1 // RUN: fir-opt -split-input-file -verify-diagnostics %s
4 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
6 %pinned = fir.alloca i1
7 %4:2 = hlfir.declare %0 {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFsub1Ea"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>)
8 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
9 %s = fir.load %1 : !fir.ref<i32>
10 // expected-error@+1{{'cuf.allocate' op pinned and stream cannot appears at the same time}}
11 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> stream(%s : i32) pinned(%pinned : !fir.ref<i1>) {data_attr = #cuf.cuda<device>} -> i32
17 func.func @_QPsub1() {
19 // expected-error@+1{{'cuf.allocate' op expect box to be a reference to a class or box type value}}
20 %2 = cuf.allocate %1 : !fir.ref<i32> {data_attr = #cuf.cuda<device>} -> i32
26 func.func @_QPsub1() {
27 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
28 %4:2 = hlfir.declare %0 {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFsub1Ea"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>)
29 %c100 = arith.constant 100 : index
30 %7 = fir.alloca !fir.char<1,100> {bindc_name = "msg", uniq_name = "_QFsub1Emsg"}
31 %8:2 = hlfir.declare %7 typeparams %c100 {uniq_name = "_QFsub1Emsg"} : (!fir.ref<!fir.char<1,100>>, index) -> (!fir.ref<!fir.char<1,100>>, !fir.ref<!fir.char<1,100>>)
32 %9 = fir.embox %8#1 : (!fir.ref<!fir.char<1,100>>) -> !fir.box<!fir.char<1,100>>
33 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
34 %16 = fir.convert %9 : (!fir.box<!fir.char<1,100>>) -> !fir.box<none>
35 // expected-error@+1{{'cuf.allocate' op expect stat attribute when errmsg is provided}}
36 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> errmsg(%16 : !fir.box<none>) {data_attr = #cuf.cuda<device>} -> i32
42 func.func @_QPsub1() {
43 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
44 %4:2 = hlfir.declare %0 {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFsub1Ea"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>)
46 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
47 // expected-error@+1{{'cuf.allocate' op expect errmsg to be a reference to/or a box type value}}
48 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> errmsg(%1 : !fir.ref<i32>) {data_attr = #cuf.cuda<device>, hasStat} -> i32
54 func.func @_QPsub1() {
56 // expected-error@+1{{'cuf.deallocate' op expect box to be a reference to class or box type value}}
57 %2 = cuf.deallocate %1 : !fir.ref<i32> {data_attr = #cuf.cuda<device>} -> i32
63 func.func @_QPsub1() {
64 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
65 %4:2 = hlfir.declare %0 {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFsub1Ea"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>)
67 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
68 // expected-error@+1{{'cuf.deallocate' op expect errmsg to be a reference to/or a box type value}}
69 %13 = cuf.deallocate %11 : !fir.ref<!fir.box<none>> errmsg(%1 : !fir.ref<i32>) {data_attr = #cuf.cuda<device>, hasStat} -> i32
75 func.func @_QPsub1() {
76 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
77 %4:2 = hlfir.declare %0 {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QFsub1Ea"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>, !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>)
78 %c100 = arith.constant 100 : index
79 %7 = fir.alloca !fir.char<1,100> {bindc_name = "msg", uniq_name = "_QFsub1Emsg"}
80 %8:2 = hlfir.declare %7 typeparams %c100 {uniq_name = "_QFsub1Emsg"} : (!fir.ref<!fir.char<1,100>>, index) -> (!fir.ref<!fir.char<1,100>>, !fir.ref<!fir.char<1,100>>)
81 %9 = fir.embox %8#1 : (!fir.ref<!fir.char<1,100>>) -> !fir.box<!fir.char<1,100>>
82 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
83 %16 = fir.convert %9 : (!fir.box<!fir.char<1,100>>) -> !fir.box<none>
84 // expected-error@+1{{'cuf.deallocate' op expect stat attribute when errmsg is provided}}
85 %13 = cuf.deallocate %11 : !fir.ref<!fir.box<none>> errmsg(%16 : !fir.box<none>) {data_attr = #cuf.cuda<device>} -> i32
91 func.func @_QPsub1() {
92 %0 = cuf.alloc f32 {bindc_name = "r", data_attr = #cuf.cuda<device>, uniq_name = "_QFsub1Er"} -> !fir.ref<f32>
93 // expected-error@+1{{'cuf.free' op expect device, managed, pinned or unified cuda attribute}}
94 cuf.free %0 : !fir.ref<f32> {data_attr = #cuf.cuda<constant>}
100 func.func @_QPsub1(%arg0: !fir.ref<!fir.array<?xf32>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "adev"}, %arg1: !fir.ref<!fir.array<?xf32>> {fir.bindc_name = "ahost"}, %arg2: !fir.ref<i32> {fir.bindc_name = "n"}, %arg3: !fir.ref<i32> {fir.bindc_name = "m"}) {
101 %0 = fir.dummy_scope : !fir.dscope
102 %1:2 = hlfir.declare %arg2 dummy_scope %0 {uniq_name = "_QFsub1En"} : (!fir.ref<i32>, !fir.dscope) -> (!fir.ref<i32>, !fir.ref<i32>)
103 %2:2 = hlfir.declare %arg3 dummy_scope %0 {uniq_name = "_QFsub1Em"} : (!fir.ref<i32>, !fir.dscope) -> (!fir.ref<i32>, !fir.ref<i32>)
104 %3 = fir.load %1#0 : !fir.ref<i32>
105 %4 = fir.load %2#0 : !fir.ref<i32>
106 %5 = arith.muli %3, %4 : i32
107 %6 = fir.convert %5 : (i32) -> i64
108 %7 = fir.convert %6 : (i64) -> index
109 %c0 = arith.constant 0 : index
110 %8 = arith.cmpi sgt, %7, %c0 : index
111 %9 = arith.select %8, %7, %c0 : index
112 %10 = fir.shape %9 : (index) -> !fir.shape<1>
113 %11:2 = hlfir.declare %arg0(%10) dummy_scope %0 {data_attr = #cuf.cuda<device>, uniq_name = "_QFsub1Eadev"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>, !fir.dscope) -> (!fir.box<!fir.array<?xf32>>, !fir.ref<!fir.array<?xf32>>)
114 %12 = fir.load %1#0 : !fir.ref<i32>
115 %13 = fir.load %2#0 : !fir.ref<i32>
116 %14 = arith.muli %12, %13 : i32
117 %15 = fir.convert %14 : (i32) -> i64
118 %16 = fir.convert %15 : (i64) -> index
119 %c0_0 = arith.constant 0 : index
120 %17 = arith.cmpi sgt, %16, %c0_0 : index
121 %18 = arith.select %17, %16, %c0_0 : index
122 %19 = fir.shape %18 : (index) -> !fir.shape<1>
123 %20:2 = hlfir.declare %arg1(%19) dummy_scope %0 {uniq_name = "_QFsub1Eahost"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>, !fir.dscope) -> (!fir.box<!fir.array<?xf32>>, !fir.ref<!fir.array<?xf32>>)
124 // expected-error@+1{{'cuf.data_transfer' op shape can only be specified on data transfer with references}}
125 cuf.data_transfer %20#0 to %11#0, %19 : !fir.shape<1> {transfer_kind = #cuf.cuda_transfer<host_device>} : !fir.box<!fir.array<?xf32>>, !fir.box<!fir.array<?xf32>>
131 module attributes {gpu.container_module} {
132 gpu.module @cuda_device_mod {
133 gpu.func @_QPsub_device1() {
137 llvm.func internal @__cudaFortranConstructor() {
138 %0 = cuf.register_module @cuda_device_mod -> !llvm.ptr
139 // expected-error@+1{{'cuf.register_kernel' op only kernel gpu.func can be registered}}
140 cuf.register_kernel @cuda_device_mod::@_QPsub_device1(%0 : !llvm.ptr)
147 module attributes {gpu.container_module} {
148 gpu.module @cuda_device_mod {
149 gpu.func @_QPsub_device1() {
153 llvm.func internal @__cudaFortranConstructor() {
154 %0 = cuf.register_module @cuda_device_mod -> !llvm.ptr
155 // expected-error@+1{{'cuf.register_kernel' op device function not found}}
156 cuf.register_kernel @cuda_device_mod::@_QPsub_device2(%0 : !llvm.ptr)
163 module attributes {gpu.container_module} {
164 llvm.func internal @__cudaFortranConstructor() {
165 %0 = cuf.register_module @cuda_device_mod -> !llvm.ptr
166 // expected-error@+1{{'cuf.register_kernel' op gpu module not found}}
167 cuf.register_kernel @cuda_device_mod::@_QPsub_device1(%0 : !llvm.ptr)
174 module attributes {gpu.container_module} {
175 llvm.func internal @__cudaFortranConstructor() {
176 %0 = cuf.register_module @cuda_device_mod -> !llvm.ptr
177 // expected-error@+1{{'cuf.register_kernel' op expect a module and a kernel name}}
178 cuf.register_kernel @_QPsub_device1(%0 : !llvm.ptr)
185 module attributes {gpu.container_module} {
186 gpu.module @cuda_device_mod {
187 llvm.func @_QPsub_device1() {
191 llvm.func internal @__cudaFortranConstructor() {
192 %0 = cuf.register_module @cuda_device_mod -> !llvm.ptr
193 // expected-error@+1{{'cuf.register_kernel' op only gpu.kernel llvm.func can be registered}}
194 cuf.register_kernel @cuda_device_mod::@_QPsub_device1(%0 : !llvm.ptr)