Bug: access violation in some certain conditions #110
Comments
|
Thank you for bug report and solution. Your OpenCL runtime seems to have a bug in clGetPlatformIDs. Though I adoped your pull request, I don't know that it works well always. If the listed order is bad, it may not find GPU device. Should I change |
|
Thank you for your response. It is worth noting that this is not a bug in the OpenCL platform, but rather the behavior of According to the documentation provided at https://man.opencl.org/clGetPlatformIDs.html, the parameter Based on the practice of the OpenCL SDK Context.c, it is recommended not to use a fixed value for Here is the OCLERROR_RET(clGetPlatformIDs(0, NULL, &num_platforms), err, end);
MEM_CHECK(platforms = (cl_platform_id *)malloc(sizeof(cl_platform_id)
* num_platforms),
err, end);
OCLERROR_RET(clGetPlatformIDs(num_platforms, platforms, NULL), err, plat);First, use If you would like me to modify the code to align with the mentioned practice, please let me know, and I can submit another pull request. If you find it too complex, although not the best practice, you can consider changing |
|
Thank you for detailed explanation. I misunderstood the behavior of
I'm OK. Thanks. Changing MAX_DEVICES would be easy. I will modify my code myself. |
|
Yes, however, typically a home PC has one or zero OpenCL devices corresponding to an OpenCL platform (similar to the environment of my laptop that I provided above). As a result, it is highly unlikely to encounter the occurrence of access violation with the Thank you for your response. I believe we can close this issue once the modifications are completed. |
|
I modified my source code to use more stack memory (several tens bytes), because I'm lazy to write many code for dynamic allocation. It's simple, and it will work mostly. I posted the sample .EXE files on GitHub. If you have time, please test them. Thanks many advices. |
|
Thank you very much. I have tested the program.
Since I do not have the debugging symbols for the GUI interface, I am unsure of the source code for the error. However, based on the function call stack frame, it seems that the issue lies with |
|
Thank you for test. I forgot to change a value in my source code. It failed to exit loop, when there are 4 or more OpenCL platforms. It might freeze. I fixed the bug, and updated the GitHub alpha version. I'm sorry for trouble some times. |
|
Thank you, but I tested it and found that the above problem is still not solved (the same behavior), is it possible that there are still some of the codes that have not been changed? |
I'm not sure where is wrong near |
|
Thank you for providing the code snippet. This particular section of code appears to be fine and I have personally tested it successfully. I believe the issue might lie elsewhere. To facilitate a more accurate problem localization, if convenient for you, you may consider compiling a debug version of the program and sharing the debugging symbols on OneDrive. This will allow me to debug the issue further. Alternatively, you could consider removing the section where |
Thank you for suggestion and help. There might be another bug somewhere. To compare error, I made a sample without checking GPU. I just commented out OpenCL section in the |
|
Oh, |
|
This is confusing me. It seems that there might be a problem with the Here is my code // OpenCL 関係
#include <stdio.h>
#include <windows.h>
#include <VersionHelpers.h>
#include <opencl.h>
typedef cl_int(CL_API_CALL* API_clGetPlatformIDs)(cl_uint, cl_platform_id*, cl_uint*);
typedef cl_int(CL_API_CALL* API_clGetDeviceIDs)(cl_platform_id, cl_device_type, cl_uint, cl_device_id*, cl_uint*);
typedef cl_int(CL_API_CALL* API_clGetDeviceInfo)(cl_device_id, cl_device_info, size_t, void*, size_t*);
#define MAX_DEVICE 8
// CPU の拡張機能と GPU の OpenCL driver の存在を調べる
static unsigned int check_cpu(void)
{
unsigned int CPUInfo[4], cpu_flag = 0;
HMODULE hLibOpenCL = NULL;
// CPU の拡張機能を調べる
__cpuid(CPUInfo, 1);
cpu_flag |= (CPUInfo[3] & (1 << 26)) >> 19; // SSE2 対応か
cpu_flag |= (CPUInfo[2] & (1 << 9)) >> 9; // SSSE3 対応か
cpu_flag |= (CPUInfo[2] & (1 << 19)) >> 18; // SSE4.1 対応か
cpu_flag |= (CPUInfo[2] & (1 << 20)) >> 18; // SSE4.2 対応か
cpu_flag |= (CPUInfo[2] & (1 << 1)) << 2; // CLMUL 対応か
if ((CPUInfo[2] & (1 << 28)) != 0) { // AVX 対応なら
if (IsWindows7OrGreater()) { // Windows 7 以降なら AVX2 の判定をする
__cpuid(CPUInfo, 0);
if (CPUInfo[0] >= 7) { // AVX2 用の基本命令領域があるなら
__cpuidex(CPUInfo, 7, 0);
cpu_flag |= (CPUInfo[1] & (1 << 5)) >> 1; // AVX2 対応か
}
}
}
// GPU を調べるけど、32-bit版しかチェックできない・・・
hLibOpenCL = LoadLibraryA("OpenCL.DLL");
if (hLibOpenCL != NULL) {
API_clGetPlatformIDs fn_clGetPlatformIDs;
API_clGetDeviceIDs fn_clGetDeviceIDs;
API_clGetDeviceInfo fn_clGetDeviceInfo;
int i, j;
size_t group_size;
cl_int ret;
cl_uint num_platforms = 0, num_devices = 0, param_value;
cl_platform_id platform_id[MAX_DEVICE];
cl_device_id device_id[MAX_DEVICE];
do {
fn_clGetPlatformIDs = (API_clGetPlatformIDs)GetProcAddress(hLibOpenCL, "clGetPlatformIDs");
if (fn_clGetPlatformIDs == NULL)
break;
fn_clGetDeviceIDs = (API_clGetDeviceIDs)GetProcAddress(hLibOpenCL, "clGetDeviceIDs");
if (fn_clGetDeviceIDs == NULL)
break;
fn_clGetDeviceInfo = (API_clGetDeviceInfo)GetProcAddress(hLibOpenCL, "clGetDeviceInfo");
if (fn_clGetDeviceInfo == NULL)
break;
// OpenCL 環境の数
ret = fn_clGetPlatformIDs(MAX_DEVICE, platform_id, &num_platforms);
if (ret != CL_SUCCESS)
break;
if (num_platforms > MAX_DEVICE)
num_platforms = MAX_DEVICE;
for (i = 0; i < (int)num_platforms; i++) {
// 環境内の OpenCL 対応機器の数
if (fn_clGetDeviceIDs(platform_id[i], CL_DEVICE_TYPE_GPU, MAX_DEVICE, device_id, &num_devices) != CL_SUCCESS)
continue;
if (num_devices > MAX_DEVICE)
num_devices = MAX_DEVICE;
for (j = 0; j < (int)num_devices; j++) {
// デバイスが利用可能か確かめる
ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_AVAILABLE, sizeof(cl_uint), ¶m_value, NULL);
if ((ret != CL_SUCCESS) || (param_value == CL_FALSE))
continue;
ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_COMPILER_AVAILABLE, sizeof(cl_uint), ¶m_value, NULL);
if ((ret != CL_SUCCESS) || (param_value == CL_FALSE))
continue;
// WORK_GROUP_SIZE が 256以上ないとテーブルを作れない
ret = fn_clGetDeviceInfo(device_id[j], CL_DEVICE_MAX_WORK_GROUP_SIZE, sizeof(size_t), &group_size, NULL);
if (ret != CL_SUCCESS)
continue;
if (group_size >= 256) {
cpu_flag |= 0x10000;
i = MAX_DEVICE;
break;
}
}
}
} while (0);
FreeLibrary(hLibOpenCL);
}
return cpu_flag;
}
int main() {
printf("%u", check_cpu());
return 0;
} |
|
Very well, I have identified the root cause of the issue. There is no issue with your modifications to the code. The problem lies with me. Firstly, this issue is only encountered on x86. That is also why I mentioned earlier that there were no issues detected during compilation, as I was using the default x64 compilation method. The primary concern lies in the loading mechanism of the However, it has been reported that the After deleting the registry key value Therefore, if someone happens to have both Intel OneAPI and this program installed, they may encounter the same issue. The solution is to unlink the P.s. ANOTHER SOLUTION: If a version of Indeed, our previous discussion was valuable as it successfully addressed the issue of access violation caused by having too many OpenCL platforms. Thank you very much for your kind words and appreciation. I'm glad that I could assist you. If you believe that we have addressed the issue and no further assistance is required, then we can consider closing the issue. I also apologize for taking up your valuable time. If you have any more questions or need assistance in the future, feel free to reach out. |
|
Oh, I see. I'm glad to know that the problem was solved. My MultiPar.exe is 32-bit application to support old OSes. When someone sees the same error in future, this thread will be helpful. Thank you for fixing the rare bug. If you see another bug while using MultiPar, please report again. |
|
Thank you very much for your great software. |
Short description
Hello, thank you very much for your software. I believe that this is the best alternative to the QuickPar program.
During usage, I noticed that clicking on the "Settings" button causes the program to crash. Additionally, when the GPU acceleration switch is enabled, there is a prompt that par2j64.exe has stopped working while creating the Par2 file. After carefully reading the source code, I identified a potential access violation in
source/par2j/lib_opencl.c.In lines 235-245 of the
source/par2j/lib_opencl.cfile within theinit_OpenCL()function:Please note that
fn_clGetPlatformIDs(MAX_DEVICE, platform_id, &num_platforms)is used to retrieve the number of platforms, and it is limited by MAX_DEVICE. However, in reality, this limitation does not take effect. In my computer environment,num_platforms = 5, whereas the defaultMAX_DEVICEis 3, resulting in an out-of-bounds error in the subsequent for loop.Fixing this is straightforward by modifying the condition of the for loop to
i < (int)min(num_platforms, MAX_DEVICE).I will submit a pull request to address this issue. Thank you!
Environment
The presence of five OpenCL platforms on my computer is the main cause of this issue.
Here is the output of
clinfoAdditional information
As I do not have access to the source code of the MultiPar GUI, I am unable to modify the behavior of the "Settings" button. However, based on your description, I suspect that it is also caused by the same issue. I apologize for the inconvenience, but I hope you proceed with the necessary fix.
The text was updated successfully, but these errors were encountered: