https://gcc.gnu.org/bugzilla/show_bug.cgi?id=32803
--- Comment #9 from CVS Commits ---
The master branch has been updated by Roger Sayle :
https://gcc.gnu.org/g:e742722f76c70be303248da7ca4842198d4fd1cc
commit r12-6057-ge742722f76c70be303248da7ca4842198d4fd1cc
Author: Roger Sayle
Date: Sat De
--- Comment #8 from ubizjak at gmail dot com 2009-09-17 11:32 ---
(In reply to comment #2)
> stop closing bugs you have no understanding of.
Then please explain, how you plan to avoid:
a) partial register stall in the case of mov $X, al:
put your explanation here ->
b) partial flag s
--- Comment #7 from dean at arctic dot org 2009-09-17 11:06 ---
(In reply to comment #6)
> Without accounting for the push/pop in CFI you will not be able to accurately
> process asynchronous unwind events.
>
just to be sure i understand -- you're saying a %rsp modification in a binary
--- Comment #6 from rguenth at gcc dot gnu dot org 2009-09-17 10:57 ---
Without accounting for the push/pop in CFI you will not be able to accurately
process asynchronous unwind events.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=32803
--- Comment #5 from dean at arctic dot org 2009-09-17 10:54 ---
(In reply to comment #4)
>0: b8 80 ff ff ff mov$0xff80,%eax
>5: 6a 80 pushq $0xff80
>7: 58 pop%rax
>
should be:
0: 48 c7 c0 8
--- Comment #4 from dean at arctic dot org 2009-09-17 10:51 ---
(In reply to comment #3)
> For the push/pop you need to add 2 bytes to the CFI which makes it useless
> compared to mov $imm32,reg.
>
note that my push/pop example said -128..127 and "push $imm8"... not imm32.
GNU assembl
--- Comment #3 from rguenth at gcc dot gnu dot org 2009-09-17 10:40 ---
For the push/pop you need to add 2 bytes to the CFI which makes it useless
compared to mov $imm32,reg.
--
rguenth at gcc dot gnu dot org changed:
What|Removed |Added
--
--- Comment #2 from dean at arctic dot org 2009-09-17 10:28 ---
stop closing bugs you have no understanding of.
--
dean at arctic dot org changed:
What|Removed |Added
--- Comment #1 from ubizjak at gmail dot com 2009-09-17 10:07 ---
All these sequences are _MUCH_ slower than load immediate, so the tradeoff is
not acceptable, even for -Os.
--
ubizjak at gmail dot com changed:
What|Removed |Added
-
