CHECK-NEXT: [40, 0, 42, 0, 44, 0, 46, 0] 135 // CHECK-NEXT: [48, 0, 50, 0, 52, 0, 54, 0] 136 // CHECK-NEXT: [56, 0, 58, 0, 60, 0, 62, 0] 137 call @printMemrefF32(%U): (memref<*xf32>) -> () 138 139 memref.dealloc %A : memref<8x8xf32> 140 141 return 142} 143 144func.func private @printMemrefF32(memref<*xf32>) attributes { llvm.emit_c_interface } 145