Swap get_multi_function and node_gather_link_searches things, just to less mix them/
Did not noticed difference with _mm_load_si128\
I had some trouble while implementing a Potrace curves converter since i stated from the generation of the bits image from the bool field (even had to deep into swar and multiplication method to shift 8 books to char). This makes things much more easier! Even IndexMask is not really necessary, at least before this will be a field kind of\
Wonder if this can have more integration with offset indices as just a chessboard for them\
Wonder if popcount check for all-range-or-all-not will be faster in benchmarks\
Note: the SWAP can be scaled even for 128 and 256 and 512 bits, we only need register, mask, shift and add operations. SWAR popcount is log2 asymptotic algorithm\
In the most worst case where number of the gaps is equal to the number of the set bits and distribution is perfectly uniform (1, 0, 1, 0, 1, 0, ...) number of the ranges equal to the number of the set bits and maximum that is possible (half of the possible set bits). So this seems like x2 oversized buffer.
Does we have some check that bool_ is 16-bytes aligned? Think here is have to be additional prefix handling.
static void customdata_weld( |
|
const CustomData *source, CustomData *dest, const int *src_indices, int count, int dest_index) |
|
{ |
|
if (count == 1) { |
|
CustomData_copy_data(source, dest, src_indices[0], dest_index, 1); |
|
return; |
|
} |
|
CustomData_interp(source, dest, (const int *)src_indices, nullptr, nullptr, count, dest_index); |
I do not know what is your expectations, attached file looks different from the report and i do not see any bug in the file, so i'll close this as not a bug.
Please submit a new report and…
THe fact that face is there or not is defined by combination of tracing and order of the points in curve, so the fact of different result for different positions does not seems like a bug.