| 1 |
// RUN: fir-opt --split-input-file --cuf-compute-shared-memory %s | FileCheck %s |
| 2 |
|
| 3 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, 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<f80, dense<128> : 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<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", gpu.container_module, llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.ident = "flang version 20.0.0 (https://github.com/llvm/llvm-project.git cae351f3453a0a26ec8eb2ddaf773c24a29d929e)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 4 |
gpu.module @cuda_device_mod { |
| 5 |
gpu.func @_QPdynshared() kernel { |
| 6 |
%c-1 = arith.constant -1 : index |
| 7 |
%6 = cuf.shared_memory !fir.array<?xf32>, %c-1 : index {bindc_name = "r", uniq_name = "_QFdynsharedEr"} -> !fir.ref<!fir.array<?xf32>> |
| 8 |
%7 = fir.shape %c-1 : (index) -> !fir.shape<1> |
| 9 |
%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>> |
| 10 |
gpu.return |
| 11 |
} |
| 12 |
} |
| 13 |
} |
| 14 |
|
| 15 |
// CHECK-LABEL: gpu.module @cuda_device_mod |
| 16 |
// CHECK: gpu.func @_QPdynshared() |
| 17 |
// CHECK: %{{.*}} = cuf.shared_memory[%c0{{.*}} : i32] !fir.array<?xf32>, %c-1 : index {bindc_name = "r", uniq_name = "_QFdynsharedEr"} -> !fir.ref<!fir.array<?xf32>> |
| 18 |
// CHECK: gpu.return |
| 19 |
// CHECK: } |
| 20 |
// CHECK: fir.global internal @_QPdynshared__shared_mem {alignment = 4 : i64, data_attr = #cuf.cuda<shared>} : !fir.array<0xi8> |
| 21 |
|
| 22 |
// ----- |
| 23 |
|
| 24 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, 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<f80, dense<128> : 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<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", gpu.container_module, llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.ident = "flang version 20.0.0 (https://github.com/llvm/llvm-project.git cae351f3453a0a26ec8eb2ddaf773c24a29d929e)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 25 |
gpu.module @cuda_device_mod { |
| 26 |
gpu.func @_QPshared_static() attributes {cuf.proc_attr = #cuf.cuda_proc<global>} { |
| 27 |
%0 = cuf.shared_memory i32 {bindc_name = "a", uniq_name = "_QFshared_staticEa"} -> !fir.ref<i32> |
| 28 |
%1 = fir.declare %0 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEa"} : (!fir.ref<i32>) -> !fir.ref<i32> |
| 29 |
%2 = cuf.shared_memory i32 {bindc_name = "b", uniq_name = "_QFshared_staticEb"} -> !fir.ref<i32> |
| 30 |
%3 = fir.declare %2 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEb"} : (!fir.ref<i32>) -> !fir.ref<i32> |
| 31 |
%8 = cuf.shared_memory i32 {bindc_name = "c", uniq_name = "_QFshared_staticEc"} -> !fir.ref<i32> |
| 32 |
%9 = fir.declare %8 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEc"} : (!fir.ref<i32>) -> !fir.ref<i32> |
| 33 |
%10 = cuf.shared_memory i32 {bindc_name = "d", uniq_name = "_QFshared_staticEd"} -> !fir.ref<i32> |
| 34 |
%11 = fir.declare %10 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEd"} : (!fir.ref<i32>) -> !fir.ref<i32> |
| 35 |
%12 = cuf.shared_memory i64 {bindc_name = "e", uniq_name = "_QFshared_staticEe"} -> !fir.ref<i64> |
| 36 |
%13 = fir.declare %12 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEe"} : (!fir.ref<i64>) -> !fir.ref<i64> |
| 37 |
%16 = cuf.shared_memory f32 {bindc_name = "r", uniq_name = "_QFshared_staticEr"} -> !fir.ref<f32> |
| 38 |
%17 = fir.declare %16 {data_attr = #cuf.cuda<shared>, uniq_name = "_QFshared_staticEr"} : (!fir.ref<f32>) -> !fir.ref<f32> |
| 39 |
gpu.return |
| 40 |
} |
| 41 |
} |
| 42 |
} |
| 43 |
|
| 44 |
// CHECK-LABEL: gpu.module @cuda_device_mod |
| 45 |
// CHECK: gpu.func @_QPshared_static() |
| 46 |
// CHECK: cuf.shared_memory[%c0{{.*}} : i32] i32 {bindc_name = "a", uniq_name = "_QFshared_staticEa"} -> !fir.ref<i32> |
| 47 |
// CHECK: cuf.shared_memory[%c4{{.*}} : i32] i32 {bindc_name = "b", uniq_name = "_QFshared_staticEb"} -> !fir.ref<i32> |
| 48 |
// CHECK: cuf.shared_memory[%c8{{.*}} : i32] i32 {bindc_name = "c", uniq_name = "_QFshared_staticEc"} -> !fir.ref<i32> |
| 49 |
// CHECK: cuf.shared_memory[%c12{{.*}} : i32] i32 {bindc_name = "d", uniq_name = "_QFshared_staticEd"} -> !fir.ref<i32> |
| 50 |
// CHECK: cuf.shared_memory[%c16{{.*}} : i32] i64 {bindc_name = "e", uniq_name = "_QFshared_staticEe"} -> !fir.ref<i64> |
| 51 |
// CHECK: cuf.shared_memory[%c24{{.*}} : i32] f32 {bindc_name = "r", uniq_name = "_QFshared_staticEr"} -> !fir.ref<f32> |
| 52 |
// CHECK: gpu.return |
| 53 |
// CHECK: } |
| 54 |
// CHECK: fir.global internal @_QPshared_static__shared_mem(dense<0> : vector<28xi8>) {alignment = 8 : i64, data_attr = #cuf.cuda<shared>} : !fir.array<28xi8> |
| 55 |
// CHECK: } |
| 56 |
// CHECK: } |
| 57 |
|
| 58 |
// ----- |
| 59 |
|
| 60 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, 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<f80, dense<128> : 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<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", gpu.container_module, llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.ident = "flang version 20.0.0 (https://github.com/llvm/llvm-project.git cae351f3453a0a26ec8eb2ddaf773c24a29d929e)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 61 |
gpu.module @cuda_device_mod { |
| 62 |
gpu.func @_QMmPshareddyn(%arg0: !fir.box<!fir.array<?x?xi32>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "a"}, %arg1: !fir.box<!fir.array<?x?xi32>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "b"}, %arg2: i32 {fir.bindc_name = "k"}) attributes {cuf.proc_attr = #cuf.cuda_proc<global>} { |
| 63 |
%c1_i32 = arith.constant 1 : i32 |
| 64 |
%c2_i32 = arith.constant 2 : i32 |
| 65 |
%c0 = arith.constant 0 : index |
| 66 |
%5 = fir.address_of(@_QM__fortran_builtinsE__builtin_blockdim) : !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>> |
| 67 |
%6 = fir.declare %5 {uniq_name = "_QM__fortran_builtinsE__builtin_blockdim"} : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>) -> !fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>> |
| 68 |
%15 = fir.alloca i32 |
| 69 |
%16 = fir.declare %15 {fortran_attrs = #fir.var_attrs<value>, uniq_name = "_QMmFss1Ek"} : (!fir.ref<i32>) -> !fir.ref<i32> |
| 70 |
%27 = fir.coordinate_of %6, x : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>) -> !fir.ref<i32> |
| 71 |
%28 = fir.load %27 : !fir.ref<i32> |
| 72 |
%29 = fir.convert %28 : (i32) -> i64 |
| 73 |
%30 = fir.convert %29 : (i64) -> index |
| 74 |
%31 = arith.cmpi sgt, %30, %c0 : index |
| 75 |
%32 = arith.select %31, %30, %c0 : index |
| 76 |
%33 = fir.coordinate_of %6, y : (!fir.ref<!fir.type<_QM__fortran_builtinsT__builtin_dim3{x:i32,y:i32,z:i32}>>) -> !fir.ref<i32> |
| 77 |
%34 = fir.load %33 : !fir.ref<i32> |
| 78 |
%35 = fir.convert %34 : (i32) -> i64 |
| 79 |
%36 = fir.convert %35 : (i64) -> index |
| 80 |
%37 = arith.cmpi sgt, %36, %c0 : index |
| 81 |
%38 = arith.select %37, %36, %c0 : index |
| 82 |
%39 = cuf.shared_memory !fir.array<?x?xi32>, %32, %38 : index, index {bindc_name = "s1", uniq_name = "_QMmFss1Es1"} -> !fir.ref<!fir.array<?x?xi32>> |
| 83 |
%40 = fir.shape %32, %38 : (index, index) -> !fir.shape<2> |
| 84 |
%41 = fir.declare %39(%40) {data_attr = #cuf.cuda<shared>, uniq_name = "_QMmFss1Es1"} : (!fir.ref<!fir.array<?x?xi32>>, !fir.shape<2>) -> !fir.ref<!fir.array<?x?xi32>> |
| 85 |
%42 = fir.load %16 : !fir.ref<i32> |
| 86 |
%43 = arith.muli %42, %c2_i32 : i32 |
| 87 |
%44 = fir.convert %43 : (i32) -> i64 |
| 88 |
%45 = fir.convert %44 : (i64) -> index |
| 89 |
%46 = arith.cmpi sgt, %45, %c0 : index |
| 90 |
%47 = arith.select %46, %45, %c0 : index |
| 91 |
%48 = fir.load %16 : !fir.ref<i32> |
| 92 |
%49 = fir.convert %48 : (i32) -> i64 |
| 93 |
%50 = fir.convert %49 : (i64) -> index |
| 94 |
%51 = arith.cmpi sgt, %50, %c0 : index |
| 95 |
%52 = arith.select %51, %50, %c0 : index |
| 96 |
%53 = cuf.shared_memory !fir.array<?x?xi32>, %47, %52 : index, index {bindc_name = "s2", uniq_name = "_QMmFss1Es2"} -> !fir.ref<!fir.array<?x?xi32>> |
| 97 |
gpu.return |
| 98 |
} |
| 99 |
} |
| 100 |
} |
| 101 |
|
| 102 |
// CHECK: gpu.func @_QMmPshareddyn(%arg0: !fir.box<!fir.array<?x?xi32>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "a"}, %arg1: !fir.box<!fir.array<?x?xi32>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "b"}, %arg2: i32 {fir.bindc_name = "k"}) attributes {cuf.proc_attr = #cuf.cuda_proc<global>} { |
| 103 |
// CHECK: %[[EXTENT0:.*]] = arith.select |
| 104 |
// CHECK: %[[EXTENT1:.*]] = arith.select |
| 105 |
// CHECK: %[[SIZE_EXTENT:.*]] = arith.muli %c4{{.*}}, %[[EXTENT0]] : index |
| 106 |
// CHECK: %[[DYNSIZE:.*]] = arith.muli %[[SIZE_EXTENT]], %[[EXTENT1]] : index |
| 107 |
// CHECK: cuf.shared_memory[%c0{{.*}} : i32] !fir.array<?x?xi32>, %9, %15 : index, index {bindc_name = "s1", uniq_name = "_QMmFss1Es1"} -> !fir.ref<!fir.array<?x?xi32>> |
| 108 |
// CHECK: %[[CONV_DYNSIZE:.*]] = fir.convert %[[DYNSIZE]] : (index) -> i32 |
| 109 |
// CHECK: cuf.shared_memory[%[[CONV_DYNSIZE]] : i32] !fir.array<?x?xi32>, %26, %31 : index, index {bindc_name = "s2", uniq_name = "_QMmFss1Es2"} -> !fir.ref<!fir.array<?x?xi32>> |
| 110 |
|
| 111 |
// ----- |
| 112 |
|
| 113 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr, 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<f80, dense<128> : 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<i128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<"dlti.endianness", "little">, #dlti.dl_entry<"dlti.stack_alignment", 128 : i64>>, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", gpu.container_module, llvm.data_layout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-i128:128-f80:128-n8:16:32:64-S128", llvm.ident = "flang version 20.0.0 (https://github.com/llvm/llvm-project.git cae351f3453a0a26ec8eb2ddaf773c24a29d929e)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 114 |
gpu.module @cuda_device_mod { |
| 115 |
gpu.func @_QPnoshared() kernel { |
| 116 |
gpu.return |
| 117 |
} |
| 118 |
} |
| 119 |
} |
| 120 |
|
| 121 |
// CHECK-LABEL: gpu.func @_QPnoshared() |
| 122 |
// CHECK-NOT: fir.global internal @_QPnoshared__shared_mem |
| 123 |
|
| 124 |
|