Mercurial > hg > CbC > CbC_llvm

comparison clang/test/CodeGenCUDA/device-stub.cu @ 236:c4bab56944e8 llvm-original

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM 16
author kono
date Wed, 09 Nov 2022 17:45:10 +0900
parents 79ff65ed7e25
children 1f2b6ac9f198
comparison
equal deleted inserted replaced
232:70dce7da266c 236:c4bab56944e8
1 // RUN: echo "GPU binary would be here" > %t 1 // RUN: echo -n "GPU binary would be here." > %t
2 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 2 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
3 // RUN: -target-sdk-version=8.0 -fcuda-include-gpubinary %t -o - \ 3 // RUN: -target-sdk-version=8.0 -fcuda-include-gpubinary %t -o - \
4 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \ 4 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \
5 // RUN: --check-prefixes=ALL,LNX,NORDC,CUDA,CUDANORDC,CUDA-OLD 5 // RUN: --check-prefixes=ALL,LNX,NORDC,CUDA,CUDANORDC,CUDA-OLD
6 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 6 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
7 // RUN: -target-sdk-version=8.0 -fcuda-include-gpubinary %t \ 7 // RUN: -target-sdk-version=8.0 -fcuda-include-gpubinary %t \
8 // RUN: -o - -DNOGLOBALS \ 8 // RUN: -o - -DNOGLOBALS \
9 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \ 9 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \
10 // RUN: -check-prefixes=NOGLOBALS,CUDANOGLOBALS 10 // RUN: -check-prefixes=NOGLOBALS,CUDANOGLOBALS
11 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 11 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
12 // RUN: -target-sdk-version=8.0 -fgpu-rdc -fcuda-include-gpubinary %t \ 12 // RUN: -target-sdk-version=8.0 -fgpu-rdc -fcuda-include-gpubinary %t \
13 // RUN: -o - \ 13 // RUN: -o - \
14 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \ 14 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \
15 // RUN: --check-prefixes=ALL,LNX,RDC,CUDA,CUDARDC,CUDA-OLD 15 // RUN: --check-prefixes=ALL,LNX,RDC,CUDA,CUDARDC,CUDA-OLD
16 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 16 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
17 // RUN: -target-sdk-version=8.0 -o - \ 17 // RUN: -target-sdk-version=8.0 -o - \
18 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefix=NOGPUBIN 18 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefix=NOGPUBIN
19 19
20 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 20 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
21 // RUN: -target-sdk-version=9.2 -fcuda-include-gpubinary %t -o - \ 21 // RUN: -target-sdk-version=9.2 -fcuda-include-gpubinary %t -o - \
22 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \ 22 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \
23 // RUN: --check-prefixes=ALL,LNX,NORDC,CUDA,CUDANORDC,CUDA-NEW 23 // RUN: --check-prefixes=ALL,LNX,NORDC,CUDA,CUDANORDC,CUDA-NEW
24 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 24 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
25 // RUN: -target-sdk-version=9.2 -fcuda-include-gpubinary %t -o - -DNOGLOBALS \ 25 // RUN: -target-sdk-version=9.2 -fcuda-include-gpubinary %t -o - -DNOGLOBALS \
26 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \ 26 // RUN: | FileCheck -allow-deprecated-dag-overlap %s \
27 // RUN: --check-prefixes=NOGLOBALS,CUDANOGLOBALS 27 // RUN: --check-prefixes=NOGLOBALS,CUDANOGLOBALS
28 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 28 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
29 // RUN: -target-sdk-version=9.2 -fgpu-rdc -fcuda-include-gpubinary %t -o - \ 29 // RUN: -target-sdk-version=9.2 -fgpu-rdc -fcuda-include-gpubinary %t -o - \
30 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \ 30 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \
31 // RUN: --check-prefixes=ALL,LNX,RDC,CUDA,CUDARDC,CUDA-NEW 31 // RUN: --check-prefixes=ALL,LNX,RDC,CUDA,CUDARDC,CUDA-NEW
32 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s -std=c++17 \ 32 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s -std=c++17 \
33 // RUN: -target-sdk-version=9.2 -fcuda-include-gpubinary %t -o - \ 33 // RUN: -target-sdk-version=9.2 -fcuda-include-gpubinary %t -o - \
34 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \ 34 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \
35 // RUN: --check-prefixes=ALL,LNX,NORDC,CUDA,CUDANORDC,CUDA-NEW,LNX_17,NORDC17 35 // RUN: --check-prefixes=ALL,LNX,NORDC,CUDA,CUDANORDC,CUDA-NEW,LNX_17,NORDC17
36 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s -std=c++17 \ 36 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s -std=c++17 \
37 // RUN: -target-sdk-version=9.2 -fgpu-rdc -fcuda-include-gpubinary %t -o - \ 37 // RUN: -target-sdk-version=9.2 -fgpu-rdc -fcuda-include-gpubinary %t -o - \
38 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \ 38 // RUN: | FileCheck %s -allow-deprecated-dag-overlap \
39 // RUN: --check-prefixes=ALL,LNX,RDC,CUDA,CUDARDC,CUDA-NEW,LNX_17,RDC17 39 // RUN: --check-prefixes=ALL,LNX,RDC,CUDA,CUDARDC,CUDA-NEW,LNX_17,RDC17
40 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 40 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
41 // RUN: -target-sdk-version=9.2 -o - \ 41 // RUN: -target-sdk-version=9.2 -o - \
42 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefix=NOGPUBIN 42 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefix=NOGPUBIN
43 43
44 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 44 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
45 // RUN: -fcuda-include-gpubinary %t -o - -x hip\ 45 // RUN: -fcuda-include-gpubinary %t -o - -x hip\
46 // RUN: | FileCheck -allow-deprecated-dag-overlap %s --check-prefixes=ALL,LNX,NORDC,HIP,HIPEF 46 // RUN: | FileCheck -allow-deprecated-dag-overlap %s --check-prefixes=ALL,LNX,NORDC,HIP,HIPEF
47 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 47 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
48 // RUN: -fcuda-include-gpubinary %t -o - -DNOGLOBALS -x hip \ 48 // RUN: -fcuda-include-gpubinary %t -o - -DNOGLOBALS -x hip \
49 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefixes=NOGLOBALS,HIPNOGLOBALS 49 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefixes=NOGLOBALS,HIPNOGLOBALS
50 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s \ 50 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s \
51 // RUN: -fgpu-rdc -fcuda-include-gpubinary %t -o - -x hip \ 51 // RUN: -fgpu-rdc -fcuda-include-gpubinary %t -o - -x hip \
52 // RUN: | FileCheck -allow-deprecated-dag-overlap %s --check-prefixes=ALL,LNX,RDC,HIP,HIPEF 52 // RUN: | FileCheck -allow-deprecated-dag-overlap %s --check-prefixes=ALL,LNX,RDC,HIP,HIPEF
53 // RUN: %clang_cc1 -triple x86_64-linux-gnu -emit-llvm %s -o - -x hip\ 53 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-linux-gnu -emit-llvm %s -o - -x hip\
54 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefixes=ALL,LNX,NORDC,HIP,HIPNEF 54 // RUN: | FileCheck -allow-deprecated-dag-overlap %s -check-prefixes=ALL,LNX,NORDC,HIP,HIPNEF
55 55
56 // RUN: %clang_cc1 -triple x86_64-pc-windows-msvc -aux-triple amdgcn -emit-llvm %s \ 56 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-pc-windows-msvc -aux-triple amdgcn -emit-llvm %s \
57 // RUN: -fcuda-include-gpubinary %t -o - -x hip\ 57 // RUN: -fcuda-include-gpubinary %t -o - -x hip\
58 // RUN: | FileCheck -allow-deprecated-dag-overlap %s --check-prefixes=ALL,WIN 58 // RUN: | FileCheck -allow-deprecated-dag-overlap %s --check-prefixes=ALL,WIN
59 59
60 // RUN: %clang_cc1 -no-opaque-pointers -triple x86_64-pc-windows-msvc -aux-triple amdgcn -emit-llvm %s \
61 // RUN: -o - -x hip\
62 // RUN: | FileCheck -allow-deprecated-dag-overlap %s --check-prefixes=ALL,WIN,HIP,HIPNEF
63
60 #include "Inputs/cuda.h" 64 #include "Inputs/cuda.h"
65
66 // HIPNEF: $__hip_gpubin_handle = comdat any
61 67
62 #ifndef NOGLOBALS 68 #ifndef NOGLOBALS
63 // NORDC-DAG: @device_var = internal global i32 69 // NORDC-DAG: @device_var = internal global i32
64 // RDC-DAG: @device_var = global i32 70 // RDC-DAG: @device_var = global i32
65 // WIN-DAG: @"?device_var@@3HA" = internal global i32 71 // WIN-DAG: @"?device_var@@3HA" = internal global i32
151 // ALL: @1 = private unnamed_addr constant [11 x i8] c"device_var\00" 157 // ALL: @1 = private unnamed_addr constant [11 x i8] c"device_var\00"
152 // ALL: @2 = private unnamed_addr constant [13 x i8] c"constant_var\00" 158 // ALL: @2 = private unnamed_addr constant [13 x i8] c"constant_var\00"
153 // ALL: @3 = private unnamed_addr constant [19 x i8] c"ext_device_var_def\00" 159 // ALL: @3 = private unnamed_addr constant [19 x i8] c"ext_device_var_def\00"
154 // ALL: @4 = private unnamed_addr constant [21 x i8] c"ext_constant_var_def\00" 160 // ALL: @4 = private unnamed_addr constant [21 x i8] c"ext_constant_var_def\00"
155 // * constant unnamed string with GPU binary 161 // * constant unnamed string with GPU binary
156 // CUDA: @[[FATBIN:.*]] = private constant{{.*GPU binary would be here.*}}\00", 162 // CUDA: @[[FATBIN:.*]] = private constant{{.*}} c"GPU binary would be here.",
157 // HIPEF: @[[FATBIN:.*]] = private constant{{.*GPU binary would be here.*}}\00",{{.*}}align 4096 163 // HIPEF: @[[FATBIN:.*]] = private constant{{.*}} c"GPU binary would be here.",{{.*}}align 4096
158 // HIPNEF: @[[FATBIN:__hip_fatbin]] = external constant i8, section ".hip_fatbin" 164 // HIPNEF: @[[FATBIN:__hip_fatbin]] = external constant i8, section ".hip_fatbin"
159 // CUDANORDC-SAME: section ".nv_fatbin", align 8 165 // CUDANORDC-SAME: section ".nv_fatbin", align 8
160 // CUDARDC-SAME: section "__nv_relfatbin", align 8 166 // CUDARDC-SAME: section "__nv_relfatbin", align 8
161 // * constant struct that wraps GPU binary 167 // * constant struct that wraps GPU binary
162 // ALL: @__[[PREFIX:cuda|hip]]_fatbin_wrapper = internal constant 168 // ALL: @__[[PREFIX:cuda|hip]]_fatbin_wrapper = internal constant
249 // HIP: exit: 255 // HIP: exit:
250 // HIP-NEXT: load i8**, i8*** @__hip_gpubin_handle 256 // HIP-NEXT: load i8**, i8*** @__hip_gpubin_handle
251 // CUDANORDC-NEXT: call void @__[[PREFIX]]_register_globals 257 // CUDANORDC-NEXT: call void @__[[PREFIX]]_register_globals
252 // HIP-NEXT: call void @__[[PREFIX]]_register_globals 258 // HIP-NEXT: call void @__[[PREFIX]]_register_globals
253 // * In separate mode we also register a destructor. 259 // * In separate mode we also register a destructor.
254 // CUDANORDC-NEXT: call i32 @atexit(void (i8*)* @__[[PREFIX]]_module_dtor) 260 // CUDANORDC-NEXT: call i32 @atexit(void ()* @__[[PREFIX]]_module_dtor)
255 // HIP-NEXT: call i32 @atexit(void (i8*)* @__[[PREFIX]]_module_dtor) 261 // HIP-NEXT: call i32 @atexit(void ()* @__[[PREFIX]]_module_dtor)
256 262
257 // With relocatable device code we call __[[PREFIX]]RegisterLinkedBinary%NVModuleID% 263 // With relocatable device code we call __[[PREFIX]]RegisterLinkedBinary%NVModuleID%
258 // CUDARDC: call{{.*}}__[[PREFIX]]RegisterLinkedBinary[[MODULE_ID]]( 264 // CUDARDC: call{{.*}}__[[PREFIX]]RegisterLinkedBinary[[MODULE_ID]](
259 // CUDARDC-SAME: __[[PREFIX]]_register_globals, {{.*}}__[[PREFIX]]_fatbin_wrapper 265 // CUDARDC-SAME: __[[PREFIX]]_register_globals, {{.*}}__[[PREFIX]]_fatbin_wrapper
260 // CUDARDC-SAME: [[MODULE_ID_GLOBAL]] 266 // CUDARDC-SAME: [[MODULE_ID_GLOBAL]]