fix(gpu): change a type to avoid possible issues when compression is executed for a large batch of LWEs #3043
Conversation
|
|
||
| // Modulus switch | ||
| int size = num_glwes * compression_params.glwe_dimension * | ||
| auto size = num_glwes * compression_params.glwe_dimension * |
There was a problem hiding this comment.
uint32_t seems to be fine to capture the result of that mul, otherwise we might need to cast one element to uint64_t
There was a problem hiding this comment.
This looks ok indeed, the first prod will always remain small and total_lwe_bodies_count would have to be insanely big to make int insufficient for this 🤔
There was a problem hiding this comment.
int is not terrible. We still can fit a lot of LWEs in there before overflowing. However, we don't usually use that type in the rest of the library and doubling the space for size would not be bad.
There was a problem hiding this comment.
yup i am a bit paranoid with the int too, prefer the uint32_t at least.
There was a problem hiding this comment.
Yes let's go for uint32_t instead of int or auto
…executed for a large batch of LWEs
pdroalves
force-pushed
the
pa/fix/sizetype
branch
from
5e9b10c to
6bc3230
Compare
3 participants
closes: please link all relevant issues
PR content/description
Check-list: