Fix SIMDWrapper::scalar::types::loop_mask for out of range

demonstrator

in the scalar namespace loop_mask returns always true. In other namespaces, it checks if the access is out-of-range. Consistent behaviour would be if loop_mask(1,1) returned false.

from my side good for testing.

approved this merge request

good point for consistency. but is there a use case where you can actually call scalar::types::loop_mask() with i>=n ?

ran into it when replacing <an LHCb::span that is all-but-last-element of a std::vector> with . And I did that by replacing the proxy.loop_mask() by ...::loop_mask(proxy.offset(), proxy.size() -1).

That then replaces

std::none_of(the_span.begin(), the_span.end(), predicate);

by (quasi code)

std::all_of( zip.begin(), zip.end(), [](const auto proxy)->mask_v { return none( loop_mask(proxy.offset(), proxy.offset() -1) && predicate(...) ); });

(assuming predicate can return mask_v)

In the last iteration of all_of one may encounter a partially empty loop_mask, an empty loop mask (unlucky case where minus one just reached the boundary of the previous simd chunk), or a false (scalar case).

ok :) thanks for the fix then

/ci-test --merge

resolved all threads

assigned to @rmatev

approved this merge request

changed title from bugfix SIMDWrapper::scalar::types::loop_mask for out of range to Fix SIMDWrapper::scalar::types::loop_mask for out of range

added Utilities bug fix labels

unassigned @rmatev and @ahennequ

merged

mentioned in commit e46cf4b5