We had to fix the way we read the SNP buffers due to USAN issues. Specifically, we cannot read the buffer in any arbitrary location and interpret as a certain type since the address must be allgned. As a fix, we create a stack variable and perform std::memcpy. This seemed to introduce a minor (but non-negligible) slowdown. It would be nice to not rely on memcpy and just reinterpret the buffer. The only true solution is to pad the buffer cleverly then.
We had to fix the way we read the SNP buffers due to USAN issues. Specifically, we cannot read the buffer in any arbitrary location and interpret as a certain type since the address must be allgned. As a fix, we create a stack variable and perform
std::memcpy
. This seemed to introduce a minor (but non-negligible) slowdown. It would be nice to not rely on memcpy and just reinterpret the buffer. The only true solution is to pad the buffer cleverly then.