[LLVMbugs] [Bug 209] [loadvn/inline/gcse] Slow optimizations with extremely large basic blocks

bugzilla-daemon at cs.uiuc.edu bugzilla-daemon at cs.uiuc.edu
Wed Feb 4 16:46:44 PST 2004


sabre at nondot.org changed:

           What    |Removed                     |Added
             Status|ASSIGNED                    |RESOLVED
         Resolution|                            |FIXED
            Summary|Slow optimizations with     |[loadvn/inline/gcse] Slow
                   |extremely large basic blocks|optimizations with extremely
                   |                            |large basic blocks

------- Additional Comments From sabre at nondot.org  2004-02-04 18:46 -------
Overall, the patches listed in this trail of this bug have fixed all of the
noted efficiency problems in optimizers when they handle large basic blocks.

Specifically, the inliner sped up from 21.7s -> 0.82s, SRoA sped up from 25.5s
-> 0.46s, and loadvn/GCSE sped up from 246.3s -> 43.3s.  This reduced the total
time to gccas the testcase from >300s to 45s.   All of these numbers are with a
release build.

As noted in the bug, the timings for loadvn/GCSE are still quite unreasonable,
but this isn't specific to large-basic-block programs, and requires
infrastructure improvements to fix (specifically GCSE needs a way to update
VN/AA analyses on the fly).  This will have to happen sometime in the future.


------- You are receiving this mail because: -------
You are on the CC list for the bug, or are watching someone who is.

More information about the llvm-bugs mailing list