| Issue |
171894
|
| Summary |
[flang] Missed optimizations with conditional do concurrent
|
| Labels |
flang
|
| Assignees |
|
| Reporter |
foxtran
|
Assuming one have two implementations of filling triangle in matrix:
```
subroutine trfill_docon(x, n)
implicit none
integer, intent(in) :: n
real(8), intent(inout) :: x(n,n)
integer :: i, j
do concurrent (i = 1:n, j = 1:n, .not. j <= i)
x(i,j) = x(j,i)
end do
end subroutine trfill_docon
subroutine trfill_dodo(x, n)
implicit none
integer, intent(in) :: n
real(8), intent(inout) :: x(n,n)
integer :: i, j
do i = 1, n
do j = 1, n
if (j <= i) continue
x(i,j) = x(j,i)
end do
end do
end subroutine trfill_dodo
```
Resulting ASM for znver5: https://godbolt.org/z/sEM93f57b
The variant with `do concurrent` has much more branches in comparison with `do-do` example. I would expect the same asm for both examples. Also, asm in GCC is much shorter for `trfill_dodo` routine.
![]()
_______________________________________________
llvm-bugs mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
