About runtime error nvJitLinkAddData

[lintianfang]

Hello,

I am using vs 2022, win 10, cuda 12.4, the solution has been compiled successfully and I met an issue when I start to run it, the following error is shown in the commandline window:

error: nvJitLinkAddData(handle, NVJITLINK_INPUT_LTOIR, (void *)module->ltoir, module->ltoirSize, "module label") failed with error 6
error: ERROR 89 in LTOIR arch newer than target arch

May I ask should I change some settings during cmake or build?

Thank you very much for your help!

Best wishes,

Tianfang.

[Justin-Akridge]

In the file include/CudaModularProgram.h comment out line --gpu-architexture=compute_89 and then uncomment line --gpu-architecture=computer_86

nvrtcCreateProgram(&prog, source.c_str(), name.c_str(), 0, NULL, NULL); std::vector<const char*> opts = { // "--gpu-architecture=compute_89", "--gpu-architecture=compute_86", }

Assuming that you have 86 gpu architecture.

rebuild and run

[m-schuetz]

Thanks @Justin-Akridge !

Here the line: https://github.com/m-schuetz/SimLOD/blob/main/include/CudaModularProgram.h#L85

Best is probably to automatically pick your GPU's arch by taking this piece of code and paste it over to line 85:

SimLOD/include/CudaModularProgram.h

Line 222 in eb9fde2

int arch = major * 10 + minor;

Something like:

int major = 0;
int minor = 0;
cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice);
cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice);

string strArch = format("--gpu-architecture=compute_{}{}", major, minor);

nvrtcProgram prog;
string source = readFile(path);
nvrtcCreateProgram(&prog, source.c_str(), name.c_str(), 0, NULL, NULL);
std::vector<const char*> opts = { 
	strArch.c_str(),
	//"--gpu-architecture=compute_89",
	// "--gpu-architecture=compute_86",
	"--use_fast_math",
	"--extra-device-vectorization",
	"-lineinfo",
	cuda_include.c_str(),
	optInclude.c_str(),
	"-I ./",
	"--relocatable-device-code=true",
	"-default-device",
	"-dlto", 
	"--std=c++20",
	"--disable-warnings",
};

So that the same architecture is used in both places.

[lintianfang]

Dear all,

Thank you very much for your answers! @Justin-Akridge comment out that line and uncommet the line below do not work for my machine, but automatically picking my GPU's arch works @m-schuetz .

Thank you very much again for your help!

Best wishes,

Tianfang.

[lintianfang]

Sorry again, one more issue when I added code in line 85, the comamnd line shows infinite loop of error：invalid resource handle

// this cuDevice is also copied from line 222, otherwise the cuDevice cannot be recognized.
CUdevice cuDevice;
cuDeviceGet(&cuDevice, 0);

int major = 0;
int minor = 0;
cuDeviceGetAttribute(&major, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MAJOR, cuDevice);
cuDeviceGetAttribute(&minor, CU_DEVICE_ATTRIBUTE_COMPUTE_CAPABILITY_MINOR, cuDevice);

string strArch = format("--gpu-architecture=compute_{}{}", major, minor);

nvrtcProgram prog;
string source = readFile(path);
nvrtcCreateProgram(&prog, source.c_str(), name.c_str(), 0, NULL, NULL);
std::vector<const char*> opts = {
strArch.c_str(),
//"--gpu-architecture=compute_89",
// "--gpu-architecture=compute_86",
"--use_fast_math",
"--extra-device-vectorization",
"-lineinfo",
cuda_include.c_str(),
optInclude.c_str(),
"-I ./",
"--relocatable-device-code=true",
"-default-device",
"-dlto",
"--std=c++20",
"--disable-warnings",
};

The GUI is open, and I drag&drop a point cloud, the issue happened after drop the point cloud.

Best wishes,

Tianfang.