[Bug-Report|缺陷反馈][工具检测]: simd_simt_gather_and_adds存在内存越界-asc-devkit-AtomGit

已关闭

[Bug-Report|缺陷反馈][工具检测]: simd_simt_gather_and_adds存在内存越界 #1743

liusa002创建于 8 天前关闭于 6 天前

liusa002

8 天前创建

Thanks for sending an issue! Please fill in the following template to help quickly solve your problem.

Describe the current behavior / 问题描述 (Mandatory / 必填)

使用内存检测拉起算子，报了512条非法写，非误报，需要算子修复。

Environment / 环境信息 (Mandatory / 必填)

HDK 版本：25.1.rc2
CANN 版本：9.2.0
硬件形态：Ascend950DT
操作系统版本：openEuler 24.03

Steps to reproduce the issue / 重现步骤 (Mandatory / 必填)

1.下载并安装CANN包
https://cmc-szv.clouddragon.huawei.com/cmcversion/index/releaseView?deltaId=14913606472500480&isSelect=Software

bash Ascend-cann-toolkit_9.2.0_linux-aarch64.run --install --install-path=/home/Ascend-B070 --quiet
bash Ascend-cann-950-ops_9.2.0_linux-aarch64.run --install --install-path=/home/Ascend-B070 --quiet
source /home/Ascend-B070/cann/set_env.sh

2.下载asc-devkit并切换到对应版本：
git clone https://gitcode.com/cann/asc-devkit.git
git checkout c65e4db9
3.切到算子路径：
cd asc-devkit/examples/05_simd_simt_hybrid/00_introduction/simd_simt_gather_and_adds
4.在CMakeLists.txt中加上检测编译选项--cce-enable-sanitizer -g：

5.编译算子并使用检测拉起：

mkdir -p build && cd build
cmake -DCMAKE_ASC_ARCHITECTURES=dav-3510 ..;make -j36
python3 ../scripts/gen_data.py
mssanitizer -t memcheck --log-level=error --full-backtrace=yes --log-file=san_error.log ./demo
python3 ../scripts/verify_result.py output/output.bin output/golden.bin

6.能够看到类似这样的告警:

====== ERROR: illegal write of size 4
======    at 0x800 on UB in "gather_and_adds_kernel(float*, unsigned int*, float*, unsigned int, unsigned int)"
======    by thread (512,0,0) in block aiv(0-7) on device 3
======    code in pc current 0x458 (serialNo:19)
======    #0 /home/asc-devkit/examples/05_simd_simt_hybrid/00_introduction/simd_simt_gather_and_adds/gather_and_adds.asc:59:32
======    #1 /home/Ascend/cann-9.2.0/tools/bisheng_compiler/lib/clang/15.0.5/include/__clang_cce_simt.h:398:3
======    #2 /home/Ascend-B070/cann-9.2.0/asc/include/simt_api/common_functions.h:39:5
======    #3 /home/asc-devkit/examples/05_simd_simt_hybrid/00_introduction/simd_simt_gather_and_adds/gather_and_adds.asc:107:5

Describe the expected behavior / 预期结果 (Mandatory / 必填)

预期没有非法读告警。

算子168行分配的动态UB大小只有2048字节，而后续105行和111行申请UB空间各需要4096字节，因此gather_output前512个thread写满2048字节，后512个threads越界，每个thread越界写4字节。将168行改为4096就不会报越界，考虑到111行也申请了UB空间，建议改为8192。