1// RUN: mlir-opt %s -test-linalg-greedy-fusion -split-input-file | FileCheck %s 2 3#map0 = affine_map<(d0)[s0] -> (2, -d0 + s0)> 4#map1 = affine_map<(d0)[s0] -> (4, -d0 + s0)> 5#map2 = affine_map<(d0)[s0] -> (3, -d0 + s0)> 6#map3 = affine_map<(d0, d1) -> (2, d0 - d1)> 7#map4 = affine_map<(d0, d1) -> (3, d0 - d1)> 8 9func @matmul_tensors(%arg0: tensor<?x?xf32>, %arg1: tensor<?x?xf32>, %arg2: tensor<?x?xf32>) -> tensor<?x?xf32> { 10 %t0 = linalg.matmul ins(%arg0, %arg1: tensor<?x?xf32>, tensor<?x?xf32>) 11 init(%arg2: tensor<?x?xf32>) 12 -> tensor<?x?xf32> 13 14 %c4 = constant 4 : index 15 %c2 = constant 2 : index 16 %c0 = constant 0 : index 17 %c3 = constant 3 : index 18 %c1 = constant 1 : index 19 %0 = dim %t0, %c0 : tensor<?x?xf32> 20 %1 = dim %t0, %c1 : tensor<?x?xf32> 21 %2 = dim %arg1, %c1 : tensor<?x?xf32> 22 %3 = scf.for %arg3 = %c0 to %0 step %c2 iter_args(%arg4 = %arg2) -> (tensor<?x?xf32>) { 23 %4 = scf.for %arg5 = %c0 to %2 step %c3 iter_args(%arg6 = %arg4) -> (tensor<?x?xf32>) { 24 %5 = scf.for %arg7 = %c0 to %1 step %c4 iter_args(%arg8 = %arg6) -> (tensor<?x?xf32>) { 25 %6 = subtensor %t0[%arg3, %arg7][%c2, 4][1, 1] : tensor<?x?xf32> to tensor<?x4xf32> 26 %7 = subtensor %arg1[%arg7, %arg5][4, %c3][1, 1] : tensor<?x?xf32> to tensor<4x?xf32> 27 %8 = subtensor %arg8[%arg3, %arg5][%c2, %c3][1, 1] : tensor<?x?xf32> to tensor<?x?xf32> 28 %9 = linalg.matmul ins(%6, %7 : tensor<?x4xf32>, tensor<4x?xf32>) init(%8 : tensor<?x?xf32>) -> tensor<?x?xf32> 29 %10 = subtensor_insert %9 into %arg8[%arg3, %arg5] [%c2, %c3] [1, 1] : tensor<?x?xf32> into tensor<?x?xf32> 30 scf.yield %10 : tensor<?x?xf32> 31 } 32 scf.yield %5 : tensor<?x?xf32> 33 } 34 scf.yield %4 : tensor<?x?xf32> 35 } 36 return %3 : tensor<?x?xf32> 37} 38 39// CHECK-LABEL: func @matmul_tensors( 40// CHECK-SAME: %[[A:[0-9a-z]*]]: tensor<?x?xf32> 41// CHECK-SAME: %[[B:[0-9a-z]*]]: tensor<?x?xf32> 42// CHECK-SAME: %[[C:[0-9a-z]*]]: tensor<?x?xf32> 43// CHECK-DAG: %[[C0:.*]] = constant 0 : index 44// CHECK-DAG: %[[C1:.*]] = constant 1 : index 45// CHECK-DAG: %[[dA1:.*]] = dim %[[A]], %[[C1]] : tensor<?x?xf32> 46// CHECK: scf.for %[[I:[0-9a-z]*]] 47// CHECK: %[[stA:.*]] = subtensor %[[A]][%[[I]], 0] [2, %[[dA1]]] [1, 1] : tensor<?x?xf32> to tensor<2x?xf32> 48// CHECK-NEXT: scf.for %[[J:[0-9a-z]*]] 49// CHECK-NEXT: scf.for %[[K:[0-9a-z]*]] {{.*}} iter_args(%[[RES:[0-9a-z]*]] 50// CHECK-DAG: %[[stB1:.*]] = subtensor %[[B]][%[[K]], %[[J]]] [4, 3] [1, 1] : tensor<?x?xf32> to tensor<4x3xf32> 51// CHECK-DAG: %[[stF:.*]] = subtensor %[[RES]][%[[I]], %[[J]]] [2, 3] [1, 1] : tensor<?x?xf32> to tensor<2x3xf32> 52// 53// subtensors of the producing matmul. 54// CHECK-DAG: %[[stB2:.*]] = subtensor %[[B]][0, %[[K]]] [%[[dA1]], 4] [1, 1] : tensor<?x?xf32> to tensor<?x4xf32> 55// CHECK-DAG: %[[stC:.*]] = subtensor %[[C]][%[[I]], %[[K]]] [2, 4] [1, 1] : tensor<?x?xf32> to tensor<2x4xf32> 56// CHECK: %[[stD:.*]] = linalg.matmul ins(%[[stA]], %[[stB2]] : tensor<2x?xf32>, tensor<?x4xf32>) init(%[[stC]] : tensor<2x4xf32>) -> tensor<2x4xf32> 57// CHECK-NEXT: %[[stG:.*]] = linalg.matmul ins(%[[stD]], %[[stB1]] : tensor<2x4xf32>, tensor<4x3xf32>) init(%[[stF]] : tensor<2x3xf32>) -> tensor<2x3xf32> 58// CHECK-NEXT: subtensor_insert %[[stG]] into %[[RES]][%[[I]], %[[J]]] 59