Open Jiaao-Bai opened 2 weeks ago
if "& (capacity - 1)" clear the lower bits, then step 0 run twice
I'm not the author of this code, but I can give some comments. In my understanding, this is a variant of the linear probing method for hash. Line 117 is used to probe different hash addresses. Since the author assumes that the capacity is a power of 2, "& (capacity - 1)" is equivalent to "% capacity". I don't think the step 0 will run twice, it should probe a different address when the loop continue.
Describe the bug in gpu_cache/src/static_hash_table.cu, line 117, why slot accumulate again? it seems like slot already in next group before line 117![企业微信截图_89159d0f-506f-4a92-a122-33ecd72b4782](https://github.com/NVIDIA-Merlin/HugeCTR/assets/60597682/7029347b-0100-40bc-bd0d-7d640ab894ca)
To Reproduce
Expected behavior
Screenshots
Environment (please complete the following information):
Additional context