<html>
<head>
<base href="https://bugs.llvm.org/">
</head>
<body><table border="1" cellspacing="0" cellpadding="8">
<tr>
<th>Bug ID</th>
<td><a class="bz_bug_link
bz_status_NEW "
title="NEW - Kernel launch failure with lambdas in CUDA on Windows"
href="https://bugs.llvm.org/show_bug.cgi?id=48866">48866</a>
</td>
</tr>
<tr>
<th>Summary</th>
<td>Kernel launch failure with lambdas in CUDA on Windows
</td>
</tr>
<tr>
<th>Product</th>
<td>clang
</td>
</tr>
<tr>
<th>Version</th>
<td>unspecified
</td>
</tr>
<tr>
<th>Hardware</th>
<td>PC
</td>
</tr>
<tr>
<th>OS</th>
<td>Windows NT
</td>
</tr>
<tr>
<th>Status</th>
<td>NEW
</td>
</tr>
<tr>
<th>Severity</th>
<td>normal
</td>
</tr>
<tr>
<th>Priority</th>
<td>P
</td>
</tr>
<tr>
<th>Component</th>
<td>CUDA
</td>
</tr>
<tr>
<th>Assignee</th>
<td>unassignedclangbugs@nondot.org
</td>
</tr>
<tr>
<th>Reporter</th>
<td>joachim@joameyer.de
</td>
</tr>
<tr>
<th>CC</th>
<td>llvm-bugs@lists.llvm.org
</td>
</tr></table>
<p>
<div>
<pre>Given multiple (nested) lambdas as CUDA kernel parameters, the mangling/naming
seems to be inconsistent between host and device and thus `cudaLaunchKernel`
fails with error code 98. Might be related to the lambda numbering issues,
solved in Clang 10 (<a href="https://reviews.llvm.org/D68818">https://reviews.llvm.org/D68818</a>)?
A rather minimal reproducer is:
```
#include <exception>
#include <iostream>
template <class T>
__global__ void kernel(int *t, int n, T lambda)
{
if(threadIdx.x < n)
t[threadIdx.x] = lambda(t[threadIdx.x]);
}
int main(){
int* t=nullptr;
cudaMalloc(&t,12);
auto lambda = [](int t){
return t++;
};
kernel<<<1, 32>>>(t, 3, lambda);
[&]()
{
auto lambda2 = [](int t){
return t++;
};
kernel<<<1, 32>>>(t, 3, lambda2);
}();
if(cudaGetLastError())
{
std::cout << "err";
cudaFree(t);
std::terminate();
}
cudaFree(t);
return 0;
}
```</pre>
</div>
</p>
<hr>
<span>You are receiving this mail because:</span>
<ul>
<li>You are on the CC list for the bug.</li>
</ul>
</body>
</html>