Debian Codegraph @

flang/test/Fir/cuf.mlir

raw
1 // RUN: fir-opt --split-input-file %s | fir-opt --split-input-file | FileCheck %s
2
3 // Simple round trip test of operations.
4
5 func.func @_QPsub1() {
6 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
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 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> {data_attr = #cuf.cuda<device>} -> i32
10 %14 = cuf.deallocate %11 : !fir.ref<!fir.box<none>> {data_attr = #cuf.cuda<device>} -> i32
11 return
12 }
13
14 // CHECK: cuf.allocate %{{.*}} : !fir.ref<!fir.box<none>> {data_attr = #cuf.cuda<device>} -> i32
15 // CHECK: cuf.deallocate %{{.*}} : !fir.ref<!fir.box<none>> {data_attr = #cuf.cuda<device>} -> i32
16
17 // -----
18
19 func.func @_QPsub1() {
20 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
21 %1 = fir.alloca i64
22 %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>>>>)
23 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
24 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> stream(%1 : !fir.ref<i64>) {data_attr = #cuf.cuda<device>} -> i32
25 return
26 }
27
28 // CHECK: cuf.allocate %{{.*}} : !fir.ref<!fir.box<none>> stream(%{{.*}} : !fir.ref<i64>) {data_attr = #cuf.cuda<device>} -> i32
29
30 // -----
31
32 func.func @_QPsub1() {
33 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
34 %1 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "b", uniq_name = "_QFsub1Eb"}
35 %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>>>>)
36 %5:2 = hlfir.declare %1 {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>>>>)
37 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
38 %12 = fir.convert %5#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
39 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> source(%12 : !fir.ref<!fir.box<none>>) {data_attr = #cuf.cuda<device>} -> i32
40 return
41 }
42
43 // CHECK: cuf.allocate %{{.*}} : !fir.ref<!fir.box<none>> source(%{{.*}} : !fir.ref<!fir.box<none>>) {data_attr = #cuf.cuda<device>} -> i32
44
45 // -----
46
47 func.func @_QPsub1() {
48 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
49 %pinned = fir.alloca i1
50 %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>>>>)
51 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
52 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> pinned(%pinned : !fir.ref<i1>) {data_attr = #cuf.cuda<device>} -> i32
53 return
54 }
55
56 // CHECK: cuf.allocate %{{.*}} : !fir.ref<!fir.box<none>> pinned(%{{.*}} : !fir.ref<i1>) {data_attr = #cuf.cuda<device>} -> i32
57
58 // -----
59
60 func.func @_QPsub1() {
61 %0 = fir.alloca !fir.box<!fir.heap<!fir.array<?xf32>>> {bindc_name = "a", uniq_name = "_QFsub1Ea"}
62 %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>>>>)
63 %c100 = arith.constant 100 : index
64 %7 = fir.alloca !fir.char<1,100> {bindc_name = "msg", uniq_name = "_QFsub1Emsg"}
65 %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>>)
66 %9 = fir.embox %8#1 : (!fir.ref<!fir.char<1,100>>) -> !fir.box<!fir.char<1,100>>
67 %11 = fir.convert %4#1 : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<none>>
68 %16 = fir.convert %9 : (!fir.box<!fir.char<1,100>>) -> !fir.box<none>
69 %13 = cuf.allocate %11 : !fir.ref<!fir.box<none>> errmsg(%16 : !fir.box<none>) {data_attr = #cuf.cuda<device>, hasStat} -> i32
70 %14 = cuf.deallocate %11 : !fir.ref<!fir.box<none>> errmsg(%16 : !fir.box<none>) {data_attr = #cuf.cuda<device>, hasStat} -> i32
71 return
72 }
73
74 // CHECK: cuf.allocate %{{.*}} : !fir.ref<!fir.box<none>> errmsg(%{{.*}} : !fir.box<none>) {data_attr = #cuf.cuda<device>, hasStat} -> i32
75 // CHECK: cuf.deallocate %{{.*}} : !fir.ref<!fir.box<none>> errmsg(%{{.*}} : !fir.box<none>) {data_attr = #cuf.cuda<device>, hasStat} -> i32
76
77 // -----
78
79 func.func @_QPsub1() {
80 %0 = cuf.alloc f32 {bindc_name = "r", data_attr = #cuf.cuda<device>, uniq_name = "_QFsub1Er"} -> !fir.ref<f32>
81 cuf.free %0 : !fir.ref<f32> {data_attr = #cuf.cuda<device>}
82 return
83 }
84
85 // CHECK: cuf.alloc
86 // CHECK: cuf.free
87
88 // -----
89
90 gpu.module @cuda_device_mod {
91 gpu.func @_QPdynshared() kernel {
92 %c-1 = arith.constant -1 : index
93 %6 = cuf.shared_memory !fir.array<?xf32>, %c-1 : index {bindc_name = "r", uniq_name = "_QFdynsharedEr"} -> !fir.ref<!fir.array<?xf32>>
94 %7 = fir.shape %c-1 : (index) -> !fir.shape<1>
95 %8 = fir.declare %6(%7) {data_attr = #cuf.cuda<shared>, uniq_name = "_QFdynsharedEr"} : (!fir.ref<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.ref<!fir.array<?xf32>>
96 gpu.return
97 }
98 }
99
100 // CHECK: cuf.shared_memory
101
102 // -----
103
104 gpu.module @cuda_device_mod {
105 gpu.func @_QPshared_static() attributes {cuf.proc_attr = #cuf.cuda_proc<global>} {
106 %0 = cuf.shared_memory i32 {bindc_name = "a", uniq_name = "_QFshared_staticEa"} -> !fir.ref<i32>
107 %1 = fir.declare %0 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEa"} : (!fir.ref<i32>) -> !fir.ref<i32>
108 %2 = cuf.shared_memory i32 {bindc_name = "b", uniq_name = "_QFshared_staticEb"} -> !fir.ref<i32>
109 %3 = fir.declare %2 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEb"} : (!fir.ref<i32>) -> !fir.ref<i32>
110 gpu.return
111 }
112 }
113
114 // CHECK-COUNT-2: cuf.shared_memory