35 lines · plain
1// RUN: mlir-opt -test-linalg-elementwise-fusion-patterns=fuse-multiuse-producer -split-input-file %s | FileCheck %s2 3#map = affine_map<(d0, d1) -> (d0, d1)>4func.func @multi_use_producer(%arg0 : tensor<?x?xf32>, %arg1 : tensor<?x?xf32>,5 %arg2 : tensor<?x?xf32>, %arg3 : tensor<?x?xf32>, %arg4 : tensor<?x?xf32>)6 -> (tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>) {7 %0:2 = linalg.generic {8 indexing_maps = [#map, #map, #map],9 iterator_types = ["parallel", "parallel"]}10 ins(%arg0 : tensor<?x?xf32>)11 outs(%arg1, %arg2 : tensor<?x?xf32>, tensor<?x?xf32>) {12 ^bb0(%b0: f32, %b1 : f32, %b2 : f32):13 %1 = arith.addf %b0, %b1 : f3214 linalg.yield %1, %1 : f32, f3215 } -> (tensor<?x?xf32>, tensor<?x?xf32>)16 %2 = linalg.generic {17 indexing_maps = [#map, #map, #map],18 iterator_types = ["parallel", "parallel"]}19 ins(%0#1, %arg3 : tensor<?x?xf32>, tensor<?x?xf32>)20 outs(%arg4 : tensor<?x?xf32>) {21 ^bb0(%b0 : f32, %b1 : f32, %b2 : f32):22 %3 = arith.mulf %b0, %b1 : f3223 linalg.yield %3 : f3224 } -> tensor<?x?xf32>25 return %0#0, %0#1, %2 : tensor<?x?xf32>, tensor<?x?xf32>, tensor<?x?xf32>26}27// CHECK: func @multi_use_producer(28// CHECK-SAME: %[[ARG0:[a-zA-Z0-9]+]]: tensor<?x?xf32>29// CHECK-SAME: %[[ARG1:[a-zA-Z0-9]+]]: tensor<?x?xf32>30// CHECK-SAME: %[[ARG2:[a-zA-Z0-9]+]]: tensor<?x?xf32>31// CHECK-SAME: %[[ARG3:[a-zA-Z0-9]+]]: tensor<?x?xf32>32// CHECK-SAME: %[[ARG4:[a-zA-Z0-9]+]]: tensor<?x?xf32>)33// CHECK: %[[RESULT:.+]]:3 = linalg.generic34// CHECK: return %[[RESULT]]#0, %[[RESULT]]#1, %[[RESULT]]#235