http://gcc.gnu.org/bugzilla/show_bug.cgi?id=51499
--- Comment #8 from Ira Rosen <irar at il dot ibm.com> 2011-12-12 11:03:59 UTC --- (In reply to comment #6) > While investigating pr51597, I have found that vectorized loops in programs as > simple as > > subroutine spmmult(x,b,ad) > implicit none > integer, parameter :: nxyz=1008315 > real(8),dimension(nxyz):: x,b,ad > b = ad*x > end subroutine spmmult > !========================================= > > has always an additional non-vectorized loop, This loop has a prologue loop for alignment purposes. > L5: > movsd -8(%rdi,%rax,8), %xmm0 > mulsd -8(%rdx,%rax,8), %xmm0 > movsd %xmm0, -8(%rsi,%rax,8) > addq $1, %rax > cmpq %rcx, %rax > jne L5 > > even when the above loops are unrolled. How can the loop L5 be unrolled if it > is only there for a "scalar epilogue"? It can't be unrolled, since the alignment is unknown, so we don't know the number of iterations of the prologue loop, and, therefore, we don't know the number of iterations of the epilogue.
