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, this 5// config could be moved to lit.local.cfg. However, there are downstream users that 6// do not use these LIT config files. Hence why this is kept inline. 7// 8// DEFINE: %{sparsifier_opts} = enable-runtime-library=true 9// 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_utils 13// DEFINE: %{run_libs_sve} = -shared-libs=%native_mlir_runner_utils,%native_mlir_c_runner_utils 14// DEFINE: %{run_opts} = -e main -entry-point-result=void 15// 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 %s 22// 23// Do the same run, but now with direct IR generation. 24// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false enable-buffer-initialization=true 25// RUN: %{compile} | %{run} | FileCheck %s 26// 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=true 29// RUN: %{compile} | %{run} | FileCheck %s 30 31// Product reductions - kept in a separate file as these are not supported by 32// the AArch64 SVE backend (so the set-up is a bit different to 33// sparse_reducitons.mlir) 34 35#SparseVector = #sparse_tensor.encoding<{map = (d0) -> (d0 : compressed)}> 36#CSR = #sparse_tensor.encoding<{map = (d0, d1) -> (d0 : dense, d1 : compressed)}> 37#CSC = #sparse_tensor.encoding<{ 38 map = (d0, d1) -> (d1 : dense, d0 : compressed) 39}> 40 41// 42// Traits for tensor operations. 43// 44 45#trait_mat_reduce_rowwise = { 46 indexing_maps = [ 47 affine_map<(i,j) -> (i,j)>, // A (in) 48 affine_map<(i,j) -> (i)> // X (out) 49 ], 50 iterator_types = ["parallel", "reduction"], 51 doc = "X(i) = PROD_j A(i,j)" 52} 53 54module { 55 func.func @redProdLex(%arga: tensor<?x?xf64, #CSR>) -> tensor<?xf64, #SparseVector> { 56 %c0 = arith.constant 0 : index 57 %cf1 = arith.constant 1.0 : f64 58 %d0 = tensor.dim %arga, %c0 : tensor<?x?xf64, #CSR> 59 %xv = tensor.empty(%d0): tensor<?xf64, #SparseVector> 60 %0 = linalg.generic #trait_mat_reduce_rowwise 61 ins(%arga: tensor<?x?xf64, #CSR>) 62 outs(%xv: tensor<?xf64, #SparseVector>) { 63 ^bb(%a: f64, %b: f64): 64 %1 = sparse_tensor.reduce %a, %b, %cf1 : f64 { 65 ^bb0(%x: f64, %y: f64): 66 %2 = arith.mulf %x, %y : f64 67 sparse_tensor.yield %2 : f64 68 } 69 linalg.yield %1 : f64 70 } -> tensor<?xf64, #SparseVector> 71 return %0 : tensor<?xf64, #SparseVector> 72 } 73 74 func.func @redProdExpand(%arga: tensor<?x?xf64, #CSC>) -> tensor<?xf64, #SparseVector> { 75 %c0 = arith.constant 0 : index 76 %cf1 = arith.constant 1.0 : f64 77 %d0 = tensor.dim %arga, %c0 : tensor<?x?xf64, #CSC> 78 %xv = tensor.empty(%d0): tensor<?xf64, #SparseVector> 79 %0 = linalg.generic #trait_mat_reduce_rowwise 80 ins(%arga: tensor<?x?xf64, #CSC>) 81 outs(%xv: tensor<?xf64, #SparseVector>) { 82 ^bb(%a: f64, %b: f64): 83 %1 = sparse_tensor.reduce %a, %b, %cf1 : f64 { 84 ^bb0(%x: f64, %y: f64): 85 %2 = arith.mulf %x, %y : f64 86 sparse_tensor.yield %2 : f64 87 } 88 linalg.yield %1 : f64 89 } -> tensor<?xf64, #SparseVector> 90 return %0 : tensor<?xf64, #SparseVector> 91 } 92 93 94 // Driver method to call and verify vector kernels. 95 func.func @main() { 96 %c0 = arith.constant 0 : index 97 98 // Setup sparse matrices. 99 %m1 = arith.constant sparse< 100 [ [0,0], [0,1], [1,0], [2,2], [2,3], [2,4], [3,0], [3,2], [3,3] ], 101 [ 1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0 ] 102 > : tensor<4x5xf64> 103 %m2 = arith.constant sparse< 104 [ [0,0], [1,3], [2,0], [2,3], [3,1], [4,1] ], 105 [6.0, 5.0, 4.0, 3.0, 2.0, 11.0 ] 106 > : tensor<5x4xf64> 107 %sm1 = sparse_tensor.convert %m1 : tensor<4x5xf64> to tensor<?x?xf64, #CSR> 108 %sm2r = sparse_tensor.convert %m2 : tensor<5x4xf64> to tensor<?x?xf64, #CSR> 109 %sm2c = sparse_tensor.convert %m2 : tensor<5x4xf64> to tensor<?x?xf64, #CSC> 110 111 // Call sparse matrix kernels. 112 %1 = call @redProdLex(%sm1) : (tensor<?x?xf64, #CSR>) -> tensor<?xf64, #SparseVector> 113 %2 = call @redProdExpand(%sm2c) : (tensor<?x?xf64, #CSC>) -> tensor<?xf64, #SparseVector> 114 115 // 116 // Verify the results. 117 // 118 // CHECK: ---- Sparse Tensor ---- 119 // CHECK-NEXT: nse = 9 120 // CHECK-NEXT: dim = ( 4, 5 ) 121 // CHECK-NEXT: lvl = ( 4, 5 ) 122 // CHECK-NEXT: pos[1] : ( 0, 2, 3, 6, 9 ) 123 // CHECK-NEXT: crd[1] : ( 0, 1, 0, 2, 3, 4, 0, 2, 3 ) 124 // CHECK-NEXT: values : ( 1, 2, 3, 4, 5, 6, 7, 8, 9 ) 125 // CHECK-NEXT: ---- 126 // CHECK: ---- Sparse Tensor ---- 127 // CHECK-NEXT: nse = 6 128 // CHECK-NEXT: dim = ( 5, 4 ) 129 // CHECK-NEXT: lvl = ( 5, 4 ) 130 // CHECK-NEXT: pos[1] : ( 0, 1, 2, 4, 5, 6 ) 131 // CHECK-NEXT: crd[1] : ( 0, 3, 0, 3, 1, 1 ) 132 // CHECK-NEXT: values : ( 6, 5, 4, 3, 2, 11 ) 133 // CHECK-NEXT: ---- 134 // CHECK: ---- Sparse Tensor ---- 135 // CHECK-NEXT: nse = 4 136 // CHECK-NEXT: dim = ( 4 ) 137 // CHECK-NEXT: lvl = ( 4 ) 138 // CHECK-NEXT: pos[0] : ( 0, 4 ) 139 // CHECK-NEXT: crd[0] : ( 0, 1, 2, 3 ) 140 // CHECK-NEXT: values : ( 2, 3, 120, 504 ) 141 // CHECK-NEXT: ---- 142 // CHECK: ---- Sparse Tensor ---- 143 // CHECK-NEXT: nse = 5 144 // CHECK-NEXT: dim = ( 5 ) 145 // CHECK-NEXT: lvl = ( 5 ) 146 // CHECK-NEXT: pos[0] : ( 0, 5 ) 147 // CHECK-NEXT: crd[0] : ( 0, 1, 2, 3, 4 ) 148 // CHECK-NEXT: values : ( 6, 5, 12, 2, 11 ) 149 // CHECK-NEXT: ---- 150 // 151 sparse_tensor.print %sm1 : tensor<?x?xf64, #CSR> 152 sparse_tensor.print %sm2r : tensor<?x?xf64, #CSR> 153 sparse_tensor.print %1 : tensor<?xf64, #SparseVector> 154 sparse_tensor.print %2 : tensor<?xf64, #SparseVector> 155 156 // Release the resources. 157 bufferization.dealloc_tensor %sm1 : tensor<?x?xf64, #CSR> 158 bufferization.dealloc_tensor %sm2r : tensor<?x?xf64, #CSR> 159 bufferization.dealloc_tensor %sm2c : tensor<?x?xf64, #CSC> 160 bufferization.dealloc_tensor %1 : tensor<?xf64, #SparseVector> 161 bufferization.dealloc_tensor %2 : tensor<?xf64, #SparseVector> 162 return 163 } 164} 165