Closed gogo2 closed 1 year ago
Curious.
After some investigation I think it is an issue with the HLS backend. I filled an internal issue about it. But the SYCL implementation is not using the HLS backend in a supported way. So I don't have a clear idea of whether they will fix it or when the will fix it.
After some investigation I think it is an issue with the HLS backend. I filled an internal issue about it. But the SYCL implementation is not using the HLS backend in a supported way. So I don't have a clear idea of whether they will fix it or when the will fix it.
If this was HLS issue, it seems to be fixed in 2022.2, at least for this test case.
I'm running some simple test with recent
unified/next
and have encountered what's described below. Last time I run those tests was about a year ago and, as far as I remember, results were correct (I believe it was still on opencl flow)Describe the bug At the end of the program
vec_c[].c
should contain the sequence70 74 78 ... 194
, but instead it is78 82 ... 194 0 0
When I comment anything in the kernel code other thanc_acc[id].x = ...
(for loop, other assignments), the results invec_c[].c
become OK. I didn't check if other values are correct.To Reproduce
Environment