1// RUN: mlir-opt %s \ 2// RUN: --sparsification --sparse-tensor-conversion \ 3// RUN: --convert-linalg-to-loops --convert-vector-to-scf --convert-scf-to-std \ 4// RUN: --func-bufferize --tensor-constant-bufferize --tensor-bufferize \ 5// RUN: --std-bufferize --finalizing-bufferize \ 6// RUN: --convert-vector-to-llvm --convert-std-to-llvm | \ 7// RUN: TENSOR0="%mlir_integration_test_dir/data/wide.mtx" \ 8// RUN: mlir-cpu-runner \ 9// RUN: -e entry -entry-point-result=void \ 10// RUN: -shared-libs=%mlir_integration_test_dir/libmlir_c_runner_utils%shlibext | \ 11// RUN: FileCheck %s 12 13!Filename = type !llvm.ptr<i8> 14 15#SparseMatrix = #sparse_tensor.encoding<{ 16 dimLevelType = [ "dense", "compressed" ], 17 pointerBitWidth = 8, 18 indexBitWidth = 8 19}> 20 21#matvec = { 22 indexing_maps = [ 23 affine_map<(i,j) -> (i,j)>, // A 24 affine_map<(i,j) -> (j)>, // b 25 affine_map<(i,j) -> (i)> // x (out) 26 ], 27 iterator_types = ["parallel", "reduction"], 28 doc = "X(i) += A(i,j) * B(j)" 29} 30 31// 32// Integration test that lowers a kernel annotated as sparse to 33// actual sparse code, initializes a matching sparse storage scheme 34// from file, and runs the resulting code with the JIT compiler. 35// 36module { 37 // 38 // A kernel that multiplies a sparse matrix A with a dense vector b 39 // into a dense vector x. 40 // 41 func @kernel_matvec(%arga: tensor<?x?xi32, #SparseMatrix>, 42 %argb: tensor<?xi32>, 43 %argx: tensor<?xi32>) -> tensor<?xi32> { 44 %0 = linalg.generic #matvec 45 ins(%arga, %argb: tensor<?x?xi32, #SparseMatrix>, tensor<?xi32>) 46 outs(%argx: tensor<?xi32>) { 47 ^bb(%a: i32, %b: i32, %x: i32): 48 %0 = muli %a, %b : i32 49 %1 = addi %x, %0 : i32 50 linalg.yield %1 : i32 51 } -> tensor<?xi32> 52 return %0 : tensor<?xi32> 53 } 54 55 func private @getTensorFilename(index) -> (!Filename) 56 57 // 58 // Main driver that reads matrix from file and calls the sparse kernel. 59 // 60 func @entry() { 61 %i0 = constant 0 : i32 62 %c0 = constant 0 : index 63 %c1 = constant 1 : index 64 %c4 = constant 4 : index 65 %c256 = constant 256 : index 66 67 // Read the sparse matrix from file, construct sparse storage. 68 %fileName = call @getTensorFilename(%c0) : (index) -> (!Filename) 69 %a = sparse_tensor.new %fileName : !llvm.ptr<i8> to tensor<?x?xi32, #SparseMatrix> 70 71 // Initialize dense vectors. 72 %bdata = memref.alloc(%c256) : memref<?xi32> 73 %xdata = memref.alloc(%c4) : memref<?xi32> 74 scf.for %i = %c0 to %c256 step %c1 { 75 %k = addi %i, %c1 : index 76 %j = index_cast %k : index to i32 77 memref.store %j, %bdata[%i] : memref<?xi32> 78 } 79 scf.for %i = %c0 to %c4 step %c1 { 80 memref.store %i0, %xdata[%i] : memref<?xi32> 81 } 82 %b = memref.tensor_load %bdata : memref<?xi32> 83 %x = memref.tensor_load %xdata : memref<?xi32> 84 85 // Call kernel. 86 %0 = call @kernel_matvec(%a, %b, %x) 87 : (tensor<?x?xi32, #SparseMatrix>, tensor<?xi32>, tensor<?xi32>) -> tensor<?xi32> 88 89 // Print the result for verification. 90 // 91 // CHECK: ( 889, 1514, -21, -3431 ) 92 // 93 %m = memref.buffer_cast %0 : memref<?xi32> 94 %v = vector.transfer_read %m[%c0], %i0: memref<?xi32>, vector<4xi32> 95 vector.print %v : vector<4xi32> 96 97 // Release the resources. 98 memref.dealloc %bdata : memref<?xi32> 99 memref.dealloc %xdata : memref<?xi32> 100 101 return 102 } 103} 104