[llvm-bugs] [Bug 39466] mca hangs on sqrt-rsqrt-rcp-memop.s

[via llvm-bugs]

https://bugs.llvm.org/show_bug.cgi?id=39466

Greg Bedwell <greg.bedwell at sony.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |FIXED
             Status|ASSIGNED                    |RESOLVED

--- Comment #3 from Greg Bedwell <greg.bedwell at sony.com> ---
Yeah.  This a script bug, when it's attempting to align identical blocks. 
Thanks for reporting it!

I reduced the failing tool output to:

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[0] Code Region - test_sqrtss

Timeline view:
                    0123456789          0
Index     0123456789          0123456789 

[1] Code Region - test_sqrtsd

Timeline view:
                    0123456789          0123456
Index     0123456789          0123456789       
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

and 

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
[0] Code Region - test_sqrtss

Timeline view:
                    0123456789    
Index     0123456789          0123

[1] Code Region - test_sqrtsd

Timeline view:
                    0123456789          0
Index     0123456789          0123456789 
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~


The problem is that the block:

```
Timeline view:
                    0123456789          0
Index     0123456789          0123456789 
```

just happens to exist in both sets of blocks so gets identified as a matching
block which it attempts to align even though they aren't really related.  This
has the affect of knocking all of the intermediate blocks out of alignment,
which it then attempts to re-align, thus knocking that one out of alignment,
and so on.

Fixed in r345499 by preventing it from trying to align something it's already
previously ever tried to align.

Note that your example will still fail as currently written because there is
already a BDVER2-prefixed line in the test so it will error out due to not
being able to find a unique prefix to represent the output.  Giving the line a
unique prefix instead produces a nice looking test file now.

No other test file output was affected by this change.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20181029/f4f062a5/attachment-0001.html>