109 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=false25// 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 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#CSC = #sparse_tensor.encoding<{35 map = (d0, d1) -> (d1 : dense, d0 : compressed)36}>37 38module {39 //40 // Column-wise storage forces the ijk loop to permute into jki41 // so that access pattern expansion (workspace) needs to be42 // done along dimension with size 8.43 //44 func.func @matmul(%A: tensor<8x2xf64, #CSC>,45 %B: tensor<2x4xf64, #CSC>) -> tensor<8x4xf64, #CSC> {46 %C = tensor.empty() : tensor<8x4xf64, #CSC>47 %D = linalg.matmul48 ins(%A, %B: tensor<8x2xf64, #CSC>, tensor<2x4xf64, #CSC>)49 outs(%C: tensor<8x4xf64, #CSC>) -> tensor<8x4xf64, #CSC>50 return %D: tensor<8x4xf64, #CSC>51 }52 53 //54 // Main driver.55 //56 func.func @main() {57 %c0 = arith.constant 0 : index58 %d1 = arith.constant -1.0 : f6459 60 // Initialize various dense matrices for stress testing.61 %da = arith.constant dense<[62 [ 1.1, 2.1 ],63 [ 1.2, 2.2 ],64 [ 1.3, 2.3 ],65 [ 1.4, 2.4 ],66 [ 1.5, 2.5 ],67 [ 1.6, 2.6 ],68 [ 1.7, 2.7 ],69 [ 1.8, 2.8 ]70 ]> : tensor<8x2xf64>71 %db = arith.constant dense<[72 [ 10.1, 11.1, 12.1, 13.1 ],73 [ 10.2, 11.2, 12.2, 13.2 ]74 ]> : tensor<2x4xf64>75 76 // Convert all these matrices to sparse format.77 %x1 = sparse_tensor.convert %da : tensor<8x2xf64> to tensor<8x2xf64, #CSC>78 %x2 = sparse_tensor.convert %db : tensor<2x4xf64> to tensor<2x4xf64, #CSC>79 80 // Call kernels with dense.81 %x3 = call @matmul(%x1, %x2)82 : (tensor<8x2xf64, #CSC>,83 tensor<2x4xf64, #CSC>) -> tensor<8x4xf64, #CSC>84 85 //86 // CHECK: ---- Sparse Tensor ----87 // CHECK-NEXT: nse = 3288 // CHECK-NEXT: dim = ( 8, 4 )89 // CHECK-NEXT: lvl = ( 4, 8 )90 // CHECK-NEXT: pos[1] : ( 0, 8, 16, 24, 32 )91 // CHECK-NEXT: crd[1] : ( 0, 1, 2, 3, 4, 5, 6, 7, 0, 1, 2, 3, 4, 5, 6, 7, 0,92 // CHECK-SAME: 1, 2, 3, 4, 5, 6, 7, 0, 1, 2, 3, 4, 5, 6, 7 )93 // CHECK-NEXT: values : ( 32.53, 34.56, 36.59, 38.62, 40.65, 42.68, 44.71, 46.74,94 // CHECK-SAME: 35.73, 37.96, 40.19, 42.42, 44.65, 46.88, 49.11, 51.34,95 // CHECK-SAME: 38.93, 41.36, 43.79, 46.22, 48.65, 51.08, 53.51, 55.94,96 // CHECK-SAME: 42.13, 44.76, 47.39, 50.02, 52.65, 55.28, 57.91, 60.54 )97 // CHECK-NEXT: ----98 //99 sparse_tensor.print %x3 : tensor<8x4xf64, #CSC>100 101 // Release the resources.102 bufferization.dealloc_tensor %x1 : tensor<8x2xf64, #CSC>103 bufferization.dealloc_tensor %x2 : tensor<2x4xf64, #CSC>104 bufferization.dealloc_tensor %x3 : tensor<8x4xf64, #CSC>105 106 return107 }108}109