brintos

brintos / llvm-project-archived public Read only

0
0
Text · 7.8 KiB · 25218b8 Raw
191 lines · plain
1//--------------------------------------------------------------------------------------------------2// WHEN CREATING A NEW TEST, PLEASE JUST COPY & PASTE WITHOUT EDITS.3//4// Set-up that's shared across all tests in this directory. In principle, this5// config could be moved to lit.local.cfg. However, there are downstream users that6//  do not use these LIT config files. Hence why this is kept inline.7//8// DEFINE: %{sparsifier_opts} = enable-runtime-library=true9// DEFINE: %{sparsifier_opts_sve} = enable-arm-sve=true %{sparsifier_opts}10// DEFINE: %{compile} = mlir-opt %s --sparsifier="%{sparsifier_opts}"11// DEFINE: %{compile_sve} = mlir-opt %s --sparsifier="%{sparsifier_opts_sve}"12// DEFINE: %{run_libs} = -shared-libs=%mlir_c_runner_utils,%mlir_runner_utils13// DEFINE: %{run_libs_sve} = -shared-libs=%native_mlir_runner_utils,%native_mlir_c_runner_utils14// DEFINE: %{run_opts} = -e main -entry-point-result=void15// DEFINE: %{run} = mlir-runner %{run_opts} %{run_libs}16// DEFINE: %{run_sve} = %mcr_aarch64_cmd --march=aarch64 --mattr="+sve" %{run_opts} %{run_libs_sve}17//18// DEFINE: %{env} =19//--------------------------------------------------------------------------------------------------20 21// RUN: %{compile} | %{run} | FileCheck %s22//23// Do the same run, but now with direct IR generation.24// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false enable-buffer-initialization=true25// RUN: %{compile} | %{run} | FileCheck %s26//27// Do the same run, but now with direct IR generation and vectorization.28// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false enable-buffer-initialization=true vl=2 reassociate-fp-reductions=true enable-index-optimizations=true29// RUN: %{compile} | %{run} | FileCheck %s30//31// Do the same run, but now with direct IR generation and VLA vectorization.32// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %}33 34// Reduction in this file _are_ supported by the AArch64 SVE backend35 36#SparseVector = #sparse_tensor.encoding<{map = (d0) -> (d0 : compressed)}>37#CSR = #sparse_tensor.encoding<{map = (d0, d1) -> (d0 : dense, d1 : compressed)}>38#CSC = #sparse_tensor.encoding<{39  map = (d0, d1) -> (d1 : dense, d0 : compressed)40}>41 42//43// Traits for tensor operations.44//45#trait_matmul = {46  indexing_maps = [47    affine_map<(i,j,k) -> (i,k)>, // A48    affine_map<(i,j,k) -> (k,j)>, // B49    affine_map<(i,j,k) -> (i,j)>  // C (out)50  ],51  iterator_types = ["parallel", "parallel", "reduction"],52  doc = "C(i,j) = SUM_k A(i,k) * B(k,j)"53}54 55module {56  func.func @min_plus_csrcsr(%arga: tensor<?x?xf64, #CSR>,57                             %argb: tensor<?x?xf64, #CSR>) -> tensor<?x?xf64, #CSR> {58    %c0 = arith.constant 0 : index59    %c1 = arith.constant 1 : index60    %maxf = arith.constant 1.0e999 : f6461    %d0 = tensor.dim %arga, %c0 : tensor<?x?xf64, #CSR>62    %d1 = tensor.dim %argb, %c1 : tensor<?x?xf64, #CSR>63    %xm = tensor.empty(%d0, %d1) : tensor<?x?xf64, #CSR>64    %0 = linalg.generic #trait_matmul65       ins(%arga, %argb: tensor<?x?xf64, #CSR>, tensor<?x?xf64, #CSR>)66        outs(%xm: tensor<?x?xf64, #CSR>) {67        ^bb(%a: f64, %b: f64, %output: f64):68          %1 = sparse_tensor.binary %a, %b : f64, f64 to f6469            overlap = {70              ^bb0(%x: f64, %y: f64):71                %3 = arith.addf %x, %y : f6472                sparse_tensor.yield %3 : f6473            }74            left={}75            right={}76          %2 = sparse_tensor.reduce %1, %output, %maxf : f64 {77              ^bb0(%x: f64, %y: f64):78                %cmp = arith.cmpf "olt", %x, %y : f6479                %3 = arith.select %cmp, %x, %y : f6480                sparse_tensor.yield %3 : f6481            }82          linalg.yield %2 : f6483    } -> tensor<?x?xf64, #CSR>84    return %0 : tensor<?x?xf64, #CSR>85  }86 87  func.func @min_plus_csrcsc(%arga: tensor<?x?xf64, #CSR>,88                             %argb: tensor<?x?xf64, #CSC>) -> tensor<?x?xf64, #CSR> {89    %c0 = arith.constant 0 : index90    %c1 = arith.constant 1 : index91    %maxf = arith.constant 1.0e999 : f6492    %d0 = tensor.dim %arga, %c0 : tensor<?x?xf64, #CSR>93    %d1 = tensor.dim %argb, %c1 : tensor<?x?xf64, #CSC>94    %xm = tensor.empty(%d0, %d1) : tensor<?x?xf64, #CSR>95    %0 = linalg.generic #trait_matmul96       ins(%arga, %argb: tensor<?x?xf64, #CSR>, tensor<?x?xf64, #CSC>)97        outs(%xm: tensor<?x?xf64, #CSR>) {98        ^bb(%a: f64, %b: f64, %output: f64):99          %1 = sparse_tensor.binary %a, %b : f64, f64 to f64100            overlap = {101              ^bb0(%x: f64, %y: f64):102                %3 = arith.addf %x, %y : f64103                sparse_tensor.yield %3 : f64104            }105            left={}106            right={}107          %2 = sparse_tensor.reduce %1, %output, %maxf : f64 {108              ^bb0(%x: f64, %y: f64):109                %cmp = arith.cmpf "olt", %x, %y : f64110                %3 = arith.select %cmp, %x, %y : f64111                sparse_tensor.yield %3 : f64112            }113          linalg.yield %2 : f64114    } -> tensor<?x?xf64, #CSR>115    return %0 : tensor<?x?xf64, #CSR>116  }117 118  // Driver method to call and verify vector kernels.119  func.func @main() {120    %c0 = arith.constant 0 : index121 122    // Setup sparse matrices.123    %m1 = arith.constant sparse<124       [ [0,0], [0,1], [1,0], [2,2], [2,3], [2,4], [3,0], [3,2], [3,3] ],125         [ 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0 ]126    > : tensor<4x5xf64>127    %m2 = arith.constant sparse<128       [ [0,0], [1,3], [2,0], [2,3], [3,1], [4,1] ],129         [6.0, 5.0, 4.0, 3.0, 2.0, 11.0 ]130    > : tensor<5x4xf64>131    %sm1 = sparse_tensor.convert %m1 : tensor<4x5xf64> to tensor<?x?xf64, #CSR>132    %sm2r = sparse_tensor.convert %m2 : tensor<5x4xf64> to tensor<?x?xf64, #CSR>133    %sm2c = sparse_tensor.convert %m2 : tensor<5x4xf64> to tensor<?x?xf64, #CSC>134 135    // Call sparse matrix kernels.136    %5 = call @min_plus_csrcsr(%sm1, %sm2r)137      : (tensor<?x?xf64, #CSR>, tensor<?x?xf64, #CSR>) -> tensor<?x?xf64, #CSR>138    %6 = call @min_plus_csrcsc(%sm1, %sm2c)139      : (tensor<?x?xf64, #CSR>, tensor<?x?xf64, #CSC>) -> tensor<?x?xf64, #CSR>140 141    //142    // Verify the results.143    //144    // CHECK:      ---- Sparse Tensor ----145    // CHECK-NEXT: nse = 9146    // CHECK-NEXT: dim = ( 4, 5 )147    // CHECK-NEXT: lvl = ( 4, 5 )148    // CHECK-NEXT: pos[1] : ( 0, 2, 3, 6, 9 )149    // CHECK-NEXT: crd[1] : ( 0, 1, 0, 2, 3, 4, 0, 2, 3 )150    // CHECK-NEXT: values : ( 1, 2, 3, 4, 5, 6, 7, 8, 9 )151    // CHECK-NEXT: ----152    // CHECK:      ---- Sparse Tensor ----153    // CHECK-NEXT: nse = 6154    // CHECK-NEXT: dim = ( 5, 4 )155    // CHECK-NEXT: lvl = ( 5, 4 )156    // CHECK-NEXT: pos[1] : ( 0, 1, 2, 4, 5, 6 )157    // CHECK-NEXT: crd[1] : ( 0, 3, 0, 3, 1, 1 )158    // CHECK-NEXT: values : ( 6, 5, 4, 3, 2, 11 )159    // CHECK-NEXT: ----160    // CHECK:      ---- Sparse Tensor ----161    // CHECK-NEXT: nse = 9162    // CHECK-NEXT: dim = ( 4, 4 )163    // CHECK-NEXT: lvl = ( 4, 4 )164    // CHECK-NEXT: pos[1] : ( 0, 2, 3, 6, 9 )165    // CHECK-NEXT: crd[1] : ( 0, 3, 0, 0, 1, 3, 0, 1, 3 )166    // CHECK-NEXT: values : ( 7, 7, 9, 8, 7, 7, 12, 11, 11 )167    // CHECK-NEXT: ----168    // CHECK:      ---- Sparse Tensor ----169    // CHECK-NEXT: nse = 9170    // CHECK-NEXT: dim = ( 4, 4 )171    // CHECK-NEXT: lvl = ( 4, 4 )172    // CHECK-NEXT: pos[1] : ( 0, 2, 3, 6, 9 )173    // CHECK-NEXT: crd[1] : ( 0, 3, 0, 0, 1, 3, 0, 1, 3 )174    // CHECK-NEXT: values : ( 7, 7, 9, 8, 7, 7, 12, 11, 11 )175    // CHECK-NEXT: ----176    //177    sparse_tensor.print %sm1 : tensor<?x?xf64, #CSR>178    sparse_tensor.print %sm2r : tensor<?x?xf64, #CSR>179    sparse_tensor.print %5 : tensor<?x?xf64, #CSR>180    sparse_tensor.print %6 : tensor<?x?xf64, #CSR>181 182    // Release the resources.183    bufferization.dealloc_tensor %sm1 : tensor<?x?xf64, #CSR>184    bufferization.dealloc_tensor %sm2r : tensor<?x?xf64, #CSR>185    bufferization.dealloc_tensor %sm2c : tensor<?x?xf64, #CSC>186    bufferization.dealloc_tensor %5 : tensor<?x?xf64, #CSR>187    bufferization.dealloc_tensor %6 : tensor<?x?xf64, #CSR>188    return189  }190}191