130 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: %{sparsifier_opts} = enable-runtime-library=false22// RUN: %{compile} | %{run} | FileCheck %s23//24// Do the same run, but now with direct IR generation and vectorization.25// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false vl=2 reassociate-fp-reductions=true enable-index-optimizations=true26// RUN: %{compile} | %{run} | FileCheck %s27//28// Do the same run, but now with direct IR generation and VLA vectorization.29// RUN: %if mlir_arm_sve_tests %{ %{compile_sve} | %{run_sve} | FileCheck %s %}30 31#SparseVector = #sparse_tensor.encoding<{map = (d0) -> (d0 : compressed)}>32 33#trait_op = {34 indexing_maps = [35 affine_map<(i) -> (i)>, // a (in)36 affine_map<(i) -> (i)>, // b (in)37 affine_map<(i) -> (i)> // x (out)38 ],39 iterator_types = ["parallel"],40 doc = "x(i) = a(i) OP b(i)"41}42 43module {44 func.func @cadd(%arga: tensor<?xcomplex<f64>, #SparseVector>,45 %argb: tensor<?xcomplex<f64>, #SparseVector>)46 -> tensor<?xcomplex<f64>, #SparseVector> {47 %c = arith.constant 0 : index48 %d = tensor.dim %arga, %c : tensor<?xcomplex<f64>, #SparseVector>49 %xv = tensor.empty(%d) : tensor<?xcomplex<f64>, #SparseVector>50 %0 = linalg.generic #trait_op51 ins(%arga, %argb: tensor<?xcomplex<f64>, #SparseVector>,52 tensor<?xcomplex<f64>, #SparseVector>)53 outs(%xv: tensor<?xcomplex<f64>, #SparseVector>) {54 ^bb(%a: complex<f64>, %b: complex<f64>, %x: complex<f64>):55 %1 = complex.add %a, %b : complex<f64>56 linalg.yield %1 : complex<f64>57 } -> tensor<?xcomplex<f64>, #SparseVector>58 return %0 : tensor<?xcomplex<f64>, #SparseVector>59 }60 61 func.func @cmul(%arga: tensor<?xcomplex<f64>, #SparseVector>,62 %argb: tensor<?xcomplex<f64>, #SparseVector>)63 -> tensor<?xcomplex<f64>, #SparseVector> {64 %c = arith.constant 0 : index65 %d = tensor.dim %arga, %c : tensor<?xcomplex<f64>, #SparseVector>66 %xv = tensor.empty(%d) : tensor<?xcomplex<f64>, #SparseVector>67 %0 = linalg.generic #trait_op68 ins(%arga, %argb: tensor<?xcomplex<f64>, #SparseVector>,69 tensor<?xcomplex<f64>, #SparseVector>)70 outs(%xv: tensor<?xcomplex<f64>, #SparseVector>) {71 ^bb(%a: complex<f64>, %b: complex<f64>, %x: complex<f64>):72 %1 = complex.mul %a, %b : complex<f64>73 linalg.yield %1 : complex<f64>74 } -> tensor<?xcomplex<f64>, #SparseVector>75 return %0 : tensor<?xcomplex<f64>, #SparseVector>76 }77 78 // Driver method to call and verify complex kernels.79 func.func @main() {80 // Setup sparse vectors.81 %v1 = arith.constant sparse<82 [ [0], [28], [31] ],83 [ (511.13, 2.0), (3.0, 4.0), (5.0, 6.0) ] > : tensor<32xcomplex<f64>>84 %v2 = arith.constant sparse<85 [ [1], [28], [31] ],86 [ (1.0, 0.0), (2.0, 0.0), (3.0, 0.0) ] > : tensor<32xcomplex<f64>>87 %sv1 = sparse_tensor.convert %v1 : tensor<32xcomplex<f64>> to tensor<?xcomplex<f64>, #SparseVector>88 %sv2 = sparse_tensor.convert %v2 : tensor<32xcomplex<f64>> to tensor<?xcomplex<f64>, #SparseVector>89 90 // Call sparse vector kernels.91 %0 = call @cadd(%sv1, %sv2)92 : (tensor<?xcomplex<f64>, #SparseVector>,93 tensor<?xcomplex<f64>, #SparseVector>) -> tensor<?xcomplex<f64>, #SparseVector>94 %1 = call @cmul(%sv1, %sv2)95 : (tensor<?xcomplex<f64>, #SparseVector>,96 tensor<?xcomplex<f64>, #SparseVector>) -> tensor<?xcomplex<f64>, #SparseVector>97 98 //99 // Verify the results.100 //101 // CHECK: ---- Sparse Tensor ----102 // CHECK-NEXT: nse = 4103 // CHECK-NEXT: dim = ( 32 )104 // CHECK-NEXT: lvl = ( 32 )105 // CHECK-NEXT: pos[0] : ( 0, 4 )106 // CHECK-NEXT: crd[0] : ( 0, 1, 28, 31 )107 // CHECK-NEXT: values : ( ( 511.13, 2 ), ( 1, 0 ), ( 5, 4 ), ( 8, 6 ) )108 // CHECK-NEXT: ----109 //110 // CHECK-NEXT: ---- Sparse Tensor ----111 // CHECK-NEXT: nse = 2112 // CHECK-NEXT: dim = ( 32 )113 // CHECK-NEXT: lvl = ( 32 )114 // CHECK-NEXT: pos[0] : ( 0, 2 )115 // CHECK-NEXT: crd[0] : ( 28, 31 )116 // CHECK-NEXT: values : ( ( 6, 8 ), ( 15, 18 ) )117 // CHECK-NEXT: ----118 //119 sparse_tensor.print %0 : tensor<?xcomplex<f64>, #SparseVector>120 sparse_tensor.print %1 : tensor<?xcomplex<f64>, #SparseVector>121 122 // Release the resources.123 bufferization.dealloc_tensor %sv1 : tensor<?xcomplex<f64>, #SparseVector>124 bufferization.dealloc_tensor %sv2 : tensor<?xcomplex<f64>, #SparseVector>125 bufferization.dealloc_tensor %0 : tensor<?xcomplex<f64>, #SparseVector>126 bufferization.dealloc_tensor %1 : tensor<?xcomplex<f64>, #SparseVector>127 return128 }129}130