https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81357
Qing Zhao <qing.zhao at oracle dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |qing.zhao at oracle dot com
--- Comment #1 from Qing Zhao <qing.zhao at oracle dot com> ---
one question to the submitter:
what optimization level you used to show this issue?
I tried-O0, -O1, -O2, and -O3. I did not see this issue with -O0 and -O1.
starting to see the issue from -O2 and above. (I am using the upstream GCC on
9/27/2017).
the assembly sequence I observed with -O1 (GOOD one) is:
test1:
add w0, w0, 1
uxtw x2, w0
adrp x1, d
str x2, [x1, #:lo12:d]
ret
the assembly sequence I observed with -O2 (BAD one) is:
test1:
adrp x2, d
add w1, w0, 1
mov w0, w1
str x1, [x2, #:lo12:d]
ret
which shows the issue that mentioned by Andrew Pinski (i.e the "mov" insn might
be
unnecessary if the "add" insn is "add w0, w0, 1".
