131 lines · plain
1// RUN: fir-opt --split-input-file --target-rewrite="target=x86_64-unknown-linux-gnu" %s | FileCheck %s2 3module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", 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.target_triple = "x86_64-unknown-linux-gnu"} {4 5gpu.module @testmod {6 gpu.func @_QPvcpowdk(%arg0: !fir.ref<complex<f64>> {cuf.data_attr = #cuf.cuda<device>, fir.bindc_name = "a"}) attributes {cuf.proc_attr = #cuf.cuda_proc<global>} {7 %0 = fir.alloca i648 %1 = fir.load %0 : !fir.ref<i64>9 %2 = fir.load %arg0 : !fir.ref<complex<f64>>10 %3 = fir.call @_FortranAzpowk(%2, %1) fastmath<contract> : (complex<f64>, i64) -> complex<f64>11 gpu.return12 }13 func.func private @_FortranAzpowk(complex<f64>, i64) -> complex<f64> attributes {fir.bindc_name = "_FortranAzpowk", fir.runtime}14}15 16// CHECK-LABEL: gpu.func @_QPvcpowdk17// CHECK: %{{.*}} = fir.call @_FortranAzpowk(%{{.*}}, %{{.*}}, %{{.*}}) fastmath<contract> : (f64, f64, i64) -> tuple<f64, f64>18// CHECK: func.func private @_FortranAzpowk(f64, f64, i64) -> tuple<f64, f64> attributes {fir.bindc_name = "_FortranAzpowk", fir.runtime}19}20 21// -----22 23module attributes {fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", 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.target_triple = "x86_64-unknown-linux-gnu"} {24 25gpu.module @testmod {26 gpu.func @_QPtest(%arg0: complex<f64>) -> (complex<f64>) {27 gpu.return %arg0 : complex<f64>28 }29}30}31 32// CHECK-LABEL: gpu.func @_QPtest33// CHECK-SAME: (%arg0: f64, %arg1: f64) -> tuple<f64, f64> {34// CHECK: gpu.return %{{.*}} : tuple<f64, f64>35 36// -----37module attributes {gpu.container_module, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", 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.target_triple = "x86_64-unknown-linux-gnu"} {38 39gpu.module @testmod {40 gpu.func @_QPtest(%arg0: complex<f64>) -> () kernel {41 gpu.return42 }43}44 45func.func @main(%arg0: complex<f64>) {46 %0 = llvm.mlir.constant(0 : i64) : i6447 %1 = llvm.mlir.constant(0 : i32) : i3248 gpu.launch_func @testmod::@_QPtest blocks in (%0, %0, %0) threads in (%0, %0, %0) : i64 dynamic_shared_memory_size %1 args(%arg0 : complex<f64>) {cuf.proc_attr = #cuf.cuda_proc<global>}49 return50}51 52}53 54// CHECK-LABEL: gpu.func @_QPtest55// CHECK-SAME: (%arg0: f64, %arg1: f64) kernel {56// CHECK: gpu.return57// CHECK: gpu.launch_func @testmod::@_QPtest blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) : i64 dynamic_shared_memory_size %{{.*}} args(%{{.*}} : f64, %{{.*}} : f64) {cuf.proc_attr = #cuf.cuda_proc<global>}58 59// -----60 61module attributes {gpu.container_module, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", 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.target_triple = "x86_64-unknown-linux-gnu"} {62 gpu.module @testmod {63 gpu.func @_QMbarPfoo(%arg0: f32, %arg1: !fir.ref<!fir.array<100xf32>>, %arg2: !fir.boxchar<1>) workgroup(%arg3 : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}) {64 %c0 = arith.constant 0 : index65 memref.store %arg0, %arg3[%c0] : memref<1xf32, #gpu.address_space<workgroup>>66 gpu.return67 }68// CHECK-LABEL: gpu.func @_QMbarPfoo(69// CHECK-SAME: %{{.*}}: f32, %{{.*}}: !fir.ref<!fir.array<100xf32>>, %[[CHAR:.*]]: !fir.ref<!fir.char<1,?>>, %[[LENGTH:.*]]: i64) workgroup(%[[WORKGROUP:.*]] : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}) {70// CHECK: %{{.*}} = fir.emboxchar %[[CHAR]], %[[LENGTH]] : (!fir.ref<!fir.char<1,?>>, i64) -> !fir.boxchar<1>71// CHECK: memref.store %{{.*}}, %[[WORKGROUP]][%{{.*}}] : memref<1xf32, #gpu.address_space<workgroup>>72 73 gpu.func @_QMbarPfoo2(%arg0: f32, %arg1: !fir.ref<!fir.array<100xf32>>, %arg2: !fir.boxchar<1>) workgroup(%arg3 : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}, %arg4 : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}) {74 %c0 = arith.constant 0 : index75 memref.store %arg0, %arg3[%c0] : memref<1xf32, #gpu.address_space<workgroup>>76 memref.store %arg0, %arg4[%c0] : memref<1xf32, #gpu.address_space<workgroup>>77 gpu.return78 }79// CHECK-LABEL: gpu.func @_QMbarPfoo2(80// CHECK-SAME: %{{.*}}: f32, %{{.*}}: !fir.ref<!fir.array<100xf32>>, %[[CHAR:.*]]: !fir.ref<!fir.char<1,?>>, %[[LENGTH:.*]]: i64) workgroup(%[[WG1:.*]] : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}, %[[WG2:.*]] : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}) {81// CHECK: %{{.*}} = fir.emboxchar %[[CHAR]], %[[LENGTH]] : (!fir.ref<!fir.char<1,?>>, i64) -> !fir.boxchar<1>82// CHECK: memref.store %{{.*}}, %[[WG1]][%{{.*}}] : memref<1xf32, #gpu.address_space<workgroup>>83// CHECK: memref.store %{{.*}}, %[[WG2]][%{{.*}}] : memref<1xf32, #gpu.address_space<workgroup>>84 85 gpu.func @_QMbarPprivate(%arg0: f32, %arg1: !fir.boxchar<1>) workgroup(%arg2 : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}) private(%arg3 : memref<1xf32, #gpu.address_space<private>> {llvm.align = 16 : i32}) {86 %c0 = arith.constant 0 : index87 memref.store %arg0, %arg2[%c0] : memref<1xf32, #gpu.address_space<workgroup>>88 memref.store %arg0, %arg3[%c0] : memref<1xf32, #gpu.address_space<private>>89 gpu.return90 }91// CHECK-LABEL: gpu.func @_QMbarPprivate(92// CHECK-SAME: %{{.*}}: f32, %[[CHAR:.*]]: !fir.ref<!fir.char<1,?>>, %[[LENGTH:.*]]: i64) workgroup(%[[WG:.*]] : memref<1xf32, #gpu.address_space<workgroup>> {llvm.align = 16 : i32}) private(%[[PRIVATE:.*]] : memref<1xf32, #gpu.address_space<private>> {llvm.align = 16 : i32}) {93// CHECK: %{{.*}} = fir.emboxchar %[[CHAR]], %[[LENGTH]] : (!fir.ref<!fir.char<1,?>>, i64) -> !fir.boxchar<1>94// CHECK: memref.store %{{.*}}, %[[WG]][%{{.*}}] : memref<1xf32, #gpu.address_space<workgroup>>95// CHECK: memref.store %{{.*}}, %[[PRIVATE]][%{{.*}}] : memref<1xf32, #gpu.address_space<private>>96 97 gpu.func @test_with_char_proc(%arg0: f32, %arg1: tuple<() -> (), i64> {fir.char_proc}) workgroup(%arg2 : memref<1xf32, #gpu.address_space<workgroup>>) {98 %c0 = arith.constant 0 : index99 memref.store %arg0, %arg2[%c0] : memref<1xf32, #gpu.address_space<workgroup>>100 gpu.return101 }102// CHECK-LABEL: gpu.func @test_with_char_proc(103// CHECK-SAME: %{{.*}}: f32, %[[CHARPROC:.*]]: () -> () {fir.char_proc}, %[[LENGTH:.*]]: i64) workgroup(%[[WG:.*]] : memref<1xf32, #gpu.address_space<workgroup>>) {104// CHECK: %{{.*}} = fir.undefined tuple<() -> (), i64>105// CHECK: %{{.*}} = fir.insert_value %{{.*}}, %[[CHARPROC]], [0 : index] : (tuple<() -> (), i64>, () -> ()) -> tuple<() -> (), i64>106// CHECK: %{{.*}} = fir.insert_value %{{.*}}, %[[LENGTH]], [1 : index] : (tuple<() -> (), i64>, i64) -> tuple<() -> (), i64>107// CHECK: memref.store %{{.*}}, %[[WG]][%{{.*}}] : memref<1xf32, #gpu.address_space<workgroup>>108 }109}110 111// -----112 113module attributes {gpu.container_module, fir.defaultkind = "a1c4d8i4l4r4", fir.kindmap = "", 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.target_triple = "x86_64-unknown-linux-gnu"} {114 gpu.module @testmod {115 gpu.func @_QPtest(%arg0: complex<f32>) -> () kernel {116 gpu.return117 }118 }119 func.func @main(%arg0: complex<f32>) {120 %0 = llvm.mlir.constant(0 : i64) : i64121 %1 = llvm.mlir.constant(0 : i32) : i32122 %2 = fir.alloca i64123 %3 = cuf.stream_cast %2 : !fir.ref<i64>124 %4 = gpu.launch_func async [%3] @testmod::@_QPtest blocks in (%0, %0, %0) threads in (%0, %0, %0) : i64 dynamic_shared_memory_size %1 args(%arg0 : complex<f32>) {cuf.proc_attr = #cuf.cuda_proc<global>}125 return126 }127}128 129// CHECK-LABEL: func.func @main130// CHECK: %{{.*}} = gpu.launch_func async [%{{.*}}] @testmod::@_QPtest blocks in (%{{.*}}, %{{.*}}, %{{.*}}) threads in (%{{.*}}, %{{.*}}, %{{.*}}) : i64 dynamic_shared_memory_size %{{.*}} args(%{{.*}} : !fir.vector<2:f32>) {cuf.proc_attr = #cuf.cuda_proc<global>}131