1// RUN: mlir-opt %s -test-linalg-elementwise-fusion-patterns=fuse-generic-ops-control -split-input-file | FileCheck %s 2 3#map0 = affine_map<(d0, d1) -> (d0, d1)> 4#binary2Dpointwise = { 5 indexing_maps = [#map0, #map0, #map0], 6 iterator_types = ["parallel", "parallel"] 7} 8#ternary2Dpointwise = { 9 indexing_maps = [#map0, #map0, #map0, #map0], 10 iterator_types = ["parallel", "parallel"] 11} 12func.func @test_fusion_limit( 13 %arg0 : tensor<?x?xf32>, %arg1 : tensor<?x?xf32>, %arg2 : tensor<?x?xf32>, 14 %arg3 : tensor<?x?xf32>, %arg4 : tensor<?x?xf32>, %arg5 : tensor<?x?xf32>) 15 -> tensor<?x?xf32> { 16 %c0 = arith.constant 0 : index 17 %c1 = arith.constant 1 : index 18 %d0 = tensor.dim %arg0, %c0 : tensor<?x?xf32> 19 %d1 = tensor.dim %arg0, %c1 : tensor<?x?xf32> 20 %init = tensor.empty(%d0, %d1) : tensor<?x?xf32> 21 %0 = linalg.generic #binary2Dpointwise 22 ins(%arg0, %arg1 : tensor<?x?xf32>, tensor<?x?xf32>) 23 outs(%init : tensor<?x?xf32>) { 24 ^bb0(%arg6 : f32, %arg7 : f32, %arg8 : f32): 25 %1 = arith.mulf %arg6, %arg7 : f32 26 linalg.yield %1 : f32 27 } -> tensor<?x?xf32> 28 %2 = linalg.generic #binary2Dpointwise 29 ins(%arg2, %arg3 : tensor<?x?xf32>, tensor<?x?xf32>) 30 outs(%init : tensor<?x?xf32>) { 31 ^bb0(%arg6 : f32, %arg7 : f32, %arg8 : f32): 32 %3 = arith.mulf %arg6, %arg7 : f32 33 linalg.yield %3 : f32 34 } -> tensor<?x?xf32> 35 %4 = linalg.generic #binary2Dpointwise 36 ins(%arg4, %arg5 : tensor<?x?xf32>, tensor<?x?xf32>) 37 outs(%init : tensor<?x?xf32>) { 38 ^bb0(%arg6 : f32, %arg7 : f32, %arg8 : f32): 39 %5 = arith.mulf %arg6, %arg7 : f32 40 linalg.yield %5 : f32 41 } -> tensor<?x?xf32> 42 %6 = linalg.generic #ternary2Dpointwise 43 ins(%0, %2, %4 : tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>) 44 outs(%init : tensor<?x?xf32>) { 45 ^bb0(%arg6 : f32, %arg7 : f32, %arg8 : f32, %arg9 : f32): 46 %7 = arith.addf %arg6, %arg7 : f32 47 %8 = arith.addf %7, %arg8 : f32 48 linalg.yield %8 : f32 49 } -> tensor<?x?xf32> 50 return %6 : tensor<?x?xf32> 51} 52 53// CHECK-LABEL: func @test_fusion_limit 54// CHECK-SAME: %[[ARG0:[a-zA-z0-9_]+]]: tensor<?x?xf32> 55// CHECK-SAME: %[[ARG1:[a-zA-z0-9_]+]]: tensor<?x?xf32> 56// CHECK-SAME: %[[ARG2:[a-zA-z0-9_]+]]: tensor<?x?xf32> 57// CHECK-SAME: %[[ARG3:[a-zA-z0-9_]+]]: tensor<?x?xf32> 58// CHECK-SAME: %[[ARG4:[a-zA-z0-9_]+]]: tensor<?x?xf32> 59// CHECK-SAME: %[[ARG5:[a-zA-z0-9_]+]]: tensor<?x?xf32> 60// CHECK: %[[OP1:.+]] = linalg.generic {{.+}} ins(%[[ARG2]], %[[ARG3]] 61// CHECK: %[[OP2:.+]] = linalg.generic {{.+}} ins(%[[ARG4]], %[[ARG5]] 62// CHECK: %[[OP3:.+]] = linalg.generic {{.+}} ins(%[[ARG0]], %[[ARG1]], %[[OP1]], %[[OP2]] 63// CHECK: return %[[OP3]] 64