[llvm-bugs] [Bug 26483] New: CUDA __device__ lambdas generating incorrect code
via llvm-bugs
llvm-bugs at lists.llvm.org
Thu Feb 4 13:52:40 PST 2016
https://llvm.org/bugs/show_bug.cgi?id=26483
Bug ID: 26483
Summary: CUDA __device__ lambdas generating incorrect code
Product: clang
Version: unspecified
Hardware: PC
OS: Linux
Status: NEW
Severity: normal
Priority: P
Component: -New Bugs
Assignee: unassignedclangbugs at nondot.org
Reporter: justin.lebar at gmail.com
CC: llvm-bugs at lists.llvm.org
Classification: Unclassified
The following CUDA code, which passes a __device__ functor to a kernel,
compiles fine but does not give the correct output.
When I compile and run it, I get the following output:
Error 3: invalid device function
Error: 0 2883584
Error: 1 0
Error: 2 2883584
Error: 3 0
Error: 4 3105024
Error: 5 0
Error: 6 0
Error: 7 0
Error: 8 0
Error: 9 0
Error: 10 0
Error: 11 0
Error: 12 0
Error: 13 0
Error: 14 0
Error: 15 0
The "error 3 invalid device function" is probably the root of the problem; it
seems like we may not be running any device code at all.
#include <cuda_runtime.h>
#include <cuda.h>
#include <cstdio>
template<class Functor>
__global__ void foo(const Functor f) {
f(blockIdx.x*blockDim.x+threadIdx.x);
}
template<class Functor>
void run(const Functor& f) {
foo<Functor> <<<256,256>>> (f);
}
int main() {
int* h_a = new int[256*256];
int d_c = 9;
int* d_a;
cudaMalloc(&d_a,256*256*sizeof(int));
run([=] (int i) __device__ {
d_a[i] = d_c;
});
cudaMemcpy(h_a,d_a,256*256*sizeof(int),cudaMemcpyDeviceToHost);
printf("Error 3: %s\n",cudaGetErrorString(cudaGetLastError()));
for(int i=0;i<16;i++)
if(h_a[i]!=d_c)
printf("Error: %i %i\n",i,h_a[i]);
}
--
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20160204/4c08fdd8/attachment.html>
More information about the llvm-bugs
mailing list