> This patch brought back PR libgomp/78468, which had caused its > predecessor to be backed out of gcc-7.
Yes, it's exactly the same mistake: + /* Since the stack is presumed to be aligned before this allocation, + we only need to increase the size of the allocation if the required + alignment is more than the stack alignment. */ The stack is aligned before the allocation but it gets misaligned during the allocation because the dynamic offset is not a multiple of STACK_BOUNDARY. The code had been realigning the stack pointer for more than 2 decades to enforce STACK_BOUNDARY but suddenly stopped again with Wilco's patch. The failure mode is very nasty (random corruption of the stack contents) and there are very likely other affected targets among the ~50 supported ones. -- Eric Botcazou
