thread - warp - block

Accelerated Computing CUDA CUDA Programming and Performance
1

hi!

given a block with dim 32, 16:

is it true that i can rely on every thread with the same y is in one warp?
and threads with thread.idx%32 0…15 being in halfwarp 1, 16…31 in halfwarp 2?

but i can´t predict the order in which warps are executed, thus i can´t rely on y being 0,1,2,…,15 while my block gets executed?

thanks for any comments on this,
downforme