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// REDEFINE: %{sparsifier_opts} = enable-runtime-library=true 22// RUN: %{compile} | %{run} | FileCheck %s 23// 24// Do the same run, but now with direct IR generation. 25// REDEFINE: %{sparsifier_opts} = enable-runtime-library=false enable-buffer-initialization=true 26// RUN: %{compile} | %{run} | FileCheck %s 27 28#CCCC = #sparse_tensor.encoding<{ map = (d0, d1, d2, d3) -> (d0 : compressed, d1 : compressed, d2 : compressed, d3 : compressed), posWidth = 32, crdWidth = 32 }> 29 30func.func @pooling_nhwc_sum_CCCC(%input: tensor<1x4x4x1xf32, #CCCC>, %filter: tensor<2x2xf32>) -> tensor<1x3x3x1xf32, #CCCC> { 31 %init = tensor.empty() : tensor<1x3x3x1xf32, #CCCC> 32 %0 = linalg.pooling_nhwc_sum {dilations = dense<1> : tensor<2xi64>, 33 strides = dense<1> : tensor<2xi64>} 34 ins (%input, %filter: tensor<1x4x4x1xf32, #CCCC>, tensor<2x2xf32>) 35 outs (%init: tensor<1x3x3x1xf32, #CCCC>) -> tensor<1x3x3x1xf32, #CCCC> 36 return %0 : tensor<1x3x3x1xf32, #CCCC> 37} 38 39func.func @pooling_nhwc_sum(%input: tensor<1x4x4x1xf32>, %filter: tensor<2x2xf32>) -> tensor<1x3x3x1xf32> { 40 %init = arith.constant dense<[[ [[0.0], [0.0], [0.0]], 41 [[0.0], [0.0], [0.0]], 42 [[0.0], [0.0], [0.0]] ]]> : tensor<1x3x3x1xf32> 43 %0 = linalg.pooling_nhwc_sum {dilations = dense<1> : tensor<2xi64>, 44 strides = dense<1> : tensor<2xi64>} 45 ins (%input, %filter: tensor<1x4x4x1xf32>, tensor<2x2xf32>) 46 outs (%init: tensor<1x3x3x1xf32>) -> tensor<1x3x3x1xf32> 47 return %0 : tensor<1x3x3x1xf32> 48} 49 50 51func.func @main() { 52 %c0 = arith.constant 0 : index 53 %zero = arith.constant 0.00000e+00 : f32 54 55 %filter = arith.constant dense< 56 [[ 1.0, 0.0], 57 [ 0.0, 1.0]] 58 > : tensor<2x2xf32> 59 60 %in_dense = arith.constant dense< 61 [[[[1.0], [2.0], [1.0], [2.0]], 62 [[1.0], [2.0], [1.0], [2.0]], 63 [[1.0], [2.0], [1.0], [2.0]], 64 [[1.0], [2.0], [1.0], [2.0]]]] 65 > : tensor<1x4x4x1xf32> 66 67 %in_CCCC = sparse_tensor.convert %in_dense : tensor<1x4x4x1xf32> to tensor<1x4x4x1xf32, #CCCC> 68 69 %dense_ret = call @pooling_nhwc_sum(%in_dense, %filter) : (tensor<1x4x4x1xf32>, tensor<2x2xf32>) -> tensor<1x3x3x1xf32> 70 %CCCC_ret = call @pooling_nhwc_sum_CCCC(%in_CCCC, %filter) : (tensor<1x4x4x1xf32, #CCCC>, tensor<2x2xf32>) -> tensor<1x3x3x1xf32, #CCCC> 71 72 // CHECK: ( ( ( ( 6 ), ( 6 ), ( 6 ) ), ( ( 6 ), ( 6 ), ( 6 ) ), ( ( 6 ), ( 6 ), ( 6 ) ) ) ) 73 %dense_v = vector.transfer_read %dense_ret[%c0, %c0, %c0, %c0], %zero 74 : tensor<1x3x3x1xf32>, vector<1x3x3x1xf32> 75 vector.print %dense_v : vector<1x3x3x1xf32> 76 77 // 78 // Sparse pooling should have the same output. 79 // 80 // CHECK: ---- Sparse Tensor ---- 81 // CHECK-NEXT: nse = 9 82 // CHECK-NEXT: dim = ( 1, 3, 3, 1 ) 83 // CHECK-NEXT: lvl = ( 1, 3, 3, 1 ) 84 // CHECK-NEXT: pos[0] : ( 0, 1 ) 85 // CHECK-NEXT: crd[0] : ( 0 ) 86 // CHECK-NEXT: pos[1] : ( 0, 3 ) 87 // CHECK-NEXT: crd[1] : ( 0, 1, 2 ) 88 // CHECK-NEXT: pos[2] : ( 0, 3, 6, 9 ) 89 // CHECK-NEXT: crd[2] : ( 0, 1, 2, 0, 1, 2, 0, 1, 2 ) 90 // CHECK-NEXT: pos[3] : ( 0, 1, 2, 3, 4, 5, 6, 7, 8, 9 ) 91 // CHECK-NEXT: crd[3] : ( 0, 0, 0, 0, 0, 0, 0, 0, 0 ) 92 // CHECK-NEXT: values : ( 6, 6, 6, 6, 6, 6, 6, 6, 6 ) 93 // CHECK-NEXT: ---- 94 // 95 sparse_tensor.print %CCCC_ret : tensor<1x3x3x1xf32, #CCCC> 96 97 // Releases resources. 98 bufferization.dealloc_tensor %in_CCCC : tensor<1x4x4x1xf32, #CCCC> 99 bufferization.dealloc_tensor %CCCC_ret : tensor<1x3x3x1xf32, #CCCC> 100 bufferization.dealloc_tensor %dense_ret : tensor<1x3x3x1xf32> 101 return 102} 103