| 1 |
// RUN: fir-opt --split-input-file --cuf-convert %s | FileCheck %s |
| 2 |
|
| 3 |
|
| 4 |
module attributes {gpu.container_module, 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>>} { |
| 5 |
gpu.module @cuda_device_mod { |
| 6 |
gpu.func @_QPsub_device1() kernel { |
| 7 |
cf.br ^bb1 |
| 8 |
^bb1: // pred: ^bb0 |
| 9 |
gpu.return |
| 10 |
} |
| 11 |
gpu.func @_QPsub_device2(%arg0: !fir.ref<f32>) kernel { |
| 12 |
cf.br ^bb1(%arg0 : !fir.ref<f32>) |
| 13 |
^bb1(%0: !fir.ref<f32>): // pred: ^bb0 |
| 14 |
%1 = fir.declare %0 {uniq_name = "_QFsub1Ei"} : (!fir.ref<f32>) -> !fir.ref<f32> |
| 15 |
%cst = arith.constant 2.000000e+00 : f32 |
| 16 |
fir.store %cst to %1 : !fir.ref<f32> |
| 17 |
gpu.return |
| 18 |
} |
| 19 |
} |
| 20 |
|
| 21 |
func.func @_QQmain() attributes {fir.bindc_name = "main"} { |
| 22 |
%0 = fir.alloca f32 |
| 23 |
// CHECK: %[[ALLOCA:.*]] = fir.alloca f32 |
| 24 |
%c1 = arith.constant 1 : index |
| 25 |
%c11_i32 = arith.constant 11 : i32 |
| 26 |
%c1024_i32 = arith.constant 1024 : i32 |
| 27 |
%c6_i32 = arith.constant 6 : i32 |
| 28 |
%c1_i32 = arith.constant 1 : i32 |
| 29 |
// CHECK: gpu.launch_func @cuda_device_mod::@_QPsub_device1 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %c0{{.*}} {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 30 |
cuf.kernel_launch @cuda_device_mod::@_QPsub_device1<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32>>>() |
| 31 |
|
| 32 |
// CHECK: gpu.launch_func @cuda_device_mod::@_QPsub_device1 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %c1024{{.*}} {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 33 |
cuf.kernel_launch @cuda_device_mod::@_QPsub_device1<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1024_i32>>>() |
| 34 |
|
| 35 |
// CHECK: gpu.launch_func @cuda_device_mod::@_QPsub_device2 blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %c0{{.*}} args(%[[ALLOCA]] : !fir.ref<f32>) {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 36 |
cuf.kernel_launch @cuda_device_mod::@_QPsub_device2<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32>>>(%0) : (!fir.ref<f32>) |
| 37 |
return |
| 38 |
} |
| 39 |
|
| 40 |
} |
| 41 |
|
| 42 |
// ----- |
| 43 |
|
| 44 |
module attributes {gpu.container_module, 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>>} { |
| 45 |
gpu.module @cuda_device_mod { |
| 46 |
gpu.func @_QMmod1Psub1(%arg0: !fir.ref<!fir.array<10xi32>>) kernel { |
| 47 |
gpu.return |
| 48 |
} |
| 49 |
} |
| 50 |
|
| 51 |
func.func @_QMmod1Psub1(%arg0: !fir.ref<!fir.array<10xi32>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "adev"}) attributes {cuf.cluster_dims = #cuf.cluster_dims<x = 2 : i64, y = 2 : i64, z = 1 : i64>, cuf.proc_attr = #cuf.cuda_proc<global>} { |
| 52 |
return |
| 53 |
} |
| 54 |
func.func @_QMmod1Phost_sub() { |
| 55 |
%c10 = arith.constant 10 : index |
| 56 |
%0 = cuf.alloc !fir.array<10xi32> {bindc_name = "adev", data_attr = #cuf.cuda<device>, uniq_name = "_QMmod1Fhost_subEadev"} -> !fir.ref<!fir.array<10xi32>> |
| 57 |
%1 = fir.shape %c10 : (index) -> !fir.shape<1> |
| 58 |
%2:2 = hlfir.declare %0(%1) {data_attr = #cuf.cuda<device>, uniq_name = "_QMmod1Fhost_subEadev"} : (!fir.ref<!fir.array<10xi32>>, !fir.shape<1>) -> (!fir.ref<!fir.array<10xi32>>, !fir.ref<!fir.array<10xi32>>) |
| 59 |
%c1_i32 = arith.constant 1 : i32 |
| 60 |
%c10_i32 = arith.constant 10 : i32 |
| 61 |
cuf.kernel_launch @_QMmod1Psub1<<<%c1_i32, %c1_i32, %c1_i32, %c10_i32, %c1_i32, %c1_i32>>>(%2#1) : (!fir.ref<!fir.array<10xi32>>) |
| 62 |
return |
| 63 |
} |
| 64 |
} |
| 65 |
|
| 66 |
// CHECK-LABEL: func.func @_QMmod1Phost_sub() |
| 67 |
// CHECK: gpu.launch_func @cuda_device_mod::@_QMmod1Psub1 clusters in (%c2{{.*}}, %c2{{.*}}, %c1{{.*}}) {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 68 |
|
| 69 |
// ----- |
| 70 |
|
| 71 |
module attributes {gpu.container_module, 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>>} { |
| 72 |
gpu.module @cuda_device_mod { |
| 73 |
gpu.func @_QMdevptrPtest(%arg0: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>) kernel { |
| 74 |
gpu.return |
| 75 |
} |
| 76 |
} |
| 77 |
fir.global @_QMdevptrEdev_ptr {data_attr = #cuf.cuda<device>} : !fir.box<!fir.ptr<!fir.array<?xf32>>> { |
| 78 |
%c0 = arith.constant 0 : index |
| 79 |
%0 = fir.zero_bits !fir.ptr<!fir.array<?xf32>> |
| 80 |
%1 = fir.shape %c0 : (index) -> !fir.shape<1> |
| 81 |
%2 = fir.embox %0(%1) {allocator_idx = 2 : i32} : (!fir.ptr<!fir.array<?xf32>>, !fir.shape<1>) -> !fir.box<!fir.ptr<!fir.array<?xf32>>> |
| 82 |
fir.has_value %2 : !fir.box<!fir.ptr<!fir.array<?xf32>>> |
| 83 |
} |
| 84 |
func.func @_QMdevptrPtest(%arg0: !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "dp"}) attributes {cuf.proc_attr = #cuf.cuda_proc<global>} { |
| 85 |
return |
| 86 |
} |
| 87 |
func.func @_QQmain() { |
| 88 |
%c1_i32 = arith.constant 1 : i32 |
| 89 |
%c4 = arith.constant 4 : index |
| 90 |
%0 = cuf.alloc !fir.array<4xf32> {bindc_name = "a_dev", data_attr = #cuf.cuda<device>, uniq_name = "_QFEa_dev"} -> !fir.ref<!fir.array<4xf32>> |
| 91 |
%1 = fir.shape %c4 : (index) -> !fir.shape<1> |
| 92 |
%2 = fir.declare %0(%1) {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<target>, uniq_name = "_QFEa_dev"} : (!fir.ref<!fir.array<4xf32>>, !fir.shape<1>) -> !fir.ref<!fir.array<4xf32>> |
| 93 |
%3 = fir.address_of(@_QMdevptrEdev_ptr) : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>> |
| 94 |
%4 = fir.declare %3 {data_attr = #cuf.cuda<device>, fortran_attrs = #fir.var_attrs<pointer>, uniq_name = "_QMdevptrEdev_ptr"} : (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>) -> !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>> |
| 95 |
%5 = fir.embox %2(%1) : (!fir.ref<!fir.array<4xf32>>, !fir.shape<1>) -> !fir.box<!fir.ptr<!fir.array<?xf32>>> |
| 96 |
fir.store %5 to %4 : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>> |
| 97 |
cuf.sync_descriptor @_QMdevptrEdev_ptr |
| 98 |
cuf.kernel_launch @_QMdevptrPtest<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32>>>(%4) : (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>) |
| 99 |
cuf.free %2 : !fir.ref<!fir.array<4xf32>> {data_attr = #cuf.cuda<device>} |
| 100 |
return |
| 101 |
} |
| 102 |
} |
| 103 |
|
| 104 |
// CHECK-LABEL: func.func @_QQmain() |
| 105 |
// CHECK: _FortranACUFSyncGlobalDescriptor |
| 106 |
// CHECK: %[[ADDROF:.*]] = fir.address_of(@_QMdevptrEdev_ptr) : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>> |
| 107 |
// CHECK: %[[CONV_ADDR:.*]] = fir.convert %[[ADDROF]] : (!fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>) -> !fir.llvm_ptr<i8> |
| 108 |
// CHECK: %[[DEVADDR:.*]] = fir.call @_FortranACUFGetDeviceAddress(%[[CONV_ADDR]], %{{.*}}, %{{.*}}) : (!fir.llvm_ptr<i8>, !fir.ref<i8>, i32) -> !fir.llvm_ptr<i8> |
| 109 |
// CHECK: %[[CONV_DEVADDR:.*]] = fir.convert %[[DEVADDR]] : (!fir.llvm_ptr<i8>) -> !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>> |
| 110 |
// CHECK: gpu.launch_func @cuda_device_mod::@_QMdevptrPtest blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %{{.*}} args(%[[CONV_DEVADDR]] : !fir.ref<!fir.box<!fir.ptr<!fir.array<?xf32>>>>) {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 111 |
|
| 112 |
// ----- |
| 113 |
|
| 114 |
module attributes {gpu.container_module, 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>>} { |
| 115 |
gpu.module @cuda_device_mod { |
| 116 |
gpu.func @_QMdevptrPtest() kernel { |
| 117 |
gpu.return |
| 118 |
} |
| 119 |
} |
| 120 |
func.func @_QMdevptrPtest() attributes {cuf.proc_attr = #cuf.cuda_proc<grid_global>} { |
| 121 |
return |
| 122 |
} |
| 123 |
func.func @_QQmain() { |
| 124 |
%c1_i32 = arith.constant 1 : i32 |
| 125 |
cuf.kernel_launch @_QMdevptrPtest<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32>>>() |
| 126 |
return |
| 127 |
} |
| 128 |
} |
| 129 |
|
| 130 |
// CHECK-LABEL: func.func @_QQmain() |
| 131 |
// CHECK: gpu.launch_func @cuda_device_mod::@_QMdevptrPtest blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) dynamic_shared_memory_size %{{.*}} {cuf.proc_attr = #cuf.cuda_proc<grid_global>} |
| 132 |
|
| 133 |
// ----- |
| 134 |
|
| 135 |
module attributes {gpu.container_module, 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>>} { |
| 136 |
gpu.module @cuda_device_mod { |
| 137 |
gpu.func @_QMdevptrPtest() kernel { |
| 138 |
gpu.return |
| 139 |
} |
| 140 |
} |
| 141 |
func.func @_QMdevptrPtest() attributes {cuf.proc_attr = #cuf.cuda_proc<grid_global>} { |
| 142 |
return |
| 143 |
} |
| 144 |
func.func @_QQmain() { |
| 145 |
%0 = fir.alloca i64 {bindc_name = "stream", uniq_name = "_QMtest_callFhostEstream"} |
| 146 |
%1:2 = hlfir.declare %0 {uniq_name = "_QMtest_callFhostEstream"} : (!fir.ref<i64>) -> (!fir.ref<i64>, !fir.ref<i64>) |
| 147 |
%c1_i32 = arith.constant 1 : i32 |
| 148 |
%c0_i32 = arith.constant 0 : i32 |
| 149 |
cuf.kernel_launch @_QMdevptrPtest<<<%c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c1_i32, %c0_i32, %1#0 : !fir.ref<i64>>>>() |
| 150 |
return |
| 151 |
} |
| 152 |
} |
| 153 |
|
| 154 |
// CHECK-LABEL: func.func @_QQmain() |
| 155 |
// CHECK: %[[STREAM:.*]] = fir.alloca i64 {bindc_name = "stream", uniq_name = "_QMtest_callFhostEstream"} |
| 156 |
// CHECK: %[[DECL_STREAM:.*]]:2 = hlfir.declare %[[STREAM]] {uniq_name = "_QMtest_callFhostEstream"} : (!fir.ref<i64>) -> (!fir.ref<i64>, !fir.ref<i64>) |
| 157 |
// CHECK: %[[TOKEN:.*]] = cuf.stream_cast %[[DECL_STREAM]]#0 : !fir.ref<i64> |
| 158 |
// CHECK: gpu.launch_func [%[[TOKEN]]] @cuda_device_mod::@_QMdevptrPtest |