Debian Codegraph @

flang/test/Fir/CUDA/cuda-global-addr.mlir

raw
1 // RUN: fir-opt --split-input-file --cuf-convert %s | FileCheck %s
2
3 module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<f80, dense<128> : vector<2xi64>>, #dlti.dl_entry<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<f64, dense<64> : vector<2xi64>>, #dlti.dl_entry<f16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>} {
4 fir.global @_QMmod1Eadev {data_attr = #cuf.cuda<device>} : !fir.array<10xi32> {
5 %0 = fir.zero_bits !fir.array<10xi32>
6 fir.has_value %0 : !fir.array<10xi32>
7 }
8 func.func @_QQmain() attributes {fir.bindc_name = "test"} {
9 %c14_i32 = arith.constant 14 : i32
10 %c6_i32 = arith.constant 6 : i32
11 %c4 = arith.constant 4 : index
12 %c1_i32 = arith.constant 1 : i32
13 %c0_i32 = arith.constant 0 : i32
14 %c10 = arith.constant 10 : index
15 %1 = fir.shape %c10 : (index) -> !fir.shape<1>
16 %3 = fir.address_of(@_QMmod1Eadev) : !fir.ref<!fir.array<10xi32>>
17 %4 = fir.declare %3(%1) {data_attr = #cuf.cuda<device>, uniq_name = "_QMmod1Eadev"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> !fir.ref<!fir.array<10xi32>>
18 %5 = fir.alloca i32 {bindc_name = "i", uniq_name = "_QFEi"}
19 %6 = fir.declare %5 {uniq_name = "_QFEi"} : (!fir.ref<i32>) -> !fir.ref<i32>
20 fir.store %c0_i32 to %6 : !fir.ref<i32>
21 %7 = fir.array_coor %4(%1) %c4 : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>, index) -> !fir.ref<i32>
22 cuf.data_transfer %c1_i32 to %7 {transfer_kind = #cuf.cuda_transfer<host_device>} : i32, !fir.ref<i32>
23 return
24 }
25
26 }
27
28 // CHECK-LABEL: func.func @_QQmain()
29 // CHECK: fir.address_of(@_QMmod1Eadev) : !fir.ref<!fir.array<10xi32>>
30 // CHECK: %[[ADDR:.*]] = fir.address_of(@_QMmod1Eadev) : !fir.ref<!fir.array<10xi32>>
31 // CHECK: %[[ADDRPTR:.*]] = fir.convert %[[ADDR]] : (!fir.ref<!fir.array<10xi32>>) -> !fir.llvm_ptr<i8>
32 // CHECK: %[[DEVICE_ADDR:.*]] = fir.call @_FortranACUFGetDeviceAddress(%[[ADDRPTR]], %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.ref<i8>, i32) -> !fir.llvm_ptr<i8>
33 // CHECK: %[[DEVICE_ADDR_CONV:.*]] = fir.convert %[[DEVICE_ADDR]] : (!fir.llvm_ptr<i8>) -> !fir.ref<!fir.array<10xi32>>
34 // CHECK: %[[DECL:.*]] = fir.declare %[[DEVICE_ADDR_CONV]](%{{.*}}) {data_attr = #cuf.cuda<device>, uniq_name = "_QMmod1Eadev"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> !fir.ref<!fir.array<10xi32>>
35 // CHECK: %[[ARRAY_COOR:.*]] = fir.array_coor %[[DECL]](%{{.*}}) %c4{{.*}} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>, index) -> !fir.ref<i32>
36 // CHECK: %[[ARRAY_COOR_PTR:.*]] = fir.convert %[[ARRAY_COOR]] : (!fir.ref<i32>) -> !fir.llvm_ptr<i8>
37 // CHECK: fir.call @_FortranACUFDataTransferPtrPtr(%[[ARRAY_COOR_PTR]], %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.llvm_ptr<i8>, i64, i32, !fir.ref<i8>, i32) -> ()
38
39 // -----
40
41 module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<f80, dense<128> : vector<2xi64>>, #dlti.dl_entry<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<f64, dense<64> : vector<2xi64>>, #dlti.dl_entry<f16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>} {
42
43 fir.global @_QMdevmodEdarray {data_attr = #cuf.cuda<device>} : !fir.box<!fir.heap<!fir.array<?xf32>>> {
44 %c0 = arith.constant 0 : index
45 %0 = fir.zero_bits !fir.heap<!fir.array<?xf32>>
46 %1 = fir.shape %c0 : (index) -> !fir.shape<1>
47 %2 = fir.embox %0(%1) {allocator_idx = 2 : i32} : (!fir.heap<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.box<!fir.heap<!fir.array<?xf32>>>
48 fir.has_value %2 : !fir.box<!fir.heap<!fir.array<?xf32>>>
49 }
50 func.func @_QQmain() attributes {fir.bindc_name = "arraysize"} {
51 %0 = fir.address_of(@_QMiso_c_bindingECc_int) : !fir.ref<i32>
52 %1 = fir.declare %0 {fortran_attrs = #fir.var_attrs<parameter>, uniq_name = "_QMiso_c_bindingECc_int"} : (!fir.ref<i32>) -> !fir.ref<i32>
53 %2 = fir.address_of(@_QMdevmodEdarray) : !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>
54 %3 = fir.declare %2 {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<allocatable>, uniq_name = "_QMdevmodEdarray"} : (!fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<!fir.heap<!fir.array<?xf32>>>>
55 %4 = fir.alloca i32 {bindc_name = "exp", uniq_name = "_QFEexp"}
56 %5 = fir.declare %4 {uniq_name = "_QFEexp"} : (!fir.ref<i32>) -> !fir.ref<i32>
57 %6 = fir.alloca i32 {bindc_name = "hsize", uniq_name = "_QFEhsize"}
58 %7 = fir.declare %6 {uniq_name = "_QFEhsize"} : (!fir.ref<i32>) -> !fir.ref<i32>
59 return
60 }
61 fir.global @_QMiso_c_bindingECc_int constant : i32
62 }
63
64 // We cannot call _FortranACUFGetDeviceAddress on a constant global.
65 // There is no symbol for it and the call would result into an unresolved reference.
66 // CHECK-NOT: fir.call {{.*}}GetDeviceAddress
67
68 // -----
69
70 module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<f80, dense<128> : vector<2xi64>>, #dlti.dl_entry<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<f64, dense<64> : vector<2xi64>>, #dlti.dl_entry<f16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>} {
71 fir.global @_QMmod1Eadev {data_attr = #cuf.cuda<device>} : !fir.array<10xi32> {
72 %0 = fir.zero_bits !fir.array<10xi32>
73 fir.has_value %0 : !fir.array<10xi32>
74 }
75 func.func @_QQmain() attributes {fir.bindc_name = "test"} {
76 %dim = arith.constant 1 : index
77 gpu.launch blocks(%bx, %by, %bz) in (%grid_x = %dim, %grid_y = %dim, %grid_z = %dim)
78 threads(%tx, %ty, %tz) in (%block_x = %dim, %block_y = %dim, %block_z = %dim) {
79 %c10 = arith.constant 10 : index
80 %1 = fir.shape %c10 : (index) -> !fir.shape<1>
81 %3 = fir.address_of(@_QMmod1Eadev) : !fir.ref<!fir.array<10xi32>>
82 %4 = fir.declare %3(%1) {data_attr = #cuf.cuda<device>, uniq_name = "_QMmod1Eadev"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> !fir.ref<!fir.array<10xi32>>
83 gpu.terminator
84 }
85 return
86 }
87
88 // CHECK-LABEL: func.func @_QQmain()
89 // CHECK: gpu.launch
90 // CHECK-NOT: fir.call {{.*}}GetDeviceAddress
91
92 }