64 lines · plain
1// RUN: mlir-opt -lower-vector-multi-reduction="lowering-strategy=inner-reduction" -split-input-file %s | FileCheck %s --check-prefixes=ALL,INNER-REDUCTION2// RUN: mlir-opt -lower-vector-multi-reduction="lowering-strategy=inner-parallel" -split-input-file %s | FileCheck %s --check-prefixes=ALL,INNER-PARALLEL3// RUN: mlir-opt -lower-vector-multi-reduction -split-input-file %s | FileCheck %s --check-prefixes=ALL,INNER-PARALLEL4 5func.func @vector_multi_reduction(%arg0: vector<2x4xf32>, %acc: vector<2xf32>) -> vector<2xf32> {6 %0 = vector.multi_reduction <mul>, %arg0, %acc [1] : vector<2x4xf32> to vector<2xf32>7 return %0 : vector<2xf32>8}9// ALL-LABEL: func @vector_multi_reduction10// ALL-SAME: %[[INPUT:.+]]: vector<2x4xf32>, %[[ACC:.*]]: vector<2xf32>)11// INNER-REDUCTION-DAG: %[[RESULT_VEC_0:.+]] = arith.constant dense<{{.*}}> : vector<2xf32>12// INNER-REDUCTION: %[[V0:.+]] = vector.extract %[[INPUT]][0]13// INNER-REDUCTION: %[[ACC0:.+]] = vector.extract %[[ACC]][0]14// INNER-REDUCTION: %[[RV0:.+]] = vector.reduction <mul>, %[[V0]], %[[ACC0]] : vector<4xf32> into f3215// INNER-REDUCTION: %[[RESULT_VEC_1:.+]] = vector.insert %[[RV0:.+]], %[[RESULT_VEC_0]] [0] : f32 into vector<2xf32>16// INNER-REDUCTION: %[[V1:.+]] = vector.extract %[[INPUT]][1]17// INNER-REDUCTION: %[[ACC1:.+]] = vector.extract %[[ACC]][1]18// INNER-REDUCTION: %[[RV1:.+]] = vector.reduction <mul>, %[[V1]], %[[ACC1]] : vector<4xf32> into f3219// INNER-REDUCTION: %[[RESULT_VEC:.+]] = vector.insert %[[RV1:.+]], %[[RESULT_VEC_1]] [1] : f32 into vector<2xf32>20// INNER-REDUCTION: return %[[RESULT_VEC]]21 22// INNER-PARALLEL: %[[TRANSPOSED:.+]] = vector.transpose %[[INPUT]], [1, 0] : vector<2x4xf32> to vector<4x2xf32>23// INNER-PARALLEL: %[[V0:.+]] = vector.extract %[[TRANSPOSED]][0] : vector<2xf32> from vector<4x2xf32>24// INNER-PARALLEL: %[[RV0:.+]] = arith.mulf %[[V0]], %[[ACC]] : vector<2xf32>25// INNER-PARALLEL: %[[V1:.+]] = vector.extract %[[TRANSPOSED]][1] : vector<2xf32> from vector<4x2xf32>26// INNER-PARALLEL: %[[RV01:.+]] = arith.mulf %[[V1]], %[[RV0]] : vector<2xf32>27// INNER-PARALLEL: %[[V2:.+]] = vector.extract %[[TRANSPOSED]][2] : vector<2xf32> from vector<4x2xf32>28// INNER-PARALLEL: %[[RV012:.+]] = arith.mulf %[[V2]], %[[RV01]] : vector<2xf32>29// INNER-PARALLEL: %[[V3:.+]] = vector.extract %[[TRANSPOSED]][3] : vector<2xf32> from vector<4x2xf32>30// INNER-PARALLEL: %[[RESULT_VEC:.+]] = arith.mulf %[[V3]], %[[RV012]] : vector<2xf32>31// INNER-PARALLEL: return %[[RESULT_VEC]] : vector<2xf32>32 33// -----34 35func.func @vector_multi_reduction_parallel_middle(%arg0: vector<3x4x5xf32>, %acc: vector<4xf32>) -> vector<4xf32> {36 %0 = vector.multi_reduction <add>, %arg0, %acc [0, 2] : vector<3x4x5xf32> to vector<4xf32>37 return %0 : vector<4xf32>38}39 40// ALL-LABEL: func @vector_multi_reduction_parallel_middle41// ALL-SAME: %[[INPUT:.+]]: vector<3x4x5xf32>, %[[ACC:.+]]: vector<4xf32>42// INNER-REDUCTION: vector.transpose %[[INPUT]], [1, 0, 2] : vector<3x4x5xf32> to vector<4x3x5xf32>43// INNER-PARALLEL: vector.transpose %[[INPUT]], [0, 2, 1] : vector<3x4x5xf32> to vector<3x5x4xf32>44 45// -----46 47func.func @vector_multi_reduction_masked(%arg0: vector<2x4xf32>, %acc: vector<2xf32>, %mask: vector<2x4xi1>) -> vector<2xf32> {48 %0 = vector.mask %mask { vector.multi_reduction <mul>, %arg0, %acc [1] : vector<2x4xf32> to vector<2xf32> } : vector<2x4xi1> -> vector<2xf32>49 return %0 : vector<2xf32>50}51 52// ALL-LABEL: func @vector_multi_reduction_masked53// ALL-SAME: %[[INPUT:.+]]: vector<2x4xf32>, %[[ACC:.+]]: vector<2xf32>, %[[MASK:.+]]: vector<2x4xi1>54// INNER-REDUCTION: %[[INNERVEC:.+]] = vector.extract %[[INPUT]][0] : vector<4xf32> from vector<2x4xf32>55// INNER-REDUCTION: %[[INNERACC:.+]] = vector.extract %[[ACC]][0] : f32 from vector<2xf32>56// INNER-REDUCTION: %[[INNERMASK:.+]] = vector.extract %[[MASK]][0] : vector<4xi1> from vector<2x4xi1>57// INNER-REDUCTION: vector.mask %[[INNERMASK]] { vector.reduction <mul>, %[[INNERVEC]], %[[INNERACC]] : vector<4xf32> into f32 } : vector<4xi1> -> f3258// INNER-PARALLEL: %[[TPMASK:.+]] = vector.transpose %[[MASK]], [1, 0] : vector<2x4xi1> to vector<4x2xi1>59// INNER-PARALLEL: %[[TPINPUT:.+]] = vector.transpose %[[INPUT]], [1, 0] : vector<2x4xf32> to vector<4x2xf32>60// INNER-PARALLEL: %[[INNERVEC:.+]] = vector.extract %[[TPINPUT]][0] : vector<2xf32> from vector<4x2xf32>61// INNER-PARALLEL: %[[INNERMASK:.+]] = vector.extract %[[TPMASK]][0] : vector<2xi1> from vector<4x2xi1>62// INNER-PARALLEL: %[[REDUCED:.+]] = arith.mulf %[[INNERVEC]], %[[ACC]] : vector<2xf32>63// INNER-PARALLEL: arith.select %[[INNERMASK]], %[[REDUCED]], %[[ACC]] : vector<2xi1>, vector<2xf32>64