brintos

brintos / llvm-project-archived public Read only

0
0
Text · 5.9 KiB · cf8b271 Raw
153 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// REDEFINE: %{env} = TENSOR0=%mlir_src_dir/test/Integration/data/mttkrp_b.tns22// RUN: %{compile} | env %{env} %{run} | FileCheck %s23//24// Do the same run, but now with direct IR generation.25// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false26// RUN: %{compile} | env %{env} %{run} | FileCheck %s27//28// Do the same run, but now with direct IR generation and vectorization.29// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true30// RUN: %{compile} | env %{env} %{run} | FileCheck %s31//32// Do the same run, but now with direct IR generation and, if available, VLA33// vectorization.34// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | env %{env} %{run_sve} | FileCheck %s %}35 36!Filename = !llvm.ptr37 38#SparseTensor = #sparse_tensor.encoding<{39  map = (d0, d1, d2) -> (d0 : compressed, d1 : compressed, d2 : compressed)40}>41 42#mttkrp = {43  indexing_maps = [44    affine_map<(i,j,k,l) -> (i,k,l)>, // B45    affine_map<(i,j,k,l) -> (k,j)>,   // C46    affine_map<(i,j,k,l) -> (l,j)>,   // D47    affine_map<(i,j,k,l) -> (i,j)>    // A (out)48  ],49  iterator_types = ["parallel", "parallel", "reduction", "reduction"],50  doc = "A(i,j) += B(i,k,l) * D(l,j) * C(k,j)"51}52 53//54// Integration test that lowers a kernel annotated as sparse to55// actual sparse code, initializes a matching sparse storage scheme56// from file, and runs the resulting code with the JIT compiler.57//58module {59  func.func private @printMemrefF64(%ptr : tensor<*xf64>)60 61  //62  // Computes Matricized Tensor Times Khatri-Rao Product (MTTKRP) kernel. See63  // http://tensor-compiler.org/docs/data_analytics/index.html.64  //65  func.func @kernel_mttkrp(%argb: tensor<?x?x?xf64, #SparseTensor>,66                           %argc: tensor<?x?xf64>,67                           %argd: tensor<?x?xf64>,68                           %arga: tensor<?x?xf64>)69                               -> tensor<?x?xf64> {70    %0 = linalg.generic #mttkrp71      ins(%argb, %argc, %argd:72            tensor<?x?x?xf64, #SparseTensor>, tensor<?x?xf64>, tensor<?x?xf64>)73      outs(%arga: tensor<?x?xf64>) {74      ^bb(%b: f64, %c: f64, %d: f64, %a: f64):75        %0 = arith.mulf %b, %c : f6476        %1 = arith.mulf %d, %0 : f6477        %2 = arith.addf %a, %1 : f6478        linalg.yield %2 : f6479    } -> tensor<?x?xf64>80    return %0 : tensor<?x?xf64>81  }82 83  func.func private @getTensorFilename(index) -> (!Filename)84 85  //86  // Main driver that reads matrix from file and calls the sparse kernel.87  //88  func.func @main() {89    %f0 = arith.constant 0.0 : f6490    %cst0 = arith.constant 0 : index91    %cst1 = arith.constant 1 : index92    %cst2 = arith.constant 2 : index93 94    // Read the sparse input tensor B from a file.95    %fileName = call @getTensorFilename(%cst0) : (index) -> (!Filename)96    %b = sparse_tensor.new %fileName97          : !Filename to tensor<?x?x?xf64, #SparseTensor>98 99    // Get sizes from B, pick a fixed size for dim-2 of A.100    %isz = tensor.dim %b, %cst0 : tensor<?x?x?xf64, #SparseTensor>101    %jsz = arith.constant 5 : index102    %ksz = tensor.dim %b, %cst1 : tensor<?x?x?xf64, #SparseTensor>103    %lsz = tensor.dim %b, %cst2 : tensor<?x?x?xf64, #SparseTensor>104 105    // Initialize dense input matrix C.106    %c = tensor.generate %ksz, %jsz {107    ^bb0(%k : index, %j : index):108      %k0 = arith.muli %k, %jsz : index109      %k1 = arith.addi %k0, %j : index110      %k2 = arith.index_cast %k1 : index to i32111      %kf = arith.sitofp %k2 : i32 to f64112      tensor.yield %kf : f64113    } : tensor<?x?xf64>114 115    // Initialize dense input matrix D.116    %d = tensor.generate %lsz, %jsz {117    ^bb0(%l : index, %j : index):118      %k0 = arith.muli %l, %jsz : index119      %k1 = arith.addi %k0, %j : index120      %k2 = arith.index_cast %k1 : index to i32121      %kf = arith.sitofp %k2 : i32 to f64122      tensor.yield %kf : f64123    } : tensor<?x?xf64>124 125    // Initialize dense output matrix A.126    %a = tensor.generate %isz, %jsz {127    ^bb0(%i : index, %j: index):128      tensor.yield %f0 : f64129    } : tensor<?x?xf64>130 131    // Call kernel.132    %0 = call @kernel_mttkrp(%b, %c, %d, %a)133      : (tensor<?x?x?xf64, #SparseTensor>,134        tensor<?x?xf64>, tensor<?x?xf64>, tensor<?x?xf64>) -> tensor<?x?xf64>135 136    // Print the result for verification.137    //138    // CHECK:      {{\[}}[16075,   21930,   28505,   35800,   43815],139    // CHECK-NEXT: [10000,   14225,   19180,   24865,   31280]]140    //141    %u = tensor.cast %0: tensor<?x?xf64> to tensor<*xf64>142    call @printMemrefF64(%u) : (tensor<*xf64>) -> ()143 144    // Release the resources.145    bufferization.dealloc_tensor %b : tensor<?x?x?xf64, #SparseTensor>146    bufferization.dealloc_tensor %c : tensor<?x?xf64>147    bufferization.dealloc_tensor %d : tensor<?x?xf64>148    bufferization.dealloc_tensor %a : tensor<?x?xf64>149 150    return151  }152}153