Details
-
Type:
Improvement
-
Status:
Closed
-
Priority:
Major
-
Resolution: Fixed
-
Affects Version/s: None
-
Fix Version/s: 3.0
-
Component/s: Compiler: Baseline
-
Labels:None
-
Number of attachments :
Description
Profiling indicates that a significant fraction (perhaps as high as 20%) of execution time in bloat is spent in the out-of-line routines used by the baseline compiler as part of its invokeinterface sequence.
The inline sequence for the common case is only a few instructions, so this should have minimal impact on code space/compile time and get some nice speedups for programs that have flat execution profiles and heavy use of interfaces (ie, bloat, eclipse).
Activity
David Grove
made changes -
| Field | Original Value | New Value |
|---|---|---|
| Status | Open [ 1 ] | In Progress [ 3 ] |
David Grove
made changes -
| Resolution | Fixed [ 1 ] | |
| Status | In Progress [ 3 ] | Closed [ 6 ] |
David Grove
made changes -
| Resolution | Fixed [ 1 ] | |
| Status | Closed [ 6 ] | Reopened [ 4 ] |
David Grove
made changes -
| Fix Version/s | 3.0 [ 13530 ] | |
| Fix Version/s | 2.9.4 [ 14162 ] |
David Grove
made changes -
| Resolution | Fixed [ 1 ] | |
| Status | Reopened [ 4 ] | Closed [ 6 ] |
committed in 14522.
Running bloat on a prototype image on ppc64-aix I see a nice speedup (1144 seconds down to 920 seconds). Unclear how much this will help with a production image, but it should have at least some impact on first iteration time (and is a step towards getting bloat on prototype/prototype-opt to run fast enough that the regression test won't time out).