79 lines · plain
1// RUN: mlir-opt %s -convert-gpu-to-rocdl='runtime=HIP' -split-input-file | FileCheck %s2 3// CHECK-LABEL: gpu.module @test_module4gpu.module @test_module {5 // CHECK-DAG: llvm.mlir.global internal constant @[[$PRINT_GLOBAL0:[A-Za-z0-9_]+]]("Hello, world\0A\00")6 // CHECK-DAG: llvm.mlir.global internal constant @[[$PRINT_GLOBAL1:[A-Za-z0-9_]+]]("Hello: %d\0A\00")7 // CHECK-DAG: llvm.func @__ockl_printf_append_args(i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i648 // CHECK-DAG: llvm.func @__ockl_printf_append_string_n(i64, !llvm.ptr, i64, i32) -> i649 // CHECK-DAG: llvm.func @__ockl_printf_begin(i64) -> i6410 11 // CHECK-LABEL: func @test_const_printf12 gpu.func @test_const_printf() {13 // CHECK: %[[CST0:.*]] = llvm.mlir.constant(0 : i64) : i6414 // CHECK-NEXT: %[[DESC0:.*]] = llvm.call @__ockl_printf_begin(%0) : (i64) -> i6415 // CHECK-NEXT: %[[FORMATSTR:.*]] = llvm.mlir.addressof @[[$PRINT_GLOBAL0]] : !llvm.ptr16 // CHECK-NEXT: %[[FORMATSTART:.*]] = llvm.getelementptr %[[FORMATSTR]][0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<14 x i8>17 // CHECK-NEXT: %[[FORMATLEN:.*]] = llvm.mlir.constant(14 : i64) : i6418 // CHECK-NEXT: %[[ISLAST:.*]] = llvm.mlir.constant(1 : i32) : i3219 // CHECK-NEXT: %[[ISNTLAST:.*]] = llvm.mlir.constant(0 : i32) : i3220 // CHECK-NEXT: %{{.*}} = llvm.call @__ockl_printf_append_string_n(%[[DESC0]], %[[FORMATSTART]], %[[FORMATLEN]], %[[ISLAST]]) : (i64, !llvm.ptr, i64, i32) -> i6421 gpu.printf "Hello, world\n"22 gpu.return23 }24 25 26 // CHECK-LABEL: func @test_printf27 // CHECK: (%[[ARG0:.*]]: i32)28 gpu.func @test_printf(%arg0: i32) {29 // CHECK: %[[CST0:.*]] = llvm.mlir.constant(0 : i64) : i6430 // CHECK-NEXT: %[[DESC0:.*]] = llvm.call @__ockl_printf_begin(%0) : (i64) -> i6431 // CHECK-NEXT: %[[FORMATSTR:.*]] = llvm.mlir.addressof @[[$PRINT_GLOBAL1]] : !llvm.ptr32 // CHECK-NEXT: %[[FORMATSTART:.*]] = llvm.getelementptr %[[FORMATSTR]][0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<11 x i8>33 // CHECK-NEXT: %[[FORMATLEN:.*]] = llvm.mlir.constant(11 : i64) : i6434 // CHECK-NEXT: %[[ISLAST:.*]] = llvm.mlir.constant(1 : i32) : i3235 // CHECK-NEXT: %[[ISNTLAST:.*]] = llvm.mlir.constant(0 : i32) : i3236 // CHECK-NEXT: %[[DESC1:.*]] = llvm.call @__ockl_printf_append_string_n(%[[DESC0]], %[[FORMATSTART]], %[[FORMATLEN]], %[[ISNTLAST]]) : (i64, !llvm.ptr, i64, i32) -> i6437 // CHECK-NEXT: %[[NARGS1:.*]] = llvm.mlir.constant(1 : i32) : i3238 // CHECK-NEXT: %[[ARG0_64:.*]] = llvm.zext %[[ARG0]] : i32 to i6439 // CHECK-NEXT: %{{.*}} = llvm.call @__ockl_printf_append_args(%[[DESC1]], %[[NARGS1]], %[[ARG0_64]], %[[CST0]], %[[CST0]], %[[CST0]], %[[CST0]], %[[CST0]], %[[CST0]], %[[ISLAST]]) : (i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i6440 gpu.printf "Hello: %d\n", %arg0 : i3241 gpu.return42 }43}44 45// -----46 47// The bulitin.module we're targetting is wrapped in a fake gpu.module48// because the convert-gpu-to-rocdl pass only runs an `gpu.module` ops,49// even though the printf patterns could run in other contexts.50 51// CHECK-LABEL: gpu.module @fake_gpu_module_for_test52// CHECK-LABEL: builtin.module @test_module53gpu.module @fake_gpu_module_for_test {54builtin.module @test_module {55 // CHECK-DAG: llvm.mlir.global internal constant @[[$PRINT_GLOBAL1:[A-Za-z0-9_]+]]("Hello: %d\0A\00")56 // CHECK-DAG: llvm.func @__ockl_printf_append_args(i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i6457 // CHECK-DAG: llvm.func @__ockl_printf_append_string_n(i64, !llvm.ptr, i64, i32) -> i6458 // CHECK-DAG: llvm.func @__ockl_printf_begin(i64) -> i6459 60 // CHECK-LABEL: llvm.func @test_printf61 // CHECK: (%[[ARG0:.*]]: i32)62 llvm.func @test_printf(%arg0: i32) {63 // CHECK: %[[CST0:.*]] = llvm.mlir.constant(0 : i64) : i6464 // CHECK-NEXT: %[[DESC0:.*]] = llvm.call @__ockl_printf_begin(%0) : (i64) -> i6465 // CHECK-NEXT: %[[FORMATSTR:.*]] = llvm.mlir.addressof @[[$PRINT_GLOBAL1]] : !llvm.ptr66 // CHECK-NEXT: %[[FORMATSTART:.*]] = llvm.getelementptr %[[FORMATSTR]][0, 0] : (!llvm.ptr) -> !llvm.ptr, !llvm.array<11 x i8>67 // CHECK-NEXT: %[[FORMATLEN:.*]] = llvm.mlir.constant(11 : i64) : i6468 // CHECK-NEXT: %[[ISLAST:.*]] = llvm.mlir.constant(1 : i32) : i3269 // CHECK-NEXT: %[[ISNTLAST:.*]] = llvm.mlir.constant(0 : i32) : i3270 // CHECK-NEXT: %[[DESC1:.*]] = llvm.call @__ockl_printf_append_string_n(%[[DESC0]], %[[FORMATSTART]], %[[FORMATLEN]], %[[ISNTLAST]]) : (i64, !llvm.ptr, i64, i32) -> i6471 // CHECK-NEXT: %[[NARGS1:.*]] = llvm.mlir.constant(1 : i32) : i3272 // CHECK-NEXT: %[[ARG0_64:.*]] = llvm.zext %[[ARG0]] : i32 to i6473 // CHECK-NEXT: %{{.*}} = llvm.call @__ockl_printf_append_args(%[[DESC1]], %[[NARGS1]], %[[ARG0_64]], %[[CST0]], %[[CST0]], %[[CST0]], %[[CST0]], %[[CST0]], %[[CST0]], %[[ISLAST]]) : (i64, i32, i64, i64, i64, i64, i64, i64, i64, i32) -> i6474 gpu.printf "Hello: %d\n", %arg0 : i3275 llvm.return76 }77}78}79