| 1 |
// RUN: fir-opt --split-input-file --cuf-gpu-convert-to-llvm %s | FileCheck %s |
| 2 |
|
| 3 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i128, 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<"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 ([email protected]:clementval/llvm-project.git ddcfd4d2dc17bf66cee8c3ef6284118684a2b0e6)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 4 |
llvm.func @_QMmod1Phost_sub() { |
| 5 |
%0 = llvm.mlir.constant(1 : i32) : i32 |
| 6 |
%1 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr |
| 7 |
%2 = llvm.mlir.constant(40 : i64) : i64 |
| 8 |
%3 = llvm.mlir.constant(16 : i32) : i32 |
| 9 |
%4 = llvm.mlir.constant(25 : i32) : i32 |
| 10 |
%5 = llvm.mlir.constant(21 : i32) : i32 |
| 11 |
%6 = llvm.mlir.constant(17 : i32) : i32 |
| 12 |
%7 = llvm.mlir.constant(1 : index) : i64 |
| 13 |
%8 = llvm.mlir.constant(27 : i32) : i32 |
| 14 |
%9 = llvm.mlir.constant(6 : i32) : i32 |
| 15 |
%10 = llvm.mlir.constant(1 : i32) : i32 |
| 16 |
%11 = llvm.mlir.constant(0 : i32) : i32 |
| 17 |
%12 = llvm.mlir.constant(10 : index) : i64 |
| 18 |
%13 = llvm.mlir.addressof @_QQclX91d13f6e74caa2f03965d7a7c6a8fdd5 : !llvm.ptr |
| 19 |
%14 = llvm.call @_FortranACUFMemAlloc(%2, %11, %13, %6) : (i64, i32, !llvm.ptr, i32) -> !llvm.ptr |
| 20 |
%15 = llvm.mlir.constant(10 : index) : i64 |
| 21 |
%16 = llvm.mlir.constant(1 : index) : i64 |
| 22 |
%17 = llvm.alloca %15 x i32 : (i64) -> !llvm.ptr |
| 23 |
%18 = llvm.mlir.undef : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 24 |
%19 = llvm.insertvalue %17, %18[0] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 25 |
%20 = llvm.insertvalue %17, %19[1] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 26 |
%21 = llvm.mlir.constant(0 : index) : i64 |
| 27 |
%22 = llvm.insertvalue %21, %20[2] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 28 |
%23 = llvm.insertvalue %15, %22[3, 0] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 29 |
%24 = llvm.insertvalue %16, %23[4, 0] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 30 |
%25 = llvm.extractvalue %24[1] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 31 |
%26 = llvm.mlir.undef : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 32 |
%27 = llvm.insertvalue %25, %26[0] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 33 |
%28 = llvm.insertvalue %25, %27[1] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 34 |
%29 = llvm.mlir.constant(0 : index) : i64 |
| 35 |
%30 = llvm.insertvalue %29, %28[2] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 36 |
%31 = llvm.mlir.constant(10 : index) : i64 |
| 37 |
%32 = llvm.insertvalue %31, %30[3, 0] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 38 |
%33 = llvm.mlir.constant(1 : index) : i64 |
| 39 |
%34 = llvm.insertvalue %33, %32[4, 0] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 40 |
%35 = llvm.mlir.constant(1 : index) : i64 |
| 41 |
%36 = llvm.mlir.constant(11 : index) : i64 |
| 42 |
%37 = llvm.mlir.constant(1 : index) : i64 |
| 43 |
llvm.br ^bb1(%35 : i64) |
| 44 |
^bb1(%38: i64): // 2 preds: ^bb0, ^bb2 |
| 45 |
%39 = llvm.icmp "slt" %38, %36 : i64 |
| 46 |
llvm.cond_br %39, ^bb2, ^bb3 |
| 47 |
^bb2: // pred: ^bb1 |
| 48 |
%40 = llvm.mlir.constant(-1 : index) : i64 |
| 49 |
%41 = llvm.add %38, %40 : i64 |
| 50 |
%42 = llvm.extractvalue %34[1] : !llvm.struct<(ptr, ptr, i64, array<1 x i64>, array<1 x i64>)> |
| 51 |
%43 = llvm.getelementptr %42[%41] : (!llvm.ptr, i64) -> !llvm.ptr, i32 |
| 52 |
llvm.store %11, %43 : i32, !llvm.ptr |
| 53 |
%44 = llvm.add %38, %37 : i64 |
| 54 |
llvm.br ^bb1(%44 : i64) |
| 55 |
^bb3: // pred: ^bb1 |
| 56 |
%45 = llvm.call @_FortranACUFDataTransferPtrPtr(%14, %25, %2, %11, %13, %5) : (!llvm.ptr, !llvm.ptr, i64, i32, !llvm.ptr, i32) -> !llvm.struct<()> |
| 57 |
gpu.launch_func @cuda_device_mod::@_QMmod1Psub1 blocks in (%7, %7, %7) threads in (%12, %7, %7) : i64 dynamic_shared_memory_size %11 args(%14 : !llvm.ptr) {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 58 |
%46 = llvm.call @_FortranACUFDataTransferPtrPtr(%25, %14, %2, %10, %13, %4) : (!llvm.ptr, !llvm.ptr, i64, i32, !llvm.ptr, i32) -> !llvm.struct<()> |
| 59 |
%47 = llvm.call @_FortranAioBeginExternalListOutput(%9, %13, %8) {fastmathFlags = #llvm.fastmath<contract>} : (i32, !llvm.ptr, i32) -> !llvm.ptr |
| 60 |
%48 = llvm.mlir.constant(9 : i32) : i32 |
| 61 |
%49 = llvm.mlir.zero : !llvm.ptr |
| 62 |
%50 = llvm.getelementptr %49[1] : (!llvm.ptr) -> !llvm.ptr, i32 |
| 63 |
%51 = llvm.ptrtoint %50 : !llvm.ptr to i64 |
| 64 |
%52 = llvm.mlir.undef : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 65 |
%53 = llvm.insertvalue %51, %52[1] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 66 |
%54 = llvm.mlir.constant(20240719 : i32) : i32 |
| 67 |
%55 = llvm.insertvalue %54, %53[2] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 68 |
%56 = llvm.mlir.constant(1 : i32) : i32 |
| 69 |
%57 = llvm.trunc %56 : i32 to i8 |
| 70 |
%58 = llvm.insertvalue %57, %55[3] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 71 |
%59 = llvm.trunc %48 : i32 to i8 |
| 72 |
%60 = llvm.insertvalue %59, %58[4] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 73 |
%61 = llvm.mlir.constant(0 : i32) : i32 |
| 74 |
%62 = llvm.trunc %61 : i32 to i8 |
| 75 |
%63 = llvm.insertvalue %62, %60[5] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 76 |
%64 = llvm.mlir.constant(0 : i32) : i32 |
| 77 |
%65 = llvm.trunc %64 : i32 to i8 |
| 78 |
%66 = llvm.insertvalue %65, %63[6] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 79 |
%67 = llvm.mlir.constant(0 : i64) : i64 |
| 80 |
%68 = llvm.mlir.constant(1 : i64) : i64 |
| 81 |
%69 = llvm.insertvalue %68, %66[7, 0, 0] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 82 |
%70 = llvm.insertvalue %12, %69[7, 0, 1] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 83 |
%71 = llvm.insertvalue %51, %70[7, 0, 2] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 84 |
%72 = llvm.mul %51, %12 : i64 |
| 85 |
%73 = llvm.insertvalue %25, %71[0] : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> |
| 86 |
llvm.store %73, %1 : !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)>, !llvm.ptr |
| 87 |
llvm.return |
| 88 |
} |
| 89 |
llvm.func @_QMmod1Psub1(!llvm.ptr) -> () |
| 90 |
llvm.mlir.global linkonce constant @_QQclX91d13f6e74caa2f03965d7a7c6a8fdd5() {addr_space = 0 : i32} : !llvm.array<2 x i8> { |
| 91 |
%0 = llvm.mlir.constant("a\00") : !llvm.array<2 x i8> |
| 92 |
llvm.return %0 : !llvm.array<2 x i8> |
| 93 |
} |
| 94 |
llvm.func @_FortranAioBeginExternalListOutput(i32, !llvm.ptr, i32) -> !llvm.ptr attributes {fir.io, fir.runtime, sym_visibility = "private"} |
| 95 |
llvm.func @_FortranACUFMemAlloc(i64, i32, !llvm.ptr, i32) -> !llvm.ptr attributes {fir.runtime, sym_visibility = "private"} |
| 96 |
llvm.func @_FortranACUFDataTransferPtrPtr(!llvm.ptr, !llvm.ptr, i64, i32, !llvm.ptr, i32) -> !llvm.struct<()> attributes {fir.runtime, sym_visibility = "private"} |
| 97 |
llvm.func @_FortranACUFMemFree(!llvm.ptr, i32, !llvm.ptr, i32) -> !llvm.struct<()> attributes {fir.runtime, sym_visibility = "private"} |
| 98 |
gpu.binary @cuda_device_mod [#gpu.object<#nvvm.target, "">] |
| 99 |
} |
| 100 |
|
| 101 |
// CHECK-LABEL: _QMmod1Phost_sub |
| 102 |
// CHECK: %[[STRUCT:.*]] = llvm.alloca %{{.*}} x !llvm.struct<(ptr)> : (i32) -> !llvm.ptr |
| 103 |
// CHECK: %[[PARAMS:.*]] = llvm.alloca %{{.*}} x !llvm.ptr : (i32) -> !llvm.ptr |
| 104 |
// CHECK: %[[ZERO:.*]] = llvm.mlir.constant(0 : i32) : i32 |
| 105 |
// CHECK: %[[STRUCT_PTR:.*]] = llvm.getelementptr %[[STRUCT]][%{{.*}}, {{.*}}] : (!llvm.ptr, i32) -> !llvm.ptr, !llvm.struct<(ptr)> |
| 106 |
// CHECK: llvm.store %{{.*}}, %[[STRUCT_PTR]] : !llvm.ptr, !llvm.ptr |
| 107 |
// CHECK: %[[PARAM_PTR:.*]] = llvm.getelementptr %[[PARAMS]][%[[ZERO]]] : (!llvm.ptr, i32) -> !llvm.ptr, !llvm.ptr |
| 108 |
// CHECK: llvm.store %[[STRUCT_PTR]], %[[PARAM_PTR]] : !llvm.ptr, !llvm.ptr |
| 109 |
// CHECK: %[[KERNEL_PTR:.*]] = llvm.mlir.addressof @_QMmod1Psub1 : !llvm.ptr |
| 110 |
// CHECK: %[[NULL:.*]] = llvm.mlir.zero : !llvm.ptr |
| 111 |
// CHECK: llvm.call @_FortranACUFLaunchKernel(%[[KERNEL_PTR]], {{.*}}, %[[PARAMS]], %[[NULL]]) |
| 112 |
|
| 113 |
// ----- |
| 114 |
|
| 115 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<f64, dense<64> : vector<2xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<f16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<f80, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i128, dense<128> : 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 ([email protected]:clementval/llvm-project.git 4116c1370ff76adf1e58eb3c39d0a14721794c70)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 116 |
llvm.func @_FortranACUFLaunchClusterKernel(!llvm.ptr, i64, i64, i64, i64, i64, i64, i64, i64, i64, !llvm.ptr, i32, !llvm.ptr, !llvm.ptr) attributes {sym_visibility = "private"} |
| 117 |
llvm.func @_QMmod1Psub1() attributes {cuf.cluster_dims = #cuf.cluster_dims<x = 2 : i64, y = 2 : i64, z = 1 : i64>} { |
| 118 |
llvm.return |
| 119 |
} |
| 120 |
llvm.func @_QQmain() attributes {fir.bindc_name = "test"} { |
| 121 |
%0 = llvm.mlir.constant(1 : index) : i64 |
| 122 |
%1 = llvm.mlir.constant(2 : index) : i64 |
| 123 |
%2 = llvm.mlir.constant(0 : i32) : i32 |
| 124 |
%3 = llvm.mlir.constant(10 : index) : i64 |
| 125 |
gpu.launch_func @cuda_device_mod::@_QMmod1Psub1 clusters in (%1, %1, %0) blocks in (%3, %3, %0) threads in (%3, %3, %0) : i64 dynamic_shared_memory_size %2 {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 126 |
llvm.return |
| 127 |
} |
| 128 |
gpu.binary @cuda_device_mod [#gpu.object<#nvvm.target, "">] |
| 129 |
} |
| 130 |
|
| 131 |
// CHECK-LABEL: llvm.func @_QQmain() |
| 132 |
// CHECK: %[[KERNEL_PTR:.*]] = llvm.mlir.addressof @_QMmod1Psub1 |
| 133 |
// CHECK: llvm.call @_FortranACUFLaunchClusterKernel(%[[KERNEL_PTR]], {{.*}}) |
| 134 |
|
| 135 |
// ----- |
| 136 |
|
| 137 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i128, 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<"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 ([email protected]:clementval/llvm-project.git ddcfd4d2dc17bf66cee8c3ef6284118684a2b0e6)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 138 |
llvm.func @_QMmod1Phost_sub() { |
| 139 |
%0 = llvm.mlir.constant(1 : i32) : i32 |
| 140 |
%1 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr |
| 141 |
%2 = llvm.mlir.constant(40 : i64) : i64 |
| 142 |
%3 = llvm.mlir.constant(16 : i32) : i32 |
| 143 |
%4 = llvm.mlir.constant(25 : i32) : i32 |
| 144 |
%5 = llvm.mlir.constant(21 : i32) : i32 |
| 145 |
%6 = llvm.mlir.constant(17 : i32) : i32 |
| 146 |
%7 = llvm.mlir.constant(1 : index) : i64 |
| 147 |
%8 = llvm.mlir.constant(27 : i32) : i32 |
| 148 |
%9 = llvm.mlir.constant(6 : i32) : i32 |
| 149 |
%10 = llvm.mlir.constant(1 : i32) : i32 |
| 150 |
%11 = llvm.mlir.constant(0 : i32) : i32 |
| 151 |
%12 = llvm.mlir.constant(10 : index) : i64 |
| 152 |
%13 = llvm.mlir.addressof @_QQclX91d13f6e74caa2f03965d7a7c6a8fdd5 : !llvm.ptr |
| 153 |
%14 = llvm.call @_FortranACUFMemAlloc(%2, %11, %13, %6) : (i64, i32, !llvm.ptr, i32) -> !llvm.ptr |
| 154 |
gpu.launch_func @cuda_device_mod::@_QMmod1Psub1 blocks in (%7, %7, %7) threads in (%12, %7, %7) : i64 dynamic_shared_memory_size %11 args(%14 : !llvm.ptr) {cuf.proc_attr = #cuf.cuda_proc<grid_global>} |
| 155 |
llvm.return |
| 156 |
} |
| 157 |
llvm.func @_QMmod1Psub1(!llvm.ptr) -> () |
| 158 |
llvm.mlir.global linkonce constant @_QQclX91d13f6e74caa2f03965d7a7c6a8fdd5() {addr_space = 0 : i32} : !llvm.array<2 x i8> { |
| 159 |
%0 = llvm.mlir.constant("a\00") : !llvm.array<2 x i8> |
| 160 |
llvm.return %0 : !llvm.array<2 x i8> |
| 161 |
} |
| 162 |
llvm.func @_FortranACUFMemAlloc(i64, i32, !llvm.ptr, i32) -> !llvm.ptr attributes {fir.runtime, sym_visibility = "private"} |
| 163 |
llvm.func @_FortranACUFMemFree(!llvm.ptr, i32, !llvm.ptr, i32) -> !llvm.struct<()> attributes {fir.runtime, sym_visibility = "private"} |
| 164 |
gpu.binary @cuda_device_mod [#gpu.object<#nvvm.target, "">] |
| 165 |
} |
| 166 |
|
| 167 |
// CHECK-LABEL: llvm.func @_QMmod1Phost_sub() |
| 168 |
// CHECK: llvm.call @_FortranACUFLaunchCooperativeKernel |
| 169 |
|
| 170 |
// ----- |
| 171 |
|
| 172 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<f64, dense<64> : vector<2xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<f16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<f80, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i128, dense<128> : 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 ([email protected]:clementval/llvm-project.git 4116c1370ff76adf1e58eb3c39d0a14721794c70)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 173 |
llvm.func @_QMmod1Psub1() attributes {cuf.cluster_dims = #cuf.cluster_dims<x = 2 : i64, y = 2 : i64, z = 1 : i64>} { |
| 174 |
llvm.return |
| 175 |
} |
| 176 |
llvm.func @_QQmain() attributes {fir.bindc_name = "test"} { |
| 177 |
%0 = llvm.mlir.constant(1 : index) : i64 |
| 178 |
%stream = llvm.alloca %0 x i64 : (i64) -> !llvm.ptr |
| 179 |
%1 = llvm.mlir.constant(2 : index) : i64 |
| 180 |
%2 = llvm.mlir.constant(0 : i32) : i32 |
| 181 |
%3 = llvm.mlir.constant(10 : index) : i64 |
| 182 |
%token = cuf.stream_cast %stream : !llvm.ptr |
| 183 |
gpu.launch_func [%token] @cuda_device_mod::@_QMmod1Psub1 blocks in (%3, %3, %0) threads in (%3, %3, %0) : i64 dynamic_shared_memory_size %2 {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 184 |
llvm.return |
| 185 |
} |
| 186 |
gpu.binary @cuda_device_mod [#gpu.object<#nvvm.target, "">] |
| 187 |
} |
| 188 |
|
| 189 |
// CHECK-LABEL: llvm.func @_QQmain() |
| 190 |
// CHECK: %[[STREAM:.*]] = llvm.alloca %{{.*}} x i64 : (i64) -> !llvm.ptr |
| 191 |
// CHECK: %[[KERNEL_PTR:.*]] = llvm.mlir.addressof @_QMmod1Psub1 |
| 192 |
// CHECK: llvm.call @_FortranACUFLaunchKernel(%[[KERNEL_PTR]], %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %[[STREAM]], %{{.*}}, %{{.*}}, %{{.*}}) : (!llvm.ptr, i64, i64, i64, i64, i64, i64, !llvm.ptr, i32, !llvm.ptr, !llvm.ptr) -> () |
| 193 |
|
| 194 |
// ----- |
| 195 |
|
| 196 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<i1, dense<8> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<i128, 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<"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 ([email protected]:clementval/llvm-project.git ddcfd4d2dc17bf66cee8c3ef6284118684a2b0e6)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 197 |
llvm.func @_QMmod1Phost_sub() { |
| 198 |
%0 = llvm.mlir.constant(1 : i32) : i32 |
| 199 |
%one = llvm.mlir.constant(1 : i64) : i64 |
| 200 |
%1 = llvm.alloca %0 x !llvm.struct<(ptr, i64, i32, i8, i8, i8, i8, array<1 x array<3 x i64>>)> {alignment = 8 : i64} : (i32) -> !llvm.ptr |
| 201 |
%stream = llvm.alloca %one x i64 : (i64) -> !llvm.ptr |
| 202 |
%2 = llvm.mlir.constant(40 : i64) : i64 |
| 203 |
%3 = llvm.mlir.constant(16 : i32) : i32 |
| 204 |
%4 = llvm.mlir.constant(25 : i32) : i32 |
| 205 |
%5 = llvm.mlir.constant(21 : i32) : i32 |
| 206 |
%6 = llvm.mlir.constant(17 : i32) : i32 |
| 207 |
%7 = llvm.mlir.constant(1 : index) : i64 |
| 208 |
%8 = llvm.mlir.constant(27 : i32) : i32 |
| 209 |
%9 = llvm.mlir.constant(6 : i32) : i32 |
| 210 |
%10 = llvm.mlir.constant(1 : i32) : i32 |
| 211 |
%11 = llvm.mlir.constant(0 : i32) : i32 |
| 212 |
%12 = llvm.mlir.constant(10 : index) : i64 |
| 213 |
%13 = llvm.mlir.addressof @_QQclX91d13f6e74caa2f03965d7a7c6a8fdd5 : !llvm.ptr |
| 214 |
%14 = llvm.call @_FortranACUFMemAlloc(%2, %11, %13, %6) : (i64, i32, !llvm.ptr, i32) -> !llvm.ptr |
| 215 |
%token = cuf.stream_cast %stream : !llvm.ptr |
| 216 |
gpu.launch_func [%token] @cuda_device_mod::@_QMmod1Psub1 blocks in (%7, %7, %7) threads in (%12, %7, %7) : i64 dynamic_shared_memory_size %11 args(%14 : !llvm.ptr) {cuf.proc_attr = #cuf.cuda_proc<grid_global>} |
| 217 |
llvm.return |
| 218 |
} |
| 219 |
llvm.func @_QMmod1Psub1(!llvm.ptr) -> () |
| 220 |
llvm.mlir.global linkonce constant @_QQclX91d13f6e74caa2f03965d7a7c6a8fdd5() {addr_space = 0 : i32} : !llvm.array<2 x i8> { |
| 221 |
%0 = llvm.mlir.constant("a\00") : !llvm.array<2 x i8> |
| 222 |
llvm.return %0 : !llvm.array<2 x i8> |
| 223 |
} |
| 224 |
llvm.func @_FortranACUFMemAlloc(i64, i32, !llvm.ptr, i32) -> !llvm.ptr attributes {fir.runtime, sym_visibility = "private"} |
| 225 |
llvm.func @_FortranACUFMemFree(!llvm.ptr, i32, !llvm.ptr, i32) -> !llvm.struct<()> attributes {fir.runtime, sym_visibility = "private"} |
| 226 |
gpu.binary @cuda_device_mod [#gpu.object<#nvvm.target, "">] |
| 227 |
} |
| 228 |
|
| 229 |
// CHECK-LABEL: llvm.func @_QMmod1Phost_sub() |
| 230 |
// CHECK: %[[STREAM:.*]] = llvm.alloca %{{.*}} x i64 : (i64) -> !llvm.ptr |
| 231 |
// CHECK: llvm.call @_FortranACUFLaunchCooperativeKernel(%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %[[STREAM]], %{{.*}}, %{{.*}}, %{{.*}}) : (!llvm.ptr, i64, i64, i64, i64, i64, i64, !llvm.ptr, i32, !llvm.ptr, !llvm.ptr) -> () |
| 232 |
|
| 233 |
// ----- |
| 234 |
|
| 235 |
module attributes {dlti.dl_spec = #dlti.dl_spec<#dlti.dl_entry<!llvm.ptr<272>, dense<64> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr, dense<64> : vector<4xi64>>, #dlti.dl_entry<i64, dense<64> : vector<2xi64>>, #dlti.dl_entry<!llvm.ptr<270>, dense<32> : vector<4xi64>>, #dlti.dl_entry<!llvm.ptr<271>, dense<32> : vector<4xi64>>, #dlti.dl_entry<f64, dense<64> : vector<2xi64>>, #dlti.dl_entry<f128, dense<128> : vector<2xi64>>, #dlti.dl_entry<f16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i32, dense<32> : vector<2xi64>>, #dlti.dl_entry<f80, dense<128> : vector<2xi64>>, #dlti.dl_entry<i8, dense<8> : vector<2xi64>>, #dlti.dl_entry<i16, dense<16> : vector<2xi64>>, #dlti.dl_entry<i128, dense<128> : 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 ([email protected]:clementval/llvm-project.git 4116c1370ff76adf1e58eb3c39d0a14721794c70)", llvm.target_triple = "x86_64-unknown-linux-gnu"} { |
| 236 |
llvm.func @_FortranACUFLaunchClusterKernel(!llvm.ptr, i64, i64, i64, i64, i64, i64, i64, i64, i64, !llvm.ptr, i32, !llvm.ptr, !llvm.ptr) attributes {sym_visibility = "private"} |
| 237 |
llvm.func @_QMmod1Psub1() attributes {cuf.cluster_dims = #cuf.cluster_dims<x = 2 : i64, y = 2 : i64, z = 1 : i64>} { |
| 238 |
llvm.return |
| 239 |
} |
| 240 |
llvm.func @_QQmain() attributes {fir.bindc_name = "test"} { |
| 241 |
%0 = llvm.mlir.constant(1 : index) : i64 |
| 242 |
%1 = llvm.mlir.constant(2 : index) : i64 |
| 243 |
%2 = llvm.mlir.constant(0 : i32) : i32 |
| 244 |
%3 = llvm.mlir.constant(10 : index) : i64 |
| 245 |
%stream = llvm.alloca %0 x i64 : (i64) -> !llvm.ptr |
| 246 |
%token = cuf.stream_cast %stream : !llvm.ptr |
| 247 |
%4 = gpu.launch_func async [%token] @cuda_device_mod::@_QMmod1Psub1 blocks in (%3, %3, %0) threads in (%3, %3, %0) : i64 dynamic_shared_memory_size %2 {cuf.proc_attr = #cuf.cuda_proc<global>} |
| 248 |
llvm.return |
| 249 |
} |
| 250 |
gpu.binary @cuda_device_mod [#gpu.object<#nvvm.target, "">] |
| 251 |
} |
| 252 |
|
| 253 |
// CHECK-LABEL: llvm.func @_QQmain() |
| 254 |
// CHECK: %[[STREAM:.*]] = llvm.alloca %{{.*}} x i64 : (i64) -> !llvm.ptr |
| 255 |
// CHECK: llvm.call @_FortranACUFLaunchKernel(%{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %{{.*}}, %[[STREAM]], %{{.*}}, %{{.*}}, %{{.*}}) |