brintos

brintos / llvm-project-archived public Read only

0
0
Text · 4.7 KiB · c1fcbd1 Raw
91 lines · plain
1// RUN: mlir-opt -split-input-file -canonicalize -cse %s | FileCheck %s2 3// This test verifies the simplification of IR patterns that emerge when4// lowering high-level element-wise ops with unranked tensor inputs. Consider5// the following function incrementing and doubling the value of an input6// unranked tensor using ops in a hypothetical high-level dialect called 'hl':7//8//  func.func @f(%input: tensor<*xf32>) -> tensor<*xf32> {9//    %0 = hl.inc %input : tensor<*xf32>10//    %1 = hl.double %0 : tensor<*xf32>11//    return %1 : tensor<*xf32>12//  }13//14// A possible strategy to lower 'hl.inc' consists in reshaping its operand into15// a 1D tensor, creating a 1D tensor splat with the same total size as the input16// operand and with value 1.0, adding both 1D tensors using 'arith.addf', and17// reshaping the result back into the original input shape. A similar process18// applies for 'hl.double', except with a tensor splat with value 2.0 and an19// 'arith.mulf' op. The body of the function in the test below contains the full20// sequence.21//22// Since such lowering process would operate on individual 'hl' ops in a23// context-oblivious manner, the emitted code produces a redundant IR pattern24// where the result of 'arith.addf' is reshaped into an unranked tensor, just25// for it to be immediately reshaped back into the 1D tensor consumed by26// 'arith.mulf'. This entails the overhead of re-computing the unranked tensor27// shape ('shape.shape_of') and size ('shape.num_elements').28//29// This test verifies that the consecutive application of a canonicalization and30// a CSE pass successfully simplifies this emerging pattern, leading to a31// version of the code in which the result of the emitted 'arith.addf' op32// associated with 'hl.inc' is directly consumed by the 'arith.mulf' op33// associated with 'hl.double', as observed in the FileCheck directives. The34// main rewrite patterns at play are 'shape.shape_of' canonicalization,35// 'tensor.reshape' canonicalization, and 'shape.num_elements' subexpression36// elimination.37//38 39// CHECK-LABEL: @unranked_tensor_lowering40// CHECK-SAME: %[[INPUT:.*]]: tensor<*xf32>41 42// CHECK-DAG: %[[ONE:.*]] = arith.constant 1.000000e+00 : f3243// CHECK-DAG: %[[TWO:.*]] = arith.constant 2.000000e+00 : f3244 45// CHECK: %[[INPUT_SHAPE:.*]] = shape.shape_of %[[INPUT]] : tensor<*xf32> -> tensor<?xindex>46// CHECK: %[[INPUT_SIZE:.*]] = shape.num_elements %[[INPUT_SHAPE]] : tensor<?xindex> -> index47// CHECK: %[[INPUT_COLLAPSED_SHAPE:.*]] = tensor.from_elements %[[INPUT_SIZE]] : tensor<1xindex>48// CHECK: %[[INPUT_COLLAPSED:.*]] = tensor.reshape %[[INPUT]](%[[INPUT_COLLAPSED_SHAPE]]) : (tensor<*xf32>, tensor<1xindex>) -> tensor<?xf32>49 50// CHECK: %[[ONE_SPLAT:.*]] = tensor.splat %[[ONE]]{{\[}}%[[INPUT_SIZE]]] : tensor<?xf32>51// CHECK: %[[SUM_COLLAPSED:.*]] = arith.addf %[[INPUT_COLLAPSED]], %[[ONE_SPLAT]] : tensor<?xf32>52 53// CHECK: %[[TWO_SPLAT:.*]] = tensor.splat %[[TWO]]{{\[}}%[[INPUT_SIZE]]] : tensor<?xf32>54// CHECK: %[[PRODUCT_COLLAPSED:.*]] = arith.mulf %[[SUM_COLLAPSED]], %[[TWO_SPLAT]] : tensor<?xf32>55 56// CHECK: %[[PRODUCT:.*]] = tensor.reshape %[[PRODUCT_COLLAPSED]](%[[INPUT_SHAPE]]) : (tensor<?xf32>, tensor<?xindex>) -> tensor<*xf32>57// CHECK: return %[[PRODUCT]] : tensor<*xf32>58 59func.func @unranked_tensor_lowering(%input: tensor<*xf32>) -> tensor<*xf32> {60 61  // Collapse input62  %input_shape = shape.shape_of %input : tensor<*xf32> -> tensor<?xindex>63  %input_size = shape.num_elements %input_shape : tensor<?xindex> -> index64  %input_collapsed_shape = tensor.from_elements %input_size : tensor<1xindex>65  %input_collapsed = tensor.reshape %input(%input_collapsed_shape) : (tensor<*xf32>, tensor<1xindex>) -> tensor<?xf32>66 67  // Second operand for sum68  %one = arith.constant 1.0 : f3269  %one_splat = tensor.splat %one[%input_size] : tensor<?xf32>70 71  // Compute sum and expand it72  %sum_collapsed = arith.addf %input_collapsed, %one_splat : tensor<?xf32>73  %sum = tensor.reshape %sum_collapsed(%input_shape) : (tensor<?xf32>, tensor<?xindex>) -> tensor<*xf32>74 75  // Collapse sum76  %sum_shape = shape.shape_of %sum : tensor<*xf32> -> tensor<?xindex>77  %sum_size = shape.num_elements %sum_shape : tensor<?xindex> -> index78  %sum_collapsed_shape = tensor.from_elements %sum_size : tensor<1xindex>79  %sum_collapsed_0 = tensor.reshape %sum(%sum_collapsed_shape) : (tensor<*xf32>, tensor<1xindex>) -> tensor<?xf32>80 81  // Second operand for product82  %two = arith.constant 2.0 : f3283  %two_splat = tensor.splat %two[%sum_size] : tensor<?xf32>84 85  // Compute product and expand it86  %product_collapsed = arith.mulf %sum_collapsed_0, %two_splat : tensor<?xf32>87  %product = tensor.reshape %product_collapsed(%sum_shape) : (tensor<?xf32>, tensor<?xindex>) -> tensor<*xf32>88 89  return %product : tensor<*xf32>90}91