1// Note: We run CSE here to make the pattern matching more direct. 2 3// RUN: mlir-opt %s -test-lower-to-llvm -cse -canonicalize | FileCheck %s 4 5// RUN: mlir-opt %s \ 6// RUN: -transform-preload-library="transform-library-paths=%p/lower-to-llvm-transform-symbol-def.mlir" \ 7// RUN: -transform-interpreter="debug-payload-root-tag=payload" \ 8// RUN: -test-transform-dialect-erase-schedule -cse -canonicalize \ 9// RUN: | FileCheck %s 10 11module attributes {transform.target_tag="payload"} { 12 13// Check that we properly lower to llvm memref operations that require to be 14// expanded first, like `memref.subview`. 15func.func @subview(%0 : memref<64x4xf32, strided<[4, 1], offset: 0>>, %arg0 : index, %arg1 : index, %arg2 : index) 16-> memref<?x?xf32, strided<[?, ?], offset: ?>> { 17 // CHECK-LABEL: @subview 18 // CHECK-SAME: %[[BASE:[^:]*]]: !llvm.ptr 19 // CHECK-SAME: %[[BASE_ALIGNED:[^:]*]]: !llvm.ptr, 20 // CHECK-SAME: %[[BASE_OFFSET:[^:]*]]: i64, 21 // CHECK-SAME: %[[BASE_STRIDE0:[^:]*]]: i64, 22 // CHECK-SAME: %[[BASE_STRIDE1:[^:]*]]: i64, 23 // CHECK-SAME: %[[BASE_SIZE0:[^:]*]]: i64, 24 // CHECK-SAME: %[[BASE_SIZE1:[^:]*]]: i64, 25 // CHECK-SAME: %[[ARG0:[^:]*]]: i64, 26 // CHECK-SAME: %[[ARG1:[^:]*]]: i64, 27 // CHECK-SAME: %[[ARG2:[^:]*]]: i64) 28 // CHECK-SAME: -> !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64> 29 30 // CHECK-DAG: %[[STRIDE0:.*]] = llvm.mlir.constant(4 : index) : i64 31 // CHECK-DAG: %[[DESCSTRIDE0:.*]] = llvm.mul %[[ARG0]], %[[STRIDE0]] overflow<nsw> : i64 32 // CHECK-DAG: %[[OFF2:.*]] = llvm.add %[[DESCSTRIDE0]], %[[ARG1]] : i64 33 // CHECK-DAG: %[[DESC:.*]] = llvm.mlir.undef : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 34 35 // Base address and algined address. 36 // CHECK-DAG: %[[DESC0:.*]] = llvm.insertvalue %[[BASE]], %[[DESC]][0] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 37 // CHECK-DAG: %[[DESC1:.*]] = llvm.insertvalue %[[BASE_ALIGNED]], %[[DESC0]][1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 38 39 // Offset. 40 // CHECK: %[[DESC2:.*]] = llvm.insertvalue %[[OFF2]], %[[DESC1]][2] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 41 // Size 0. 42 // CHECK: %[[DESC3:.*]] = llvm.insertvalue %[[ARG0]], %[[DESC2]][3, 0] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 43 // Stride 0 == 4 * %arg0. 44 // CHECK: %[[DESC4:.*]] = llvm.insertvalue %[[DESCSTRIDE0]], %[[DESC3]][4, 0] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 45 // Size 1. 46 // CHECK: %[[DESC5:.*]] = llvm.insertvalue %[[ARG1]], %[[DESC4]][3, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 47 // Stride 1 == 1 * %arg1. 48 // CHECK: %[[DESC6:.*]] = llvm.insertvalue %[[ARG1]], %[[DESC5]][4, 1] : !llvm.struct<(ptr, ptr, i64, array<2 x i64>, array<2 x i64>)> 49 50 %1 = memref.subview %0[%arg0, %arg1][%arg0, %arg1][%arg0, %arg1] : 51 memref<64x4xf32, strided<[4, 1], offset: 0>> 52 to memref<?x?xf32, strided<[?, ?], offset: ?>> 53 return %1 : memref<?x?xf32, strided<[?, ?], offset: ?>> 54} 55 56} // transform payload 57 58module @named_inclusion_in_named attributes { transform.with_named_sequence } { 59 transform.named_sequence private @lower_to_cpu(!transform.any_op {transform.consumed}) -> !transform.any_op 60 61 transform.named_sequence @__transform_main(%toplevel_module: !transform.any_op {transform.consumed}) { 62 %m2 = transform.include @lower_to_cpu failures(suppress) (%toplevel_module) 63 : (!transform.any_op) -> (!transform.any_op) 64 transform.yield 65 } 66} 67