62 lines · plain
1// RUN: mlir-opt %s -convert-linalg-to-std -split-input-file | FileCheck %s2 3func.func private @printMemrefF32(memref<*xf32>)4 5// CHECK: func private @linalg_fill_f32_viewsxsxf32(f32, memref<?x?xf32, {{.*}}>) attributes {llvm.emit_c_interface}6// CHECK: func private @linalg_matmul_viewsxsxf32_viewsxsxf32_viewsxsxf32(memref<?x?xf32, {{.*}}>, memref<?x?xf32, {{.*}}>, memref<?x?xf32, {{.*}}>) attributes {llvm.emit_c_interface}7 8func.func @matmul(%A: memref<?x?xf32>, %B: memref<?x?xf32>) -> (memref<?x?xf32>) {9 %c0 = arith.constant 0 : index10 %c1 = arith.constant 1 : index11 %f0 = arith.constant 0.0 : f3212 %x = memref.dim %A, %c0 : memref<?x?xf32>13 %y = memref.dim %B, %c1 : memref<?x?xf32>14 %C = memref.alloc(%x, %y) : memref<?x?xf32>15 16 // CHECK: call @linalg_fill_f32_viewsxsxf32({{.*}}) : (f32, memref<?x?xf32, {{.*}}>)17 linalg.fill ins(%f0 : f32) outs(%C : memref<?x?xf32>)18 19 // CHECK: call @linalg_matmul_viewsxsxf32_viewsxsxf32_viewsxsxf32({{.*}}) : (memref<?x?xf32, {{.*}}>, memref<?x?xf32, {{.*}}>, memref<?x?xf32, {{.*}}>) -> ()20 linalg.matmul ins(%A, %B: memref<?x?xf32>, memref<?x?xf32>)21 outs(%C: memref<?x?xf32>)22 return %C : memref<?x?xf32>23}24 25// -----26 27#accesses = [28 affine_map<(d0) -> (d0)>,29 affine_map<(d0) -> (d0)>,30 affine_map<(d0) -> (d0)>31 ]32#trait = {33 doc = "...",34 indexing_maps = #accesses,35 library_call = "test",36 iterator_types = ["parallel"]37}38 39// CHECK: func.func private @linalg_copy_view32xf16as1_view32xf16as6(memref<32xf16, strided<[?], offset: ?>, 1>, memref<32xf16, strided<[?], offset: ?>, 6>) attributes {llvm.emit_c_interface}40// CHECK: func.func private @linalg_copy_view32xf16as6_view32xf16as1(memref<32xf16, strided<[?], offset: ?>, 6>, memref<32xf16, strided<[?], offset: ?>, 1>) attributes {llvm.emit_c_interface}41 42module {43 func.func @helper(%arg7: memref<32xf16, 1>, %arg8: memref<32xf16, 1>, %arg9: memref<32xf16, 1>) {44 %localA = memref.alloca() : memref<32xf16, 6>45 %localB = memref.alloca() : memref<32xf16, 6>46 %localOut = memref.alloca() : memref<32xf16, 6>47 linalg.copy ins(%arg8 : memref<32xf16, 1>) outs(%localA : memref<32xf16, 6>)48 linalg.copy ins(%arg9 : memref<32xf16, 1>) outs(%localB : memref<32xf16, 6>)49 50 linalg.generic #trait51 ins(%localA, %localB : memref<32xf16, 6>, memref<32xf16, 6>)52 outs(%localOut : memref<32xf16, 6>) {53 ^bb0(%0: f16, %1: f16, %2: f16) :54 %e = arith.addf %1, %0: f1655 linalg.yield %e : f1656 }57 58 linalg.copy ins(%localOut : memref<32xf16, 6>) outs(%arg7 : memref<32xf16, 1>)59 return60 }61}62