// RUN: mlir-hlo-opt --transform-unranked-hlo --cse --split-input-file %s | FileCheck %s // Check the validity of expected IR. // CHECK-LABEL: @sqr_transform_result func @sqr_transform_result(%a: tensor<*xf32>) -> tensor<*xf32> { // Flatten operand shape. %shape = shape.shape_of %a : tensor<*xf32> -> tensor %num_elements = shape.num_elements %shape : tensor -> index %flat_shape = tensor_from_elements %num_elements : tensor<1xindex> %flat_a = "mhlo.dynamic_reshape"(%a, %flat_shape) : (tensor<*xf32>, tensor<1xindex>) -> tensor // Apply operation. %flat_b = "mhlo.sqrt"(%flat_a) : (tensor) -> tensor // Restore original shape. %b = "mhlo.dynamic_reshape"(%flat_b, %shape) : (tensor, tensor) -> tensor<*xf32> return %b : tensor<*xf32> } // ----- // Check transformation of unranked code. // CHECK-LABEL: @sqrt // CHECK-SAME: (%[[A:.*]]: tensor<*xf32>) func @sqrt(%a: tensor<*xf32>) -> tensor<*xf32> { // CHECK-NEXT: %[[SHAPE:.*]] = shape.shape_of %[[A]] : tensor<*xf32> -> tensor // CHECK-NEXT: %[[NUM_ELEMENTS:.*]] = shape.num_elements %[[SHAPE]] // CHECK-NEXT: %[[FLAT_SHAPE:.*]] = tensor_from_elements %[[NUM_ELEMENTS]] : tensor<1xindex> // CHECK-NEXT: %[[FLAT_A:.*]] = "mhlo.dynamic_reshape"(%[[A]], %[[FLAT_SHAPE]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK-NEXT: %[[FLAT_B:.*]] = "mhlo.sqrt"(%[[FLAT_A]]) : (tensor) -> tensor // CHECK-NEXT: %[[B:.*]] = "mhlo.dynamic_reshape"(%[[FLAT_B]], %[[SHAPE]]) : (tensor, tensor) -> tensor<*xf32> // CHECK-NEXT: return %[[B]] : tensor<*xf32> %b = "mhlo.sqrt"(%a) : (tensor<*xf32>) -> tensor<*xf32> return %b : tensor<*xf32> } // ----- // Not transformed when ranked. // CHECK-LABEL: @sqrt_ranked // CHECK-SAME: (%[[A:.*]]: tensor<3x?xf32>) func @sqrt_ranked(%a: tensor<3x?xf32>) -> tensor<3x?xf32> { // CHECK-NEXT: %[[B:.*]] = "mhlo.sqrt"(%[[A]]) : (tensor<3x?xf32>) -> tensor<3x?xf32> // CHECK-NEXT: return %[[B]] : tensor<3x?xf32> %b = "mhlo.sqrt"(%a) : (tensor<3x?xf32>) -> tensor<3x?xf32> return %b : tensor<3x?xf32> } // ----- // Not transformed when statically shaped. // CHECK-LABEL: @sqrt_static // CHECK-SAME: (%[[A:.*]]: tensor<2x3xf32>) func @sqrt_static(%a: tensor<2x3xf32>) -> tensor<2x3xf32> { // CHECK-NEXT: %[[B:.*]] = "mhlo.sqrt"(%[[A]]) : (tensor<2x3xf32>) -> tensor<2x3xf32> // CHECK-NEXT: return %[[B]] : tensor<2x3xf32> %b = "mhlo.sqrt"(%a) : (tensor<2x3xf32>) -> tensor<2x3xf32> return %b : tensor<2x3xf32> } // ----- // CHECK-LABEL: @add_unranked // CHECK-SAME: (%[[A:.*]]: tensor<*xf32>, %[[B:.*]]: tensor<*xf32>) -> tensor<*xf32> func @add_unranked(%a : tensor<*xf32>, %b : tensor<*xf32>) -> tensor<*xf32> { // CHECK: %[[SHAPE_A:.*]] = shape.shape_of %[[A]] // CHECK: %[[SHAPE_B:.*]] = shape.shape_of %[[B]] // CHECK: %[[SHAPE:.*]] = shape.any %[[SHAPE_A]], %[[SHAPE_B]] // CHECK: %[[NUM_ELEMENTS:.*]] = shape.num_elements %[[SHAPE]] // CHECK: %[[FLAT_SHAPE:.*]] = tensor_from_elements %[[NUM_ELEMENTS]] : tensor<1xindex> // CHECK: %[[FLAT_A:.*]] = "mhlo.dynamic_reshape"(%[[A]], %[[FLAT_SHAPE]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK: %[[FLAT_B:.*]] = "mhlo.dynamic_reshape"(%[[B]], %[[FLAT_SHAPE]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK: %[[FLAT_RESULT:.*]] = mhlo.add %[[FLAT_A]], %[[FLAT_B]] : tensor // CHECK: %[[RESULT:.*]] = "mhlo.dynamic_reshape"(%[[FLAT_RESULT]], %[[SHAPE]]) : (tensor, tensor) -> tensor<*xf32> // CHECK: return %[[RESULT]] : tensor<*xf32> %result = mhlo.add %a, %b : tensor<*xf32> return %result : tensor<*xf32> } // ----- // CHECK-LABEL: @tan // CHECK-SAME: (%[[A:.*]]: tensor<*xf32>) -> tensor<*xf32> func @tan(%a : tensor<*xf32>) -> tensor<*xf32> { // CHECK: %[[SHAPE:.*]] = shape.shape_of %[[A]] : tensor<*xf32> -> tensor // CHECK: %[[NUM_ELEMENTS:.*]] = shape.num_elements %[[SHAPE]] // CHECK: %[[FLAT_SHAPE:.*]] = tensor_from_elements %[[NUM_ELEMENTS]] : tensor<1xindex> // CHECK: %[[FLAT_A:.*]] = "mhlo.dynamic_reshape"(%[[A]], %[[FLAT_SHAPE]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK: %[[FLAT_B:.*]] = chlo.tan %[[FLAT_A]] : tensor // CHECK: %[[B:.*]] = "mhlo.dynamic_reshape"(%[[FLAT_B]], %[[SHAPE]]) : (tensor, tensor) -> tensor<*xf32> // CHECK: return %[[B]] : tensor<*xf32> %result = chlo.tan %a : tensor<*xf32> return %result : tensor<*xf32> } // ----- func @addScalarUnranked(%arg0: tensor, %arg1: tensor<*xf32>) -> tensor<*xf32> { %0 = chlo.broadcast_add %arg0, %arg1 : (tensor, tensor<*xf32>) -> tensor<*xf32> return %0 : tensor<*xf32> } // CHECK-LABEL: func @addScalarUnranked( // CHECK-SAME: %[[ARG_0:.*]]: tensor, // CHECK-SAME: %[[ARG_1:.*]]: tensor<*xf32> // CHECK-SAME: ) -> tensor<*xf32> { // First handle the dynamic reshaping of the unranked operand // to a 1D tensor. // CHECK-NEXT: %[[SHAPE_1:.*]] = shape.shape_of %[[ARG_1]] : tensor<*xf32> // CHECK-NEXT: %[[NUM_ELEMENTS:.*]] = shape.num_elements %[[SHAPE_1]] : tensor -> index // CHECK-NEXT: %[[SIZE_TENSOR:.*]] = tensor_from_elements %[[NUM_ELEMENTS]] : tensor<1xindex> // CHECK-NEXT: %[[RESHAPED:.*]] = "mhlo.dynamic_reshape"(%[[ARG_1]], %[[SIZE_TENSOR]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK-NEXT: %[[BROADCASTED_RESULT:.*]] = chlo.broadcast_add %[[ARG_0]], %[[RESHAPED]] : (tensor, tensor) -> tensor // As part of the unranked logic, the result is reshaped back // to an unranked tensor. // CHECK-NEXT: %[[RESHAPED_RESULT:.*]] = "mhlo.dynamic_reshape"(%[[BROADCASTED_RESULT:.*]], %[[SHAPE_1]]) : (tensor, tensor) -> tensor<*xf32> // CHECK-NEXT: return %[[RESHAPED_RESULT]] : tensor<*xf32> // CHECK-NEXT: } // ----- func @addUnrankedScalar(%arg0: tensor<*xf32>, %arg1: tensor) -> tensor<*xf32> { %0 = chlo.broadcast_add %arg0, %arg1 : (tensor<*xf32>, tensor) -> tensor<*xf32> return %0 : tensor<*xf32> } // CHECK-LABEL: func @addUnrankedScalar( // CHECK-SAME: %[[ARG_0:.*]]: tensor<*xf32>, // CHECK-SAME: %[[ARG_1:.*]]: tensor) -> tensor<*xf32> { // First handle the dynamic reshaping of the unranked operand // to a 1D tensor. // CHECK-NEXT: %[[SHAPE_0:.*]] = shape.shape_of %[[ARG_0]] : tensor<*xf32> // CHECK-NEXT: %[[NUM_ELEMENTS:.*]] = shape.num_elements %[[SHAPE_0]] : tensor -> index // CHECK-NEXT: %[[SIZE_TENSOR:.*]] = tensor_from_elements %[[NUM_ELEMENTS]] : tensor<1xindex> // CHECK-NEXT: %[[RESHAPED:.*]] = "mhlo.dynamic_reshape"(%[[ARG_0]], %[[SIZE_TENSOR]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // The assuming region is part of the second stage of lowering // with ranked broadcasting logic. // CHECK-NEXT: %[[BROADCASTED_RESULT:.*]] = chlo.broadcast_add %[[RESHAPED]], %[[ARG_1]] : (tensor, tensor) -> tensor // As part of the unranked logic, the result is reshaped back // to an unranked tensor. // CHECK-NEXT: %[[RESHAPED_RESULT:.*]] = "mhlo.dynamic_reshape"(%[[BROADCASTED_RESULT:.*]], %[[SHAPE_0]]) : (tensor, tensor) -> tensor<*xf32> // CHECK-NEXT: return %[[RESHAPED_RESULT]] : tensor<*xf32> // CHECK-NEXT: } // ----- func @addUnrankedUnranked( %arg0: tensor<*xf32>, %arg1: tensor<*xf32>) -> tensor<*xf32> { %0 = chlo.broadcast_add %arg0, %arg1 : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32> return %0 : tensor<*xf32> } // CHECK-LABEL: func @addUnrankedUnranked( // CHECK-SAME: %[[LHS:.*]]: tensor<*xf32>, // CHECK-SAME: %[[RHS:.*]]: tensor<*xf32>) -> tensor<*xf32> { // CHECK-NEXT: %[[LHS_SHAPE:.*]] = shape.shape_of %[[LHS]] : tensor<*xf32> -> tensor // CHECK-NEXT: %[[RANK_LHS:.*]] = shape.rank %[[LHS_SHAPE]] : tensor -> index // CHECK-NEXT: %[[C0:.*]] = constant 0 : index // CHECK-NEXT: %[[LHS_IS_SCALAR:.*]] = cmpi "eq", %[[RANK_LHS]], %[[C0]] : index // Handle scalar LHS case // CHECK-NEXT: %[[VAL_8:.*]] = scf.if %[[LHS_IS_SCALAR]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[SCALAR_LHS:.*]] = tensor_cast %[[LHS]] : tensor<*xf32> to tensor // CHECK-NEXT: %[[RHS_SHAPE_1:.*]] = shape.shape_of %[[RHS]] : tensor<*xf32> -> tensor // CHECK-NEXT: %[[NUM_RHS:.*]] = shape.num_elements %[[RHS_SHAPE_1]] : tensor -> index // CHECK-NEXT: %[[NUM_TENS_RHS:.*]] = tensor_from_elements %[[NUM_RHS]] : tensor<1xindex> // CHECK-NEXT: %[[RESHAPED_RHS:.*]] = "mhlo.dynamic_reshape"(%[[RHS]], %[[NUM_TENS_RHS]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK-NEXT: %[[LHS_SCALAR_RESULT:.*]] = chlo.broadcast_add %[[SCALAR_LHS]], %[[RESHAPED_RHS]] : (tensor, tensor) -> tensor // CHECK-NEXT: %[[RESHAPED_LHS_SCALAR_RESULT:.*]] = "mhlo.dynamic_reshape"(%[[LHS_SCALAR_RESULT]], %[[RHS_SHAPE_1]]) : (tensor, tensor) -> tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESHAPED_LHS_SCALAR_RESULT]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %[[RHS_SHAPE:.*]] = shape.shape_of %[[RHS]] : tensor<*xf32> -> tensor // CHECK-NEXT: %[[RANK_RHS:.*]] = shape.rank %[[RHS_SHAPE]] : tensor -> index // CHECK-NEXT: %[[RHS_IS_SCALAR:.*]] = cmpi "eq", %[[RANK_RHS]], %[[C0]] : index // Handle scalar RHS case // CHECK-NEXT: %[[VAL_14:.*]] = scf.if %[[RHS_IS_SCALAR]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[SCALAR_RHS:.*]] = tensor_cast %[[RHS]] : tensor<*xf32> to tensor // CHECK-NEXT: %[[NUM_LHS:.*]] = shape.num_elements %[[LHS_SHAPE]] : tensor -> index // CHECK-NEXT: %[[NUM_TENS_LHS:.*]] = tensor_from_elements %[[NUM_LHS]] : tensor<1xindex> // CHECK-NEXT: %[[RESHAPED_LHS:.*]] = "mhlo.dynamic_reshape"(%[[LHS]], %[[NUM_TENS_LHS]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK-NEXT: %[[RHS_SCALAR_RESULT:.*]] = chlo.broadcast_add %[[RESHAPED_LHS]], %[[SCALAR_RHS]] : (tensor, tensor) -> tensor // CHECK-NEXT: %[[RESHAPED_RHS_SCALAR_RESULT:.*]] = "mhlo.dynamic_reshape"(%[[RHS_SCALAR_RESULT:.*]], %[[LHS_SHAPE]]) : (tensor, tensor) -> tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESHAPED_RHS_SCALAR_RESULT]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %[[SHAPES_EQ:.*]] = shape.shape_eq %[[LHS_SHAPE]], %[[RHS_SHAPE]] : tensor, tensor // Handle equal shapes case // CHECK-NEXT: %[[VAL_18:.*]] = scf.if %[[SHAPES_EQ]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[ANY_SHAPE:.*]] = shape.any %[[LHS_SHAPE]], %[[RHS_SHAPE]] : tensor, tensor -> tensor // CHECK-NEXT: %[[ANY_NUM:.*]] = shape.num_elements %[[ANY_SHAPE]] : tensor -> index // CHECK-NEXT: %[[ANY_TENSOR:.*]] = tensor_from_elements %[[ANY_NUM]] : tensor<1xindex> // CHECK-NEXT: %[[FLATTENED_LHS:.*]] = "mhlo.dynamic_reshape"(%[[LHS]], %[[ANY_TENSOR]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK-NEXT: %[[FLATTENED_RHS:.*]] = "mhlo.dynamic_reshape"(%[[RHS]], %[[ANY_TENSOR]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor // CHECK-NEXT: %[[FLATTENED_RESULT:.*]] = mhlo.add %[[FLATTENED_LHS]], %[[FLATTENED_RHS]] : tensor // CHECK-NEXT: %[[RESHAPED_SAME_RESULT:.*]] = "mhlo.dynamic_reshape"(%[[FLATTENED_RESULT]], %[[ANY_SHAPE]]) : (tensor, tensor) -> tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESHAPED_SAME_RESULT]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %[[LHS_RANK:.*]] = rank %[[LHS_SHAPE]] : tensor // CHECK-NEXT: %[[RHS_RANK:.*]] = rank %[[RHS_SHAPE]] : tensor // CHECK-NEXT: %[[LHS_RANK_GREATER:.*]] = cmpi "sgt", %[[LHS_RANK]], %[[RHS_RANK]] : index // CHECK-NEXT: %[[GREATEST_RANK:.*]] = select %[[LHS_RANK_GREATER]], %[[LHS_RANK]], %[[RHS_RANK]] : index // CHECK-NEXT: %[[C2:.*]] = constant 2 : index // CHECK-NEXT: %[[GREATEST_RANK_IS_2:.*]] = cmpi "eq", %[[GREATEST_RANK]], %[[C2]] : index // Handle rank 2 specialization // CHECK-NEXT: %[[VAL_26:.*]] = scf.if %[[GREATEST_RANK_IS_2]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[CONST_SHAPE_2:.*]] = shape.const_shape [1, 1] // CHECK-NEXT: %[[BROADCASTED_LHS_2:.*]] = shape.broadcast %[[LHS_SHAPE]], %[[CONST_SHAPE_2]] : tensor, tensor<2xindex> -> tensor // CHECK-NEXT: %[[CASTED_LHS_2:.*]] = tensor_cast %[[BROADCASTED_LHS_2]] : tensor to tensor<2xindex> // CHECK-NEXT: %[[BROADCASTED_RHS_2:.*]] = shape.broadcast %[[RHS_SHAPE]], %[[CONST_SHAPE_2]] : tensor, tensor<2xindex> -> tensor // CHECK-NEXT: %[[CASTED_RHS_2:.*]] = tensor_cast %[[BROADCASTED_RHS_2]] : tensor to tensor<2xindex> // CHECK-NEXT: %[[RESHAPED_LHS_2:.*]] = "mhlo.dynamic_reshape"(%[[LHS]], %[[CASTED_LHS_2]]) : (tensor<*xf32>, tensor<2xindex>) -> tensor // CHECK-NEXT: %[[RESHAPED_RHS_2:.*]] = "mhlo.dynamic_reshape"(%[[RHS]], %[[CASTED_RHS_2]]) : (tensor<*xf32>, tensor<2xindex>) -> tensor // CHECK-NEXT: %[[RESULT_RANK_2:.*]] = chlo.broadcast_add %[[RESHAPED_LHS_2]], %[[RESHAPED_RHS_2]] : (tensor, tensor) -> tensor // CHECK-NEXT: %[[RESULT_2:.*]] = tensor_cast %[[RESULT_RANK_2]] : tensor to tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESULT_2]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %[[C3:.*]] = constant 3 : index // CHECK-NEXT: %[[GREATEST_RANK_IS_3:.*]] = cmpi "eq", %[[GREATEST_RANK]], %[[C3]] : index // Handle rank 3 specialization // CHECK-NEXT: %[[VAL_34:.*]] = scf.if %[[GREATEST_RANK_IS_3]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[CONST_SHAPE_3:.*]] = shape.const_shape [1, 1, 1] // CHECK-NEXT: %[[BROADCASTED_LHS_3:.*]] = shape.broadcast %[[LHS_SHAPE]], %[[CONST_SHAPE_3]] : tensor, tensor<3xindex> -> tensor // CHECK-NEXT: %[[CASTED_LHS_3:.*]] = tensor_cast %[[BROADCASTED_LHS_3]] : tensor to tensor<3xindex> // CHECK-NEXT: %[[BROADCASTED_RHS_3:.*]] = shape.broadcast %[[RHS_SHAPE]], %[[CONST_SHAPE_3]] : tensor, tensor<3xindex> -> tensor // CHECK-NEXT: %[[CASTED_RHS_3:.*]] = tensor_cast %[[BROADCASTED_RHS_3]] : tensor to tensor<3xindex> // CHECK-NEXT: %[[RESHAPED_LHS_3:.*]] = "mhlo.dynamic_reshape"(%[[LHS]], %[[CASTED_LHS_3]]) : (tensor<*xf32>, tensor<3xindex>) -> tensor // CHECK-NEXT: %[[RESHAPED_RHS_3:.*]] = "mhlo.dynamic_reshape"(%[[RHS]], %[[CASTED_RHS_3]]) : (tensor<*xf32>, tensor<3xindex>) -> tensor // CHECK-NEXT: %[[RESULT_RANK_3:.*]] = chlo.broadcast_add %[[RESHAPED_LHS_3]], %[[RESHAPED_RHS_3]] : (tensor, tensor) -> tensor // CHECK-NEXT: %[[RESULT_3:.*]] = tensor_cast %[[RESULT_RANK_3]] : tensor to tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESULT_3]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %[[C4:.*]] = constant 4 : index // CHECK-NEXT: %[[GREATEST_RANK_IS_4:.*]] = cmpi "eq", %[[GREATEST_RANK]], %[[C4]] : index // Handle rank 4 specialization // CHECK-NEXT: %[[VAL_42:.*]] = scf.if %[[GREATEST_RANK_IS_4]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[CONST_SHAPE_4:.*]] = shape.const_shape [1, 1, 1, 1] // CHECK-NEXT: %[[BROADCASTED_LHS_4:.*]] = shape.broadcast %[[LHS_SHAPE]], %[[CONST_SHAPE_4]] : tensor, tensor<4xindex> -> tensor // CHECK-NEXT: %[[CASTED_LHS_4:.*]] = tensor_cast %[[BROADCASTED_LHS_4]] : tensor to tensor<4xindex> // CHECK-NEXT: %[[BROADCASTED_RHS_4:.*]] = shape.broadcast %[[RHS_SHAPE]], %[[CONST_SHAPE_4]] : tensor, tensor<4xindex> -> tensor // CHECK-NEXT: %[[CASTED_RHS_4:.*]] = tensor_cast %[[BROADCASTED_RHS_4]] : tensor to tensor<4xindex> // CHECK-NEXT: %[[RESHAPED_LHS_4:.*]] = "mhlo.dynamic_reshape"(%[[LHS]], %[[CASTED_LHS_4]]) : (tensor<*xf32>, tensor<4xindex>) -> tensor // CHECK-NEXT: %[[RESHAPED_RHS_4:.*]] = "mhlo.dynamic_reshape"(%[[RHS]], %[[CASTED_RHS_4]]) : (tensor<*xf32>, tensor<4xindex>) -> tensor // CHECK-NEXT: %[[RESULT_RANK_4:.*]] = chlo.broadcast_add %[[RESHAPED_LHS_4]], %[[RESHAPED_RHS_4]] : (tensor, tensor) -> tensor // CHECK-NEXT: %[[RESULT_4:.*]] = tensor_cast %[[RESULT_RANK_4]] : tensor to tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESULT_4]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %[[C5:.*]] = constant 5 : index // CHECK-NEXT: %[[GREATEST_RANK_IS_5:.*]] = cmpi "eq", %[[GREATEST_RANK]], %[[C5]] : index // Handle rank 5 specialization // CHECK-NEXT: %[[VAL_50:.*]] = scf.if %[[GREATEST_RANK_IS_5]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[CONST_SHAPE_5:.*]] = shape.const_shape [1, 1, 1, 1, 1] // CHECK-NEXT: %[[BROADCASTED_LHS_5:.*]] = shape.broadcast %[[LHS_SHAPE]], %[[CONST_SHAPE_5]] : tensor, tensor<5xindex> -> tensor // CHECK-NEXT: %[[CASTED_LHS_5:.*]] = tensor_cast %[[BROADCASTED_LHS_5]] : tensor to tensor<5xindex> // CHECK-NEXT: %[[BROADCASTED_RHS_5:.*]] = shape.broadcast %[[RHS_SHAPE]], %[[CONST_SHAPE_5]] : tensor, tensor<5xindex> -> tensor // CHECK-NEXT: %[[CASTED_RHS_5:.*]] = tensor_cast %[[BROADCASTED_RHS_5]] : tensor to tensor<5xindex> // CHECK-NEXT: %[[RESHAPED_LHS_5:.*]] = "mhlo.dynamic_reshape"(%[[LHS]], %[[CASTED_LHS_5]]) : (tensor<*xf32>, tensor<5xindex>) -> tensor // CHECK-NEXT: %[[RESHAPED_RHS_5:.*]] = "mhlo.dynamic_reshape"(%[[RHS]], %[[CASTED_RHS_5]]) : (tensor<*xf32>, tensor<5xindex>) -> tensor // CHECK-NEXT: %[[RESULT_RANK_5:.*]] = chlo.broadcast_add %[[RESHAPED_LHS_5]], %[[RESHAPED_RHS_5]] : (tensor, tensor) -> tensor // CHECK-NEXT: %[[RESULT_5:.*]] = tensor_cast %[[RESULT_RANK_5]] : tensor to tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESULT_5]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %[[C6:.*]] = constant 6 : index // CHECK-NEXT: %[[GREATEST_RANK_IS_6:.*]] = cmpi "eq", %[[GREATEST_RANK]], %[[C6]] : index // Handle rank 6 specialization // CHECK-NEXT: %[[VAL_58:.*]] = scf.if %[[GREATEST_RANK_IS_6]] -> (tensor<*xf32>) { // CHECK-NEXT: %[[CONST_SHAPE_6:.*]] = shape.const_shape [1, 1, 1, 1, 1, 1] // CHECK-NEXT: %[[BROADCASTED_LHS_6:.*]] = shape.broadcast %[[LHS_SHAPE]], %[[CONST_SHAPE_6]] : tensor, tensor<6xindex> -> tensor // CHECK-NEXT: %[[CASTED_LHS_6:.*]] = tensor_cast %[[BROADCASTED_LHS_6]] : tensor to tensor<6xindex> // CHECK-NEXT: %[[BROADCASTED_RHS_6:.*]] = shape.broadcast %[[RHS_SHAPE]], %[[CONST_SHAPE_6]] : tensor, tensor<6xindex> -> tensor // CHECK-NEXT: %[[CASTED_RHS_6:.*]] = tensor_cast %[[BROADCASTED_RHS_6]] : tensor to tensor<6xindex> // CHECK-NEXT: %[[RESHAPED_LHS_6:.*]] = "mhlo.dynamic_reshape"(%[[LHS]], %[[CASTED_LHS_6]]) : (tensor<*xf32>, tensor<6xindex>) -> tensor // CHECK-NEXT: %[[RESHAPED_RHS_6:.*]] = "mhlo.dynamic_reshape"(%[[RHS]], %[[CASTED_RHS_6]]) : (tensor<*xf32>, tensor<6xindex>) -> tensor // CHECK-NEXT: %[[RESULT_RANK_6:.*]] = chlo.broadcast_add %[[RESHAPED_LHS_6]], %[[RESHAPED_RHS_6]] : (tensor, tensor) -> tensor // CHECK-NEXT: %[[RESULT_6:.*]] = tensor_cast %[[RESULT_RANK_6]] : tensor to tensor<*xf32> // CHECK-NEXT: scf.yield %[[RESULT_6]] : tensor<*xf32> // CHECK-NEXT: } else { // CHECK-NEXT: %false = constant false // CHECK-NEXT: assert %false // CHECK-NEXT: scf.yield %[[LHS]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: scf.yield %[[VAL_64:.*]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: scf.yield %[[VAL_65:.*]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: scf.yield %[[VAL_66:.*]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: scf.yield %[[VAL_67:.*]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: scf.yield %[[VAL_68:.*]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: scf.yield %[[VAL_69:.*]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: scf.yield %[[VAL_70:.*]] : tensor<*xf32> // CHECK-NEXT: } // CHECK-NEXT: return %[[VAL_71:.*]] : tensor<*xf32> // CHECK-NEXT: }