Setting the environment variable AF_CUDA_MAX_JIT_LEN to any value less than 2 makes any ArrayFire code produce SIGSEGV (Segmentation Fault).
Description
- The problem also occurs with the OpenCL backend and
AF_OPENCL_MAX_JIT_LEN.
- The problem does not occur with the CPU backend and
AF_CPU_MAX_JIT_LEN.
- Using a debugger, I can see the error appears to be some form of infinite recursion:
Thread 1 received signal SIGSEGV, Segmentation fault.
0x00007ffcfa6ae179 in afcuda!??5af@@YA?AVarray@0@AEBV10@AEBE@Z () from C:\Program Files\ArrayFire\v3\lib\afcuda.dll
(gdb) bt
#0 0x00007ffcfa6ae179 in afcuda!??5af@@YA?AVarray@0@AEBV10@AEBE@Z () from C:\Program Files\ArrayFire\v3\lib\afcuda.dll
#1 0x00007ffcfa6a9b2b in afcuda!??5af@@YA?AVarray@0@AEBV10@AEBE@Z () from C:\Program Files\ArrayFire\v3\lib\afcuda.dll
#2 0x00007ffcfa6a9c1c in afcuda!??5af@@YA?AVarray@0@AEBV10@AEBE@Z () from C:\Program Files\ArrayFire\v3\lib\afcuda.dll
#3 0x00007ffcf9516e1d in afcuda!??5af@@YA?AVarray@0@AEBV10@AEBE@Z () from C:\Program Files\ArrayFire\v3\lib\afcuda.dll
#4 0x00007ffcf9517009 in afcuda!??5af@@YA?AVarray@0@AEBV10@AEBE@Z () from C:\Program Files\ArrayFire\v3\lib\afcuda.dll
...
- Did you build ArrayFire yourself or did you use the official installers: Official installers
- Which backend is experiencing this issue: CUDA and OpenCL
- Do you have a workaround: No (besides the obvious "don't do that")
- Can the bug be reproduced reliably on your system: Yes
- A clear and concise description of what you expected to happen: I expect it to either disable the JIT entirely, or if this is not possible, at least throw an exception to signal an invalid environment variable
- Run your executable with AF_TRACE=all and AF_PRINT_ERRORS=1 environment
variables set.
- Screenshot or terminal output of the results:
[platform][1608204098][023044] [ ..\src\backend\common\DependencyModule.cpp(99) ] Attempting to load: forge.dll
[platform][1608204098][023044] [ ..\src\backend\common\DependencyModule.cpp(102) ] Found: forge.dll
[platform][1608204098][023044] [ ..\src\backend\cuda\device_manager.cpp(428) ] CUDA Driver supports up to CUDA 10.1 ArrayFire CUDA Runtime 10.1
[platform][1608204098][023044] [ ..\src\backend\cuda\device_manager.cpp(495) ] Found 1 CUDA devices
[platform][1608204098][023044] [ ..\src\backend\cuda\device_manager.cpp(521) ] Found device: GeForce GTX 1070 (8 GB | ~6256.71 GFLOPs | 15 SMs)
[platform][1608204098][023044] [ ..\src\backend\cuda\device_manager.cpp(556) ] AF_CUDA_DEFAULT_DEVICE:
[platform][1608204098][023044] [ ..\src\backend\cuda\device_manager.cpp(575) ] Default device: 0(GeForce GTX 1070)
[mem][1608204099][023044] [ ..\src\backend\cuda\memory.cpp(158) ] nativeAlloc: 1 KB 0xb04e00000
[jit][1608204099][023044] [ ..\src\backend\cuda\compile_module.cpp(430) ] {18389128331537752529 : loaded from C:\Users\CORENT~1.SCH\AppData\Local\Temp\\ArrayFire\KER18389128331537752529_CU_61_AF_37.cubin for GeForce GTX 1070 }
Reproducible Code and/or Steps
Compile the following with afcuda and run the following code with the environment variable AF_CUDA_MAX_JIT_LEN set to 1
#include <arrayfire.h>
#include <iostream>
int main() {
try {
af::array a = af::constant(1, 10);
af::array b = af::constant(2, 10);
af::array c = a * b;
std::cout << af::sum<double>(c) << std::endl;
} catch (const af::exception& e) {
std::cerr << e.what() << std::endl;
} catch (const std::exception& e) {
std::cerr << e.what() << std::endl;
} catch (...) {
std::cerr << "Unknown error" << std::endl;
}
return 1;
}System Information
Please provide the following information:
- ArrayFire version: 3.7.2
- Devices installed on the system: Nvidia GeForce GTX 1070 (8GB)
- (optional) Output from the af::info() function if applicable.
ArrayFire v3.7.2 (CUDA, 64-bit Windows, build 218dd2c)
Platform: CUDA Runtime 10.1, Driver: 10010
[0] GeForce GTX 1070, 8192 MB, CUDA Compute 6.1
- Output from the following scripts:
Windows: nvidia-smi
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 432.00 Driver Version: 432.00 CUDA Version: 10.1 |
|-------------------------------+----------------------+----------------------+
| GPU Name TCC/WDDM | Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce GTX 1070 WDDM | 00000000:65:00.0 Off | N/A |
| 0% 49C P8 14W / 151W | 559MiB / 8192MiB | 2% Default |
+-------------------------------+----------------------+----------------------+
Checklist
Setting the environment variable
AF_CUDA_MAX_JIT_LENto any value less than 2 makes any ArrayFire code produce SIGSEGV (Segmentation Fault).Description
AF_OPENCL_MAX_JIT_LEN.AF_CPU_MAX_JIT_LEN.variables set.
Reproducible Code and/or Steps
Compile the following with
afcudaand run the following code with the environment variableAF_CUDA_MAX_JIT_LENset to1System Information
Please provide the following information:
Windows: nvidia-smi
Checklist